Molecular profiling of long-term responders to immune checkpoint inhibitors in advanced non-small cell lung cancer Joan Frigola1, Alejandro Navarro2, Caterina Carbonell1, Ana Callejo2, Patricia Iranzo2, Susana s2, Alex Martinez-Marti2, Nuria Pardo2, Nadia Saoudi-Gonzalez2, Debora Martinez3,4, Jose Cedre Jimenez5, Irene Sansano6, Francesco M. Mancuso7, Paolo Nuciforo5, Luis M. Montuenga8,9,10,11, Montse Sanchez-Cespedes12, Aleix Prat3,4, Ana Vivancos7, Enriqueta Felip1,2 and Ramon Amat1 1 Thoracic Cancers Translational Genomics Unit, Hebron Institute of Oncology (VHIO), Vall d, Barcelona, Spain 2 Oncology Department, Vall d’Hebron University Hospital & Vall d’Hebron Institute of Oncology (VHIO), Barcelona, Spain 3 Department of Medical Oncology, Hospital Clinic, Barcelona, Spain 4 Translational Genomics and Targeted Therapies in Solid Tumors, IDIBAPS, Barcelona, Spain 5 Molecular Oncology Group, Vall d’Hebron Institute of Oncology (VHIO), Barcelona, Spain 6 Pathology Unit, Vall d’Hebron University Hospital, Barcelona, Spain 7 Cancer Genomics Laboratory, Vall d’Hebron Institute of Oncology (VHIO), Barcelona, Spain 8 Program in Solid Tumors, Center for Applied Medical Research (CIMA), Pamplona, Spain 9 Department of Pathology, Anatomy and Physiology, School of Medicine, University of Navarra, Pamplona, Spain 10 Centro de Investigacion Biomedica en Red de Cancer (CIBERONC), Madrid, Spain 11 Navarra Health Research Institute (IDISNA), Pamplona, Spain 12 Cancer Genetics Group, Josep Carreras Leukaemia Research Institute (IJC), Campus ICO-Germans Trias i Pujol, Badalona, Barcelona, Spain Keywords Immunotherapy has transformed advanced non-small cell lung cancer chromosomal alterations burden; copy (NSCLC) treatment strategies and has led to unprecedented long-lasting number alterations; immune checkpoint responses in some patients. However, the molecular determinants driving inhibitors; long-term benefit; NSCLC; PD-L1; these long-term responses remain elusive. To address this issue, we per- tumor mutational burden formed an integrative analysis of genomic and transcriptomic features of Correspondence long-term immune checkpoint inhibitors (ICIs)-associated responders. We Enriqueta Felip, Vall d’Hebron University assembled a cohort of 47 patients with NSCLC receiving ICIs that was Hospital, P. Vall d’Hebron 119-129, 08035, enriched in long-term responders [>18 months of progression-free survival Barcelona, Spain, and Dr. Ramon Amat, Vall (PFS)]. We performed whole-exome sequencing from tumor samples, esti- d’Hebron Institute of Oncology (VHIO), C/ mated the tumor mutational burden (TMB), and inferred the somatic copy Nazaret 115-117, 08035, Barcelona, Spain number alterations (SCNAs). We also obtained gene transcription data for Emails: efelip@vhio.net and ramat@vhio.net a subset of patients using Nanostring, which we used to assess the tumor Joan Frigola and Alejandro Navarro immune infiltration status and PD-L1 expression. Our results indicate that contributed equally to this study there is an association between TMB and benefit to ICIs, which is driven by those patients with long-term response. Additionally, high SCNAs bur- (Received 28 October 2020, revised 8 den is associated with poor response and negatively correlates with the December 2020, accepted 14 December presence of several immune cell types (B cells, natural killers, regulatory T 2020) cells or effector CD8 T cells). Also, CD274 (PD-L1) expression is increased doi:10.1002/1878-0261.12891 in patients with benefit, mainly in those with long-term response. In our cohort, combined assessment of TMB and SCNAs burden enabled identifi- cation of long-term responders (considering PFS and overall survival). Notably, the association between TMB, SCNAs burden, and PD-L1 Abbreviations FFPE, formalin-fixed paraffin-embedded; FGA, fraction genome altered; HLA, human leukocyte antigen; ICIs, immune checkpoint inhibitors; NSCLC, non-small cell lung cancer; OS, overall survival; PD1, programmed death 1; PD-L1, programmed death-ligand 1; PFS, progression- free survival; SCNAs, somatic copy number alterations; TMB, tumor mutational burden; WES, whole-exome sequencing. Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 1 This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. Molecular determinants of ICI-associated long-term response J. Frigola et al. expression with the outcomes of ICIs treatment was validated in two public datasets of ICI-treated patients with NSCLC. Thus, our data indicate that TMB is associated with long-term benefit following ICIs treatment in NSCLC and that TMB, SCNAs burden, and PD-L1 are complementary determinants of response to ICIs. melanoma, high levels of aneuploidy (SCNAs encom- 1. INTRODUCTION passing whole chromosome arms or entire chromo- Lung cancer is the world’s leading cause of cancer-re- somes) or an overall increase of SCNAs burden were lated deaths due to its high incidence and low survival. associated with poorer response to ICIs [17,19], which Fortunately, immune checkpoint inhibitors (ICIs), could be explained by the fact that highly aneuploid antibodies that block programmed death 1 (PD1) tumors exhibit lower levels of immune-related tran- receptor or its ligand programmed death-ligand 1 (PD- scriptional signatures [19]. Conversely, TMB does not L1), have shown great efficacy leading to responses of seem to correlate with tumor immune infiltration, nor unprecedented duration in some patients [1–4]. Never- with SCNAs burden or PD-L1 [8,11,20]. Therefore, theless, the majority of patients fail to respond to this models combining TMB and PD-L1 expression or type of immunotherapy, and many that do eventually immune-related signatures have been shown to better develop resistance [5]. Therefore, the identification and predict response to ICIs [11,20]. validation of biomarkers of ICIs response, and specifi- Nevertheless, most efforts have been directed toward cally of sustained benefit, are highly relevant to the identifying biomarkers of ICIs response, that is dur- management of NSCLC patients. able clinical benefit (PFS > 6 months), while studies Assessment of PD-L1 expression by immunohisto- assessing their value in predicting long-term benefit are chemistry is the only validated test for ICIs first-line scarce, owing to the lack of long-lasting clinical fol- treatment decisions in NSCLC [6,7]. Additionally, PD- low-up and the low representation of long-term L1 expression has been recently associated with long- responders in unselected cohorts. Moreover, studies term response to pembrolizumab, an anti-PD-1 drug often focus on just one or two biomarkers. Thus, there [8]. However, PD-L1 determination presents limita- is a need for integrative studies analyzing multiple tions such as analysis variability or intra-tumor hetero- molecular biomarkers in the same set of individuals geneity. Moreover, favorable responses to ICIs in presenting prolonged response to understand ICI-asso- patients with no apparent expression of PD-L1 have ciated long-term benefit. Here, we gathered a cohort of also been observed. Hence, there is a need to find ICI-treated patients with advanced NSCLC, some of additional biomarkers or combinations of them to bet- whom exhibited outstanding long-term responses to ter predict response. In this direction, the quantitative these therapies, and assessed tumor-intrinsic genomic analysis of the presence of different immune cell types biomarkers and extrinsic biomarkers (immune infiltra- infiltrated within the tumor, which can be inferred tion) to study their interplay and utility to discern ICIs from gene transcriptional data, has been shown to be response and long-term benefit. Finally, we validated indicative of benefit [9–13]. our findings in two publicly available independent Tumor mutational burden (TMB) has also been cohorts [14,20]. associated with ICIs response in several tumor types such as melanoma and lung cancer [10,14,15]. Concep- 2. Materials and methods tually, a higher number of somatic mutations increase the amount of potentially immunogenic neoantigens that could be recognized by the adaptive immune sys- 2.1. Patients data tem. Nevertheless, the association between TMB and ICIs response has not been observed in other tumor All patients included in our cohort were diagnosed types such as renal cell carcinoma [16]. Moreover, the with advanced NSCLC and treated at the Vall suitability of TMB as biomarker of ICIs response, d’Hebron Hospital. Written informed consent was especially in melanoma, has been questioned [17,18]. A obtained from all patients before enrollment and the less explored feature in relation to ICIs is the tumor’s Hospital Institutional Review Board approved this somatic copy number alterations (SCNAs) burden. In study (PR(AG)139/2014). The study methodologies 2 Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. J. Frigola et al. Molecular determinants of ICI-associated long-term response conformed to the standards set by the Declaration of 2.6. Gene expression assessment Helsinki. Detailed clinical information regarding the cohort and each patient can be found in Tables S1 Gene expression by Nanostring was carried out as pre- and S2. Archive tumor samples were obtained prior to viously described in Prat et al [13,25]. In fact, 8 out of ICIs treatment. Patients were retrospectively collected. 22 samples were previously published in Prat [13]. Nanostring results were then normalized following the Nanostring Gene Expression Data Analysis Guide- 2.2. Whole-exome sequencing lines. Whole-exome sequencing was performed on DNA General transcriptional signatures associated with T- extracted from formalin-fixed paraffin-embedded cell activity in the tumor were obtained from different (FFPE) tumor samples (Maxwell 16 FFPEPlus LEV publications [9,12,19,26]. Gene expression signatures DNA Purification Kit, Promega, Madison, WI, USA) characteristic of different immune cell populations in addition to freshly obtained peripheral blood or were obtained from Davoli [19]. Those signatures with normal tissue. WES Libraries were prepared according less than 50% of the genes represented in the Nanos- to manufacturer’s protocol (SureSelect XT Human All tring PanCancer Immune Profiling Panel were dis- Exon v5, Agilent, Santa Clara, CA, USA). Finally, carded. A list of genes used in each signature is shown libraries were sequenced in a HiSeq2500 (Illumina, San in Table S3. Diego, CA, USA), 2X100 Paired-end. Reads were The value of each of these signatures per sample aligned against the hg19 reference genome. was computed as the geometric mean of the expression values of all genes included in the signature. Next, the values of all samples for each signature were standard- 2.3. Mutation calling ized by subtracting the mean value of the signature in Mutation calling and subsequent filtering were per- the cohort and dividing by the standard deviation. formed using the Mutect2-GATK pipeline [21]. During the filtering process, cross-sample contamination was 2.7. Statistical analysis assessed. Those samples with a percentage of cross- sample contamination greater than 1% were discarded. Mann–Whitney Wilcoxon (MWW) tests were per- Resulting mutations were annotated using ANNO- formed using the Statannot python library. MWW VAR [22]. tests comparing TMB and immune infiltration scores across ICIs benefit groups were left-sided. MWW tests comparing SCNAs burden across ICIs benefit group 2.4. SCNAs calling were right-sided. Finally, MWW tests regarding TMB Given the difficulties of estimating SCNAs from WES or SCNAs burden differences across clinical features data, SCNAs assessment was performed using two were two-sided. independent methods: CNVkit [23] and Sequenza [24]. Survival analyses were performed using the lifelines CNVkit was run providing the tumor purity estimated python package. Univariate Cox proportional hazards by the pathologist as input. In the case of Sequenza, models were built for each feature separately. Also, sample purity was estimated by the tool itself. Only multivariate models were built combining features of regions found to bear SCNAs alterations by both interest. Additionally, multivariate models were strati- methods were considered for further analysis. Addi- fied by histology and smoking history of the patient. tionally, due to the limitations of assessing SCNAs in Variables correlation. The relationship between ana- low purity samples, those with an estimated tumor lyzed features was assessed using both Spearman’s purity by Sequenza of < 20 and a tumor purity rank and Pearson correlations. assessed by the pathologist of < 40 were discarded. 2.8. Validation cohorts 2.5. TMB and SCNAs burden The Non-Small Cell Lung Cancer MSKCC cohort Tumor mutational burden was computed as the sum (NSCLC-MSK) [20] was downloaded from cBioPortal of all exonic nonsynonymous mutations, insertions, [27] (https://www.cbioportal.org/study/summary?id= and deletions per sample. nsclc_pd1_msk_2018). SCNAs burden per patient was computed as the Patients with available TMB, FGA (estimation of sum of the sizes of all genomic regions affected by SCNAs burden), and PD-L1 expression data were SCNAs. selected and stratified into three groups based on their Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 3 Molecular determinants of ICI-associated long-term response J. Frigola et al. PFS following ICIs treatment (<6 months, 6- histology, sex, smoking status, and drug’s target 18 months, or > 18 months of PFS). Patients with a (Fig. S1). As expected, patients with smoking history PFS < 18 months but with no confirmed progression presented significantly higher TMB than never smokers (censored) were discarded. Then, TMB, SCNAs bur- (median 387 and 95, respectively; Mann–Whitney–Wil- den, and PD-L1 were compared between groups. coxon test (MWW) P = 0.004). Additionally, TMB, FGA, and PD-L1 expression We next interrogated the association between TMB was incorporated into a Cox proportional hazards and response to ICIs. Patients who presented durable model together with sex, age, and smoking status. clinical benefit (PFS > 6 months) had slightly higher Unlike the previous analysis, progression-free patients TMB than those with no clinical benefit (MWW, with a PFS under 18 months were included. This P = 0.029) (Fig. 1A). Patients were then stratified into model was stratified by tumor histology. three groups based on their PFS: no benefit (< A second cohort was used as further validation. 6 months), moderate (6-18 months), and long-term Data were downloaded from the supplementary mate- benefit (> 18 months). When comparing TMB distribu- rials of the publication [14]. Patients with both TMB tion across groups, patients with long-term benefit had and PD-L1 data were selected and stratified into three substantially higher TMB than those with moderate or groups based on their PFS as described above. Next, no benefit (MWW, P = 0.01 and 0.003, respectively) the distribution of TMB and PD-L1 expression (Fig. 1B), while no statistically significant differences between these groups was compared. were observed between these two groups (Fig. 1B). Additionally, Kaplan–Meier analysis of patients strati- fied into TMB tertiles indicated that patients in the 3. RESULTS upper tertile had longer PFS than those in the lower one (log-rank test, P = 0.003) (Fig. 1C). Somatic copy number alterations burden has also 3.1. Patients and clinical setting been reported to influence ICIs response [17,19,28]. To We assembled a cohort of 47 patients diagnosed with determine its importance in our cohort, we first inferred advanced NSCLC and treated with ICIs as monother- SCNAs for those samples with enough tumor purity (40 apy. Detailed clinical description is shown in Table S1 samples, see methodology) and computed the sum of and S2. 27 patients exhibited durable clinical benefit the size of all altered regions of the genome. Then, we defined as a progression-free survival (PFS) > related this score to patients’ PFS. No statistically sig- 6 months, while 20 progressed within this period. Our nificant differences were observed when stratifying cohort included 15 patients who exhibited a patients into the three categories of benefit defined PFS > 18 months, exceeding 36 months in 8 cases, above, even though we could observe that those with and 6 of whom remained without signs of progression clinical benefit tend to have lower SCNAs burden at the time of closing this study (reaching 60 months (Fig. 1D,E). Indeed, Kaplan–Meier analysis dividing in 3 cases). patients into tertiles based on their tumor’s SCNAs bur- den revealed that those in the upper tertile had statisti- cally significant shorter PFS than those in the 3.2. Tumor-intrinsic features and response to ICIs intermediate and lower tertiles (log-rank test, P = 0.029 We performed whole-exome sequencing (WES) from and 0.023, respectively) (Fig. 1F). These data suggest tumor samples of all 47 patients and their paired-nor- that high levels of SCNAs are associated with decreased mal sample. After appropriate quality controls, 44 PFS, while patients with moderate and low SCNAs bur- samples were found to be suitable for further analysis. den levels present no differences in response to ICIs. We determined the TMB for these 44 samples and cor- Finally, we assessed whether mutations or copy related it with different clinical features including number alterations in individual genes were associated Fig. 1. Tumor genomic characterization and response to ICIs. (A, B) Tumor mutational burden (TMB) distribution across groups of ICIs benefit. Color indicates best response to ICIs (CR: complete response, PR: partial response, SD: stable disease, and PD: progressive disease). (C) Kaplan–Meier plot dividing the cohort into TMB tertiles. (D, E) Somatic copy number alterations (SCNAs) burden distribution across groups of ICIs benefit. Color indicates best response. (F) Kaplan–Meier plot survival curves dividing the cohort into SCNAs burden tertiles. (G) Genomic alterations in selected genes. Columns and rows represent patients and genes, respectively. Mann–Whitney–Wilcoxon tests have been used to determine differences between ICIs benefit groups (A, B, D, E). Log-rank tests have been used to determine differences between TMB groups and between SCNAs groups (C, F). ns: 0.05 < P ≤ 1.0; *: 0.01 < P <0.05; **: 0.001 < P ≤ 0.01. 4 Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. J. Frigola et al. Molecular determinants of ICI-associated long-term response ** A B * CR C * ns PR SD 1.0 1200 1200 PD TMB high NA TMB mid 0.8 1000 1000 TMB low PFS probability 800 800 0.6 TMB TMB 600 600 0.4 400 400 ns ** 200 200 0.2 ns 0 0 0.0 <6 >6 <6 6–18 > 18 0 500 1000 1500 2000 ICIs PFS (months) ICIs PFS (months) Time (days) E ns D ns F CR ns ns PR SD 1e9 1e9 1.0 1.4 1.4 PD SCNAs high NA 1.2 1.2 SCNAs mid 0.8 SCNAs low 1.0 1.0 PFS probability SCNAs burden SCNAs burden 0.8 0.6 0.8 0.6 0.6 0.4 0.4 0.4 ns 0.2 0.2 0.2 * * 0.0 0.0 0.0 <6 >6 <6 6–18 > 18 0 500 1000 1500 2000 ICIs PFS (months) ICIs PFS (months) Time (days) G ICIs PFS Best response Histology TP53 ······· · ··· · ·· · ··· ····· · · KRAS ·· ·· ·· · · · · · STK11 · · · · · · KEAP1 · ·· · · CDKN2A · ·· · · ARID1A · · · · PTEN TAP1 · TAP2 TMB SCNAs burden ICIs benefit Histology Best ICIs response Gene alterations > 18 ADK CR Deleted 6–18 SCC PR Amplified <6 LCNEC SD Copy number unaltered · NOS PD Copy number not available NA Mutation Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 5 Molecular determinants of ICI-associated long-term response J. Frigola et al. with response (complete list of somatic alterations correlation between CD274 and PD-L1 was high Supplementary Tables S4 and S5). After appropriate (Spearman’s rank correlation, P = 0.0002, Fig. S2D). multiple-test correction, no alterations were found to be significantly associated with response. We also 3.4. Interplay between tumor-intrinsic and explored the distribution of somatic alterations in tumor-extrinsic features known lung cancer-related genes and genes associated with ICIs response (Fig. 1G). However, no single gene To better understand the determinants of response to was found to be significantly enriched in any of the ICIs, we also studied the interplay between tumor-in- groups (data not shown). trinsic and tumor microenvironment’s immune features. We found that SCNAs burden negatively correlates with most of the immune-related transcriptional signa- 3.3. Tumor immune-related transcriptional tures evaluated but not with CD274 (PD-L1) expression, signatures and response to ICIs while TMB appeared to be unrelated to all features ana- The amount and type of immune cells infiltrated within lyzed (Fig. 3A, Fig. S3A, S3B). As TMB and SCNAs the tumor have been suggested to influence response to burden are likely independent, we evaluated how their ICIs [9,12,13]. Thus, we obtained RNA expression data combination related to ICIs benefit. Thereby, we repre- from 22 patient’s biopsies by using a Nanostring panel sented patients’ TMB, SCNAs burden, and PFS and enriched in immune-related gene transcripts. We evalu- then divided the cohort into four groups by using the ated whether transcriptional signatures previously asso- SCNAs burden and TMB means as thresholds. This ciated with activated T cells were related to ICIs analysis revealed that almost all long-term responders response [9,12,19,26]. Indeed, patients with clinical ben- had above-average TMB and below-average SCNAs efit exhibited higher levels of these signatures compared burden, while patients whose tumors present opposite to those with lack of benefit; some were statistically sig- characteristics were nonresponders (Fig. 3B). Also, nificant while others were close to significance (TEFF patients with moderate benefit tended to have either score, MWW, P = 0.046) (Fig. 2A, Fig. S2A, S2B). high TMB or low SCNAs burden, but not both When stratifying patients into the three categories of (Fig. 3B). We observed three patients presenting high benefit, no differences were observed between moderate TMB and low SCNAs burden who did not benefit from and long-term benefit (Fig. 2B, Figs S2A, S2B), sug- ICIs treatment. Interestingly, in-depth inspection of gesting that the results in Fig. 2A are not likely driven these patients revealed that one of them had prolonged only by long-term responders. Of note, this analysis has benefit despite radiological progression (499 days of limited statistical power due to sample size limitations benefit after progression) and another presented a Large and should be taken cautiously. cell neuroendocrine carcinoma (LCNEC) tumor known Since the Nanostring panel encompasses a broad to respond poorly to ICIs. range of gene transcripts, we analyzed immune cell-type- Furthermore, Kaplan–Meier analysis based on these specific signatures [19] and found that B cells, CD8 effec- same four categories indicated that patients with high tive cells, natural killers, and T-reg were significantly TMB and low SCNAs exhibited statistically significant enriched in patients with benefit (MWW, P = 0.011, longer PFS than those with low TMB and high 0.035, 0.035, and 0.014, respectively), whereas M2 SCNAs burden, those with high TMB and high macrophages or other scores related to immunosuppres- SCNAs and near statistically significant than those sive microenvironments were balanced between both with low TMB and low SCNAs burden (log-rank test, groups or even exhibited a trend toward being higher in P = 0.006, 0.029, and 0.067, respectively) (Fig. 3C). nonresponders (Fig. 2C,D, Fig. S2B, S2C). Notably, high SCNAs burden seems to curtail benefit’s Finally, we examined CD274 (PD-L1) expression duration regardless of TMB levels, as none of the from the Nanostring data in these 22 patients and patients with outstanding response exhibited high found that high levels of CD274 expression were asso- SCNAs (Fig. 3C). ciated with clinical benefit (MWW, P = 0.040) Finally, TMB and SCNAs burden were integrated (Fig. 2E). Notably, stratifying benefit into moderate as continuous variables, together with patient’s sex and long term revealed that high CD274 expression and ICIs’ target, in a multivariate Cox proportional levels were most evident in long-term responders com- hazards model stratified by histology and smoking sta- pared to patients with no benefit (MWW, P = 0.014) tus. This analysis indicated that both TMB and (Fig. 2F). It is worth highlighting that for a few SCNAs combined are significantly associated with ICIs patients with Nanostring data, PD-L1 expression by response even when correcting by potential confound- immunohistochemistry was also available. In those, ing factors (Fig. 3D). 6 Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. J. Frigola et al. Molecular determinants of ICI-associated long-term response ns ns ns ns A ns B ns * ns 3 3 2.0 2.0 Cytotoxic immune score Cytotoxic immune score 1.5 2 1.5 2 TEFF score TEFF score 1.0 1.0 1 1 0.5 0.5 0.0 0.0 0 0 −0.5 −0.5 −1.0 −1 −1.0 −1 <6 >6 <6 >6 <6 6–18 > 18 <6 6–18 > 18 ICIs PFS (months) ICIs PFS (months) ICIs PFS (months) ICIs PFS (months) CR PR SD PD NA C D * * 2 2 B cell Treg 1 NK cells 1 B cells CD8 effector 0 NK 0 Macrophages M1 −1 −1 Macrophages M2 <6 >6 <6 >6 ICIs benefit Immune score ICIs PFS (months) ICIs PFS (months) > 18 6–18 <6 2.5 −2.5 * ns * * E * F ns 2.0 500 500 2 1.5 CD274 expression CD274 expression 400 400 CD8 effector 1.0 1 Treg 0.5 300 300 0 0.0 200 200 −0.5 100 100 −1 −1.0 0 0 <6 >6 <6 >6 <6 >6 <6 6–18 > 18 ICIs PFS (months) ICIs PFS (months) ICIs PFS (months) ICIs PFS (months) CR PR SD PD NA CR PR SD PD NA Fig. 2. Immune expression profiling of tumor’s microenvironment. (A, B) TEFF and cytotoxic score distribution across groups of ICIs benefit. For each patient, the value of each signature has been computed as the geometric mean of the expression of all genes in the signature. Next, for each of the scores, all values have been standardized. Color indicates best response to ICIs (CR: complete response, PR: partial response, SD: stable disease, and PD: progressive disease). (C) Distribution of expression signatures belonging to four different immune cell populations across groups of benefit. Values computed as in panel A. Color indicates best response to ICIs. (D) Representation of different immune populations across 22 patients. Columns and rows correspond to patients and specific immune cell populations, respectively. Patients have been clustered using a hierarchical clustering algorithm based on their degree of similarity considering amount and type of tumor immune infiltration. (E) CD274 expression distribution according to indicated group of benefit. Color indicates best response. Mann– Whitney–Wilcoxon tests have been used to determine differences between ICIs benefit groups (A, B, C, E, F); ns: 0.05 < P ≤ 1.0; *: 0.01 < P <0.05; **: 0.001 < P ≤ 0.01. Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 7 Molecular determinants of ICI-associated long-term response J. Frigola et al. A Correlation SCNAs 0.08 coefficient burden 0.8 B ICIs benefit TEFF 0.99 0.002 1e9 0.4 1.4 > 18 6–18 0.0 1.2 B cells 0.19 0.005 0.0026 <6 −0.4 1.0 Treg 1 0.0081 7.9e-08 0.0014 SCNAs burden −0.8 0.8 NK cells 0.94 0.00049 3.5e-09 0.00011 3.3e-06 0.6 0.4 CD8 effector 0.83 0.0035 2.9e-07 0.00041 7.9e-07 3.4e-08 0.2 CD274 0.73 0.11 0.0021 0.061 0.00026 0.037 0.0065 0.0 0 200 400 600 800 1000 1200 en g lls lls or B FF e TM ct ce ce TMB rd Tr TE fe bu B K ef N As 8 D N C SC Univariate Multivariate C D 1.0 Q1 Q2 Pval SCNAs ** burden * ns 0.8 ns ns ns ICIs target PFS probability (PD-1) 0.6 0.4 Sex (Male) 0.2 TMB 0.0 0 500 1000 1500 2000 −1 0 1 −1 0 1 Time (days) log (HR) (95% CI) log (HR) (95% CI) TMB high SCNA high TMB low SCNA high P < 0.05 TMB low SCNA low TMB high SCNA low Fig. 3. Interplay between biomarkers of response to ICIs. (A) Spearman correlation between indicated features. Color represents the correlation coefficient. The P value of each correlation is shown within each cell. (B) Scatter plot representation of TMB and SCNAs burden. Horizontal and vertical gray lines represent the SCNAs burden and TMB mean, respectively. Dot color represents the group of benefit. (C) Kaplan–Meier plot survival curves dividing the cohort into four groups based on the quadrants defined in panel 3B. Long-rank tests have been used to determine differences between TMB-SCNAs groups. ns: 0.05 < P ≤ 1.0; *: 0.01 < P <0.05; **: 0.001 < P ≤ 0.01. (D) Univariate and multivariate Cox proportional hazards model for SCNAs burden, ICIs target, sex, and TMB. Multivariate analysis has been stratified by histology and smoking history. Features represented in blue are statistically significant (P < 0.05). Importantly, similar results were obtained when per- 3.5. Validation through analysis of publicly forming the analysis above using patient’s overall sur- available data vival (OS)—instead of PFS—as a measure of benefit (Fig. S4A, S4B). As stated in the introduction, NSCLC long-term Altogether, these data indicate the utility of both responders to ICIs remain poorly molecularly charac- biomarkers to discriminate patients with benefit to ICIs terized and indeed integrative studies using genomic in a cohort of patients enriched in long-term responders. and transcriptomic data are almost nonexistent. 8 Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. J. Frigola et al. Molecular determinants of ICI-associated long-term response Nevertheless, we sought to validate some of our find- reduced number of patients with CD274 (PD-L1) gene ings by reanalyzing a publicly available cohort expression data. Nevertheless, we could create it using (NSCLC-MSK) for which genomic data (by targeted the NSCLC-MSK cohort. Thus, we incorporated next-generation sequencing) as well as PD-L1 expres- TMB, SCNAs burden, and PD-L1 expression, together sion data (by IHC) were available [20]. Of note, both with other clinically relevant features to a multivariate TMB and SCNAs burden (referred to as Fraction of Cox proportional hazards model, which was stratified Genome Altered (FGA) in the original manuscript) by tumor histology. Our analysis indicates that all were estimated from a gene panel, which owing to the three features are statistically associated with patient’s size of the fraction of the genome included in the panel PFS (Fig. 4D). and the type of genes included (mostly cancer-related Finally, the increased TMB and PD-L1 levels in genes) is likely less accurate than WES or whole-genome long-term responders were further validated using a sequencing (especially when estimating SCNAs burden). second independent cohort treated with PD-1 plus We stratified patients into the three same categories CTLA-4 blockade [14] (Fig. S5A, S5B). based on their PFS (no benefit, moderate, and long- Thus, the analysis of two independent datasets con- term benefit) as we did in Fig. 1B,E. As observed in firms our findings, highlighting the importance of our cohort, patients with long-term benefit had sub- TMB to discriminate long-term responders. Addition- stantially higher TMB than those with moderate or no ally, we showed the value of combining TMB, PD-L1, benefit (MWW, P = 0.036 and 0.012, respectively) and SCNAs burden as complementary determinants of (Fig. 4A), while no statistically significant differences ICIs response. were observed between these two groups (Fig. 4A). Additionally, we observed that long-term responders 4. DISCUSSION had substantially higher PD-L1 expression (MWW, P = 0.013) and lower SCNAs burden (MWW, Advanced NSCLC treatment strategies have substan- P = 0.023) than patients with no benefit (Fig. 4B,C). tially changed since the emergence of ICIs. However, The integrative analysis performed on our cohort only a fraction of patients benefits from this type of (Fig. 3A) suggests the potential value of combining immunotherapy, and most that do eventually acquire TMB, SCNAs burden, and PD-L1 expression. In our resistance. Thus, there is a need to find biomarkers to cohort, we did not build such model due to the identify patients who will benefit from ICIs, A B FGA (SCNAs burden) TMB ICIs PFS (months) ICIs PFS (months) D Fig. 4. NSCLC-MSK cohort analysis. (A) FGA (SCNAs burden) TMB, (B) FGA (SCNAs burden), and (C) PD-L1 distribution across groups of ICIs C Age benefit. Mann–Whitney–Wilcoxon tests have been used to determine differences TMB between clinical benefit groups (A, B, C). ns: 0.05 < P ≤ 1.0; *: 0.01 < P < 0.05. (D) Smoker (yes) PD-L1 Multivariate Cox proportional hazards Sex (male) model for FGA (SCNAs burden), age, P < 0.05 TMB, smoking status, sex, and PD-L1, PD-L1 P > 0.05 stratified by histology. Features represented in blue are statistically significant (P < 0.05). ICIs PFS (months) Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 9 Molecular determinants of ICI-associated long-term response J. Frigola et al. specifically those who will present a long-lasting in previous studies, such as individual gene mutations response which, despite a recent publication [8], remain and copy number alterations [29,32,33]. However, we largely unexplored. could not identify a single gene associated with To this end, we gathered a new cohort of NSCLC response or long-term benefit after appropriate statisti- patients treated with ICIs enriched in long-term cal correction; even though this might not be surpris- responders and analyzed genomic and transcriptomic ing as the inferred number of patients required to features aiming to better understand the determinants identify individual genes was predicted to be substan- of ICI-associated long-term response. Additionally, we tial [29]. The degree of homozygosis of human leuko- validated our findings by reanalyzing public data from cyte antigen (HLA) genes and the HLA allele types previous studies [14,20]. has been also shown to influence ICIs response [34]. The relevance of TMB as biomarker to predict While we did study whether there was an association response upon ICIs treatment is under debate between HLA homozygosis and response to ICIs in [8,15,18,20,29–31]. We believe that our data shed light our cohort, we did not observe a trend toward more on this topic, as we found that the association between heterozygosis in any group (data not shown). clinical benefit and TMB is mainly driven by patients We also presented gene transcription data, which with long-term response (>18 months), whose TMB is indicates that T-cell activity-related transcriptional sig- much higher than the rest of patients, while we did not natures are associated with clinical benefit to ICIs, observe substantial differences regarding TMB between consistent with previous reports [9,10,12,13]; however, patients with no or moderate clinical benefit. This it does not seem to discriminate between moderate and observation, originally based on our cohort, was fur- long-term response. Additionally, we identified specific ther validated by reanalyzing two publicly available immune cell-type populations enriched in responders, cohorts [14,20]. Thus, in an unselected cohort, in which some of which (CD8, natural killers, B cells) have been the proportion of individuals with long-lasting benefit related to the ability of the immune system to elimi- would be small, the association between TMB and ICIs nate tumor cells [35,36]. Notably, a recent report response could easily go unnoticed. Our data indicate found that PD-L1 expression was the most reliable that TMB might be a biomarker of long-term response biomarker associated with long-term overall survival (PFS > 18 months) rather than of durable clinical ben- upon PD-1 blockade [8]. Consistent with this study, efit (PFS > 6 months). Thus, there is a subgroup of our results evaluating CD274 (PD-L1) gene transcript patients who benefit from ICIs treatment (those with a levels from a Nanostring panel indicate a similar PFS between 6 and 18 months), but do not exhibit result. Similarly, we found that patients with long-last- higher TMB than those without benefit, a feature that ing benefit following ICIs exhibited higher levels of seems to be specific of long-term responders. These PD-L1 in two independent cohorts. patients, with moderate benefit, must present other fea- We also investigated the interplay between tumor-in- tures that would explain this positive response to the trinsic features and immune infiltration. It is worth treatment. highlighting that we found that immune-related signa- In fact, we evaluated the influence of SCNAs burden, tures negatively correlated with SCNAs burden as which remains less studied particularly in advanced reported from TCGA data [19], which could explain NSCLC. Our data highlight its importance as determi- why patients whose tumors bore high levels of chro- nant of response to ICIs, as patients with high SCNAs mosomal aberrations responded poorly to ICIs. In burden exhibited a poorer response than the rest. In contrast to SCNAs burden and consistent with other contrast, no response differences seem to exist between studies [8,11,20], our data indicate that TMB does not patients with medium or low SCNAs burden. Alto- correlate with immune-related transcriptional signa- gether it suggests a detrimental effect of high levels of tures nor with PD-L1. It is worth mentioning that chromosomal aberrations—presumably chromosomal CD274 (PD-L1) expression, in contrast to other instability—upon ICIs treatment, which could be at immune-related signatures, did not seem to correlate least partially explained by its negative correlation with with SCNAs burden. Tumors may evade the immune immune infiltration as discussed below. system through different mechanisms such as express- Furthermore, TMB and SCNAs are independent ing PD-L1 or by preventing infiltration of immune biomarkers and our data indicate the value of combin- cells within the tumor, thereby an elevated burden of ing both features to discriminate those patients who SCNAs might be indicative of the latter. will achieve sustained benefit upon ICIs treatment. Thus, both SCNAs burden and PD-L1 expression Beyond these features, other tumor genomic alter- might provide complementary information, which in ations have been reported to be indicators of response combination with TMB can enhance our ability to 10 Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. J. Frigola et al. Molecular determinants of ICI-associated long-term response identify patients who will exhibit sustained response to PFS or OS) better than each feature individually. ICIs. In our cohort, we could not build a model Notably, our observation that TMB is strongly including the three features together due to sample size enriched in long-term responders to ICIs was validated limitations. Nevertheless, we could evaluate them in in two independent cohorts, suggesting that TMB the NSCLC-MSK cohort and we showed, using multi- might be a biomarker of long-lasting benefit rather variate analysis, that the three factors are indeed sig- than durable clinical benefit (6 months of PFS). nificantly associated with ICIs response. Finally, using the NSCLC-MSK cohort, we described Similarly, the relatively low number of patients with that TMB, SCNAs burden, and PD-L1 are signifi- expression data limits our capacity to discriminate cantly associated with clinical benefit following ICIs between medium and long-term benefit and curtails treatment. the use of this data in survival analyses (i.e., multivari- ate model). Thus, we could not assess the value of Acknowledgements combining these gene transcription-based scores together with TMB to better identify patients with We acknowledge the CELLEX Foundation for provid- long-term benefit. ing research facilities and equipment. We thank Javier Another limitation of our study is that we could not Gonzalo from VHIO for project management support. evaluate intra-tumoral heterogeneity within the pri- We thank our patients and their families for accepting mary tumor [37], as well as possible differences to participate in this study. This work was supported between primary tumor and metastases. Certainly, by the Fundacion Cientifica Asociacion Espa~ nola Con- analysis of multiple sites of the same primary tumor tra el Cancer-AECC [grant number GCB14142170 to and/or metastases would provide a higher accuracy LMM, MS-C, and EF]; Instituto Carlos III-ISCIII when estimating TMB, SCNAs burden, and tumor [FIS PI17/00938 to EF and AV]; the Catalan Govern- immune infiltration. However, obtaining these samples ment/AGAUR [2017–SGR–1738 to EF]. is undoubtedly challenging in advanced NSCLC patients. Additionally, WES might not be the best Conflict of interest approach to determine SCNAs along the genome; hence, other techniques such as low pass whole-gen- EF reports the following conflicts of interest: advisory ome sequencing could offer more accurate results. role or speaker’s bureau: AbbVie, AstraZeneca, Ber- Finally, our cohort is strongly enriched in patients GenBio, Blueprint medicines, Boehringer Ingelheim, with long-term benefit upon ICIs treatment, allowing a Bristol-Meyers Squibb, Celgene, Eli Lilly, Guardant detailed study of this subgroup, which was the primary Health, Janssen, Medscape, Merck KGaA, Merck aim of our study. Therefore, our cohort does not fully Sharp & Dohme, Novartis, Pfizer, priME Oncology, recapitulate the clinical reality in advanced NSCLC, as Roche, Samsung, Takeda, Touchtime. Board: Grifols, in a cohort of unselected patients, long-term ICIs independent member. Research funding: Fundaci on responders would be few. Nevertheless, we validated Merck Salud, Grant for Oncology Innovation EMD our findings by analyzing two independent cohorts Serono. AV reports advisory role: Sysmex, Novartis, with publicly data available. Merck, Roche, Bristol-Meyers Squibb, Guardant Health, and research funding: Bristol-Meyers Squibb, Novartis, Debio, Sysmex, Cellestia Biotech, Roche. 5. CONCLUSION PN has consulted for Bayer, Novartis, MSD, and Tar- Our data indicate that high TMB moderately associ- gos, and received compensation. IS reports advisory ates with durable clinical benefit to ICIs, while it is role, speaker’s bureau or travel compensation: Roche strongly correlated with long-term response. Con- Farma, Abbvie, Roche Diagnostics, Merck Sharp & versely, high SCNAs burden is indicative of lack of Dohme, Pfizer, Takeda, Bristol-Myers Squibb, Lilly, response. Additionally, patients who benefit from ICIs Sysmex, Boehringer Ingelheim. CC has been partially treatment present higher levels of immune infiltration supported by Grant for Oncology EMD Serono signatures, even though this does not seem suggestive research funding to EF. AMM provided consultation, of benefit’s duration. Additionally, CD274 expression attended advisory boards and/or speaker’s bureau for is particularly high in long-term responders. TMB is the following organizations: BMS, Roche, MSD, Pfi- independent of SCNAs burden or tumor immune infil- zer, Boehringer Ingelheim, AstraZeneca. AN reports tration, which are negatively correlated. Combining advisory role, speaker’s bureau or travel compensa- TMB and SCNAs burden allows to discriminate tion: Bristol-Myers Squibb, F. Hoffmann La Roche patients with ICI-associated long-term benefit (either AG, Pfizer, Boehringer Ingelheim, Oryzon Genomics, Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 11 Molecular determinants of ICI-associated long-term response J. Frigola et al. Merck Sharp & Dohme. SC Bristol-Myers Squibb REFERENCES Recipient F, Hoffmann La Roche AG, Pfizer, Boeh- ringer Ingelheim, MSD Oncology, Amphera. AC 1 Borghaei H, Paz-Ares L, Horn L, Spigel DR, Steins M, reports advisory role and/or travel compensation: Bris- Ready NE, Chow LQ, Vokes EE, Felip E, Holgado E tol-Myers Squibb Recipient, F. Hoffmann La Roche et al. (2015) Nivolumab versus docetaxel in advanced AG, Pfizer, Boehringer Ingelheim, MSD Oncology, nonsquamous non-small-cell lung cancer. N Engl J Med Kyowa Kirin, Celgene. PI reports advisory role and/or 373, 1627–1639. travel compensation: Bristol-Myers Squibb Recipient, 2 Herbst RS, Baas P, Kim DW, Felip E, Perez-Gracia JL, F. Hoffmann, La Roche AG, Merck Sharp & Dohme, Han J-Y, Molina J, Kim J-H, Arvis CD, Ahn M-J et al. (2016) Pembrolizumab versus docetaxel for Boehringer Ingelheim, MSD Oncology, Rovi, Yowa previously treated, PD-L1-positive, advanced non-small- Kirin, Grunenthal Pharma S.A., Pfizer. NP reports cell lung cancer (KEYNOTE-010): A randomised advisory role and/or travel compensation: Bristol- controlled trial. Lancet 387, 1540–1550. Myers Squibb Recipient, F. Hoffmann La Roche AG, 3 Garon EB, Rizvi NA, Hui R, Balmanoukian AS, Eder Pfizer, Boehringer Ingelheim. AP reports advisory role JP, Patnaik A, Aggarwal C, Gubens M, Horn L, and/or travel compensation: Pfizer, Novartis, Roche, Carcereny E et al. (2015) Pembrolizumab for the MSD Oncology, Lilly, Daiichi Sankyo, Amgen, Boeh- treatment of non-small-cell lung cancer. N Engl J Med ringer, PUMA, Oncolytics Biotech, Abbvie, Nanos- 372, 2018–2028. tring Technologies. All remaining authors have 4 Garon EB, Hellmann MD, Rizvi NA, Carcereny E, declared no conflicts of interest. Leighl NB, Ahn MJ, Eder JP, Balmanoukian AS, Aggarwal C, Horn L et al. (2019) Five-year overall Author contributions survival for patients with advanced non-small-cell lung cancer treated with pembrolizumab: Results from the Joan Frigola involved in conceptualization, data cura- phase i KEYNOTE-001 study. J Clin Oncol 37, 2518– tion, formal analysis, methodology, visualization, writ- 2527. ing—review and editing, software. Alejandro Navarro 5 Kalbasi A & Ribas A (2020) Tumour-intrinsic resistance involved in resources, data curation, writing—review to immune checkpoint blockade. Nat Rev Immunol 20, and editing. Caterina Carbonell involved in resources, 25–39. investigation, writing—review &and editing. Ana Call- 6 Planchard D, Popat S, Kerr K, Novello S, Smit EF, ejo involved in resources, writing—review and editing. Faivre-Finn C, Mok TS, Reck M, Van Schil PE, Patricia Iranzo involved in resources, writing—review Hellmann MD et al. (2018) Metastatic non-small cell and editing. Alex Martinez-Marti involved in resources, lung cancer: ESMO Clinical Practice Guidelines for writing—review and editing. Nadia Saoudi-Gonzalez diagnosis, treatment and follow-up. Ann Oncol 29 involved in resources, writing—review and editing. (Supplement, 4):iv192–iv237. Nuria Pardo involved in resources, writing—review 7 Reck M, Rodriguez-Abreu D, Robinson AG, Hui R, and editing. Susana Cedres involved in resources, writ- Cs}oszi T, F€ul€ op A, Gottfried M, Peled N, Tafreshi A, ing—review and editing. Debora Martinez involved in Cuffe S et al. (2016) Pembrolizumab versus investigation, writing—review and editing. Jose Chemotherapy for PD-L1-Positive Non-Small-Cell Lung Jimenez involved in resources, writing—review and Cancer. N Engl J Med 375, 1823–1833. 8 Hu-Lieskovan S, Lisberg A, Zaretsky JM, Grogan TR, editing. Irene Sansano involved in resources, writing— Rizvi H, Wells DK, Carroll J, Cummings A, review and editing. Francesco M Mancuso involved in Madrigal J, Jones B et al. (2019) Tumor characteristics data curation, formal analysis, writing—review and associated with benefit from pembrolizumab in editing. Paolo Nuciforo involved in resources, writing advanced non–small cell lung cancer. Clin Cancer Res —review and editing. Luis M Montuenga involved in 25, 5061–5068. writing—review and editing and formal analysis. 9 Herbst RS, Soria JC, Kowanetz M, Fine GD, Hamid Montserrat S anchez-Cespedes involved in writing—re- O, Gordon MS, Sosman JA, McDermott DF, Powderly view and editing, formal analysis. Aleix Prat involved JD, Gettinger SN et al. (2014) Predictive correlates of in writing—review and editing, resources, investigation. response to the anti-PD-L1 antibody MPDL3280A in Ana Vivancos involved in investigation, writing—re- cancer patients. Nature 515, 563–567. view and editing, resources. Enriqueta Felip involved 10 Van Allen EM, Miao D, Schilling B, Shukla SA, Blank in conceptualization, funding acquisition, supervision, C, Zimmer L, Sucker A, Hillen U, Foppen MHG, writing—review and editing. Ramon Amat involved in Goldinger SM et al. (2016). Genomic correlates of conceptualization, investigation, supervision, writing— response to CTLA-4 blockade in metastatic melanoma. original draft, writing—review and editing. Science 352, 207–212. 12 Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. J. Frigola et al. Molecular determinants of ICI-associated long-term response 11 Cristescu R, Mogg R, Ayers M, Albright A, Murphy E, Hollmann T, Chatila W et al. (2018) Molecular Yearley J, Sher X, Liu XQ, Lu H, Nebozhyn M et al. determinants of response to anti-programmed cell death (2018). Pan-tumor genomic biomarkers for PD-1 (PD)-1 and anti-programmed death-ligand 1 (PD-L1) checkpoint blockade-based immunotherapy. Science. blockade in patients with non-small-cell lung cancer 362, 197–207. https://doi.org/10.1126/science.aar3593 profiled with targeted next-generation sequencing. J 12 Ayers M, Ribas A, Mcclanahan TK, Murphy E, Clin Oncol 36, 633–641. Loboda A, Kaufman DR, Albright A, Cheng JD, Kang 21 Cibulskis K, Lawrence MS, Carter SL, Sivachenko A, SP, Shankaran V et al. (2017) IFN-c–related mRNA Jaffe D, Sougnez C, Gabriel S, Meyerson M, Lander profile predicts clinical response to PD-1 blockade Find ES, Getz G. (2013). Sensitive detection of somatic point the latest version : IFN-c–related mRNA profile mutations in impure and heterogeneous cancer samples. predicts clinical response to PD-1 blockade. J Clin Nature Biotechnology. 31, 213–219. http://dx.doi.org/10. Invest 127, 2930–2940. 1038/nbt.2514 13 Prat A, Navarro A, Pare L, Reguart N, Galvan P, 22 Wang K, Li M & Hakonarson H (2010) ANNOVAR: Pascual T, Martınez A, Nuciforo P, Comerma L, Alos Functional annotation of genetic variants from high- L et al. (2017) Immune-related gene expression profiling throughput sequencing data. Nucleic Acids Res 38, 1–7. after PD-1 blockade in non–small cell lung carcinoma, 23 Talevich E, Shain AH, Botton T & Bastian BC (2016) head and neck squamous cell carcinoma, and CNVkit: Genome-Wide Copy Number Detection and melanoma. Cancer Res 77, 3540–3550. Visualization from Targeted DNA Sequencing. PLoS 14 Hellmann MD, Nathanson T, Rizvi H, Creelan BC, Comput. Biol 12, 1–18. Sanchez-Vega F, Ahuja A, Ni A, Novik JB, Mangarin 24 Favero F, Joshi T, Marquard AM, Birkbak NJ, LMB, Abu-Akeel M et al. (2018) Genomic features of Krzystanek M, Li Q, Szallasi Z, Eklund AC. (2015). response to combination immunotherapy in patients Sequenza: allele-specific copy number and mutation with advanced non-small-cell lung cancer. Cancer Cell profiles from tumor sequencing data. Annals of 33, 843–852.e4. Oncology. 26, 64–70. http://dx.doi.org/10.1093/annonc/ 15 Rizvi NA, Hellmann MD, Snyder A, Kvistborg P, mdu479. Makarov V, Havel JJ, Lee W, Yuan J, Wong P, Ho 25 Pare L, Pascual T, Seguı E, Teixid o C, Gonzalez-Cao TS. (2015). Mutational landscape determines sensitivity M, Galvan P, Rodrıguez A, Gonzalez B, Cuatrecasas to PD-1 blockade in non-small cell lung cancer. M, Pineda E et al. (2018) Association between PD1 Science. 348, 124–128. http://dx.doi.org/10.1126/scie mRNA and response to anti-PD1 monotherapy across nce.aaa1348 multiple cancer types. Ann Oncol 29, 2121–2128. 16 McDermott DF, Huseni MA, Atkins MB, Motzer RJ, 26 Karachaliou N, Gonzalez-Cao M, Crespo G, Rini BI, Escudier B, Fong L, Joseph RW, Pal SK, Drozdowskyj A, Aldeguer E, Gimenez-Capitan A, Reeves JA et al. (2018) Clinical activity and molecular Teixido C, Molina-Vila MA, Viteri S, De Los Llanos correlates of response to atezolizumab alone or in Gil M et al. (2018). Interferon gamma, an important combination with bevacizumab versus sunitinib in renal marker of response to immune checkpoint blockade in cell carcinoma. Nat Med 24, 749–757. non-small cell lung cancer and melanoma patients. Ther 17 Roh W, Chen PL, Reuben A, Spencer CN, Prieto PA, Adv Med Oncol 10, 1–23. https://doi.org/10.1177/ Miller JP, Gopalakrishnan V, Wang F, Cooper ZA, 1758834017749748 Reddy SM et al. (2017) Integrated molecular analysis of 27 Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, tumor biopsies on sequential CTLA-4 and PD-1 Aksoy BA, Jacobsen A, Byrne CJ, Heuer ML, Larsson blockade reveals markers of response and resistance. E, Antipin Y, Reva B, Goldberg AP, Sander C, Schultz Sci Transl Med 9, eaah3560. N. (2012). The cBio Cancer Genomics Portal: An Open 18 Liu D, Schilling B, Liu D, Sucker A, Livingstone E, Platform for Exploring Multidimensional Cancer Jerby-Arnon L, Zimmer L, Gutzmer R, Satzger I, Genomics Data: Figure 1. Cancer Discovery. 2, Loquai C, et al. (2019) Integrative molecular and 401–404. http://dx.doi.org/10.1158/2159-8290.cd-12-0095 clinical modeling of clinical outcomes to PD1 blockade 28 Liu L, Bai X, Wang J, Tang XR, Wu DH, Du SS, Du in patients with metastatic melanoma. Nat Med 25, XJ, Zhang Y-W, Zhu HB, Fang Y, Guo ZQ, Zeng Q, 1916–1927. Guo XJ, Liu Z, Dong ZY. (2019). Combination of 19 Davoli T, Uno H, Wooten EC, Elledge SJ. (2017). TMB and CNA Stratifies Prognostic and Predictive Tumor aneuploidy correlates with markers of immune Responses to Immunotherapy Across Metastatic evasion and with reduced response to immunotherapy. Cancer. Clinical Cancer Research. 25, 7413–7423. http:// Science.355, eaaf8399. http://dx.doi.org/10.1126/scie dx.doi.org/10.1158/1078-0432.ccr-19-0558 nce.aaf8399 29 Miao D, Margolis CA, Vokes NI, David Liu D, 20 Rizvi H, Sanchez-Vega F, Jonsson P, Halpenny D, Taylor-Weiner A, Wankowicz SM, Adeegbe D, Keliher Plodkowski A, Long N, Sauter JL, Rekhtman N, D, Schilling B, Tracy A et al. (2018) Genomic Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. 13 Molecular determinants of ICI-associated long-term response J. Frigola et al. correlates of response to immune checkpoint blockade metastasis, and survival in colorectal cancer. N Engl J in microsatellite-stable solid tumors. Nat Genet 50, Med 353, 2654–2666. 1271–1281. 36 Helmink BA, Reddy SM, Gao J, Zhang S, Basar R, 30 Carbone DP, Reck M, Paz-Ares L, Creelan B, Horn L, Thakur R, Yizhak K, Sade-Feldman M, Blando J, Han Steins M, Felip E, van den Heuvel MM, Ciuleanu TE, G et al. (2020) B cells and tertiary lymphoid structures Badin F et al. (2017) First-line nivolumab in stage IV promote immunotherapy response. Nature 577, 549– or recurrent non-small-cell lung cancer. N Engl J Med 555. 376, 2415–2426. 37 Jamal-Hanjani M, Wilson GA, McGranahan N, 31 Hellmann MD, Paz Ares L, Bernabe Caro R, Zurawski Birkbak NJ, Watkins TBK, Veeriah S, Shafi S, Johnson B, Kim SW, Carcereny Costa E, Park K, Alexandru A, DH, Mitter R, Rosenthal R et al. (2017) Tracking the Lupinacci L, de la Mora Jimenez E et al. (2019) evolution of non-small-cell lung cancer. N Engl J Med Nivolumab plus ipilimumab in advanced non-small-cell 376, 2109–2121. lung cancer. N Engl J Med 381, 2020–2031. 32 Shen J, Ju Z, Zhao W, Wang L, Peng Y, Ge Z, Nagel ZD, Zou J, Wang C, Kapoor P et al. (2018) ARID1A Supporting information deficiency promotes mutability and potentiates Additional supporting information may be found therapeutic antitumor immunity unleashed by immune online in the Supporting Information section at the end checkpoint blockade. Nat Med 24, 556–562. of the article. 33 Wong K-K, Gainor JF, Wolchok JD, Hellmann MD, Fig S1. Patients’ TMB distribution across clinical fea- Awad MM, Gainor JF, Schrock AB, Hartmaier RJ, tures. Trabucco SE, Gay L et al. (2018) STK11/LKB1 Fig S2. Expression profiling of tumor microenviron- mutations and PD-1 inhibitor resistance in KRAS - ment. mutant lung adenocarcinoma. Cancer Discov 8, 822– Fig S3. Correlations between features. 835. Fig S4. Overall survival analysis. 34 Chowell D, Morris LGT, Grigg CM, Weber JK, Fig S5. Cancer Cell validation cohort analysis. Samstein RM, Makarov V, Kuo F, Kendall SM, Requena D, Riaz N et al. (2018). Patient HLA class I Table S1. Cohort clinical data summary. genotype influences cancer response to checkpoint Table S2. Patient information and clinical outcome. blockade immunotherapy. Science. 359, 582–587. Table S3. Gene list for gene expression signatures. 35 Pages F, Berger A, Camus M, Sanchez-Cabo F, Costes Table S4. Mutations per patient. A, Molidor R, Mlecnik B, Kirilovsky A, Nilsson M, Table S5. Copy number alterations per patient. Damotte D et al. (2005) Effector memory T cells, early Supplementary Material 14 Molecular Oncology (2021) ª 2020 The Authors. Published by FEBS Press and John Wiley & Sons Ltd.