*Article* **Association between Plasma Trace Element Concentrations in Early Pregnancy and Gestational Diabetes Mellitus in Shanghai, China**

**Ting Wu 1,2,†, Tao Li 1,2,†, Chen Zhang 1,2, Hefeng Huang 1,2,3,4,5,\* and Yanting Wu 3,4,\***


**Abstract:** (1) Background: Trace elements play important roles in gestational diabetes mellitus (GDM), but the results from reported studies are inconsistent. This study aimed to examine the association between maternal exposure to V, Cr, Mn, Co, Ni, and Se in early pregnancy and GDM. (2) Methods: A nested case-control study with 403 GDM patients and 763 controls was conducted. Trace elements were measured using inductively coupled plasma-mass spectrometry in plasma collected from pregnant women in the first trimester of gestation. We used several statistical methods to explore the association between element exposure and GDM risk. (3) Results: Plasma V and Ni were associated with increased and decreased risk of GDM, respectively, in the single-element model. V and Mn were found to be positively, and Ni was found to be negatively associated with GDM risk in the multi-element model. Mn may be the main contributor to GDM risk and Ni the main protective factor against GDM risk in the quantile g computation (QGC). 6.89 μg/L~30.88 μg/L plasma Ni was identified as a safe window for decreased risk of GDM. (4) Conclusions: V was positively associated with GDM risk, while Ni was negatively associated. Ni has dual effects on GDM risk.

**Keywords:** gestational diabetes mellitus; nickle; trace elements; restricted cubic spline; LASSO regression; quantile g-computation; BKMR models

between Plasma Trace Element Concentrations in Early Pregnancy and Gestational Diabetes Mellitus in Shanghai, China. *Nutrients* **2023**, *15*, 115. https://doi.org/10.3390/ nu15010115

**Citation:** Wu, T.; Li, T.; Zhang, C.; Huang, H.; Wu, Y. Association

Academic Editor: Annunziata Lapolla

Received: 23 November 2022 Revised: 13 December 2022 Accepted: 23 December 2022 Published: 27 December 2022

**Copyright:** © 2022 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**

Gestational diabetes mellitus (GDM), which refers to diabetes diagnosed for the firsttime during pregnancy, is one of the most common medical complications of pregnancy [1]. It is associated with substantial short- and long-term adverse complications for both mother and child. The documented prevalence of GDM varies substantially worldwide, ranging from 1% to >30% [2]. The incidence rate of GDM has been increasing worldwide and is approximately 14.8% (95% CI 12.8, 16.7%) in China according to the latest meta-analysis involving 79,064 Chinese participants [3]. In addition to traditional risk factors, such as advanced maternal age, ethnicity, a previous history of gestational diabetes, and a family history of type 2 diabetes mellitus (T2DM), trace elements may play important roles in the development of diabetes [4].

Certain trace elements, such as chromium (Cr), have been suggested to participate in increasing insulin binding and insulin receptor number [5]. Vanadium (V) was found to participate in inhibiting glucose release, improving gluconeogenesis-related enzyme activity, and exerting an insulin-sensitizing effect [6]. Meanwhile, some essential elements, such as manganese (Mn) were found to be associated with a higher risk of hyperglycemia by inhibiting glucose-stimulated insulin secretion and inducing inflammation and oxidative stress [7]. However, not all human studies support the results from laboratory studies. An adult cohort study from Southern Spain suggested that concentrations of certain trace elements (such as Cr) in adipose tissue are associated with the risk of incident T2DM, while V might have a protective effect [8]. A case-control study in China indicated that higher levels of serum selenium (Se) were associated with increased T2DM risk [9].

Some trace elements have recently been suggested to be associated with the risk of GDM in epidemiologic studies. A prospective study demonstrated that increased concentrations of urinary nickel (Ni), Cobalt (Co), and V in early pregnancy are associated with an elevated risk of GDM [10]. In contrast to the results of the above research, two case-control studies indicated an inverse association of V exposure with GDM [11,12], which was reflected by plasma V concentrations and meconium V concentrations. No significant association was found between blood Ni and GDM in the single-metal model in a Chinese birth cohort study [13]. Moreover, a nested case-control study in Xiamen, China, measured Cr concentrations in meconium from newborns delivered by mothers with GDM (137 cases) and without GDM and found a positive association between Cr concentration and GDM prevalence in a dose-dependent manner [14]. One recent meta-analysis showed that the serum Se level of patients with GDM was lower than that in healthy pregnant women. However, no association was found between plasma Se, Cr, and GDM in another nested case-control study [15]. A higher concentration of Mn within a certain range before 24 weeks gestation was demonstrated to impair fasting plasma glucose during pregnancy in a retrospective study [16]. Additionally, a French mother-child cohort study did not find a significant association between blood Mn and the prevalence of GDM [17].

Thus far, the results and conclusions on the relationship between the six trace elements— V, Cr, Mn, Co, Ni, and Se—and GDM are limited and contradictory. In addition, it is essential to study the joint effects of trace elements on GDM risk because elements in the environment exist in the form of co-exposure, and the specific elements included in the analysis individually could be potentially confounded by other elements to which pregnant women are also exposed from the same source. However, when exploring the effects of a multielement mixture in a traditional way, highly unstable results may be obtained if incorporating two or more highly correlated (collinear) elements in a regression model [18]. In recent years, various interdisciplinary methods [19] have been developed to address such issues.

In the present study, we aimed to explore the relationship between these six plasma trace element concentrations before 14 gestational weeks and the risk of GDM. We used least absolute shrinkage and selection operator (LASSO) regression, quantile g computation (QGC), and Bayesian Kernel Machine Regression (BKMR) to screen out independent variables, assess the joint effect of elements on GDM risk and determine the contribution of each element on GDM risk, restricted cubic spline (RCS) was employed to explore the dose-response relationship between elements exposure and GDM risk, with the hope to provide new insights for the prevention of GDM.

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

#### *2.1. Study Population*

This case-control study was nested in a prospective study initiated in Shanghai, China. From November 2020 to February 2021, pregnant women who visited the International Peace Maternal and Child Hospital (IPMCH) for the first prenatal examination between 8 and 14 gestational weeks and provided enough blood samples were included in the study (*n* = 2069).

The excluded participants were those: (1) who had multiple births (*n* = 52); (2) who were diagnosed with T2DM and other metabolic diseases before pregnancy (*n*= 31); (3) who had serious medical diseases such as cancer (*n* = 12); and (4) who had missing information on birth outcomes (*n* = 292) and missing blood samples (*n* = 92).

Among 1724 finally included pregnant women, 403 pregnant women were diagnosed with GDM and included in the GDM group, and a total of 763 controls were randomly selected from the remaining participants by maternal pre-pregnancy BMI and maternal age (case/control = 1:2 for 360 cases and case/control = 1:1 for 43 cases).

All of the participants in the study signed informed consent forms. This study was approved by the ethics committee of the IPMCH.

#### *2.2. Data Collection*

Baseline information was obtained from electronic medical records, including maternal age, ethnic group, pre-pregnancy body mass index (BMI), reproductive history, family and personal disease history, smoking exposure, alcohol consumption, education levels, household income, delivery method, and fetal sex. Maternal BMI was calculated using the formula BMI = weight (kg)/height (m2). Gestational age was calculated based on the gestational week of delivery and the first day of the last menstrual period. In the present study, smoking exposure was defined as positive if the mother had a smoking history, and alcohol consumption was considered positive if the mother had a drinking history.

#### *2.3. Laboratory Measurements*

Plasma concentrations of total cholesterol (CHOL), triglycerides (TG), high-density lipoprotein cholesterol (HDL), low-density lipoprotein cholesterol (LDL), apolipoprotein-A (APO-A), apolipoprotein-B (APO-B) and fasting plasma insulin (FPI) were measured by an automatic chemistry analyzer (BeckmanDXI800, Beckman, Bria, CA, USA). The homeostasis model of assessment-insulin resistance (HOMA-IR) score was obtained according to the following formula: HOMA-IR = FPG (mmol/L) × FPI (μU/mL)/22.5.

Inductively coupled plasma-mass spectrometry (ICP-MS) was used for the determination of the six trace elements. ICP-MS is a quadrupole mass spectrometer, consists of basic components, including the peristaltic pump, nebulizer, spray chamber, ICP torch, interface cones, ion optics, quadrupole, and detector. It has been considered the gold standard analytical method for element measurements in biological samples which meet the interference elimination of the determination of different elements in the sample. We used the NexION 300X device (PerkinElmer, Waltham, MA, USA) and the stander mode for the measurements [20]. Blood was collected between 8 and 14 gestational weeks, and plasma was collected in EDTA tubes after centrifugation at 2000 rpm for 20 min. All plasma samples were frozen at −80 ◦C for storage and transferred to a 4 ◦C refrigerator the night before detection. Several standard curves were prepared by diluting the element standard solution (PerkinElmer, Waltham, MA, USA), and the value of the limit of detection (LOD) of each element was calculated. Plasma (100 μL) was diluted 20 times with sample diluent (1% TMAH1 + % nitric acid) and fully vibrated before detection. See Table S1 for the LOD and detection rate of each element. When the plasma element concentration was below the LOD, LOD/√2 was used instead. Standard samples were detected in each batch (30 samples) for quality control purposes.

#### *2.4. Diagnosis of GDM*

At 24–28 weeks of gestation, an oral glucose tolerance test (OGTT) was implemented by a 75 g glucose challenge. A diagnosis of GDM was made if fasting plasma glucose was ≥5.1 mmol/L (≥92 mg/dL), 1-h plasma glucose was ≥10.0 mmol/L (≥180 mg/dL), or 2-h plasma was ≥8.5 mmol/L (≥153 mg/dL), according to the recommendations from the Diabetes and Pregnancy Study Group (IADPSG) [21].

#### *2.5. Statistical Analysis*

The control group was matched for the GDM group by maternal age and pre-pregnancy BMI using the propensity score matching method (PSM) [22]. Basic demographic characteristics, plasma microelement concentrations, and clinical indicators of the study population were represented using N (%) for categorical variables and median and interquartile range (IQR) for continuous variables. Comparison between case and control groups was determined by the Wilcoxon rank sum test (for continuous variables) or Chi-square (χ2) test (for categorical variables). The concentrations of trace elements were natural log-transformed [Ln(X)] to normalize their distribution. The pairwise correlations among multiple elements were calculated by Spearman's rank correlation analysis and a correlation-matrix heatmap was plotted. Conditional logistic regression was adopted to evaluate the association between the concentration of trace elements and the risk of GDM by odds ratios (ORs) and 95% confidence intervals (CIs). We chose covariates based on the literature review, stepwise regression, best subset selection, and biological reliability. Potential confounding factors and factors with significant differences between the case and control groups in univariate analysis were included in Model 4, including age (continuous variable), pre-pregnancy BMI (<18.5, 18.5–24, >24), family history of diabetes (yes or no), education level (<10, 10–12, ≥13 years), ethnic groups (Ethnic Han or others), household income level (<0.1 million, 0.2–0.3 million, >0.3 million), TG (continuous variable), CHOL (continuous variable), LDL-cholesterol (continuous variable), HDL-cholesterol (continuous variable) and APOB (continuous variable). Covariates screened by stepwise regression were included in Model 2, including education level, ethnic groups, TG, LDL-cholesterol, HDL-cholesterol, and APOB. Covariates including family history of diabetes, education level, ethnic groups, TG, LDL-cholesterol, and APOB, which were screened by best subset selection, were included in Model 3. The potential nonlinearity of the association of plasma trace elements with odds of GDM, OGTT value, and FPI was further examined using RCS with three knots at the 25th, 50th, and 75th percentiles of Ln (plasma element concentrations) assessed via R version 4.2.0 software ("rms" package).

LASSO regression, QGC, and BKMR models were used to screen out independent variables, assess the joint effect of elements on GDM risk, and determine the contribution of each element to GDM risk. In these analyses, we adjusted for the same variables as in Model 3 of the conditional logistic regression analysis. The 11 covariates and six elements were included in the LASSO regression, and the independent variables with greater influence on the dependent variable were screened when the regression coefficient was compressed to zero. These selected elements were included simultaneously in the multiple-element model adjusted or not adjusted for covariates selected by LASSO (family history of diabetes, education level, ethnic groups, household income level, TG, LDLcholesterol, HDL-cholesterol, APOB). Quantile g computation, an adaptive adaptation modeling method with weighted quantile sum regression, was used to evaluate the different directions of mixed effects for individual elements and rank important constituents [23]. QGC was conducted using R version 4.2.0 with the "qgcomp" package. BKMR [24] was also used to assess the joint effect of all elements on the risk of GDM and the effect of an individual element as part of the element mixture via the R version 4.2.0 software ("bkmr" package). A PIP (prosterior inclusion probabilities) threshold of 0.5 was considered to be relatively important for individual element exposure to GDM risk.

All statistical analyses were performed using the SPSS 26.0 and R version 4.2.0 software. A *p*-value (two-tailed) < 0.05 was considered significant.

#### **3. Results**

#### *3.1. Characteristics of the Study Population*

The characteristics of the study population are presented in Table 1. The median age of the included pregnant women was 32 years. The study population was well-educated, with around 71.78% of educational level reaching university and higher, and the women who developed GDM were less educated than the women in the control group.


**Table 1.** Characteristics of the study population.

GDM, gestational diabetes mellitus; BMI, body mass index; \* *p* < 0.05.

#### *3.2. Levels of Plasma Trace Elements and Glucose and Lipid Metabolism Indices*

The exposure levels of the six trace elements in the case and control groups are summarized in Table 2. There were significantly increased levels of plasma V in the GDM group but significantly lower plasma concentrations of Cr and Se. Correlations between trace elements ranged from 0.07–0.82 in Spearman's rank correlation analysis (Figure S1). As shown in Table S2, despite Apo-A, other glucose and lipid metabolism indices were significantly different between the case and control groups, with FPG, OGTT-1h, OGTT-2h, FPI, HOMA-IR, CHOL, TG, LDL-cholesterol, and Apo-B increased significantly and HDL-cholesterol decreased significantly in the GDM group.

**Table 2.** Profiling of trace elements in maternal plasma of the case-control group.


GDM, gestational diabetes mellitus; V, Vanadium; Cr, Chromium; Mn, Manganese; Co, Cobalt; Ni, Nickel; Se, Selenium; \* *p* < 0.05, \*\* *p* < 0.01.

#### *3.3. Association between Plasma Trace Elements and Risk of GDM*

The results of the conditional logistic regression are shown in Table 3. The plasma level of V was positively associated with the risk of GDM, and every unit increase in the natural log of V exposure was associated with 39% (OR = 1.39 (95% CI 1.14, 1.69)) a higher risk of GDM. In contrast, the concentration of plasma Ni was negatively associated with the risk of GDM, and every unit increase in the natural log of Ni exposure was associated with 14% (OR = 0.86 (95% CI 0.77, 0.97)) a lower risk of GDM. Elevated plasma concentrations of Cr, Mn, Co, and Se were not associated with the risk of GDM.

**Table 3.** Associations between plasma trace element exposure and GDM risk.


GDM, gestational diabetes mellitus; OR, odds ratio; CI, confidence interval; \* *p* < 0.05, \*\* *p* < 0.01, \*\*\* *p* < 0.001. Model 1 adjusted by education level, ethnic groups, TG, LDL cholesterol, HDL cholesterol, and APOB. Model 2 adjusted by family history of diabetes, education level, ethnic group, TG, LDL cholesterol, and APOB. Model 3 adjusted by age, pre-pregnancy BMI, family history of diabetes, education level, ethnic group, household income level, TG, CHOL, LD, HDL cholesterol, and APOB. Adjusted model adjusted by family history of diabetes, education level, ethnic groups, household income level, TG, LDL cholesterol, HDL cholesterol, and APOB.

*3.4. Dose-Response Association of Plasma Trace Element Exposure with GDM Risk, Glucose, and Insulin Level*

The potential nonlinearity of the relation between Ln- Ni (*p* overall < 0.001, *p* nonlinearity = 0.003) and the risk of GDM was observed in the restricted cubic spline model (Figure 1). In the relatively low levels (<6.89 μg/L) and higher levels (>30.88 μg/L) of plasma Ni, a positive correlation was found between plasma Ni and GDM risk. U-shaped exposure relationships were observed between Ni and FPI (*p* = 0.038), FPG (*p* = 0.006), OGTT-1h (*p* < 0.001), and OGTT-2h (*p* = 0.036) (Figure S2). Additionally, positive dose-response relationships were also observed between V and FPI (*p* = 0.005) and FPG (*p* = 0.004) (Figure S3), Mn and FPI (*p* = 0.018) (Figure S4), and Co and FPI (*p* = 0.039), OGTT-1h (*p* = 0.024) and OGTT-2h (*p* = 0.044) (Figure S5). Nonlinear relationships were not observed between Cr and Se and glucose or insulin levels (Figures S6 and S7).

**Figure 1.** The dose-response relationship of plasma Ni level with GDM risk. RCS regression was used to analyze the dose-response relationship of Ln-Ni with GDM risk after adjusting for family history of diabetes, education level, ethnic groups, TG, LDL-cholesterol, and APO-B. The knots were located at the 25th, 50th, and 75th percentiles. The red line and black dotted line represent the OR value and 95% CI, respectively. The black points represent OR = 1, and the corresponding value of plasma Ni concentration is presented.

#### *3.5. Associations of Metallic Elements Screened by LASSO Regression and Their Coexposure with GDM Risk*

The results of LASSO regression showed that all six trace elements had a strong effect on GDM risk (Figure S8). Next, we fitted a logistic regression model and brought all six elements into the model, and additionally adjusted covariates selected by the LASSO regression (Table 3). In this part, increasing Ln-V (OR = 1.27 (95% CI 1.01, 1.60)) and Ln-Ni (OR = 0.72 (95% CI 0.60, 0.86)) were positively and negatively related to the increased risk of GDM, respectively. In addition, increasing Ln-Mn was also observed to be positively associated with an increased risk of GDM (OR = 1.70 (95% CI 1.22, 2.36)).

#### *3.6. Quantile G-Computation Analyses*

QGC analysis showed that increasing the Ln-element mixture by one unit was not associated with an increased risk of GDM (OR = 0.97, 95% CI: 0.84, 1.13). The individual weights for each trace element mixture component are shown in Figure 2: Mn (59.1%), Co (31.6%), and V (9.3%) had a positive contribution, and Ni (61.5%), Cr (20.8%) and Se (17.7%) had a negative contribution.

**Figure 2.** Association between trace element levels and GDM based on quantile g-computation analyses. The estimated weights of each element in the mixture were presented by bootstrapping in either direction.

#### *3.7. Bayesian Kernel Machine Regression Analyses*

The PIP of elements varied between 0.7 and 1 (V: 0.8666, Cr: 0.7746, Mn: 1.0000, Co: 0.7742, Se: 1.0000, Ni: 0.8062), and thus, all the elements could be considered important. The univariate relationship between each element and GDM risk is shown in Figure S8. Positive trends were observed between V, Co, and GDM risk (Figure S9a,d), and Cr showed a negative relationship with GDM risk (Figure S9b). A J-shaped relationship between Ni and GDM risk (Figure S9e) and an inverted J-shaped relationship between Mn, Se, and GDM risk (Figure S9c,f) were observed. Considering the comparable positive weight and negative weight of elements on GDM risk, the cumulative effect was not statistically significant, as shown in Figure 3a. Figure 3b illustrates the estimated contribution of individual exposures to the cumulative effect by comparing the GDM risk when a single element was at the 75th percentile compared to when it was at its 25th percentile, where all of the remaining elements were fixed to a particular quantile, such as P25, P50, and P75. When the other element percentiles increased, the higher percentiles of Ni and Mn (P75) presented a more significant negative effect and positive effect on GDM risk compared to their lower concentrations (P25), respectively. The trends of each element in the bivariate

exposure-response analysis (Figure S10) were consistent with the univariate relationship analysis, except for the effect of Se on the GDM risk, which was inversely changed when Mn was at a high concentration (P75).

**Figure 3.** The joint effect of the trace element mixture on GDM by the BKMR model (**a**) The overall effects of element mixtures (estimates and 95% CI) in elements fixed to different percentiles compared to when they were at their medians (P50). (**b**) The effects of single exposure when an individual element was at its 75th percentile compared to when that exposure was at its 25th percentile, where all other exposures were fixed to a particular quantile (P25, P50, and P75). The results were adjusted by family history of diabetes, education level, ethnic groups, TG, and LDL-cholesterol.

#### **4. Discussion**

In this study, we found that a higher level of plasma V and a lower level of plasma Ni before 14 weeks of pregnancy may be prospectively related to a higher risk of GDM in both single- and multiple-element models. Plasma Mn was found to be positively associated with an increased risk of GDM only in the multiple-element model, and there was no significant relationship between Cr, Co, and Se in the conditional logistic regression. A J-shaped relationship between plasma Ni concentrations and GDM and a U-shaped exposure-response relationship between plasma Ni and OGTT values and FPI were found in RCS analysis. The joint effect of element mixtures on the risk of GDM was not observed in the QGC analysis or the BKMR model. The results from QGC analysis indicate that Mn (59.1%), Co (31.6%), and V (9.3%) had a positive contribution, and Ni (61.5%), Cr (20.8%), and Se (17.7%) had a negative contribution to the risk of GDM. The association of incident GDM with V, Ni, and Mn was consistent in the outcomes of conditional logistic regression, QGC analysis, and BKMR analyses.

V was found to participate in inhibiting glucose release, improving gluconeogenesisrelated enzyme activity, and exerting an insulin-sensitizing effect [6]. Two case-control studies showed a positive association between V in plasma [25] and serum [9] with diabetes risk. Two case-control studies based on pregnant women also reported that V exposure, reflected by meconium [12] and plasma V [11] concentrations, respectively, was inversely associated with the odds of GDM. Therefore, we expected that V would reduce the risk for GDM in our study, but the results indicated the opposite. The results from a prospective cohort study conducted in Wuhan, China, were close to those of our study, where there was a significant and positive association between urinary V and GDM based on singlemetal models (OR = 1.28, 95% CI: 1.05–1.55) [10]. However, the biological sample of the abovementioned study was inconsistent with our study.

Notably, the median value of plasma V in the present study (6.25 μg/L) was much higher than that in previous studies (plasma V: 0.191 μg/L [25], 0.73 μg/L in GDM cases and 0.80 μg/L in controls [11]), which may account for the inconsistent result. Although V has been found to exert a beneficial effect on the metabolism of carbohydrates, the effective therapeutic dose is difficult to establish, and excess concentrations may lead to several toxic effects [5]. In addition, the health hazards of V, especially when it is at the highest oxidation state (+5), cannot be ignored. V can act as a strong pro-oxidant and pro-apoptotic factor, damage the antioxidant barrier, exacerbate lipid peroxidation (LPO), and lead to programmed cell death (apoptosis) [26]. Therefore, more research is warranted to further explore the effect of V on the risk of GDM and determine the safe exposure range of V.

The epidemiological evidence of Ni in glucose metabolism is limited and inconsistent, although Ni was suggested to adversely affect glucose metabolism by inducing hyperglycemia and glycogenolysis in laboratory studies [27]. Two studies based on Chinese adults and U.S. adults showed that increased urinary Ni concentration is associated with an elevated prevalence of diabetes [28,29]. However, a multisite and multiethnic cohort study of midlife women did not find an association between urinary Ni and an elevated risk of diabetes in midlife women [30]. Another nested case-control study obtained a similar result, and no significant associations were found between plasma Ni and incident diabetes [31]. The evidence of an association between Ni exposure and diabetes in pregnant women was insufficient, and no significant association was found between blood Ni and GDM in the single-metal model in a Chinese birth cohort study [13]. Nevertheless, another Chinese cohort study demonstrated that increased concentrations of urinary Ni in early pregnancy are associated with an elevated risk of GDM [10]. In the present study, a relationship between elevated maternal plasma concentrations of Ni and decreased risk of GDM was observed when evaluated individually or as an element mixture. Additionally, a J-shaped exposure relationship of Ln-Ni with the OR of GDM and U-shaped exposure relationships between Ln-Ni and the three OGTT values and FPI were all observed in the RCS analysis.

When interpreting our study results, the different study biological materials and the much higher concentration of Ni measured in our study should be considered (median (IQR) 30.67 (17.34–48.58) μg/L) when compared to previously published studies (median ranged from 2.48 to 6.484 μg/L [13,31,32]). Urine was a more commonly used biological material in previous studies because of the short half-life period of Ni. Considering that the main exposure source of Ni (from drinking water and food) may be stable in the pregnancy period, our findings may provide new insight into the effect of Ni on glucose metabolism. To the best of our knowledge, this is the first study to demonstrate the dual effect of plasma Ni exposure on the risk of GDM and provide a safety window value of Ni exposure (6.89 μg/L~30.88 μg/L) with potential clinical significance. Interestingly, the cutoff value of 30.88 μg/L was close to the median value of the study population (30.67 μg/L). Thus, we still recommend low levels of Ni exposure in daily life because Ni is potentially essential to the human body, but at high doses is toxic. More research is warranted to further verify our findings and explore the underlying mechanism.

Mn is both an essential nutrient and a potential toxicant, depending on the level of exposure. Mn supplementation may protect mitochondria and islets from ROS by enhancing MnSOD activity and protecting against diabetes [33], but it was also suggested that Mn can inhibit glucose-stimulated insulin secretion in β-cells by impairing mitochondrial function [7]. A cross-sectional study based on coke oven workers indicated that urinary Mn levels were positively associated with hyperglycemia but not with diabetes risk [34]. A Ushaped association between plasma manganese and T2DM was reported by a case-control study [35]. Nevertheless, no significant association was observed between second-trimester blood Mn and GDM in a French mother-child cohort study [17]. A retrospective cohort study from South China demonstrated that serum Mn may prospectively increase the late second trimester OGTT0 but not GDM risk [16].

We measured a relatively low concentration of Mn [median (IQR) 5.79 (3.51–8.90) μg/L] compared with other studies (median ranged from 6.52–21.85 μg/L) [15,17,35]. We found

no significant association between Mn and GDM in the single-element model, but interestingly, when we included all the elements in the conditional logistic model, Mn showed a significant positive association with GDM. Additionally, Mn was positively associated with FPI level and was found to be the greatest contributor (59.1%) to GDM in the QGC analysis, which was similar to the results of QGC analysis from a large Japanese study (Mn: 47.4%) [36]. A similar result can also be observed in the BMKR model; when other element percentiles increased, Mn showed a more obvious positive association with GDM. We can speculate that Mn can promote the development of GDM through interactions with other elements such as Se as indicated in the bivariate exposure-response analysis of BKMR models but more evidence is needed to validate the speculation.

Cr, Co, and Se are essential trace elements in the human body. Cr was found to play a significant role in glucose metabolism and have beneficial effects on insulin sensitivity and lipid parameters [37]. Co is an important component of vitamin B12, and Se plays a critical role mainly as a selenoprotein.

Nevertheless, the role of Cr and Co in the development of diabetes mellitus in human studies remains controversial. A positive association was reported between Cr in adipose tissue with T2DM in a 16-year follow-up period prospective adult cohort study [8] and between Co in urine with T2DM in a study based on the National Health and Nutrition Examination Survey (NHANES, 1999–2010) [38]. In a case-control study involving 1471 patients with newly diagnosed T2DM, 682 individuals with newly diagnosed pre-DM indicated that plasma Cr concentrations were inversely associated with T2DM and pre-DM [39]. A negative linear relationship between urinary Co and FPG was found in an ongoing occupational cohort study in China [40]. A large case-control study elucidated a U-shaped relationship between plasma Co concentrations and newly diagnosed T2DM [41].

For pregnant women, Cr in meconium was found to be positively associated with GDM prevalence in a dose-dependent manner in a nested case-control study [14], while data from another two nested case-control studies showed no significant association between Cr levels and the risk of GDM in pregnant women [15,42]. Studies exploring the relationship between Co exposure and GDM are limited; in a prospective cohort study, Co was shown to be significantly and positively associated with GDM [10]. The relationship between Se and GDM has been well established, and most studies support the negative association between Se and the risk of GDM [43]. A recent meta-analysis involving 27 studies showed that the serum Se level of patients with GDM was lower than that in healthy pregnant women [43].

In our present study, although no significant association was observed between Cr, Co, and Se concentrations and GDM in either a single-element or element coexposure logistic regression model. The plasma Cr and Se levels of patients with GDM were lower than those in the control group in our present study. In addition, the positive association between Co and GDM and the negative association between Se and GDM were consistent in the BKMR model and QGC analysis, and Co showed a positive non-linear relationship with FPI, OGTT-1h, and OGTT-2h in RCS analysis.

Our present results showed a much higher concentration of plasma Cr and Co and a comparable concentration of Se than those of previously published studies. Several previous studies showed that the median values varied from 0.2 to 3.97 μg/L for Cr [15,39,44], 1.68–1.9 mg/dL for Co [41,45], and 29.43–94.73 μg/L for Se [15,41,46]. The discrepancies between study populations remain to be elucidated because, aside from Cd, Cr and Co were reported to be the greatest heavy metal pollutant (Cr > Cd > Co > Zn > Ti > Cu) in the surface sediments of the Yangtze River Estuary [47], and there may indeed be a much higher level of metal/element exposure in the Shanghai population. In addition, Cr (III) and Se have been considered to have nutritional or pharmacological effects on the human body [48,49], ranging from antioxidant and anti-inflammatory effects to improving symptoms of insulin resistance, and Se is part of, for instance, Novalac Prenatal pills. The higher plasma level of Se and Cr in pregnant women in the non-GDM group may be the result of their using supplements containing Cr and Se before or during pregnancy. Further

well-designed studies should be carried out to explore the role of Cr, Co, and Se in the occurrence, development, and treatment of GDM.

We adopted the BKMR method in our study to determine the joint effects of elements, but the results showed that the increasing percentile of element mixtures was not related to an increased risk of GDM. We speculate that the main explanation is that the contributing effect and protective effect of these six metallic elements on GDM offset each other, as shown in the QGC analysis. Notably, the association between Ni with GDM and Mn with GDM becomes statistically significant along with the increasing percentile of other elements in the BKMR model. There may exist a relatively strong interaction between Ni, Mn, and other elements.

Our study has several strengths. First, we collected blood samples during the first period of pregnancy, which may reflect the causal relationship between element exposure and the risk of GDM. Second, the exposure levels of metallic elements were reflected by a continuous variable (Ln-concentration), which avoids the data loss caused by classification variable conversion. Third, we used several statistical methods to assess the joint effect of all six elements and the independent contribution of each element on the risk of GDM, and the results were stable among these models.

Limitations should also be considered. First, detailed information regarding other potential confounding factors of GDM, such as physical activity and the occupational status during pregnancy, was not well collected. Second, plasma is not the best biological material to reflect body exposure to some elements, such as Ni. Third, we did not take exposure sources such as dietary patterns and residential environment into consideration.

#### **5. Conclusions**

In conclusion, our results suggest a positive association between V exposure and a negative association between Ni exposure in early pregnancy with subsequent risk of GDM, regardless of whether they are evaluated individually or as elements mixtures. Plasma Mn was found to be positively associated with an increased risk of GDM in the multiple-element model. In addition, we demonstrate a J-shaped exposure-response relationship between plasma Ni concentrations and GDM and a U-shaped exposure-response relationships between plasma Ni concentrations and FPI, FPG, OGTT-1h, and OGTT-2h. Further studies are warranted to confirm these associations and explore the potential mechanism.

**Supplementary Materials:** The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/nu15010115/s1, Figure S1: Correlations of the plasma concentration of the six trace elements. Spearman's rank correlation was used to analyze the correlations between the Ln-transformed trace elements. Correlation coefficients are presented in the lower left part, and the upper right part is the heatmap of the correlation coefficients between Ln-transformedtrace elements. Blue represents a positive correlation, and the darker the color is, the greater the correlation coefficient. \* *p* < 0.05, \*\* *p* < 0.01, \*\*\* *p* < 0.001. Figure S2: The dose-response relationships of plasma Ni with FPI and OGTT values. The blue line and gray shadow represent the OR value and 95% CI, respectively. (a) The dose-response relationship of Ln-Ni with FPI. (b) The dose-response relationship of Ln-Ni with FPG. (c) The dose-response relationship of Ln-Ni with OGTT-1h. (d) The dose-response relationship of Ln-Ni with OGTT-2h. Figure S3: The dose-response relationships of plasma V with FPI and OGTT values. The blue line and gray shadow represent the OR value and 95% CI, respectively. (a) The dose-response relationship of Ln-V with FPI. (b) The dose-response relationship of Ln-V with FPG. (c) The dose-response relationship of Ln-V with OGTT-1h. (d) The dose-response relationship of Ln-V with OGTT-2h. Figure S4: The dose-response relationships of plasma Mn with FPI and OGTT values. The blue line and gray shadow represent the OR value and 95% CI, respectively. (a) The dose-response relationship of Ln-Mn with FPI. (b) The dose-response relationship of Ln-Mn with FPG. (c) The dose-response relationship of Ln-Mn with OGTT-1h. (d) The dose-response relationship of Ln-Mn with OGTT-2h. Figure S5: The dose-response relationships of plasma Co with FPI and OGTT values. The blue line and gray shadow represent the OR value and 95% CI, respectively. (a) The dose-response relationship of Ln-Co with FPI. (b) The dose-response relationship of Ln-Co with FPG. (c) The dose-response relationship of Ln-Co with OGTT-1h. (d) The dose-response relationship of

Ln-Co with OGTT-2h. Figure S6: The dose-response relationships of plasma Cr with FPI and OGTT values. The blue line and gray shadow represent the OR value and 95% CI, respectively. (a) The dose-response relationship of Ln-Cr with FPI. (b) The dose-response relationship of Ln-Cr with FPG. (c) The dose-response relationship of Ln-Cr with OGTT-1h. (d) The dose-response relationship of Ln-Cr with OGTT-2h. Figure S7: The dose-response relationships of plasma Se with FPI and OGTT values. The blue line and gray shadow represent the OR value and 95% CI, respectively. (a) The doseresponse relationship of Ln-Se with FPI. (b) The dose-response relationship of Ln-Se with FPG. (c) The dose-response relationship of Ln-Se with OGTT-1h. (d) The dose-response relationship of Ln-Se with OGTT-2h. Figure S8: LASSO regression analysis diagram. Maternal age, pre-BMI, ethnic group, educational level, household income level, family history of diabetes, TG, CHOL, LDL, HDL, APO-B, and 6 trace elements were included in the LASSO regression model for analysis. (a) The changing trajectory of misclassification error with the penalty parameter (logλ) (estimates and 95% CI). The dotted line represents the optimal λ value selected after cross-validation. (b) Cross-validation plot for the penalty term. Supplementary Figure S9: Effect of single trace element exposure on GDM by BKMR model Univariate exposure-response functions of each element (95% CI) with others fixed at their medians (P50). The results were adjusted by family history of diabetes, education level, ethnic groups, TG, LDL-cholesterol, and APO-B. Figure S10: Bivariate cross-section effects of trace element exposure on GDM by BKMR model Bivariate cross-section effects of the exposure-response function of a single element where the second element was fixed at P25, P50, and P75. Table S1: Profiling of trace elements in maternal plasma (*n* = 1166) Table S2: Level of glucose and lipid metabolism indices.

**Author Contributions:** Conceptualization, T.W. and T.L.; methodology, T.W.; software, T.W.; validation, Y.W. and H.H.; formal analysis, T.W.; investigation, T.L.; resources, T.L.; data curation, T.W.; writing—original draft preparation, T.W.; writing—review and editing, C.Z., H.H. and Y.W.; visualization, T.W. and C.Z.; supervision, H.H. and Y.W; funding acquisition, H.H. and Y.W. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the National Natural Science Foundation of China (82088102), CAMS Innovation Fund for Medical Sciences (2019-I2M-5-064), Collaborative Innovation Program of Shanghai Municipal Health Commission (2020CXJQ01), Clinical Research Plan of SHDC (SHDC2020CR1008A) and Shanghai Frontiers Science Research Base of Reproduction and Development.

**Institutional Review Board Statement:** The study was conducted in accordance with the Declaration of Helsinnki and approved by the Ethics Committee of the International Peace Maternal and Child Hospital (IPMCH)((GKLW) 2019-51).

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

**Data Availability Statement:** The data presented in this study are available on reasonable request from the corresponding author.

**Acknowledgments:** We would like to thank the participants and the medical staff of the International Peace Maternal and Child Hospital (IPMCH). We would like to acknowledge the School of Public Health, Fudan University, for providing the detection platform. We would like to thank Yiming Dai, Qiang Liu, and Zhijun Zhou. We thank them for the instrument maintenance during element detection. We would like to thank Chun Xia, a graduate student from Tongji University, for helping us test the samples and process the raw data.

**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**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

### *Article* **Vitamin A Concentration in Human Milk: A Meta-Analysis**

**Huanmei Zhang 1,2, Xiangnan Ren 1,2, Zhenyu Yang 1,2 and Jianqiang Lai 2,3,\***

	- Beijing 100050, China

**Abstract:** Humans require vitamin A (VA). However, pooled VA data in human milk is uncommon internationally and offers little support for dietary reference intake (DRIs) revision of infants under 6 months. As a result, we conducted a literature review and a meta-analysis to study VA concentration in breast milk throughout lactation across seven databases by August 2021. Observational or intervention studies involving nursing mothers between the ages of 18 and 45, with no recognized health concerns and who had full-term infants under 48 months were included. Studies in which retinol concentration was expressed as a mass concentration on a volume basis and determined using high-, ultra-, or ultra-fast performance liquid chromatography (HPLC, UPLC, or UFLC) were chosen. Finally, 76 papers involving 9171 samples published between 1985 and 2021 qualified for quantitative synthesis. Results from the random-effects model showed that the VA concentration of healthy term human milk decreased significantly as lactation progressed. VA (μg/L) with 95% CI at the colostrum, transitional, early mature and late mature stages being 920.7 (744.5, 1095.8), 523.7 (313.7, 733.6), 402.4 (342.5, 462.3) and 254.7 (223.7, 285.7), respectively (X2 = 71.36, *p* < 0.01). Subgroup analysis revealed no significant differences identified in VA concentration (μg/L) between Chinese and non-Chinese samples at each stage, being 1039.1 vs. 895.8 (*p* = 0.64), 505.7 vs. 542.2(*p* = 0.88), 408.4 vs. 401.2 (*p* = 0.92), 240.0 vs. 259.3 (*p* = 0.41). The findings have significant implications for the revision of DRIs for infants under six months.

**Keywords:** vitamin A; retinol; human milk; full-term infant; lactation stage

#### **1. Introduction**

VA involves various physiological processes including retinal vision, gene expression, immune strength, reproduction, embryonic development, and growth [1,2]. Latest research shows that VA may have protective effects on outcomes of some viral infections such as HPV and measles [3]. Breastfed infants, particularly exclusively breastfed infants, acquire VA through human milk to achieve such needs. Breast-milk VA is a good indication of infants and lactating mothers' VA status, according to the World Health Organization (WHO) [4,5]. The WHO, European Food Safety Authority (EFSA), and other competent scientific organizations have established dietary reference VA values for infants and lactating women [2,6,7]. However, such recommendations by the WHO or EFSA were based on hypotheses, carried out with limited research and few subjects [8]. There have been quite a few publications about human milk VA concentrations worldwide; however, there is a wide range as shown in publications [9–11]. Meta-analysis methodology has been viewed as a beneficial method for statistically combining and summarizing the results from various studies, so as to obtain pooled data or estimates that may better represent what is true in the population [12]. However, relatively few systematic reviews and meta-analyses have been undertaken to synthesize VA concentration in human milk using international data [13,14], resulting in scarcity of updated and solid breast milk VA levels. Such data is

**Citation:** Zhang, H.; Ren, X.; Yang, Z.; Lai, J. Vitamin A Concentration in Human Milk: A Meta-Analysis. *Nutrients* **2022**, *14*, 4844. https:// doi.org/10.3390/nu14224844

Academic Editor: Robert B. Rucker

Received: 30 October 2022 Accepted: 11 November 2022 Published: 16 November 2022

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

**Copyright:** © 2022 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/).

critical for establishing dietary reference intakes (DRIs) for groups, particularly newborns and breastfeeding mothers [8].

To inform DRI revision for the group of healthy full-term infants aged 0 to 6 months, we conducted a meta-analysis study to analyze human milk's VA concentration on volume base, determined by advanced methods (HPLC, UPLC, or UFLC), and to explore the influence of potential confounders using meta-regression.

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

#### *2.1. Literature Search*

Articles in English were searched through PubMed, Web of Science, Embase and Cochrane Central Register of Controlled Trials employing matching keywords "vitamin A" or "retinol" with "human \* milk", "woman \* milk", "mother \* milk", "breast \* milk", "lactation" or "lactating". Articles in Chinese were searched through the China National Knowledge Internet (CNKI), Wan Fang Database, or China Science and Technology Journal Database (CSTJ) utilizing matching Chinese keywords covering "human milk" and "vitamin A", the former keyword including "human milk", "breast milk", or "lactating mother" and the latter including "vitamins", "vitamin A", "retinol", "nutrients" or "nutritional composition". Articles published by 21 August 2021 were taken into account.

#### *2.2. Study Selection and Screening*

Studies were considered if they matched the following criteria: (1) intervention or observational studies that reported the level of VA in human milk; (2) mothers were aged between 18 to 45 years; (3) healthy lactating mothers free of degenerative or metabolic illnesses; (4) full-term infants aged 0 to 48 months; and (5) high-, ultra-, or ultra-fast performance liquid chromatography (HPLC, UPLC or UFLC) determination method. Criteria for exclusion: (1) VA supplementation or particular diet intervention was used; (2) studies were presented as a review, case report, conference abstract, or proceedings without full-text articles, communication letters, texts described in language other than English or Chinese, duplicate publications, or full-text inaccessible; (3) preterm milk studies or data derived from a blend of preterm and term breast milk; (4) research with identical samples, or concentration data not in volume unit, inconsistent data, or unusual data; (5) no clear identification of lactation stage. EndNote version 20.3 (Clarivate Analytics (UK) Limited, London, UK) was used to screen and choose studies.

Data of Orhon et al. [15] at the colostrum stage, Vaisman et al. [16] at the transitional phase and Redeuil et al. [9] at the transitional and late mature stages were removed due to unusual data distribution. That of Eagle-Stone et al. [17] was omitted because participants in some regions likely took high-dose VA supplements three months before the survey. The supplementing effect on raised VA levels of human milk was considered durable within 6 months [4].

#### *2.3. Data Extraction*

Papers were chosen in the order of title, abstract and entire contents based on the inclusion and exclusion criteria stated above. Four detectives extracted and double-checked the data. A fifth investigator was consulted for assistance if there was any doubt throughout the selection process. First author, publication year, study design, study location, analysis methods, participant characteristics (sample size, mother age, lactation stage) and VA concentration of human milk sample were all extracted.

Data from various lactating stages were retrieved and included in cohort studies. A median point was chosen if more than one human milk sample subgroup was tested within the same lactation stage. In intervention studies, baseline data from the intervention and control groups were retrieved, and follow-up data from the control group were treated similarly to cohort research. When both full breast-milk samples and random samples were used in the same study, data of the former was extracted. All extracted data was recorded using Excel.

#### *2.4. Data Analysis*

The VA concentration data were given as mean± standard deviation of retinol. If retinyl palmitate levels were also given, they were converted in proportion. If provided as geometric mean and 95% CI, geometric mean was deemed identical to arithmetic mean values, and 95% CI was regarded arithmetic mean values. Zhang et al. [18] were consulted on the data transformation. The transformation of VA concentration in human milk from mass base to volume was multiplied by a factor of 1.032. Individual sample size formation was used in the computation when many samples were used in a single study, and weighted mean and standard deviation were required. The following are the relevant functions:

$$\text{Mean} = (\text{n1} \times \text{M1} + \text{n2} \times \text{M2} + \text{n3} \times \text{M3} + \dots \dots + \text{ni} \times \text{Mi})/(\text{n1} + \text{n2} + \text{n3} + \dots \dots + \text{ni}) \tag{1}$$

$$\mathbf{A}\_{\mathrm{i}} = \mathrm{S}\_{\mathrm{i}}^{\,2} \left( \mathbf{n}\_{\mathrm{i}} - \mathbf{1} \right) + \mathrm{M}\_{\mathrm{i}}^{\,2} \times \mathbf{n}\_{\mathrm{i}} \tag{2}$$

$$\text{SD} = \sqrt{\frac{\sum A\_i - \frac{\left[\sum \left(M\_i n\_i\right)\right]^2}{N}}{N - 1}} \tag{3}$$

The individual mean value, sample size and standard deviation of personal research are represented by Mi, ni and Si, respectively.

A weighted mean and standard deviation were determined for a common lactation stage. If the VA result of an individual study exceeded the range of Mean ± 2SD, the data was considered an outlier and was excluded. The other studies were then incorporated and integrated using meta-analysis with a random-effects model because there are differences between studies in both population and performance. Subgroup analysis at each lactation stage was conducted by countries: Chinese and non-Chinese studies.

The I<sup>2</sup> test with a significance level of α = 0.05 was used to visually analyze heterogeneity among studies regarding the human milk VA level and to quantify the magnitude of heterogeneity. Individual trials were examined using sensitivity analysis at each stage of breastfeeding. Sources of heterogeneity were assessed using meta-regression analysis, which included research design, publication year, mother's age, country, sampling time, sampling volume, and whether the breast was empty after sampled. R packages version 4.1.3 (10 March 2022) were used for meta-analysis and meta-regression.

#### **3. Results**

#### *3.1. Study Identification*

Database searching yielded 12,887 entries, including 11,089 abstracts in English and 1798 in Chinese (Figure 1). After all duplicates were removed, 9558 records were tested against the title and abstract. In addition, three articles were included during the paperchasing procedure [18–20]. To establish eligibility for inclusion in the review, we evaluated 118 full-text studies. Finally, 76 studies from 33 countries met the inclusion and exclusion criteria. They were assessed for data review, with seventy-one full-texts in English [9–11,16,20–87], four in Chinese [88–91], and one in Spanish but with abstract in English [74].

#### *3.2. Study Characteristics*

There was one human milk bank study, forty cross-sectional studies, sixteen randomizedcontrol studies, four cohort studies, eleven intervention studies, one cross-sectional study in parallel to one intervention study, and three longitudinal studies (Table 1), with six studies involving Chinese participants, sixty-nine studies involving non-Chinese participants and one study involving multinational participants. A total of 9171 human milk samples were included, with VA concentrations in colostrum, transitional and mature human milk determined in 2170, 719 and 6282 models, respectively. There were 4082 and 950 samples included as early mature and late mature human milk, respectively, yet 999 specimens with no clear indication whether they were early or late mature. There were 2053 Chinese and 5602 non-Chinese participants. These studies were published between Year 1985–2021.

**Figure 1.** Flow diagram of literature review.

#### *3.3. VA Concentration in Human Milk*

The VA concentration (μg/L) with 95% CI of human milk at colostrum, transitional, early and late mature stages was 920.7 (744.5, 1096.8), 523.7 (313.7, 733.6), 402.4 (342.5, 462.3) and 254.7 (223.7, 285.7) for all samples, respectively (Table 2). The VA concentration with 95% CI of mature human milk was 385.3 (339.4, 431.3) μg/L. Subgroup analysis by lactation stage showed there were significant difference between the colostrum, transitional and mature stages (X2 = 170.02, *p* < 0.01) (Figure S1) and between the colostrum, transitional, early and late mature stages (X<sup>2</sup> = 71.36, *p* < 0.01) (Figure S2).

At the colostrum stage, which is within 7 days following delivery, there were five studies performed on 429 Chinese participants and twenty-two studies completed on 1741 non-Chinese subjects (Figure 2). The VA content with 95% CI was 1039.1 (470.3, 1607.8) μg/L in Chinese specimens and 895.75 (714.1, 1077.4) in non-Chinese. There was no statistically significant variance between the two population groups (X<sup>2</sup> = 0.22, *p* = 0.64).


Summary Characteristics of included studies.

> **Table 1.**

#### *Nutrients* **2022** , *14*, 4844


**Table 1.** *Cont.*


**Table 1.** *Cont.* single sampling and specified daytime but not clarify exact sampling hours.




**Figure 2.** Forest plot of colostrum VA concentration and subgroup analysis between Chinese and non-Chinese samples.

At the transitional stage, which is postpartum 8–14 days, there were three studies carried out among 356 Chinese subjects and three studies among 363 non-Chinese subjects (Figure 3). The VA concentration with 95% CI was 505.7 (118.0, 893.4) μg/L for Chinese and 542.2 (278.9, 805.6) for non-Chinese samples, respectively. There was no significant difference between the two populations (X<sup>2</sup> = 0.02, *p* = 0.88).

There were seven studies conducted among 1268 Chinese subjects and fifty-three studies among 5014 non-Chinese participants at the mature human milk stage (Figure S3), covering seven studies with 1112 Chinese and thirty-eight studies with 3221 non-Chinese at early mature stage (Figure 4), one study with 156 Chinese and five studies with 794 non-Chinese subjects at late mature stage (Figure 5). The VA concentration with 95% CI between Chinese and non-Chinese participants was 386.4 (270.6, 502.3) μg/L vs. 385.2 (335.1, 435.3) μg/L at the mature stage (X2 = 0.00, *p* = 0.98), 408.4 (282.6, 534.1) μg/L vs. 401.2 (333.6, 468.8) μg/L

at the early mature stage (X2 = 0.01, *p* = 0.92) and 240.0 (214.9, 265.1) μg/L vs. 259.3 (220.8, 297.8) μg/L at the late mature stage (X<sup>2</sup> = 0.68, *p* = 0.41). There was no significant difference when comparing population subgroups at the mature, early mature or late mature lactation stage.

**Figure 3.** Forest plot of transitional human milk VA concentration and subgroup analysis between Chinese and non-Chinese samples.

#### *3.4. Heterogeneity and Sensitity Analysis*

All analyses revealed substantial heterogeneity (I2 in 85~100%). Following sensitivity testing, no significant change in the combined effect of VA levels was seen at each lactation stage, indicating that all respective synthesized results was stable.

#### *3.5. Meta-Regression*

The results of the univariate meta-regression analysis revealed that none of the following, i.e., publication year, sampling time, whether emptying breast or not after sampling, whether Chinese or not, or study design type at each lactation stage, were significantly associated with heterogeneity between studies (all *p* > 0.05) except maternal age (≥30 years vs. <30 years) and nationality (Table A1). The explained heterogeneity of country changed very little following correction for maternal age, i.e., 54.58% to 54.20% at early mature human milk stage (both *p* < 0.0001), but the effect of maternal age changed from significant to insignificant, i.e., its *p* value being increased from 0.025 to 0.051. Equally, at the colostrum stage, the explained heterogeneity of country after correction resulted in a minute change, i.e., 21.56% to 25.94% but with the *p* value decreasing from 0.34 to 0.044, whereas the impact of maternal age changed from significant to insignificant again, with the *p* value increasing from 0.041 to 0.48. The results of the multivariate meta-regression study results suggested that country was a source of heterogeneity, while maternal age was not.


**Figure 4.** Forest plot of early mature human milk VA concentration and subgroup analysis between Chinese and non-Chinese samples.

**Figure 5.** Forest plot of late mature human milk VA concentration and subgroup analysis between Chinese and non-Chinese samples.

#### **4. Discussion**

In this study, we compiled previously published data on retinol concentrations in term human milk at each step of the four lactation stages and compared them between Chinese and non-Chinese studies. Our research comprised 76 articles, including 9171 participants from 33 countries from 1985 to 2021, for calculating human milk VA levels determined by HPLC, UPLC, or UFLC. At the colostrum, transitional, early mature and late mature stages of human milk, the VA levels were 920.7 μg/L (3.21 μmol/L), 523.7 μg/L (1.83 μmol/L), 402.4 μg/L (1.40 μmol/L), and 254.7 μg/L (0.89 μmol/L), respectively. There was no significant difference in the VA levels between Chinese and non-Chinese human milk at each lactation stage. This research has crucial implications for DRIs VA modification.

#### *4.1. Data Interpretation*

Our findings were compatible with previous meta-analysis findings on VA levels in human milk and the declining tendency with the lactation stage. Dror et al. [14] examined retinol levels in colostrum in four included studies and the mature stage in twenty-four studies (21~365-day lactation). The systematic approach chose the retinol-to-fat ratio (μmol/g fat) as the primary outcome measure which led to nearly two-thirds of the relevant literature being excluded for the meta-analysis. Despite this, the outcomes of our research at the two segmental stages were similar to those of Dror et al., namely 920.7 μg/L vs. 999.7 μg/L, 385.4 μg/L vs. 383.8 μg/L. de Vries et al. [13] conducted a systematic review of 11 studies on the relationship between colostrum VA and maternal serum (plasma) vitamin concentration but did not carry out a meta-analysis. As a result, our findings have greater precision and comprehensiveness.

The respective wide data distribution could explain the similar VA level of human milk at between Chinese and non-Chinese individuals at each lactation stage. Typically, the samples by Zhang et al. [91] were from 20 counties in 11 provinces across China, including urban and rural locations. In contrast, the non-Chinese samples came from 32 nations, comprising both developed, developing, and under-developed ones. Our multivariate meta-regression results, on the other hand, showed that the country factor explained more than 50% of the heterogeneity, implying that the remaining variation between studies could be due to factors such as VA intake, maternal status or sampling protocol rather than the insignificant factors such as study design, publication year, and so on that we analyzed here. Previous research has shown that inadequate dietary VA consumption, maternal VA status during pregnancy and lactation all contribute to clinical heterogeneity, while breast milk sampling protocol accounts for methodological heterogeneity [53,63,66,70,92].

#### *4.2. Implications of Our Results for DRIs Revision*

Two studies reported mean liver VA concentration of perinatal normal-weight newborns [93,94], one being 17.3 ± 17.4 μg/g liver in Thai fetuses in gestational age of 37–40 week (*n* = 10), the other being 22 ± 26 μg/g liver in USA infants aged 0–6 days (*n* = 22). Assuming that the liver represents 4.3% of body weight and the liver VA concentration is 20 μg/g, a 3.2-kg full-term newborn has stores of 2.8 mg VA. In contrast, an exclusively breast-fed infant consumes approximately 54.3 mg of VA from mother's milk (402.4 μg/L × 0.75 L/day × 180 days). About 19.4 times more VA is transferred from a mother to a baby during the 6 months of lactation than is accumulated by the fetus during 9 months of gestation. Obviously, the VA in breast milk is of paramount importance for maintaining adequate VA status in early postnatal life of infants as compared to accumulation of VA in the liver prenatally.

A proper estimation of human milk VA level is critical for reference setting in terms of dietary adequate intake requirement for the population of exclusively breastfed infants under six months of age to guarantee optimal growth and development of the newborns. Accurate VA adequacy information for newborns and nursing mothers is desperately needed [8]. As a result, we proposed that the VA content in early mature human milk expressed as a mean with a 95% CI of 402.4 (342.5, 462.3) μg/L or 1.40 (1.20, 1.61) μmol/L, could be used as data support for the purpose. First, a VA level in human milk greater than >1.05 μmol/L may prevent clinical VA deficiency during the first six months of infancy [5]. Second, in this investigation, the synthesis result of VA level at the early mature stage showed less variance than colostrum and transitional phase and had a greater level than at later mature stage, indicating a better representative of human milk VA level. Third, an equilibrium of VA secretion appeared to be obtained in early mature human milk for human beings, as evidenced by the relatively comparable levels of VA in both Chinese and non-Chinese participants at this stage. It is worth noting that 923 mother-infant dyad participants in the Zhang 2021 [91] study had generally adequate nutrition and health status. The comparable VA concentration in early mature human milk is 0.25 mg/L (0.87 μmol/L). This threshold is far lower than the level proposed here. Fourth, past values presented by authoritative groups might have been overstated. The current acceptable intake level of VA for infants aged 0 to 6 months established by EFSA or IOM was based on the average amount of VA consumed in humans [1,2]. However, if the figures are derived from a small number of articles, there is a risk of overestimating the average demand for this group. The EFSA limit of 530 μg/L, chosen as the midpoint of a range of averages (229–831) μg/L, was based on five studies conducted in western countries that did not differentiate between early mature and later-stage human milk. Based on four investigations [19–21,32], the IOM established 485 μg/L (1.70 μmol/L) as the VA level in human milk in 2001 and adopted the level from one of the studies, which was undertaken among three healthy, wellnourished mothers within 75~277 days postpartum [21]. Our data analysis included these four research studies, whereas one study [19] was omitted due to outdated methodology. According to EFSA [7], the average levels of total VA concentration in western countries have generally been estimated to be between 450 and 600 μg/L. Nevertheless, we proposed a reference concentration range of VA in human milk of 402.4 (95% CI: 342.5, 462.3) μg/L for DRIs VA modification for infants ≤ 6 months of age. The range could also be helpful to nursing women and to optimizing the VA level of infant formula.

#### *4.3. Study Limitations*

Since the availability of the individual studies limited our meta-analysis of studies at the transitional and later mature lactation stages, the influence of country variability on human milk VA level at transitional stage may be weak, and it was not possible to assess it at the later mature stage. As a result, the meta-regression results for these two stages should be interpreted with caution.

#### **5. Conclusions**

The current study found that synthesized human milk VA levels decreased as breastfeeding progressed and that there was no significant difference in human milk VA levels between China and other countries, even though country played a vital role in the variation. Our findings have important implications for DRIs VA revision for the population of exclusively breastfed infants under six months.

**Supplementary Materials:** The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/nu14224844/s1, Figure S1: Forest plot of VA in human milk by 3 lactation stages; Figure S2: Forest plot of VA in human milk by 4 lactation stages; Figure S3: Forest plot of VA in mature human milk by population.

**Author Contributions:** H.Z. worked on the study design, data extraction, analysis, interpretation and manuscript writing. X.R. contributed to the study design, data extraction, data check and editing. Z.Y. contributes the study design, data check, data analysis, data interpretation and manuscript review. J.L. contributed to funding acquisition, study design and manuscript review. All authors have read and agreed to the published version of the manuscript.

**Funding:** CNS Research Fund for DRIs.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available in the inserted articles.

**Acknowledgments:** We are thankful for the assistance by Ai Zhao and Yuandi Xi on data extraction.

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

#### **Appendix A**

**Stage 1 Stage 2 Stage 3 Stage 4** *p* **R<sup>2</sup>** *p* **R<sup>2</sup>** *p* **R<sup>2</sup>** *p* **R2** Country 0.066 21.56% 0.49 0.00% <0.0001 54.58% - & - Population (Chinese vs. Non-Chinese) 0.59 0.00% 0.81 0.00% 0.97 0.00% 0.68 0.00% Publication year 0.94 0.00% 0.91 0.00% 0.88 0.00% 0.82 0.00% Sampling time 0.24 3.39% 0.06 33.90% 0.29 1.88% 0.11 45.00% Study design 0.15 6.59% 0.052 49.13% 0.53 0.00% 0.295 1.62% Whether emptying breast or not 0.23 2.18% 0.49 0.00% 0.37 0.00% 0.12 36.45% Maternal age (<30 years vs. ≥30 years) 0.019 17.65% - \$ - 0.025 9.50% 0.015 66.48%

**Table A1.** Univariate Meta-regression of enrolled studies for heterogeneity source.

Note: Lactating stage: 1 colostrum; 2 transitional human milk; 3 early mature human milk; 4 late mature human milk. R2: Accounted for amount of heterogeneity. & Unable to make analysis due to limited number of studies. \$ Unable to make analysis due to same age stratification.

#### **References**

