A Gene Mutation Signature Predicting Immunotherapy Benefits in Non-Small Cell Lung Cancer Patients

Abstract

Background

Identification of patients who can benefit from immune checkpoint blockade (ICB) therapy is key for improved clinical outcome. Recently, US FDA approved tumor mutational load high (TMB-H, or TMB≥10) as a biomarker for pembrolizumab treatment of solid tumors. We intend to test the hypothesis that mutations in select genes may be a better predictor of non-small cell lung cancer (NSCLC) response to ICB therapy than TMB-H.

Methods

We compiled a list of candidate genes that may predict for benefits from ICB treatment by use of data from a recently published cohort of 350 NSCLC patients. We then evaluated the influences of different mutation signatures in the candidate genes on ICB efficacy. They were also compared with TMB-H. The predictive powers of different mutation signatures were then examined in an independent cohort of ICB-treated non-small cell lung cancer patients.

Results

A compound mutation signature, where two or more of the 52 candidate genes were mutated, accounted for 145 of 350 NSCLC patients and was associated with significant ICB treatment benefits. Specifically, the median duration of overall survival (OS) was 36 vs 8 months in NSCLC in those with two or more vs none of the 52 genes mutated. Moreover, those patients with the compound mutation signature but low TMB (<10) achieved significant OS benefits when compared with those without the signature but TMB-H (≥10). Finally, in an independent cohort of 156 ICB-treated NSCLC patients the median duration of progression free survival was 8.3 months vs 3.5 months in those with the compound mutation signature vs those with none mutated in the 52 genes.

Conclusion

A genetic signature with mutations in at least 2 of 52 candidate genes was superior than TMB-H in predicting clinical benefits for ICB therapy in NSCLC patients.

Introduction

There is great excitement over immune checkpoint blockade (ICB) treatment for cancer because of the existence of durable responders among advanced stage patients^1–6. However, only a minority responds to ICB treatment among eligible candidates⁷. Because of the costs associated with ICB treatments and potentially severe side effects in some treated patients^{8, 9}, there is an unmet need to discover biomarkers that allow for more precise identification of patients who can benefit.

The most widely used biomarker for ICB therapy is PD-L1. Both first-line¹⁰ and second-line¹¹ pembrolizumab treatment of NSCLC was approved by FDA based on PD-L1 expression in tumor cells. However, other ICB agents such as nivolumab, atezolizumab, and durvalumab showed significant clinical benefits in NSCLC treatment without requirement for specific PD-L1 levels^12–17. In the first line setting, nivolumab did not improve PFS in patients with high PD-L1 levels¹⁸. Therefore, PD-L1 level has so far been inconsistent in predicting ICB efficacy.

Another well-established genetic predictor of tumor response to ICB treatment is MSI (microsatellite instability)^19–21. Microsatellite instability occurs in a subset of malignancies because of deficiencies in mismatch repair genes. MSI⁺ tumors were associated with significantly better response to ICB therapy.^{19, 20} As such, MSI⁺ status has been approved as a clinical biomarker for ICB therapy irrespective of tumor origin. The utility of MSI⁺ status as a biomarker for ICB therapy was consistent with the idea that neoantigens were responsible for positive responses to immunotherapy²². The premise is that more mutations in the cancer genome were more likely to generate more tumor-specific neoantigens that could promote CD8⁺T cell mediated tumor cell killing²³. Based on this concept, tumors without clear evidence of MSI but nonetheless had high levels of tumor mutational burden (TMB) were examined for their responses to ICB therapy. Indeed, it was demonstrated that within many different types of cancer, high TMB predicted for better response to ICB therapy in multiple cancer types^24–30. However, compared to MSI status, TMB as a biomarker for ICB treatment has been difficult to implement because high or low TMB is relative for each tumor type³¹ and the cut-offs appeared to be different for various types of cancers²⁵. Despite these issues, the US FDA recently approved pembrolizumab treatment of adult and pediatric solid tumors with high TMB-H (≥10)³². However, given ambiguities in previous studies, the suitability of TMB-H for selecting patients requires additional clinical evidence.

In the present study, we evaluated the hypothesis that mutations in certain genes could promote anti-tumor immune response beyond merely serving as sources of neoantigens and could thus predict ICB efficacy better than TMB. Our goal was to identify candidate genes whose mutations can predict for ICB clinical benefits more effectively than TMB.

Methods

Source of patient treatment and mutation data

All patient data were previously published and obtained from the cBioPortal database (https://www.cbioportal.org)³³, which includes more than 160 studies from The Cancer Genome Atlas (TCGA), International Cancer Genome Consortium (ICGC), Memorial Sloan-Kettering Cancer Center (MSKCC) and other sources within the timeframe of our study (11/17/2019–7/15/2020).

We identified candidate genes that might influence ICB treatment of NSCLC by use of data from a recently published cohort of patients treated with ICB therapy at the Memorial Sloan Kettering Cancer Center (MSKCC)²⁵. It included 1661 patients who underwent ICB treatment and we referred to it as the MSK-TMB cohort hereinafter. We also analyzed mutation and survival data from some of the lung cancer patients in the 10,336-patient MSK-IMPACT Clinical Sequencing Cohort. Mutations in both the MSK-TMB and MSK-IMPACT cohorts were derived by several different versions of the IMPACT targeted sequencing assay^{25, 34, 35}.

In addition to patient clinical and genomic data from the MSK-TMB and MSK-IMPACT cohorts, we also used data from several other published studies on ICB treatment of NSCLC^{28, 34, 36, 37} to validate the predictive power of the candidate genes. Additional tumor mutation and transcriptome profiling data were obtained from NSCLC patients in the TCGA Pan-Cancer Atlas cohort³⁸.

Mutation signatures and survival analysis

To analyze if mutations in 52 candidate genes could predict ICB treatment efficacy, we grouped patients into those with no mutations in the 52 genes (wild type, or wt group), those with mutations in one or more of the 52 genes (mut group), those with only a single mutation in any of the 52 genes (single mutation, or single mut group), and those with 2 or more mutations in the 52 genes (compound mutation, or compound mut group). Overall or progression-free survival of patients with different mutation signatures were then compared in the MSK-TMB cohort²⁵ and additional cohorts^{28, 36, 37}.

The relationship between the compound mutation signature and PD-L1 expression

We determined potential interactions between PD-L1 expression and the compound mutation signature in influencing patient response to ICB therapy by examining a sub-cohort of the MSK-TMB NSCLC cohort that was published. Kaplan-Meier survival analysis was conducted of patients stratified by both their mutation signature and PD-L1 expression levels³⁷.

CIBERSORT estimate of the intratumoral lymphocyte infiltration

To determine if different mutation signatures correlate with specific characteristics of the tumor microenvironment, we carried out CIBERSORT³⁹ analysis of NSCLC patients based on their RNA expression data. An published online tool TIP⁴⁰, based on CIBERSORT principles, was used for the immune cell subset analysis of the tumor samples.

Statistical analysis

For survival analysis, Kaplan-Meier survival curves were generated by use of the statistical software GraphPad Prism (version 8.2). Nonparametric Mantel-Cox logrank test was used to determine the differences among different patient groups. Hazards ratio (HR) and the 95% confidence interval (CI) were calculated by use of the logrank (Mantel-Cox) test and adopting the Cox proportional-hazards model.

Results

We hypothesized that gene mutations not only produced neoantigens but also could functionally affect the outcome of immunotherapy. Based on this hypothesis, we attempted to identify genes whose mutations could positively influence ICB treatment efficacy by investigating 350 NSCLC patients among a recently published cohort ICB-treated patients²⁵ (see Appendix Table A.1 for more patient information). These patients were sequenced with the MSK-IMPACT targeted sequencing panel^{34, 35}, which contained around 400 genes known to be involved in cancer development. To compile our list of candidate genes, we used the following criteria: 1) the gene has to be mutated in at least 3 or more cancer patients (out of a total of 350); 2) mutations in the gene has to be correlated with a survival benefit, e.g. its mutation frequency has to be significantly higher among the surviving patients (at the end of clinical observation period) than that among the deceased patients (with a p value ≤0.10 instead of p<0.05 due to small number of patients for some genes). We only focused on nonsysnomous mutations and excluded gene fusions and amplifications. Applying these criteria to the targeted sequencing gene panel used in the study, we were able to obtain a 52 gene panel (Table 1). Among the 350 NSCLC patients, 230 had mutations in at least one of the 52 genes.

Table 1.

List of 52 genes whose mutations influence immunotherapy outcomes.

Gene name (% with mutations in MSK-TMB NSCLC cohort)
ABL1(2.0%), ASXL1(3.4%), ATM(6.6%), BCOR(4.0%), BRCA2(4.3%), BRIP1(3.7%), CARD11(4.0%), CD79B(1.4%), CDC73(0.9%), CIC(3.7%), EPHA3(10.6%), EPHA5(8.3%), EPHA7(5.1%),	EPHB1(3.7%), ERBB4(6.0%), ERCC4(2.0%), FGFR4(2.6%), FLT3(2.6%), FOXL2(1.4%), HGF(6.3%), INHBA(2.9%), JAK3(3.1%), MAX(2.0%), MDC1(4.0%), MED12(3.4%), MET(4.0%),	MGA(5.7%), MRE11(2.0%), MSH2(2.0%), NF2(3.4%), NFKBIA(0.9%), NOTCH1(4.0%), NOTCH2(3.1%), NTRK3(5.7%), NUF2(0.9%), PARP1(1.1%), PAX5(1.1%), PGR(4.3%), PIK3C2G(4.0%),	PIK3C3(3.7%), PIK3CG(7.7%), PIM1(0.9%), POLE(4.6%), PPM1D(1.4%), PPP2R1A(2.0%), PTPRD(12.3%), RET(2.9%), STAT3(0.9%), TENT5C(0.9%), TET1(2.3%), TSC2(2.6%), ZFHX3(7.7%)

We next determined the predictive power of different types of mutation signatures in the 52 genes by dividing the NSCLC patients in the MSK-TMB cohorts into the following three sub-cohorts with different mutation signature cohorts: those without mutations in the 52 genes (wt), those with only 1 of the 52 genes mutated (single), and those with 2 or more of the 52 genes mutated (compound). We compared the OS among the three cohorts (Fig. 1A). Our data indicated that the median OS was 36 months among the 145 patients in the compound mutation group (vs 8 months for wt group, p=3.7e⁻⁸, logrank test). Importantly, the single mutation group was not significantly different from the wt group in terms of overall survival (p=0.29, log rank test). These data therefore showed significant overall survival advantages of the compound mutation group when compared with the wt or single mutation groups.

Fig. 1. The compound mutation signature in the 52-gene panel predicts for significantly better response to ICB treatment in NSCLC.

A). OS of 350 ICB-treated MSK-TMB NSCLC patients with wt, single, and compound mutation signatures. Compound vs wt: Hazard ratio (HR) = 0.43 (95% confidence interval [CI]: 0.31–0.59).

B). PFS of 69 ICB-treated NSCLC patients with wt, single, and compound mutation signatures. These patients belong to a sub-cohort of the MSK-TMB NSCLC cohort in A for whom PFS data are available. Compound vs wt: HR = 0.41 (95% [CI]: 0.22–0.79).

Wt: patients with no mutations in the 52 gene panel; single: patients with only a single gene mutation in the 52-gene panel; compound: patients with two or more of genes mutated in the 52-gene panel.

P-values calculated by use of logrank test.

We further compared the influences of the mutation signatures on profession free survival (PFS) in a sub-cohort 69 MSK-TMB NSCLC patients where PFS data were available³⁷. Our analysis indicated that 77 patients with the compound mutation signature had a median PFS of 8.33 months (vs 2.52 months for the wt group, p=0.0013, logrank test) (Fig. 1B). On the other hand, PFS for the single mutation group was 2.37 months (p=0.77 when compared with wt group). Therefore the association patterns between the mutation signatures and PFS was similar to that of the OS data.

We next attempted to validate the predictive power of the compound mutation signature in an independent cohort of 156 NSCLC patients from previously published studies (see Appendix Table A.2 for more patient information)^{28, 36, 37}. Different from the MSK-TMB cohort, these ICB-treated patients were sequenced by WES (whole exome sequencing) for gene mutations. We then conducted survival analysis of patients with different mutation signatures in the 52-gene panel. Our results showed patients with the compound mutation signature showed significant survival advantages (Fig. 2). The median progression free survival (PFS) was 3.5 months, 3.7 months, and 8.3 months in the wt, single, and compound mutation groups in NSCLC patients (Fig. 2). These were remarkably similar to the PFS data obtained from the MSK-TMB cohort (Fig. 1B), thereby validating the predictive power of the compound mutation signature independently.

Figure 2. Validation of the predictive value of the compound mutation signature in an independent cohort of ICB-treated NSCLC patients.

Kaplan-Meier survival curve of 156 ICB-treated NSCLC patients with wt, single, or compound mutation signatures. HR (Compound vs wt): 0.42 (95% [CI]: 0.24–0.71). P values calculated by use of logrank test.

One important question is whether the observed clinical benefits in the compound mutation group were genetically inherent for the patients and therefore prognostic instead of predictive of clinical benefits from ICB treatment. In order to answer this important question, we studied a sub-cohort of NSCLC patients in the MSK-IMPACT cohort^{34, 35} who had not been treated with immunotherapy but were otherwise very similar in patient characteristics to the MSK-TMB lung cancer cohort (Appendix Table A.3). Our analysis indicated that in the absence of ICB treatment, the survival benefits observed in the compound mutation group were not observed (Fig. 3). In fact, the compound and single mutation group had worse survival than the wild type group. Therefore, we conclude that the compound signature was predictive for NSCLC ICB therapy instead of prognostic for NSCLC in general.

Fig. 3. Non-prognostic nature of the compound mutation signature in non-ICB treated NSCLC patients from the MSK-IMPACT cohort.

Kaplan-Meier analysis of the OS levels in a sub-cohort of 1193 non-ICB treated NSCLC patients from the MSK-IMACT cohort with different mutation signatures. HR (compound vs wt): 1.7 (95% [CI]: 1.32–2.20). P values calculated by use of logrank test.

We also examined the relationships between different gene mutation signatures and genes whose mutations previously associated with negative outcome from ICB therapy: EGFR, ALK, KRAS, and LKB1. Our analysis indicated that mutations in these genes were not mutually exclusive with the compound mutation signature (data not shown). In fact, they are quite evenly distributed across the three mutation groups.

We further examined if the mutation signature were preferentially associated with two clinical features: tumor tissue histology or patient smoking history. Our analysis indicated that the compound mutation signature did not preferentially enrich in either lung cancer with adenocarcinoma histology or squamous cell carcinoma histology (Appendix Table A1–A3). However, we found that patients who are previous or current smokers did have a significantly higher chance of possessing the compound mutation signature (Appendix Table A1–A3). However, smoking status alone was not predictive of response to ICB therapy (Appendix Figure A1A and B). Therefore, higher fractions of smokers tend to have the compound mutation signature but smoking status alone did not predict for superior response to ICB.

We next compared with relative predictive powers of the compound mutation signature vs TMB. The FDA recently approved pembrolizumab treatment for pediatric and adult solid tumors that are TMB-H (≥10). We thus evaluated the predictive power of TMB in the MSK-TMB cohort of NSCLC patients. Our data indicated among 350 NSCLC patients, 115 are TMB-H. When compared with those with TMB<10, TMB-H patients had a clear survival benefits (Fig. 4A), with median OS at 18 months (vs 11 months in the TMB-L group). To compare the predictive power of TMB-H vs the compound mutation signature, we stratified patients with high or low TMB by the compound mutation signature and compared their OS levels (Fig. 4B). Our results indicated those patients low TMB (<10) but possessing the compound mutation signature had a significant survival advantage (median OS unreached) when compared with those with no compound mutation signatures with either high (median OS 5.0 months) or low (median OS 9.0 month) TMB. Furthermore, the compound mutation signature captured 145 patients vs 115 patients captured by TMB-H. Therefore, our analysis clearly demonstrated the superiority of the compound mutation signature in predicting ICB benefits compared with TMB (≥10) in NSCLC patients.

Figure. 4. Comparing the predictive powers of TMB-H vs the compound signature in ICB-treated NSCLC patients.

A). OS levels of the MSK-TMB patients with high (≥10) and low (<10) TMB values. HR=0.71 (95% [CI]:0.52–0.94).

B). OS levels in ICB treated MSK-TMB NSCLC patients with TMB-H (≥10) and TMB-L (<10) further stratified according to groups with no compound (NC) or with compoud mutation signature (C).

P-values calculated by use of logrank test.

We next evaluated the relationship between the compound mutation signature and PD-L1 levels in predicting NSCLC response to ICB treatment because the latter is currently the most widely used biomarker to select NSCLC patients for immunotherapy. In order to carry out the analysis, we used a sub-cohort of the MSK-TMB cohort of NSCLC patients for whom the PD-L1 expression data was published³⁷. Patients with all three mutation signatures had a range of PD-L1 expression levels with the compound mutation group possessing the highest average level (but not statistically different when compared with the other two gorups) (Fig. 3A). Importantly, our survival analysis indicated that those patients with the compound mutation signature had a significantly better OS level irrespective of whether they have high or low PD-L1 expression levels (Fig. 3B). Therefore, the compound mutation signature predicts for ICB benefits independent of the PD-L1 expression levels.

What is the biological underpinning of the significant ICB benefits observed in patients with the compound mutation signature ? Based on our hypothesis that gene mutations not only provide neoantigens that might be recognized by the immune system but may also influence the biology of the tumors, we postulated that the compound mutation signature may be associated with altered tumor immune microenvironment. In order to test this hypothesis, we analyzed gene expression data from 501 NSCLC patients from the TCGA pan cancer atlas cohort (Appendix Table A.3) based on their mutation signatures using the CIBERSORT method³⁹ by use of the online tool TIP⁴⁰. The CIBERSORT methods can accurately enumerate different immunoeffector cellular subsets in tumor tissues based on RNA expression data. Our analysis indicated that in the TCGA NSCLC patient cohort, tumors with the compound mutation signature had significantly more intratumoral infiltration of CD8⁺ T cells (Fig. 6A), NK cells (Fig. 6B), and eosinophils (Fig. 6C). On the other hand, the numbers of intratumoral Tregs were significantly reduced (Fig. 6D). These data thus suggest that the compound mutation signature in the 52-gene panel predicts for a pro-inflammatory tumor immune microenvironment in NSCLC that is conducive to ICB treatment.

Fig. 6. Association of the compound mutation signature with a pro-inflammatory tumor microenvironment.

A cohort of NSCLC patients from the TCGA Pan-Cancer Atlas were analyzed for their CIBERSORT immune effector scores based on their mutation signatures.

A). Recruitment scores for CD8+ T cells; 95% confidence intervals [CI]: Wt 0.61 to 1.05, single 0.78 to 1.28, compound 1.04 to 1.37.

B). Recruitment scores for NK cells; 95% CI : Wt 0.72 to 1.21, single 0.98 to 1.56, compound 1.31 to 1.70.

C). Recruitment scores for eosinophils; 95% CI: Wt 0.92 to −0.63, single 0.76 to −0.46, compound −0.57 to −0.36.

D). Recruitment scores for Treg cells; 95% CI: Wt 1.00 to −0.68, single 1.00 to −0.68, compound 1.17 to −0.93.

P-values calculated by use of t-test (unpaired, two-tailed). Wt: 139 patients with no mutations in the 52 gene panel; single: 119 patients with only a single gene mutation in the 52-gene panel; compound: 243 patients a compound mutation signature in the 52-gene panel. Median, lower and upper quartile, minimum and maximum values were shown in the box and whisker plots.

Discussion

Although there is an abundance of evidence to associate high TMB with NSCLC response to ICB treatment, the use of TMB as a biomarker in the clinic has been hampered by the lack of clearly defined cut-offs because of the great variability in TMB among different cancer patient cohorts²⁵. Thus the recent FDA approval of TMB≥10 for pembrolizumab treatment was a significant step in the use of TMB for selecting patients for treatment. On the other hand, our finding that the compound mutation signature in the 52-gene panel predicted NSCLC ICB therapy response more favorably than those with high TMB (≥10) showed that there is significant space for improvement in using TMB to select patients, at least in NSCLC. Our results provided a rationale for analyzing additional cohorts of ICB-treated patients or conducting new prospective clinical trials to validate if the compound mutation signature may serve as a biomarker for selecting patients suitable for ICB therapy.

Our 52-gene panel was selected using criteria that are agnostic of their biological functions. We used the standard that the difference of mutation frequencies in the surviving vs deceased patients have to be significant. We chose p≤0.1 as a compromise standard to include more genes on the list. This was necessitated by the relative small size of the patient cohort. One can imagine with more patients, we can have a more stringent criteria and improve the quality of the gene panel, which may potentially improve the predictive power of our approach. From this perspective, our gene panel should be the starting point of an evolving list that can be modified with the availability of future clinical gene sequencing data.

A closer examination of the gene list and published literature indicate that some of the genes, such as ATM/MRE11/BRCA2/PARP1, are well known genes involved in DNA double strand break repair. These genes have previously been implicated to play significant roles regulating cellular innate immunity,^{41, 42} and mutations in them have been associated with better response to immunotherapy⁴³. MSH2 mutations are known to cause microsatellite instability, which is associated with better treatment outcome for immunotherapy^{19, 44}. In addition, POLE mutations are associated with better immunotherapy outcome in multiple cancer types⁴⁵, presumably because its key roles in in maintaining genetic stability. On the other hand, we do not have any mechanistic understanding of the how other genes in our 52-gene panel may be involved in tumor response to immunotherapy. Nonetheless, they should be good candidates to be evaluated in wet lab studies

Our findings that the compound mutation signature may be associated with a pro-inflammatory tumor immune microenvironment (Fig. 6) provided some hints into biological underpinning of the compound mutation signature. We speculate that it may be that the compound mutation signature ensures both a reasonable high TMB and functionally relevant mutations for ICB treatment.

At a practical level, the compound mutation signature as a predictive biomarker has some potential advantages. At present, the most commonly used biomarker to select for ICB treatment of NSCLC patients is PD-L1 expression. However, it is subject to challenges in tissue sample acquisition, preservation, preparation, and ambiguities in its predictive value. In comparison, compound mutation identification is a straightforward process that can be achieved with different platform technologies, including archival FFPE tissues and liquid biopsies such as patients’ blood, saliva, or urine samples. Therefore, the compound mutation signature is relatively easier to implement clinically as a biomarker. However, its utility in the clinic, especially when compared with TMB or more established markers such as PD-L1, needs to be evaluated in future prospective clinical trials.

In conclusion, our study reveals that a compound mutational signature with 2 or more of the 52-gene panel mutated predicted for response to ICB therapy in retrospectively in several published cohorts. Future clinical trials should be conducted to provide validating evidence before it can be used as a predictive biomarker for ICB patient selection.

Figure. 5. Independence of the compound mutation signature from PD-L1 in predicting ICB treatment response.

A). PD-L1 expression levels in NSCLC patients with different mutation signatures.

B). OS levels of patients stratified according to their PD-L1 expression levels and mutation signature status. PL-NC: PD-L1 level low(<1%), no compound signature; PH-NC, PD-L1 high (≥1%), no compound signature; PL-C, PD-L1 low (<1%), compound signature; PH-C, PD-L1 high (≥1%,) compound mutation signature. P-values calculated by use of unpaired t-test (A) or logrank test (B).