**Women's Intentions to Engage in Risk-Reducing Behaviours after Receiving Personal Ovarian Cancer Risk Information: An Experimental Survey Study**

### **Ailish Gallagher 1, Jo Waller 2, Ranjit Manchanda 3,4, Ian Jacobs <sup>5</sup> and Saskia Sanderson 1,6,\***


Received: 5 October 2020; Accepted: 24 November 2020; Published: 27 November 2020

**Simple Summary:** Risk stratification using genetic testing to identify women at increased risk of ovarian cancer may increase the number of patients to whom risk-reducing surgery (e.g., salpingo-oophorectomy) may be offered. However, little is known about public acceptability of such approaches. Our online experimental survey aimed to explore whether women aged 45–75 in the general population are willing to undergo ovarian cancer risk assessment, including genetic testing, and whether women's potential acceptance of risk-reducing surgery differs depending on their estimated risk. We looked at whether psychological and cognitive factors mediated women's decision-making. The majority of participants would be interested in having genetic testing. In response to our hypothetical scenarios, a substantial proportion of participants were open to the idea of surgery to reduce risk of ovarian cancer, even if their absolute lifetime risk is only increased from 2% to 5 or 10%.

**Abstract:** Risk stratification using genetic and/or other types of information could identify women at increased ovarian cancer risk. The aim of this study was to examine women's potential reactions to ovarian cancer risk stratification. A total of 1017 women aged 45–75 years took part in an online experimental survey. Women were randomly assigned to one of three experimental conditions describing hypothetical personal results from ovarian cancer risk stratification, and asked to imagine they had received one of three results: (a) 5% lifetime risk due to single nucleotide polymorphisms (SNPs) and lifestyle factors; (b) 10% lifetime risk due to SNPs and lifestyle factors; (c) 10% lifetime risk due to a single rare mutation in a gene. Results: 83% of women indicated interest in having ovarian cancer risk assessment. After receiving their hypothetical risk estimates, 29% of women stated they would have risk-reducing surgery. Choosing risk-reducing surgery over other behavioural responses was associated with having higher surgery self-efficacy and perceived response-efficacy, but not with perceptions of disease threat, i.e., perceived risk or severity, or with experimental condition. A substantial proportion of women age 45–75 years may be open to the idea of surgery to reduce risk of ovarian cancer, even if their absolute lifetime risk is only increased to as little as 5 or 10%.

**Keywords:** risk stratification; genomics; questionnaires; attitudes

#### **1. Introduction**

Ovarian cancer is the sixth most common cancer among women in the UK. The general population lifetime risk of developing ovarian cancer is approximately 2%, and incidence is predicted to rise by 26% in the UK, 14% in Europe, and by 55% worldwide over the next two decades [1]. The risk of ovarian cancer rises with age, increasing significantly in women over 45 years [2]. DNA variants in a number of cancer susceptibility genes are known to be associated with ovarian cancer: women with a high penetrance genetic variant, such as a *BRCA1* or *BRCA2* mutation, are considered to be at high risk for developing breast and ovarian cancer [3–5]. Historically, genetic testing for ovarian cancer risk has been clinically indicated only for women with a strong family history of breast and/or ovarian cancer. However, using a family history based approach misses over half the cancer susceptibility gene (CSG) carriers at risk [6,7], and is associated with restricted access and limited utilization of genetic testing [8]. Additionally, the majority of cases of ovarian cancer do not occur in affected families [9]. There is increasing interest in the idea of adopting a risk-stratified approach to ovarian cancer prevention by offering genetic testing to all women regardless of family history [10–12].

In addition to rare variants of high penetrance, genome-wide association studies have to date identified a number of common single nucleotide polymorphisms (SNPs) associated with slightly increased risk of ovarian cancer [13]. SNP-based information and certain lifestyle factors each increase ovarian cancer risk by a small amount individually, but this becomes clinically significant when the information is combined, e.g., from 2% to between 5% and 10% lifetime risk [9,14,15]. Surgical prevention has been shown to be cost-effective at the 4–5% ovarian cancer risk threshold [16,17]. Newer risk models and recently validated intermediate risk genes can identify individuals at these risk thresholds. Risk stratification using multigene testing to identify women at increased risk of ovarian cancer is potentially more cost- and time-effective than single gene testing and increases the number of patients to whom risk-reducing surgery (e.g., salpingo-oophorectomy) may be offered [18]. While clinical practice has gradually begun to change [19], data on public acceptability of such approaches are limited.

An initial quantitative study assessing attitudes towards population-based genetic testing for ovarian cancer risk in a general population sample found high levels of support for risk-stratified ovarian cancer screening based on prior genetic risk assessment [20]. There is good evidence to suggest that population-wide genetic testing for ovarian cancer is acceptable, feasible and cost-effective amongst Ashkenazi Jewish populations [6,7,21–23]. Preliminary data from the general population also indicate that population-based personalised ovarian cancer risk stratification is feasible, acceptable, has high satisfaction, reduces cancer worry/risk perception, and does not negatively impact psychological health or quality of life [12].

Bilateral risk-reducing salpingo-oophorectomy (surgical removal of the ovaries and fallopian tubes, hereafter referred to as "risk-reducing salpingo-oophorectomy" or "RRSO") is currently recommended as the main and most effective preventative strategy for ovarian cancer in women at increased risk of ovarian cancer such as *BRCA* mutation carriers. RRSO can reduce ovarian cancer risk by 85–90% [24]. Traditionally the most common group of women undergoing surgical prevention have been *BRCA* carriers, who have a 17–44% lifetime risk of ovarian cancer [5,25]. In the UK, women with an estimated lifetime ovarian cancer risk of greater than 10%, who have completed their families, have traditionally been offered risk-reducing surgery [15]. Undertaking surgery on the basis of family history alone in the absence of a known mutation (at lower than *BRCA* levels of risk) has thus been clinical practice in the UK and other countries for many years [25,26]. Recently, the 10% threshold was relaxed to 4–5% [14,15]. A number of new ovarian cancer risk genes have been identified, such as *RAD51C* (lifetime risk 11%) [27], *RAD51D* (lifetime risk 13%), *PALB2* (lifetime risk 5%) [28], and *BRIP1* (lifetime risk 5.8%) [29], testing for which is part of routine clinical practice. RRSO is now offered and being undertaken for these CSGs too. Thus a number of clinical teams now offer RRSO to women in the "intermediate" risk category (5–10%) as well as those in the "high" risk category (over 10%) [15]. Additionally, more complex models using SNP profiles, in combination with other epidemiological

and genetic risk factors, are being validated, which will provide absolute lifetime risk estimates in these ranges in the not too distant future [12].

National screening programmes for ovarian cancer are unavailable. In a large randomised control trial designed to establish the effect of early detection by ovarian screening in the general low-risk population, no conclusive significant impact on mortality from ovarian cancer was found [30], and definitive mortality data are awaited in 2021. Surveillance for those identified as high-risk (or in some cases moderate-to-high-risk) for ovarian cancer currently consists of serial 3–4 monthly serum CA125 (Cancer Antigen 125 protein; a tumour marker) measurement (and annual transvaginal ultrasound) aiming to detect pre-symptomatic cancer in the earlier stages and/or low volume disease where treatment is more effective [31]. This 4 monthly longitudinal CA125 biomarker driven surveillance strategy, using the risk of the ovarian cancer (ROCA) algorithm, may be beneficial in women at high risk of ovarian cancer [31]. We have shown that this is associated with a significant stage shift, which can be a surrogate for improved survival [31]. Identifying those at increased risk using a population wide risk-stratified approach may result in more timely risk reduction options and could have a significant impact on disease burden: modelling suggests that 13% of the female UK population have greater than 4% lifetime risk and 9% have greater than 5% lifetime risk [15]. Manchanda et al. (2018) suggest that, based on National Institute for Health and Care Excellence (NICE) cost-effectiveness guidelines, risk-reducing surgery may be cost effective for postmenopausal women over the age of 50, with a lifetime ovarian cancer risk of ≥5%. Wider implementation of targeted surgical prevention for women at greater than 4–5% lifetime risk threshold provides a huge opportunity for cost-effective targeted primary prevention.

Offering risk stratification to women in the general population, including communicating personal ovarian cancer risk information and offering risk-reducing surgery, has the potential to be a feasible way to reduce ovarian cancer mortality and reduce the population burden of the disease. However, risk stratification will only lead to improved ovarian cancer prevention and early diagnosis if women whose results indicate increased risk take action to reduce their risk. Women with a family history of breast and ovarian cancer have been found to opt for risk reduction surgery, e.g., among *BRCA1* and *BRCA2* mutation carriers, the majority underwent risk-reducing surgery (salpingo-oophorectomy) after their risk was communicated to them [25,32]. However, although quite a lot is known about how genetic risk information impacts psychological wellbeing and behaviours among women from families affected with ovarian (and/or breast) cancer, less is known about how women in the wider non-Jewish population might react to being informed they have an increased genetic risk of ovarian cancer [33–37]. Further research is needed to determine how women in the general population might respond if presented with ovarian cancer risk information indicating they are at high risk based on genetic as well as other risk factors.

Based on research prior to 2016, the evidence does not support the hypothesis that communicating CSG-based risk estimates motivates lifestyle behaviour changes [33,38]. CSG-based risk information also has not been associated with negative psychological outcomes [7,33,38,39]. More recently, a nested study within the Predicting Risk of Cancer at Screening (PROCAS) study was conducted comparing the psychological impact of providing women with personalised breast cancer risk estimates based on: (a) the Tyrer–Cuzick (T–C) risk algorithm including breast density, or (b) T–C including breast density plus SNPs, versus (c) comparison women awaiting results. This study found little evidence of either psychological harm or of differences between women provided with risk estimates based on SNPs versus others. However, women categorised as high-risk were excluded from the study, so no conclusions could be drawn regarding high-risk results specifically. It remains to be seen whether the source of the risk may have impacted psychological factors or if it had an effect on acceptance of the risk information in this study [40]. In another recent study that examined the impact of returning secondary findings (including *BRCA1*/*2*) from genomic sequencing to unselected populations, few adverse psychological effects were found [41].

As an initial step to providing some empirical data on the question of how women in the general population might respond to personal ovarian cancer risk information indicating increased risk (as against moderate risk [40]), we conducted an experimental survey study with women in the general population, using the Extended Parallel Process Model (EPPM) [42] as our theoretical framework, and to inform our selection of variables and measures. The EPPM is a social cognition model of information processing and behaviour: it posits that how individuals react to threatening information is informed by (a) their perceptions of the threat (perceived risk or susceptibility, and perceived severity), and (b) their perceptions of the recommended action to reduce the threat (self-efficacy, i.e., their confidence in their ability to carry out the recommended behaviour, and perceived response efficacy, i.e., their confidence that the recommended behaviour will effectively reduce the threat to their health).

Our specific aims were to: (1) explore whether women in the general population are willing to undergo ovarian cancer risk assessment which includes genetic testing; (2) examine whether women's potential acceptance of risk-reducing surgery differs depending on whether their estimated risk is 5% or 10%; (3) examine whether women's potential acceptance of risk-reducing surgery differs depending on whether their estimated risk is based on a single rare genetic variant of high penetrance or a more complex combination of genetic and non-genetic factors. We also explored whether threat and efficacy cognitions mediated any observed between-group differences, and examined the associations between these cognitions (threat, efficacy) and acceptance of risk-reducing surgery in the sample overall.

#### **2. Results**

#### *2.1. Sample Characteristics*

Table 1 provides an overview of the participant characteristics. Age ranged from 45 to 75 years with a mean of 57.50 (SD = 8.13). The majority were White (95.6%) with 3.8% from other ethnic backgrounds. Educational attainment was fairly evenly split between General Certificate of Secondary Education (GCSE) or equivalent (34.6%), A levels or equivalent (23.8%), and undergraduate degree or equivalent (24.1%). The majority (85.2%) of women were either perimenopausal (beginning menopause) or post-menopausal. See Figure 1 for the Consolidated Standards of Reporting Trials (CONSORT) flow diagram of participants throughout the study.


**Table 1.** Sample characteristics and interest in genetic testing overall and in each randomised experimental group (total *n* = 1017).


**Table 1.** *Cont.*

SNPs: single nucleotide polymorphisms; SD: standard deviation; GCSE: General Certificate of Secondary Education.

**Figure 1.** CONSORT 2010 Flow Diagram.

#### *2.2. Interest in Ovarian Cancer Risk Assessment*

Overall, 83.2% of women indicated they would "yes definitely" (38.0%) or "yes probably" (45.2%) have an ovarian cancer risk assessment if it was offered to them by their general practitioner (GP) on the National Health Service (NHS) (see Table 1 and Figure 2).

**Figure 2.** Interest in ovarian cancer risk assessment.

#### *2.3. Behavioural Response to Personalised Ovarian Cancer Risk Information*

After receiving their hypothetical risk result, 28.5% of women said they would opt for risk-reducing surgery, 33.9% for increased surveillance (transvaginal ultrasound), and 20.9% would make lifestyle changes (e.g., quitting smoking, maintaining a healthy weight; see Figure 3 and Table S1).

**Figure 3.** Behavioural intentions after exposure to hypothetical risk scenario compared between groups.

#### *2.4. Di*ff*erences by Experimental Condition*

Women's intentions to have risk-reducing surgery did not differ significantly between the 5% and 10% multifactorial SNPs + lifestyle groups (27.9% vs. 26.2%, respectively) (χ2(1) = 0.314, *p* = 0.61). Women who received a 10% risk result based on a rare genetic variant were no more likely to opt for RRSO over other risk-reducing options than women who received a 10% risk result based on multifactorial SNPs + lifestyle factors (31.4% vs. 26.2%, respectively) (χ2(1) = 2.512, *p* = 0.13).

#### *2.5. EPPM Variables*

The mean (M) and standard deviation (SD) of the EPPM variables were: perceived risk (M = 3.51, SD = 0.82), perceived severity (M = 4.52, SD = 0.58), perceived response-efficacy (M = 4.03, SD = 0.78), perceived self-efficacy (M = 2.98, SD = 1.34). Means by exposure group are shown in the Supplementary Materials (Table S2).

#### *2.6. Intention to Have Risk-Reducing Surgery (RRSO) versus Other Risk-Management Options*

A binary logistic regression was conducted to investigate what factors were associated with hypothetical intention to have risk-reducing surgery vs. other behavioural options. Independent variables included in the model were age, ethnicity, educational attainment, previous breast and cervical screening participation, experimental group and EPPM variables (perceived risk, perceived severity, self-efficacy, perceived response-efficacy). In unadjusted analyses, women reporting higher perceived risk of ovarian cancer and higher perceived severity of ovarian cancer (i.e., the perceived threat variables), and higher surgery self-efficacy and perceived response-efficacy (i.e., variables relating to perceptions of the risk-reducing behaviour) were more likely than other women to opt for risk-reducing surgery. In the multivariable model, perceived response-efficacy (odds ratio (OR) = 2.22; 95% confidence interval (CI): 1.64–3.00) and self-efficacy (OR = 1.90; 95% CI: 1.63–2.22) remained significantly associated, whereas the perceived threat variables were no longer significantly associated, with choosing risk-reducing surgery over other behavioural options. None of the measured socio-demographic or health-related factors were significantly associated with intention to have surgery (see Table 2).


**Table 2.** Logistic regression predicting likelihood of intending to have risk-reducing surgery vs. other behavioural response (*n* = 1017).


**Table 2.** *Cont.*

\*\* Predictor significant at the 0.01 level (2-tailed), \* Predictor significant at the 0.05 level (2-tailed), CI = Confidence Interval. Ref = reference category.

#### **3. Discussion**

A high proportion (83%) of women in this sample indicated they would be interested in having an ovarian cancer risk assessment if offered by their GP on the NHS. This is consistent with previous research by Meisel et al. (2016), who found that 88% of a general population sample of women in the UK would be interested in genetic testing for ovarian cancer risk if it were offered by the NHS, and included information about breast cancer risk, echoing previous support from qualitative research for the availability of genetic testing and risk-stratified screening [43]. It is also consistent with uptake of genetic testing in our population-based studies [12,21].

We also found that a substantial proportion of British women over the age of 45 years might be open to the idea of having RRSO, even if their absolute lifetime risk were increased from a general population risk of 2% to as little as 5% or 10%. In addition, we also demonstrated in multivariable analyses that perceptions of risk-reducing surgery (self-efficacy and perceived response-efficacy) were independently associated with choosing surgery over other options, whereas the perceived threat of ovarian cancer (perceived risk and perceived severity) was not.

Although over a quarter (29%) of women opted for RRSO, slightly more women opted for surveillance (34%). The observed preference for surveillance may be due to the invasiveness of surgery, and could also potentially be due to the generally positive attitude towards cancer screening in the UK [44]. Lack of detailed information on the efficacy of each risk management option due to the hypothetical nature of this study may have resulted in participants deciding on the less invasive option, i.e., surveillance. Research suggests perceptions about risk-reducing surgery and surveillance are potentially modifiable: Mai et al. (2017) identified misperceptions about ovarian cancer risk and benefits of screening as important factors influencing decisions about risk-reducing surgery versus surveillance [45]. The concept of common genetic variants of low penetrance single nucleotide polymorphisms (SNPs) may be unfamiliar to the majority of individuals in the general public. For example, in a study by French et al. (2018), there was considerable variation in understanding of test results. The role of SNPs in cancer risk may be less familiar to individuals than more widely publicised

rare genetic variants such as those in the *BRCA* genes [40]. Additionally, lifestyle factors may be perceived as being under greater personal control and, therefore, less serious than rare genetic variants.

Our study findings suggest that individuals interpreted the two levels of risk (5% vs. 10%) similarly, with the difference in communicated risk having a non-significant impact on participants' intentions to have RRSO. This supports previous research exploring the effect of risk information on behaviour, suggesting there is not a simple linear relationship between increments in risk and risk perception [39,40].

In addition to the lack of impact on perceived risk of ovarian cancer, we similarly found that different presentations of risk in the hypothetical scenarios (5% SNPs + lifestyle vs. 10% SNPs + lifestyle risk; 10% rare genetic variant vs. 10% SNPs + lifestyle) did not lead to differences in any other cognitive factors considered in the EPPM framework (i.e., perceived severity of ovarian cancer, perceived response-efficacy of risk-reducing surgery, self-efficacy to undertake risk-reducing surgery).

In contrast, we found that, in the sample overall, opting for risk-reducing surgery was associated with higher self-efficacy and higher perceived response-efficacy of risk-reducing surgery. According to the Extended Parallel Processing Model [46], higher perceptions of self-efficacy and/or response-efficacy relating to the recommended behaviour are associated with greater likelihood that systematic processing of threatening (risk) information will occur. Conversely, when perceptions of efficacy are low, people are more likely to avoid threatening risk information. Together, our findings suggest that if women perceive or believe that RRSO is being recommended to them clinically, this may have a greater impact on their decision-making than the details of their risk result (i.e., whether their risk is 5% or 10%, and whether that risk is based on a single rare genetic variant or a more complex combination of SNPs and lifestyle factors). The observation that psychological variables had a greater impact on intentions than the absolute risk numbers suggest that this might be important to consider in any potential future national rollouts. Offering psychological support for those who need it as part of the RRSO discussion and decision-making process is part of routine clinical practice in many centres today. Our study highlights the importance of incorporating this into future national guidelines.

In previous research using hypothetical scenarios, there has been some evidence to suggest that genetic information leads to more deterministic responses than non-genetic information [33,38,47–49]. Our study did not include a non-genetic condition and so does not speak to this aspect of how people respond to personal genetic versus non-genetic information.

The present study had several limitations. The cross-sectional design of the study did not allow for causation to be inferred. However, exploratory experimental studies such as this one can be valuable in informing hypotheses before moving on to study designs designed to trial real risk assessments. The use of hypothetical scenarios was both a strength and a limitation. This study attempted to model a "real life" scenario in which genetic risk information was provided to the general population. However, many of the contextual details and additional resources that accompany risk information are not available in hypothetical scenarios, which may limit the ecological validity of the study. This study was concerned with behavioural intention, as ovarian cancer population surveillance or population-wide genetic testing is not currently clinically available, so actual behaviour could not be measured. The presence of a potential intention behaviour gap is well established for other clinical interventions and cannot be excluded here.

The measures used in this study are adapted from previous research; however, they were almost all single-item measures, which may not be sensitive enough to adequately represent the underlying construct being measured due to the lack of psychometric information (e.g., test-retest reliability, discriminant or convergent validity). Prior knowledge may have an influence on how individuals appraise threatening health information [50]: this study did not measure previous ovarian cancer and genetic risk knowledge or previous genetic testing, which may have had an impact on behavioural intention (however this may be unlikely given this type of genetic testing is not widely available in the UK). In addition, we did not assess understanding of the information provided, and it is possible that some concepts (e.g., SNPs) may not have been well understood.

The sample was predominately White British and, therefore, may not be generalisable to other ethnic groups, given decisions about risk-reducing surgery and psychological effects may differ cross-culturally. In addition, the restricted age-range of the sample limits the generalisability of the findings (e.g., to younger women age 35 years and over who may also be offered risk-reducing surgery if they are at high risk). However, most women from the general population who are at increased risk of ovarian cancer will fall in the intermediate risk (5–10% lifetime risk) category [9]. RRSO at intermediate ovarian cancer risk levels (including for moderate penetrance CSGs) is recommended to be undertaken over the age of 45–50 years [15]. The sample was self-selected and may have had greater interest in the topic than the wider general population; the generalizability of the findings is therefore uncertain. Finally, as with any experimental study, we are unable to rule out the possibility of demand characteristics, i.e., participants responding in a way they think is expected according to their perceptions of the aim of the study. Despite the limitations, this study provides insights on the effects of experimentally manipulating genetic risk information for ovarian cancer on outcomes, comparing different sources and levels of risk on risk management behavioural intentions and psychological variables in the general population.

The information provided before being exposed to the hypothetical risk scenario on ovarian cancer, risk factors, and the efficacy of risk management was necessarily basic and brief, which may be an additional limitation. However, previous research did not find any difference in behavioural outcomes between use of gist and extended versions of decision aids in relation to ovarian cancer risk management [51]. Future research might usefully provide more detailed information containing details about the efficacy of a particular risk management behaviour to encourage "danger control" cognitive processing. This may aid in changing risk management preferences.

Future research might also benefit from including measures of other psychological and cognitive factors as potential predictors of risk management, e.g., causal beliefs. In addition, further research is needed on communicating risk information incorporating genetics to people in the general population outside of traditional clinical genetics department settings. Furthermore, a control group where participants are given a general population-based risk estimate would be useful for future research, as we were unable to compare between the general population risk and increased risk in this study.

The mean age of participants in this study was 57 years, with the majority of participants having completed having children and/or being past childbearing age, with most participants reporting they had begun menopause or were post-menopausal. Future research should explore the psychological and cognitive effects of ovarian cancer risk information being offered to younger women. In addition, there were relatively few women in the oldest age group (71–75 years) in this study, and so it is possible the apparent trend of increasing age being less associated with interest in surgery was due to the study being underpowered. This could also be a topic of investigation in future research.

Risk-reducing surgery, specifically RRSO, is at present the most effective risk management option available to women at increased risk of ovarian cancer. Our findings suggest there are a number of cognitive factors that influence intention to have ovarian cancer risk-reducing clinical interventions, beyond perceptions of risk. Future research should explore other possible factors that may have an impact on decision-making about risk management strategies. It is imperative to identify whether and/or how genetic risk information about common complex diseases will be translated into public health benefit: this is arguably especially urgent for diseases, such as ovarian cancer, which are characterised by being notoriously difficult to detect early and by having a high prevalence of late-stage diagnosis. Combined testing for multiple genetic factors together with lifestyle and other risk factors may lead to the ability to stratify the population for ovarian cancer risk for targeted prevention thus potentially saving lives.

Population testing provides a new paradigm for ovarian cancer prevention and can prevent thousands more cancers than the current clinical approach [52]. Jewish population studies support population testing for CSGs [53]. Our pilot study shows that population testing for lifetime risk of ovarian cancer is feasible, acceptable and has high satisfaction in general population women [12]. However, there is now need for large implementation studies, with long term outcomes, to provide real world evidence and develop context-specific models for implementing this approach for women in the general population. This will valuably inform future policy decisions regarding population-wide risk stratified approaches for risk-adapted ovarian cancer screening and prevention.

#### **4. Materials and Methods**

#### *4.1. Overview*

Participants (*n* = 1017) were women aged 45–75 years recruited via online survey company Survey Sampling International (SSI) in July 2017. An email containing a web link was sent to SSI panellists who fit the study criteria with respect to gender and age, inviting them to take part. The email did not contain information about the topic of the study. Those responding were directed to a short screening questionnaire. Eligible participants were then presented with a consent form. Incentive points, which can be exchanged for shopping vouchers, were awarded to SSI panellists for their time (equivalent to ~£0.50 for this 10 min study).

All participants were given information about ovarian cancer, including that on average around 2% of women will develop ovarian cancer in their lifetime; asked to imagine that they had had an ovarian cancer risk assessment via the NHS; and asked to imagine they had received a result indicating they were at increased risk of ovarian cancer (see Appendix A).

Women were randomly assigned to one of three experimental conditions using a software algorithm (See Figure 1). They were asked to imagine they had undergone an ovarian cancer risk assessment and had received this personalised risk estimate from their GP: (a) 5% ovarian cancer risk due to common genetic variants and lifestyle factors; (b) 10% ovarian cancer risk due to common genetic variants and lifestyle factors; or (c) 10% ovarian cancer risk due to a single rare variant in a cancer susceptibility gene such as *BRCA2* (see Appendix B).

The study was approved by the University College London ethics committee (Project ID Number: 10251/001).

#### *4.2. Inclusion Criteria*

Eligible participants were women aged 45–75 years, with no previous history of breast or ovarian cancer diagnosis. Women who indicated they were unsure of, or had not completed childbearing, were excluded from analyses (*n* = 13).

#### *4.3. Measures*

All measures are shown in Appendix C.

#### 4.3.1. Interest in Ovarian Cancer Risk Assessment

Interest was assessed before exposure to the hypothetical test results, with the item, "If your GP offered you this ovarian cancer risk assessment on the NHS, would you take up the offer?" (adapted from [20]. Response options were "no, definitely not"; "no probably not"; "yes, probably" and "yes, definitely". The information women read before answering the question explained that the risk assessment would involve providing lifestyle information as well as a blood sample for genetic testing (see Appendix A).

#### 4.3.2. Perceived Risk of Ovarian Cancer

Perceived risk was measured using a single item, "If I had just received this personal ovarian cancer risk result, I would feel that my risk of developing ovarian cancer was" adapted from [54]. Responses for these questions were recorded on a 5-point Likert scale with response options ranging from "much lower than other women of my age" to "much higher than women of my age". A higher score on the 5-point scale indicated greater perceived risk.

#### 4.3.3. Perceived Severity of Ovarian Cancer

Perceived severity was measured using two questions adapted from [55]: "Developing ovarian cancer would have major consequences on my life" and "ovarian cancer is a serious condition" with five response options ranging from "strongly agree" to "strongly disagree". A higher score on the (possible scores 1–5) scale indicated greater perceived severity.

#### 4.3.4. Self-Efficacy for Risk-Reducing Surgery

One item, adapted from [56], assessed participants' confidence in their ability to have risk-reducing surgery. Individuals were asked "How confident are you that you would go through with risk-reducing surgery if you were motivated to do so". The response options ranged from "not at all confident" to "extremely confident". A higher score on the scale indicated greater perceived self-efficacy.

#### 4.3.5. Perceived Response-Efficacy of Risk-Reducing Surgery

For perceived response-efficacy of risk-reducing surgery, participants were asked to indicate how effective they felt risk-reducing surgery would be in lowering their ovarian cancer risk using a single item adapted from [56]: "Having surgery to remove your ovaries and fallopian tubes is an effective way to lower your risk of ovarian cancer". The response options were "strongly agree" to "strongly disagree". Items were reverse coded: a higher score on the (possible scores 1–5) scale indicated greater perceived response-efficacy.

#### 4.3.6. Behavioural Intention

To assess women's potential behavioural reactions to their risk results, they were asked: "If I had just received this personal ovarian cancer risk result, I would choose to...". The response options were: "have risk-reducing surgery to remove my ovaries"; "have surveillance such as regular ultrasound scans"; "make lifestyle changes"; "do nothing"; and "I am not sure what I would do".

#### 4.3.7. Demographic and Health Characteristic Measures

Information on demographics was collected from all participants including: age, ethnicity, educational attainment, relationship status, health characteristics, family history of cancer, personal history of cancer, menopause status, and breast and cervical screening attendance. Ethnicity (White vs. other ethnic group), menopause status (pre-menopausal vs. peri/post-menopause) and breast and cervical screening attendance (regular vs. irregular or not yet eligible) were dichotomised.

#### *4.4. Data Analyses*

A power calculation based on the primary binary outcome, intention to have risk-reducing surgery, taking into account group comparisons of three groups, suggested a sample size of 782 was required (medium effect size, power of 90%, alpha of 0.05). All statistical analyses of the data were carried out using SPSS 24. Analyses of variance (ANOVAs) and chi-square tests were conducted to explore between-group differences. Logistic regression was used to explore predictors of willingness to have risk-reducing surgery (vs. other behavioural responses to the risk information). Unadjusted and adjusted models were examined to explore the predictive effect of the experimental group and psychological variables on intention to have surgery and address possible demographic and health-related covariates.

#### **5. Conclusions**

The findings of this study contribute to a growing body of risk stratification research exploring the potential usefulness and clinical utility of population-wide risk assessment incorporating genetic testing alongside other risk factors. The need for risk stratification is perhaps particularly urgent for diseases, such as ovarian cancer, where survival outcomes are poor and population-wide screening for the disease is not currently recommended. We provide initial evidence, suggesting that a substantial proportion of women aged 45 years and over are open to the idea of risk stratification and having surgery to reduce their risk of ovarian cancer in response to increased risk results, even if their absolute lifetime risk is only increased by a few percentage points in absolute terms. Our findings do not speak to other barriers that might prevent women's behavioural intentions or preferences being translated into actions—barriers such as lack of timely access to healthcare services.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6694/12/12/3543/s1, Table S1: Data for Figure 3, Table S2: Mean EPPM variables by experimental group.

**Author Contributions:** Conceptualization, S.S. and J.W.; methodology, J.W., A.G., S.S., R.M.; formal analysis, A.G., J.W., S.S.; resources, I.J., R.M., J.W.; data curation, A.G., S.S., J.W.; writing—original draft preparation, A.G.; writing—review and editing, J.W., S.S., I.J., R.M.; visualization, A.G., J.W., S.S.; supervision, J.W., S.S., I.J., R.M.; project administration, A.G.; funding acquisition, J.W. All authors have read and agreed to the published version of the manuscript.

**Funding:** The fieldwork for this study was funded by Cancer Research UK as part of a programme grant awarded to Professor Jane Wardle.

**Acknowledgments:** We are grateful to the women who took part in the study.

**Conflicts of Interest:** Ian Jacobs is a Director and shareholder in Abcodia, Ltd., a company focused on biomarkers for early detection of cancer. He is a co-inventor of the Risk of Ovarian Cancer Algorithm, which has been licensed to Abcodia by Massachusetts General Hospital, and has a right to a royalty stream. The other authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **Appendix A**

*Ovarian cancer and risk information*

#### *Section 1: Hypothetical scenario*

Please imagine that you have gone to your GP, and they have offered you a new approach to assessing your risk of developing ovarian cancer in the future. When your GP offered this to you, they gave you some written information to help you decide whether or not you want to have the risk assessment done. This information is below. Please read the information, and then answer the question that follows.

*Assessing ovarian cancer risk*

Ovarian cancer is the sixth most common cancer among women in the UK: 2% of women will be diagnosed with ovarian cancer during their lifetime. Ovarian cancer is caused by many genetic and non-genetic factors. Currently, ovarian cancer is often detected at a late stage because symptoms are hard to spot. This means that it is often very hard to treat effectively.

A new ovarian cancer risk assessment has been developed. This risk assessment combines lots of different types of information about you to estimate how likely you are to develop ovarian cancer in your lifetime. The types of information included in the risk assessment include lifestyle factors, rare genetic variants, and common genetic variants.

Lifestyle factors: Lifestyle factors that may increase a woman's risk of ovarian cancer include tobacco smoking and being overweight.

Common genetic variants: Single nucleotide polymorphisms, frequently called SNPs (pronounced "snips"), are the most common type of genetic variation among people. SNPs occur normally throughout a person's DNA. Most SNPs have no effect on health, but some are important to a person's health. Some SNPs can influence a woman's risk of developing ovarian cancer. Individually, each one of these SNPs only influences ovarian cancer risk by a tiny amount, but if a woman has a large number of these SNPs then her risk of ovarian cancer may be increased.

Rare genetic variants: Some ovarian cancers are caused by a rare variant in a person's DNA. These rare variants can have quite a strong effect on a woman's risk of developing ovarian cancer. For example, variants in the *BRCA2* gene can increase a woman's lifetime risk of ovarian cancer up to between 10% and 20%. If you want to have this ovarian cancer risk assessment carried out, you will need to provide your GP with the information they request, including about your lifestyle. You will also need to have a blood test, so that the scientists can see whether you have any of the genetic variants that increase your risk.

### **Appendix B**

#### *Generic Risk Scenarios*

Next, regardless of how you answered in the previous question, please imagine that you have had the ovarian cancer risk assessment done, and your GP has now given you the result from the assessment. Please read your hypothetical result below.

*Your personal ovarian cancer risk assessment: results [Group 1 only]*

Your result indicates that your lifetime risk of developing ovarian cancer is 5%. This is higher than the average risk for women, which is 2%.

Your risk of ovarian cancer is higher than average because you have been found to have at least one lifestyle factor and several common genetic variants which are known to put women at increased risk of developing ovarian cancer.

### *OR*

*Your personal ovarian cancer risk assessment: results [Group 2 only]*

Your result indicates that your lifetime risk of developing ovarian cancer is 10%. This is higher than the average risk for women, which is 2%.

Your risk of ovarian cancer is higher than average because you have been found to have at least one lifestyle factor and several common genetic variants which are known to put women at increased risk of developing ovarian cancer.

#### *OR*

*Your personal ovarian cancer risk assessment: results [Group 3 only]*

Your result indicates that your lifetime risk of developing ovarian cancer is 10%. This is higher than the average risk for women, which is 2%.

Your risk of ovarian cancer is higher than average because you have been found to have a rare genetic variant which is known to put women at increased risk of developing ovarian cancer.

#### AND

#### *[All groups]*

There are several options for women who are at higher than average risk of ovarian cancer.

Risk-reducing surgery involves removing the ovaries and fallopian tubes to prevent ovarian cancer from developing. However, removing the ovaries has downsides. For example, it causes a woman who has not yet been through her menopause to start her menopause (a natural process that usually happens in a woman's early 50 s).

Surveillance includes having a regular (e.g., annual) ultrasound of your ovaries to see if a tumour is present. This ultrasound is usually an internal (transvaginal) ultrasound. Effective screening has not been established. Lifestyle changes include maintaining a healthy weight and quitting smoking. These types of lifestyle changes may reduce a woman's risk of developing ovarian cancer.

#### **Appendix C**

#### *Questionnaire*

Please carefully imagine what you would think and how you would feel if you had received this personal result from the ovarian cancer risk assessment. Now please answer the following questions.

#### *Intention for Genetic Screening*

**Q1. If your GP o**ff**ered you this ovarian cancer risk assessment on the NHS, would you take up the o**ff**er?**


#### **Q2. How likely do you think you are to develop ovarian cancer in your lifetime?**


#### *Behavioural outcome*

#### **Q3. If I had just received this personal ovarian cancer risk result, I would choose to** ...


(Please select one option only)

### *Perceived risk*

### **Q4. If I had just received this personal ovarian cancer risk result, I would feel that my risk of developing ovarian cancer was** ...


#### *Perceived Response e*ffi*cacy*

How much do you agree or disagree with the following statement based on how you would feel if you had received this personal ovarian cancer risk result from the ovarian cancer risk assessment.

#### **Q7. There is little that can be done to prevent ovarian cancer.**


#### **Q8. Having surgery to remove your ovaries and fallopian tubes is an e**ff**ective way to lower your risk of ovarian cancer.**


#### *Perceived Response e*ffi*cacy*

**Q9. Regular screening through transvaginal ultrasound is an e**ff**ective way to lower your risk of ovarian cancer.**


#### *Perceived severity*

#### **Q10. Developing ovarian cancer would have major consequences on my life**


#### *Perceived severity*

#### **Q11. Ovarian cancer is a serious condition.**


#### *Self-e*ffi*cacy*

### **Q12. How confident are you that you could go through with having risk-reducing surgery if you were motivated to do so?**


#### **Q13. Have you ever been diagnosed with cancer? (Please select one)**

Yes No Not sure

**Q14. If yes, what type of cancer is it**/**was it?**

... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... .

#### **Q15. Do you have a family history of cancer?**

*Have anyfirst-degree family member(parents, brothers, sisters, children) orsecond-degree(aunts, uncles, nieces, nephews, grandparents, grandchildren) been diagnosed with cancer.*

Yes

No Not sure

#### **Q15a. If yes, what type(s) of cancer?**

#### ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... .. **Q16. What is your current menstrual status? (Please select one)**

Premenopausal (before menopause; having regular periods) Perimenopause (menopause transition—changes in periods, but have not gone 12 months in a row without a period) Postmenopausal (After menopause; periods have stopped for at least 12 months) Don't know

**Q17. If you have indicated that you have not had a period in the previous 12 months, what age were you at your last period?**

**Q18. If you are still having periods, how often do they occur? (Please respond in days)**

**Q19. Is there a recent change in how often you have periods?**

Yes/No

### **Q20. Was your menopause:**

Spontaneous ("natural") Surgical (removal of both ovaries) Due to chemotherapy or radiation therapy Other n/a—haven't yet reached menopause *(If indicated they are still having periods don't ask this question)*

**Q21. Women aged 25–49 years are invited for cervical screening (also called a smear or Pap test) every 3 years, and women aged 50–64 are invited every 5 years. Which of these statements best describes you?**

I'm up to date with cervical screening I'm overdue for cervical screening I've never been for cervical screening I'm 65 or over so I'm not invited any more

**Q22. Women aged 50–70 years are invited for breast screening (also called a mammogram or mammography) every 3 years. Which of these statements best describes you?**

I'm up to date with breast screening I'm overdue for breast screening I've never been for breast screening I'm under 50 or over 70 so I'm not eligible for breast screening.

#### **Q23. How old are you?**

\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_

### **Q24. How would you describe your ethnic background? (Please select one)**

White British White non-British Black Asian Mixed Other Do not wish to answer

#### **Q25. What is the highest level of education you have achieved? (Please select one)**

No formal qualifications GCSEs/O levels or equivalent A-Levels or equivalent Undergraduate degree or equivalent Postgraduate degree or equivalent Other (please state)

#### **Q26. What is your relationship status? (Please select one)**

Single In a relationship Living with a partner Married Separated/divorced/widowed

#### **Q27. How many children do you have? (Please select one)**

### **Q28. Do you plan to have any (more) children in the future? (Please select one)**

Yes No Not sure

*END OF SURVEY*

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **Identifying Ovarian Cancer in Symptomatic Women: A Systematic Review of Clinical Tools**

**Garth Funston 1,\*, Victoria Hardy 1, Gary Abel 2, Emma J. Crosbie 3,4, Jon Emery 1,5, Willie Hamilton <sup>2</sup> and Fiona M. Walter 1,5**


Received: 12 November 2020; Accepted: 4 December 2020; Published: 8 December 2020

**Simple Summary:** Most women with ovarian cancer are diagnosed after they develop symptoms identifying symptomatic women earlier has the potential to improve outcomes. Tools, ranging from simple symptom checklists to diagnostic prediction models that incorporate tests and risk factors, have been developed to help identify women at increased risk of undiagnosed ovarian cancer. In this review, we systematically identified studies evaluating these tools and then compared the reported diagnostic performance of tools. All included studies had some quality concerns and most tools had only been evaluated in a single study. However, four tools were evaluated in multiple studies and showed moderate diagnostic performance, with relatively little difference in performance between tools. While encouraging, further large and well-conducted studies are needed to ensure these tools are acceptable to patients and clinicians, are cost-effective and facilitate the early diagnosis of ovarian cancer.

**Abstract:** In the absence of effective ovarian cancer screening programs, most women are diagnosed following the onset of symptoms. Symptom-based tools, including symptom checklists and risk prediction models, have been developed to aid detection. The aim of this systematic review was to identify and compare the diagnostic performance of these tools. We searched MEDLINE, EMBASE and Cochrane CENTRAL, without language restriction, for relevant studies published between 1 January 2000 and 3 March 2020. We identified 1625 unique records and included 16 studies, evaluating 21 distinct tools in a range of settings. Fourteen tools included only symptoms; seven also included risk factors or blood tests. Four tools were externally validated—the Goff Symptom Index (sensitivity: 56.9–83.3%; specificity: 48.3–98.9%), a modified Goff Symptom Index (sensitivity: 71.6%; specificity: 88.5%), the Society of Gynaecologic Oncologists consensus criteria (sensitivity: 65.3–71.5%; specificity: 82.9–93.9%) and the QCancer Ovarian model (10% risk threshold—sensitivity: 64.1%; specificity: 90.1%). Study heterogeneity precluded meta-analysis. Given the moderate accuracy of several tools on external validation, they could be of use in helping to select women for ovarian cancer investigations. However, further research is needed to assess the impact of these tools on the timely detection of ovarian cancer and on patient survival.

**Keywords:** ovarian cancer; symptoms; early detection; risk assessment; diagnostic prediction model; triage tool; ovarian cancer symptoms

#### **1. Introduction**

Ovarian cancer is the eighth most common cancer to affect women worldwide, accounting for over 384,000 deaths in 2018 [1]. Outcomes are strongly linked to stage at diagnosis, with five-year survivals of 90% and 4% for UK women diagnosed at stages I and IV, respectively [2]. Given this, large ovarian cancer screening trials have been conducted, but these have so far failed to demonstrate a significant reduction in long-term mortality [3,4]. In the absence of effective screening programs, the majority of ovarian cancers are diagnosed following symptomatic presentation [5,6], and a focus has been placed on the early detection of symptomatic disease [7].

While once regarded as a 'silent killer', many studies have demonstrated that a range of symptoms are more common in women with ovarian cancer than in control subjects and that symptoms occur at all stages of the disease [8]. Clinical guidelines in countries around the world recommend that patients presenting with symptoms of possible ovarian cancer undergo investigation, although debate remains over which symptoms are indicative of disease and should be included in guidelines [7]. To facilitate the early detection of symptomatic cancer, researchers have developed a number of symptom-based checklists for use either when patients first present in the clinical setting or in 'symptom-triggered screening' programs, in which symptoms are proactively solicited [9–11]. More sophisticated tools, which can take the form of diagnostic prediction models [12], have also been developed to incorporate test results and ovarian cancer risk factors alongside symptoms, in a bid to improve tool performance. Several of these tools have been incorporated into clinical computer systems, which, then, automatically alert the clinician to consider ovarian cancer investigations when relevant symptoms are present or when the risk of undiagnosed cancer reaches a certain level. However, the relative limitations and merits of the various available tools remain unclear. In this systematic review, we aimed to identify and compare the diagnostic performances of symptom-predicated tools for the detection of ovarian cancer.

#### **2. Methods**

#### *2.1. Eligibility Criteria and Searches*

This review was conducted and is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines (Table S1); a study protocol was registered with PROSPERO [CRD42020149879]. We searched MEDLINE, EMBASE and Cochrane CENTRAL for keywords relating to ovarian cancer, symptoms and prediction/diagnostic tools to identify papers published between 1 January 2000 and 3 March 2020 (Text S1). The start date was chosen to predate the publication of key ovarian cancer symptom papers [13,14]. No language restrictions or restrictions on methodological design were applied. No restrictions were placed on study setting, so studies conducted in the general population or in primary, secondary, or tertiary care were all eligible for inclusion. Reference lists of included papers were screened to identify any additional relevant papers.

Studies were included if they (a) described the development and or evaluation of a multivariable tool designed to identify patients with undiagnosed ovarian cancer and (b) provided the sensitivity and specificity of the tool or gave sufficient data to allow these metrics to be calculated. For the purposes of this review, we defined a multivariable tool as a combination of three or more variables used to detect or predict the risk of undiagnosed ovarian cancer. This broad definition encompasses traditional multivariable diagnostic prediction models and clinical prediction rules [12,15]. We considered variable 'checklists', in which any one variable in the list needed to be present for a positive result, to be a form of multivariable tool. As the focus of this review was on symptom-based tools, the tool under investigation had to include at least one symptom for a study to be eligible. No other restrictions were placed on the type of variable that could be included in a tool. Studies on tools intended to estimate future risk of developing ovarian cancer rather than the current risk of having an undiagnosed ovarian cancer were excluded, as were studies on tools that solely provide an indication of the risk of relapse or recurrence. We excluded studies in which all participants had a pelvic mass—as this represents a highly selected high-risk population—and studies undertaken solely in paediatric (<18 years) populations. Non-primary research studies were also excluded.

#### *2.2. Study Selection*

The online Rayyan software was used to facilitate abstract screening and study selection [16]. Following removal of duplicates, two reviewers (G.F. and V.H.) independently screened titles and abstracts against eligibility criteria. Potentially eligible papers identified at the screening stage were obtained and the full texts were independently examined against eligibility criteria by two reviewers (G.F. and V.H.). Any disagreements were resolved by consensus.

#### *2.3. Data Extraction and Synthesis*

Data extraction was performed by one reviewer (G.F.) and checked against full-text papers by a second reviewer (V.H.) to ensure accuracy. Using a predeveloped template, information was extracted on study characteristics (year of publication and location), study design (methodology, population, data source and outcome definition), tools (variables and tool development methods), and tool performance metrics (sensitivity, specificity and other diagnostic metrics). Where a study evaluated multiple tools, data relating to each tool were extracted separately.

Sensitivity and specificity were used to compare tool accuracy. For diagnostic prediction models, area under the receiver operator characteristic curve (AUC) was used to compare discrimination (the ability of a tool to identify those with a condition from those without a condition) and calibration (agreement between estimated and observed outcomes). Due to the marked heterogeneity of included studies in terms of the study designs, populations, variable definitions, outcome definitions and use of different tool thresholds, and the failure of multiple studies to report numbers of patients with true positive/true negative/false positive/false negative results, we were unable to perform any meta-analyses. Instead, performance characteristics were summarised in tabular form and using a narrative synthesis approach. When synthesising data, we paid particular attention to several study and tool characteristics. First, the source of participant recruitment. For example, whether controls were recruited from the general population or after entry into healthcare, as symptoms may be more common in clinical controls than population controls, which could influence measures of tool sensitivity and specificity [17]. Second, whether the measures of tool accuracy were obtained directly from the patient sample in which the tool was developed (apparent performance), by applying internal validation methods, such as splitting the sample into development and validation sets or using cross-validation techniques (internal validation), or from a separate analysis in a distinct population (external validation) [12]. Tools usually exhibit poorer diagnostic performance in external validation studies than when evaluated in the original development sample, and external validation of tools is recommended before they are used in clinical practice [12]. Third, we considered whether tools consisted solely of symptoms or symptoms in addition to other variables, as this is likely to impact the clinical utility of the tool.

#### *2.4. Risk of Bias Assessment*

The Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool was used to assess the risk of bias and applicability of the included studies [18]. QUADAS-2 includes signalling questions (intended to identify areas of potential bias or concern over study applicability) covering four domains: (1) patient selection, (2) index test(s), (3) reference standard and (4) flow and timing. Each domain was rated as having "high", "low" or "unclear" (where insufficient information is provided) risk of bias. Domains 1–3 were also rated for applicability as "high", "low" or "unclear" concern. Two reviewers (G.F. and V.H.) independently assessed each study using QUADAS-2. Ratings were compared and disagreements were resolved by consensus.

#### **3. Results**

#### *3.1. Study Selection*

In total, 2331 records were identified from database searches, of which 708 were duplicates. Two additional records were identified from examination of reference lists. A total of 1625 titles and abstracts were screened, and 35 full-text papers were examined. Sixteen studies met the eligibility criteria and were included (Figure 1).

**Figure 1.** Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) flow diagram illustrating the study selection process.

#### *3.2. Study Characteristics*

The characteristics of the included studies are summarised in Table 1 and additional exclusion criteria are detailed in Supplementary Material Table S2. Three studies were population-based [19–21], five studies were based in a primary care setting [14,22–25], four studies were entirely hospital-based [26–29] and four studies were hospital-based but also recruited controls from screening studies [30–33]. All populationand hospital-based studies were of case-control design. Two of the studies that recruited from the hospital setting included a proportion of controls with benign ovarian pathology [26,28]. Three of the five primary care studies were of cohort design [22–24], and the remaining two were of case-control design [14,25]. The studies used a variety of data sources for variables, including pre-existing routinely collected primary care data (*n* = 6), information from surveys or patient interviews (*n* = 11) and blood samples (*n* = 4). Study sizes varied markedly, with 75–1,908,467 participants and 24–1885 women with ovarian cancer per study. While all studies used ovarian cancer as an outcome, how this was defined differed, with some only including invasive epithelial cancer or specifically stating that they excluded borderline tumours [19–21,26–29], and others apparently including both invasive and borderline epithelial tumours or all ovarian cancers [14,22–25,30–33]. One study included ovarian cancer alongside other common cancers in a composite outcome, but tool performance characteristics for each cancer were given separately [23]. Seven studies developed entirely new tools [14,19,22,23,25,30,33], six modified existing tools [26–29,31,32] and eight externally validated existing tools [20,21,24,26–29,33].


**Table 1.** Study characteristics.


**Table 1.** *Cont.*



their families, participating in the Ovarian Cancer Early Detection Study (OCEDS) [37]. c Numbers varied by study component: questionnaire (191 cases, 268 controls), telephoneinterview (111 cases, 125 controls) and GP notes (171 cases, 227 controls). d Controls with benign gynaecological disease were also included in study but are excluded from the review, asperformance was examined separately to healthy controls and no overall specificity measure was given. Study design and Objectives denoted by "•". Abbreviations: OC = Ovarian cancer; NOS =Not otherwise specified; GP=General practice; US =Ultrasound.

#### *3.3. Risk of Bias*

The main potential sources of bias were identified in the "patient selection" and the "index test" domains (Figure 2). As the case-control design can lead to overestimation of test performance [18], 13 studies were flagged as being at high risk of bias for patient selection. Key potential sources of bias identified for studies in the "index test" domain included failing to pre-define the tool threshold and retrospectively administering the tool after the outcome had been determined, e.g., questioning participants after the ovarian cancer diagnosis had been made. The risk of bias was generally judged as low for the "reference standard" and "flow and timing" domains. However, all primary care studies were flagged as being at high risk of bias in the "reference standard" domain as they relied on general practitioner (GP) records to identify ovarian cancer diagnoses, supplemented in two studies by death registration data [22,23] rather than hospital or cancer registry histological diagnoses. Concern over the applicability of studies was judged as low, save for the "reference standard" domain of one study which used a composite cancer outcome [23].


**Figure 2.** QUADAS-2 Risk of Bias Assessment. Green = "low", orange = "high", blue = "unclear" risk of bias.

#### **4. Tool Variables**

The studies evaluated a total of 21 distinct tools, of which five were diagnostic prediction models developed using appropriate statistical methods from which variable weights were derived [12]. We grouped variables included in the tools into four categories: (1) patient demographics, (2) personal and family medical history, (3) symptoms and (4) test results (Table 2). By definition, all tools included symptoms, with 14 including only symptoms. Four tools incorporated demographics, two incorporated personal and family medical history and six incorporated test results. Five symptoms (abdominal pain, pelvic pain, distension, bloating and appetite loss) were included in more than half (≥11) of the tools and a further six symptoms (feeling full quickly, difficulty eating, postmenopausal bleeding, urinary frequency, palpable abdominal mass/lump and rectal bleeding) were included in at least a quarter (≥6) of the tools. Six tools were based on an existing tool—the Goff Symptom Index (SI)—which was modified to include additional symptom or test result variables. Specifications of each tool, including how variables were defined, are included in the Supplementary Material Table S3.



**Test Results** **CA125 HE4**


**Table 2.** *Cont.* of a variableand durationand frequency criteria, are included in Supplementary Material Table S3. Abbreviations: PMH = past medical history; FH = family history; Abdo. = abdominal; Distens. = distension;Bloat. = bloating; Postmen. = postmenopausal; bleed. = bleeding; Freq. = frequency; Hb = haemoglobin; CA125 = cancer antigen 125; HE4 = human epididymis protein 4; SI = symptom index; OC = ovarian cancer; BMI = body mass index; endomet. = endometrial; T2DM = type 2 diabetes mellites; COPD = chronic obstructive pulmonary disease; GI = gastrointestinal;CIBH = change in bowel habit.

#### *4.1. Evaluation of Tool Performance*

The diagnostic performance of the included tools is summarised in Table 3. Measures of diagnostic performance for the majority of the tools were obtained directly from the patient sample with which the tool was developed (apparent performance) or by applying internal validation methods, such as splitting the sample into development and validation sets (internal validation), with only four tools—the Society of Gynaecologic Oncology (SGO) consensus criteria, Goff SI, QCancer Ovarian, Modified Goff SI 1—undergoing independent validation with an external dataset. Although the Goff SI in combination with CA125 was evaluated in several studies, the CA125 thresholds used varied markedly, so no studies were considered to have externally validated the same combination. There was overlap in evaluation of tools between healthcare settings, but no tool evaluated in primary care was evaluated in another setting or vice versa.

The most widely studied tool was the Goff SI, which was evaluated in nine studies [20,21,26,27,29–33], but two of these used data from subsets of women in the original tool development study [31,32]. Apparent deviations from the original Goff SI in how variables were defined were noted in several studies (Table S4). The Goff SI was the only tool to be externally validated in groups of women recruited from more than one setting.

#### *4.2. Tool Diagnostic Accuracy*

#### 4.2.1. Hospital Setting

All but two tools evaluated in hospital populations incorporated the Goff SI. Two of these underwent external evaluation—the original Goff SI and a modified version incorporating additional symptoms (Modified Goff SI 1). The Goff SI, which was externally validated in six studies, demonstrated sensitivities which ranged from 56.9% to 83.3% (an outlier result) and specificities from 48.3% (an outlier result) to 98.9%. A modified version of the Goff SI (Modified Goff SI 1) demonstrated a sensitivity of 71.6% and a specificity of 88.5% in a single external validation study.

Augmenting symptom checklists with baseline risk factors and test results generally led to a reduction in sensitivity and an increase in specificity, or vice versa, depending on the threshold used. For example, the addition of the serum ovarian cancer biomarker CA125 to the Goff SI by Anderson et al. (2008) led to a reduction in tool sensitivity—if both variables were required to be abnormal for a positive tool result—or in tool specificity—if only one was required to be abnormal for a positive tool result [31].

#### 4.2.2. Population Setting

In women recruited from the population setting, two symptom checklists were externally validated side by side—the Goff SI and the SGO consensus criteria. While the sensitivities and specificities of the tools differed between the studies, within each study, they were similar, with an in-study maximum difference in sensitivity of 3.4% and specificity of 2.4% between the tools.

#### 4.2.3. Primary Care

A single tool (QCancer Ovarian), which took the form of a prediction model and combined symptom variables with demographics, family history and routine blood test results, underwent external validation in a primary care setting. When the threshold for abnormality was set to include the 5% of women at the highest predicted risk, QCancer Ovarian had a sensitivity of 43.8% and a specificity of 95%, while when the threshold was set to include women at the 10% highest risk, the sensitivity increased to 64.1% but the specificity fell to 90.1%. Several scores, developed by Grewal et al., demonstrated higher sensitivities and specificities than QCancer Ovarian at the 5% risk threshold (OC Score B ≥ 4) and 10% risk threshold (OC Score C ≥ 4), but diagnostic accuracy measures were derived from the same dataset used in score development.


**Table** 



**Table 3.** *Cont.*



(questionnaire, telephone interview, GP notes). f Biomarker level (CA125, HE4) dichotomised at 95th percentile in control group—levels above that deemed abnormal. The Recruitmentsetting and the source of accuracy estimate are denoted by "•". Abbreviations: OC = ovarian cancer; CI = confidence interval; AUC = area under the receiver operator characteristiccurve; PPV=positive predictive values; yrs =years.

Discrimination was reported for five tools (Table 3), all of which had similar AUCs within the 'good' range (0.84–0.89), with QCancer Ovarian exhibiting an AUC of 0.86 on external validation. Tool calibration was assessed for QCancer tools by graphically comparing the predicted cancer risk at two years with the observed risk by predicted risk deciles [22–24]. Authors reported good calibration on internal validation. On external validation, QCancer Ovarian had reasonable calibration but overpredicted risk, particularly in older women [24].

#### 4.2.4. Positive Predictive Values

The three cohort studies conducted in primary care reported positive predictive values (PPV) for QCancer tools at a range of thresholds (Table 3). The PPVs at any given risk threshold were similar—for example, values ranged from 0.5 to 0.8% when the threshold was set to identify the 10% of women at highest risk. Two case control studies (Rossing et al. and Jordan et al.) used external disease prevalence figures from screening studies and available population-level statistics to estimate the PPVs of the Goff SI and SGO consensus criteria—if they were to be used in general populations. The tools had similar estimated PPVs within each study, but PPVs were higher in Rossing et al. (0.63–1.12%) than in Jordan et al. (<55 years: 0.04–0.05%, ≥55 years: 0.18–0.31%).

#### **5. Discussion**

To our knowledge, this is the first systematic review to compare the diagnostic performance of existing symptom-based tools for ovarian cancer detection. We identified 21 symptom-based tools designed to help identify women with undiagnosed ovarian cancer. These tools comprised simple symptom checklists, checklists which included both symptoms and tests and more complex diagnostic prediction models which incorporated symptoms, test results and baseline risk factors. While the diagnostic performances of most tools were evaluated solely within the study development datasets, four tools were independently externally validated, with one being validated in multiple population settings. Externally validated tools demonstrated similar moderate diagnostic performances. Our findings should inform future studies evaluating the clinical impact of validated symptom-based tools when implemented in clinical practice.

#### *5.1. Study Strengths and Limitations*

The main strengths of this study were its systematic approach, broad search strategy and liberal eligibility criteria, which enabled us to identify and compare the performances of a wide variety of tools. However, the identified studies were extremely heterogeneous in their designs, populations, variable definitions, outcome definitions and thresholds, which ultimately precluded any meaningful meta-analyses. For example, although the Goff SI was evaluated in nine studies, there was overlap between the participants in three studies, control groups ranged from apparently healthy general population participants to hospital gynaecology patients (with or without benign pathology), ovarian cancer definitions differed and deviations in the parameters of the SI itself, in terms of symptom duration and frequency criteria, were noted in several studies. While meta-analysis was not deemed appropriate, our results demonstrate how the Goff SI performs under different conditions. An additional limitation was that all included studies were at high risk of bias in at least one QUADAS-2 domain, which limits the conclusions that can be drawn.

#### *5.2. Comparison of Tools*

Although all tools were symptom-based and designed to help identify women with ovarian cancer, they varied markedly in the symptoms they included. This mirrors discrepancies in the literature and within national guidelines as to which symptoms are associated with the disease and probably reflects differences in study methodologies and study populations [7]. Despite this, the symptoms with the highest positive likelihood ratios for ovarian cancer in a recent systematic review (distension, bloating, abdominal or pelvic pain) were incorporated into the majority of tools [8]. The more cancer-associated symptoms that are included in a checklist, the higher the sensitivity of the tool is likely to be, but at the

cost of reducing specificity, as demonstrated by several of the included studies [19,26,33]. This was cited by Goff et al. as a rationale for not including urinary symptoms in the Goff SI [30]. Ultimately, variation in which additional symptoms a tool includes may have limited impact on tool performance; on external validation, two studies reported similar diagnostic accuracy metrics for the Goff SI and the SGO criteria (which differed on several symptoms), and on internal validation, Lim et al. concluded that changing several of the symptoms made relatively little difference to tool diagnostic accuracy [33].

In multiple studies, symptom checklists were augmented by ovarian cancer biomarkers with the aim of improving tool diagnostic accuracy. This approach naturally led to a reduction in tool specificity (where either symptoms or an abnormal test resulted in a positive tool) or sensitivity (where symptoms and an abnormal test were needed for a positive tool). If ovarian cancer biomarkers are to be included alongside symptoms within tools, this loss of performance could be avoided by incorporating them within prediction models, as per the inclusion of anaemia in QCancer Ovarian. As the prediction model threshold can be set at a desired risk level, biomarkers, such as CA125 and HE4, could be incorporated without harming tool performance. However, this would require women to have specialist ovarian cancer markers performed in order for the tool to be used, which significantly limits clinical utility. A more practical approach would be to incorporate tools within a two-step pathway in which symptom-based tools (which do not include specialist test variables) are used to help select higher-risk women for specialist ovarian cancer tests.

Variation in the reported sensitivity and specificity of the most widely evaluated tool, the Goff SI, was noted between studies. This variation is likely to be due, in part, to the marked differences in study design, populations and outcome definitions which precluded meta-analysis across these studies. Despite these differences, in 5 of the 6 external validation studies (including two large population-based studies), the Goff SI had a sensitivity in excess of 60%, and in all but the smallest study, which included only 24 ovarian cancers and 31 controls, its specificity exceeded 85%. The sensitivities and specificities of the two other externally validated symptom checklists—the SGO consensus criteria and the modified Goff SI 1—were similar, as were those of the only externally validated diagnostic prediction model—QCancer Ovarian (applying a 10% risk threshold). Given the similarity in performance of the various existing validated tools, future research efforts may be better directed at evaluating the impact of using available tools in practice rather than developing further tools consisting of different symptom combinations.

#### *5.3. Clinical Relevance*

Two distinct uses for tools were identified by the authors of the included studies: (1) assessment of women presenting symptomatically in the standard clinical setting to identify those at higher risk of undiagnosed cancer and to inform decision making and further investigation, and (2) proactive 'symptom-triggered screening' programs in which women are actively screened using the tool, with further testing for ovarian cancer occurring if the tool is positive. Several of the tools identified in this review are already available for use within the standard clinical setting in the form of electronic clinical decision support tools (eCDSTs). QCancer tools are integrated within some UK general practice IT systems and alert the clinician if the risk of ovarian cancer in an individual reaches a certain level, prompting them to consider ovarian cancer as a possible diagnosis. eCDSTs have been shown to improve practitioner performance and patient care, but there are multiple barriers to their implementation and they do not always lead to improved outcomes [43,44]. Therefore, even if eCDSTs are deemed to have acceptable diagnostic accuracy, their cost-effectiveness, acceptability to patients and clinicians and their impact on timely ovarian cancer detection and survival need to be evaluated. Currently, a large, clustered, randomised control trial is seeking to help to address this by investigating the clinical impact of implementing a suite of electronic cancer risk assessment tools (including an electronic version of the Hamilton ovarian SI) in UK general practice [45]. Studies have also sought to evaluate the impact of using tools as part of 'symptom-triggered screening' programs, but none have taken the form of randomised control trials—the gold standard approach—and so findings should

be interpreted with caution. In one study, 5000 women were approached in primary care clinics and screened for symptoms using the Goff SI, with further investigations performed if the Goff SI was positive [11]. However, conclusions were limited as only two ovarian cancers were identified in the study window. The Diagnosing Ovarian and Endometrial Cancer Early (DOvEE) trial also employs a proactive symptom-triggered testing approach, supported by media campaigns, in which women can self-refer and are screened for range of symptoms prior to study inclusion. Although the final DOvEE results are yet to be published, a pilot study reported that participants had lower tumour burden and more resectable disease than women diagnosed via the standard clinical pathway [9].

When considering the clinical utility of a tool, it is important to assess the proportion of women who are 'tool-positive' who ultimately have ovarian cancer, i.e., the PPV. Primary care cohort studies indicated that between 1 in 200 and 1 in 100 women who were QCancer tool-positive (5% or 10% risk) had the disease. Although these figures may appear low, evidence indicates that patients would opt for cancer testing at PPVs of 1% [46]. Further, having a positive tool result in the clinical setting does not necessarily mean that further investigation will automatically occur, as there may be a clear alternative cause for the symptoms—the tool is simply intended as a diagnostic aid to highlight the risk of ovarian cancer to the clinician. In addition, the most common follow-up tests—CA125 and transvaginal ultrasound—are relatively non-invasive, and CA125 is known to perform well when used in a symptomatic primary care population [47], although invasive investigations/surgery may ultimately be needed to determine whether ovarian cancer is present. In proactive symptom-triggered screening programs, the tool is more than just a diagnostic aid—it is the initial screening step which will dictate whether further ovarian cancer tests take place. The two population studies reporting PPVs relied on external ovarian cancer prevalence figures, but their PPV estimates were similar to that reported in the pilot DOvEE study (0.76% in women ≥ 50 years) [9]. Further research is needed to help determine whether, given this PPV, follow-up testing in proactive symptom-triggered testing programs is acceptable to women and improves outcomes. The definitive diagnosis of ovarian cancer often involves invasive procedures/surgery, which has contributed to patient morbidity in key ovarian cancer screening trials [3,39]. Although initial findings indicate that proactive symptom triggered testing approaches lead to minimal unnecessary surgery [9,11], large trials are needed to confirm that the implementation of symptom-based tools in clinical practice does not lead to significant excess morbidity.

#### **6. Conclusions**

Over 20 symptom-based tools have been developed in different populations to help assess women for ovarian cancer, but the majority have not been validated. Four symptom-based tools—the Goff SI, a modified version of the Goff symptom Index, SGO consensus criteria and QCancer Ovarian—have undergone independent external validation and exhibit similar sensitivities and specificities. These tools could have an important role to play in the detection of ovarian cancer, but further large well-conducted studies are needed to assess their cost-effectiveness, their acceptability, their effect on the timeliness of ovarian cancer diagnosis and their impact on clinical outcomes, including patient survival.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6694/12/12/3686/s1, Table S1: PRISMA Checklist. Text S1: MEDLINE search strategy. Table S2: Specific study exclusions. Table S3: Tool specifications. Table S4: Deviations from the original Goff SI in validation studies.

**Author Contributions:** Conceptualization, G.F.; methodology, G.F., F.M.W. and W.H.; formal analysis, G.F. and V.H.; data curation, G.F. and V.H.; writing—original draft preparation, G.F.; writing—review and editing, G.F., V.H., G.A., E.J.C., W.H., F.M.W. and J.E.; supervision, F.M.W., G.A., E.J.C. and W.H.; funding acquisition, F.M.W. and W.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research arises from the CanTest Collaborative, which is funded by Cancer Research UK [C8640/A23385], of which G.F. is Clinical Research Fellow, V.H. is a PhD student, G.A. is the Senior Statistician, J.E. is Associate Director, and W.H. and F.M.W. are Directors. E.J.C. is supported through the NIHR Manchester Biomedical Research Centre (IS-BRC-1215-20007). The funders of this study had no role in the study design, data collection, data analysis, data interpretation, or writing of the report.

**Conflicts of Interest:** Two studies included in this review were conducted by W.H. W.H. played no role in study selection or quality assessment. All other authors declare no conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Long-Term Evaluation of Women Referred to a Breast Cancer Family History Clinic (Manchester UK 1987–2020)**

**Anthony Howell 1,2,3,\*, Ashu Gandhi 1,2,3, Sacha Howell 1,2,3, Mary Wilson 1, Anthony Maxwell 1,4,\*, Susan Astley 1,2,4, Michelle Harvie 1, Mary Pegington 1,3, Lester Barr 1, Andrew Baildam 1,5, Elaine Harkness 1,4, Penelope Hopwood 1,6, Julie Wisely 1,7, Andrea Wilding 1, Rosemary Greenhalgh 1, Jenny A**ff**en 1, Andrew Maurice 1, Sally Cole 1, Julia Wiseman 1, Fiona Lalloo 1,8, David P. French <sup>9</sup> and D. Gareth Evans 1,2,4,8,10,11**


Received: 22 November 2020; Accepted: 5 December 2020; Published: 9 December 2020

**Simple Summary:** This study reports the management of women at high risk for breast cancer over a 33 years period. The aim was to summarize the numbers seen and to report the results of our studies on gene testing, the outcomes of screening and the success of preventive methods including lifestyle change, chemoprevention and risk-reducing mastectomy. We also discuss how the clinical Family History Service may be improved in the future.

**Abstract:** Clinics for women concerned about their family history of breast cancer are widely established. A Family History Clinic was set-up in Manchester, UK, in 1987 in a Breast Unit serving a population of 1.8 million. In this review, we report the outcome of risk assessment, screening and prevention strategies in the clinic and propose future approaches. Between 1987–2020, 14,311 women were referred, of whom 6.4% were from known gene families, 38.2% were at high risk (≥30% lifetime risk), 37.7% at moderate risk (17–29%), and 17.7% at an average/population risk who were discharged. A total of 4168 (29.1%) women were eligible for genetic testing and 736 carried pathogenic variants, predominantly in *BRCA1* and *BRCA2* but also other genes (5.1% of direct referrals). All women at high or moderate risk were offered annual mammographic screening between ages 30 and 40 years old: 646 cancers were detected in women at high and moderate risk (5.5%) with a detection rate of 5 per 1000 screens. Incident breast cancers were largely of good prognosis and resulted in a predicted survival advantage. All high/moderate-risk women were offered lifestyle prevention advice and 14–27% entered various lifestyle studies. From 1992–2003, women were offered entry into IBIS-I (tamoxifen) and IBIS-II (anastrozole) trials (12.5% of invitees joined). The NICE guidelines ratified the use of tamoxifen and raloxifene (2013) and subsequently anastrozole (2017) for prevention; 10.8% women took up the offer of such treatment between 2013–2020. Since 1994, 7164 eligible women at ≥25% lifetime risk of breast cancer were offered a discussion of risk-reducing breast surgery and 451 (6.2%) had surgery. New approaches in all aspects of the service are needed to build on these results.

**Keywords:** family history; breast cancer; risk; genes; screening; prevention

#### **1. Introduction**

In the 1980s, the rising incidence of breast cancer (BC) and the introduction in the UK of the NHS National Health Service Breast Screening Programme (NHSBSP) led women with a family history of the disease to seek advice concerning management of their personal risk. In response to concerns expressed by primary care physicians and colleagues within our breast oncology service, we established a referral Family History Clinic (FHC) in Manchester, UK, in 1987 with a cancer genetics service (CGS) initiated in 1990 (DGRE, FL). The clinic serves a catchment population of 1.8 million, (just over half the population of Greater Manchester), although women at high-risk may be specifically referred to the centre from a population of approximately 5 million in North West England.

The aims of the FHC were to introduce a service for the estimation and management of BC risk for women with familial risk and to evaluate the short- and long-term effectiveness of the clinic. At presentation, an individual's risk was explained, annual breast screening initiated and advice given concerning diet and lifestyle factors which might affect risk. Later, chemoprevention (1992 as part of the IBIS I clinical trial), genetic testing (1994) and risk-reducing surgery (1994) were introduced. In 1994, we published local guidelines for the management of women with a family history of BC [1]. These were followed by national guidelines for the UK [2,3] and the USA [4,5].

Management was undertaken by a multi-disciplinary team. Following referral, each woman was sent a questionnaire to assess family history and breast factors and, if eligible, offered a clinic appointment. Women were initially seen by geneticists (DGRE, FL) or medical oncologists (AH, SJH). Breast examination was undertaken and advice given by specially trained nurses (RG, JW, AW) and annual mammography and MRIs performed by radiologists within the Breast Unit (represented by MW and AM). Risk-reducing surgery was performed by a team of surgeons (represented by AB, AG and LB). Quality of life aspects of risk communication, mutation testing and risk-reducing mastectomy (RRM) were an integrated part of the FHC clinical and research agenda provided by psycho-oncologists and a health psychologist (PH, JW, DF).

The aims of this paper were to present the results of each aspect of the service and to suggest potential future improvements. The results include the numbers of referrals, estimation of their BC risks and results of genetic testing, screening and uptake to lifestyle prevention, chemoprevention and risk-reducing surgery interventions. The main quality of life outcomes are also reported. The second half of this paper then suggests improvements to the service based on our own studies and those of others.

#### **2. Results**

#### *2.1. Referrals to the Clinic*

Over the period from September 1987 to September 2020, 14,311 women were referred to the clinic. Referrals were from primary care (GPs, 55.9%), from secondary care (mainly breast surgeons, 15.1%) and from the local Clinical Genetics Service for further follow up after risk assessment (20.9%). Women were also referred from a research study (notably, 3.1% from PROCAS—Prediction Risk Of Cancer At Screening, [6]) or from "other" sources such as relatives at risk attending with a proband (5.0%). The age at entry ranged from 16–81 years (median 39.9; interquartile range (IQR) 33.9–46.9, with 83% of women (11,878/14,311) below age 50 (Table 1). The number of referrals by year is shown in Figure 1 and ranged between approximately 300–700 per year following the initial 3 years lead in the period after the clinic was established.

**Figure 1.** Annual referrals to the Manchester FHC between 1987 and 2020. Increases in referral were seen during the period when the first breast cancer genes, *TP53* and *BRCA1*/*2*, were identified and also related to the publicity surrounding Angelina Jolie when she indicated that she was a *BRCA1* PV carrier. The increase in median body mass index (BMI) and median age at first full-term pregnancy (FFTP) over this period are shown. Over the period of 33 years, lifetime risk of BC in the population increased from 1 in 12 to 1 in 8, an increase presumed to be associated with change in modifiable risk factors. These trends were apparent in the FHC. For example, the median age of first full-term pregnancy increased from 24 years to 27 years (*p* < 0.001), and median BMI at clinic entry increased from 23.7 to 24.8 kg/m2 (*p* < 0.001).


generalpractitioner;PROCAS, PredictionRiskOfCancerAtScreening.

#### *2.2. Estimated Lifetime BC Risk*

Risk was estimated initially by a modification of the Claus method with the addition of hormonal and lifestyle factors, such as age of first pregnancy and BMI, and by the Tyrer–Cuzick and BOADICEA models from 2004 [7–9]. We demonstrated that the modified Claus and the Tyrer–Cuzick models gave similar distributions of risk accurately, but the Gail model underpredicted risk [10,11].

Risk was reported as moderate (17–29% lifetime risk) and high (30%+) according to our original clinic guidelines (Evans 1994) and thereafter using the NICE Guidelines risk categories (2004) [1,3]. Overall, 44.6% of women were at high-risk (including 6.4% either with a pathogenic variant (PV) or a known PV in a close family member), 37.7% at moderate risk and 17.7% at average risk (Table 1). Following assessment of the referral questionnaire, women at average risk were referred back to primary care. It was clear that more higher risk women were referred to the Clinical Genetics Service reflecting a referral pathway from primary care to the Clinical Genetics Service for the highest risk women. (Table 1). After excluding referrals at average risk, 43.7% (*n* = 3397) of women directly referred to the clinic from primary and secondary care were estimated to be at high-risk and 56.7% (*n* = 4373) were at moderate risk.

#### *2.3. Risk Perception and Cancer Worry*

Risk was uniformly given as a proportion (e.g., 1 in 4 or 1 in 5). Our early studies reported that women had a relatively inaccurate perception of personal and population risk at presentation which improved when reassessed after risk counselling [12,13]. In another study, we demonstrated that the proportion of women with accurate personal risk perceptions significantly improved after risk counselling from 12% pre-counselling to 67% 3 months post-counselling (*p* < 0.001), which was maintained for 1 year [14]. Reassuringly, this improvement in accuracy of women's risk appraisal was not associated with increased anxiety. A subsequent analysis of questionnaire data from 500 FHC attendees over time indicated that BC risk counselling reduced self-reported cancer worry in women who initially overestimated their risk, with no significant change in levels for other risk perception groups, even if the risk was greater than they had estimated pre-counselling [15,16].

#### *2.4. Genetic Testing*

Genetic testing in the clinic began in 1991 just after the discovery of *TP53* and then the *BRCA1* (1994) and *BRCA2* (1995) genes [17–19]. Initially, testing was by single-strand confirmation polymorphism (SSCP) and protein truncation testing (PTT) [20], then, from 2001, Sanger sequencing all coding exons [20], and from 2013, next generation sequencing [21]. All samples, including retrospectively, were tested for large deletions and duplications by multiple ligation-dependent probe amplification (MLPA). All mutations detected by PTT or SSCP were confirmed by sequencing. From 2004, the probability of a *BRCA1* or *BRCA2* PV in the family was estimated using the Tyrer–Cuzick model [8] the BOADICEA model [9] or the Manchester Score [20,21]. The NICE Guidelines (2004) [3] initially indicated that the proband required the probability of a PV in BRCA1/2 of ≥20% for genetic testing in England and Wales [3]. In 2013, this was reduced to a ≥10% likelihood [22]. At present, gene testing in the UK is restricted to the *BRCA1, BRCA2,* and *PALB2* genes (National Genomic Test Directory (2020)) [23].

Of the 14,311 referrals to the FHC, *BRCA1* and *BRCA2* testing was completed for 4168 individuals (29.1%) or their affected family member. A total of 736 women (5.1% of the whole FHC cohort and 17.6% of those tested) were identified as *BRCA* PV carriers (*BRCA1* = 364, *BRCA2* = 372). However, only 2.5% of unaffected direct referrals to the FHC without a known gene in the family subsequently tested positive. No systematic approach to testing for PVs in other genes was undertaken. However, 35 potential BC and other risk genes were tested on a research basis in a subpopulation of 808 women unaffected by breast cancer representative of the risk distribution of direct referrals to the FHC [24]. Of the 808 tests there were 29 (3.6%) with PVs in other potentially actionable genes (*ATM* = 11, *CHEK2* = 11,

*PALB2* = 7). These data indicate that approximately 6.1% of women directly referred to the FHC carry a PV in one of the actionable BC-associated genes.

#### *2.5. Mammographic Screening*

Annual mammography and clinical breast examination were offered from the inception of the clinic. Women were screened between the ages of 35 and 50 years (moderate risk) or 35 to 60 years (high risk), plus from 5 years younger than the earliest family diagnosis of BC. These screening periods were in accordance with an initial in-house protocol [25] and later, NICE guidelines [3,22]. Women with *BRCA1* and *BRCA2* PVs and others of equivalent high risk were screened by annual MRI between age 30 and 50 and annual mammography from age 30 to 70 according to NICE Guidelines [3,22].

Between 1987 and 2020, there were 129,119 women years of follow up; 646 BCs occurred prospectively, giving an annual incidence rate of 5.0 per 1000 which was approximately 1.7 times higher than the general population's annual rate of 3 per 1000 women aged 50–75 in the NHSBSP. Three hundred and ninety-four breast cancers occurred whilst on the screening programme or within 18 months of a screen. The majority of invasive cancers were lymph node negative (72.9%), small (≤20 mm, 73.2%) and stage 1 (61.4%). Cancers in women with *BRCA1* and *BRCA2* PVs were smaller overall with 75.0% and 85.4% being ≤20 mm, respectively, reflecting the benefits of MRI screening [26]. Breast cancer-specific survival at 10 years was 91.3% (87.4–94.0), compared with the current 10 year survival from BC in England, from 2013–2017, of 75.9% (74.9–77.0). Overall, 30.5% (92/322) of invasive cancers were oestrogen receptor negative (ER–) and 11/51 (21.5%) of carcinoma in situ (not assessed 21) were ER–.

#### *2.6. Lifestyle Prevention*

Excess weight, weight gain, sedentary lifestyle and high alcohol intake were established risk factors for BC before initiation to the FHC, and so verbal and standard written lifestyle advice was given to all women referred to the clinic. Lifestyle risk factors were common amongst high-risk women in the FHC with a similar prevalence of unhealthy lifestyles to women in the general population (57% are overweight/obese, 30% report <150 min moderate intensity physical activity/week and 45% have alcohol intakes of >14 units per week) [27]. Adult weight gain is a well-documented risk factor for BC (6% increased risk per 5 kg gain) [28]. We and others have reported that maintained weight loss of 5% or more before or after menopause in an unselected population cohort is associated with reductions of 39% and 23%, respectively, in the risk of post-menopausal BC [29]. Thus, women were advised to avoid weight gain if at a healthy weight and reduce weight by at least 5% if overweight/obese. Current guidelines include advising at least 150 min of moderate exercise per week and no more than 14 units of alcohol per week.

Since 2001 we conducted a series of randomised studies to determine the optimal methods for introducing and supporting weight control and other lifestyle changes amongst women in the FHC. These indicated that intermittent energy restriction (2 days of 50–60% energy restriction and 5 days of normal healthy eating/week) was associated with greater reductions in weight and insulin resistance than continuous energy restriction [30,31]. We also demonstrated that remotely supported (i.e., web and phone) weight loss/lifestyle behaviour change programmes are feasible and particularly effective amongst higher risk women producing ≥5% weight loss in approximately 60% of women at 12 months [32]. Uptake to these weight loss trials amongst moderate and high-risk women was between 14% and 27%.

#### *2.7. Chemoprevention*

Chemoprevention was not available at the inception of the FHC. However, women were randomised to tamoxifen or placebo in the IBIS-I trial (ISRCTN91879928). Between 1992 and 2001 and to anastrozole or placebo in the IBIS-II trial (SRCTN31488319). Between 2003 and 2012 [33,34]. Of the 7865 women invited to join these trials, 1003 women (12.8%) agreed (Table 2). These and other

studies resulted in tamoxifen use being advised by NICE in the 2013 guidelines and anastrozole in the 2017 guidelines [22]. From 2013, 5121 women were invited to take either drug as part of management. To date, 282 have chosen chemoprevention and a further 284 in clinical trials so that overall 10.8% of eligible women have accepted treatment since 2013 (Table 2).

#### *2.8. Risk-Reducing Mastectomy*

Risk-reducing mastectomies were being performed, particularly in the USA, at the time of inception of the clinic. In 1994, we decided to offer this service in the FHC. Since then, 7164 women with a lifetime BC risk ≥25%, including *BRCA1*/*2* PV carriers, were offered a discussion concerning RRM according to a published protocol [35]. Of a total cohort of 7195 women at a ≥25% lifetime risk of BC, 451 (6.2%) without a current or previous breast cancer diagnosis elected to undergo RRM. Uptake of RRM was 49.3% in 479 *BRCA1*/*2* PV carriers and 5.2% in 6685 in non-carriers (9.4% for 1261 women at a ≥40% lifetime risk (non-BRCA), 4.9% in 3561 women at 30–39% risk and 3.0% in 1783 women at 25–29% lifetime risk).

In Cox regression analyses, factors which independently predicted risk-reducing mastectomy uptake included either the death of a sister with BC <50 years or mother <60 years, having children, having a breast biopsy or younger age at assessment (<30 years).

Of the 451 women who underwent RRM, four developed post-surgery BCs (all in *BRCA1*/*2* PV carriers) compared to 94 expected over a period of follow up of 7894 years, giving a risk reduction of 95.8%.

Twenty women (5.7%) had no reconstruction, whereas 352 (78%) had implant-based reconstruction (nipple sparing in 31% of these) and 63 (14%) flap-based reconstruction. The number of planned surgical procedures per patient was 2.41 ± 1.11 SD [36].

Two studies assessing psychological distress in our FHC patients undergoing risk-reducing mastectomy have been published [37,38]. Between 1995 and 1999, quality of life was assessed in 52 of 76 (79%) women undergoing surgery one-year post-operatively. At this point, 1 in 6 women had high scores for mental health problems on the General Health Questionnaire but for most, psychological distress appeared to be comparable with women at high risk who did not have surgery. Body image changes on the Body Image Scale were generally minor in degree with the most frequently reported changes reported in sexual attractiveness (55% responders), feeling less physically attractive (53%) and self-consciousness about appearance (53%). For the majority of women there was no evidence of significant mental health or body image problems in the first 3 years following RRM. Careful pre-operative preparation and long-term monitoring was advocated. In a second study, 79 women who chose to have surgery were compared with 64 women who declined surgery [38]. The main findings were that risk-reducing mastectomy reduced psychological morbidity and anxiety and did not have a significant detrimental impact on women's body image or sexual functioning.


**Table2.**Numberofwomen(A)randomisedintotheIBISIorIBISIItrial1992and2012;(B)numberofwomenprescribedchemoprevention(CP)from2013

#### **3. Discussion**

We summarised the developments that occurred since the inception of the clinic including updated models for risk and gene testing estimation (i.e., Tyrer-Cuzick v8 2017 [39], BOADICEA-V2019 [40], the Manchester Score [20,21] consistent use of mammography and MRI (NICE 2013 [22], a more po-active approach to lifestyle change [32], chemoprevention [41] and RRM [42]. We now suggest potential improvements to the service in each of the areas considered above based on our own studies and those of others.

#### *3.1. Referrals*

The numbers of referrals have remained relatively stable over the years and are still mainly instituted by women concerned about their family history and primary and secondary care clinicians who refer according the NICE familial breast FHC guidelines [3]. Our own and the reports of others indicate that approximately 10% of women in the UK have a first degree relative with BC; however, we estimate that <20% of these are referred [43]. It is interesting that referrals to FHCs and Clinical Genetics Services increased over two-fold after Angelina Jolie made her *BRCA1* PV carrier status and breast and ovarian surgery public [44,45]. This suggests that lack of awareness of the services may be an issue. Lack of uptake may also be related to a complex referral system which requires a visit to primary care for referral and the completion of extensive questionnaires. The latter is illustrated in our study of risk estimation in women undergoing breast screening. Thirty-seven percent of women invited completed a two-page questionnaire concerning their risk factors [6]. The cohort included 13% of women with a first-degree relative with breast cancer. Of 673 women found to be at high risk and invited for counselling and treatment at the FHC, 500 (74.3%) attended [46]. These data indicate a greater interest in risk management if the system is streamlined, suggesting that progress may best be made by more effort to align risk estimation with screening programmes. Referral would then be less dependent on health care professionals and would make it a more routine and streamlined service.

#### *3.2. Risk Estimation*

We demonstrated that the modified Claus model used in the clinic before 2004 gave similar results to the Tyrer–Cuzick model, suggesting consistent risk estimation for the duration of the clinic to date [10,11]. However, several studies indicate that the accuracy of risk estimation increases with the number of risk factors that are incorporated into the models used [47–50]. Recently, mammographic density (MD) and polygenic risk scores (PRS) based on single nucleotide polymorphism (SNP) results) have been added to risk models such as Tyrer–Cuzick (v8) and BOADICEA V [39,40]. In patients under follow up at the FHC, we assessed the effect on risk of incorporating the first 18 BC risk-associated SNPs discovered into the Tyrer–Cuzick model [51,52]. Adding SNP18 resulted in a change to the original given risk using Tyrer–Cuzick (version 6) in half the population of women: 25% had an increase in risk and 27% had a decrease in risk, indicating the potential importance of additional risk factors [53].

In the screening population we demonstrated that when both mammographic density and a PRS score were added to the Tyrer–Cuzick (v8), the proportion of women at elevated risk (>5% 10 years risk) increased from 12% to 18%. Ten percent of women changed from average to high risk and 4% from high to average [54,55]. These studies illustrate that using standard models may give erroneous risks and adding more risk factors may result in more appropriate management. However, more work is required to routinely apply optimal risk models in the clinic and deal with change in risk estimation over time.

#### *3.3. Genes*

Of the women directly referred to the FHC in which there was no currently known PVs, 2.5% were *BRCA1*/*2* PV carriers and 3.6% were found to have PVs in other genes after multigene panel testing [24]. This low pickup rate reflects that over 50% of referrals were at moderate risk and many of those at

higher risk had undergone testing in themselves or their family (Table 1). At the FHC, much time was used to calculate the probability that the proband or her family are likely to carry a breast cancer gene PV. At present our Clinical Genetics Service forwards primary care referrals for women unlikely to carry a PV immediately to the FHC; conversely, we refer relatives of known PV carriers to the Clinical Genetics Service. A simple method where primary and secondary care physicians could estimate PV risk and refer appropriately would be invaluable. A more widespread use of the simple Manchester Score needs to be evaluated in this regard [20,21]. Currently, the NHS guidelines to the UK genetics departments allow estimation of *BRCA1*/*2* and *PALB2* (as well as syndromic genes where indicated such as *PTEN* in Cowden disease) [23]. The recent report illustrating the nine genes (*BRCA1, BRCA2*, *PALB2, CHEK2, ATM, CDH1, STK11, PTEN, TP53)* in which PVs more than doubles the risk of BC and three genes at the two-fold threshold (*RAD51C, RAD51D, BARD1)* might allow better selection for appropriate gene testing and a reduction in the need for multigene testing which, in the UK, is in the commercial sector [24].

#### *3.4. Breast Screening*

Mammographic screening in this at-risk population detects more cancers annually (5/1000 screened) than in the national programme (3/1000) as expected. It also results in the detection of smaller, better risk cancers. Three studies in the UK, one in our clinic and two in association with other clinics in the UK indicated that screening at-risk women results in a survival advantage compared (in non-randomised trials) with age matched populations [56–58]. The latest study was designed to assess the value of screening both moderate- and high-risk women from age 35–39 and confirmed a survival advantage even in the moderate-risk group [58].

Countries where national screening programmes begin at age 40 will already be screening in the high-risk groups. A review of the value of screening from age 40 in the general population concluded it was of equivocal value [59]. In the UK, screening is every 3 years for all from the age of 50, but a recent study now suggests a survival advantage in the general population when screening begins annually at the age of 40. This may lead to a change in UK policy [60]. The results of two randomised trials of risk adapted screening will inform a potential change in policy since they both screen women from age 40 onwards (WISDOM [61] MyPebs UNICANCER 2018 [62]). Further, a programme of work carried out in Manchester is considering how best to implement risk adapted screening to optimise the ratio of benefits and harms, including how to include ethnically minority women and to minimise harms of screening for women at low risk [63].

In countries where screening begins at 50 (e.g., the UK), consideration should be given to offering all women a one-off mammogram at age 40, together with risk estimation to determine future screening frequency. Mammographic density could be assessed automatically using artificial intelligence methods [64,65] and SNPs used only to determine precise risk where needed. The trials of risk and density-adapted screening and determination of the value of supplemental imaging techniques, such as whole breast ultrasound, contrast-enhanced mammography and abbreviated MRI, would further refine the management advice offered to women with high MD. Currently, at the FHC, women are offered an MRI if they carry a PV of a high-risk gene or if they have a 10 year risk of ≥8% aged 30 or ≥12% aged 40, based on the finding in trials that MRIs detect smaller tumours and may offer a survival advantage [66–70]. However, neither MRIs nor ultrasounds are routinely available (or proven) for a large group of women at increased risk outside those in very high-risk groups.

#### *3.5. Lifestyle Advice*

Women at high risk who have a high BMI [71], low physical activity levels, high alcohol intake [72] and smoke [73] have proportionately higher BC risks than similar women at population risk [71–73]. These potentially modifiable risk factors have also been associated in women at high risk due to the fact of family history or high PRS [72,73]. Thus, there is a rationale for focussing on lifestyle change in

this higher risk group, and it is probable that the more avenues to promote change that are pursued, the greater the likelihood of success [74].

Women at high risk present a challenge for achieving lifestyle behaviour change. Firstly, some women can view their BC risk as unchangeable because of their family history [75]. This is consistent with a large body of literature that indicates education around disease risk does not alter behaviour by itself and that people require an appropriate level of support to achieve and sustain lifestyle behavioural change [76]. Lifestyle behaviour change programmes need to address the often complex psychological issues amongst women who have a high burden of cancer diagnoses and bereavements in their family. For many women, the majority of excess weight is acquired between the age of 18 and 35 years [76], indicating that lifestyle programmes should begin at younger ages. Interviews with young, high-risk women (under 35 years) in our FHC indicate that those women require a supportive weight control lifestyle programme that is remotely accessible, provides a point of contact within the high-risk service and promotes general wellbeing as well as cancer risk reduction [76]. There is a need for wider testing of low-cost programmes for lifestyle prevention which can reach and engage the maximum number of women across the network of UK FHCs.

#### *3.6. Chemoprevention*

The reduction of risk of BC by 30–50% by the use of SERMs and AIs such as tamoxifen, raloxifene and anastrozole are well known [77]. More recently, long-term follow up of the IBIS I and IBIS II trials indicate that risk reduction continues long after the usual five-year prescription period [41,78]. A recent analysis by NICE indicates that the use of anastrozole, in particular, is cost saving to the NHS in women at moderate to high risk of BC. However, whilst reduction of risk is of benefit, none of the trials to date have shown a survival benefit. This has led to the suggestion that premarin should be used for women at least 5 years post-menopausal and without a uterus, since in this group the Women's Health Initiative trial use was associated with a survival advantage [79].

Our report of recent uptake of chemoprevention being relatively low at 10% is consistent with many but not all studies [80,81]. Part of the reason for the low uptake concerns the side effects, although our own and other studies show that the observed frequency of side effects are comparable to controls [33,34,81–84]. We found four themes associated with low uptake: the perceived impact of side effects, the impact of others' experience on beliefs about tamoxifen, tamoxifen as a "cancer drug" and the daily reminder of cancer risk [80]. These reasons are understandable and consistent with other studies. Future developments require better communication of the pros and cons of therapy and alternative approaches including low dose or topical tamoxifen [85]. New agents such as antiprogestins [86], and denosumab [87] are currently being trialled in the FHC and elsewhere.

#### *3.7. Risk-Reducing Surgery*

Historically, our unit performed approximately 10–12 operations per year. With recent increases in publicity surrounding RRM [44], this has increased approximately three-fold in our own and other units [45]. The seminal paper by Hartmann [88] indicated BC a risk reduction of 92%, similar to our observed reduction of 95.8%. Over the years, our surgical approaches have evolved to reflect refinements in surgical technique and improved technology. Initially, mastectomy inevitably involved sacrifice of the nipple areolar complex, and immediate reconstructions relied exclusively on submuscular implant placement or use of the transverse rectus abdominus flap. In recent years, with the increasing appreciation of patient reported outcome measures in women undergoing risk reducing surgery [89], surgeons have sought more aesthetically focussed reconstruction options whilst not compromising risk-reduction principles. This has allowed the safe introduction of skin sparing and nipple sparing mastectomy [90] and autologous reconstruction with deep inferior epigastric perforator flaps [91] or single-stage prepectoral implant-based reconstruction [92]. The use of acellular dermal matrices has revolutionised implant-based reconstruction, allowing structural support of implants within a

reconstruction to mimic natural breast ptosis [93]. Further improvements may come from the use of lipomodelling to improve aesthetics and thus patient satisfaction [94].

In *BRCA* PV carriers, RRM results in an improvement in survival, especially in women with *BRCA1* PVs, a result also found by others [95,96]. There may also be an improvement in women with *BRCA2* PVs with longer follow up [97]. Our previous studies indicated good acceptance and psychological health after RRM [37,38] More recent overviews have emphasised the enormous importance of excellent pre-surgical explanation, the presence of a psychologist in the multidisciplinary team and improved surgical techniques have been emphasised (Braude 2017) [98].

#### *3.8. Summary*

Here, we summarised the updated results from the Manchester FHC which spans the period from the inception of such clinics in the UK up to the present. A large proportion of these clinics in the UK are associated with Breast Units and work in conjunction with local Clinical Genetics Services.

The question remains regarding how their services may be improved (Table 3). At present, relatively small numbers of women are referred, partly because of the emphasis on family history for referral. Inclusion in FHCs of women at increased risk due to the presence of non-familial risk factors awaits the widespread introduction of MD and SNPs to risk prediction models. The introduction of new risk factors, such as MD and SNPs, is particularly important, as there is evidence that without them women are currently being given erroneous risk estimates that may result in imprecise treatment stratification. Clinical Genetics Services would be helped by more precise prediction of PVs. Consideration might be given to abandoning large panel tests and focussing on the nine genes in which the PVs are associated with a two-fold or more risk of BC [24].


**Table 3.** Summary of "current practice" and issues to be addressed for each of the interventions discussed.

We are currently testing the feasibility of introducing identification and referral of higher risk women as part of routine screening, i.e., research with a focus on implementation as part of routine care in a NHSBSP could bring about a "step change" if implemented [6]. There is already some evidence that communicating breast cancer risk estimates as part of routine screening does not produce the harms that have been anticipated [99]. For instance, communicating risk estimates in this setting did not produce adverse emotional effects or effects on screening uptake [100].

It appears that it is timely to consider introduction of the service into primary care as is seen in the USA [101]. A challenge to implementation is that the risk estimation and treatment algorithms have become more complex and efforts to introduce the two models we have focussed on here have led to legitimate difficulties amongst busy primary care physicians [102,103]. Even the mainstream estimation of cardiovascular risk on practice computer systems is apparently only applied to half of those in need and only half of these who need it are treated (Q-RISK; Hippisley-Cox 2017) [104] suggesting difficulties with the primary care approach.

A possible approach, pioneered in Melbourne, is to develop a simplified version of the Tyrer–Cuzick model, called iPrevent, and to make it widely available to all women. The model is user friendly and provides suggested treatment pathways in addition to an individual's risk. If made widely available this could lead to patient-initiated referral for initial screening and SNP estimation to define definitive management [105]. Other measures may be to establish one-off breast density assessment for all women at a certain age (e.g., 40 years, as suggested above) to estimate BC risk and introduce further screening and preventive measures for those found to be at high risk [99,106].

#### **4. Conclusions**

We reported the activity in a clinic designed for referral of women concerned about their family history of breast cancer. The long period of the clinic illustrates the changes in risk estimation and management over the years. The time span also allows for multiple studies on the effectiveness of management, for example, the effectiveness of breast screening. It also allows for the study of and introduction of preventive approaches such as use of tamoxifen and anastrozole.

A major aim of the clinic is to reduce the incidence of and deaths from breast cancer. These will be reduced by screening, lifestyle change and chemoprevention. Improvements in their effectiveness depends upon more widespread introduction not only into the current at-risk population but also into the large proportion of women unknowingly at high risk

**Author Contributions:** Conceptualisation, A.H. and D.G.E.; Data curation, D.G.E.; Formal analysis, E.H. and D.G.E.; Investigation, A.G., S.H., M.W., A.M. (Anthony Maxwell), S.A., M.H., M.P., L.B., A.B., E.H., P.H., J.W. (Julie Wisely), A.W., R.G., J.A., A.M. (Andrew Maurice), S.C., J.W. (Julia Wiseman), F.L., D.P.F. and D.G.E.; Project administration, A.H. and D.G.E.; Writing—original draft, A.H., A.G., S.H., M.W., A.W., S.A., M.H., M.P., L.B., E.H., P.H., J.W. (Julie Wisely), F.L., D.P.F. and D.G.E.; Writing—review and editing, A.H., A.G., S.H., M.W., A.M. (Anthony Maxwell), S.A., M.H., M.P., L.B., A.B., E.H., P.H., J.W. (Julie Wisely), A.W., R.G., J.A., A.M. (Andrew Maurice), J.W., (Julia Wiseman), F.L. and D.G.E. All authors have read and agreed to the published version of the manuscript.

**Funding:** This paper received no external funding.

**Acknowledgments:** D.G.E., E.F.H., S.J.H., D.P.F. and A.H. are supported by the National Institute for Health Research (NIHR) BRC Manchester (Grant Reference Number: 1215-200074). This work was also supported by Prevent Breast Cancer and Breast Cancer Now. We thank all the women referred to the clinic, the many unnamed individuals who have helped run the clinic over the years and the many investigators who have assisted with our clinical studies.We also thank Lorna McWilliam for comments on the manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Aberrant Dyskerin Expression Is Related to Proliferation and Poor Survival in Endometrial Cancer**

**Rafah Alnafakh 1,2,3, Gabriele Saretzki 4, Angela Midgley 5, James Flynn 6, Areege M. Kamal 2,7, Lucy Dobson 1,2, Purushothaman Natarajan 1,2, Helen Stringfellow 8, Pierre Martin-Hirsch 8, Shandya B. DeCruze 1, Sarah E. Coupland <sup>9</sup> and Dharani K. Hapangama 1,2,\***


**Simple Summary:** Telomeres are the protective caps at the ends of chromosomes, and they are maintained by an enzyme called telomerase. Telomerase activity allows rapid reproduction of the cells (proliferation) of the lining of the womb (endometrium). Telomerase levels are high in cancers in general, including in endometrial cancer. Dyskerin is one of the main components of the telomerase enzyme. While the other main components of telomerase have been studied in endometrial cancer, there are no previous studies on dyskerin in the endometrium. Our study shows that dyskerin levels are significantly lower in endometrial cancer and levels are linked to the survival of women. Experimentally increasing dyskerin protein in endometrial cells in the laboratory reduces the rate of cell proliferation. Consequently, we propose that dyskerin may be a regulator of endometrial cancer cell proliferation, and further studies are required to test if it can be targeted to develop new therapies for endometrial cancer.

**Abstract:** Dyskerin is a core-component of the telomerase holo-enzyme, which elongates telomeres. Telomerase is involved in endometrial epithelial cell proliferation. Most endometrial cancers (ECs) have high telomerase activity; however, dyskerin expression in human healthy endometrium or in endometrial pathologies has not been investigated yet. We aimed to examine the expression, prognostic relevance, and functional role of dyskerin in human EC. Endometrial samples from a cohort of 175 women were examined with immunohistochemistry, immunoblotting, and qPCR. The EC cells were transfected with Myc-DDK-DKC1 plasmid and the effect of dyskerin overexpression on EC cell proliferation was assessed by flow cytometry. Human endometrium expresses dyskerin (*DKC1*) and dyskerin protein levels are significantly reduced in ECs when compared with healthy postmenopausal endometrium. Low dyskerin immunoscores were potentially associated with worse outcomes, suggesting a possible prognostic relevance. Cancer Genome Atlas (TCGA) ECs dataset (*n* = 589) was also interrogated. The TCGA dataset further confirmed changes in *DKC1* expression in EC with prognostic significance. Transient dyskerin overexpression had a negative effect on EC cell proliferation. Our data demonstrates a role for dyskerin in normal endometrium for the first time

**Citation:** Alnafakh, R.; Saretzki, G.; Midgley, A.; Flynn, J.; Kamal, A.M.; Dobson, L.; Natarajan, P.; Stringfellow, H.; Martin-Hirsch, P.; DeCruze, S.B.; et al. Aberrant Dyskerin Expression Is Related to Proliferation and Poor Survival in Endometrial Cancer. *Cancers* **2021**, *13*, 273. https://doi.org/10.3390/ cancers13020273

Received: 1 November 2020 Accepted: 8 January 2021 Published: 13 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional clai-ms in published maps and institutio-nal affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

and confirms aberrant expression with possible prognostic relevance in EC. Interventions aimed at modulating dyskerin levels may provide novel therapeutic options in EC.

**Keywords:** dyskerin; *DKC1*; endometrial cancer; telomerase; proliferation; telomeres

#### **1. Introduction**

Telomeres are specialised nucleoprotein complexes consisting of tandem repeats of TTAGGG and associated specific shelterin proteins [1]. They prevent chromosomal ends from being identified as DNA damage and protect them from degradation and end to end fusion [2,3]. With each round of cell division, telomeric DNA is lost due to the end replication problem as well as oxidative stress [4,5]. In mitotic cells, critical shortening of telomeres induces apoptosis and senescence [6]. Telomerase is a specialised reversetranscriptase which maintains and elongates telomeres [7] and is composed of: (i) the template-containing telomerase RNA component (TERC), (ii) the catalytic component of the enzyme, human telomerase reverse transcriptase (hTERT) and (iii) the protein dyskerin as one of the main core components [8]. In most human somatic cells, telomerase activity (TA) is either undetectable or very low [9]. However, human cells with high replicative demand such as lymphocytes [10] epithelial cells [11] and tissue stem cells have active or inducible telomerase [12]. The human endometrium is a highly regenerative tissue with a dynamic TA corresponding to epithelial proliferation [13]. Most cancer cells express constitutively high TA, providing them with an indefinite proliferative ability [14].

EC is the commonest gynaecological malignancy in developed countries, with an increasing incidence [15]. In an era of decreasing cancer-related deaths reported for most other cancers, mortality due to EC is expected to increase [16]. Therefore, novel biomarkers to stratify high-risk patients for therapy as well as novel therapeutic targets are urgently required to reduce the rising EC-associated mortality and morbidity.

High TA has been reported in over 90% of all ECs [17]. hTERT and hTERC expression levels and TA measured by Telomere Repeat Amplification Protocol (TRAP) assay have been previously reported in the healthy endometrium [13] and in ECs [17,18]. However, dyskerin, which forms the foundation of the H/ACA lobe structure of the telomerase holo-enzyme, has not been studied in normal or pathological endometrium. Dyskerin protein is encoded by the *DKC1* gene located on the X chromosome [19] and it stabilises hTERC and enhances TA [20]. Dyskerin also has an extra-telomerase function in ribosomal biogenesis [21,22].

Available evidence suggests either the gain or loss of dyskerin to be carcinogenic [23,24]. High dyskerin levels have been reported in breast and prostate cancers [21,25,26] while decreased levels of dyskerin had been linked to carcinogenesis in the pituitary gland [27]. Low dyskerin levels observed in dyskeratosis congenita (DC) [28] have also been associated with an increased cancer-susceptibility before the age of 30 due to prematurely shortened telomers [29]. This observation is also in agreement with the only available animal model, where half of the hypomorphic *DKC1* mutant (*DKC1m*) mice (with decreased *DKC1* expression) developed various malignancies [22]. We, therefore, aimed to explore the role of dyskerin in endometrial carcinogenesis.

#### **2. Results**

*2.1. In Silico Interrogation of the Cancer Genome Atlas (TCGA) Endometrioid and Serous EC Dataset Demonstrates Dysregulation of DKC1 to Be Associated with Poor Survival*

Analysis of the TCGA dataset demonstrated a more than 2-fold upregulation of DKC1 RNA levels in 69/477 (14.65%) of the endometrioid and serous ECs compared with a set of normal endometrial samples obtained from 35 EC patients, at 2–3 cm distance from the cancer margin [30]. High *DKC1* expression was significantly associated with poor prognosis (*<sup>p</sup>* = 2 × <sup>10</sup><sup>−</sup>5, Cox-regression = 0.91) (Figure 1).

**Figure 1.** Kaplan-Meier survival curve for the association between DKC1 mRNA levels and overall survival (*<sup>p</sup>* = 2 <sup>×</sup> <sup>10</sup>−5, Cox-regression = 0.91) in The Cancer Genome Atlas (TCGA) dataset (endometrioid and serous endometrial cancer) {*n* = 477}.

The mutation frequency of the *DKC1* gene in ECs was low (9/235, 3.69%) and consisted of mainly missense mutations that occurred without any *TERC* gene mutations (Figure S1). Patients with ECs harboring a mutant *DKC1* gene seemed to have a better clinical outcome, compared with cancers carrying a wild-type *DKC1* gene (Figure S2). Twenty out of 464 (4.31%) ECs also demonstrated a copy number variation (mostly loss) of the *DKC1* gene. However, the TCGA dataset did not show a correlation of DKC1 RNA levels with the tumour grade (r2 = 0.19, *<sup>p</sup>* = 9.61 × <sup>10</sup>−23) or clinical stage (r2 = 0.02, *<sup>p</sup>* = 7.27 × <sup>10</sup>−4), (Figure S3A,B). Similarly, there was no correlation between RNA levels of *DKC1* with steroid receptor genes, *TERT* (r2 = 0.03, *<sup>p</sup>* = 3.43 × <sup>10</sup>−4), (Figure S4A) or *TERC* (r2 = 0.04, *<sup>p</sup>* = 1.62 × <sup>10</sup><sup>−</sup>4), (Figure S4B). High DKC1 RNA levels were observed in *TP53* mutated ECs (*<sup>p</sup>* = 1.23 × <sup>10</sup>−8) (Figure S5A) while in contrast, lower DKC1 RNA levels were observed in *FGFR2* (*<sup>p</sup>* = 7.90 × <sup>10</sup>−3) (Figure S5B), *PTEN* (*<sup>p</sup>* = 2.90 × <sup>10</sup>−6), *PIK3R1* (*<sup>p</sup>* = 0.02), (Figure S6A,B) and *CTNNB1* (*<sup>p</sup>* = 1.67 × <sup>10</sup>−3) mutated ECs. No significant difference in *DKC1* RNA level was observed in *TERC*, *TERT*, *POLE*, *PIK3CA*, *KRAS*, and *ARID1A* mutated ECs compared with un-mutated EC samples.

#### *2.2. Study Cohort*

Patients' demographic details are detailed in Table 1. Women with high-grade EC (HGEC) were significantly older than those with low-grade EC (LGEC) and healthy postmenopausal (PM) women (*p* < 0.001, *p* = 0.002, respectively). A significantly higher body mass index (BMI) was observed in the endometrial hyperplasia with a cytological atypia (EHA) group compared with the healthy PM women (*p* < 0.001) and in the EC group. There was an apparent trend for the LGEC group to have a higher BMI compared with the HGEC group (*p* = 0.06).


**Table 1.** Demographic features of study groups.

Abbreviations: Body mass index (BMI); high-grade endometrial carcinoma (HGEC); low-grade endometrial cancer (LGEC); \* Data expressed as median (range). \*\* BMI data were available for only 161 cases.

#### *2.3. Dyskerin mRNA Was Lower in ECs Compared with Normal PM Endometrium*

In contrast to the TCGA data in our patient samples, DKC1 mRNA levels showed a tendency towards downregulation in ECs in comparison with endometrium from healthy PM women (*p* = 0.06), (Figure 2A). No difference in the DKC1 mRNA level was observed between LGEC and HGEC.

#### *2.4. Dyskerin Protein is Significantly Reduced in ECs When Compared with Healthy PM Control Endometrium*

When EC samples were compared with healthy PM endometrium, immunoblotting demonstrated significantly reduced dyskerin protein levels (normalised to the epithelial marker pancytokeratin (*p* = 0.02, Figure 2B and Figure S7A), but significantly higher TA (*p* = 0.009, Figure 2C). IHC staining revealed the presence of dyskerin protein at a cellular level. In both epithelial and stromal cells of the healthy PP and PM endometrium, immunostaining was primarily localised in the nucleus and/or nucleolus (Figure 2D) and epithelial cells displayed stronger staining than the stroma. Dyskerin immunoscores were significantly lower in PP compared with PM (*p* = 0.03, Figure 2E). However, neither dyskerin quickscores nor DKC1 mRNA levels correlated with TA (Spearman r = −0.12, *p* = 0.16 and Spearman r = 0.04, *p* = 0.77, respectively).

**Figure 2.** DKC1 mRNA and dyskerin protein in human endometrium. (**A**) DKC1 mRNA is normalised to geometric means of PPIA and YWHAZ and measured by qPCR in endometrial tissue samples: healthy postmenopausal (PM) (*n* = 6) and endometrial cancer (EC) (*n* = 22). Mann-Whitney test. (**B**) The amount of dyskerin protein was evaluated by immuno-blotting in healthy PM (*n* = 4) and EC (*n* = 4), Glyceraldehyde 3-Phosphate Dehydrogenase (GAPDH) was used to ensure equal loading of protein. Dyskerin protein levels in epithelial cells of tissue samples were analysed by normalising to pancytokeratin (panck). Mann-Whitney test, \* *p* < 0.05. (**C**) Telomerase activity (TA) in healthy endometrial PM (*n* = 6) and EC (*n* = 32) was measured using a Telomere Repeat Amplification Protocol (TRAP) assay, Mann-Whitney test, \*\* *p* < 0.01. AU: arbitrary units (**D**) Representative microphotographs illustrating dyskerin IHC staining at the cellular level in endometrial samples in (**1**) normal proliferative phase (PP) endometrium, (**2**) healthy PM endometrium, (**3**) endometrial hyperplasia with cytological atypia (EHA) and (**4**) EC. Positive staining appears brown. Magnification 400×. Scale bar 50 μm. (**E**) Immunostaining quickscores for dyskerin protein in the human endometrium, healthy PP (*n* = 16), PM (*n* = 30), EHA (*n* = 15), EC (*n* = 109). Kruskal-Wallis test, \* *p* < 0.05, \*\*\*\* *p* < 0.0001.

#### *2.5. Loss of Dyskerin Was a Feature of Precancerous and Cancerous Endometrial Epithelial Cells*

Dyskerin immunoscores were significantly lower in EHA and EC compared with normal PM endometrial epithelium (*p* = 0.01 and *p* < 0.0001, respectively, Figure 2E). All ECs in this cohort (Figure 3A) showed lower dyskerin scores compared with healthy PM endometrial tissue (Figure 3B), the difference was significant in endometrioid, carcinosarcoma and clear cell EC (*p* < 0.0001, *p* < 0.0001, and *p* = 0.002, respectively) and this reduction remained significant even when the histological LGEC (*p* < 0.001) and HGEC (*p* < 0.001) were considered separately (Figure 3C). There were no significant differences in dyskerin immunostaining among different EC subtypes or between LGEC and HGEC (Figure 3B,C). Metastatic lesions (Figure 4A) had significantly higher dyskerin immunoscores compared

with their matched primary tumours (*p* = 0.003, Figure 4B), whereas ECs at advanced clinical stages (FIGO stages III&IV) had significantly lower dyskerin immunoscores compared with those at early stages (FIGO stages I&II, *p* = 0.04, Figure 4C).

**Figure 3.** Immunostaining of dyskerin in endometrial cancer subtypes (*n* = 109). (**A**) Representative microphotographs of dyskerin in human ECs. (**1**–**3**) grade 1–3 endometrioid carcinoma, (**4**) serous subtype, (**5**) Carcinosarcoma and (**6**) clear cell carcinoma. Positive staining appears brown. Magnification 400×. Scale bar 50 μm. (**B**) Dyskerin immunoscores in healthy PM (*n* = 30) and various EC subtypes including endometrioid (E) (*n* = 65), Serous (S) (*n* = 12), carcinosarcoma (CS) (*n* = 19), clear cell carcinoma (**C**) (*n* = 10), mixed cell adenocarcinoma (M) (*n* = 2) and dedifferentiated EC (DD) (*n* = 1). \*\* *p* < 0.01, \*\*\*\* *p* < 0.0001. Kruskal-Wallis test. (**C**) Dyskerin immunoscores in human endometrial epithelium of healthy PM (*n* = 30), LGEC (*n* = 53) and HGEC (*n* = 56). \*\*\* *p* < 0.001. Kruskal-Wallis test.

**Figure 4.** Dyskerin immunostaining in endometrial cancers. (**A**) Representative microphotographs illustrating dyskerin immunohistochemical staining in primary endometrial cancer (EC) (**1**) and matched metastatic lesion (**2**). Positive staining appears in brown. Magnification 400×, Scale bar 50 μm (**B**) Difference in dyskerin immunoscores in primary EC samples versus matched metastatic lesions (*n* = 30) each, \*\* *p* < 0.01. (**C**) Difference in dyskerin immunoscores between early-stage ECs (FIGO stage I–II) (*n* = 63) and advanced stage ECs (FIGO stage III–IV) (*n* = 43). Mann-Whitney test, \* *p* < 0.05.

#### *2.6. Endometrial Epithelial Dyskerin Immunoscores Correlate with ERβ Scores and Inversely with the Ki67 Proliferation Index (PI)*

Dyskerin immunoscores in endometrial samples correlated with ERβ immunoscores (Spearman *r* = 0.46, *p* < 0.0001), while an inverse correlation was found with the Ki67 PI (Spearman *r* = −0.34, *p* < 0.0001). No correlation was identified with other steroid receptors' immunoscores (Table S1). Figure S7B shows immunostaining of dyskerin, Ki67 and steroid recepters.

#### *2.7. Survival Analysis*

According to the national guidance, patients were followed-up for at least 3 years after primary surgery in the two recruiting centers during the study period. By March 2020, follow-up data were available for 108 out of 109 women in our cohort [31]. During this follow-up period, there were 10 recurrent tumours and 38 deaths (27 as a result of

disease progression and 11 from other causes). Worse outcomes were found in women with low dyskerin immunoscores. All outcomes analysed, including disease-free survival (DFS), cancer-specific survival (CSS), and overall survival (OS) suggested high dyskerin immunoscores to be potentially favourable (*p* = 0.08, *p* = 0.07, and *p* = 0.06, respectively, Figure 5A–C). For low dyskerin scores, the DFS hazard ratio (HR) = 1.92, 95% CI of HR (0.9200–4.006), CSS HR = 1.991, 95% CI of HR (0.9300–4.261), and OS HR = 1.841, 95% CI of HR (0.9667–3.506). When we only considered the endometrioid and serous ECs (similar to the selected TCGA EC dataset), low dyskerin immunoscores were still possibly suggestive of worse clinical outcomes with HR = 2.169, 95% CI of HR (0.7999–5.882), HR = 1.762, 95% CI of HR (0.5607–5.539) and HR = 1.698, 95% CI of HR (0.6925–4.165) for DFS, CSS, and OS, (Figure S7C–E), respectively. However, the *p* values were not significant (DFS, CSS, and OS; *p* = 0.1, *p* = 0.3, and *p* = 0.2, respectively) and confidence intervals were wide. These findings therefore need to be interpreted with caution and require future validation.

**Figure 5.** Kaplan Meier survival curves for the correlation between dyskerin immunoscores and patient outcome. (**A**) Disease-free survival (DFS), the median DFS time is undefined for low dyskerin and high dyskerin endometrial cancer groups. Hazard ratio (HR) = 1.92, 95% CI of the ratio (0.9200–4.006) (**B**) Cancer-specific survival (CSS), the median CSS time was undefined for low dyskerin and high dyskerin endometrial cancer groups. HR = 1.991, 95% CI of HR (0.9300–4.261) and (**C**) Overall survival (OS) in endometrial cancer samples (*n* = 109). Median OS time: Low dyskerin protein 8.00 months, High dyskerin protein 2.00 months. Low dyskerin/high dyskerin median survival Ratio: 0.5217, 95% CI of ratio (0.004444–1.039) HR = 1.841, 95% CI of HR (0.9667–3.506). A quickscore of 6 was chosen as the cut-off point. The *p* values relevant to the difference between low and high dyskerin protein levels in endometrial cancer groups that is visually represented in Kaplan Meier survival curves from the log-rank test.

When clinicopathological features were considered, dyskerin immunoscores inversely correlated with cervical invasion (*p* = 0.01, Table S2).

#### *2.8. In Vitro Transient Transfection of ISK Cells with the DKC1 Gene Resulted in Successful Overexpression of Dyskerin Protein*

A positive band corresponding to endogenous dyskerin was observed in negative controls (empty vector and non-transfected cells) and in transfected Ishikawa (ISK) cells at 6, 24, and 48 h after transfection (Figure 6A and Figure S8A). Exogenous dyskerin protein was first observed at 24 h and was still present at 48 h (although was decreased) in the *DKC1* transfected cells (Figure 6A and Figure S8B).

**Figure 6.** Transient overexpression of DKC1 in ISK cells. The plasmid and the empty vector (EV) used were tagged with the synthetic DYKDDDDK (DDK) protein to discern the transfected cells by using an anti-DDK antibody. (**A**) Immunoblot showing the level of dyskerin protein in DKC1 and EV transfected and non-transfected (NT) ISK cells. Cells were harvested 6, 24, and 48 h following transfection. Endogenous and exogenous dyskerin bands were present at the molecular weight of 58 and 60 KDa (red and blue arrows, respectively). DDK bands (yellow arrows) were observed at 60 KDa. Glyceraldehyde 3-Phosphate Dehydrogenase (GAPDH) bands were at 37 KDa. (**B**) Flow cytometric histogram showing the level of DDK tag protein in ISK cells. Cells positively stained with anti-DDK tag antibody represent transfected cells. (**C**) Cell proliferation was analysed using flow cytometry. ISK cells were stained with CellTrace Carboxyfluorescein Diacetate Succinimidyl Ester (CFSE) and fluorochrome-conjugated DDK Tag Antibody. Transfected cells (blue curve) and non-transfected cells (red curve). Higher proliferation is suggested when the curve was shifted to the left. (**D**) The difference in median fluorescence index (MFI) between transfected (T) and non-transfected ISK cells. \*\* *p* < 0.01, Wilcoxon signed-rank test.

A signal at the correct molecular weight (MW) demonstrates that the DDK tag peptide was present in transfected cells at 24 and 48 h only (Figure 6A and Figure S8C). Immunofluorescent staining with an anti-dyskerin antibody demonstrated the presence of endogenous dyskerin, characterised by a punctate pattern that was exclusively localised in the nuclei of all cells (cells transfected with *DKC1* and empty vector and in non-transfected cells) (Figure S8D) Exogenous dyskerin was located both in the nucleus and in the cytoplasm and observed only in *DKC1* transfected cells (Figure S8D).

Flow cytometric analysis of ISK cells 48 h after transfection revealed the transfection efficiency to be 18.1% in the dyskerin transfected cells (Figure 6B), 11% in the empty vectortransfected cells (Figure S9A), and 1.76% in the non-transfected control (false positive level) (Figure S9B). Figure S10 shows different negative staining controls used in the transient transfection experiment and Figure S11 shows the empty vector control map.

#### *2.9. Transient Overexpression of the DKC1 Gene Reduced ISK Cell Proliferation In Vitro*

Overexpression of *DKC1* reduced cellular proliferation rates (Figure 6C), as demonstrated by a significantly higher median fluorescence intensity (MFI) of Carboxyfluorescein Diacetate Succinimidyl Ester (CFSE) staining in *DKC1* transfected cells compared with the non-transfected cells (*p* = 0.007) (Figure 6D). Dyskerin transfected cells also have a lower proliferation rate compared with those transfected with the empty vector using an immune-staining method (Figure S12).

#### **3. Discussion**

To our knowledge, this is the first study to examine the expression of the telomerase core-component dyskerin in human endometrium. It validates the findings of our in silico interrogation of a published, large TCGA *DKC1* gene alteration profile of endometrioid and serous ECs, using a cohort of human ECs containing all histological EC-subtypes with transcriptional and protein data. We have demonstrated that healthy PP and PM endometrium express the *DKC1* gene and have detectable dyskerin protein levels. Importantly, EC samples have significantly lower dyskerin protein levels when compared with healthy PM controls. Our findings are important for the following reasons: (i) we examined the endometrial dyskerin protein levels with immunoblotting and at the cellular level with IHC for the first time; (ii) our local patient cohort consisted of all EC subtypes, including carcinosarcoma, dedifferentiated, mixed-cell adenocarcinoma and clear cell cancer types, precancerous EH samples and metastatic EC lesions, as well as external control healthy endometrium (both healthy PM and PP samples) to increase the generalisability of the data; (iii) Importantly, our data suggests a possible better clinical outcome in ECs containing high levels of dyskerin protein in comparison with those with lower dyskerin levels. Our data, therefore, fill the gaps in the current literature, including the TCGA dataset.

Sufficient dyskerin levels are required for competent TA to overcome telomere attrition [28]. *DKC1* dysregulation is associated with a high incidence of cancers in DC patients (reduced *DKC1*) and in *DKC1* hypomorphic mice [22], but no reports are available of DC associated with EC. Although high TA in over 90% of ECs had been reported, that is usually associated with short telomeres [32].

Examination of the TCGA dataset only identified *DKC1* out of the three core telomerase components to have an altered gene expression, with a prognostic relevance in ECs. Our data also suggests that dyskerin protein levels in ECs correlate with differences in patient outcomes. Variable dyskerin levels are also reported in other cancers [25,33]. Data from our cohort and the TCGA dataset jointly suggests a dysregulation of dyskerin in ECs. However, our cohort results differ from the TCGA data, and this discrepancy may be due to different "normal controls" used in the two studies and the fact that we examined protein rather than only mRNA levels. It is important to appreciate that endometrioid/serous ECs included in the TCGA data usually originate from a background of EHA or endometrial intraepithelial neoplasia (EIN). Thus, the normal tissue within 2 cm from the tumour included in the TCGA data as normal endometrium is likely to include

hyperplastic tissue or EIN lesions. Our cohort data is more generalisable since we histologically confirmed our external healthy control tissue obtained from a well-characterised and age-matched population.

Many studies on different cancer types reported a high expression of the *DKC1* gene and dyskerin protein to be associated with poor prognosis [25,34]. For example, in contrast to our results on ECs, reports on prostate, hepatocellular carcinoma, and colorectal cancers showed that high *DKC1* is commonly associated with an extensive tumour growth pattern [25,34,35]. Recently, Elsharawy et al. showed that high DKC1 mRNA or protein levels in breast cancer associated with poor patient outcome and unfavourable clinicopathological characteristics [26]. In a recent study in breast cancer, *DKC1* over-expression associated with unfavourable clinicopathological characteristics and poor outcome [26]. Two publicly available "Breast Cancer Gene-Expression Miner v4.3" [26] and TCGA breast cancer datasets revealed high DKC1 mRNA levels to significantly correlate with larger tumour size, higher tumour grades, and poor prognosis. At the protein level, high dyskerin protein levels, whether in the nucleus and/or nucleoli, were reported to be associated with aggressive features in breast cancer [26]. In those tissues, however, carcinogenesis is associated with reactivation of TA compared with healthy tissues [36], whereas high TA is a feature of healthy PP endometrium [37]. Therefore, we suggest that ECs are different in this respect, and consequently, endometrial carcinogenesis seems to be associated with a reduction of dyskerin protein and *DKC1* gene expression.

Advanced primary ECs (stage-III and IV) had significantly lower dyskerin protein levels compared with early stages, suggesting that dyskerin protein may be useful in stratifying EC patients for further therapy after primary surgery. Prior reports have suggested metastatic EC lesions to demonstrate a regressed phenotype when compared with the matched primary tumour [38] which agrees well with our dyskerin data. The observed dyskerin loss we report may also produce a pro-oxidant environment in EC cells as demonstrated in other cancer cells [39]. Therefore, reduced dyskerin protein in the context of the excessive cellular division in ECs may contribute to genomic instability that is known to be present, particularly in more advanced ECs. Dyskerin deficiency may also contribute to carcinogenesis by adversely influencing the translational machinery via affecting the balance in ribosomal proteins [33] and by modifying the splicing of specific pre-mRNAs, or by altering the level of certain snoRNAs [40,41]. These mechanistic aspects need to be examined in future studies.

The healthy quiescent PM endometrium with absent cellular proliferative activity had high dyskerin levels. TA positively correlated with endometrial epithelial proliferation [13] and the downregulation of dyskerin protein in ECs in comparison with the healthy PM endometrium we observe, occurred in a background of high TA and Ki67 levels [18]. This suggests a tumour suppressor function [22] and an inhibitory effect on endometrial epithelial cell proliferation for dyskerin in ECs. Therefore, we sought to examine the functional consequence of overexpressing the *DKC1* gene on cell proliferation using a cell line that reflecting low grade ECs. Dyskerin knock-out is lethal, and thus all cells (independent of detectable TA) express the dyskerin gene/protein. The available *in vitro DKC1* gene manipulation studies had only examined knocking down of the *DKC1* gene [25] but not over-expression and they also did not examine cellular proliferation as an outcome. Knock-down studies in prostate carcinoma cells demonstrated dyskerin to be crucial in protein biosynthesis [25]. Both high and low dyskerin is associated with carcinogenesis [25], which fundamentally demonstrates the cardinal feature of excessive cellular proliferation. Our data demonstrates a consequential reduction in cell proliferation when dyskerin is overexpressed in the EC cell line, therefore establishing a functional effect of dyskerin on cell proliferation for the first time.

Reduction in dyskerin rendered human breast cancer cells to be more prone to incorrect codon recognition and induced a defect in rRNA uridine modification resulting in altered ribosome activity [42]. Low dyskerin expression levels correlated with poor overall survival of Chronic Lymphocytic Leukaemia (CLL) patients following chemotherapy [33].

The authors proposed that reduced dyskerin may cause a reduction of the synthesis of subsets of ribosomal proteins, and selectively alters the translatome of the cancer cells to increase their aggressiveness [33]. Loss of dyskerin dysregulates initiation of translation of tumour suppressor proteins such as p53 and p27 and thus may promote carcinogenesis [27,43]. In addition, dysregulation of p53 translation has been reported in DC patients with reduced dyskerin function via its internal ribosome entry segment being impaired resulting in increased cellular proliferation [44,45]. The exact mechanistic pathway by which dyskerin exerts this observed anti-proliferative effect on EC cells remains to be explored in future studies.

We used opportunistic recruitment and available archived samples in our study to answer our research question. This meant inclusion of retrospectively collected patient samples, and only a small proportion of cases seen in the centres over that time period were included in the study. Although this is a limitation of our study, since no previous data available for the levels of dyskerin protein in the EC, our study, which included a relatively sizeable EC cohort with associated important clinical details, fills the current gap in the literature, and provides significantly different results to inform sample sizes for adequately powered studies in the future.

Another limitation to our study is that we have included a similar number of LGEC and HGEC, meaning the stage distribution was skewed towards metastatic disease in the local cohort. This caused our sample to be deviated from the real incidence of nonendometrioid EC; however, this offers us a better assessment of HGECs, which are usually associated with poor prognosis. Although we have recruited women without known endometrial pathology as normal controls for EC samples, a potential limitation would be that all these control women were undergoing hysterectomy for a non-cancerous pathology, thus they may not represent asymptomatic and completely healthy normal women. Therefore, our findings require further validation in future prospective studies.

Endometrial TA and hTERT levels have been shown to be under hormonal regulation [13] and correspondingly, endometrial dyskerin immunoscores revealed a significant positive correlation with ERβ immunostaining. This may suggest dyskerin expression to be under estrogen regulation mainly via ERβ. ERβ is known to harness the estrogen-driven mitotic effect of ERα [46], therefore inducing dyskerin levels may also be a part of the ERβ-associated inhibition of the endometrial epithelial proliferation. Further studies are required to examine the hormonal regulation of dyskerin in human endometrium.

#### **4. Materials and Methods**

*4.1. Study Groups:*

#### 4.1.1. TCGA Database Cohort

The publicly-available TCGA cohort of uterine cancers included data for RNA levels (*n* = 477), copy number variation (*n* = 464), and somatic mutation (*n* = 235); for *DKC1*, the data were interrogated using Illumina's Base Space Cohort Analyzer application (BSCA) [47] (Software; https://www.illumina.com/informatics/research/biological-datainterpretation/nextbio.html; Illumina, San Diego, CA, USA) [48]. The normal endometrial controls were obtained from 35 EC patients at 2–3 cm distance from the cancer margin [30].

#### 4.1.2. Local Study Cohort

The study was performed in accordance with the Declaration of Helsinki. The Liverpool and Cambridge Adult Research Ethics Committees (LREC 09/H1005/55, 11/H1005/4 and CREC 10/H0308/75) approved the study. A total of 175 endometrial samples collected from women undergoing hysterectomy in the Liverpool Women's Hospital (LWH) and Lancashire Teaching Hospitals Trusts from 2009 to 2017 were included. Our cohort included a total of 15 endometrial samples with histological hyperplasia and cytological atypia were collected from patients undergoing hysterectomy at LWH. Out of these, three women had prior histological evidence of hyperplasia in an endometrial biopsy with ongoing symptoms of irregular or heavy menstrual bleeding; another 12 samples were from paraffin

blocks of hyperplastic changes adjacent to EC that were retrieved from the Histopathology Department archive at the Royal Liverpool University Hospital.

Additionally, a total of 109 histologically confirmed EC samples from patients who underwent staging operations at LWH or at Lancashire Teaching Hospitals during the period between 2009 and 2017 were also recruited to the current study. Out of those 109 samples, 60 were pipelle biopsies collected at the time of their hysterectomy as part of their primary surgical treatment for EC. The remaining samples were paraffin blocks retrieved from the Histopathology Department archives at the Royal Liverpool University Hospital, or Lancaster Teaching Hospital. Paraffin blocks of 30 metastatic lesions from some of these women with ECs that were obtained during the same primary surgery were also studied. The sites of metastases were as follows: lymph nodes (*n* = 11), omentum (*n* = 7), parametrium (*n* = 5), soft tissue (*n* = 4), fallopian tube (*n* = 1), cervix (*n* = 1), and urinary bladder (*n* = 1).

None of the included EHA or EC patients had received hormonal treatment, chemotherapy, or pelvic radiation prior to surgery when the endometrial samples were harvested.

Demographic data are shown in Table 1. Experienced gynaecological pathologists confirmed the histological type and grade of EC specimens according to FIGO classification [49]. Considering the clinical relevant outcome, we further categorised the EC samples as low-grade (LGEC), consisting of grade 1 and grade 2 endometrioid EC or high-grade (HGEC), including grade 3 endometrioid, serous, clear cell carcinomas, carcinosarcoma, Mixed cell adenocarcinoma, and dedifferentiated ECs [43,50] as shown in Table 1. Healthy endometrial tissue specimens were collected from women undergoing hysterectomy for benign gynaecological pathologies such as prolapse or heavy bleeding without a known endometrial pathology (a full-thickness samples). Since EC is a disease mainly affecting postmenopausal (PM) women, 35 age-matched healthy endometrial tissue samples were included as an external control group. Some previous authors have suggested that the proliferative phase (PP) control samples were more suitable as a healthy comparator because EC is a proliferative disease; therefore, we also included a second external control group of 16 normal healthy premenopausal endometrial PP samples. Samples from healthy women were thus assigned to premenopausal (PP) and postmenopausal (PM) groups according to the last menstrual date and histological criteria [51].

#### *4.2. Collection of Endometrial Samples*

Once the uterus was removed at hysterectomy, in theatre, endometrial biopsies were collected by a trained member of the research team or the operating surgeon. Full-thickness endometrial biopsies were obtained from healthy women undergoing a hysterectomy, as previously described by cutting a thin slice of endometrium attached to underlying myometrium after opening the anterior uterine aspect in the coronal plane [52]. In order to avoid interference with pathological diagnosis and staging, samples from women undergoing primary surgery for EC were collected by using a pipelle suction curette (Laboratoire C.C.D., Paris, France). Each sample was split into two to three containers: (i) 15 mL 10% neutral buffered formalin (10% NBF) (Sigma, Dorset, UK) for immunohistochemistry study; (ii) 0.5 mL RNAlater (Sigma, Dorset, UK) for RNA extraction and PCR analysis; (iii) Immediately snap-frozen for immunoblotting and TRAP analysis.

#### *4.3. Immunohistochemistry (IHC)*

IHC was performed on 3 μm serial sections of formalin-fixed, paraffin-embedded endometrial tissue employing heat-induced antigen retrieval, and the ImmPRESS Polymerized Reporter Enzyme Staining System (Vector Laboratories, Peterborough, UK) as previously described [38]. The primary antibody sources, concentrations, and incubation conditions are detailed in Table S3.

Immunoreactivity for nuclear dyskerin was assessed using a modified quick score as previously described [53]. The four steroid receptors were evaluated semi-quantitatively using a four-tiered Liverpool endometrial steroid quick score (LESQS) as previously described [38]; the Ki67 proliferative index (PI) was evaluated as the percentage of positive cells of any intensity [38].

#### *4.4. Real-Time qPCR*

RNA was extracted, quantified, and reverse transcribed as previously described [53]. cDNA was amplified using iTaq universal SYBR Green supermix and CFX Connect Real-Time System (Bio-Rad, Hertfordshire, UK). Primers and reaction conditions are listed in Table S4 [35,54,55]. The 2−ΔΔCt method was used to calculate relative transcript level. *DKC1* expression was normalised to *YWHAZ* and *PPIA* reference genes [56,57].

#### *4.5. TRAP Assay*

TA was measured using a TeloTAGGG™ TRAP assay (Sigma-Aldrich, Dorset, UK) according to the manufacturers' manual and as previously described [13]. Absorbance was measured at 450 nm in an Omega spectrophotometer (BMG, Labtech, UK) and presented as arbitrary units (AU). A total of 1 μg of protein was used per sample, and negative controls without protein were included and their absorption was subtracted from those of the samples.

#### *4.6. Cell Culture*

Cultured ISK cells were maintained in Dulbecco modified Eagle medium/F12 (DMEM/ F12) supplemented with 10% (*v*/*v*) fetal bovine serum (FBS), L-glutamine, and penicillin/streptomycin at 37 ◦C in a 5% CO2 atmosphere. All cell culture reagents were purchased from Sigma-Aldrich (Dorset, UK) as previously described [58].

#### *4.7. Transient Transfection*

Transfection of ISK cells was performed twenty-four hours after seeding cells on 6 well plates at a density of 0.5 × <sup>10</sup><sup>6</sup> cells/well by using a mixture of MYC-DDK tagged Dyskerin plasmid (OriGene Technologies, Rockville, MD, USA, 3 μL) with Lipofectamine 2000 (Thermo Fisher Scientific, Loughborough, UK, 9 μL). The plasmid or the Lipofectamine was diluted in 250 μL of Gibco Opti-MEM I (Thermo Fischer Scientific, Loughborough, UK). Empty vector (Myc-DDK tagged pCMV6-Entry) (OriGene Technologies, Rockville, MD, USA) and non-transfected cells were used as negative controls. The diluted plasmids and Lipofectamine were incubated for 20 min at room temperature. In the meantime, DMEM/F12 culture medium with supplements (FBS, L-glutamine and antibiotics) was replaced with the same medium but without antibiotics. A total of 4–6 h after transfection, the medium containing transfection reagents was removed and replaced with a fresh one supplemented with FBS, L-glutamine, and antibiotics. The cells were incubated at 37 ◦C, 5% CO2. The plasmids used were tagged with the synthetic DYKDDDDK Tag (DDK) Tag protein to discern the transfected cells by using an anti-DDK antibody.

#### *4.8. SDS-PAGE and Immunoblotting*

Protein lysates from homogenised tissues and cultured cells were extracted using a Radioimmunoprecipitation assay (RIPA) buffer (Sigma-Aldrich, Dorset, UK) supplemented with protease inhibitor (Sigma-Aldrich, Dorset, UK) and phosphatase inhibitor (PhosSTOP, Roche Diagnostics Ltd., Burgess Hill, UK). Lysates were analysed by SDS-PAGE under reducing conditions on precast 12% gels (Mini-PROTEAN TGX, Bio-Rad, Hertfordshire, UK) and transferred to an Immune-Blot polyvinylidene difluoride (PVDF) membrane (Bio-Rad, Hertfordshire, UK). The primary antibody sources, concentrations, and incubation conditions are detailed in Table S3. Horseradish peroxidase (HRP)-linked secondary antibodies were from Thermo Fisher Scientific (Loughborough, UK). Signal detection was performed using SuperSignal West Dura Extended Duration chemiluminescent Substrate (Thermo Fisher Scientific, Loughborough, UK) and CL-Xposure film (Thermo Fisher Scientific, Loughborough, UK).

#### *4.9. Immunofluorescence*

In order to differentiate between endogenous dyskerin and exogenous overexpressed protein, immunofluorescent staining of dyskerin was performed, allowing examination of their respective location within ISK cell. Rabbit anti-dyskerin antibody (Santa Cruz Biotechnology, Dallas, TX, USA, 1:200) was added to the fixed cells, which were seeded onto coverslips in a 6 well plate. The secondary antibody was Alexa Fluor Anti-rabbit IgG (H + L), (Alexa Fluor 488 Conjugate), (Cell Signalling Technology, London, UK, 1:1000). The cells were mounted in DAPI containing medium (Vector Laboratories, Peterborough, UK,). Fluorescence was visualised with a Nikon Eclipse 50i microscope using NIS elements F software (Nikon, Tokyo, Japan). Rabbit and mouse isotype control antibodies were used as negative controls. Antibody details are provided in Table S3.

#### *4.10. CFSE Labelling and Flow Cytometry*

ISK Cells were initially labelled with CellTrace CFSE (Thermo Fisher Scientific, Loughborough, UK) according to manufacturers' guidelines, then fixed, permeabilised, and labelled with fluorochrome-conjugated primary antibody (anti-DYKDDDDK (DDK) Tag antibody [iFluor 647], Genscript, Piscataway, NJ, USA) and the corresponding fluorochromeconjugated isotype control antibody (Alexa Fluor 647 antibody, Biolegend, UK). The cells were then incubated (1 h at 37 ◦C in the dark). A Guava EasyCyte flow cytometer (Millipore, Watford, UK) was used to perform flow cytometry and FlowJo v10 (Becton Dickinson, Franklin Lakes, NJ, USA) was used for data analysis.

#### *4.11. Statistical Analysis*

Statistical differences between groups were calculated by non-parametric tests (Kruskal–Wallis or Mann-Whitney U-test) using the Statistical Package for the Social Sciences (SPSS) version 24 (IBM Corp, Armonk, NY, USA). Descriptive values were presented as median and range. Graphs were plotted using GraphPad prism 5 (GraphPad Software, San Diego, CA, USA). The correlation between immunostaining scores was determined with a Spearman test and the association between dyskerin immunoscores and the multiple clinicopathological parameters were evaluated by Pearson's Chi-square test. The duration of DFS was measured from the date of surgery to the date of EC recurrence or death from EC, while the CSS duration was calculated from the date of surgery to the date of death from EC. OS duration was measured from the date of surgery to the date of death caused by any reason. All the observations were censored at the last date at which the patient was seen. Kaplan-Meier survival curves were constructed. Cumulative proportions of survivors in the high and low level of dyskerin protein were compared using Log-rank test. A significant difference between groups was only achieved with *p* value < 0.05. Significance values have been adjusted by Bonferroni correction for multiple tests.

#### **5. Conclusions**

Taking these observations together, we concluded that dyskerin protein and the *DKC1* gene are expressed in healthy endometrium [59] and in ECs. Low dyskerin immunoscores were potentially associated with worse outcomes, suggesting a possible prognostic relevance. Furthermore, increased dyskerin protein levels in ISK cells seem to inhibit cell proliferation, and therefore, the observed loss of dyskerin in endometrial cancer tissue may contribute to the increased cell proliferation and the progression of these ECs.

The detailed role of dyskerin in normal endometrial regeneration as well as in pathological conditions such as EC in the context of telomerase biology is yet to be determined. Since TA is known to play an intricate role in endometrial epithelial cellular proliferation, further studies elucidating the associated telomerase and other functions of dyskerin in the human endometrium and in EC are warranted.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2072 -6694/13/2/273/s1, Table S1: Correlation between dyskerin quick scores with steroid receptors immunoscores and Ki67 proliferative index (PI) in endometrial samples. Table S2: Association between dyskerin protein immunoscores and clinicopathological parameters in EC samples. Table S3: Primary antibodies and conditions for IHC, immuno blotting and immunofluorescence. Table S4: Primer sequences used for qPCR amplification. Figure S1: Multi co-occurrence plot of somatic mutation, RNA level and copy number variation of *DKC1, TERT, and TERC* genes. Figure S2: Kaplan-Meier survival curve for the association between mutation status of the DKC1 gene and overall survival in endometrial cancers (ECs). Figure S3: Violin plot demonstrating the correlation between DKC1 RNA levels with tumour grades and stages. Figure S4: DKC1 RNA levels correlation with TERT and TERC RNA levels in The Cancer Genome Atlas (TCGA) dataset (endometrioid and serous endometrial cancers). Figure S5: Violin plot showing the association between DKC1 RNA levels with the mutation status: normal, mutant, or wild type + silence of TP53 and FGFR2 genes. Figure S6: Violin plot showing the association between DKC1 RNA levels and mutation status: normal, mutant or wildtype + silenceof PTEN and PIK3R1genes. Figure S7: Whole immunoblots, dyskerin, Ki67, and steroid receptors' immunohistochemical staining and the association of dyskerin immunoscores with survival outcome in endometrioid and serous EC samples {n = 77}. Figure S8: Transient overexpression of *DKC1* gene in ISK (Ishikawa) cells (Immunoblotting and immunofluorescence experiments). Figure S9: Negative controls used in the transient transfection experiment. Figure S10: Negative staining controls used in transient transfection experiment. Figure S11: pCMV6-Entry vector map. Figure S12: Transient overexpression of DKC1 in ISK cells.

**Author Contributions:** D.K.H. and P.M.-H. obtained the Ethical approval, and D.K.H. conceived the study design. D.K.H., R.A., and G.S. formulated experiments, analysed and interpreted data, produced figures, and produced the first draft. J.F. conducted the in silico study. Experimental data were produced by R.A., A.M., A.M.K., and G.S. with support from D.K.H. and S.E.C. The samples and outcome data were collected by D.K.H., L.D., P.N., S.B.D., P.M.-H., and H.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** The authors would like to acknowledge the support from Wellbeing of Women project grant RG1487 and RG2137 (DKH) and Higher Committee for Education Development in Iraq (R.A.).

**Institutional Review Board Statement:** The study was performed in accordance with the Declaration of Helsinki. The Liverpool and Cambridge Adult Research Ethics Committees (LREC 09/H1005/55, 11/H1005/4 and CREC 10/H0308/75) approved the study.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** Data is contained within the article or supplementary material.

**Acknowledgments:** The authors are grateful for Steven Lane for statistical advice, Josephine Drury and Kishen Popat for their help with immunohistochemistry, Stuart Ruthven of Royal Liverpool Hospital with supporting sample procurement, Helen Cox and Sarah Northey for assistance in preparing tissue sections, Lisa Heathcote for assistance with the BCA protein assay, Dada Pisconti for assistance with transient transfection, Phil Rudland, Stephane Gross, and Anthony Valentijn for assistance with immunoblotting, and Meera Adishesh for help with patient outcome data.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Article* **Metabolomic Biomarkers for the Detection of Obesity-Driven Endometrial Cancer**

**Kelechi Njoku 1,2,3, Amy E. Campbell 3, Bethany Geary 3, Michelle L. MacKintosh 1,2, Abigail E. Derbyshire 1,2, Sarah J. Kitson 1,2, Vanitha N. Sivalingam 1,2, Andrew Pierce 4, Anthony D. Whetton 3,4,\*,† and Emma J. Crosbie 1,2,\*,†**


**Simple Summary:** Endometrial cancer is the commonest cancer of the female genital tract and obesity is its main modifiable risk factor. Over 80% of endometrial cancers develop in the context of obesityinduced metabolic changes. This study focuses on the potential of plasma-based metabolites to enable the early detection of endometrial cancer in a cohort of women with body mass index (BMI) <sup>≥</sup> 30 kg/m2. Specific lipid metabolites including phospholipids and sphingolipids (sphingomyelins) demonstrated good accuracy for the detection of endometrial cancer, especially when combined in a diagnostic model. This study advances our knowledge of the role of metabolomics in endometrial cancer and provides a basis for the minimally invasive screening of women with elevated BMI.

**Abstract:** Endometrial cancer is the most common malignancy of the female genital tract and a major cause of morbidity and mortality in women. Early detection is key to ensuring good outcomes but a lack of minimally invasive screening tools is a significant barrier. Most endometrial cancers are obesity-driven and develop in the context of severe metabolomic dysfunction. Blood-derived metabolites may therefore provide clinically relevant biomarkers for endometrial cancer detection. In this study, we analysed plasma samples of women with body mass index (BMI) <sup>≥</sup> 30 kg/m<sup>2</sup> and endometrioid endometrial cancer (cases, *n* = 67) or histologically normal endometrium (controls, *n* = 69), using a mass spectrometry-based metabolomics approach. Eighty percent of the samples were randomly selected to serve as a training set and the remaining 20% were used to qualify test performance. Robust predictive models (AUC > 0.9) for endometrial cancer detection based on artificial intelligence algorithms were developed and validated. Phospholipids were of significance as biomarkers of endometrial cancer, with sphingolipids (sphingomyelins) discriminatory in postmenopausal women. An algorithm combining the top ten performing metabolites showed 92.6% prediction accuracy (AUC of 0.95) for endometrial cancer detection. These results suggest that a simple blood test could enable the early detection of endometrial cancer and provide the basis for a minimally invasive screening tool for women with a BMI <sup>≥</sup> 30 kg/m2.

**Keywords:** endometrial cancer; obesity; metabolomics; liquid biopsy; mass spectrometry; plasma biomarkers; artificial intelligence

**Citation:** Njoku, K.; Campbell, A.E.; Geary, B.; MacKintosh, M.L.; Derbyshire, A.E.; Kitson, S.J.; Sivalingam, V.N.; Pierce, A.; Whetton, A.D.; Crosbie, E.J. Metabolomic Biomarkers for the Detection of Obesity-Driven Endometrial Cancer. *Cancers* **2021**, *13*, 718. https://doi.org/10.3390/ cancers13040718

Academic Editor: Eduardo Nagore Received: 30 December 2020 Accepted: 6 February 2021 Published: 10 February 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

#### **1. Introduction**

Endometrial cancer is the most common gynaecological malignancy in the United Kingdom, where its incidence is rising in parallel with the obesity epidemic [1]. Obesity is the major risk factor for type I cancers of low-grade endometrioid morphology, with every 5 kg/m2 increase in body mass index (BMI) linked to a 60% increased cancer risk [2]. Almost half of all endometrial cancers are attributed to overweight (BMI ≥ 25 kg/m2) and obesity (BMI ≥ 30 kg/m2) [3]. The strong dose–response relationship portends a 10–15% lifetime risk of endometrial cancer in women with class III obesity (BMI ≥ 40 kg/m2) compared with a population average of 2% [4]. Whilst its aetiological importance is clear, the biology underpinning obesity-driven endometrial carcinogenesis is incompletely understood [5]. Adipose tissue is a rich source of oestrogens that stimulate endometrial proliferation, particularly when unopposed by progesterone in postmenopausal and anovulatory states [6]. Metabolically unhealthy obesity, rather than excess bodyweight per se, is of particular aetiological significance, with impaired glucose tolerance and chronic insulin resistance acting synergistically to increase endometrial cancer risk [7]. Type 2 diabetes mellitus is associated with a 62% upsurge [8], and uncontrolled diabetes mellitus a nearly five-fold greater susceptibility to endometrial cancer [9].

A recent study found occult endometrial abnormalities in 14% of women with class III obesity referred for weight loss management [10]. All but one had low-grade early-stage endometrial cancer or its precursor lesion, atypical hyperplasia. The early identification of these abnormalities in asymptomatic women could enable conservative management strategies that preserve fertility and/or reduce the morbidity of surgery [11,12]. Yet, no current screening programme exists for these high-risk women, partly because current diagnostics are invasive with low acceptability profiles and/or poor diagnostic accuracy [13]. A simple, minimally invasive endometrial cancer screening tool that can triage high-risk women for diagnostic workup, whilst safely reassuring those at low risk, would represent a major advance in the field [14,15].

High-throughput technologies and machine learning techniques have emerged as powerful tools for biomarker discovery and validation [15–19]. Metabolomics studies the downstream products of genomic, transcriptomic, and proteomic processes and best mirrors the human phenotype [20,21]. Thus, metabolomics has great potential to deliver clinically relevant biomarkers for endometrial cancer detection [22]. A blood-based test for cancer has broad appeal, being rated the second most important research priority for detecting cancer early in our recent James Lind Alliance Priority Setting Partnership [23]. A significant challenge is identifying cancer-relevant biomarkers within the context of severe metabolic dysfunction that characterises endometrial cancer risk. Here, we investigate the potential of plasma-based metabolites to detect endometrial cancer in a cohort of women with class III obesity, using a mass spectrometry-based metabolomics approach.

#### **2. Materials and Methods**

#### *2.1. Study Population*

This study included women with BMI ≥ 30 kg/m2 participating in clinical research, who donated blood samples and gave written, informed consent for their pseudo-anonymised data to be used for future research. The primary research studies received approval from the North West and Cambridge East Research Ethics Committees and were conducted according to the principles of the Declaration of Helsinki. Cases and controls were recruited at Manchester University and Salford Royal NHS Foundation Trusts, United Kingdom. Cases were confirmed to have endometrioid endometrial cancer based on specialist histopathological assessment of biopsy and/or hysterectomy specimens [24,25]. Controls were women referred for weight loss management and confirmed to have normal histology on endometrial biopsy [10]. Clinicopathological data included age, BMI, smoking status, menopausal status, parity, type 2 diabetes mellitus status and medications used. All tissue specimens were assessed by at least two specialist gynaecological pathologists reporting according to UK Royal College of Pathology standards. Blood samples were collected following an

overnight fast. Study investigators were blinded to the clinical information and biopsy results of subjects during acquisition of metabolomics data.

#### *2.2. Metabolomic Profiling*

Blood samples were collected in standard EDTA tubes, centrifuged at 2000 rpm for 10 min and the supernatant (plasma) was collected and stored at −80 ◦C. The samples were subsequently shipped to Metabolon Inc®, Durham, NC, USA, on dry ice and maintained at −80 ◦C until processed. Non-targeted MS metabolomic analysis was performed by Metabolon Inc®, according to company protocols and is summarised below.

#### 2.2.1. Sample Preparation

Sample preparation was carried out using the automated MicroLab STAR® liquid handling system (Hamilton Company, Reno, NV, USA). Recovery standards were added to the samples prior to extraction for quality control purposes. To optimise the recovery of chemically diverse metabolites, proteins were removed by precipitation with methanol under vigorous shaking GenoGrinder 2000 by Glen Mills Inc., Clifton, NJ, USA) followed by centrifugation. The resulting extract was split into four aliquots and prepared for subsequent analysis using solvents compatible with the various separation and detection methods. Zymark TurboVap concentration evaporator (SOTAX AG, Aesch, Switzerland) was used to remove organic solvents.

#### 2.2.2. Metabolite Separation and Detection

Multiple methods were used for metabolite separation and identification to maximise the number of metabolites detected. All methods were performed using a Waters AC-QUITY ultra-performance liquid chromatography (UPLC) system (Waters Corporation, Milford, MA, USA) and a Thermo Scientific Q-Exactive high resolution/accurate mass spectrometer (ThermoFisher Scientific, Waltham, MA, USA). This was interfaced with a heated electrospray ionisation (HESI-II) source and Orbitrap mass analyzer operating at 35,000 mass resolution. Three sample extract aliquots were analysed using reversed phase UPLC with tandem mass spectrometry (RP UHPLCMS/MS). A positive ion mode electrospray ionisation (ESI) was used for two aliquots chromatographically optimised for more hydrophilic and more hydrophobic compounds, respectively, and a negative ion mode ESI for the third aliquot. The fourth aliquot was analysed using negative ion mode ESI following elution from a hydrophilic interaction liquid chromatography column (HILIC UPLCMS/MS). The chromatographic conditions used and optimised for the various metabolite species are summarised in Table S1.

#### 2.2.3. Metabolite Identification

Raw data including molecular and fragment ions were searched against a reference library of over 14,000 metabolites based on authenticated standards. Metabolites were identified based on their chromatographic features (including MS/MS spectra), retention time/index (RI) and mass-to-charge ratio (*m/z*). The specific criteria used for biochemical identification included a retention index within a narrow window of the proposed identification and an accurate mass match to the library ± 10 ppm. MS/MS forward and reverse scores were used to control for false discovery rates. Ions that lacked a definite biochemical identity were given a numerical designation. Data curation was carried out by Metabolon, Inc, Durham, NC, USA data analysts to ensure accurate and consistent identification of metabolites as well as removal of artefacts, misassignments and background noise. Peak quantification was carried out using area under the curve analysis. Comparison of the peak area of a given metabolite in the sample to the peak area of a standard of known concentration was used to determine the metabolite concentration.

#### 2.2.4. Data Pre-Processing

Metabolite concentrations were reported in the form of standardised intensities. Each metabolite concentration was rescaled to set the median equal to 1 (by dividing the concentration of each metabolite by the median). Thus, the concentration of a given metabolite in a given sample was made relative to the median concentration of all the samples processed as part of the study. The presence of missing values in this study was indicated by the concentration of a given metabolite falling below an assay's limit of detection (LOD). Missing metabolite concentrations were imputed with a standardised intensity set at the minimum detected value for that compound.

#### *2.3. Data Analysis*

All statistical analyses were performed using R version 3.2.5 (R Development Core Team, Vienna, Austria), STATA version 16, and MetaboAnalyst 4.0. The Shapiro–Wilk test was used to assess normality of continuous variables. Descriptive analyses of the study demographic data (continuous and categorical) were performed using means (±standard deviations) and counts (%), respectively, with differences between groups assessed using Student's *t*-test for continuous variables and the chi-square test for categorical variables. The majority of the metabolite concentrations (median scaled standardised intensity) were not normally distributed. As such, non-parametric tests were used in subsequent analysis. Specifically, the Mann–Whitney *U* test was used to compare metabolite concentrations in the cancer group versus control group and for other group comparisons made. We applied a false discovery rate adjustment for multiple testing using the Benjamini–Hochberg correction method (q = 0.05). A computation of the ratio of metabolite concentrations in cases and controls was used to identify the direction and degree of fold change and allowed for the identification of the groups of metabolites with unidirectional alterations. Principal component analysis (PCA) and t-distributed stochastic neighbour embedding (t-SNE) plots were used to assess degree of separation between groups. Random forest modelling was used to identify the best-performing biomarkers and to develop predictive models for the detection of endometrial cancer. Eighty per cent of the samples were randomly selected to serve as a "training set" and the remaining 20% were used to test the model. Heat maps were generated based on hierarchical clustering of the top discriminatory metabolites using the Euclidean distance measure and the Ward algorithm. Row scaling (heat maps) was performed for each metabolite by the subtraction of the mean from each feature and then dividing by the standard deviation. Area under the receiver-operator characteristic curves (AUC) and the 95% confidence intervals were computed for both metabolites and metabolomics signatures. The selection of cut-off points was based on the Youden Index (J = max {Sensitivity + Specificity − 1}).

An overview of the study workflow is summarised in Figure S1.

#### **3. Results**

#### *3.1. Participant Demographics*

The study comprised 136 women with BMI ≥ 30kg/m2 of whom 67 had endometrioid endometrial cancer (cases) and 69 had histologically normal endometrium (controls). The median age and BMI for the cohort was 54 years (IQR 43, 65) and 46 kg/m<sup>2</sup> (IQR 39, 52) respectively. Cases were older and more likely to be post-menopausal and nulliparous while controls were more obese. The majority of the endometrial cancers were low-grade (91.0% grades I/II), early-stage (88.0% stage I) cancers with lymphovascular space invasion occurring in only 12 women (18.0% of cases) (Table 1). Participant demographics and clinicopathological characteristics are summarised in Table 1.


**Table 1.** Clinicopathological characteristics of the cohort.

*3.2. Metabolomic Analysis of Plasma Samples*

A total of 1137 metabolites were quantified in the study plasma samples of which 733 (64.5%) were biochemically defined. These included amino acids, fatty acids, biogenic amines, sphingolipids, steroids, hexoses, nucleotides, phospholipids, vitamins and xenobiotics. The remaining 35.5% were unnamed biochemical entities, the pathways of which are unknown. We performed classical univariate ROC curve analyses of individual biomarkers to identify putative biomarkers for the discrimination of endometrial cancer from controls (Figure 1). In this analysis, 1-Lignoceroyl GPC (24:0), 1-(1-enyl-stearoyl)-2-linoleoyl-GPE (P-18:0/18:2) and 1-linolenoyl-GPC (18:3) were the most discriminatory biomarkers with AUCs of 0.91 (95%CI 0.86–0.95), 0.85 (95%CI 0.78–0.91) and 0.84 (95% CI 0.78–0.91), respectively. Phosphatidylcholines (PCs) thus feature as potentially important biomarkers. Other discriminatory biomarkers included 3-hydroxylbyryl carnitine and 3-hydroxybutyrate with AUCs of 0.83 and 0.82, respectively (see Figures 1 and 2). Principal component analysis (PCA) and t-distributed stochastic neighbour embedding (t-SNE) were employed and showed some discrimination between cancers and controls (Figure 3a,b). Random forest machine learning was then applied and identified the top 20 discriminatory biomarkers. These were ranked by their contributions to the classification accuracy based on the mean decrease accuracy metric and the mean decrease gini index (Figure 4). A PCA and t-SNE plot based on the top ten discriminatory biomarkers showed a strong degree of separation between cancers and controls (Figure 3c,d). Hierarchical clustering was subsequently performed based on the top 10 discriminatory biomarkers and a heat map was generated (Figure 5). The random forest algorithm was used to split the samples 80:20, 80% for the training set and 20% for testing. The algorithm demonstrated an accuracy of 86.2% (OOB error rate of 13.76%) in the training set, 92.6% prediction accuracy in the testing set and an AUC of 0.95 for endometrial cancer detection (Tables 2 and 3). Biochemical identities, super-pathways and sub-pathways of discriminatory metabolites for EC detection are summarized in Table S2. ROC curves based on the Random Forest diagnostic algorithms are shown in Figure S2.

**Figure 1.** Receiver operating characteristic (ROC) curves of the promising endometrial cancer diagnostic biomarkers from different classes based on the area under the curve (AUC) analyses of *n* = 67 cancers and *n* = 69 controls. The optimal cut-off was based on the closest to the top left corner principle and is indicated by the red dot in the ROC curves. Metabolites starting with X are unnamed; the pathways of these are unknown. GPC—Glycerophosphocholine. GPE—Glycerophosphoethanolamine.

**Figure 2.** Box plot distribution of promising endometrial cancer diagnostic metabolites based on analyses of *n* = 67 cancers and *n* = 69 controls. The black dots along the Y axis in the box plots represent the concentrations of each metabolite while the yellow diamond represents the mean concentration for the group. The notch represents the 95% confidence interval around the median of each group. The horizontal red lines represent the optimal cut-off. Metabolites starting with X are unnamed; the pathways of these are unknown.

**Figure 3.** Analysis of sample separation using the training set (*n* = 109, cancers = 54, controls = 55) based on principal component (PCA) (**a**,**c**) and t-distributed stochastic neighbour embedding (t-SNE) (**b**,**d**) analyses using all identified metabolites (**a**,**b**) and the top 10 discriminatory metabolites (**c**,**d**) identified by random forest machine learning technique. t-SNE (perplexity: 5, iteration: 10,000).

**Figure 4.** Top 20 discriminatory metabolites identified by random forest machine learning technique and ranked by their contribution to classification accuracy using mean decrease accuracy and mean decrease gini index (node impurity) based on the training set (*n* = 109, cancers = 54, controls = 55). Metabolites starting with X are unnamed; the pathways of these are unknown.

**Figure 5.** Hierarchical clustering using the top 10 discriminatory metabolites in the training set (*n* = 109, cancers = 54, controls = 55) based on mean decreasing accuracy. The difference in intensities of the top 10 metabolites by cancer-control status is shown. Each coloured cell in the map represents scaled/relative concentration of indicated metabolite. Metabolites are clustered along the vertical axis while subjects are clustered along the horizontal axis. Hierarchical clustering was based on the Euclidean distance measure and the Ward algorithm.

**Table 2.** Random forest diagnostic accuracy based on the training set made from 80% cases and controls (*n* = 109, cancers = 54, controls = 55).


OOB Error rate: 13.76%. Number of Trees: 1000. Number of variables tried at each split: 33. Sensitivity: 88.9%, specificity: 83.6%.

**Table 3.** Random forest prediction accuracy applied on the testing set made from 20% of cases and controls (*n* = 27, cancers = 13, controls = 14).


OOB Error rate: 7.41%. Prediction accuracy: 92.6%. AUC: 0.95.

#### *3.3. Metabolomic Analysis for the Detection of Early-Stage Endometrial Cancer*

It is important that plasma metabolites used for the identification of endometrial cancer can detect early-stage, not just advanced-stage, disease. We therefore sought to identify metabolites able to distinguish stage 1 endometrial cancer (*n* = 59) from controls (*n* = 69). PCA and t-SNE analyses showed good discrimination between stage 1 disease and controls on all study metabolites (Figure 6a,b) and based on the top 10 metabolites identified using random forest modelling (Figure 6c,d). The top 20 metabolites that distinguished stage 1 endometrial cancer from controls based on random forest algorithm are summarised in Figure 7 and their contribution to the classification accuracy ranked by the mean decrease accuracy and mean decrease gini index. Glycerophospholipids remained important predic-

tors of stage 1 disease, however, the top discriminatory metabolites were uncharacterised chemical entities. Hierarchical clustering using the top 10 metabolites was performed and the generated heat map presented in Figure 8. This showed good discrimination between stage 1 endometrial cancer and controls based on selected metabolites. The study samples were subsequently split 80:20 (80% training set and 20% testing set) using random forest algorithm. The diagnostic algorithm demonstrated an OOB error rate of 14.7% in the training set, a prediction accuracy of 84.6% in the testing set and an AUC of 0.98 for stage 1 endometrial cancer detection (Tables 4 and 5).

**Figure 6.** Analysis of sample separation (comparing early-stage (stage 1) endometrial cancer versus controls (*n* = 102, cancers = 47, controls = 55) based on PCA (**a**,**c**) and t-distributed stochastic neighbour embedding (t-SNE) (**b**,**d**) analyses using all identified metabolites (**a**,**b**) and the top 10 discriminatory metabolites (**c**,**d**) identified by random forest machine learning technique. t-SNE (perplexity: 5, iteration: 10,000).

**Figure 7.** Top 20 discriminatory metabolites for the detection of early-stage endometrial cancer based on the training set (*n* = 102, cancers = 47, controls = 55) identified by random forest machine learning technique and ranked by their contribution to classification accuracy using mean decrease accuracy and mean decrease gini index. Metabolites starting with X are unnamed; the pathways of these are unknown.

**Figure 8.** Hierarchical clustering using the top 10 discriminatory metabolites for the detection of early-stage endometrial cancer in the training set (*n* = 102, cancers = 47, controls = 55) based on mean decreasing accuracy using random forest classification algorithm. The difference in intensities of the top 10 metabolites by cancer-control status is shown. Each coloured cell in the map represents scaled/relative concentration of indicated metabolite. Metabolites are clustered along the vertical axis while subjects are clustered along the horizontal axis. Metabolites starting with X are unnamed; the pathways of these are unknown.

> **Table 4.** Random forest diagnostic accuracy developed based on the training set made from 80% of stage 1 endometrial cancer cases and controls (*n* = 102, cancers = 47, controls = 55).


OOB Error rate: 14.71%. Number of Trees: 1000. Number of variables tried at each split: 22. Sensitivity: 87.2%, specificity: 83.6%.

**Table 5.** Random forest prediction accuracy applied on the testing set made from 20% of stage 1 endometrial cancer cases and controls (*n* = 26, cancers = 12, controls = 14).


OOB Error rate: 15.4%. Prediction accuracy: 84.6%.

#### *3.4. Metabolomic Biomarkers for Predicting Deep Myometrial Invasion and LVSI*

Lymphovascular space invasion (LVSI) and deep myometrial invasion are important endometrial cancer prognostic biomarkers. However, their characterisation in clinical practice is performed by histopathologists with moderate interobserver reproducibility. Metabolites with the potential to predict deep myometrial invasion and LVSI will significantly improve endometrial cancer prognostic characterisation. We therefore sought to identify metabolites that can predict LVSI (*n* = 12) and deep myometrial invasion (*n* = 12) in women with endometrioid endometrial cancer. We limited our analysis to univariate

ROC curve analysis and identified specific glycerophosphoethanolamines, glycerophosphocholines, heme and hydroxybutyrate as important predictors of LVSI with AUCs ranging from 0.75–0.83 (Figure 9). A number of unnamed metabolites were noted to predict deep myometrial invasion in addition to Homovanillate, 3-OH-isobutyrate and Tigloylglycine with AUCs ranging between 0.73 and 0.82 (Figure 10).

**Figure 9.** ROC curves of the promising biomarkers for the prediction of lymphovascular space invasion (*n* = 12) based on AUC analyses of *n* = 67 cancers. The optimal cut-off was based on the closest to the top left corner principle and is indicated by the red dot in the ROC curves. Metabolites starting with X are unnamed; the pathways of these are unknown.

**Figure 10.** ROC curves of the promising biomarkers for the prediction of deep myometrial invasion (*n* = 12) based on AUC analyses of *n* = 67 cancers. The optimal cut-off was based on the closest to the top left corner principle and is indicated by the red dot in the ROC curves. Metabolites starting with X are unnamed; the pathways of these are unknown.

#### *3.5. Consideration of Potential Confounding Factors*

In order to confirm that the discriminatory power of the metabolite signature was due to the presence and absence of endometrial cancer and not confounding variables, we carried out further analyses, taking into consideration the effects of age, BMI, menopausal and diabetic status. First, we performed unsupervised exploratory analyses using score plots generated from PCAs to identify differences between groups (Figure 11). The PCA score plots showed a mild segregation pattern in the confounding factor comparisons suggesting that age, menopausal and diabetic status could potentially have influenced the diagnostic performance within groups of samples (Figure 11). However, these analyses were limited by small numbers within groups. Next, we performed pairwise Spearman's correlation analysis with Bonferroni correction looking at the correlation between age, BMI and selected metabolites (Table 6). There was no evidence of a strong correlation between the metabolite concentrations and age, BMI or parity. Correlation coefficients ranged between 0.25–0.45 for age-based comparisons, 0.33–0.58 for BMI-based comparisons and 0.21–0.32 for parity-based comparisons, suggesting weak correlations between age, BMI, parity and selected metabolite concentrations. While the glycerophospholipids (GPC, GPE) had a positive correlation with age and a negative correlation with BMI/parity, the reverse was the case for the hydroxybutyrates.

**Figure 11.** Score plots generated after unsupervised PCA to visualise differences and similarities according to confounding factors. (**a**,**b**) Score plots according to age (<60 years; ≥60 years) for cancers (**a**) and controls (**b**). (**c**,**d**) Score plots according to menopausal status for cancers (**c**) and controls (**d**). (**e**,**f**) Score plots according to diabetes (present; not present) for cancers (**e**) and controls (**f**).


*Cancers* **2021**, *13*, 718

**Table 6.** Pairwise correlation analysis for selected metabolites

 with age and BMI.

We then applied an exclusion principle by eliminating women with type 2 diabetes mellitus, leaving 50 cancers and 40 controls. There was still a difference between cases and controls by menopausal status. The list of the top-performing metabolites remained largely similar (Figure 12) based on our machine learning (ML) approaches, suggesting that diabetic status did not significantly affect the diagnostic performance of the metabolites. A receiver characteristics curve analysis of these metabolites gave an AUC of 0.94, 0.90 and 0.89 for 1-Lignoceroyl GPC, 1-Steroyl GPC and 1-1 Enyl-Steroyl-2-Linoleoyl-GPE, respectively (Figure 13). The PCA analyses and heat maps also showed good discrimination between cancer cases and controls (Figures 14 and 15), confirming that diabetes status was not a significant confounder in the study analyses, especially with respect to the diagnostic performance of the glycerophospholipids. However, we noted that the hydroxybutyrates and their derivatives were no longer important discriminators of cancers from controls following exclusion of women with type 2 diabetes mellitus (Figure 12), suggesting that their diagnostic ability may be related to their association with diabetes mellitus. The samples of women with no clinical or biochemical evidence of diabetes mellitus were split 80:20 (80% training set and 20% testing set) with the training data used to build a model to separate cancers from controls. The random forest model had an OOB error rate of 11.1% and when tested using the remaining 20% data, it gave a prediction accuracy of 88.9% (Tables 7 and 8).

**Figure 12.** Top 20 discriminatory metabolites for the detection of endometrial cancer following exclusion of women with type 2 diabetes mellitus (training set: *n* = 72, cancers = 40, controls = 32) Metabolites were identified by random forest machine learning technique and ranked by their contribution to classification accuracy using mean decrease accuracy and mean decrease gini index. Metabolites starting with X are unnamed; the pathways of these are unknown.

**Figure 13.** ROC curves of selected metabolites for endometrial cancer detection after exclusion of women with type 2 diabetes mellitus (*n* = 90, cases = 50, controls = 40) based on AUC analysis. The optimal cut-off was based on the closest to the top left corner principle and is indicated by the red dot in the ROC curves.

**Figure 14.** Analysis of sample separation after exclusion of women with type 2 diabetes mellitus (training set: *n* = 72, cancers = 40, controls = 32) based on PCA (**a**,**c**) and t-distributed stochastic neighbour embedding (t-SNE) (**b**,**d**) analyses using all identified metabolites (**a**,**b**) and the top 10 discriminatory metabolites (**c**,**d**) identified by random forest machine learning. t-SNE (perplexity: 5, iteration: 10,000).

**Figure 15.** Hierarchical clustering using the top 10 discriminatory metabolites for the detection of endometrial cancer after exclusion of women with type 2 diabetes mellitus (training set: *n* = 72, cancers = 40, controls = 32). Discriminatory metabolites were based on mean decreasing accuracy metric from random forest analysis. The difference in intensities of the top 10 metabolites by cancer-control status is shown. Each coloured cell in the map represents the scaled/relative concentration of indicated metabolite. Metabolites are clustered along the vertical axis and subjects along the horizontal axis. Metabolites starting with X are unnamed with unknown pathways.

**Table 7.** Random forest diagnostic accuracy developed based on the training set made from 80% of endometrial cancer cases and controls after exclusion of those with type 2 diabetes mellitus (*n* = 72, cancers = 40, controls = 32).


OOB Error rate: 11.11%. Number of Trees: 1000. Number of variables tried at each split: 73. Sensitivity = 95%, Specificity = 81%.

**Table 8.** Random forest prediction accuracy applied on the testing set made from 20% of endometrial cancer cases and controls after exclusion of women with type 2 diabetes mellitus (*n* = 18, cancers = 10, controls = 8).


OOB Error rate 11.11%. Prediction accuracy 88.9%.

Finally, we restricted the analysis to post-menopausal women (*n* = 77, cases = 56, controls = 21). There was still a difference according to diabetes status between cancers and controls in this cohort (*p* = 0.001). The PCA and t-SNE plots showed good discrimination between cancers and controls based on all study metabolites and on the top 10 discriminatory metabolites (Figure 16). The glycerophospholipids remained important predictors of endometrial cancer. The 3-hydroxybutyrate derivatives were also important predictors of endometrial cancer (ranked in the top 10 based on random forest mean decrease accuracy and mean decrease gini index) (Figure 17), confirming their likely association with type 2

diabetes mellitus. Importantly, we noticed the sphingolipids, specifically sphingomyelins, to be well represented in the top 10 discriminatory biomarkers in post-menopausal women (Figure 17). Tricosanoyl and Behenoyl sphingomyelins, in particular, demonstrated AUCs of 0.83 and 0.78, respectively (Figure 18). Hierarchical clustering also showed good discrimination based on the top 10 metabolites in this cohort (Figure 19).

**Figure 16.** Analysis of sample separation for post-menopausal women (*n* = 77, cases = 56, controls = 21) based on PCA (**a**,**c**) and t-distributed stochastic neighbour embedding (t-SNE) (**b**,**d**) analyses using all identified metabolites (**a**,**b**) and the top 10 discriminatory metabolites (**c**,**d**) identified by random forest machine learning. t-SNE (perplexity: 5, iteration: 10,000).

**Figure 17.** Top 20 discriminatory metabolites for the detection of endometrial cancer in post-menopausal women (*n* = 77, cases = 56, controls = 21). Metabolites were identified by random forest machine learning and ranked by their contribution to classification accuracy using mean decrease accuracy metric and mean decrease gini index. Metabolites starting with X are unnamed; the pathways of these are unknown.

**Figure 18.** ROC and box-plot distributions of selected metabolites (sphingomyelins) for endometrial cancer detection in post-menopausal women (*n* = 77, cases = 56, controls = 21) based on AUC analysis. The optimal cut-off was based on the closest to the top left corner principle and is indicated by the red dot in the ROC figures. The black dots in the box plots represent the concentrations of each metabolite, while the red diamond represents the mean concentration for the group. The notch represents 95% confidence interval around the median of each group.

**Figure 19.** Hierarchical clustering using the top 10 discriminatory metabolites for the detection of endometrial cancer in post-menopausal women (*n* = 77, cases = 56, controls = 21). Discriminatory metabolites were based on mean decrease accuracy metric using random forest analysis. The difference in intensities of the top 10 metabolites by cancer-control status is shown. Each coloured cell in the map represents the scaled concentration of indicated metabolite. Metabolites are clustered along the vertical axis while subjects are clustered along the horizontal axis. Metabolites starting with X are unnamed; the pathways of these are unknown.

#### **4. Discussion**

In this study, we evaluated the potential of plasma-based metabolomic biomarkers to detect endometrial cancer in women with class III obesity. Top-performing metabolites, particularly glycerophospholipids and hydroxybutyrates, showed good accuracy for endometrial cancer detection, with AUCs > 0.80. An algorithm combining the ten most discriminatory metabolites was even more successful, with AUCs > 0.90. Potential sources of confounding, particularly age, BMI and diabetes status, did not demonstrate strong correlations with individual metabolites, with the exception of hydroxybutyrates and type

2 diabetes mellitus. These data suggest that a simple blood test could offer a minimally invasive endometrial cancer detection tool for women with class III obesity.

The rising prevalence of endometrial cancer has stimulated an interest in biomarker discovery alongside minimally invasive sampling technologies for its early detection [11]. Many studies have explored the possibility of detecting endometrial cancer in blood using genetic biomarkers (including tumour DNA [26], epigenetic modifications [27] and transcripts [28,29]), proteins [18,30] and metabolites [19,22] through genomic, epigenomic, transcriptomic, proteomic, spectroscopic and metabolomic approaches. The metabolome reflects the functional human phenotype and as such, has enormous potential to deliver clinically relevant biomarkers for endometrial cancer detection [20,31]. Indeed, metabolic reprogramming is a defining hallmark of carcinogenesis [32]. Pertubations in critical pathways involving fatty acid metabolism, choline metabolism, tricarboxylic acid cycle and glycolysis have all been described in the pathogenesis of cancer [21,33,34]. Metabolomic biomarkers have shown promise for the early detection of several cancers, including those of the breast [35], colon [36] and prostate [37], and may be particularly relevant in endometrial cancer, given its strong association with obesity, insulin resistance and type 2 diabetes mellitus [38].

Our finding that glycerophospholipids are important diagnostic biomarkers in endometrial cancer is consistent with published data [39–42]. Glycerophospholipids are the main components of biological membranes and, alongside fatty acids, glycerolipids, sphingolipids and sterols, have been linked to cancer development [43]. The upregulation of phospholipid biosynthetic pathways in cancer cells is a direct consequence of accelerated growth and enhanced membrane biosynthesis that accompanies tumorigenesis [44]. A recent systematic review by our group identified choline derivatives, specifically glycerophosphocholines and phosphocholines, as promising biomarkers for endometrial cancer detection [22]. Altered choline metabolism is a hallmark of carcinogenesis and is linked to mitogenic signal transduction, the regulatory mechanism that modulates cell proliferation, differentiation, metabolism and death [34,45,46]. Up-regulation of choline-containing precursors, including phosphocholines and total choline-containing compounds, is caused by the overexpression and activation of several key enzymes involved in choline metabolism by cancer cells. These processes are mediated by oncogenic signalling pathways, including RAS and PI3K-AKT [46,47]. Trousil and colleagues found that altered choline metabolism in endometrial cancer is caused by an overexpression of choline kinase alpha and hyperactivation of the deacylation pathway [48]. Choline derivatives are detectable in blood, tumour and vaginal fluid in women with endometrial cancer [39–41]. They have also been described in breast, prostate and other solid tumours [46]. 3-hydroxybutyrate and its derivatives have also shown promise for endometrial cancer detection [49,50]. Bahado-Singh found that 3-OH butyrate was an important endometrial cancer biomarker even after adjusting for diabetes [49]. In the current study, 3-OH butyrate and its derivatives did not significantly discriminate between cases and controls after excluding women with type 2 diabetes mellitus. This may relate to the strong association between 3-OH butyrate and diabetes, with multiple studies suggesting that 3-OH butyrate is an early marker of insulin resistance, even in non-diabetic populations [51–53]. 3-OH butyrate has also been identified as a potential biomarker of low-grade female papillary thyroid cancer [54] and high-grade serous carcinoma of the ovary [55]. Knapp and colleagues found sphinganine, sphingosine, dihydroceramide and ceramide levels to be significantly elevated in endometrial cancer tissue compared to healthy endometrium [56]. Audet-Delage and colleagues reported sphingolipids to be significantly elevated in the serum of women with recurrent non-endometrioid endometrial cancer [39]. Sphingolipids are involved in inflammation, proliferation, cell migration and apoptosis [57]. Here, we found tricosanoyl and behenoyl sphingomyelins to be upregulated in the plasma of post-menopausal women with endometrial cancer. Further studies are needed to validate the utility of these biomarkers for endometrial cancer detection.

Metabolomic biomarkers that can identify aggressive endometrial cancer phenotypes are important for directing therapy. Here, several metabolites were shown to have potential for establishing tumour stage, the presence of LVSI and deep myometrial invasion (Figures 9 and 10, respectively). Glycerophosphocholines, glycerophosphoethanolamines, heme and 3-OH butyrate were important predictors of LVSI while X-12847, X-17337, Homovanillate (HVA), X-23644, 3-OH butyrate and Tigloylglycine were important predictors of deep myometrial invasion. These results must be interpreted with caution given the small sample sizes. Heme, an iron-containing porphyrin, is an important source of electrons for electron transfer and has been shown to be elevated in the clinically aggressive type II endometrial cancer [39,58]. Homovanillate, a metabolite of dopamine, is a neurotransmitter originating from tyrosine [59]. We did not find any prior studies identifying HVA as a marker of deep myometrial invasion in endometrial cancer. These markers warrant validation in an independent cohort and their mechanistic links to endometrial cancer should be elucidated prior to clinical translation.

This study has several strengths. Our metabolomics methodology, using multiple approaches for metabolite separation and identification (Reverse Phase Liquid Chromatography and Hydrophilic Interaction Liquid Chromatography), helped maximise the number of metabolites identified. The use of artificial intelligence to select the best-performing metabolites and to qualify their performance in an independent sub-group of samples is a further strength, as this minimises the unwanted inflation of performance that occurs in the absence of independent testing. Identified metabolites showed sufficient accuracy for endometrial cancer detection (including early-stage tumours), especially when combined in a biomarker panel, and thus have good potential for clinical utility. Indeed, many of these metabolites have mechanistic links with the malignant transformation process. The use of obese controls maximises the chance that discriminatory metabolites are cancerspecific rather than obesity-related and sets our study apart from previous studies where apparently healthy controls (i.e., women with normal BMI) were used.

A limitation of our study design is that our metabolite panel may not identify nonendometrioid-/non-obesity-related tumours. It is also unclear how well the biomarkers will perform in other high-risk groups such as the elderly, those with postmenopausal bleeding or Lynch syndrome. The relatively small sample size and the attendant difficulty in controlling for potential confounding factors is another limitation. Several discriminatory metabolites could not be biochemically identified, which limits their clinical implementation.

#### **5. Conclusions**

We found specific plasma metabolites to have potential for the detection of endometrial cancer in a cohort of women with class III obesity. A metabolomic signature based on the top ten performing metabolites showed good promise. Glycerophospholipids, specifically glycerophosphocholines and glycerophosphoethanolamines, were particularly important in differentiating endometrioid endometrial cancer from controls. These findings suggest that a simple blood-based test has the potential to enable the early detection of endometrial cancer and provides a basis for a minimally invasive screening tool for women with class III obesity. Further studies are needed to validate the biomarker candidates and elucidate their role in endometrial carcinogenesis.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2072-669 4/13/4/718/s1, Figure S1: Overview of study workflow, Figure S2: ROC curves based on Random Forest algorithms for the detection of endometrial cancer of all stages (a) and stage 1 endometrial cancer (b) using 80% of study samples and based on the top 10 discriminatory biomarkers, Table S1: Description of liquid chromatographic columns and mode of ionisation used in metabolite extraction based on protocols by Metabolon Inc, Table S2: Biochemical identities, super-pathways and sub-pathways of discriminatory metabolites for EC detection.

**Author Contributions:** K.N. analysed the data, carried out data interpretation and wrote the first draft of the manuscript including creation of figures. E.J.C. and A.D.W. conceptualised the study, supervised study execution, contributed to data interpretation and wrote the manuscript. A.E.C. and B.G. contributed to data analysis and interpretation. M.L.M., A.E.D., S.J.K., V.N.S. and E.J.C. recruited for the study and performed study procedures. A.P. contributed to data interpretation. E.J.C. and A.D.W. are Principal Investigators and obtained funding for the study. All authors provided critical comment. All authors have read and agreed to the published version of the manuscript.

**Funding:** K.N. is supported by Cancer Research UK (CRUK) Manchester Cancer Research Centre Clinical Research Fellowship and the Wellcome Trust Manchester Translational Informatics Training Scheme. S.J.K. and V.N.S. are supported by the National Institute for Health Research (NIHR) Academic Clinical Lectureship. The Stoller Biomarker Discovery Centre was established with an award from the Medical Research Council (MR/M008959/1). This work was supported by the CRUK Manchester Centre award (C5759/A25254) and Bloodcancer UK (Award 19007 to AP and ADW). ADW and EJC are supported by the National Institute for Health Research Manchester Biomedical Research Centre (IS-BRC-1215-20007).

**Institutional Review Board Statement:** PREMIUM study (North West Research Ethics Committee ref. 14/NW/1236), Weight loss study (North West Research Ethics Committee ref. 12/NW/0050), Metformin study (North West Research Ethics Committee ref. 11/NW/0442), and PROTEC study (Cambridge East Research Ethics Committee ref. 15/EE/0063).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** Data are available through the corresponding author upon reasonable request.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Article* **Performance Characteristics of the Ultrasound Strategy during Incidence Screening in the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS)**

**Jatinderpal Kalsi 1, Aleksandra Gentry-Maharaj 2, Andy Ryan 2, Naveena Singh 3, Matthew Burnell 2, Susan Massingham 2, Sophia Apostolidou 2, Aarti Sharma 4, Karin Williamson 5, Mourad Seif 6, Tim Mould 7, Robert Woolas 8, Stephen Dobbs 9, Simon Leeson 10, Lesley Fallowfield 11, Steven J. Skates 12, Mahesh Parmar 2, Stuart Campbell 13, Ian Jacobs 1,14, Alistair McGuire <sup>15</sup> and Usha Menon 2,\***


**Simple Summary:** The United Kingdom Collaborative Trial of Ovarian Cancer Screening was undertaken to assess whether screening postmenopausal women from the general population might result in detection of ovarian/tubal cancers at an earlier stage and thus save lives. One of the screening strategies tested was a yearly transvaginal ultrasound scan of the ovaries (USS). Following the initial screen, 44,799 of the 50,639 women in the USS group went on to have a further 280,534 annual scans during April 2002–December 2011. Abnormalities leading to surgery were detected in 960 women of whom 113 (80 invasive epithelial) had ovarian/tubal cancer. Ovarian/tubal cancer was missed in 52 (50 invasive epithelial) women. Of the screen-detected cancers, 37.5% and missed cancers 6% were early stage(I/II). The number (detection rate 61.5%; 80/130) and advanced stage of the missed invasive cancers suggests that a yearly ultrasound scan may not be suitable for screening average risk women for ovarian cancer.

**Abstract:** Randomised controlled trials of ovarian cancer (OC) screening have not yet demonstrated an impact on disease mortality. Meanwhile, the screening data from clinical trials represents a rich resource to understand the performance of modalities used. We report here on incidence screening in the ultrasound arm of UKCTOCS. 44,799 of the 50,639 women who were randomised to annual screening with transvaginal ultrasound attended annual incidence screening between 28 April 2002

**Citation:** Kalsi, J.; Gentry-Maharaj, A.; Ryan, A.; Singh, N.; Burnell, M.; Massingham, S.; Apostolidou, S.; Sharma, A.; Williamson, K.; Seif, M.; et al. Performance Characteristics of the Ultrasound Strategy during Incidence Screening in the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS). *Cancers* **2021**, *13*, 858. https://doi.org/10.3390/ cancers13040858

Academic Editor: Sara Gutierrez-Enriquez

Received: 5 January 2021 Accepted: 9 February 2021 Published: 18 February 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

and 31 December 2011. Transvaginal ultrasound was used both as the first and the second line test. Participants were followed up through electronic health record linkage and postal questionnaires. Out of 280,534 annual incidence screens, 960 women underwent screen-positive surgery. 113 had ovarian/tubal cancer (80 invasive epithelial). Of the screen-detected invasive epithelial cancers, 37.5% (95% CI: 26.9–49.0) were Stage I/II. An additional 52 (50 invasive epithelial) were diagnosed within one year of their last screen. Of the 50 interval epithelial cancers, 6.0% (95% CI: 1.3–16.5) were Stage I/II. For detection of all ovarian/tubal cancers diagnosed within one year of screen, the sensitivity, specificity, and positive predictive values were 68.5% (95% CI: 60.8–75.5), 99.7% (95% CI: 99.7–99.7), and 11.8% (95% CI: 9.8–14) respectively. When the analysis was restricted to invasive epithelial cancers, sensitivity, specificity and positive predictive values were 61.5% (95% CI: 52.6–69.9); 99.7% (95% CI: 99.7–99.7) and 8.3% (95% CI: 6.7–10.3), with 12 surgeries per screen positive. The low sensitivity coupled with the advanced stage of interval cancers suggests that ultrasound scanning as the first line test might not be suitable for population screening for ovarian cancer. Trial registration: ISRCTN22488978. Registered on 6 April 2000.

**Keywords:** ovarian cancer; screening; ultrasound; TVS; early detection; trial; randomised controlled trial; UKCTOCS

#### **1. Introduction**

Transvaginal ultrasonography (TVS) is considered the best modality for pelvic imaging, and is used routinely in the clinic for investigating women with suspected ovarian cancer. Based on its ability to assess ovarian volume and morphology, it has been used in large randomised trials of ovarian cancer screening as the primary screen. In the ovarian arm of the Prostate Lung Colorectal and Ovarian (PLCO) Cancer Screening trial [1], it was used in combination with the serum biomarker CA125 while in the ultrasound arm (USS) of the United Kingdom Collaborative Trial of Ovarian Cancer Screening (UKC-TOCS), it was used as the sole primary screening test [2]. In both trials, there was no difference in the proportion of women detected with Stage I/II disease or deaths due to ovarian/tubal/peritoneal cancer between the ultrasound arm and the no screening (control) arm [2].

The data collected during these trials provides a rich resource to understand the performance characteristics of TVS in the setting of multicentre, general population screening. We have previously reported on the results of the initial (prevalence) USS screen [3]. We now report on the performance characteristics of USS screening in UKCTOCS during the 10 years of incidence screening.

#### **2. Results**

Following the initial (prevalence) screen, of the 50,639 women randomised to the USS arm 49,610 were eligible for incidence screening. Of them, 1029 were ineligible as both ovaries had been removed (896), death (131), moved away (2). Overall, 44,799 (88.5%) of those randomized to the USS arm underwent incidence screening (Figure 1).

In total the women underwent 280,534 annual incidence screens between 28 April 2002 and 31 December 2011. Of these screens, 257,337 (91.8%) were TVS, 20,707 (7.4%) transabdominal, 2309 (0.8%) both and for nine data on mode were missing. Individual women attended between 1 and 10 incidence screens with the median number per woman being 7 (IQR 5–8). The baseline characteristics of these women have been previously reported [2,3]. Median age of the women at the last annual incidence screen was 67 (IQR 62.6–72.0) years.

Overall, 99.4% (278,851/280,534) of the screens resulted in women being returned to annual screening. Two percent (5497/280,534) of screens involving 4256 (9.5%; 4256/44,799) women resulted in referral for clinical evaluation. Of these women 960 (0.34% of screens; 960/280,534) were screen positive and had surgery (Figure 2 and Table 1). This figure includes one woman with a simple ovarian cyst who underwent surgery against protocol recommendation.

**Figure 2.** Ultrasound screening (USS) algorithm and outcome of incidence screening.


**Table 1.** Results of annual incidence screens performed in USS group.

Data is number (%). \* Denominators for header rows are numbers of annual screens. Denominators for subsequent rows are number who underwent specific screen. † Difference in numbers between those recommended tests and number who underwent test is due to non-compliance. ‡ 123 women were clinically assessed following a level 1 screen. § Seven women went on to have laparotomy as a second procedure.

Of the 960 surgical procedures, 69% (662/960) were laparoscopic or vaginal. 113 (11.8%) women were diagnosed with ovarian/tubal cancers (Table 2). This included 80 (70.8%) invasive epithelial ovarian or tubal (iEOC), 29 (25.7%) borderline (low malignant potential) epithelial ovarian, and 4 (3.5%) non-epithelial ovarian cancers.

**Table 2.** Pathologic findings in screen positive women and those with interval cancers (screen negative).



**Table 2.** *Cont*.


Data are numbers. \* Includes one volunteer who withdrew consent for accessing medical records and two volunteers where the ovaries were not identified due to extensive adhesions arising from a previous hysterectomies. \*\* Cancers of colorectal (3) breast (1), stomach (1), lymphoma (1), carcinoid small bowel (1).

Of the 29 borderline epithelial ovarian cancers, 28 (96.5%) were Stage I/II as were 3 of 4 (75%) non-epithelial ovarian cancers. Of the screen detected iEOC, 37.5% (30/80) were Stage I/II (Table 3). Of the iEOC 80% (64/80) were Type II and 18.8% (15/80) were Type I. Majority (86.7%; 13/15) of Type I were Stage I/II. Of Type II, only 26.6% (17/64) were Stage I/II. The median time from Level 1 annual screen to surgery for screen detected iEOC was 12.6 weeks (IQR 8.7 to 20.5).

**Table 3.** Stage and type of invasive epithelial ovarian and tubal cancers as per WHO 2014 classification.


Date are numbers unless otherwise stated. \* Morphology could not be determined as only peritoneal fluid cytology was undertaken.

Of the 960 women who had screen positive surgery, 831 had benign pathology or normal adnexa (Table 2). In this subgroup, 35 (4.2%) women had a major complication (with significant sequelae) (Table S1).

Median follow up from the end of incidence screening to cancer registration update in 2015 (25 March 2015 England and Wales, 15 April 2015 Northern Ireland) was 3.9 (IQR 3.6–5.0) years. Only 5 of 44,799 (0.01%) women had follow-up of less than 2 years after their last screen. An additional 52 women were diagnosed with ovarian/tubal cancer (screen negative/interval cancer) within 1 year of the last incidence screen scan (Table 2). This included 2 borderline and 50 iEOC. Of the latter, 6% (3/50) were diagnosed at Stage I/II (Table 3).

The sensitivity, specificity, and positive predictive values (PPV) were 68.5% (95% CI: 60.8–75.5), 99.7% (95% CI: 99.7–99.7), and 11.8% (95% CI: 9.814) respectively for all ovarian and tubal cancers with 8.5 operations per case detected during incidence screening. When the analysis was restricted to iEOC, sensitivity, specificity and positive predictive values were 61.5% (95% CI: 52.6–69.9); 99.7% (95% CI: 99.7–99.7); and 8.3% (95% CI: 6.7–10.3) with 12 surgeries per screen positive (Table 4).

**Table 4.** Performance characteristics of incidence USS screening for detection of ovarian and tubal cancers (WHO 2014 classification) within one year of screen.


Data are numbers or % (95% CI) \* excludes non epithelial and borderline epithelial ovarian neoplasms).

Combining incidence and prevalence screening [3] of UKCTOCS, the sensitivity, specificity and positive predictive values were 72.3% (95% CI: 65.9–78.0), 99.5% (99.5–99.5), and 9.1% (95% CI: 7.8–10.5) for all ovarian and tubal cancers with 11.0 operations per case detected. When the analysis was restricted to iEOC, sensitivity, specificity, and positive predictive values were 63.3% (55.4, 70.6), 99.5% (95% CI: 99.5–99.5), and 5.8% (4.78–7) with 17.2 surgeries per screen positive.

#### **3. Discussion**

#### *3.1. Principal Findings*

The performance characteristics of ultrasound screening in the largest ovarian cancer screening trial suggests that USS may not be suitable as a first line test for population screening. While the PPV was significantly higher (11.8% vs. 5.3%; *p* < 0.0001) with fewer operations (8.5 vs. 18.8; *p* < 0.0001) required to detect an ovarian/tubal cancer during incidence screening compared to the prevalence [3], the sensitivity was lower (68.5% versus 84.9%; *p* = 0.02). For invasive epithelial cancers, while over one-third (38%) of the screen detected invasive cancers were early stage, the majority (94.0%) of the interval cancers were advanced (Stage III/IV). The latter, coupled with the low sensitivity (61.5%) resulted in no overall difference (24% USS versus 26% Control; *p* = 0.57) in low volume (Stage I, II, IIIa) invasive epithelial disease between USS and control arm on the previously reported intention to treat analysis [1,2].

#### *3.2. Results in Context*

While TVS is integral to all ovarian cancer screening strategies to date, its use as the primary screening test, as described here, has only been assessed in one other study, the University of Kentucky Ovarian Cancer Screening Trial (UKOCST). The latter study involved a slightly higher risk population with just under one fourth having a family

history of ovarian and over 40% of breast cancer. It is a single arm single-centre prospective study and involved 46,101 women who underwent a mean of 6.5 annual screens [4]. Overall sensitivity for detecting ovarian cancers (85.5% vs. 72%) was higher than in the USS arm of our trial. TVS has a significant subjective component that is likely to be the key contributor to the differences noted. UKOCST involves a single centre, with all scans performed by a small group of highly experienced ultrasonologists. UKCTOCS involved over 200 Level I ultrasonologists [5] (certified sonographers or doctors with experience in gynaecological scanning in the National Health Service) across 13 centres undertaking ~45,000 scans every year. The latter is more akin to a general population screening programme which would require annual scans for millions of women.

The sensitivity in our USS arm was significantly higher than that sensitivity of TVS alone (44.6%; 33/74) noted during four rounds of screening in the PLCO trial [6]. In the latter trial, the overall sensitivity was higher as the annual screen involved CA125 in addition to a scan, with abnormalities in both tests triggering additional investigations (combined strategy).

In comparison to a CA125-based strategy, PPV of ultrasound screening is low. The number of operations per ovarian cancer decreased from 18.8 during prevalence screening in our USS arm to 8.5 during incidence screening. This latter is similar to the 7.4 operations per case reported in the Kentucky study [4]. It is not possible to calculate a comparable estimate in the PLCO trial as a combined strategy was used.

In our trial, 10,000 complex adnexal masses were detected during the annual incidence screen. Through a process of repeat scanning for persistence of lesion and evaluation of ultrasound features by Level 2 expert sonologists, we were able to restrict surgery to just below 1000 of these women. Both the Kentucky and International Ovarian Tumour Analysis (IOTA) groups have over the years developed increasingly sophisticated rules/scoring systems to improve risk stratification of these adnexal masses and encourage conservative management. In the most recent international IOTA5 study of women with adnexal masses, they were able to avoid surgery in one-third on the basis of low risk ultrasound features [7].

A key requirement to impact on the high ovarian cancer mortality is detection of invasive epithelial ovarian/tubal cancer at a sufficiently early stage. A similar proportion of screen detected ovarian cancers were invasive epithelial both in our analysis (71%: 80/113) and in the Kentucky study (75.5%; 71/94). However, only 37.5% (95% CI: 26.9, 49.0) of screen detected invasive epithelial cancers were early stage (I and II) in our trial compared to 51% (45/71; 95% CI: 51.1, 74.5) in the latest report of the Kentucky study [4]. In the latter, this together with increased sensitivity is likely responsible for the significantly higher 5-year disease-specific survival of women with ovarian (including interval) cancers in the screening group (79 ± 4%) compared to unscreened women with clinically detected epithelial ovarian cancer treated at the same centre during the same time period (45 ± 2%).

In comparison to a CA125 based approach [3], an ultrasound-based strategy detects a larger proportion of borderline ovarian cancers. This was similar in the Kentucky study (15.5%; 17/124, 95% CI: 9.3,23.6) and during incidence screening in UKCTOCS (18.8%; 29/165,95% CI: 13.1, 25.6). In our prevalence screen, it was higher (37.7%, 20/53, 95% CI: 24.7, 52.1). The lower incidence with time is likely due to increasing conservative management of less complex asymptomatic adnexal masses.

#### *3.3. Clinical and Research Implications*

The performance characteristics suggest that ultrasound as a first line test is not suitable for population ovarian cancer screening. The subjective nature of TVS, the challenges in identifying normal postmenopausal ovaries [8] that diminish in size with age and the low disease prevalence (1 in 2500) means that detection of disease early requires significant expertise coupled with constant attention to detail. In the course of the trial, we developed an accreditation programme for scanning postmenopausal ovaries [5]. However, our performance characteristics suggest that we were not able to replicate in the Level 1 ultrasonographers, the expertise available at a specialist centre such as Kentucky. The IOTA

group have shown in multicentre studies that the performance of ultrasound prediction models/rules can be maintained in sonographers with varying levels of experience [9]. However, this is in the context of evaluation of adnexal masses, which is equivalent to a Level 2 rather than Level 1 screen during population screening. First line TVS screening of the population is always going to be a challenge given the size of the workforce required. The ideal is a less subjective, automated, and more reproducible test. In cervical screening, this has translated to HPV DNA testing increasingly replacing the older resource intensive and skill dependent cytology in many population-screening programmes.

Incidental adnexal findings are on the rise given the widespread use of ultrasound. The unnecessary surgery rates seen in our and the other ultrasound screening trials are relevant to the clinical management of these asymptomatic masses. Our findings suggest that many with low-risk features can be managed conservatively [10].

#### *3.4. Strengths and Limitations*

The key strengths of our study are the scale of the trial, high compliance with screening, the multicentre setting and detailed screening protocols and automated management algorithms, implemented by a dedicated central team. Completeness of data on screen-negative cancers was ensured by flagging of the trial cohort through cancer, death, and hospital administrative registries as well as postal follow-up of all women. All potential ovarian cancer cases were reviewed by an independent, blinded outcomes review committee.

A key limitation relates to use of self-reported visualisation of postmenopausal ovaries as a quality assurance measure during the trial. A retrospective audit of random, grey scale TVS images showed only moderate agreement for visualisation of normal ovaries between experts and sonographers and between expert reviewers alone [8]. This was despite a robust accreditation programme established within the trial for visualisation of postmenopausal ovaries. This again highlights the subjectivity of ultrasound scanning, use of video recordings of the ultrasound examination would probably have been a betterquality assurance measure. During the 14 years of trial, there have been significant advances in our understanding of the origin and heterogeneity of ovarian cancer. Our scanning protocol focused on evaluation of the ovary. However, we now know that at least half of high-grade serous cancers arise in the fallopian tube [11] making tubal evaluation critical. The Kentucky group has recently described and assessed such a protocol in older normal women and reported a 77% visualisation rate [12]. Furthermore, in the last decade, there has been significant improvement in the resolution of ultrasound machines and their ability to detect subtle changes as a result of advances in ultrasound transducer technology and electronics.

#### **4. Materials and Methods**

#### *4.1. Ethical Approval*

The trial (ISRCTN22488978, ClinicalTrials.gov NCT00058032) was approved by the UK North West Multicentre Research Ethics Committees (North West MREC 00/8/34) with site specific approval from the local regional ethics committees and the Caldicott guardians (data controllers) of the primary care trusts. All participants provided written consent.

#### *4.2. Subjects and Screening Strategy*

The trial design has been described previously [2,3,13]. Briefly, 202,638 postmenopausal women aged 50 to 74, from the general population were recruited through 13 regional trial centres located in NHS Trusts in England, Wales and Northern Ireland, between April 2001 and October 2005. Overall, 1.6% of women had a maternal history of ovarian cancer and 6.3% a maternal history of breast cancer [3]. Women at increased risk of familial ovarian cancer were excluded from the study. The participants were randomised 1:1:2 to annual screening (until 31 December 2011) with serum CA125 (MMS: 50, 640) or TVS (USS: 50, 639) or no screening (control C: 101, 359). The full trial protocol is accessible at http://ukctocs.mrcctu.ucl.ac.uk/media/1066/ukctocs-protocol\_v90\_19feb2020.pdf (accessed on 4 February 2021). In the USS arm, 48,230 women underwent an initial (prevalence) screen [3].

Scans were performed by trial sonographers, the majority of whom worked in the NHS providing gynaecological scanning. All trial sonographers underwent additional training for assessment of postmenopausal ovaries and from 2008, formal accreditation [5]. Annual (Level 1) scans were performed by Type 1 (certified sonographers, trained midwives, or doctors with experience in gynaecological scanning) or Type 2 (experienced gynaecologists/radiologists, or senior sonographers, usually superintendent grade with particular expertise in gynaecological scanning) ultrasonographers. Repeat scans on detection of an abnormality (Level 2 scans) were only undertaken by Type 2 sonographers. Most scans during 2002–2008 were done on a dedicated Kretz SA9900 ultrasound machine (Medison, Seoul, Korea) and from 2008–2011 on Acquvix (Medison, Seoul, Korea).

At the annual transvaginal scan (Level 1), ovarian morphology and dimensions were assessed, and ovarian volume calculated. Ovarian morphology was classified as normal, simple cyst (single, thin walled, anechoic cyst with no septa or papillary projections) or complex (ovary had any non-uniform ovarian echogenicity excluding single simple or inclusion cyst). The number and size of cysts, wall regularity, presence and thickness of septae, size of papillations, and echogenicity of the fluid contents were recorded. The cysts were initially classified using the Kentucky screening trial morphology index [14] and from 2003, the International Ovarian Tumour Analysis (IOTA) classification [15]. Where an ovary was not visualised, the sonographer documented 'good view' if 3–5 cms of iliac vessels with well-defined walls and a clear anechoic centre was seen or 'poor view' and stated the reason such as bowel, fibroids, pelvic varicosities, or other. Ascites was defined as a vertical pool of fluid measuring >10 mm in the Pouch of Douglas.

Ultrasound scans were classified based on the morphology of the adnexa and visualisation of the surrounding tissue as follows: (a) normal—where both ovaries had normal morphology or simple cysts were <60 cm3, or were not visualised but a good view of the iliac vessels was obtained; (b) unsatisfactory—where one or both ovaries were not visualised due to a poor view); (c) abnormal—where one or both ovaries had complex morphology or simple cysts were >60 cm3, or ascites was present. Based on these results the women were returned to annual screening (normal scan), repeat Level 1 scan (unsatisfactory scan) or Level 2 scan (abnormal scan). In women where adnexal masses had been previously managed conservatively and remained unchanged in morphology or volume (complex unchanged) on repeat annual screens, there was the option for clinical review of results and return to annual screening without undergoing Level 2. Women with an abnormal Level 2 scan were referred for clinical assessment.

This was undertaken at the regional centre by a designated trial clinician and included clinical evaluation and investigations as appropriate. Latter included serum CA125, repeat transvaginal scans and Doppler studies, CT/MRI of the abdomen and pelvis, and occasionally assessment of other tumour markers. A decision was made either to offer surgery or manage conservatively, taking into account the views of the woman, any significant comorbidity, morphological features of the ultrasound-detected lesion, previous hysterectomy, or major pelvic surgery that could contribute to false-positive ultrasound findings. The surgery in most cases involved removal of both ovaries and fallopian tubes using a laparoscopic approach where possible. If pelvic adhesions increased the risk of complications, the clinician could opt to remove only the 'abnormal' ovary. Hysterectomy was only undertaken where there was clear clinical indication. Women found to have ovarian or tubal cancer at a primary laparoscopic procedure underwent a subsequent staging procedure. Where there was high suspicion of ovarian cancer, laparotomy was undertaken. For those managed conservatively, the follow up plan usually involved a TVS and serum CA125 at 3 months with a possible repeat at 6 months, and return to annual screening if the findings were unchanged (unchanged complex).

#### *4.3. Follow-Up*

Follow up involved electronic health record linkage for cancer and death registration and hospital admissions using the NHS number through the appropriate national agencies. Cancer registrations received until 25 March 2015 (England and Wales) and 15 April 2015 (Northern Ireland) were used for this analysis. In addition, women were sent postal questionnaires, 3–5 years post randomisation, and again in April 2014 after the end of screening [2].

#### *4.4. Confirmation of Diagnosis*

Copies of medical notes were retrieved for all women who had surgery as a consequence of a positive screening test as previously described [2]. Where cancer was diagnosed, additional information e.g., multidisciplinary team meeting notes, discharge summaries, and other relevant correspondence was also collated. The above were also obtained for all women where a notification was received either through linked electronic health records, follow-up questionnaire, or personal communication of a possible ovarian/tubal/peritoneal cancer. The case notes of all of these individuals were reviewed by an Outcomes Review Committee blinded to the randomisation group. They confirmed primary site, stage, morphology, and—where possible—classified invasive epithelial cancer into Type I (low-grade serous, low-grade endometrioid, mucinous, and clear cell cancers) or Type II (high-grade serous, high-grade endometrioid, carcinosarcomas and undifferentiated carcinoma) cancers [16]. Primary site was originally classified according to WHO 2003 [17] and more recently revised using WHO 2014 classification [18]. As a result, cancers initially classified as peritoneal have been reclassified for this analysis as ovarian/tubal. Stage for all cases included in this analysis have been re-reviewed by the Outcomes Committee and assigned as per FIGO 2014 criteria [19].

#### *4.5. Analysis*

This analysis is limited to annual screens that followed the initial (prevalence screen). An annual screen as previously defined is a single or series of scans culminating in surgery (screen positive) or return to annual screening (screen negative). For this analysis, women were censored at one year following the last scan performed as part of their last screening episode on the trial. The primary outcome measure was ovarian or fallopian tube cancer as per WHO 2014 classification [18] diagnosed within 12 months of the last scan. Sensitivity (proportion of ovarian/tubal cancers diagnosed within one year that were detected by screening), specificity (proportion of those without ovarian/tubal cancer who had a negative screen) and positive predictive value (proportion with a positive test result who actually had ovarian/tubal cancer) of incidence screening was calculated. Subgroup analysis of invasive epithelial cancers (borderline epithelial and non-epithelial ovarian cancers were excluded) was undertaken. Proportion of cancers detected in early (I/II) stage were calculated.

#### **5. Conclusions**

The performance characteristics suggest that ultrasound as the first line test may not be suitable for population screening.

**Supplementary Materials:** The following table is available online at https://www.mdpi.com/2072 -6694/13/4/858/s1, Table S1: Surgical complications in women with benign adnexal masses.

**Author Contributions:** Conceptualisation—I.J., U.M., S.C., L.F., S.J.S., M.P., and A.M.; Data curation— A.R. and U.M., Formal analysis—A.R., U.M., J.K., A.G.-M., and M.B.; Funding acquisition—I.J., U.M., M.P., L.F., S.J.S., S.C., and A.M.; Investigation—U.M., A.G.-M., A.R., A.S., K.W., M.S., T.M., R.W., S.D., S.L.; J.K., N.S., S.M., and S.A., Methodology—U.M., I.J., S.C., A.G.-M., A.R., and N.S., Project administration—U.M., J.K., A.G.-M., A.R., K.W., M.S., T.M., R.W., S.D., and S.L.; Resources—U.M., I.J., K.W., M.S., T.M., R.W., S.D., and S.L.; Software—U.M. and A.R.; Supervision—U.M., S.C., and I.J.; Validation—U.M., A.G.-M., A.R., S.M., A.S., and S.C.; Visualisation—A.R.; Writing—original draft

J.K., A.G.-M., A.R., and U.M.; Writing—review and editing—all. All authors have read and agreed to the published version of the manuscript.

**Funding:** The current analysis is supported by National Institute for Health Research (NIHR) HTA grant (16/46/01) and The Eve Appeal. UKCTOCS was funded by Medical Research Council (G9901012 and G0801228), Cancer Research UK (C1479/A2884), and the Department of Health, with additional support from The Eve Appeal. Researchers at UCL are supported by the NIHR University College London Hospitals (UCLH) Biomedical Research Centre and MRC CTU at UCL core funding (MR\_UU\_12023).

**Institutional Review Board Statement:** The trial was conducted according to the guidelines of the Declaration of Helsinki, and Good Clinical Practice. The trial was approved by the UK North West Multicentre Research Ethics Committees (North West MREC 00/8/34) on 21 June 2000 with site-specific approval from the local regional ethics committees and the Caldicott guardians (data controllers) of the primary care trusts. The current trial protocol is located at http://ukctocs.mrcctu. ucl.ac.uk/ukctocs/documents/ (accessed date 1 Feberuray 2021).

**Informed Consent Statement:** All trial participants provided written informed consent.

**Data Availability Statement:** The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

**Acknowledgments:** The authors thank the volunteers without whom the trial would not have been possible and everyone involved in conduct and oversight, especially the Ultrasound Sub-Committee (http://ukctocs.mrcctu.ucl.ac.uk/ukctocs/committees/ (accessed date 1 Feberuray 2021)).

**Conflicts of Interest:** U.M. has stock ownership and has received research funding from Abcodia. I.J.J. reports personal fees from and stock ownership in Abcodia as the non-executive director and consultant. He has a patent for the Risk of Ovarian Cancer algorithm and an institutional license to Abcodia with royalty agreement. He is a trustee (2012–2014) and Emeritus Trustee (2015 to present) for The Eve Appeal. SJS reports personal fees from the LUNGevity Foundation and SISCAPA Assay Technologies as a member of their Scientific Advisory Boards, Abcodia as a consultant, and AstraZeneca as a speaker honorarium. He has a patent for the Risk of Ovarian Cancer algorithm and an institutional license to Abcodia. All other authors declare no competing interests.

**Disclaimer:** Views expressed in this publication are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care.

#### **References**


## *Article* **Tumor Signature Analysis Implicates Hereditary Cancer Genes in Endometrial Cancer Development**

**Olga Kondrashova 1,†, Jannah Shamsani 1,†, Tracy A. O'Mara 1, Felicity Newell 1, Amy E. McCart Reed 2, Sunil R. Lakhani 2,3, Judy Kirk 4,5, John V. Pearson 1, Nicola Waddell 1,† and Amanda B. Spurdle 1,\*,†**


**Simple Summary:** Women with a family history of cancer are at increased risk of cancer, including endometrial cancer (affecting the womb lining). In some of the women with such family history, the risk can be explained by deleterious changes in mismatch repair genes that cause Lynch syndrome. This study explored the role of other genes in risk of endometrial cancer, using several approaches. The number and type of changes in gene sequence information in women with endometrial cancer was compared to that from individuals in the general population. Gene sequence changes in endometrial cancer patients with a family history of cancer were also analyzed. Lastly, endometrial cancers from individuals with gene changes were examined for distinctive genomic patterns expected to be seen if a gene change was driving the cancer. This study has identified several additional genes for further exploration in relation to endometrial cancer risk and therapy.

**Abstract:** Risk of endometrial cancer (EC) is increased ~2-fold for women with a family history of cancer, partly due to inherited pathogenic variants in mismatch repair (MMR) genes. We explored the role of additional genes as explanation for familial EC presentation by investigating germline and EC tumor sequence data from The Cancer Genome Atlas (*n* = 539; 308 European ancestry), and germline data from 33 suspected familial European ancestry EC patients demonstrating immunohistochemistry-detected tumor MMR proficiency. Germline variants in MMR and 26 other known/candidate EC risk genes were annotated for pathogenicity in the two EC datasets, and also for European ancestry individuals from gnomAD as a population reference set (*n* = 59,095). Ancestry-matched case–control comparisons of germline variant frequency and/or sequence data from suspected familial EC cases highlighted *ATM*, *PALB2*, *RAD51C*, *MUTYH* and *NBN* as candidates for large-scale risk association studies. Tumor mutational signature analysis identified a microsatellite-high signature for all cases with a germline pathogenic MMR gene variant. Signature analysis also indicated that germline loss-of-function variants in homologous recombination (*BRCA1*, *PALB2*, *RAD51C*) or base excision (*NTHL1*, *MUTYH*) repair genes can contribute to EC development in some individuals with germline variants in these genes. These findings have implications for expanded therapeutic options for EC cases.

**Keywords:** endometrial cancer; genomic sequencing; tumor mutational signatures; hereditary cancer genes; mismatch repair; familial cancer

**Citation:** Kondrashova, O.; Shamsani, J.; O'Mara, T.A.; Newell, F.; Reed, A.E.M.; Lakhani, S.R.; Kirk, J.; Pearson, J.V.; Waddell, N.; Spurdle, A.B. Tumor Signature Analysis Implicates Hereditary Cancer Genes in Endometrial Cancer Development. *Cancers* **2021**, *13*, 1762. https:// doi.org/10.3390/cancers13081762

Academic Editor: David Wong

Received: 1 February 2021 Accepted: 29 March 2021 Published: 7 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

#### **1. Introduction**

Endometrial cancer (EC) is the most commonly diagnosed gynecological malignancy, with an increased prevalence rate in developed countries [1]. Modifiable factors such as obesity, lifestyle, and hormone levels are associated with increased risk of EC, and women with a family history of EC or other cancers, such as colorectal, are at ~2–3 fold increased risk of EC [2]. The genetic factors identified to date are either common low-risk cancer predisposition variants that act together to cause polygenic disease, or rare highrisk pathogenic variants in cancer syndrome genes generally present in patients with a monogenic disease phenotype [3].

The major known monogenic form of EC is Lynch syndrome, caused by germline pathogenic variants impacting the mismatch repair (MMR) genes *MLH1*, *MSH2*, *MSH6*, *PMS2*, as well as *EPCAM* deletions, which impact *MSH2* expression. Lynch syndrome accounts for approximately 3–5% of EC at the population level and an increased proportion in cases with family history of colorectal, endometrial and other cancers [4]. The lifetime cumulative risk of EC for women with Lynch syndrome is 40–70%, depending on which MMR gene is disrupted [5]. EC is also a spectrum cancer of Cowden syndrome, caused by the inheritance of pathogenic *PTEN* variants. The cumulative risk of EC for women up to 60 years of age with Cowden syndrome is around 20% [6]. Studies to date suggest that *PTEN* pathogenic variants are very rarely detected in the general population, and mostly in the context of clinical features of Cowden syndrome [7].

Results from a recent study assessing risk associated with reported family history of endometrial and other cancers, after considering proband MMR proficiency and MMR germline test results, indicate that the genetic basis for a substantial fraction of familial EC patients with MMR deficient and MMR proficient tumors remains unexplained [8].

Several genes involved in other hereditary cancer syndromes have been either directly or indirectly implicated in hereditary EC, but with insufficient or conflicting support that germline DNA gene testing would provide clinically useful information for genetic counseling [4]. These include established hereditary cancer syndrome genes, such as *POLE*, *POLD1*, *MUTYH, STK11, TP53, BRCA1* and *BRCA2* [9–21]. Additionally, germline alterations in a number of other known or candidate cancer risk genes have been identified in EC patients from clinical or research studies, including homologous recombination (HR) DNA repair pathway genes (reviewed in [4]). However, because of the paucity of studies focusing on EC and limitations due to study design, there is uncertainty regarding EC risk associated with variants in these genes [4,22].

To explore which genes may influence the EC risk beyond the well-recognized MMR genes, we assessed the frequency of pathogenic variants in a total of 30 known or candidate EC risk genes in publicly available EC and population data. To assist with the interpretation of the EC driver status of pathogenic variants, we performed tumor mutational signature analysis. We also sequenced and analyzed the germline exomes or whole genomes of 33 EC cases with reported family history of endometrial and other cancer types with no evidence of tumor MMR deficiency.

#### **2. Materials and Methods**

#### *2.1. Study Participants and Data Resources*

EC cases unselected for family history were accessed from The Cancer Genome Atlas Uterine Corpus Endometrial Carcinoma study (TCGA-UCEC; *n* = 539). Germline and tumor whole exome sequencing data was used. To align with the most recent NIH genomic data sharing policy, TCGA IDs have been de-identified. For case–control variant frequency comparison, the analysis was limited to individuals of European ancestry (*n* = 308; Table S1). Ancestry was determined from SNP arrays and classified as European or Non-European [23]. Where SNP-determined ancestry was not available, cases were selected by self-reported race.

GnomAD r2.1.1 database was used as a control population (*n* = 15,708 genomes and *n* = 125,748 exomes). To overcome issues around population stratification for case–control comparison, we limited our analysis to individuals of European ancestry (gnomAD—Non-Finnish Europeans; *n* = 95,095).

Suspected familial EC cases were selected from the Australian National Endometrial Cancer Study (ANECS), a population-based study of epidemiological and genetic risk factors for EC. Details of the ANECS study design, including recruitment and data collection, are described in detail in previous publications [8,24,25]. Cases were selected for this study if they met all of the following criteria: the case provided detailed cancer report information in first, second and selected third degree relatives by structured questionnaire and follow-up interview [8]; the case (or for one individual—endometrial cancer affected sister) had previously demonstrated tumor MMR proficiency using immunohistochemistry [24,25]; the case had reported at least one affected relative with a cancer diagnosis (excluding skin cancer due to the significant role of environmental factors in Australia, and excluding EC after a breast cancer diagnosis due to possible confounding by tamoxifen exposure); and there was a germline DNA sample (extracted from whole blood) available for analysis. Germline sequencing was undertaken for 33 unrelated EC cases. The clinical features of the cohort are summarized in Table S2. Participants self-reported British/Irish heritage, and/or were confirmed to have European heritage based on genetic markers.

#### *2.2. Sequencing for Suspected Familial EC Cases*

Genomic DNA was extracted from blood using a salting out method. DNA samples from 6 cases were sequenced using whole exome sequencing and 27 samples were sequenced using whole genome sequencing. Exome libraries were prepared using the Nextera Rapid Capture Exome Kit (Illumina) and sequencing was performed on the NextSeq500 (Illumina) using 2 × 150 bp reads with an average read depth of 75× (Table S3). Whole genome sequencing was performed using HiSeq X Ten (Illumina) with an average read depth of 36× (Table S3). Tumor DNA of one ANECS EC patient (case 28) carrying a germline *MUTYH* variant was extracted from Formalin-Fixed Paraffin-Embedded (FFPE) tissue using Qiagen DNeasy Blood and Tissue kit (Qiagen, Hilden, Germany). Tumor DNA whole genome sequencing was performed using HiSeq X Ten (Illumina, San Diego, CA, USA) to an average read depth of 12×.

#### *2.3. Sequence Analysis*

TCGA-UCEC sequencing data were downloaded as aligned reads (BAM format) and converted to FASTQ format for processing.

Sequencing reads were trimmed using Cutadapt (version 1.9) [26] and aligned to the reference genome (GRCh37) with BWA-MEM (version 0.7.13) [27]. Duplicate aligned reads were marked with Picard (version 1.141) (http://picard.sourceforge.net accessed on 17 November 2015) and sorted using samtools (version 1.3) [28]. Somatic and germline variants were identified by a dual calling strategy using qSNP [29] and GATK Haplotype caller [30], as previously described [31]. For the FFPE tumor sample (case 28), single nucleotide variants (SNVs) were annotated to identify overlapping reads to prevent overcalling due to DNA fragmentation from formalin fixation. SNVs with at least 5 alternate bases after removal of overlapping reads and those absent in dbSNP were kept for signature analysis.

Germline variants were annotated using the Ensembl Variant Effect Predictor (VEP) [32], with population allele frequency based on the Exome Aggregation Consortium (ExACnonTCGA v3). The *in silico* predictions were annotated using VEP-plugins: REVEL [33] and MaxEntScan [34]. Variants were also annotated for variant pathogenicity as submitted to ClinVar [35], if present in this database.

#### *2.4. Variant Prioritization*

Analysis was focused on rare germline variants (minor allele frequency (MAF) of less than 1% in any population in the ExAC-nonTCGA) in 30 genes of interest [4], including the four MMR genes and *EPCAM* (Table S4). In this study, we excluded from analysis any variants in exons 9 and 11–15 of *PMS2*, due to homology with the *PMS2L* pseudogene in these regions [36]. For *POLE* and *POLD1* genes, only missense variants were considered [37].

For the gnomAD and TCGA-UCEC dataset analysis, only pathogenic or likely pathogenic ClinVar variants or predicted truncating variants (termed as likely pathogenic in this study) were considered (Figure S1). The proportion of pathogenic/likely pathogenic carriers in TCGA and gnomAD datasets was calculated by dividing the number of observed pathogenic/likely pathogenic variants by the total number of individuals sequenced for that gene. For the gnomAD dataset, the number of individuals sequenced was calculated by halving the highest allele number for each gene.

For the familial EC dataset, variants present in three or more samples were excluded as common variants. The remaining variants were reviewed and included if they were: (i) predicted truncating variants (nonsense, frameshift indels, and splice donor or acceptor); (ii) predicted to be deleterious by *in silico* predictions using REVEL (cutoff of ≥ 0.5) or PROVEAN (cutoff of ≤ −2.5) [38]; (iii) predicted to disrupt native donor/acceptor site or create a de novo donor splice site (including synonymous) [34]; or (iv) annotated as pathogenic, likely pathogenic or uncertain significance (VUS), with supporting evidence provided, by multiple submitters in ClinVar database. All candidate variants identified in the familial EC samples were manually reviewed in the Integrated Genome Viewer (IGV) to eliminate any artefacts. Validation of the three prioritized variants was performed by Sanger sequencing.

#### *2.5. Mutation Signature Analysis*

At least 100 somatic SNVs per sample were required for signature analysis. SNV mutational signature analysis was performed using deconstructSigs and the COSMIC v2 signature catalogue with the minimum signature contribution set to 15% [39]. Default settings were used for the familial EC case 28 (whole-genome sequencing) and the exome settings for the TCGA-UCEC cohort. *De novo* signature analysis was previously performed using SigProfiler [40].

TCGA-UCEC data were assessed for tumor mutation burden (TMB), microsatellite instability (MSI) status, tumor enrichment of the germline variant in question and additional somatic variations in same gene for *POLE* and MMR genes. TMB was calculated as a number of all somatic mutations divided by the coverage (Mb) of capture kit used (hg18 Nimblegen v2—26.2 Mb, SureSelect All Exon—44 Mb, Nimblegen SeqCap EZ v2.0— 36.5 Mb and Nimblegen SeqCap EZ v3.0—64 Mb). The level of MSI was assessed using MSIsensor (v0.2) on tumor-normal pairs [41]. The analysis was limited to the capturecovered regions. Samples with MSI scores ≥ 3.5 were classified as MSI-high. Germline variants were considered enriched in tumor if the percentage of sequence reads containing a variant was ≥60% in the tumor sample.

*POLE* somatic mutation status for TCGA-UCEC samples was determined by checking for somatic missense *POLE* mutations in exons 9–14. MMR gene somatic mutation status for TCGA-UCEC samples was assessed using the same approach as for the germline variants. *MLH1* gene methylation and *MSH2* gene deletion (copy number-based) information for TCGA-UCEC (Firehose legacy) study [42] was downloaded from cBioPortal [43,44]. *MLH1* was classified as methylated if the beta-value was >0.3.

#### *2.6. Code and Data Availability*

Scripts used for TCGA and gnomAD data analysis are available on https://github. com/okon/EC\_TCGA\_vs\_gnomAD. TCGA-UCEC data were downloaded from GDC data portal in October 2016. GnomAD variant files (r.2.1.1) were downloaded from the gnomAD portal in April 2019. ANECS sequencing data are available upon reasonable request and subject to ethics approval.

#### **3. Results**

#### *3.1. Germline Variants in Data from Publicly Available EC Cases*

We compared the frequency of germline variants between EC cases unselected for family history (TCGA-UCEC study) and the general population (gnomAD database) in a subset of 30 genes, previously highlighted as known or purported to be associated with risk of developing EC (Table S4) [4]. Pathogenic or likely pathogenic variants were selected based on ClinVar reports or predicted protein truncating effect, as outlined in Figure S1. We did not perform formal statistical comparisons because the EC cohort size (*n* = 308) was underpowered to detect significant differences for the expected rare observations, even for MMR genes.

A total of 19 distinct germline pathogenic or likely pathogenic variants were detected in 12 of 30 analyzed risk genes in 25 of 308 TCGA-UCEC cases (Table 1 and Table S5), similar to previous analyses [4,45]. The carrier frequency in the EC cases compared to the gnomAD population was more than double for three of the known MMR genes—*MSH6* (1.3% vs. 0.23%), *MSH2* (0.65% vs. 0.02%) and *PMS2* (0.32% vs. 0.13%), as well as for the HR repair genes *RAD51C* (0.97% vs. 0.1%), *PALB2* (0.32% vs. 0.14%) and *NBN* (0.32% vs. 0.15%). Pathogenic or likely pathogenic variants observed for other candidate EC risk genes occurred at less than 2-fold increased frequency or were found with a lower frequency in cases versus controls, namely: *BRCA1* (0.32% vs. 0.24%), *NTHL1* (0.65% vs. 0.45%), *FAN1* (0.32% vs. 0.31%), *SEC23B* (0.32% vs. 0.33%), *MUTYH* (1.62% vs. 1.73%) and *CHEK2* (0.97% vs. 1.86%).

#### *3.2. Role of Germline Variants in Driving EC Development in TCGA-UCEC Cases*

We explored the potential role of germline variants in known and candidate EC risk genes in cancer development by analyzing tumor sequencing data for evidence of tumor variant enrichment and presence of mutational signatures reflective of defective DNA repair pathways (e.g., HR pathway). We assessed 46 TCGA-UCEC cases, unselected by ancestry, with pathogenic or likely pathogenic germline variants (*n* = 31 distinct variants) in the 30 prioritized genes (Table S5).

Three of the eight cases with pathogenic or likely pathogenic germline variants in MMR genes had evidence of variant enrichment in tumor (one *MSH2* and two *MSH6* variants with >60% variant reads in the tumor sample; Figure 1). In three cases with *MSH2* or *MSH6* variants (one with germline variant enrichment in tumor), we detected a second somatic hit in the respective genes (Figure 1). While we did not observe tumor variant enrichment or second hits for the other three MMR-positive cases, all eight cases had high TMB (>10 Mut/Mb) indicative of MMR deficiency and MSI detected by MSIsensor. We also observed MMR-associated mutational signatures in all eight cases by *de novo* signature analysis (over 25% contribution; eight out of eight cases; Figure S2), and also by signature assignment to the 30 known COSMIC v2 signatures for two of the eight cases, further supporting the tumor driver role of MMR variants in these cases (Figure 1).

Nine cases with germline variants in HR-related genes *PALB2*, *BRCA1*, *RAD51C*, *FAN1* and *CHEK2* also showed evidence of enrichment of the germline variant in the tumor, while the other 12 cases with HR-related gene variants (seven *FAN1*, three *CHEK2*, one *BRIP1,* one *NBN*) did not (Figure 1). Using mutational signature assignment analysis, Signature 3 associated with HR deficiency, was detected in six of seven of tumors with *BRCA1, PALB2* and *RAD51C* variants. We did not observe Signature 3 in the other cases with germline alterations in HR-related genes, suggesting that they were HR pathway proficient.

One of two cases that harbored germline inactivating *NTHL1* variant (p.Gln90\*) had evidence of tumor variant enrichment (Figure 1). This case showed high TMB and presence of Signature 30, characterized by the prevalence of C>T mutations and associated with deficiency in base excision repair expected due to *NTHL1* inactivation [46]. However, this case also showed high MSI and *MLH1* methylation. Finally, no cases with the germline pathogenic *MUTYH* variant (c.1187G>A, p.Gly396Asp) showed evidence of variant enrichment in the tumor nor presence of Signature 18, associated with *MUTYH* inactivation. Of note, while three cases with *MUTYH* variants had high TMB, we attributed it to MMR deficiency in the tumor due to *MLH1* methylation or deletion of *MSH2*, supported by high MSI levels and MMR-deficient mutational signatures.

**Table 1.** Overall frequency of pathogenic and likely pathogenic variants in 30 known and candidate endometrial cancer (EC) risk genes in an EC sample set (TCGA-UCEC study) and the general population (gnomAD).


Only cases with Non-Finnish European ethnicity were included. Genes highlighted in bold had a frequency of >2 times higher in TCGA-UCEC compared with gnomAD. Carrier frequency represents the sum of all (likely) pathogenic variants in that gene. Genes highlighted in bold had more than double variant carrier frequency in the EC cases compared to the gnomAD population.

#### *3.3. Germline Variants in Suspected Familial EC Cases*

To further explore which genes may explain the etiology of familial EC beyond the well-recognized MMR genes, we sequenced the germline exomes or whole genomes of 33 familial EC cases with no evidence of tumor MMR deficiency, and reported family history of endometrial or other cancer types. The analysis was focused on the same 30 genes as in the sections above (Table S4). Out of the 33 cases, we identified three cases with candidate variants in the prioritized genes. These were a frameshift deletion in *PALB2*:c.3116delA (p.Asn1039Ilefs), an in-frame deletion in *ATM*:c.7638\_7646del (p.Arg2547\_Ser2549del) and a missense pathogenic variant *MUTYH*:c.536A>G (p.Tyr179Cys).

**Figure 1.** Somatic mutational signature analysis of the germline variant carriers in the TCGA-UCEC cohort. Tumor mutation burden (TMB), microsatellite instability (MSI) scores and mutational signatures observed in the TCGA-UCEC cases with pathogenic or likely pathogenic variants in DNA damage repair genes associated with specific mutational signatures.

> The patient (case 2) carrying the pathogenic *PALB2* frameshift variant (c.3116delA, p.Asn1039Ilefs) was diagnosed with stage 1 endometrioid EC at age 70 years. She self-reported that 17 family members had been diagnosed with various types of cancer (Figure 2A), including two with EC—diagnosed at age 60 years (mother) and age 35 years (maternal aunt). Although DNA was not available from the EC-affected mother, the pedigree analysis indicates she is an obligate carrier; genotyping of three other relatives identified two carrying the *PALB2* variant, specifically a sister with colon cancer and maternal cousin with breast cancer (Figure 2A).

> The in-frame deletion *ATM* variant (c.7638\_7646del, p.Arg2547\_Ser2549del) was predicted to be deleterious by PROVEAN and was classified as pathogenic for Ataxiatelangiectasia syndrome by multiple ClinVar submitters. The carrier (case 1) of this variant was diagnosed with stage 1 endometrioid EC at age 77 years. Two of the family members were also diagnosed with EC: mother at age 55 years and sister at age 54 years (Figure 2B). Other family members were affected with colorectal cancer at age 54 years (nephew) and cervical cancer at age 27 years (niece). DNA from relatives was not available for testing.

> The missense heterozygous *MUTYH* variant (c.536A>G, p.Tyr179Cys) was identified in a female affected with grade 2 endometrioid EC at age 62 years (case 28). This *MU-TYH* variant is a known common pathogenic missense variant known to cause MUTYHassociated polyposis (MAP) in Western populations when detected in homozygous or compound heterozygous state [47]. The proband reported seven family members affected with various cancers (Figure 2C), including a father diagnosed with melanoma, three relatives with breast cancer (maternal great aunt, paternal aunt, sister), two relatives affected with colorectal cancer (maternal grandfather, sister), and a maternal uncle with prostate cancer. A DNA sample was only available for the female sibling with breast cancer and we identified her to be a non-carrier of the *MUTYH* variant. Although no *MUTYH*-related cancers were reported for the parents of the proband, her maternal grandfather and female

sibling were both affected with colorectal cancer at relatively young age, age 52 years and 39 years, respectively.

**Figure 2.** Pedigrees of families of the endometrial cancer cases carrying candidate variants. (**A**) Family pedigree of *PALB2* p.Asn1039Ilefs carrier. (**B**) Family pedigree of *ATM* p.Arg2547\_Ser2549del carrier. (**C**) Family pedigree of the endometrial cancer case carrying candidate *MUTYH* p.Tyr179Cys. Squares symbolize males, circles symbolize females. Affected individuals are indicated by highlighted symbols, with cancer type and age at diagnosis noted below. Unaffected individuals are indicated by empty symbols. Endometrial cancer proband sequenced is indicated by black arrow below the symbol. Variant carriers are indicated by a (+) symbol and the non-carriers are indicated by a (−) symbol.

#### *3.4. Tumor Sequencing to Assess Role of MUTYH Variant in a Suspected Familial EC Case*

To explore the potential role of the germline heterozygous *MUTYH* variant in cancer development, we conducted tumor DNA sequencing of the heterozygous *MUTYH* variant carrier (case 28) to establish whether there was evidence of tumor variant enrichment and whether the *MUTYH*-associated mutational signature could be detected. We performed whole genome sequencing of an archival endometrial tumor block from the *MUTYH* carrier. The read depth was too low to accurately assess evidence of loss of heterozygosity at the *MUTYH* locus, although an increase in the percentage of variant reads from 43% in germline (16 of 37 reads) to 67% in the tumor (six of nine reads) was suggestive of tumor variant enrichment. The sequencing analysis also revealed a high proportion of C>T and C>A somatic mutations (Figure 3A). The pattern of C>T mutations is similar to COSMIC Signature 1, identified in many tumors and typically attributed to aging or deamination [48] and may be present due to formalin fixation. By performing signature assignment analysis we attributed 41% of all somatic single nucleotide variants to Signature 18 (Figure 3B), previously associated with inactivation of *MUTYH* in a series of familial colorectal cancer and adrenocortical carcinomas [49], indicating that the germline variant was driving the pattern of somatic mutations, and underlay development of EC in this individual.

**Figure 3.** Somatic mutational signature analysis of the *MUTYH* germline variant carrier (suspected familial endometrial cancer cohort). (**A**) A total of 287 somatic single nucleotide variants (SNVs) identified in the endometrial tumor and used in signature analysis, plotted as counts in a 96 trinucleotide context. (**B**) The proportion of mutations in the tumor sample which were assigned to Signature 18.

#### **4. Discussion**

Based on the existing clinical management guidelines, a previous review suggested that only six genes currently have sufficient evidence of association with EC risk to be appropriate for hereditary EC diagnostic testing; these include the MMR genes (*MLH1*, *MSH2*, *MSH6* and *PMS2*), *EPCAM* (deletions due to their effect on *MSH2*) and *PTEN* [4]. We explored the role of candidate EC risk genes [4] beyond the MMR and *PTEN* genes, by analyzing an EC sample set unselected for family history and a cohort of familial EC cases. We also performed tumor sequencing analysis to explore whether these genes are cancer drivers associated with somatic mutagenesis in endometrial tumors.

The findings suggest that variation in the following genes should be considered in future studies of EC risk: *ATM, MUTYH, PALB2, RAD51C* and *NBN*. *PALB2* was highlighted by both case–control and suspected familial EC analysis. Tumor mutational signatures provided evidence that germline variation in *BRCA1*, *PALB2*, *RAD51C, MUTYH* and *NTHL1* can be (but is not always) associated with tumor mutational signatures consistent with a functional role of these genes in endometrial tumor development.

*ATM* encodes for a cell cycle checkpoint kinase that initiates DNA damage response via error-free repair pathway, HR, for double-stranded DNA breaks [50]. The *ATM* variant identified in a suspected familial EC case was classified as pathogenic for the rare autosomal recessive ataxia-telangiectasia syndrome by multiple submitters in ClinVar. The syndrome manifests a variety of phenotypic characteristics, including high incidence of cancer. Pathogenic variants in *ATM* are associated with increased breast cancer risk. Monoallelic c.7271T>G carriers are at a significantly increased risk, a 60% cumulative risk by age 80 years, similar to penetrance conferred by pathogenic germline variants in *BRCA2* [51]. Monoallelic carriers of other loss of function variants are reported to have a moderate increased risk of developing breast cancer (3-fold; 95% CI: 2.1–4.5) [52]. A number of *ATM* variants predicted to be deleterious to ATM protein function have been identified in EC cases, in unselected as well as a familial setting [7,53]. Another recent study [22] reporting results from germline panel testing of unselected EC cases identified *ATM* pathogenic variants as among the most common alterations observed (9/1170 cases), and estimated risk for *ATM* carriers to be OR 1.86 (*p* = 0.07) by comparison of case frequency to gnomAD non-Finnish European controls. Given that *ATM* loss of function variants

are estimated to be associated with only a modest risk of breast cancer (OR 3.0, 95% CI 2.1–4.5) [52], larger well-designed studies will be required to determine if *ATM* variation confers a similar modest level of risk to EC.

*PALB2* encodes for one of the key proteins involved in the HR DNA damage repair by recruiting BRCA2 to DNA breaks [54]. The *PALB2* truncating variant identified in our familial cohort has been classified as a pathogenic variant for familial breast cancer by multiple submitters to ClinVar. *PALB2* is emerging as a gene that confers a high risk of breast cancer, with data suggesting individuals with pathogenic variants in *PALB2* have a high lifetime risk of around 32% [55]. *PALB2* variants have also been associated with increased risk of ovarian and pancreatic cancers [56]. In our study, the EC patient carrying the *PALB2* variant had a strong family history of various cancers, with carrier or obligate carrier status confirmed for relatives with breast, colon and EC. EC has been reported in relatives of breast cancer patients known to carry loss of function variants in *PALB2* [57], but carrier status was not confirmed. *PALB2* loss of function variants have also been detected in EC patients in several previous studies [45,57–61]. The results to date indicate that the role of *PALB2* loss of function variants in conferring EC risk should be further explored.

Other HR pathway genes implicated in this study were *BRCA1*, *NBN* and *RAD51C*. Interestingly, while *NBN* and *RAD51C* had a more than 2-fold increased variant frequency in the EC sample set, *BRCA1* did not. To date, the role of *BRCA1* or *BRCA2* in EC risk has been much debated, with numerous conflicting reports [4]. Overall findings indicate that increased EC risk for *BRCA1/2* carriers has been associated with tamoxifen use for breast cancer prevention or treatment (since these genes confer high breast cancer risk) comparable to the risk observed in the general population [18,19]. There is also suggestive evidence that *BRCA1* pathogenic variants may confer a modest risk EC increase in the absence of tamoxifen exposure, particularly for serous and serous-like subtype cancers [62,63]. Unfortunately, the patient cancer history or tamoxifen exposure was not well documented for the TCGA-UCEC cancer cohort used in this study, hence we were unable to assess the possible contribution of tamoxifen for *BRCA1* or other genes that confer breast cancer risk. *RAD51C* has been recently shown to confer moderate risk for breast (relative risk (RR) = 1.99, 95% CI: 1.39–2.85) and high risk for ovarian cancers (RR = 7.55, 95% CI: 5.6– 10.19) [64]; however, there have only been observational studies so far for EC [7,65]. Given the breast cancer risk, future studies on *RAD51C* and EC risk will need to account for tamoxifen exposure, same as for *BRCA1/2* genes. The role of *NBN* in EC risk has largely been unexplored. It is notable that while certain *NBN* variants have previously been reported to increase breast cancer risk [66], the most recent evidence from a large-scale case–control analysis refutes (OR 0.90, 95% CI 0.67–1.20) an association of truncating *NBN* variants with breast cancer risk [67].

In addition to considering a role of the above-mentioned HR-related genes in EC risk, we also investigated their potential role in EC development by analyzing tumor mutational signatures. We observed HR-associated mutational signature (Signature 3) in most tumors with *BRCA1*, *PALB2* and *RAD51C* pathogenic or likely pathogenic variants, but not in tumors with *BRIP1*, *CHEK2*, *FAN1* or *NBN* variants. This is consistent with previous reports in breast cancer and cell line experiments where Signature 3 was only detected for *BRCA1/2*, *PALB2* and *RAD51C* genes but not *ATM* or *CHEK2* [68,69]. The presence of Signature 3 in cases with *BRCA1*, *PALB2* and *RAD51C* variants, as well as tumor enrichment of these variants, suggest that these cancers are HR-deficient. Our observation is also supported by the report of tumor loss of heterozygosity in serous/serous-like EC with germline *BRCA1* mutations (two of three cases) [62].

Other genes implicated in this study included DNA base excision repair genes, *MU-TYH* and *NTHL1*. Signature 36 (COSMIC v3), similar to Signature 18 detected in this study (COSMIC v2), has been associated with inactivation of *MUTYH* in MAP colorectal cancer [70] and observed in 5% of pancreatic neuroendocrine tumors that bore heterozygous germline *MUTYH* variants and subsequent loss of the wildtype allele in the tumor [71]. Together these observations indicate that oxidative DNA damage due to *MUTYH* inactiva-

tion may contribute to cancer etiology in several organs. In our study, a *MUTYH* variant considered pathogenic for MAP was likely enriched in the tumor of a suspected familial EC case, which presented with a tumor mutational signature consistent with the driver status of the *MUTYH* variant. However, *MUTYH* pathogenic variants were not more common in the TCGA unselected EC cohort relative to the population reference group (1.62% vs. 1.73%), and there was no evidence for tumor enrichment or appropriate tumor mutational signature in the TCGA cases. The majority of *MUTYH* pathogenic variants identified were two well recognized common pathogenic variants identified in the Western population to cause MAP (c.536A>G (p.Tyr179Cys), as detected in the suspected familial EC case; and c.1187G>A (p.Gly368Asp)) [47]. MAP is an autosomal recessively inherited predisposition to adenomatous polyposis and colorectal cancer [72]. The cumulative colorectal cancer risk to age 70 years for biallelic carriers is reported to be 75% (95% CI: 41–97%) for males and 72% (95% CI: 41–97%) for females, and for monoallelic carriers it is estimated to be 7% (95% CI: 5–11%) for males and 6% (95% CI: 4–9%) for females [73]. Indeed, the case reported here had a family history of colorectal cancer (in two relatives, ages 39 and 52). However, the risk of extracolonic cancers for *MUTYH* monoallelic pathogenic variant carriers with a family history of colorectal cancer is still uncertain, current evidence derived from a single study estimated cumulative risk of EC age 70 to be 4% (95% CI: 2–8%) for monoallelic *MUTYH* carriers [11], with an updated analysis of the same cohort [74] reporting a modest 2-fold EC risk for carriers (95% C.I 1.1–3.9). These previous findings suggest that, *MUTYH*-associated risk of EC, if validated, is likely to be extremely modest.

We have identified several genes in this study as possible additional EC risk genes; however, these results should be considered preliminary, and require further exploration in the follow-up studies. Although this study has not provided conclusive evidence regarding the role of the aforementioned genes in EC risk, results could nevertheless be of relevance as secondary findings for the patient and their relatives. We have shown that at least some germline carriers had a tumor mutational signature supportive of the driver role of the respective gene in cancer development. Overall, 28% of carriers of HR-related gene variants had a presence of HR-deficiency associated tumor mutational signature, which increased to 86% of carriers when only well-recognized HR genes (*BRCA1*, *PALB2* and *RAD51C*) were included [75]. This has implications for patient treatment decisions, since HR-deficient cancers are known to respond to PARP inhibitors [76]. Furthermore, cases with MMR-deficient or base excision repair-deficient (*MUTYH* or *NTHL1-*driven) tumors are likely to show hypermutated profiles, and thus would be good candidates for immunotherapy treatment, given the likely increase in neo-antigen production [77]. Our findings suggest potential value in secondary tumour profiling on identification of a germline gene alteration in EC patients, irrespective of a confirmed role of that gene in EC risk. Furthermore, somatic only changes would have the same implications for treatment decisions. It will thus be important to explore the overall proportion of EC cases with actionable tumor mutation profiles to determine the clinical value of unselected tumor mutational profiling.

#### **5. Conclusions**

We used genome sequencing and tumor mutational signature analysis to explore the role of purported EC risk genes in an EC sample set unselected for family history, and to identify candidate germline variants underlying the genetic cause of familial EC without MMR defects. Ancestry-matched case–control comparisons of germline variant frequency and/or sequence data from the suspected familial EC cases proposed several preliminary candidates for future risk association studies, with *PALB2* highlighted by both approaches. Tumor analysis highlighted germline variation in HR-related repair genes, particularly *BRCA1*, *PALB2* and *RAD51C,* to have a potential driver role in EC development based on the presence of mutational signature indicative of HR deficiency. For the heterozygous germline variants in other DNA damage repair genes, *MUTYH* and *NTHL1,* the mutational

signature analysis indicates possible involvement in the etiology of EC, but only when there were indications of the germline variant being enriched in the tumor.

Inclusion of these highlighted genes in clinical testing panels for EC predisposition will require results from further large-scale studies, to assess the level of EC risk associated with loss of function variation in these genes. Such studies should preferably follow a populationbased case–control design and consider the role of other genetic and environmental factors in disease penetrance, including previous exposure to tamoxifen. While we anticipate that genes outside of MMR pathway are unlikely to explain a large component of suspected familial EC, our results indicate that additional tumor signature analysis for individuals with a germline gene alteration has potential to impact therapeutic decisions.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/cancers13081762/s1, Figure S1: Variant filtering workflow used for germline pathogenic and likely pathogenic variants included in this study for TCGA-UCEC and gnomAD datasets, Figure S2: De novo mutational signature analysis in the TCGA-UCEC cases with pathogenic or likely pathogenic variants, Table S1: Subset of European TCGA-UCEC cases included in the variant frequency comparison with the gnomAD population of Non-Finnish Europeans, Table S2: Phenotypic characteristics of the familial endometrial cancer cases, Table S3: Summary of sequencing coverage and number of variants identified for each familial endometrial cancer case prior to filtering, Table S4: List of known and purported endometrial cancer risk genes, Table S5: List of pathogenic and likely pathogenic variants identified in the genes of interest in gnomAD (non-Finnish Europeans) and TCGA-UCEC cases.

**Author Contributions:** O.K. and J.S.: Designed the methods, carried out the analysis, interpreted the data and prepared the manuscript. A.B.S.: Conceived the study, designed the methods, and supervised the work. N.W.: Designed the methods, developed genomic analysis pipelines, supervised the work. T.A.O.: Designed the methods and supervised the work. F.N.: Carried out mutational analysis. A.E.M.R.: Reviewed and processed the tumor sample. S.R.L.: Reviewed the tumor sample and supervised the work. J.K.: Provided clinical input. J.V.P.: Managed the genomic data and developed the genomic pipelines for analysis. All authors read and approved the final manuscript.

**Funding:** J.S. is supported by a QIMR Berghofer PhD scholarship. N.W. is supported by a National Health and Medical Research Council (NHMRC) of Australia Senior Research Fellowship (APP1139071). A.B.S. was supported by NHMRC Funding (APP1061779, APP117524). T.A.O. was supported by NHMRC Funding (APP1111246, APP1173170). The Australian National Endometrial Cancer Study was supported by project grants from the NHMRC (Grant No. 339435); The Cancer Council Queensland (Grant No. 4196615); Cancer Council Tasmania (Grant No. 403031 and Grant No. 457636); the Cancer Australia Priority-driven Collaborative Cancer Research Scheme (#552468), Cancer Australia (Grant No. 1010859). The sequencing component of this study was funded by a "Rio Tinto Ride to Conquer Cancer & Weekend to End Women's Cancer Research Grant" administered by QIMR Berghofer Medical Research Institute.

**Institutional Review Board Statement:** Accessing TCGA-UCEC genome data and undertaking reanalysis to detect variants in the tumour and matching normal was performed with approval from the QIMR Berghofer Human Research Ethics Committee (HREC/P2905). Sequencing and analysis of familial EC patients was undertaken with approval from QIMR Berghofer Ethics Committee (HREC/P1051).

**Informed Consent Statement:** All ANECS participants provided informed written consent for participation in EC-related research, with overall study approval by the QIMR Berghofer Medical Research Institute Human Research Ethics Committee (P853), encompassing approvals from participating hospitals and cancer registries.

**Data Availability Statement:** TCGA-UCEC data were downloaded from GDC data portal in October 2016. GnomAD variant files (r.2.1.1) were downloaded from the gnomAD portal in April 2019. ANECS sequencing data are available upon reasonable request and subject to ethics approval.

**Acknowledgments:** We thank the individuals who participated in ANECS, the institutes and clinicians who contributed to recruitment and data collection (see website: www.anecs.org.au for the full list of contributors). We thank Sullivan Nicolaides Pathology for providing the archival tumor tissue for whole genome sequencing, and Stephen Kazakoff, Michael Bowman, Katia Nones, Ann-Marie Patch for helpful advice and contributions to data cleaning and processing. We thank Kathy Tucker for critique regarding interpretation and representation of clinical relevance of findings. The results shown here are in part based upon data generated by the TCGA Research Network: https://www. cancer.gov/tcga. Information about TCGA can be found at http://cancergenome.nih.gov. The TCGA dataset used in this study was accessed through dbGaP accession number phs000178.v11.p8.

**Conflicts of Interest:** O.K. has consulted for XING Technologies. N.W. and J.V.P. are co-founders and Board members of genomiQa. The other authors declare that they have no competing interests. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


## *Article* **Is Breast Cancer Risk Associated with Menopausal Hormone Therapy Modified by Current or Early Adulthood BMI or Age of First Pregnancy?**

**Eleni Leventea 1, Elaine F. Harkness 2,3,4, Adam R. Brentnall 5, Anthony Howell 2,4,6,7, D. Gareth Evans 2,4,6,8,9,10 and Michelle Harvie 2,4,6,7,\***


**Simple Summary:** Menopausal hormone therapy (MHT) increases risk of developing breast cancer (BC), and women are often advised to avoid its use for this reason. In this analysis we examined the size of this effect using data from a large cohort of women attending breast cancer screening in Manchester, UK. We additionally explored the extent to which risk from MHT might be modified by current BMI, early adulthood body mass index (BMI) (age 20 years), and age of first pregnancy. Identifying modifying effects would help enable better estimation of risk associated with MHT for an individual woman. Results indicated that women using combined oestrogen and progestagen MHT were at greater risk than those receiving oestrogen-only MHT. The Relative risk associated with MHT was less for obese women than non-obese women. After adjustment for current BMI, the effect of MHT did not appear to be substantially modified by early BMI or age of pregnancy.

**Abstract:** Menopausal hormone therapy (MHT) has an attenuated effect on breast cancer (BC) risk amongst heavier women, but there are few data on a potential interaction with early adulthood body mass index (at age 20 years) and age of first pregnancy. We studied 56,489 women recruited to the PROCAS (Predicting Risk of Cancer at Screening) study in Manchester UK, 2009-15. Cox regression models estimated the effect of reported MHT use at entry on breast cancer (BC) risk, and potential interactions with a. self-reported current body mass index (BMI), b. BMI aged 20 and c. First pregnancy >30 years or nulliparity compared with first pregnancy <30 years. Analysis was adjusted for age, height, family history, age of menarche and menopause, menopausal status, oophorectomy, ethnicity, self-reported exercise and alcohol. With median follow up of 8 years, 1663 breast cancers occurred. BC risk was elevated amongst current users of combined MHT compared to never users (Hazard ratioHR 1.64, 95% CI 1.32–2.03), risk was higher than for oestrogen only users (HR 1.03, 95% CI 0.79–1.34). Risk of current MHT was attenuated by current BMI (interaction HR 0.80, 95% CI 0.65–0.99) per 5 unit increase in BMI. There was little evidence of an interaction between MHT use, breast cancer risk and early and current BMI or with age of first pregnancy.

**Citation:** Leventea, E.; Harkness, E.F.; Brentnall, A.R.; Howell, A.; Evans, D.G.; Harvie, M. Is Breast Cancer Risk Associated with Menopausal Hormone Therapy Modified by Current or Early Adulthood BMI or Age of First Pregnancy? *Cancers* **2021**, *13*, 2710. https://doi.org/10.3390/cancers 13112710

Academic Editor: Virgilio Sacchini

Received: 9 April 2021 Accepted: 25 May 2021 Published: 31 May 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

**Keywords:** menopausal hormone therapy; breast cancer risk; BMI; early BMI; age of pregnancy

#### **1. Introduction**

Approximately 80% of women going through the menopause experience symptoms [1]. Menopausal hormone therapy (MHT) is the most effective treatment option. MHT use halved in the early 2000s as a result of widely publicised associations between MHT use and increased risk of breast cancer and thromboembolism. Rates stabilised in the 2010s and currently there are an estimated one million MHT users in the UK each year, representing approximately 10% of women passing through the menopause [1]. A recent meta-analysis based on 58 prospective and retrospective studies, including 568,859 women and 143,887 breast cancer cases concluded that ever MHT use is associated with increased breast cancer risk (RR 1.26, 95% CI 1.24–1.28). Risk is higher for current compared to past users and increased with longer MHT use. Amongst current users, risk is greater with combined (oestrogen plus progestagen) MHT (RR for 5–14 years of use 2.08, 95% CI 2.02–2.15) compared to oestrogen only MHT (RR 1.33, 95% CI 1.28–1.37). Risk was attenuated amongst heavier women, particularly for oestrogen-only MHT, with little additional risk from oestrogen-only MHT in women who were obese [2].

MHT use increases breast cancer risk amongst women in the general population but also in women with increased risk due to familial cancer [3]. Many women, especially those at higher risk of breast cancer, are counselled to avoid or minimise MHT use [4]. Guidelines recommend an individual risk benefit assessment for prescribing MHT [5]. However there are limited data on whether MHT risk is modified by other patient characteristics which would help to inform this decision.

Increased body mass index (BMI) after the menopause (RR per 5 BMI units: 1.12, 95% CI 1.09–1.15) and weight gain throughout adult life after the age of 20 (RR per 5 kg: 1.06, 95% CI 1.05–1.08) are consistently associated with increased risk of postmenopausal breast cancer [6]. In contrast, higher weight during adolescence and early adulthood (age ≤30 years) are observed to have an inverse effect [7]. High body adiposity in early adulthood is associated with a reduced risk of postmenopausal (RR per 5 BMI units: 0.82, 95% CI 0.76–0.88) and premenopausal breast cancer (RR per 5 BMI units: 0.82, 95% CI 0.76–0.89) [6]. We recently reported higher BMI in early adulthood (>23.4 kg/m2) negated the impact on risk of high later attained BMI [8].

Late age of first pregnancy (after the age of 30 years) and nulliparity are associated with increased risk of breast cancer. Women with first pregnancy after the age of 30 have approximately twice the risk of developing breast cancer compared to women with first pregnancy before the age of 20, and nulliparous women have a 30% increased risk compared to parous women [9]. Risks associated with attained adult and early adulthood weight and age of pregnancy are all in part mediated by different exposure to oestrogen and progesterone, and may alter the hormone responsiveness of the breast [10]. Thus they may modify the risk associated with MHT use.

Here we sought to examine the association between combined and oestrogen only MHT use and breast cancer risk in a cohort of women from the National Health Service Breast Screening Programme (NHSBSP), in Greater Manchester, UK. We examined whether BC risk associated with these types of MHT are modified by BMI at study entry (age 46–84 years) and/ or early adulthood BMI (age 20), or age at first pregnancy or nulliparity.

#### **2. Methods**

#### *2.1. Population*

The Predicting Risk of Cancer At Screening (PROCAS) study has been described in detail elsewhere. In total 57,902 women aged between 46 and 84 years in the National Health Service Breast Screening programme (NHSBSP) were recruited from five areas of Greater Manchester (Manchester, Oldham, Salford, Tameside and Trafford) between October 2009

and June 2015 [11]. Recruitment was carried out in two phases: initially all women who were invited for three-yearly breast screening (October 2009–October 2012) after which only women invited to their first screen in the area (mainly aged 46–53 years) were invited to participate in the study. Participants were invited once during the recruitment period. During the initial phase uptake to screening was 68% with uptake to PROCAS 37% of attendees, in the second phase screening uptake was 58% and uptake to PROCAS was 47% of attendees (screening uptake is lower in first time invitees).

#### *2.2. Data Collection*

Data collection was based on a two-page questionnaire, which was sent to participants with a consent form between their invitations to attend screening and scheduled screening appointment (Supplementary Materials).

The self-reported questionnaire gathered information on risk factors for breast cancer including: previous breast cancer diagnosis, breast or ovarian cancer in first and second degree relatives, hormonal risk factors, including age at menarche, oophorectomy and hysterectomy, menopausal status, age at menopause, parity and age at first pregnancy, physical activity levels, alcohol intake and ethnicity. Questions in relation to MHT use included name of preparation, duration of use and when MHT was last used (if no longer on MHT). Women were also asked to record their height, current weight and recalled weight at age 20 as a proxy of early adulthood BMI. Current BMI and BMI at 20 were calculated from these variables. Completed questionnaires were collated by the study team and entered into the study database.

#### *2.3. Diagnosis of Breast Cancer*

The primary outcome was diagnosis of a new breast cancer (invasive or ductal carcinoma in situ), from entry to PROCAS onwards, as identified through the NHSBSP and the Somerset and North West Cancer Intelligence services. Follow-up (median eight years) was censored at date of breast cancer diagnosis, date of death, date lost to follow-up, e.g., moving out of the area, or date cancer databases were last checked (April 2020). The current analysis excluded women with breast cancer diagnosed prior to study entry (*n* = 895).

#### *2.4. MHT Use*

MHT use was classed as never, current and former. Never users were women who indicated they had never been on MHT at any time. Current users were women who reported they were still on MHT and gave details about length of time on MHT but did not indicate a time of stopping. Former users were women who reported no longer using MHT but indicated how long they had been using MHT and time since stopping.

Where women did not provide an MHT name, it was assumed that those who had a hysterectomy had oestrogen only MHT, whereas women who did not have hysterectomy had combined (oestrogen plus progestagen) MHT. MHT status was missing for 514 women who were excluded from any further analyses (Figure 1).

The analyses included pre/peri and postmenopausal women enrolled in the PROCAS study. Women were considered postmenopausal if they indicated on the questionnaire they had been through the menopause (i.e., not had period for 12 months) or reported they had both ovaries removed (surgical menopause) or current use of MHT or age at menopause was unknown but age at time of questionnaire completion was 55 years or over as based on criteria defined by Phipps et al. [12].

**Figure 1.** Flow diagram: Number of women in the cohort meeting the criteria for inclusion in the analysis.

#### *2.5. BMI Data*

We excluded attained BMI and early BMI values which were >60 or <16 kg/m2. For current BMI, 3943 were unknown or beyond these cut-offs. For early BMI, 6300 values were unknown or beyond these cut-offs. Median value of the cohort was assumed for missing values (total 6.8% for current and 10.9% for BMI at age 20).

#### *2.6. Statistical Analysis*

Demographic Characteristics and Breast Cancer Risk Factors amongst MHT Users and Non-Users

Differences in demographic characteristics and breast cancer risk factors across MHT groups (never, current, former) were tested using one-way ANOVA and Chi-square tests where appropriate. Adjusted comparisons of continuous measure used linear regression (with covariates for age (years), current and early BMI ( kg/m2)).

#### *2.7. MHT Use and BC Risk*

Cox (or proportional hazards) regression was used to model the relationship between MHT use and diagnosis of breast cancer. Follow-up was censored at date of breast cancer, date of death or date of last follow-up (April 2020). Results were expressed as hazard ratios (HR) and 95% Wald confidence intervals (95% CI).

Fully adjusted Cox regression models included the following established risk factors for breast cancer: age (1 year), height (5 cm), BMI (5 units), early BMI (5 units), ethnicity (white/other), age at menarche (1 year), age at first pregnancy (<20, 20–24, 25–29, 30–34, ≥35) years, parity, age at menopause (1 year) menopausal status (pre/peri or postmenopausal), oophorectomy, self-reported exercise (1 h/week) and alcohol (1 unit/week), MHT status (current, former, never), MHT type (combined, oestrogen only) and family history (first or second degree). For the fully adjusted analysis, missing values were imputed as the median value of the cohort. We did not include the available mammographic density data in the models since there is some evidence this may be part of the pathway for reduced risk alongside higher early BMI [13]. The multiple deprivation score was not included since this was not associated with risk once the variables associated with this, such as age of pregnancy were included in the model (Table S1a, S1b).

#### *2.8. MHT Use, Breast Cancer Risk and Effect Modification by Current BMI and Early BMI and Age of First Pregnancy*

To assess whether current and early BMI modified the relationship between MHT and the risk of breast cancer the following two way interaction terms were also included in the fully adjusted models: current/ former MHT use\* current BMI \* (Table S2a) and current/ former MHT use\* early BMI (Table S2b). We also tested the 3-way interactions: current/ former MHT use\*current BMI \*early BMI (Table S2c).

For presentation the relationship between MHT use and BC risk was tabulated by stratifying above and below the median for current (26.4 kg/m2) and early (21.6 kg/m2) BMI. The reference group was never use of MHT, current BMI below the median and early BMI below the median. The median was used to dichotomise results and aid with interpretation (Table 1).

The relationship between late age at first pregnancy and nulliparity, MHT status and BC risk was assessed using analysis stratified by age at first pregnancy <30 years or a combined group which included women with first pregnancy ≥30 years or who were nulliparous (Table 2 and Table S5).

Due to previous known heterogeneity by type of MHT and subtype of breast cancer we also performed these analyses according to (i) MHT type (combined oestrogen and progestagen or oestrogen only) (Table 3) and (ii) oestrogen receptor positive breast cancer. (Table S1c).


**Table 1.** MHT and BC risk and current and early BMI.

BC breast cancer; Units: age (1 year), BMI (5 BMI units), BMI at age 20 (5 BMI units), height (5 cm), age at menopause (1 year), exercise (1 h per week), alcohol (1 unit per week), age at menarche (1 year) BMI median: 26.4 kg/m2, BMI20 median: 21.6 kg/m.

**B**


**Table 2.** MHT status and BC risk and age of pregnancy > 30 and nulliparity.

BC breast cancer; Units: age (1 year), BMI (5 units), BMI at age 20 (5 units), height (5 cm), age at menopause (1 year), exercise (1 h per week), alcohol (1 unit per week), age at menarche (1 year).

**Table 3.** MHT type and BC risk and current and early BMI.




**Table 3.** *Cont.*

BC breast cancer; Units: age (1 year), BMI (5 units), BMI at age 20 (5 units), height (5 cm), age at menopause (1 year), exercise (1 h per week), alcohol (1 unit per week), age at menarche (1 year); BMI median: 26.4 kg/m2, BMI age 20 median: 21.6 kg/m2.

#### **3. Results**

#### *3.1. Flow Diagram*

From the cohort of 57,902 women, 895 were diagnosed with breast cancer prior to study entry, 4 were lost to follow up and were excluded from the analysis. Women with unknown MHT use status were also excluded from the analysis (*N* = 514). The denominator for the main analysis was 56,489 with the endpoint being a new diagnosis of breast cancer (*N* = 1663, 3.2% of the cohort) (Figure 1).

#### *3.2. Demographic Characteristics and Breast Cancer Risk Factors According to MHT Use*

Demographic characteristics and BC risk factors according to MHT groups are shown in Table 4. Of those eligible for analysis 7.8% were current MHT users, 28.6% were former users and 63.6% had never used MHT. Across all MHT groups at study entry, 35.9% were in the underweight or normal BMI category range, 39.5% in the overweight, 24.6% in the obese BMI category, 3801 women had an unknown BMI. The median age at first pregnancy was 24 years (interquartile range [IQR] 21–28 years), 26.7% of women had their first pregnancy either at or after the age of 30 years. In total, 27.9% of women had a first or second degree family history of ovarian and/ or breast cancer.

When compared to never users, current MHT users had a higher percentage of oophorectomy (27.5% vs. 5.6%), lower BMI at study entry (median: 26.0 vs. 26.4 kg/m2), were less likely to have a family history of ovarian and/ or breast cancer (26.6% vs. 28.3%), a lower deprivation score (median: 17.6 vs. 19.4), and younger age of first pregnancy (median 23 vs. 24 years). As expected, former users were the oldest group, also reflected by the higher percentage of participants being postmenopausal.


**Table 4.** Baseline Characteristics of the 56,489 women in the PROCAS cohort (2009–2015).

BMI body mass index, SD standard deviation, IQR interquartile range; MHT hormonal replacement therapy, VAS visual analogue scale; EIMD English Index of multiple deprivation. \* First or second degree relative with ovarian and/ or breast cancer; \*\* Ever vs. never MHT EIMD 2010 *p* = 0.002, EIMD current vs. never *p* < 0.001, never vs. former *p* = 0.444; age adjusted comparisons: BMI current vs. never *p* < 0.001, never vs. former *p* = 0.639; age adjusted VAS current vs. never *p* < 0.001, never vs. former *p* = 0.697; age, menopausal status, BMI adjusted VAS current vs. never *p* < 0.001, never vs. former *p* = 0.001.

> Compared with former users, current users were younger (median: 55.2 vs. 62.9 years), had longer use of MHT (median: 7 vs. 5 years), higher percentage use of oestrogen only

MHT (51.7% vs. 41.0%), lower BMI at entry (median: 26.0 vs. 26.4 kg/m2), were less likely to have family history of breast and/ or ovarian cancer (26.6% vs. 27.1%), and lower deprivation score (median: 17.6 vs. 18.7). All differences cited have *p* < 0.001 due to large sample size. Women with known MHT status were lighter at entry to PROCAS, more likely to be white, and from less deprived backgrounds than those with unknown MHT status (Table S3).

#### *3.3. MHT Use Status and Breast Cancer Risk*

Median follow up from study entry was 8 years (IQR: 7–9, minimum 5 and maximum 12 years). Compared with never use, current MHT use was positively associated with breast cancer (HR 1.35, 95% CI 1.13–1.60), while there was little evidence of an association for former users (HR 1.03, 95% CI 0.91–1.17) (Table 5). The fully adjusted model fit is shown in Table S1b.



BC breast cancer, HR hazard ratio, CI confidence interval. Fully adjusted for age at consent (1 year), BMI (5 units), BMI at age 20 (5 units), height (5 cm), age at menarche (1 year), age at menopause (1 year), menopausal status (pre/perimenopausal vs. postmenopausal), ethnicity (white vs. other), alcohol consumption (1 unit/week), exercise (1 h/week), age at first pregnancy (<20, 20–24, 25–29, 30–34, ≥35), oophorectomy (yes vs. no), family history (yes vs. no).

> Breast cancer risk was highest in current users of combined MHT (*n* = 2129) (HR 1.64, 95% CI 1.32–2.03), compared with never users. Risk amongst current users of oestrogen only was (*n* = 2278) (HR 1.03, 95% CI 0.79–1.34) compared with never users (Table 5).

> Analysis was repeated for ER+ breast cancers only, which comprised 88% of breast cancer diagnoses in the cohort. Current MHT users (combined and oestrogen only) were at increased risk of an ER+ diagnosis BC (HR 1.45, 95% CI 1.20–1.74) (Table S1c).

#### *3.4. Current BMI and MHT and Breast Cancer Risk*

Current BMI was positively associated with breast cancer risk (adjusted HR 1.23 per 5 kg/m2, 95% CI 1.16 to 1.30; Table S1). The effect of MHT in current users was attenuated by BMI (adjusted interaction 0.81, 95% CI 0.67 to 0.98; Table S1b). This is illustrated by the data in Table 1 Part A, where there is less difference in risk between never and current users of MHT in the higher BMI category than the lower BMI category.

#### *3.5. Early Adulthood BMI and MHT and Breast Cancer Risk*

BMI in young adulthood was inversely associated with breast cancer risk (adjusted HR 0.77 per 5 kg/m2, 95% CI 0.69 to 0.87; Table S1b). BMI at age 20 did not attenuate risk of current MHT use (adjusted interaction 1.05, 95% CI 0.72 to 1.53; Table S1). Table 1 Part B shows high early BMI reduced risk across never, former and current users of MHT, MHT did however increase risk across women with early BMI above and below the median.

#### *3.6. Combined Stratified Model with Current BMI, Early BMI and MHT and Breast Cancer Risk*

Risk of breast cancer amongst MHT users was stratified for ≥ and < median for current BMI (26.4 kg/m2) and ≥ and < median for BMI age 20 years (21.6 kg/m2) (Table <sup>1</sup> Part C).

Higher BMI at age 20 appears to attenuate the effects of high current BMI on BC risk both amongst women who are not using MHT and amongst current MHT users (adjusted interaction BMI \*BMI 20 HR 0.96, 95% CI 0.90–1.02; Table S1).

There is no specific interaction between MHT use, current BMI and early BMI (HR 0.99, 95% CI 0.74–1.31) (Table S2c). Risk was highest amongst MHT users with current BMI higher than the median and BMI at 20 less than the median (HR 1.84, 95% CI 1.22–2.76) (Table 1c). Analysis for ER+ breast cancer found comparable results to the overall analysis (Table S2d).

#### *3.7. Effects of Oestrogen Only and Combined MHT and Current and Early BMI and Breast Cancer Risk*

Current combined MHT use increased risk amongst women irrespective of their current and early adulthood BMI (Table 3 Part A). Oestrogen only MHT use did not significantly increase risk in current or former MHT users in any of the BMI groups (Table 3 parts A, B, and C). However, risk appeared lower amongst current users of oestrogen only MHT with early BMI > median (HR 0.67, 95% CI 1.45–1.00) compared to early BMI < median (HR 1.23, 95% CI 0.87–1.73) (Table 3 Part B). Attenuation of risk amongst current oestrogen only MHT users with current BMI > median is seen in women who also had early BMI > median (HR 0.75, 95% CI 0.44–1.28) compared to women with early BMI < median (HR 1.68, 95% CI 0.97–2.90) (Table 3 Part C).

#### *3.8. MHT Use and Age at First Pregnancy*

Women who had a late pregnancy had an increased risk of breast cancer (age of first pregnancy >35 years HR 1.38, 95% CI 1.08–1.76) (Table S1b). Late first pregnancy or nulliparity increased risk across MHT users and non-users (Table 2). There is no specific interaction between MHT use, age of pregnancy or nulliparity (HR 1.00, 95% CI 0.80–1.02) (Table S4).

#### *3.9. Effects of Oestrogen Only and Combined MHT and Age of First Pregnancy*

Risks were comparable with combined MHT use amongst women with age of first pregnancy <30 years (HR 1.74, 95% CI 1.34–2.25) and > age 30 or nulliparity (HR 1.90, 95% CI 1.16–3.13). However, risks appeared lower amongst current oestrogen only MHT users with age of first pregnancy <30 years (HR 0.96, 95% CI 0.71–1.31) compared to >age 30 or nulliparity (HR 1.81, 95% CI 0.93–3.53) (Table 2).

#### **4. Discussion**

In this cohort, women with a lower current BMI were more likely to be current MHT users compared to women with a higher BMI. We have confirmed that current use of MHT increases breast cancer risk with excess risk mainly attributed to use of combined MHT. Higher early adulthood BMI had a small reduction in risk across never, former and current MHT groups. Late first pregnancy or nulliparity increased risk across never, former and current use of MHT groups. Neither BMI at age 20 or late first pregnancy or nulliparity had a specific modifying effect on the breast cancer risk related to overall MHT use. Observations of lower risks with oestrogen only MHT amongst women with high early BMI and early age of first pregnancy are interesting and require further study in larger cohorts.

Previous studies have reported that women with higher BMI were less likely to have used MHT [14–16]. Possible reasons for less use of MHT by heavier women include: experiencing fewer menopausal symptoms although this seems unlikely as the majority of papers report increased vasomotor symptoms in heavier women [17,18]; reduced likelihood of heavier women engaging with health behaviours, and contra-indications to MHT prescription associated with higher risk of thrombosis [19].

Our findings of increased breast cancer risk with current combined MHT, particularly with ER+ cancers concur with those found by the Collaborative Group [2]. We did not however observe an increased breast cancer risk for former MHT users. There was no significant association with oestrogen only MHT and BC risk, however our reported confidence intervals of the HR for oestrogen only MHT overlap with that reported by the Collaborative Group, indicating the results are broadly similar.

In these analyses we observed that higher current BMI attenuates the BC risk associated with overall current (combined and oestrogen only) MHT use. The attenuation of BC risk associated with oestrogen only MHT amongst currently heavier women has previously been reported [2]. Postmenopausal oestrogen levels correlate with BMI since endogenous oestrogen synthesis occurs within adipose tissue. The observed attenuation of oestrogen MHT risk is thought to reflect that exogenous oestrogen do not further stimulate the breast tissue in heavier women. This is consistent with a previous stated model that proposes a threshold for free oestrogen concentration beyond which there is no additional risk of breast cancer [20].

The stratified analysis in the current study suggests the attenuation that effects of oestrogen only MHT amongst currently heavier women is mainly seen in women who were heavier at an early age, and are not seen in formerly lighter women. The confidence intervals of these associations are quite wide due to small numbers of current users. However this observation raises the possibility that the apparent attenuation of oestrogen only MHT breast cancer risk amongst currently heavier women may be related to early rather than current BMI effects. Previous reports summarised in the collaborative overview did not examine the effects of early BMI.

A previous analysis within the PROCAS cohort reported that for women with early adulthood BMI >23.4 kg/m<sup>2</sup> (top 25% centile) neither attained adult BMI nor adult weight gain was associated with breast cancer risk [8]. The observation that higher BMI in early adulthood attenuates the BC risk associated with later adiposity has been reported in a number of studies [21–23]. A significant part of BC risk associated with postmenopausal BMI is thought to be mediated by increased oestrogen levels and the associated stimulation of breast tissue proliferation [24]. Reduced breast tissue proliferation has been reported amongst pre and postmenopausal women who had been heavier at age 18 (BMI >22 kg/m2) [25]. Higher early adulthood BMI may attenuate the proliferative response of breast tissue to endogenous (associated with current BMI) or exogenous (MHT) oestrogen through a number of mechanisms. These include reducing terminal end buds and ductal elongation, and an overall reduction in the number of cells within breast tissue [25–27] and decreased expression of genes involved with both oestrogen action, i.e., ESR1 and GATA3, and cell proliferation, i.e., RPS6KB1, in breast tissue [28]. Also, increased levels of bioavailable oestrogens in early adulthood (associated with insulin levels and decreased sex hormone binding globulin) [29] could induce earlier differentiation of mammary cells [30] and expression of the BRCA1 tumour suppressor gene [31]. Higher levels of insulin-like growth factor 1 (IGF-1) during childhood and adolescence are associated with lower levels in adulthood [32]. The synergistic effects and cross talk between IGF-1 and oestrogen and their receptors are well established [33]. In addition women who are heavier in early adulthood are likely to have greater numbers of adipocytes (a hyperplastic phenotype). In contrast formerly lean women who gain weight will develop adipose hypertrophy (few but large cells) which are associated with inflammation and dysregulated metabolism [34].

A prospective study among 483,241 women and 7656 breast cancers studied whether MHT-associated BC risk is modified by life course patterns of BMI. The study reported that current users of MHT who reported being overweight at the ages of 7 and 15 (selfidentified as being heavier than their peers) were at higher risk (HR 1.68, 95% CI 1.32–2.14) compared to never users of MHT who were overweight as young. However, risks were higher amongst current MHT users who remained at normal weight throughout adult life (HR 2.25, 95% CI 1.93–2.62) or who had gained weight (HR 2.28, 95% CI 1.94–2.67) [35]. The authors reported these risks were higher than expected when adding the separate risks of BMI and MHT with respective relative excess risk due to the interaction scores of 0.52 (95% CI 0.09–0.95) and 0.37 (95% CI -0.07 – +0.08). They concluded that women who were

overweight at a young age were less susceptible to the effects of MHT than women who remained a normal weight or who gain weight in adulthood. This effect was seen amongst the whole cohort (70% combined MHT and 23% oestrogen only) and within the combined only group. They did not report the associations amongst current oestrogen only MHT users. The findings from this large cohort are of interest, however the analyses of early weight did not adjust for current BMI. Since higher weight at a younger age is likely to result in a higher current BMI, this study may not distinguish any interactions between early weight and current BMI and the effects of MHT.

We found that late age at first pregnancy increased risk amongst never, current and former MHT users. Age at first pregnancy does not modify risk associated with overall MHT use and breast cancer. Oestrogen only MHT appeared to be associated with a greater risk amongst women with a late first pregnancy or nulliparity compared to women with age of first pregnancy <30. There was no modification of the risk of combined MHT.

Risk of late pregnancy relates to a prolonged duration of undifferentiated state of the mammary tissue [36]. Synergistic effects of nulliparity and high postmenopausal BMI on BC risk in women aged >70 years have been reported [37]. Murrow et al recently reported that parity and high BMI amongst premenopausal women both decreased the oestrogen and progesterone responsiveness of the breast. These effects were associated with respective reductions in hormone signalling in hormone responsive luminal cells and reductions in the proportion of hormone responsive luminal cells within the mammary epithelium [10]. This data suggests a common pathway that could be shared between early pregnancy and high young age adiposity that is protective against breast cancer.

Our stratified analysis aimed to study the independent and combined interactions of current BMI, early BMI, MHT use and BC risk. Also whether these interactions differed with combined and oestrogen only MHT. Our sample size limits the ability to study all of these relationships with sufficient power. The observed trends in higher early weight and early pregnancy reducing the risk of oestrogen only MHT breast cancer requires further investigation in larger cohort or consortium studies.

Strengths of this analysis are that it was conducted in a large UK population. Many confounders associated with breast cancer risk were taken into consideration. The independent effects of both current BMI and early adulthood BMI were elicited in the models by including BMI at two different time points. Additionally, detailed information regarding MHT use and type were collected and breast cancer diagnosis updated on a regular basis. Sensitivity analysis showed that the models were consistent.

All information used in this analysis were self-reported, including current weight, height, weight at age 20, age at first pregnancy, name of MHT, how long MHT was used for and when MHT was stopped for former users. It is well known that there is a bias of underreporting weight and over reporting height [38]. However validation studies show self-reported BMI is highly correlated with independently measured weight and the mean difference between self-reported and measured weight is minimal [8,39]. BMI at entry and at age 20 was missing for 7.0% and 12.4% of the study population respectively. Recall bias could also occur for variables such as MHT duration. We did not update HRT usage status during follow up, and so were unable to estimate any association between duration of HRT use and risk of BC. The median value of the cohort was assumed where data was missing, and other methods could be used to account for the missing values. Breast cancer risk varies across racial groups [40] and the majority of women in the study were Caucasian thus limiting generalization of the findings to other ethnic groups. Risk factor information was only collected at baseline and it is possible that MHT status, BMI and other risk factors changed for some women. Some current MHT users are likely to have become former users and premenopausal women who had not used MHT could have become postmenopausal and started using MHT during the eight years of follow up.

Implications for practice for this research include that clinical risk assessment of suitability of a woman to commence on MHT, should include consideration of their current BMI and potentially early adulthood BMI and the type of MHT to be prescribed. The findings of this study support recommendations to maintain a healthy weight across the life course. Breast cancer risk is similar among women with higher current BMI who never used MHT and women with lower BMI who use MHT. The smaller increase in risk with MHT amongst heavier women should not deter these women from losing weight. The highest BC risk is seen amongst current MHT users with a high current BMI, especially those with a low BMI at an early age. Women at increased weight taking MHT will have higher risks of other MHT associated adverse effects including venous thromboembolism, stroke and endometrial cancer [5]. Weight loss has been shown to decrease risk of other cancers including colorectal and endometrial cancer, and additionally might help manage menopausal symptoms [41].

Future studies need to investigate the associations between current and early BMI and age of pregnancy and MHT associated breast cancer risk. These studies could determine whether there are specific BMI ranges where these effects occur and whether the range is different within different ethnic groups.

#### **5. Conclusions**

Combined oestrogen and progestagen MHT was associated with the highest BC risk. This risk was not modified by early or current BMI and age of pregnancy. Exploratory analysis amongst oestrogen only MHT users showed an attenuation of risk with early BMI greater than or equal to the median compared to less than the median and with age of first pregnancy less than 30 years compared to equal or greater than age 30 or nulliparity which require further study. Identifying characteristics which modify a woman's MHT associated BC risk will allow their individual risks and benefits to be assessed and appropriate prescription of MHT to manage troublesome menopausal symptoms.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/cancers13112710/s1, Table S1a: MHT status and breast cancer risk model without BMI interactions, Table S1b: Type of MHT and breast cancer risk fully adjusted model, Table S1c: MHT status and breast cancer risk and ER+ breast cancer; Table S2a: MHT status and breast cancer risk fully adjusted model including current BMI interaction term, Table S2b: MHT status and breast cancer risk fully adjusted model including BMI at age 20 interaction term, Table S2c: MHT status and breast cancer risk fully adjusted, including MHT use\* current BMI \* BMI at age 20 interaction term, Table S3: characteristics of women with known and unknown MHT status; Table S4: MHT status and breast cancer risk fully adjusted with interaction term for age of first pregnancy.

**Author Contributions:** Conceptualization M.H., A.H., D.G.E., A.R.B., E.F.H.; Formal Analysis E.L.; Writing—Original Draft Preparation E.L., M.H.; Writing—Review & Editing E.L., M.H., A.H., D.G.E., A.R.B., E.F.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Central Manchester Research Ethics Committee (reference 09/H1008/81).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The dataset used and analysed during the current study is available from the corresponding author on reasonable request.

**Acknowledgments:** M.H.: D.G.E. and E.F.H. are supported by the NIHR Manchester Biomedical Research Centre (IS-BRC-1215-20007). This work is supported by the National Institute for Health Research (NF-SI-0513-10076 and IS-BRC-1215-20007 to D.G.E.). We thank all PROCAS researchers, clinicians, administrative staff and the individuals who contributed to our study. We thank Mary Pegington for proofreading.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


## *Article* **The Relationship between Body Mass Index and Mammographic Density during a Premenopausal Weight Loss Intervention Study**

**Emma C. Atakpa 1, Adam R. Brentnall 1, Susan Astley 2,3,4, Jack Cuzick 1, D. Gareth Evans 2,3,5,6,7, Ruth M. L. Warren 8,9, Anthony Howell 2,3,10 and Michelle Harvie 2,\***


**Simple Summary:** This study assessed the association between short-term weight change and mammographic density in premenopausal women losing weight through diet and exercise to reduce their risk of postmenopausal breast cancer. We aimed to understand whether a reduction in body mass index affects various components of the breast, which could indicate a potential pathway for the reduction in postmenopausal breast cancer risk seen with premenopausal weight loss. Understanding this pathway is useful for monitoring the effectiveness of prevention strategies based on lifestyle advice. We found that a short-term reduction in premenopausal body mass index through diet and exercise is associated with a reduction in breast fat, but it is unlikely to have a significant effect on the quantity of breast glandular tissue. Breast cancer risk determined by changes in breast density might not capture potential weight loss-induced breast cancer risk reduction, instead falsely ascribing an increased risk due to increased percent density.

**Abstract:** We evaluated the association between short-term change in body mass index (BMI) and breast density during a 1 year weight-loss intervention (Manchester, UK). We included 65 premenopausal women (35–45 years, ≥7 kg adult weight gain, family history of breast cancer). BMI and breast density (semi-automated area-based, automated volume-based) were measured at baseline, 1 year, and 2 years after study entry (1 year post intervention). Cross-sectional (between-women) and short-term change (within-women) associations between BMI and breast density were measured using repeated-measures correlation coefficients and multivariable linear mixed models. BMI was positively correlated with dense volume between-women (*r* = 0.41, 95%CI: 0.17, 0.61), but less so within-women (*r* = 0.08, 95%CI: −0.16, 0.28). There was little association with dense area (betweenwomen *r* = −0.12, 95%CI: −0.38, 0.16; within-women *r* = 0.01, 95%CI: −0.24, 0.25). BMI and breast fat were positively correlated (volume: between *r* = 0.77, 95%CI: 0.69, 0.84, within *r* = 0.58, 95%CI: 0.36, 0.75; area: between *r* = 0.74, 95%CI: 0.63, 0.82, within *r* = 0.45, 95%CI: 0.23, 0.63). Multivariable

A.R.; Astley, S.; Cuzick, J.; Evans, D.G.; Warren, R.M.L.; Howell, A.; Harvie, M. The Relationship between Body Mass Index and Mammographic Density during a Premenopausal Weight Loss Intervention Study. *Cancers* **2021**, *13*, 3245. https://doi.org/10.3390/ cancers13133245

**Citation:** Atakpa, E.C.; Brentnall,

Academic Editors: Ranjit Manchanda and Andrea Manni

Received: 7 April 2021 Accepted: 18 June 2021 Published: 29 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

models reported similar associations. Exploratory analysis suggested associations between BMI gain from 20 years and density measures (standard deviation change per +5 kg/m2 BMI: dense area: +0.61 (95%CI: 0.12, 1.09); fat volume: −0.31 (95%CI: −0.62, 0.00)). Short-term BMI change is likely to be positively associated with breast fat, but we found little association with dense tissue, although power was limited by small sample size.

**Keywords:** mammographic density; body mass index; weight loss; breast cancer risk; breast cancer prevention; premenopausal

#### **1. Introduction**

Mammographic density (herein referred to as 'density') is an established risk factor for breast cancer. Women in the highest density category are at a 4- to 6-fold increased risk of breast cancer relative to those with little or no dense tissue [1]. When assessed by mammography, the breast is broadly characterised by two components: fibroglandular dense tissue and fatty non-dense tissue. Percent breast density is measured as the relative proportion of dense tissue in the breast, either in terms of area or volume depending on the measurement method. Visual assessment measures percent density with respect to the total breast area (TA) whilst automated and semi-automated methods can also measure the extent of dense and fatty tissue separately. Both absolute dense area (DA) and percentage dense area (PDA) are positively associated with risk of premenopausal (and postmenopausal) breast cancer [2–4], and absolute dense volume (DV) and percentage dense volume (PDV) have also shown positive associations [5,6]. Associations of breast fat area (FA) and volume (FV) with breast cancer risk are unclear, although there is some suggestion of an inverse relationship with premenopausal breast cancer risk [4,6].

In postmenopausal women, higher attained body mass index (BMI) is associated with a higher risk of breast cancer [7–9], with an estimated 40% increase in risk for every 10 kg/m<sup>2</sup> of BMI in never users of hormone replacement therapy [9]. This increase in risk is partly explained by increased aromatisation of androgens to oestrogen in peripheral adipose tissue, which promotes cell proliferation [10,11], carcinogenesis [10,11], and insulin resistance [12]. Whilst BMI is a widely accepted risk factor for breast cancer in postmenopausal women, there may be an inverse relationship in premenopausal women [13].

Weight gain across the premenopausal years has also been linked to an increased risk of postmenopausal breast cancer. Every 5 kg of adult weight gain is associated with an approximate 10% increase in risk amongst never or low-hormone replacement therapy users [14,15]. However, a number of studies (as summarised by Hardefeldt et al. [16]) suggest that these effects are reversible with efficient weight loss [16]. In particular, weight loss in the premenopausal years has been shown to reduce postmenopausal breast cancer risk [17,18]. Risk reductions of approximately 40% have also been seen with large weight losses as a result of bariatric surgery in populations of pre- and postmenopausal women [19].

The effects of short-term weight change on breast density are less well understood, particularly those as a result of dietary weight loss. Mammographic density is a dynamic phenotype and has the potential to respond to short-term weight changes, making density reduction a possible biomarker for reduction in risk as a result of weight loss. This study aims to explore the effect of short-term dietary weight change on density using both areabased and volumetric methods in a cohort of premenopausal women to ascertain whether the relationship between weight loss and reduced postmenopausal breast cancer risk could, in part, be mediated by reductions in mammographic tissue.

#### **2. Materials and Methods**

#### *2.1. Study Design and Participants*

The Lifestyle Study is a prospective non-randomised 1 year diet and exercise weight loss intervention study amongst 79 high-risk premenopausal women attending annual screening within the Breast Cancer Family History clinic at the Prevent Breast Cancer research unit at the Manchester University Hospital Foundation NHS Trust [20–23]. Attendees of our regional Family History Clinic, aged 35–45 years, received a mailed invitation to enter either a 12-month intensive diet and exercise weight loss programme or a usual care group receiving standard written advice only, depending on their proximity to the hospital. Eligibility required women to be premenopausal with regular menstrual cycles, non-smokers, have a self-reported adult weight gain ≥ 7 kg, and a sedentary lifestyle (<40 min moderate physical activity per week). All women had a family history of breast cancer (with lifetime risk 17–40% as assessed by the Tyrer–Cuzick model [24,25]), but were excluded if they had a known *BRCA1/2* mutation or a previous history of cancer. Women were also excluded if they were already successfully dieting or losing weight, were pregnant or planning to become pregnant over the next year, had used hormonal oral contraceptives in the last six months, or had psychiatric or physical co-morbidities that could affect their ability to take part in a diet and physical activity weight loss programme.

In the intervention group (*n* = 40), women followed a 12-month intensive supervised weight loss programme involving a 25% energy-restricted Mediterranean type diet and an individualised physical activity program (150 min moderate intensity physical activity and 40 min of resistance exercise per week). The usual care group (*n* = 39) received standard written advice about diet and physical activity but no additional support for weight loss. Women provided baseline information on alcohol intake (from a 4-day food diary) and physical activity (7-day recall from an interview questionnaire) at their baseline clinic visit. All subjects gave their informed consent for inclusion before they participated in the study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the South Manchester Ethics Committee (Reference no. 01/426).

The objective of this analysis was to assess the relationship between BMI and breast density in the entire cohort of women. All participants had changing BMI measures irrespective of the type of weight loss advice they received, hence the intervention and usual care groups were combined and treated as one cohort. Furthermore, to limit the effect of women contributing observations to an area-based measure or volumetric measure only, the cohort was restricted to those with both an area and volumetric density measurement at any one or more time points (*n* = 65, 82% of the cohort).

#### *2.2. Mammographic Density*

Mammographic films were digitised using a Kodak LS85 digitiser at a pixel size of 50 μm and with 12-bits (4096 grey levels) pixel depth. The images were then anonymised and randomised to ensure the radiologists remained unaware of the time point of each mammogram. Mammograms were analysed using three different methods: (1) a semiautomated area-based measure based on computer-assisted thresholding by a single expert user (Cumulus, Sunnybrook health sciences centre, Toronto, Canada, [26]); (2) an automated volumetric Stepwedge method developed at Manchester University [27]; and (3) a visual assessment score of percentage density read to the nearest 5% by two experienced readers and expressed as an average of the two scores to calculate PDA. Cumulus was used to calculate TA, DA, FA, and PDA, and the Manchester Stepwedge method calculated total volume (TV), DV, FV, and PDV. Density assessments were made at 3 time points: baseline, 1 year follow-up (at the end of the intervention) and 1 year after the end of the intervention. Baseline mammograms were taken at the point of entry to the study; for those women with a mammogram performed within one year of entry, their most recent mammogram within the last 12 months was used. Each woman had four mammographic views taken at each time point: Left Cranial-Caudal, Right Cranial-Caudal, Left Mediolateral-Oblique, and Right Mediolateral-Oblique, and a final mammographic score at each time point

was calculated using an average of the four views. The main analysis refers to Cumulus measured area-based density and Stepwedge measured volumetric density only to assess the effects of BMI on dense and non-dense tissue separately. Visually-assessed density had similar results to Cumulus-assessed PDA, so was included as a secondary density measure only. Results for TA and TV are also reported as secondary density measures in the Supplementary Materials.

#### *2.3. Body Weight and Body Composition*

Weight, BMI, and a variety of different measures of body composition were assessed at baseline, 1 year follow-up (at the end of the intervention), and 1 year after the end of the intervention. Weight (kg) and height (m) were determined using a calibrated beam balance and stadiometer and used to calculate BMI (kg/m2). Other body composition assessments were also made such as waist circumference; total body fat, fat free mass and % body fat (assessed using a DXA whole body scanner (Hologic Inc., Bedford, MA, USA) and bioelectrical impedance (Tanita TBF-300A, Tanita Europe B.V., Hoogoorddreef 56E, 1101 BE Amsterdam, The Netherlands)); and intra-abdominal and abdominal subcutaneous area (assessed using a magnetic resonance imaging (MRI) scan with a single transverse scan taken at the level of the intervertebral disc between the L2 and L3 vertebrae). Weight, BMI, waist circumference, and total body fat, fat free mass, and % body fat (impedance) were recorded at all three time points. Intra-abdominal area, abdominal subcutaneous area, and total body fat, fat free mass, and % body fat (DXA) were only measured at baseline and at 1 year. Weight at age 20 years was self-reported via questionnaire, and BMI at age 20 years was calculated using weight at age 20 years and height at study entry. Long-term adult BMI gain was calculated as the difference between baseline BMI and BMI at age 20 years. We discuss BMI as the measure of body weight throughout the main analysis because BMI is a commonly used adjustment for density and it is a well-established risk factor for breast cancer. Other body composition measures gave similar correlations with density to those of BMI and were highly correlated with BMI. Therefore, other body composition measures are included as secondary analyses in the Supplementary Materials. Weight gain during the intervention was defined as ≥+3% of baseline weight, weight loss was defined as ≤−3% of baseline weight, and a weight change >−3% to <+3% of the baseline weight was defined as a stable weight [28].

#### *2.4. Statistical Analysis*

Data were visualised using custom-made 'tadpole plots', where each tadpole represents a woman, the head plots the woman's BMI and density at her last time point, and the points on the tail plot her BMI and density at earlier time points. Correlation (*r*) between BMI and mammographic density was assessed on a cross-sectional basis (between women), and within women as their short-term BMI changed, using repeated measures methods that use all of the measurements at the same time [29,30]. Briefly, between women correlation was a weighted Pearson correlation coefficient [30], and within women correlation was based on the decomposition of sums of squares from an analysis of variance [29]. The 95% confidence intervals were estimated using an empirical bootstrap (10,000 resamples). The simultaneous association of between and within women correlations was tested using a linear mixed model adjusted for age [31] (Appendix A). To help with comparisons across different measures of breast density, the breast density values were first standardised (Appendix B). To make density measures more symmetric and approximately normallydistributed, they were transformed: a square root transformation for area measures and a cube root transformation for volumetric measures. An exploratory analysis was undertaken to assess the effect of adding BMI gain since 20 years of age to the model. An additional exploratory analysis tested whether there was an association between breast density and DXA bone density. A sensitivity analysis assessed repeated measures correlation coefficients for BMI and density stratified by intervention group.

Analysis used the statistical software R [32]. All tests were two-sided and considered significant at the 5% level.

#### **3. Results**

Baseline characteristics of the cohort are shown in Table 1. Median age was 41 years (interquartile range (IQR), 38–43 years), and the majority of women were Caucasian (*n* = 60, 92%) and parous (*n* = 55, 85%). At baseline, 27 women (42%) were classified as overweight (BMI ≥ 25 kg/m2 and <30 kg/m2), 20 (31%) were obese (BMI ≥ 30 kg/m2), and 18 (28%) were in the normal BMI range (BMI ≥ 18.5 kg/m<sup>2</sup> and <25 kg/m2). By the end of the 2 year study period (1 year post intervention), 22 women (34%) had lost weight, 16 (25%) had gained weight, and 26 (41%) maintained their original weight. Overall, women in the intervention group lost more weight than the usual care group (mean percentage of baseline weight at 1 year = −4.4% and 0.1%, respectively; mean percentage of baseline weight at 2 years = −2.9% and 2.0%, respectively).

Median PDA, DA, and FA of each woman's average density measure over the intervention were 37.1% (IQR, 2.5%–71.3%), 59.9 cm2 (IQR, 5.8–158.4 cm2) and 107.3 cm2 (IQR, 23.6–405.1 cm2), respectively. For Stepwedge measures, PDV, DV, and FV were 22.7% (IQR, 6.7%–69.4%), 191.5 cm<sup>3</sup> (IQR, 56.7–710.4 cm3), and 573.0 cm3 (IQR, 72.8–1992.1 cm3), respectively. A flow chart detailing the availability of mammographic density measures across the intervention is shown in Figure S1 (all women had BMI available at all time-points except for one woman with missing BMI at 2 years—this data point was excluded from analyses involving BMI).

Table 2 shows the repeated measures correlations. DV was positively correlated with BMI between women (*r* = 0.41, 95%CI 0.17 to 0.61) but less so within women (*r* = 0.08, 95%CI −0.16 to 0.28). There was little association between DA and BMI (between women *r* = −0.12, 95%CI −0.38 to 0.16; within women *r* = 0.01, 95%CI −0.24 to 0.25). PDV was inversely associated with BMI between and within women (between *r* = −0.48, 95%CI −0.64 to −0.33; within *r* = −0.36, 95%CI −0.54 to −0.12), and PDA was inversely associated with BMI between women (*r* = −0.58, 95%CI −0.72 to −0.42), but less so within women (*r* = −0.22, 95%CI −0.44 to 0.01). FV and FA were positively correlated with BMI between and within women (volume: between *r* = 0.77, 95%CI 0.69 to 0.84, within *r* = 0.58, 95%CI 0.36 to 0.75; area: between *r* = 0.74, 95%CI 0.63 to 0.82, within *r* = 0.45, 95%CI 0.23 to 0.63). The magnitude of correlations was stronger between women than within women. These associations were also seen in Figure 1 when data were visually assessed using tadpole plots (trends in the tadpole heads represented the between women correlations and trends in the tadpole tails represented within women correlations).

Results for repeated measures correlation coefficients were similar when evaluated in a sensitivity analysis stratifying the cohort by intervention group. Within women associations for BMI and FA or FV were slightly stronger for women following the supervised weight loss programme compared with the usual care group, but there was little association (within women) for BMI and DA or DV in both intervention groups (Table S6).

Other body fat composition measures were highly correlated with BMI (Table S3), and the associations between breast density and other body fat compositions were similar to those with BMI (Tables S1 and S2). The correlations between various mammographic density measures are also reported in the Supplementary Materials (Table S4).


**Table 1.** Participant characteristics at study entry.

BMI: Body mass index. # N (%); \* Median (interquartile range); \*\* Mean (standard deviation) % of baseline weight (kg). <sup>a</sup> Alcohol from a 4-day food diary; <sup>b</sup> Physical activity from 7-day recall. Weight loss defined as ≤−3% of baseline weight (kg); Stable weight defined as >−3% to <+3% of baseline weight (kg); Weight gain defined as ≥+3% of baseline weight (kg).


**Table 2.** Repeated measures between women and within women correlations for mammographic density and body mass index.

VAS: Visual assessment score; PDA: percent dense area; PDV: percent dense volume; FA: fat area; FV: fat volume; DA: dense area; DV: dense volume; sqrt: square root transformed; cbrt: cube root transformed; BMI: body mass index; 95%CI: 95% confidence interval. Area-based measures from Cumulus; volumetric measures from Manchester Stepwedge. Within women correlations represent trends over the entire 2 year period.

**Figure 1.** Tadpole plots showing body mass index (BMI) and density measures across the 2 year follow-up. Each tadpole represents a woman: the tadpole head shows BMI and density (if density is available) at her last follow-up and the points on the tail show BMI and density (if density is available) at her earlier follow-ups. (**a**) Dense volume; (**b**) Fat volume; (**c**) Dense area; (**d**) Fat area.

The between and within women associations for density and BMI measures were similar when estimated jointly in an age-adjusted linear mixed model (Table 3). In a sensitivity analysis, the same model was fit using weight instead of BMI, but it had a worse model fit for almost all density measures (Table S5).

**Table 3.** Multivariable linear mixed model fit results for mammographic density on body mass index (between and within women), adjusted for age (A1).


VAS: Visual assessment score; PDA: percent dense area; PDV: percent dense volume; FA: fat area; FV: fat volume; DA: dense area; DV: dense volume; sqrt: square root transformed; cbrt: cube root transformed; BMI: body mass index; 95%CI: 95% confidence interval. Area-based measures from Cumulus; volumetric measures from Manchester Stepwedge. Between women BMI calculated as the mean BMI for each woman; within women BMI calculated as the difference between each woman's BMI and her mean BMI. Density measures are standardised (see Appendix B). One woman with missing BMI at age 20 years excluded. Within women effects represent trends over the entire 2 year period.

> When a term for BMI gain since age 20 years was added to the linear mixed model, the model fit improved for PDA, PDV, FV, and DA (all ΔLR-χ<sup>2</sup> *p* < 0.05) (Table 4). After including BMI gain since age 20 years, between women associations for BMI became more strongly inversely associated with percent density (approximately −0.5 to −0.8), more strongly positively associated with breast fat (approximately 0.6 to 0.8), more strongly inversely associated with DA (−0.1 to −0.5), and less strongly positively associated with DV (0.4 to 0.2). Within women effects of BMI on density were almost unchanged when including BMI gain since age 20 years. BMI gain from age 20 years (adjusted for attained BMI) was positively associated with DA, PDA, and PDV (5 kg/m<sup>2</sup> increase in BMI gain since age 20 years was associated with 0.61 (95%CI 0.12 to 1.09), 0.61 (95%CI 0.21 to 1.02), and 0.47 (95%CI 0.05 to 0.88) standard deviation increase in breast density (β), respectively), and inversely associated with FV (*β* = −0.31, 95%CI −0.62 to 0.00), but less association was seen with DV (*β* = 0.15, 95%CI −0.29 to 0.59) and FA (*β* = −0.32, 95%CI −0.67 to 0.03).

> Finally, in tests of association between breast and bone density, there was some indication of a positive between women correlation for bone density and FV (*r* = 0.26, 95%CI, 0.00 to 0.50), DV (*r* = 0.33, 95%CI, 0.09 to 0.54), and TV (*r* = 0.31, 95%CI, 0.06 to 0.54) (Table S1), but we found little correlation within women (Table S2).



BMI calculated as the mean BMI for each woman; within women BMI calculated as the difference between each woman's BMI and her mean BMI; BMI gain from age 20 years calculated as

the difference between each woman's BMI at baseline and her BMI at age 20 years. Density measures are standardised

excluded. ΔLR-χ2 represents the difference in likelihood ratio for A1 versus A2. Within women effects represent trends over the entire 2 year period. All variables adjusted for each other

in the multivariable

 model, therefore BMI gain since 20 years of age is adjusted for current BMI through the variable for between women BMI.

 (see Appendix B). One woman with missing BMI at age 20 years

#### **4. Discussion**

The data in this study provide some support for the two main findings. First, it is possible that the higher a premenopausal woman's BMI, the higher her breast fat and dense tissue (in particular, dense volume), and the lower her percent density. Second, the data suggested that as a premenopausal woman loses weight, her breast fat reduces, dense tissue remains relatively unchanged, and percent dense tissue increases. Effective weight loss during premenopausal years has been associated with a reduced risk of postmenopausal breast cancer [16–18], but our study data suggest that risk reduction is unlikely to be mediated by a short-term reduction in dense breast tissue. This is likely to mean that incorporation of change in percent breast density into risk algorithms will not capture potential weight loss-induced breast cancer risk reduction and may falsely ascribe an increased risk due to increased percent density. Therefore, risk prediction models need to consider how best to incorporate changes in weight and mammographic density when predicting breast cancer risk.

The between women associations of attained premenopausal BMI and breast density observed in this study were consistent with previous studies. High BMI is associated with high dense volume [33–35], but the correlation between BMI and dense area is less strong, and often close to zero [36–39]. These differences are likely to be a result of volumetric measures representing breast tissue more accurately than area-based methods by accounting for breast thickness and overlapping tissue. Additionally, since the breast is a deposit for adipose tissue, high attained BMI is strongly associated with high levels of breast fat area [36–39] and breast fat volume [33,34], which in turn leads to an inverse association between BMI and both percent dense area [36–41] and percent dense volume [33–35,42,43].

There have been very few studies to assess the effect of dietary weight loss on breast density in premenopausal women. Boyd et al. reported reductions in total and dense area alongside modest weight change within an intervention trial of women on a 2-year low-fat, high-carbohydrate diet [44]. In particular, a 5.4% decrease in dense area was seen in premenopausal women in the low-fat diet group with a 0.1kg/m<sup>2</sup> BMI reduction (*n* = 249) compared with a 2.5% decrease in the control group with a 0.3kg/m<sup>2</sup> BMI gain (*n* = 264). These reductions may be associated with the large reductions in dietary fat (55 to 35 g/day) and saturated fat (19 to 12 g/day) rather than weight loss in this study. This was considerably higher than those advised and achieved in the current reported study (total fat reduced from 77 to 60 g/day and saturated fat reduced from 28 to 21 g/day). Other trials have also assessed the effect of lifestyle interventions for weight loss on breast density, although in postmenopausal women only. In the ALPHA trial, postmenopausal women on a 1-year aerobic exercise intervention lost on average 39 cm3 more breast fat than the controls, but there was little difference in the change in dense tissue between the two groups [45]. Furthermore, the DAMA trial reported a reduction in volumetric percent density of approximately 14% for postmenopausal women following a 2-year diet or exercise intervention when compared with the controls [46]. Large weight loss with bariatric surgery is also associated with large reductions in breast fat alongside relatively smaller reductions in dense tissue, and an increase in percent density [47–49].

As an exploratory analysis, we also found an association between increased BMI gain since 20 years of age and higher dense tissue and percent density. It is possible that this is a pathway for the increased risk of postmenopausal breast cancer seen with adult weight gain [7,15,50–52]. However, this association is likely to reflect the inverse association seen in previous studies between adolescent body adiposity and dense tissue in later life [38,40,53–55], since, given the adjustment for current BMI, women with greater gain in BMI will have had lower BMI at 20 years of age. This interesting observation requires further investigation in larger datasets of women. Additionally, exploratory analysis of bone density found little association with breast density, which is in agreement with previous studies [56].

Strengths of this study include the various measures of breast density including Cumulus and the Stepwedge method, which allowed for the assessment of dense and fatty

tissue separately as well as various measures of body weight to assess adiposity. The study also assessed breast density as an area-based measure and volumetrically; both of which have similar abilities for breast cancer risk prediction [57]. Additionally, all women were encouraged to lose weight, which produced data with large within women variation in BMI, in turn increasing the potential to see an effect of changing BMI on mammographic density. Furthermore, the Lifestyle Study provided a data source to assess premenopausal weight loss and density associations; something that is not possible in studies involving routine screening data. This also provided a greater ability to capture the effects of weight loss on density because this cohort of premenopausal women were likely to have had higher dense tissue at baseline (with greater ability to decrease) than screening populations involving postmenopausal women [58]. Finally, the use of repeated measures over a 2-year period allowed us to assess the association between BMI and breast density longitudinally, whilst making use of all available data simultaneously.

Limitations of the study include the small sample size, which limits statistical power. This is particularly relevant for volumetric measures, which had a moderate amount of missing data at the baseline. In addition, the study design was not powered for the analysis of mammographic density, which was a secondary analysis (the study was powered for salivary oestradiol). This was a relatively small study, and ideally, a larger study with sufficient power would be run to verify our evidence. Another limitation is the analysis of BMI gain since 20 years of age relies on self-reported information on weight at age 20 years. This may be less accurate than the measured values. However, validation studies show that self-reported BMI is highly correlated with independently measured BMI, and the mean difference between self-reported and measured weight is minimal [59,60]. Finally, breast thickness is likely to have changed whilst women lost weight during the intervention. Volumetric measures are influenced by breast thickness [61], hence there might have been larger variation in the serial compared with stable volumetric measurements, resulting in reduced ability to capture the within women effects of BMI on dense tissue volumetrically.

#### **5. Conclusions**

This study suggests that premenopausal weight loss reduces breast fat but that it does not reduce dense tissue. Short-term premenopausal weight loss is likely to be linked to lower postmenopausal breast cancer risk through reductions in adipose tissue, not fibroglandular tissue. This means that a potential breast cancer risk reduction as a result of weight loss might not be captured by changes in breast density, and the resulting increase in percent density may falsely ascribe an increase in risk. However, the study was limited by the small sample size, and more studies are required to provide evidence to confirm these results.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/cancers13133245/s1, Table S1: Complete results for repeated measures between women correlations for mammographic density and body composition measures; Table S2: Complete results for repeated measures within women correlations for mammographic density and body composition measures; Table S3: Complete results for repeated measures between women correlations for different body composition measures; Table S4: Complete results for repeated measures between women correlations for different mammographic density measures; Table S5: Multivariable linear mixed model fit results for A1 using either body mass index or weight; Table S6: Repeated measures between women and within women correlations for mammographic density and body mass index, stratified by intervention group; Figure S1: Flow chart of women included in the analysis and availability of mammographic density data.

**Author Contributions:** Conceptualization, A.H. and M.H.; Formal Analysis, E.C.A. and A.R.B.; Investigation, S.A., J.C., D.G.E., R.M.L.W., A.H., and M.H.; Data Curation, S.A., D.G.E., R.M.L.W., A.H., and M.H.; Writing—Original Draft Preparation, E.C.A., A.R.B., S.A., J.C., D.G.E., R.M.L.W., A.H., and M.H.; Writing—Review & Editing, E.C.A., A.R.B., S.A., J.C., D.G.E., R.M.L.W., A.H., and M.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** E.C.A., A.R.B. and J.C. are supported by Cancer Research UK (C569/A16891 to J.C.); S.A., D.G.E., A.H., and M.H. are supported by the NIHR Manchester Biomedical Research Centre (IS-BRC-1215-20007 to D.G.E.). The funders had no role in the design of the study; in the collection, analysis and interpretation of data; in the writing of the manuscript, or in the decision to submit for publication.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki and was approved by the South Manchester Ethics Committee on 28 January 2002 (Reference no 01/426).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The dataset used and analysed during the current study is available from the corresponding author on reasonable request.

**Acknowledgments:** We thank Hilary Graff for contributions to the collection and assembly of data (University of Manchester). We are particularly grateful to the women who participated in this study and the entire medical and administrative staff who worked on the Lifestyle Study.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **Abbreviations**


#### **Appendix A**

Linear mixed model for mammographic density on body mass index and age. A linear mixed model was used to model density and body mass index (BMI) associations in Table 3. This model allows for repeated measures and uses all of the available data (missing pairs of density and BMI are excluded). Breast density *yij* for woman *i* = 1, ... , *n* at time *j* = {1, 2, 3} is modelled as:

$$y\_{i\rangle} = \alpha + \beta \text{ag}\mathbf{e}\_{i\rangle} + \gamma \overline{\mathbf{x}}\_{i.} + \delta \left(\mathbf{x}\_{i\rangle} - \overline{\mathbf{x}}\_{i.}\right) + u\_i + \mathbf{e}\_{i\rangle};\tag{A1}$$

where *α* is an overall intercept; *ageij* is the age at baseline for woman *i* at time *j*; *β* is the slope for age; *xi*. is mean BMI for woman *i*; *γ* is the between women slope; *xij* is the BMI of woman *i* at time *j*; *δ* is the within women slope; and *eij* is an independent random error. Another term that allows for differences between women in their overall density level is the independent random intercept *ui* for woman *i*. The model is completed by assuming normal distributions for *ui* and *eij* with zero mean, unknown variances, and zero covariance. The model was fitted by maximum likelihood. To aid interpretation of the estimates across different measures of density, the density values were standardised (see Appendix B). To test *γ* = 0 (between women correlation) and *δ* = 0 (within women correlation), a Wald test was applied.

The model was extended to consider BMI gain from age 20 years in Table 4:

$$y\_{i\bar{j}} = \mathfrak{a} + \beta \text{age}\_{i\bar{j}} + \gamma \overline{\mathfrak{x}}\_{i.} + \delta \left( x\_{i\bar{j}} - \overline{\mathfrak{x}}\_{i.} \right) + u\_{i} + \varepsilon z\_{i} + \varepsilon\_{i\bar{j}};\tag{A2}$$

where *zi* is the BMI gain since age 20 years for woman *i*: calculated as the difference between baseline BMI for woman *i* and BMI at age 20 years for woman *i*, and *ε* is the slope for BMI gain since age 20 years. To test *ε* = 0, a Wald test was applied.

#### **Appendix B**

Standardisation of each mammographic density measure:

$$
\overline{\underline{x}} = \frac{\sum\_{i=1}^{n} \overline{x}\_i}{n}
$$

$$
\sigma = \sqrt{\frac{\sum\_{i=1}^{n} \left(\overline{x}\_i - \overline{x}\right)^2}{n-1}}
$$

$$
z\_{ij} = \frac{d\_{ij} - \overline{x}}{\sigma}
$$

where *xi* is the mean density for woman *i* = 1, ... , *n*; *dij* is the density measure for woman *i* = 1, . . . , *n* at time point *j* = {1, 2, 3}; and *zij* is the standardised density measure for woman *i* at time point *j*.

#### **References**


## **Implementation of Multigene Germline and Parallel Somatic Genetic Testing in Epithelial Ovarian Cancer: SIGNPOST Study**

**Dhivya Chandrasekaran 1,2, Monika Sobocan 1,2,3, Oleg Blyuss 4,5,6, Rowan E. Miller 7, Olivia Evans 1, Shanthini M. Crusz 7, Tina Mills-Baldock 8, Li Sun 1,9, Rory F. L. Hammond 10, Faiza Gaba 1, Lucy A. Jenkins 11, Munaza Ahmed 11, Ajith Kumar 11, Arjun Jeyarajah 2, Alexandra C. Lawrence 2, Elly Brockbank 2, Saurabh Phadnis 2, Mary Quigley 8, Fatima El Khouly 8, Rekha Wuntakal 12, Asma Faruqi 10, Giorgia Trevisan 10, Laura Casey 10, George J. Burghel 13, Helene Schlecht 13, Michael Bulman 13, Philip Smith 13, Naomi L. Bowers 13, Rosa Legood 9, Michelle Lockley 14, Andrew Wallace 13, Naveena Singh 10, D. Gareth Evans <sup>13</sup> and Ranjit Manchanda 1,2,9,\***

	- Sechenov First Moscow State Medical University, Moscow 119991, Russia
	- r.f.l.hammond@smd14.qmul.ac.uk (R.F.L.H.); asma.faruqi@nhs.net (A.F.); giorgia.trevisan1@nhs.net (G.T.); laura.casey5@nhs.net (L.C.); naveenasingh7@gmail.com (N.S.)

**Simple Summary:** Multigene testing in ovarian cancer has received increased support due to its' applicability for cancer treatment and the impact it has on cancer prevention in families. This study shows that multi-gene germline and somatic testing uptake after counselling by a member of the multidisciplinary cancer clinical team in women with ovarian cancer, was high (97%). A total of 15.5% of women were identified to have germline *BRCA1/BRCA2* pathogenic variants and 7.8% had somatic *BRCA1/BRCA2* pathogenic variants. A total of 2.3% patients had *RAD51C/RAD51D/BRIP1* pathogenic variants. We found that 11% of germline pathogenic variants were large-genomic-rearrangements

309

Academic Editor: Kylie Gorringe

Received: 8 July 2021 Accepted: 24 August 2021 Published: 27 August 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

and were missed by somatic testing. Our findings support prospective parallel somatic-&-germline panel testing to maximize variant identification.

**Abstract:** We present findings of a cancer multidisciplinary-team (MDT) coordinated mainstreaming pathway of unselected 5-panel germline *BRCA1/BRCA2/RAD51C/RAD51D/BRIP1* and parallel somatic *BRCA1/BRCA2* testing in all women with epithelial-OC and highlight the discordance between germline and somatic testing strategies across two cancer centres. Patients were counselled and consented by a cancer MDT member. The uptake of parallel multi-gene germline and somatic testing was 97.7%. Counselling by clinical-nurse-specialist more frequently needed >1 consultation (53.6% (30/56)) compared to a medical (15.0% (21/137)) or surgical oncologist (15.3% (17/110)) (*p* < 0.001). The median age was 54 (IQR = 51–62) years in germline pathogenic-variant (PV) versus 61 (IQR = 51–71) in *BRCA* wild-type (*p* = 0.001). There was no significant difference in distribution of PVs by ethnicity, stage, surgery timing or resection status. A total of 15.5% germline and 7.8% somatic *BRCA1/BRCA2* PVs were identified. A total of 2.3% patients had *RAD51C/RAD51D/BRIP1* PVs. A total of 11% germline PVs were large-genomic-rearrangements and missed by somatic testing. A total of 20% germline PVs are missed by somatic first *BRCA*-testing approach and 55.6% germline PVs missed by family history ascertainment. The somatic testing failure rate is higher (23%) for patients undergoing diagnostic biopsies. Our findings favour a prospective parallel somatic and germline panel testing approach as a clinically efficient strategy to maximise variant identification. UK Genomics test-directory criteria should be expanded to include a panel of OC genes.

**Keywords:** ovarian cancer; *BRCA*; genetic testing; germline; somatic; *RAD51C*; *RAD51D*; *BRIP1*

#### **1. Introduction**

Ovarian cancer (OC) is the leading cause of deaths from gynaecological cancers, with 240,000 new cases and 152,000 deaths occurring worldwide annually [1]. GLOBOCAN data suggest the number of cases from OC will increase by 26% in the UK and 47% worldwide, respectively, over the next 20 years [1]. Standard treatment approaches have been associated with limited long-term OC survival of ~30% [2]. However, the progress over the last 10–15 years has provided the foundations for a precision medicine [3] approach for OC management, involving inherited cancer susceptibility genes. *BRCA1/BRCA2* pathogenic and likely pathogenic variants (henceforth termed 'pathogenic variants' or 'PVs') account for most of the known inheritable risk of OC. Around 11–18% of OC have germline *BRCA1/BRCA2* PV and another 6–9% have a somatic *BRCA1/BRCA2* PV in the tumour tissue alone which is not inherited. Women with germline *BRCA1/BRCA2* PVs have a cumulative risk by age 80 of 17–44% for developing EOC and 69–72% for developing breast cancer (BC) [4].

Genetic testing for OC susceptibility genes has recently received an impetus through increasing applicability for cancer treatment and eligibility for clinical trials. The proteins coded by *BRCA1/BRCA2* are essential in the homologous recombination repair (HRR) of double stranded DNA breaks, whilst PARP (poly ADP ribose polymerase) is an essential component of single-strand DNA repair. Inhibition of PARP increases double strand breaks and prevents HRR deficient (HRD) tumour cells from surviving chemotherapy induced DNA damage, leading to synthetic lethality [5]. Germline as well as somatic *BRCA* mutated OC have been shown to benefit from 'PARP inhibitor' (PARP-i) therapy with improved progression free survival at both recurrent and more recently primary settings [5–9]. Therefore, knowledge of *BRCA* status at the time of diagnosis has become pivotal in the guidance of treatment options. Genetic testing for germline *BRCA1/BRCA2* PVs in EOC was commissioned by NHS-England in 2015 [10], and has been recommended by other published guidelines over the last few years [11]. More recently, the American Society of Clinical Oncology (ASCO) [12], the British Gynaecological Cancer Society (BGCS) [13] and

the European Society of Medical Oncology (ESMO) [14] have advocated for somatic testing too.

However, HRD can arise through somatic and germline PV in a wide range of OC susceptibility genes [15]. Approximately 50% of high-grade serous OC are characterised by HRD suggesting additional mechanisms other than *BRCA* mutations play a significant role [14]. HRD assays are now available and are beginning to be used in clinical practice [14]. Further moderate risk OC susceptibility genes in the HRR pathway, such as, *RAD51C, RAD51D* and *BRIP1* with lifetime OC-risks of 5.8 to 13% have been identified and their risks validated [16,17]. Testing for additional genes of clinical utility [18] can lead to wider therapeutic benefit. ASCO now recommends germline *BRCA* testing within the context of a multigene panel [12]. In addition to targeted therapy, identification of PVs offers opportunities for cancer surveillance and prevention for secondary cancers in index patients as well as cascade testing in relatives. Unaffected relatives with PVs can access relevant surgical prevention and screening options which have well established clinical benefit. This includes risk-reducing salpingo-oophorectomy (RRSO) to reduce their OC risk [19,20]; MRI/mammography screening, or risk reducing mastectomy (RRM) [21], or chemoprevention with selective oestrogen receptor modulators (SERM) to reduce their BC risk [22].

Over recent years, many models of care delivery for OC genetic testing have been implemented into clinical practice [23–25]. There has been great variation in these clinical pathways, with strategies varying with respect to (a) whom to test (unselected or restricted by histology such as for high-grade serous OC or restricted by age, such as under 70 years); (b) what to test (either germline only, or somatic only, or both) and (c) in which order to test (parallel or sequential); (d) which genes to test (*BRCA* only or multiple genes); and (e) who provides counselling and testing (genetics teams in genetics clinics, genetics professional embedded in oncology clinics, medical oncologists, surgical oncologists, or clinical nurse specialists (CNS)). Despite guidelines, historically, the overall uptake and access to genetic testing across health systems has remained poor, with only 20–30% eligible patients accessing testing [26,27]. Obstacles to introducing routine somatic testing at diagnosis have been attributed to reasons like cost, access/availability of validated somatic testing in a National Health Service (NHS) accredited laboratory and additional resources required to process tumour samples [28]. Most studies to date report clinical experience of implementing *BRCA* testing. Reports of systematic prospective parallel germline panel and somatic genetic testing are limited. We present our experience and findings of implementing a cancer multidisciplinary team (MDT) coordinated mainstreaming pathway of unselected 5-panel germline *BRCA1, BRCA2, RAD51C, RAD51D, BRIP1* and parallel somatic *BRCA1/BRCA2* testing in all women with high grade non-mucinous epithelial OC in the Systematic Genetic Testing for Personalised Ovarian Cancer Therapy (SIGNPOST) study (ISRCTN: 16988857) in women from North East London Cancer Network (NELCN). We report on the somatic testing success rates with different types of sample ascertainment. Moreover, importantly we highlight the discordance between germline and somatic testing strategies incorporating testing data from NELCN as well as the Manchester NHS Foundation trust.

#### **2. Materials and Methods**

#### *2.1. Pre-Test Counselling and Recruitment*

Women ≥18 years with high-grade non-mucinous epithelial OC, who were newly diagnosed or under follow-up in the NELCN, were offered parallel germline testing for *BRCA1, BRCA2, RAD51C, RAD51D, BRIP1* genes and concomitant *BRCA1/BRCA2* somatic genetic testing. This was undertaken through the SIGNPOST study (ISRCTN: 16988857). Newly diagnosed patients were identified from gynaecological oncology MDT meetings and consented for genetic testing during their primary treatment. Patients undergoing surveillance post-treatment, were identified through follow-up surgical and medical oncology clinics as well as pathology and clinical databases. Eligibility for genetic testing was established by the treating clinician. Patients received written pre-test education

information regarding the advantages, disadvantages and implications of genetic-testing. Pre-test genetic counselling and consent was undertaken at routine clinic visits. This was led initially by medical and surgical oncology consultants, and subsequently also undertaken by cancer CNSs. Psychological support was offered by CNSs within the cancer services.

#### *2.2. Germline and Somatic Testing*

Testing was undertaken by clinically accredited NHS laboratories. A 4 mL EDTA blood sample was taken for germline genetic testing for *BRCA1, BRCA2, RAD51C, RAD51D* and *BRIP1*. Germline testing for NELCN samples was undertaken for *BRCA1, BRCA2, RAD51C, RAD51D* and *BRIP1* at the North East Thames Regional Genomics Laboratory (Great Ormond Street Hospital), while for Manchester samples testing for *BRCA1* and *BRCA2* was undertaken at the Genomic Diagnostic Laboratory at the North West Genomic Laboratory Hub. This was carried out using next generation sequencing (NGS; Agilent SureSelect and Illumina NextSeq) of the coding region, sequenced to a minimum depth of 30 reads, including intron/exon splice boundaries. Sanger sequencing was also carried out to confirm variants detected during the NGS screen. Additionally, exon deletions/duplications in *BRCA1* and *BRCA2* genes were detected using Exome Depth. Multiplex ligation-dependent probe amplification (MLPA; MRC Holland) kits P002-D1 and P090-C1, respectively.

Somatic testing was undertaken using formalin fixed paraffin embedded (FFPE) tissue specimen from diagnostic biopsies, or up front cytoreductive surgery or postchemotherapy cytoreductive surgery as appropriate. FFPE blocks were reviewed by a consultant histopathologist to identify areas with >20% tumour content and therefore deemed suitable for somatic testing. The specimens were processed and sent as either 5 × 5 μM thick unstained sections, or as 3 mm core biopsies from paraffin blocks. Unstained slides were preferred for small volume diagnostic biopsies and in <20% neoplastic content. Tumour blocks were selected by the pathologist and graded as <20%, 20–50% and >50% neoplastic content. Testing was undertaken in two NHS accredited diagnostic laboratories. Majority NELCN and Manchester samples were analysed at the Manchester Genomics Laboratory while a few NELCN samples were also tested at the Royal Marsden Hospital laboratory. Detection of variants is dependent on the percentage of tumour infiltration, DNA input concentration and DNA quality. DNA extracted from FFPE tissue was analysed in the coding regions of *BRCA1* and *BRCA2*, using NGS and minimum variant allele depth was 10×. The analysis was performed with Molecular Diagnostics Information Management System v-4.0, based on genome hg19 or GeneRead DNAseq v2 Human Breast Cancer Panel (Qiagen) and Illumina NGS. Mutation and variant calling by custom bioinformatic analysis pipeline validated to detect SNVs and small insertion/deletion mutations (<40 bp) to 5% mutant allele frequency (MAF).

Variants were classified using the ACGS and CanVIG guidance in force (https:// www.acgs.uk.com/quality/best-practice-guidelines/ (accessed on 5 January 2021)) [29,30]. Common, high frequency benign and likely benign variants were filtered bioinformatically from a curated list of variants whilst all other variants were assessed by a registered Clinical Scientist. In case of discordance between the germline and somatic samples, a further repeat analysis was undertaken and second report issued. Reports from both germline and somatic tests were sent to the referring clinician for disclosure to the patients.

Validation of 3 mm FFPE punch biopsies for high-volume somatic testing:

Somatic testing using NGS on FFPE specimens has been validated on 5 × 5 μM thick unstained sections. [31] In order to minimise delay without compromising DNA yield, particularly for archival FFPE tissue, 3 mm punch biopsies from FFPE tumour blocks were validated for diagnostic somatic testing. Following review by a gynaecological oncology histopathologist, a 5 mm area with high tumour content (>20%) was marked on the Haematoxylin and Eosin (H&E) stain slide. Keyes punch biopsy (routinely used for skin biopsy) was used to core out 3 mm sample from corresponding area in FFPE block. Five 5 μM thick unstained sections were also cut from same block. Five matched 3 mm cores and unstained sections were compared for DNA yield.

#### *2.3. Test Result Management*

Most patients including all those diagnosed with a PV were given their test result and counselled in an outpatient clinic by their consenting and treating cancer clinician. A small proportion of patients on long-term follow up declined an additional hospital visit and were given the result by post. All patients with a PV were referred to North East Thames regional genetics service team for additional post-test genetic counselling and facilitating predictive testing in family members.

We report on testing undertaken between 01/05/2017 to 31/12/2019 across the NELCN, which provides cancer care to a ~1.7 M population covering six NHS hospitals. Patient demographic and clinical data were extracted from electronic patient records, and FH questionnaires completed by the patient. Positive (or strong) FH was defined as any index case of high-grade non-mucinous epithelial OC and breast cancer or epithelial OC in a first-degree or second-degree relative. Patients who had previously undergone genetic testing as they had been referred to clinical genetics in view of a strong FH, were excluded from mainstreaming, but are included in the analysis of prevalence estimates. For the analysis of discordance between germline and somatic *BRCA1/BRCA2* testing we also include data of 116 unselected OC cases from Manchester NHS Foundation trust who underwent parallel germline and somatic testing. The testing procedures and offer of testing was similarly undertaken in Manchester but germline testing was restricted to *BRCA1* and *BRCA2*.

Descriptive statistics were used for baseline characteristics. PV and wild type groups were compared for ethnicity, age, FH, histology, stage, timing of surgery, chemotherapy response score, and residual disease status. Variables associated with number of pre-test consultations (1 or >1) were explored for type of clinician undertaking counselling, disease status at time of counselling (new diagnosis or on follow up) and treatment status (whether undergoing active treatment or not).

Wilcoxon rank-sum test and Fisher's exact or Chi-square tests were used to test the difference in means and proportions correspondingly. Two-sided *p*-values were reported for all statistical tests. Statistical analysis was undertaken in R version 3.5.1 and SPSS version 26.

#### **3. Results**

#### *Pathway Development*

Development of the genetic testing pathway was preceded by a wide consultation with the regional clinical geneticists, genetic counsellors, surgical and medical oncologists, CNS, clinical scientists from genetic laboratories, patient representatives and *BRCA* charity leads. Patient representatives and charity leads expressed a preference for genetic testing to be provided at diagnosis, to be made available all patients including those remained under surveillance post-treatment, and for provision for adequate pre-test counselling and informed consent.

In preparation of a cancer MDT coordinated mainstreaming genetic testing service, all gynaecological cancer MDT members (surgical oncologists, medical oncologist, pathologist and CNS) attended small group teaching sessions led by the regional lead in clinical genetics and a gynaecological oncologist with a long-standing special interest and significant experience in cancer genetics, counselling and testing. This covered principles of Mendelian inheritance, OC susceptibility genes and associated cancer risks; the principles, structure and factors specific to genetic counselling; as well as the developed local testing and referral pathways. Knowledge questionnaires were completed by attendees to ensure appropriate understanding of issues. Following pathway implementation, ongoing professional support for the cancer MDT team was provided by gynaecological cancer precision prevention service, with support from the regional clinical genetics team. Pre-counselling

written information was developed in collaboration with the major stakeholders and provided to all patients. Additionally, service management meetings across the broader group with representation from medical and surgical oncologist, lead clinical geneticist, clinical scientists from genetic laboratories, lead histopathologist were held every 6–9 months.

Counselling, Recruitment and Genetic Testing:

A total of 310 patients with high-grade non-mucinous epithelial OC who were eligible for genetic testing were identified across the NELCN. This included 188 newly diagnosed women and 122 patients on follow up post-treatment. Of these women seven were excluded: four died prior to commencing treatment, one was unable to consent due to dementia and learning difficulties and two declined genetic testing. The remainder 303 untested patients remained eligible for testing and received pre-test genetic counselling. Of these patients 7/122 (6%) under surveillance had previously undergone germline *BRCA1/BRCA2* mutation testing through clinical genetics due to a strong FH of BC or OC fulfilling prior standard clinical criteria for genetic testing. They were offered and underwent extended panel testing for *RAD51C, RAD51D* and *BRIP1* along-with somatic testing. Overall, we found a 97.7% uptake of parallel multi-gene germline and somatic testing via the cancer MDT mediated mainstreaming pathway.

All of the patients were counselled and consented by a member of the cancer MDT, with 45% (*n* = 137) by a medical oncology member, 36% (*n* = 110) by a surgical oncology member and 18% (*n* = 56) by a CNS. The majority required a single pre-test consultation (78%) prior to consenting, whereas 18% (*n* = 54) required two consultations, 4% (*n* = 13) required three and one patient required four consultations prior to decision to undergo testing (Table 1). The number of pre-test counselling sessions needed varied significantly depending on the clinical professional undertaking counselling. Counselling by CNS was more frequently associated with needing more than one consultation (53.6% (30/56)) compared to counselling by a medical oncologist (15.0% (21/137)) or a surgical gynaeoncologist (15% (17/110)) (*p* < 0.001). The number of consultations required did not significantly differ whether (a) the patient was newly diagnosed or under follow up; and (b) if they were undergoing active treatment or not (Table 1).


**Table 1.** Factors associated with number of pre-test consultations.

\* Chi-square test comparing '1 consultation and >1 consultation groups' by variables of type of counselling clinician, disease status and treatment status at time of pre-test counselling.

Patient demographics and clinical characteristics are summarised in Table 2. The median age at OC diagnosis was 54 years (IQR 51–62) in germline PV compared with 61 (IQR 51–71) in *BRCA* wild type (*BRCA*-WT) (*p* = 0.001) patients. In germline *BRCA1/BRCA2*/ *RAD51C/RAD51D/BRIP1* PVs, 44.4% (24/54) had a positive FH compared to 11.3% (28/249) of sporadic tumours (*p* < 0.001) (Table 2). Thus 55.6% of PVs would have been missed by

using FH alone. Only 2/7 of *RAD51C/RAD51D/BRIP1* PVs had a positive FH. Ethnicity of OC cases included 196 (64.7%) White, 28 (9.2%) Black, 52 (17.2%) South Asian and 27 (8.9%) were classed as 'other'. In women with somatic *BRCA1/BRCA2* PV, the median age at diagnosis was 61 (IQR 59–66) and 13% (2/15) had a positive FH. Most PVs had a highgrade serous (HGS) histology except one *BRCA1* with grade 3 endometrioid carcinoma and one *BRIP1* with mixed epithelial adenocarcinoma. There was no significant difference in distribution of PVs by ethnicity, stage at diagnosis, timing of surgery or resection status (Table 2). In post-chemotherapy cytoreductive surgery specimens, chemotherapy response score (CRS) of 3 (minimal residual disease) was recorded in 13/69 (18.8%) germline and somatic PVs compared to 13/234 (5.6%) of *BRCA*-WT tumours (*p* = 0.025).


**Table 2.** Demographic and clinical characteristics NELCN cohort.


**Table 2.** *Cont.*

Pathogenic variants = class 4/5 variant in BRCA1, BRCA2, RAD51C, RAD51D, BRIP1. Family history positive = first-degree or second degree relative with ovary and/or breast cancer. HGSC = high grade serous carcinoma. Early stage = stage 1–2; advanced stage = stage 3–4. R0 = zero or nil residual disease, R1 = ≤1 cm residual disease, R2 = >1 cm residual disease. IQR = inter quartile range, PV = Pathogenic variants, VUS = Variants of uncertain significance, LGR- large genomic rearrangements. This table describes outcomes by two groups: (a) with and (b) without germline/somatic pathogenic variants. Two-sided *p*-values were reported for statistical tests comparing these two groups \* Results of somatic testing at time of analysis for 71 patients were unavailable (only 232 patients had paired samples). Of these 71 patients 9 had a germline PV.

Validation of 3 mm FFPE punch biopsies for somatic testing:

Analysis of 3 mm Keyes punch biopsy and 5 × 5 μM unstained sections from the same FFPE tumour block demonstrated comparable DNA concentration and yield; therefore, archived tumour samples of patients under follow-up were processed as 3 mm core which proved time-efficient, as it reduced consultant pathologist time needed for review, retrieval and marking of slides. This is therefore likely to be more cost-efficient (Table 3).

**Table 3.** Comparison of DNA concentration and yield from FFPE 3 mm core and unstained sections of tumour tissue.


Table 3 describes the validation data of DNA yield from FFPE 3 mm core biopsies and unstained sections of tumour tissue.

Tumour testing results were available for 232 NELCN cases. Of the 71 cases without tumour testing results, 40 cases lacked available archived tumour tissue for analysis (unable to retrieve from pathology archive or surgery at another cancer centre); and 25 archived cases lacked any tissue with adequate neoplastic content (minimal diagnostic biopsy

or post-chemotherapy tumour necrosis leaving no viable sample for analysis); and six test results were awaited at the time of analysis (delays due to COVID pandemic). Of these 71 cases without a somatic result, nine had a PV on germline genetic testing (four *BRCA1*, three *BRCA2*, one *RAD51C*, one *RAD51D*). Of the 232 NELCN tumour samples that underwent testing, 19 (8.9%) failed analysis due to fragmented DNA or low neoplastic content. Of these failed 19 cases, one had a *BRCA1* PV and one a *RAD51D* PV on germline testing. Further details on tumour tissue processing are provided in Table 4. The failure rate was higher for diagnostic biopsies (22.9%; 11/48) compared to primary cytoreductive surgical specimens (5.4%; 6/110) and post-chemotherapy surgical specimens (2.7%; 2/74). Primary-surgery specimens that failed analysis were due to fragmented DNA. There were 11 (out of 232) samples categorised with <20% neoplastic content, of which five (45%) were subsequently found to be adequate for analysis (Table 4). A majority of the samples were sent for analysis as 3 mm core biopsies from paraffin blocks (174/232, 75%) and the rest as unstained slides (58/232, 25%). Failure rates were 3/174 (1.7%) in 3 mm cores and 16/58 (27.6%) in unstained slides, respectively. However, 6/16 failed analysis in the unstained slides group had <20% neoplastic content. In our centre, tissue was preferentially sent as unstained slides if neoplastic content was <20% or the sample was a small volume diagnostic biopsy.


Primary surgery 104/110 (94.5%) 6/110 (5.5%)

surgery 72/74 (97.3%) 2/74 (2.7%) **Type of tumour sample** 3 mm core from FFPE 171/174 (98.3%) 3/174 (1.7%) 5 × 5 μM unstained slides 42/58 (72.4%) 16/58 (27.6%) **Neoplastic content** <20% 5/11 (45.5%) 6/11 (54.5%) 20–50% 33/40 (82.5%) 7/40 (17.5%)

**Table 4.** NELCN tumour tissue *BRCA1/BRCA2* next generation sequencing analysis.

>50% 175/181 (96.7%) 6/181 (3.4%) This table describes the results of BRCA testing of tumour tissue in the NELCN cohort. Results are available for 232 cases. \* Of the 19 failed analysis, one had a BRCA1 PV and one a RAD51D PV.

#### Genetic testing results:

Post-chemo cytoreductive

Following multi-gene germline testing, 54 germline PVs were identified in 303 women from the NELCN cohort (Supplementary Table S1). Of these PVs, 33 (11%) were *BRCA1*; 14 (4.6%) *BRCA2*, 2 (0.7%) *RAD51C*, 3 (1.0%) *RAD51D* and 2 (0.7%) *BRIP1*). Six PVs were large genomic rearrangements (LGR) and detected by MLPA: four in *BRCA1*, one in *BRCA2* and one in *RAD51C*. The germline VUS rate in *BRCA1/BRCA2* was 3.3% (*n* = 10) and 3.3% (*n* = 10) in *RAD51C/RAD51D* and *BRIP1* (Table 5). Germline *BRCA1/BRCA2* testing in the Manchester cases identified 11 (9.5%) PVs, of which 8 (6.9%) were *BRCA1* and 3 (2.6%) were *BRCA2* PVs (Supplementary Table S1). Additionally, one *BRCA1* VUS was identified. The median age of the Manchester cohort was 63 years (IQR = 55–72). Overall, 14 Manchester patients had a strong FH of cancer. Four of the eleven germline PV had a strong FH, while seven lacked a strong FH and would have been missed without unselected testing. Combining data from NELCN and Manchester series, the total *BRCA1/BRCA2* germline PV rate was 15.5% (65/419) and *BRCA1/BRCA2* germline VUS rate was 2.6% (11/419).


**Table 5.** Mutation Prevalence (Manchester cohort).

This table describes the prevalence of variants in the Manchester cohort. VUS—variants of uncertain significance. PV—pathogenic variants. LGR—Large genomic rearrangements.

A total of 232 tumour *BRCA1/BRCA2* results were available at the time of analysis from NELCN cases. Somatic *BRCA1/BRCA2* PVs were detected in 15 (6.6%) cases and the VUS rate was 2.2% (*n* = 5). Tumour *BRCA1/BRCA2* testing in 116 Manchester cases identified 7 (6%) *BRCA1* and 5 (4.3%) *BRCA2* somatic PVs as well as 1 (0.9%) *BRCA1* and 1 (0.9%) *BRCA2* somatic VUS each (Table 5). The total *BRCA1/BRCA2* somatic PV rate was 7.8% (27/348) and somatic VUS rate was 2% (7/348). A germline or somatic PV was identified in 22% (92/419) patients overall. The list of all the variants identified are detailed in Supplementary Table S1. PARP-i treatment was commenced in 49 (16%) NELCN women (27 following primary treatment and 22 following recurrence).

*BRCA1/BRCA2* germline and somatic PV concordance:

Concordance of *BRCA1/BRCA2* PV identified through germline and tumour testing was explored. This included 232 paired samples with results from NELCN and 116 paired samples with results from Manchester NHS Trust. There were six *BRCA1/BRCA2* PVs that showed discordance between germline and tumour testing, five in the NELCN cases and one from the Manchester cases, comprising 10.3% of all germline PVs. Five of these six *BRCA1/BRCA2* PVs were LGR that were not detected on somatic testing; one (3%) germline mutation (from NELCN cases) was initially reported in the somatic report but not in the germline. This mutation was then subsequently identified in the germline following reanalysis of the germline sample. The inability of routine somatic testing to reliably identify LGRs is an important finding with implications for those developing and/or implementing OC mainstreaming pathways and for those whose pathways currently use a somatic testing first triage mechanism. It is critical that patients with LGRs are not missed both from a cancer treatment perspective as well as for precision prevention in unaffected relatives with a PV identified through cascade testing.

Pathway improvements:

Changes to the NELCN pathway were incorporated over time to improve logistic efficiencies, communication between team members and timely communication of result to the patient. These included: agreement on a standardised format for reports received from genomic laboratories and omitting of reporting class-1 and class-2 variants. This improved interpretability by cancer clinicians and reduced unnecessary distress in patients.

Initially somatic reports were uploaded as supplementary reports to the original histology result but this caused delays in clinician receiving the information and communicating this to the patient. This was addressed by results being directly sent from the genomic laboratory creating to a shared email-box which was accessed by all members of the clinical team. Responsibility for monitoring and ensuring all results were actioned was subsequently undertaken by the lead medical oncologist.

Electronic communication with electronic request forms being sent directly to cellular pathology lead scientist rather than to the lead histopathologist, triggered the laboratory technician to pull the relevant blocks and slides for the attention of the gynaecological histopathologist, minimising the delay between clinician request and sample being sent to the genomic laboratory.

The NELCN has a Bengali speaking ethnic minority population, which varies from 3% to 33% depending on the borough. All patient facing documents were translated into Bengali to improve engagement and communication with Bengali patients and family members as well as improve decision making. Additionally, a Bengali-speaking clinical member of the extended team, acted as an advocate during genetic counselling.

#### **4. Discussion**

We demonstrate that unselected concomitant/parallel panel germline and somatic testing at OC diagnosis can be implemented within the NHS setting, and delivered by treating cancer clinicians/professionals through a cancer-MDT coordinated approach. Pretest counselling was undertaken by all members of the cancer MDT team including medical oncologists, surgical oncologists and CNSs. Consistent with other reports of high uptake rates for *BRCA* testing [23,32–34], we showed this high acceptability extends to panel germline and somatic genetic testing too, with an uptake rate of 97%. PV carriers were younger, more likely to have a strong FH of cancer, HGSC histology and a CRS of 3 at histology. PV status was independent ethnicity, stage at diagnosis, timing of surgery or resection status. We undertook genetic testing prospectively for newly diagnosed patients and also for patients undergoing follow-up. Restricting this to prospective implementation of newly diagnosed cases alone (as has been implemented in some centres) would have missed 19 (19/54, 35.1%) germline PVs which were detected in the follow-up patients, thus significantly affecting screening/prevention options for these unaffected family members. A total of 56% of PVs would have been missed by using an FH based approach alone, reconfirming the importance of unselected testing and a mainstreaming approach. This is consistent with reports from others who also showed that around 50% PVs lacked a strong FH of BC or OC [23,33]. The *BRCA* PV prevalence in our NELCN cohort was higher than the Manchester cohort. Some boroughs in North East of London are known to have an Ashkenazi Jewish (AJ) population and the presence of AJ founder mutations in seven NELCN OC cases (Supplementary Table S1) is a contributory factor towards this as *BRCA* PV are commoner in AJ compared to non-AJ general population OC cases [35]. We found seven AJ BRCA founder mutations in the NELCN cohort but none of these patients self-reported Jewish ethnicity at recruitment. These patients may have had mixed parentage or grand-parentage and been unaware of their ethnicity or may have preferred not to report/disclose Jewish ethnicity. Additionally, NELCN includes 122 women who had previously been diagnosed and were alive at the time of commencement of the study. Although short term survival for *BRCA* PV carriers is higher, we did not find the sub-group of 122 women may be enriched for PV.

Our data show that over 1 in 5 (22%) patients have a PV which can affect their treatment, and 1 in 6 have a germline PV which can also affect predictive testing and screening and prevention in unaffected family members. This is consistent with some other reports in the literature [23,33,36,37]. Testing for a panel which includes *RAD51C, RAD51D, BRIP1* is not currently part of the NHS Genomics test directory and therefore not mandatory across the UK. However, it can if implemented identify an additional 13% (7/54) PVs, with a prevalence of 2.3% in OC patients, whose families can benefit from precision prevention. Rust et al. showed a slight increase in PVs detected with additional *RAD51C/RAD51D* testing but this was not completely unselected in their cohort and was undertaken either sequentially or in those with a strong FH [33]. Our data confirm the benefit of amending the UK test directory criteria to offer multi-gene panel testing to all UK women with OC. Our multi-gene germline test includes high- and intermediate risk genes which have already proven clinical utility [38]. A number of commercially available panels

are available today which test for many more (30–100) genes. However, it is important that only genes of established clinical utility are tested for. We are against indiscriminate panel testing for genes without established clinical utility [39,40]. In addition to *RAD51C, RAD51D* and *BRIP1* genes, it would be appropriate for an OC panel to also include *PALB2* and Lynch Syndrome genes going forward. *PALB2* has recently been reported as a moderate risk OC gene [41] and Lynch Syndrome (MMR) genes may be found in another 1% OC patients [42–44]. Some initial reports suggest that cascade testing rates may be lower following mainstreaming compared to testing in clinical genetics [34]. However, all our patients with PVs are reviewed in clinical genetics teams, who are responsible for facilitating cascade testing. Additionally, cascade testing rates are likely to increase with longer follow up.

As multiple genes get incorporated into OC testing panels, the reported VUS rate will also increase. Our germline panel VUS rate was 6.6% and is comparable to that reported by others [45,46]. VUS reporting and subsequent management can pose challenges for counselling, variant monitoring and onwards risk management. This will become an increasingly important issue with widening of the panel of genes tested for [47]. Risk reducing surgery, chemoprevention, screening or downstream predictive testing for unaffected family members, is not recommended in individuals with a VUS. Our report also highlights the importance of uniform classification and standardised reporting of class 3 variants (VUS) across genetic laboratories, including the description in clinical reports issued. The Cancer Variant Interpretation Group UK (CanVIG-UK) now provides an exemplar of a multidisciplinary network addressing this nationally [30]. This improves interpretability of reports by cancer clinicians. Appropriate pre-test education of patients and providers is necessary to limit the harm that could result from VUS misinterpretation. While not of immediate direct relevance, a proportion of VUS will be reclassified in the future to PVs and then have implications for the patients and relatives. This reclassification rate has been reported as around 9% in a large cohort [48]. In our cohort, a germline mutation *BRCA1* c.442-22\_442-13del reported in somatic but missed in initial germline (identified in re-analysis of germline) was initially reported as Class 3 VUS and subsequently a year on from testing, was re-classified as a PV.

Strengths of this study include prospective design and systematic approach to include all patients including those on follow up, as well as the high acceptability and uptake rates demonstrated with our pathway and testing process. The upfront staff training implemented across the pathway and continued support provided along-with broad stakeholder engagement contributed to improved patient experience and satisfaction. The extra efforts undertaken to engage with our ethnic minority Bengali population is another strength. In order to broaden access and informed decision making we translated information sheets into local Bengali language and trained a Bangladeshi oncology team member who was instrumental in engaging them in genetic counselling. Our analysis also demonstrates likely success rates for tumour testing for different types of samples which can be helpful for counselling patients and planning services. Limitations include lack of qualitative data and long term follow up data on patient outcomes. These are being collected.

Mainstreaming models such as ours delivered by the cancer MDT team enables implementation of large-scale genetic testing at cancer diagnosis. This approach too can encompass more than one pre-test counselling session where needed. A total of 22% women needed and received more than one pre-test counselling session in our study. Most other mainstreaming studies do not report on the number of pre-test counselling sessions needed or if multiple were offered. Our clinical nurse specialists favoured utilising more appointments/consultations prior to recruitment. While we did not undertake a formal quantitative assessment of reasons for multiple consultations, colleague feedback indicates these included, some patients needing more time to assimilate information and reflect on it before deciding and/or the need to discuss further with family before decision making; as well as a clinical assessment of not overloading the patient with too much

information at the first setting especially if they were struggling with managing decision making and information related to their cancer care at the appointment. The issues of some initially long consultations and time pressures in a busy oncology clinic also contributed to this. Other examples of models used to deliver unselected genetic testing at OC diagnosis include a genetics team embedded in oncology clinics, [25] genetic nurse coordinated model [24] and medical oncology [32] delivered testing.

Validation and implementation of 3 mm cored biopsies from FFPE tumour blocks enabled time- and resource-efficient processing of archived samples. This is particularly suited for archived FFPE tissue (analysis of retrospective cases) and gave a comparable/higher DNA yield than that obtained through slides. Although, we were unable to test 21% of archived tumour samples, undertaking tumour testing at time of diagnosis for future cases will overcome this. Our pathway now incorporates pathology processing/preparation for genetic testing for all cases at the time of routine histopathology analysis of the initial diagnostic or surgical specimen itself. As a large proportion of failed analysis was pre-treatment diagnostic biopsies, we now routinely obtain additional tissue cores for all women suspected of advanced ovarian malignancy at the time of their diagnostic biopsy. This minimises additional pathology laboratory resources needed and is more cost and time efficient. We also provide estimates of failure rates of diagnostic biopsy (~23%), which is relevant for counselling and management of patients planned for neo-adjuvant chemotherapy. NHS Laboratory guidelines suggest the minimum tumour content for NGS somatic/tumour testing referrals should be 20% [13]. However, we showed benefit of undertaking tumour testing even with <20% content in 45% of such cases. Hence, tumour testing should not be held back in cases with low tumour content as it could be successful in almost half these cases, thus identifying additional women who may benefit from PARP-i treatment.

There has been debate whether both germline and somatic testing should be offered to all; whether unselected germline testing should be offered as first line, followed by somatic testing if germline is negative for PV; or whether reflex somatic testing should be done first, reserving germline if a somatic PV is identified. PVs caused by large genomic rearrangements (LGRs) are missed when PCR-based testing alone is used [49,50]. MLPA is a commonly/routinely used technique to detect LGRs and is found to be highly sensitive and inexpensive [51,52]. LGRs are far more prevalent in *BRCA1* than *BRCA2* genes and have been reported to account for a wide range of *BRCA1* (up to 27%) and *BRCA2* (up to 11%) PVs [53–55]. In a large study, LGRs were reported to constitute around 24% of *BRCA1/BRCA2* PVs in high-risk breast/ovarian cancer families, [55] while lower rates are reported in other series and in individuals without strong family histories [53,55,56]. Reports suggest significant ethnic variation in the presence of LGR-related PVs: [55] African (2.4%), Caribbean and Latin American (6.7%), Danish (9.2%) and Spanish ancestry (14.5%) [55–57]. A disadvantage of using an initial tumour/somatic testing triage strategy is the possibility of missing LGRs. The 11% LGR-rate in our cohort (6/54) is similar to the LGR rate reported in some high-risk breast and ovarian cancer families [54]. In the majority of diagnostic laboratories, NGS tumour/somatic *BRCA*-testing is not validated for detection of LGRs [50]. While sequential tumour/somatic followed by germline testing may be a less costly approach [58], this strategy runs the risk of missing some germline PVs, particularly LGRs. This can have significant consequences for cancer prevention in families which are missed. Additionally, although reflex tumour testing can identify PVs seen in the germline, up to 31% of patients found to have a PV in the tumour may not get referred for genetic counselling or germline testing [59]. This highlights a potential limitation of a somatic first strategy, and the need for more robust implementation pathways with built in quality control and fail-safe mechanisms.

In contrast to our findings, a few earlier reports suggest 100% concordance between somatic and germline testing [45,60,61]. However, the proportion of LGRs amongst the *BRCA* mutations reported in these studies is unknown, as these have not been described. It is probable/likely that these studies did not have any LGRs in their mutation spectrum. In our cohort, somatic *BRCA*-testing alone, would have missed 9.2% (4/54) of *BRCA1/BRCA2* germline PVs and seven PVs in *RAD51C/RAD51D* and *BRIP1*, which comprise 20% (11/54) of germline PVs detected from 5-gene panel testing, who can benefit from targeted therapy and downstream predictive testing.

Germline-testing alone would have missed 2% (1/54) germline *BRCA1/BRCA2* PVs, and 15 somatic PVs, comprising 23.1% (16/69) of all *BRCA1/BRCA2* PVs in this cohort, who can benefit from PARP-i treatment. The germline PV missed is an error, which is unlikely to be repeated. A germline first followed by a somatic testing strategy could be an alternative option, but this approach will lead to a longer delay in turn-around times and increase clinician counselling time for giving results as this will need to be done twice. It is also likely to increase the laboratory processing and reporting time and costs, as this is undertaken after initial diagnosis (not contemporaneously with diagnostic reporting). In our experience, a simultaneous or parallel somatic/tumour and germline strategy is a more efficient approach for patients.

#### **5. Conclusions**

We demonstrate successful implementation of unselected 5-panel germline and concomitant somatic *BRCA1/BRCA2* testing for patients with OC. *BRCA1/BRCA2* germline PVs were identified in 15.5% patients and *BRCA1/BRCA2* somatic PVs in 7.8%. *RAD51C/RAD51D/ BRIP1* PVs comprised 13% of PVs and were identified in an additional 2.3% patients. A total of 11% germline PVs are LGRs and are missed by a somatic first testing strategy. A total of 20% of germline PVs would be missed if somatic *BRCA*-testing alone was used to triage for germline testing. A total of 55.6% germline PVs would have been missed by using FH ascertainment alone. The somatic testing failure rate is higher (23%) for patients undergoing diagnostic biopsies. Retrospective archival FFPE tissue testing is feasible using 3 mm punch biopsies from tumour blocks. Our findings favour a prospective parallel somatic and germline panel testing approach as a clinically efficient strategy which maximises variant identification for clinical benefit. The UK Genomics test directory criteria should be expanded to include a panel of OC genes. Formal cost-effectiveness analysis for panel testing is needed and can facilitate wider clinical implementation.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/cancers13174344/s1, Table S1: List of variants identified through germline and somatic testing.

**Author Contributions:** Conceptualization, R.M.; methodology, R.M., N.S., R.L., M.L. and D.G.E.; formal analysis, R.M., D.C., O.B. and D.G.E.; Implementation and Investigation, R.M., D.C., R.E.M., S.M.C., A.W., L.A.J., N.S., D.G.E., R.L., A.F., A.J., E.B., A.K., M.L., R.W. and M.Q.; resources, R.M., D.C., R.E.M., M.S., O.E., S.M.C., F.G., R.F.L.H., L.S., L.A.J., M.A., A.K., A.J., A.C.L., E.B., S.P., M.Q., T.M.-B., F.E.K., A.F., L.C., G.T., G.J.B., H.S., M.B., P.S., N.L.B., A.W., N.S. and D.G.E.; data curation, D.C., RM, O.B. and D.G.E.; writing—original draft preparation, R.M., M.S., D.C. and D.G.E.; writing—review and editing, All authors; supervision, R.M.; project administration, R.M., D.C. and M.S.; funding acquisition, R.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** The study is funded by The Barts Charity, grant ECMG1B6R.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Ethics Committee of London Riverside Ethics Committee (reference number 17/LO/0405).

**Informed Consent Statement:** Informed consent was obtained from all study participants.

**Data Availability Statement:** The data that support the findings of this study are available from the corresponding author, R.M. upon reasonable request.

**Acknowledgments:** D.G.E. is supported by the Manchester NIHR Biomedical Research Centre (IS-BRC-1215-20007). OB thanks the Ministry of Science and Higher Education of the Russian Federation within the framework of state support for the creation and development of World-Class Research Center "Digital biodesign and personalized healthcare" 075-15-2020-926.

**Conflicts of Interest:** R.M. declares research funding from Barts and the London Charity and Rosetrees Trust outside this work, an honorarium for grant review from Israel National Institute for Health Policy Research and honorarium for advisory board membership from AstraZeneca/MSD/GSK. R.M. is supported by an NHS Innovation Accelerator (NIA) Fellowship for population testing. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results. FG declares research funding from The NHS Grampian Endowment Fund, Medtronic and Karl Storz outside this work.

#### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Cancers* Editorial Office E-mail: cancers@mdpi.com www.mdpi.com/journal/cancers

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com

ISBN 978-3-0365-2983-7