**3. Results**

Between 2014 and 2019, 293 patients fulfilling the inclusion criteria were analysed. The median age was 64 years, 79.9% were male, and most of the patients (70%) presented adenocarcinoma. A biopsy sample was available in 127 (43.3%) patients *(n* = 64 in Untreated Cohort (#1) and *n* = 63 in Treated Cohort (#2)). PD-L1 expression was deemed positive in 58.3% cases *(n* = 31 in Untreated Cohort (#1) and *n* = 43 in Treated Cohort (#2)). Table 2 depicts patients' characteristics in the Untreated and Treated cohorts.


**Table 2.** Clinical and pathological data according to the testing cohorts.

n.a.—not applicable; IQR – Interquartil Range.

In the Treated Cohort (#2), 18 patients whose tumours showed ≥50% PD-L1 expression were treated with pembrolizumab in first-line; 19 and 3 patients whose tumours had 1–49% PD-L1 expression were treated with pembrolizumab and nivolumab, respectively, in second-line after progression on chemotherapy. Eighteen patients with PD-L1 negative tumours were treated with nivolumab as a second-line treatment. Four patients with adenocarcinoma carried driver mutations (3 had an *EGFR* tyrosine kinase mutation and 1 had an *ALK* gene rearrangement). As such, anti-PD-1 therapy was administered as a third-line treatment, after progression on tyrosine kinase inhibitors (first-line) and chemotherapy (second-line).

Regarding molecular analysis, *RAD51Bme* levels were significantly higher in PD-L1 positive vs. negative cases in both cohorts (Untreated Cohort (#1)—*p* = 0.0216; Treated Cohort (#2)—*p* < 0.0001) (Figure 1). Patients presenting higher *RAD51Bme* levels showed a higher chance of having a positive PD-L1 immunoexpression (Untreated cohort (#1) OR: 51.68, 95% CI: 1.77–1512.04, *p* = 0.022; Treated cohort (#2) OR: 45.51, 95% CI: 5.29–391.20, *p* = 0.001), adjusting for sex, age, smoking status and histological subtype (detailed information in Table S1). No differences in *RAD51Bme* levels were found between squamous cell carcinoma and adenocarcinoma cases in both cohorts (Untreated Cohort (#1)—*p* = 0.774; Treated Cohort (#2)—*p* = 0.520).

**Figure 1.** *RAD51B* promoter methylation levels within PD-L1 negative and PD-L1 positive immunoexpression among NSCLC samples. Scatter plot representing *RAD51B* promoter methylation levels distribution obtained by qMSP for (**A**) Untreated Cohort (#1) and (**B**) Treated Cohort (#2) patients, according to negative and positive PD-L1 immunoexpression. Mann–Whitney U-test. Red horizontal line represents the median methylation levels.

*RAD51Bme* levels were significantly higher in patients submitted to immunotherapy, which demonstrated clinical benefit (*p* = 0.0390; Figure 2A). Moreover, patients with positive *RAD51Bme* levels (*RAD51Bme*<sup>+</sup> was consider when methylation levels >P75) disclosed clinical benefit independently from PD-L1 expression (Figure 2B). Additionally, *RAD51Bme* discriminated between PD-1 blockade clinical benefit and no clinical benefit with 85% specificity and 90% positive predictive value (AUC: 0.758, 95% CI: 0.626–0.889, *p* = 0.0015; Figure 2C and Table 3). Remarkably, combining *RAD51Bme*<sup>+</sup> with PD-L1<sup>+</sup> improved the sensitivity of the test (68%) to predict immunotherapy response, maintaining high specificity (85%) and increasing positive predictive value (94%).

**Table 3.** *RAD51Bme*, PD-L1 staining and the combination of the two variables performances as predictive biomarkers of PD-1 blockade response in the Treated Cohort (#2).


Abbreviations: PPV: positive predictive value, NPV: negative predictive value.

**Figure 2.** *RAD51Bme* levels and PD-L1 positivity associate with PD-1 blockade clinical benefit. (**A**) Scatter plot representing *RAD51B* promoter methylation levels distribution obtained by qMSP in patients with and without clinical benefit from immunotherapy. Mann–Whitney U-test. Red horizontal line represents the median methylation levels; (**B**) Contingency graph displaying the percentage of patients with and without PD-1 blockade clinical benefit, according to *RAD51B* promoter methylation and PD-L1 status. Chi-square test. *RAD51Bme* were considered positive when promoter methylation levels >P75; (**C**) Receiver operator characteristic (ROC) curve for discrimination between patients with and without clinical benefit from immunotherapy based on *RAD51B* promoter methylation levels distribution in the Treated Cohort (#2).

The median follow-up time for the Treated Cohort (#2) was 18 months (95% CI: 15.1–20.9). The median PFS was significantly higher in *RAD51Bme*<sup>+</sup> patients (*p* = 0.0216; Figure 3A). Furthermore, patients with *RAD51Bme*<sup>+</sup> disclosed a lower risk of disease progression (HR 0.37; 95% CI: 0.15–0.88; *p* = 0.025) compared with *RAD51Bme-*. Considering the PD-L1 expression, no significant differences were depicted for PFS (*p* = 0.2023), although PD-L1<sup>+</sup> patients disclosed a trend for higher PFS (Figure 3B). Nonetheless, PD-L1<sup>+</sup> associated with a longer OS (*p* = 0.0307) and a lower risk of death (HR 0.35; 95% CI: 0.15–0.81; *p* = 0.014). For *RAD51B*, lower methylation levels tend to associate with shorter OS, despite not being statistically significant. Also, no significant differences were observed for PFS or OS, when combining in panel PD-L1 expression and *RAD51Bme* levels.

**Figure 3.** Kaplan–Meier survival curves for progression-free survival (after first anti-PD-1 treatment) of patients according to (**A**) *RAD51Bme* status; (**B**) PD-L1 status; and (**C**) combined *RAD51Bme* and PD-L1 status. Kaplan–Meier survival curves for patients' overall survival according to (**D**) *RAD51Bme* status, (**E**) PD-L1 status, and (**F**) combined *RAD51Bme* and PD-L1 status. Log-rank test. *RAD51Bme* was considered positive when promoter methylation levels >P75.

### **4. Discussion**

Despite the improvement in lung cancer treatment over the last years, it remains a lethal disease in most cases, mostly due to diagnosis at advanced stages and suboptimal effectiveness of standard therapy. Nonetheless, the emergence of novel therapeutic strategies, including immune-based cancer therapies, has improved the prospects of patients diagnosed at advanced stages of the disease. Indeed, anti-PD-1 treatment for advanced NSCLC has improved the survival of patients [22]. Currently, the most commonly used biomarker to predict this response to anti-PD-1 therapy is PD-L1 immunostaining, although a substantial number of patients with PD-L1 positive immunostaining do not respond [21], highlighting the need for new biomarkers. In NSCLC, similar to other tumours, a higher tumour mutation burden was a strong predictor of immunotherapy efficacy [25–28]. Additionally, defects in the HRR pathway have been associated with higher expression of co-regulatory molecules such as PD-L1, suggesting that deficient homologous recombination, by disabling repair of DNA defects, may lead to neoantigens production with the recruitment of T-cells to the tumour microenvironment. This engages tumour cells to upregulate the expression of PD-L1 as an adaptive resistance mechanism [29]. A recent study demonstrated that DNA methylation profile of NSCLC might also be determinant for the efficacy of anti-PD-1 treatment in stage IV patients [30]. Furthermore, epigenetic alterations in *RAD51B*, specifically DNA promoter methylation, were associated with PD-L1 expression in squamous cell carcinomas [18]. This is a *RAD51* paralog, essential for DSB repair in the homologous recombinant pathway [17]. Thus, we sought to investigate the association of immune checkpoint PD-L1 expression and DNA methylation status of DNA repair gene *RAD51B* in non-small cell lung cancer (NSCLC), correlating with patient outcome.

Overall, the chances of positive PD-L1 expression in advanced NSCLC increased with the level of *RAD51me*+. Remarkably, a link between *RAD51Bme* and the immune response in NSCLC has been previously suggested [29]. Furthermore, *Rieke et al.* demonstrated that methylation was associated with low mRNA expression levels and with homologous recombination deficiency [18]. Additionally, a significant positive correlation between *RAD51B* methylation status and the inflammatory gene signature, particularly, interferon-gamma (IFN-γ) was disclosed [18]. Interestingly, IFN-γ is an

important inducer of PD-L1 expression, which acts via the JAK/STAT1/interferon regulatory factor (IRF) [31] in various types of cancers, including NSCLC. Furthermore, the depletion of *RAD51B* was shown to induce immune response through activation of the STAT3 pathway [32], which activates *CD274* gene/PD-L1 induction [31,33]. Therefore, our results further support the link between homologous repair deficiency by epigenetic regulation and immune checkpoint players, specifically PD-L1. Considering the available literature, assessing the inflammatory profile of these tumours might be useful to determine whether there is a direct effect between DNA repair candidate genes hypermethylation and the expression of immune checkpoint proteins.

Remarkably, *RAD51Bme*<sup>+</sup> associated with better clinical response to treatment with PD-1 blockade and to a reduction of disease progression by 60%. Conversely, *RAD51Bme-* associated with the absence of clinical benefit, which was even more relevant in negative PD-L1 expression cases. Hence, *RAD51Bme* might constitute a potential biomarker of response to anti-PD-1 therapy. Although *RAD51Bme* depicted lower sensitivity than PD-L1<sup>+</sup> as a predictive biomarker for treatment with anti-PD-1, it displayed higher specificity.

Although PD-L1 expression has not been described as a strong prognostic factor mostly due to methodological approaches variations, including diverse immunohistochemistry antibodies, dissimilar evaluation for PD-L1 positivity (cut-off % or H-score) and patients' selection [13,34], in our study, both PD-L1<sup>+</sup> and *RAD51Bme*<sup>+</sup> associated with better overall survival. Conversely, another research team suggested that *RAD51B* overexpression associates with improved OS in NSCLC patients [35]. Notwithstanding higher promoter methylation levels might entail expression downregulation, several other genetic and epigenetic mechanisms may contribute to this apparent inconsistency. Furthermore, higher *RAD51B* methylation status was depicted in patients with longer progression-free survival after anti-PD-1 treatment, supporting once more the clinical benefit of PD-1 blockade when *RAD51B* promoter is methylated. The shorter overall survival of non-smokers patients may be partially explained by the fact that these patients had a longer median time (higher than 20 months) between diagnosis and the treatment with PD-L1 inhibitors than smokers.

Therefore, PD-L1<sup>+</sup> and *RAD51Bme*<sup>+</sup> are promising biomarkers to predict response to PD-1 blockade rather than overall prognostic factors in NSCLC's patients. As such, *RAD51Bme* might represent a new predictive marker potentially assessable in liquid biopsies, allowing for a better selection of patients for anti-PD-1 treatment and eventually for monitoring patients' immunotherapy response throughout the course of the disease. Although our study paves the way for new prospective studies on the *RAD51B* promoter methylation's predictive role in patients with NSCLC treated with anti-PD-1, the retrospective design and small sample size are not neglectable limitations. Nevertheless, all the patients and samples enrolled in the study were analysed using the same criteria both for molecular biology strategies or clinical and pathological data collection. Importantly, other strengths of our research work are the fact that all patients were uniformly treated at the same institution, and all were evaluated by computed tomographic scans at specific timepoints during the course of treatment.
