Immune Checkpoint Blockade Outcome in Small-Cell Lung Cancer and Its Relationship With Retinoblastoma Mutation Status and Function

PURPOSE

Immune checkpoint blockade (ICB) in conjunction with chemotherapy is approved for the treatment of extensive-stage small-cell lung cancer (SCLC). Although specific genomic abnormalities such as KEAP1 and STK11 gene mutations are associated with resistance to ICB in non-SCLC, no genomic abnormality has been found in association with resistance to ICB in SCLC.

MATERIALS AND METHODS

We first analyzed a retrospective cohort of 42 patients with SCLC treated with single-agent ICB or ICB combination (data set A). We then validated our results in a large prospective clinical trial of 460 patients (CheckMate 032, data set B). DNA and RNA sequencing were performed.

RESULTS

In data set A, patients treated with ICB with RB1 wild-type (WT) had a median overall survival (OS) of 23.1 months (95% CI, 9 to 37.5), whereas the RB1 mutant OS was 5 months (95% CI, 2.5 to 26; P = .04). Differentially expressed gene analysis between RB1 mutant and RB1 WT samples indicated the enrichment of downregulated immune-related genes and an immune exclusion phenotype among RB1 mutant but not in the RB1 WT tumor samples. We then assessed results from 460 patients enrolled in CheckMate 032, a trial of nivolumab (NIVO) or NIVO + ipilimumab only in SCLC. In this large cohort, RB1 WT patients had significantly improved outcome with NIVO therapy compared with mutant patients (hazard ratio, 1.41; 95% CI, 1.02 to 2.01; P = .041). High RB1 loss-of-function (LOF) signature scores significantly associated with neuroendocrine subtypes (ASCL1 and NeuroD1). However, neuroendocrine subtypes did not associate with OS. Remarkably, patients with lower RB1 LOF scores had longer OS following treatment with NIVO.

CONCLUSION

SCLC patients with RB1 WT status or lower RB1 LOF signature scores by transcriptomics have better outcomes with ICB monotherapy.

INTRODUCTION

Small-cell lung cancer (SCLC) represents approximately 10%-15% of all lung cancers and is characterized by a high growth fraction, propensity for early metastases, and poor survival.^1,2 Although response rates to chemotherapy are high in SCLC (> 70%), so too are relapse rates. Recently, however, immune checkpoint blockade (ICB) has been added as a new therapeutic modality to fight SCLC. ICB is now standard of care for most patients with non–small-cell lung cancer (NSCLC),³ and predictive markers of a positive response to ICB in NSCLC include > 50% programmed death-ligand-1 (PD-L1) positivity of the tumor by immunohistochemistry and a high tumor mutation burden (TMB) by genomic sequencing.^4,5 Unfortunately, ICB in SCLC provides only limited clinical benefit as monotherapy,⁶ maintenance therapy,⁷ or third-line therapy.⁸ However, chemotherapy combined with ICB in first-line extensive-stage SCLC did show a survival benefit of 2 months and is now standard of care.^9,10 Although the overall role of ICB in SCLC remains limited, some patients demonstrate dramatic and durable tumor responses. Unlike NSCLC, there are no identified genomic alterations that predict for response or lack of response to ICB in SCLC.¹¹ SCLC does, however, demonstrate elevated TMB like NSCLC, and TMB has been shown to be a potential independent predictive marker of ICB in relapsed SCLC.^12,13 It is thought that SCLC is resistant to ICB because of an immunologically inert tumor microenvironment. This idea is supported by a study showing decreased T-cell infiltration of SCLC tumors compared with NSCLC tumors.¹⁴ However, before ICB use in SCLC, several immunology studies showed that improved outcomes were associated with increased T-cell infiltration of SCLC tumors under a variety of settings.¹⁵ Thus, it is also possible that the current checkpoint targets are not important in SCLC, leading to the search for alternatives, such as those involved in natural killer cell regulation.^16,17

CONTEXT

• Key Objective

• We sought to determine whether RB1 mutational status or RB loss-of-function signature could predict outcomes to single-agent checkpoint inhibitors in small-cell lung cancer (SCLC).

• Knowledge Generated

• SCLC patients with RB1 mutations or high loss-of-function signatures had lower survival rates and response to checkpoint inhibitors.

• Relevance

• Although next-generation sequencing has not been used to predict outcomes in SCLC treated with immunotherapy, our study suggests that determining RB1 status may be helpful.

SCLC has been shown to be composed of two different phenotypes, neuroendocrine (NE) and non-NE. Within the NE and non-NE categories, four different subgroups have been described on the basis of the expression of dominant transcription factor genes, including ASCL1 (SCLC-A), NEUROD1 (SCLC-N) and POU2F3 (SCLC-P), and YAP1 (SCLC-Y).¹⁸ We have previously shown that SCLC-Y has a predominantly non-NE profile and may be predominantly RB1 wild-type (WT),¹⁹ whereas the other three subtypes are commonly RB1-deficient.

RB1 is the second most frequent tumor suppressor gene mutation in SCLC after TP53 and is present in 70% of patients.²⁰ Transgenic mouse models of SCLC require deletion of both TP53 and RB1 genes.²¹ Although the retinoblastoma protein encoded by RB1 plays a significant role in cell cycle regulation, significant literature also exists demonstrating the involvement of RB1 in immune function; expression of immune-related genes is downregulated in preclinical models of RB1 loss, as well as decreased interferon beta, interleukin 8, and chemokines.^22-25 We postulated that an underlying oncogenic mediator of immunosuppression may be related to RB1 mutation or functional loss. Herein, we explore the relationship between RB1 mutation or functional loss and ICB benefit in SCLC. To this end, we evaluated RB1 status in a retrospective test cohort of 42 patients followed by validation in a large prospective trial of 460 patients.

MATERIALS AND METHODS

Study Population

The study was approved by the University Hospitals Seidman Cancer Center Institutional Review Board (CC006). This retrospective study was performed in accordance with the standards set by the Declaration of Helsinki. Our initial patient population (data set A) consisted of patients (sequential) seen mainly at two institutions (University Hospitals Seidman Cancer Center and Columbia University Medical Center) selected with three criteria: (1) a diagnosis of SCLC, (2) received ICB without concurrent chemotherapy, and (3) had DNA sequencing available. The cutoff date for this cohort was January 2019 as we were interested only in patients who received ICB (without concurrent chemotherapy) and had at least 6 months of follow-up. Subsequent to this cutoff date, ICB was approved for use in conjunction with chemotherapy and the use of single-agent ICB substantially declined. Additional patients were recruited from two other institutions for a total of 42 subjects (University Hospitals Seidman Cancer Center = 25, Columbia University Medical Center = 9, Memorial Sloan Kettering Cancer Center = 6, and University of Chicago = 2). Data were collected under institutional review board–approved protocols. All patients had been treated before ICB approval as first-line treatment for SCLC. Patients received either single-agent anti–programmed cell death (PD)(L)1 (n = 21) or in combination with anticytotoxic T-cell lymphocyte-4 blockade (n = 21). Molecular testing included targeted tumor next-generation sequencing (NGS, evaluated by Foundation One; n = 21); targeted plasma NGS (Guardant360; n = 4); or whole-exome/transcriptome sequencing (Personal Genome Diagnostics; n = 17). TMB was either calculated by Foundation Medicine (FM; n = 21)¹³ or determined by whole-exome sequencing (WES). To both validate and expand upon our initial DNA profiling results, we then used a significantly larger data set (data set B) from a prospective clinical trial in pretreated patients with SCLC (CheckMate 032, ClinicalTrials.gov identifier: NCT01928394). In this trial, 460 patients with SCLC were in the intent-to-treat (ITT) population with WES and RNA-sequencing (RNA-Seq) available for 279 and 286 samples, respectively.

Clinical Outcomes, Biomarker Evaluation, and Statistical Methodology

Details are provided in Appendix 1.

RESULTS

Clinical characteristics of data set A are shown in Appendix Table A1. A total of 42 patients were in this data set. Checkpoint inhibitors (without chemotherapy combination) were only used for a short period between late 2018 and March 2019 at which time the US Food and Drug Administration approved the combination of chemotherapy with checkpoint inhibition as front-line treatment for SCLC and patients were no longer being treated with single-agent checkpoint inhibitors. The majority of patients were treated with a combination of ipilimumab (IPI, low dose) every 3 weeks and nivolumab (NIVO) every 3 weeks (19/42), and all but one patient received low-dose IPI with standard-dose NIVO every 2 weeks. Additional treatment regimens received included NIVO (n = 16), pembrolizumab (n = 2), durvalumab (n = 3), and durvalumab in combination with tremelimumab (n = 2). The majority of patients received treatment in the second-line setting (27/42) and 18/42 patients had liver metastases. NGS was performed from tumor samples by FM testing in 21 patients; by matched tumor and normal WES in 17 patients; and Guardant Health plasma-based testing in four patients, where insufficient tumor was available.

Consistent with previous reports, RB1 (55%) and TP53 (79%) were the most frequently mutated genes. We observed a range of RB1 mutations including splice site (n = 8), frameshift indels (n = 7), exon loss (n = 2), and point mutations (n = 6). The overall response rate for the entire cohort was 40.5% (17/42; 95% CI, 27 to 56). We focused on single gene RB1 in the current study on the basis of the previous findings of genes relevant in extensive-stage SCLC outcome that we have published.²⁶ The overall response rate for the RB1 WT group was 68.4% versus 17.4% for the RB1 mutant group. We also looked at other genes with mutation rate > 10% in this patient cohort. Tp53, RB1, LRp1B, NKX2-1, FGFR1, and NOTCH1 were the genes with altered rate > 10%, and none of them except RB1 was associated with clinical response (P > .15). With the six tests conducted, the FDR adjusted P value for testing the predictive value of RB1 on clinical response was .048.

The median progression-free survival (PFS) for the entire cohort was 10 months (95% CI, 2.2 to 21.6). For the RB1 WT group, the median PFS was 18 months (95% CI, 7 to not reached) and for the RB1 mutant group was 3 months (95% CI, 1.5 to 24.1). This difference was remarkable (P = .05), showing the association of WT RB1 with better ICB outcome (Fig 1A). The median overall survival (OS) for the entire 42 patient cohort was 17 months (95% CI, 4.99 to 24). The RB1 WT group had a median OS of 23.1 months (95% CI, 9 to 37.5), whereas the RB1 mutant group was 5 months (95% CI, 2.5 to 26). This difference is statistically significant (P = .04; Fig 1B). Of the 18 patients with liver metastases, 13/18 (72%) had an RB1 mutation versus only 10/24 (42%) of patients without liver metastases (P = .08). TMB was determined as previously described from WES or FM testing. As two different platforms were used in this analysis (WES or FM targeted panel sequencing), we determined the median for each group separately. The median number of nonsynonymous mutations by WES was 227 mutations in responders versus 277 in nonresponders, and by FM testing was 8.7 mut/mb in responders versus 13.9 mut/mb in nonresponders. By WES, the median number of nonsynonymous mutations was 218 mutations in RB1 mutant versus 272 in RB1 WT, and by FM testing was 13.9 mut/mb in RB1 mutant versus 8.7 mut/mb in RB1 WT.

FIG 1.

(A) Kaplan-Meier estimation of PFS by RB1 status (data set A). The median PFS (95% CI) was 18.0 (7.0 to not available) months for RB1 WT versus 3.0 (1.5 to 24.1) months for RB1 mutant (P = .051). (B) Kaplan-Meier estimation of OS by RB1 status (data set A). The median OS (95% CI) was 23.1 (9.0 to 37.5) months for RB1 WT versus 5.0 (2.5 to 21.6) months for RB1 mutant (P = .037). Mut, mutant; OS, overall survival; PFS, progression-free survival; WT, wild-type.

Using RNA-seq data, we performed differentially expressed gene analysis between RB1 mutant (n = 4) and RB1 WT (n = 6) samples (Fig 2A). The results indicated a striking downregulation of immune-related genes among RB1 mutant but not in the RB1 WT tumor samples (Fig 2B). As confirmation, we also conducted IPA analysis on RB1 mutant differentially expressed genes, which revealed an immune exclusion phenotype among them (Appendix Fig A1). This observation was further validated by gene set enrichment analysis (GSEA) run against Molecular Signatures Database (MSigDB) Hallmark gene sets, as well as RNA-seq immune deconvolution through single-sample GSEA analysis against published immune signatures (Appendix Fig A2). Furthermore, an independent cohort containing RB1 WT (n = 3) and RB1 mutant (n = 5) samples again showed downregulation of immune-related genes and lower immune cell infiltration in the RB1 mutant group (Appendix Fig A3). We calculated the scores for the cytotoxic T-cell and T-cell effector function signatures, as well as for PD-L1 and PD-1, which are all known positive predictive markers for ICB therapies. Comparing the scores of RB1 mutant and WT SCLC samples revealed that RB1 mutant tumors had lower scores than RB1 WT tumors (Fig 2C).

FIG 2.

(A) Volcano plot showing differentially expressed genes (data set A). Colored genes indicate statistical significance. Red indicates genes upregulated in RB1 mutant tumors and blue indicates genes downregulated in RB1 mutated tumors. (B) A focused heatmap showing differences in immune-related gene expression between RB1 mutant and WT tumors (data set A). Immune-related genes are downregulated in RB1 mutant tumors. (C) Cytotoxic T cell, T-cell effector function, PD-L1, and PD-1 signature scores between RB1 mutant and WT tumors (data set A). Signatures are as previously defined and are described in the Methods section. mut, mutant; PD, progressive disease; PD-L1, programmed death ligand-1; PR, partial response; SD, stable disease; TMB, tumor mutation burden; WT, wild-type.

Given the limitations and risk of bias in retrospective analyses (as seen by extraordinary survival and response rate of 40% in our data set A cohort), we sought confirmation of these initial findings by interrogating the CheckMate 032 trial (data set B, clinical characteristics in Appendix Table A2). Clinical parameters as well TMB and PD-L1 expression were previously shown not to predict benefit from ICB in Checkmate 032 trial.^6,27 Of the 460 pretreated patients with SCLC in the ITT population in this study, WES and RNA-seq data were available for 279 and 286 samples, respectively. Of the 245 patients treated with NIVO, WES and RNA-seq data were available for 155 and 156 patients, respectively. Of the 215 patients treated with NIVO + IPI, WES and RNA-seq data were available for 124 and 130 patients, respectively. Patient characteristics were comparable across the biomarker-evaluable subgroups and the ITT population (Appendix Table A2). No significant differences in RB1 mutation frequency were observed between treatment arms.

Of the 279 patients with WES data available, 178 (63.8%) patients had RB1 mutations (including missense). We are excluding missense mutations called from WES as assessing if a missense variant is deleterious is challenging. Furthermore, we find that the number of missense mutations correlates with TMB and including missense mutations may inflate the true number of RB1 mutant samples. We also evaluated for loss of heterozygosity (LOH). Indeed, most RB1 mutants also had LOH. Since the number of RB1 mutants and RB1 LOH were similar, we chose to use RB1 mutants for our analysis. If we exclude the missense mutations, 166 (59.4%) patients had mutations in RB1. OS was significantly associated with RB1 mutation in the NIVO monotherapy arm (Fig 3), with a hazard ratio of 1.46 (95% CI, 1.02 to 2.01; P = .041). However, RB1 mutation status was not associated with OS in the NIVO + IPI arm, with a hazard ratio of 1.03 (95% CI, 0.68 to 1.69; P = .9). As RB1 mutational status does not equate necessarily to RB1 functional status (RB1 function may be lost by other mechanisms), we established an RB1 loss-of-function (LOF) signature on the basis of a previously published RB1 LOF signature that uses the expression of 87 genes related to E2F targets.²⁸ First, we sought to determine the association between the RB1 LOF signature with the described transcriptional subtypes of SCLC. Transcriptomically evaluable SCLC samples were classified as NE or non-NE on the basis of the relative expression levels of ASCL1, INSM1, NEUROD1, YAP1, MYC, and POU2F3, as previously described²⁹ (Fig 4A). High RB1 LOF signature scores significantly associated with the NE subtype (Fig 4B, P < .001). However, despite this association, transcriptional subtypes did not associate with OS following NIVO ± IPI (Fig 4C). As an inflamed subtype has recently been reported to potentially correlate with the benefit of ICB in combination with chemotherapy,³⁰ we looked at the relation between the four genes inflammatory signature with RB1 status or SCLC transcriptional subtype. No association was seen (Appendix Fig A4). As predicted, RB1 LOF signature scores were significantly higher (indicating loss of RB1 function) in RB1 mutant tumors compared with RB1 WT tumors (P < .001; Fig 5A). Patients with RB1 LOF scores in the middle and lower tertiles had longer OS following treatment with NIVO ± IPI compared with patients with RB1 LOF signature scores in the upper tertile (Fig 5B). This, however, was not associated with greater overall response in RB1 WT compared with RB1 mutant patients (P = .78). Given the report of possible association of TMB and response to checkpoint inhibitors in SCLC,⁵ we assessed the relation between RB1 status and TMB, and found no association (P = .22, Appendix Fig A5).

FIG 3.

Kaplan-Meier estimation of (A) PFS and (B) OS by RB1 status in the CheckMate 032 cohort (data set B). OS was significantly associated with RB1 mutation in the nivolumab monotherapy arm, with a HR ratio of 1.46 (95% CI, 1.02 to 2.01; P = .041). HR, hazard ratio; mut, mutant; OS, overall survival; PFS, progression-free survival; WT, wild-type.

FIG 4.

(A) Heatmap showing the relative expression levels of ASCL1, INSM1, NEUROD1, YAP1, MYC, and POU2F3 in small-cell lung cancer samples that were used to classify NE or non-NE status (data set B). (B) Difference in RB1 LOF signature scores in NE and non-NE subtypes (P < .001). (C) Kaplan-Meier estimation of OS by NE status following NIVO ± IPI treatments in the CheckMate 032 cohort (data set B). CR, complete response; HR, hazard ratio; IPI, ipilimumab; LOF, loss of function; MUT, mutant; NE, neuroendocrine; NIVO, nivolumab; ORR, overall response rate; OS, overall survival; PD, progressive disease; PR, partial response; RNE, radiologically not evaluable; SD, stable disease; WT, wild-type.

FIG 5.

(A) RB1 LOF signature scores in RB1 mutant tumors compared with RB1 WT tumors in data set B (P < .001). (B) Tertile analysis was used to correlate the OS with the RB1 LOF (data set B). The middle and lower tertiles had longer OS following treatment with NIVO ± IPI compared with patients with RB1 LOF signature scores in the upper tertile. HR, hazard ratio; IPI, ipilimumab; LOF, loss of function; Mut, mutant; NIVO, nivolumab; OS, overall survival; WT, wild-type.

DISCUSSION

Tumor-intrinsic oncogenic pathways have been shown to mediate T-cell exclusion from the tumor microenvironment and immunotherapy resistance.³¹ WNT/β-catenin activation and PTEN loss are both prime examples of oncogenic pathways that correlate with decreased intratumoral T-cell infiltration and resistance to ICB. In NSCLC, STK11 inactivation is common in lung adenocarcinomas and is associated with increased suppressive myeloid cells and aberrant cytokine production that can lead to decreased T-cell number and function, ultimately impeding host immune surveillance.³² RB1 has recently been shown to play a role in mediating immune response. For example, in hepatocellular carcinoma, RB1 depletion in cancer cells results in a diminished immunologic response to a variety of stimuli.³³ In bladder cancer, decreased RB1 expression was predictive of a decreased response to bacillus Calmette-Guerin therapy.³⁴ In NSCLC, RB1 alterations may be associated with lack of response to ICB.^35,36

In our retrospective analysis of patients with extensive-stage SCLC treated with ICB (data set A), we observed a significantly higher response rate, PFS, and OS in patients with RB1 WT versus RB1 mutant tumors. Our transcriptome analyses showed an impaired tumor immune surveillance phenotype in RB1 mutant SCLC. GSEA analysis results showed that the transcriptomic profile of RB1 mutant tumors is in agreement with that of the RB1 gene knockout mouse model, where downregulated RB1 gene expression is associated with compromised immune function and response to pathogens.²² The benefit of ICB in RB1 WT tumors was independent of the TMB.

Our study represents the first data to our knowledge of RB1 mutation associating with ICB resistance. This is even more striking as RB1 WT SCLC are more frequently chemorefractory and portends a worse prognosis in the absence of immunotherapy.²⁶ The response rate for the entire first data set (n = 42) was 40.5% and is higher than one would expect from published data; nevertheless, the balance of ICB responders versus nonresponders allows identification of potential predictive factors. An additional area of concern was the high prevalence of RB1 WT tumors (45%), as a previous publication suggests that RB1 gene alterations may be closer to 100%.³⁷ This genomic study, however, was enriched for early-stage resected SCLC (not a scenario typically seen in practice). The prevalence of RB1 WT tumors in our data set is consistent with other data sets of extensive-stage SCLC.^38,39 Although arguably RB1 WT by targeted NGS does not imply that the RB1 pathway is functional, it is now clear that a small fraction of SCLC tumors indeed lack NE markers and that some of these are categorized by high YAP1 expression and activation of the Hippo pathway.^19,40 Interestingly, the high YAP1 subtype of SCLC enriches for intact retinoblastoma function, which can be demonstrated with immunohistochemistry (protein), DNA sequencing, or RNA sequencing. Furthermore, RB1 WT SCLC has been associated with a chemotherapy-refractory status.¹⁹ To validate our data, we evaluated CheckMate 032, a prospective trial of NIVO ± IPI and saw a similar effect with better survival in RB1 WT patients compared with RB1 mutant. As not all RB1 WT SCLC patients may have a functional RB pathway,³⁷ we used a published signature of RB1 functional status using RNA-seq that confirmed these findings. As can be seen in Figure 5, the RB1 LOF signature correlated with RB1 mutational status, although it is also clear that some RB1 WT samples still have RB1 functional loss, indicating alternative mechanisms for RB1 dysfunction. When using RB1 LOF signature, a similar benefit can be seen in the IPI/NIVO patients as well. It must be noted that a PFS benefit on the basis of RB1 WT status was not seen in the Checkmate 032 trial. It is well known that checkpoint inhibitors may not improve PFS (or have minimal effect) in PD-L1 low expressing tumors but have a much greater effect on OS.⁴¹ We also observed in the survival curve that the effect of RB1 status in CheckMate 032 does not become apparent till 6 months at which time the curves diverge. Late separation of survival curves is common in immunotherapy trials.⁴² This may indicate a delayed action of therapy but may indicate that patients with lower burden or slower progression may be benefiting.

Given recent emphasis on different SCLC subtypes on the basis of transcriptional regulators,¹⁸ we looked at the correlation of RB1 status with these different subtypes and found that RB1 WT mainly have a non-NE phenotype (lacking ASCL1 and NeuroD1 expression), in agreement with Sonkin et al.⁴⁰ However, the use of NE subtypes does not predict response to checkpoint inhibitors in our analysis, suggesting that RB1 function is more relevant in this regard.

APPENDIX 1. Supplemental Information on Biomarker Testing and Statistical Evaluation

Clinical Outcome Evaluation

Response was defined using RECIST criteria by a thoracic oncologist at the respective institutions using RECIST v1.1. Progression-free survival was defined as the time from the start of immunotherapy to the date of disease progression or death, whichever occurred first. Patients who were alive without disease progression were censored on the date of their last disease assessment. Overall survival (OS) was defined as the time from the start of immunotherapy to death. Patients who were still alive were censored at the date of the last contact.

Biomarker Evaluation

Details concerning biomarker evaluation are provided in Appendix 1. RNA sequencing read processing was done as previously described.^43-46 Differentially expressed gene (DEG) analysis was performed with the DESeq2 R package. DEG analysis results were used in gene set enrichment analysis (GSEA).^47,48 Infiltration levels for different immune cell types were quantified using the single-sample GSEA implementation by the R package “gsva”.⁴⁹ Single-sample GSEA scores for each immune cell type were used to calculate total T-cell infiltration score and immune infiltration score, as previously described by Şenbabaoğlu et al.⁵⁰ We derived a cytotoxic T-cell signature on the basis of the geometric mean of two key cytolytic effectors, granzyme A (GZMA) and perforin (PRF1), according to the work of Rooney et al.⁵¹ Previously published signatures from McDermott et al⁵² were used to assess differences in T-cell effector function. DEG analysis results were used in performing Ingenuity Pathway Analysis (IPA analyses, QIAGEN Inc).⁵³ Filters used in picking up DEG genes for IPA include mean expression > 50, fold change > 20%, and P < .05.

With CheckMate 032, biomarker evaluation was performed in samples from patients in the nonrandomized and randomized cohorts. RB1 wild-type or mutant status and tumor mutation burden were determined by whole-exome sequencing and measuring the total number of missense mutations. Gene expression signatures were evaluated by RNA sequencing and included (1) neuroendocrine transcriptional subtype from three genes (ASCL1, INSM1, and NEUROD1) and (2) the non-neuroendocrine transcriptional subtype from three genes (YAP1, MYC, and POU2F3). For RB1 protein loss of function, 87 genes were used²⁸; and for an inflammatory signature of the antitumor immune response, four genes (CD274, CD8A, LAG3, and STAT1) were used.⁵⁴

Statistical Analysis

The association between two categorical factors was tested using the Fisher's exact test or chi-square test, and the difference of continuous measurements between groups was examined using the Wilcoxon rank-sum test. Survivor distribution was estimated using Kaplan-Meier methods and differences of OS, progression-free survival between/among groups was examined by log-rank. The association between RB1 gene mutation and clinical response was evaluated using the chi-square test. All tests were two-sided; P value for testing the association between RB1 and clinical response was adjusted on the basis of Benjamini and Hochberg's procedure,⁵⁵ and a P value ≤ .05 was considered statistically significant. In data set A (test cohort), because the response rate of the sample population selected was 40.5%, which was about twice the response rate of the overall small-cell lung cancer population (20%), we weighted our samples (with a weight of 0.5 for responders and 1.34 for nonresponders) to correct the selection bias, so the response rate of the weighted sample is about the same as the general population (20%). Statistical analysis was performed using SAS (SAS 9.4, SAS Institute Inc, Cary, NC). Associations of biomarkers with overall response rate and OS were tested using the Cox proportional hazards model and were reported with odds ratios and hazard ratios with 95% CIs, respectively.

FIG A1.

The Ingenuity Pathway Analysis on RB1 mutant differentially expressed genes showing an immune exclusion phenotype. NFAT, nuclear factor of activated T-cells; PI3K, phosphatidylinositol 3-kinase; PKC, protein kinase C.

FIG A2.

(A) The GSEA run against MSigDB Hallmark gene sets showing the enriched pathways (left) in RB1 mut versus wild-type (*P < .05.). (B) The immune deconvolution through single-sample GSEA analysis showing immune signatures (*P < .05.). aDC, activated dendritic cells; CTLA, cytotoxic T-cell lymphocyte; CYT, cytotoxic T cell; DC, dendritic cells; GSEA, gene set enrichment analysis; IDC, immature dendritic cells; IIS, immune infiltration score; mut, mutant; NES, enrichment score normalized; NK, natural killer; pDC, plasmacytoid dendritic cells; PD-1, programmed cell death-1; PD-L1, programmed death ligand-1; TIS, T-cell infiltration score; WT, wild-type.

FIG A3.

(A) Heatmap showing downregulation of immune-related gene expression in RB1 mutant tumors (n = 5) compared with RB1 WT (n = 3) tumors (independent cohort). (B) Comparison of relative abundance of immune cells in RB1 WT (n = 3) and RB1 mutant (n = 5) tumors on the basis of network-based deconvolution (ImSig R package) analysis (mean with standard deviation, *P < .05 by two-tailed Mann-Whitney U test). Mut, mutant; NK, natural killer; WT, wild-type.

FIG A4.

Correlation of inflammatory signature with (A) RB1 status (RB1 Mut v WT) and (B) SCLC subtypes (NE v non-NE). Mut, mutant; NE, neuroendocrine; SCLC, small-cell lung cancer; WT, wild-type.

FIG A5.

Correlation of TMB with RB1 status (RB1 Mut v WT). No significant difference was observed (P = .22). Mut, mutant; TMB, tumor mutation burden; WES, whole-exome sequencing; WT, wild-type.

TABLE A1.

Patient Characteristics of Data Set A

TABLE A2.

Patient Characteristics of Data Set B (CheckMate 032)

DATA SHARING STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request; however, part of the data used in this study are from Bristol Myers Squibb that are available from the authors upon reasonable request and with permission of Bristol Myers Squibb.

AUTHOR CONTRIBUTIONS

Conception and design: Afshin Dowlati, Timothy Chan, Parul Doshi, Han Chang

Financial support: Afshin Dowlati

Administrative support: Afshin Dowlati, Timothy Chan, Parul Doshi

Provision of study materials or patients: Afshin Dowlati, Timothy Chan, Brian Henick, Parul Doshi, Jyoti Patel

Collection and assembly of data: Afshin Dowlati, Timothy Chan, Brian Henick, Parul Doshi, Han Chang, David Balli

Data analysis and interpretation: All authors

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).