Optimizing cancer immunotherapy response prediction by tumor aneuploidy score and fraction of copy number alterations

Optimizing cancer immunotherapy response prediction by tumor aneuploidy score and fraction of copy number alterations | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Brief Communication Optimizing cancer immunotherapy response prediction by tumor aneuploidy score and fraction of copy number alterations Tiangen Chang, Yingying Cao, Eldad Shulman, Alejandro Schäffer, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-2644875/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 03 Jun, 2023 Read the published version in npj Precision Oncology → Version 1 posted 11 You are reading this latest preprint version Abstract Identifying patients with low tumor mutation burden (TMB) that are likely to respond to cancer immunotherapy is an important, yet highly challenging clinical need. Using 3,139 patients across 17 different cancer types, we comprehensively studied the ability of two common copy number alteration (CNA) scores – the tumor aneuploidy score (AS) and the fraction of genome encompassed by copy number alterations (FGA) – to predict survival following immunotherapy in both pan-cancer and individual cancer types. We propose an elbow-point based method to optimize the cutoff used for calling CNAs. The optimized AS and FGA scores show significantly improved predictive performance compared to the arbitrary cutoffs reported in the literature. However, our data suggests that the use of AS and FGA for predicting immunotherapy response is currently limited to only a few cancer types. Therefore, larger sample sizes are needed to evaluate the clinical utility of these measures for patient stratification in other cancer types. Biological sciences/Cancer/Tumour biomarkers Biological sciences/Computational biology and bioinformatics Biological sciences/Cancer/Cancer genomics Cancer Fraction of copy number alterations Immunotherapy response Low-TMB Tumor aneuploidy Figures Figure 1 Figure 2 Figure 3 Introduction Various studies have shown that high tumor mutation burden (TMB) may predict immunotherapy response, at least in some cancer types 1,2 . However, identifying patients with low TMB that are likely to respond to cancer immunotherapy is a challenging unmet clinical need. One promising approach to identify low-TMB responders of immunotherapy has been to study the predictive ability of other measures of genomic alterations in cancer in these patients. Two natural candidates are scores based on copy number alterations (CNAs): (a) tumor aneuploidy, which measures chromosome-level CNAs, and (b) global genomic CNAs, which quantifies the extents of both chromosomal and focal copy-number events 3 . Both tumor aneuploidy and genomic CNAs have been shown to play a role in cancer progression and to be predictive for cancer prognosis 3-5 . Recently, Spurr et al. reported that the tumor aneuploidy score (AS), defined as the fraction of chromosome arms with arm-level copy number alterations in a sample, is significantly predictive of survival following immunotherapy in low-TMB patients in a pan-cancer analysis 6 . In addition, they reported that AS had stronger predictive power than another metric conceptually related to the AS, the fraction of genome encompassed by copy number alterations (FGA) which quantifies the extent of both chromosomal and focal copy-number events 6 . Intrigued by these potentially clinically impactful findings, we set out to explore several related fundamental questions: (1) Is AS predictive of survival of low-TMB patients following immunotherapy in individual cancer types? (2) Does the choice of cutoff during CNA calling influence the predictive power? (3) Under optimal calling cutoffs, is AS or FGA a better predictor? We first re-analyzed the same data used by Spurr et al. 6 , i.e., the Samstein et al. cohort 1 from MSK-IMPACT (data acquired from the AACR Project GENIE). This study has analyzed a published cohort of 1,660 advanced cancer patients from 10 different cancer types treated with immune checkpoint blockade (ICB). As a pan-cancer Kaplan-Meier survival analysis (as performed by Spurr et al. 6 ) may be confounded by the cancer-type composition of the overall dataset, and as most clinical trials usually focus on individual cancer types, we first set out to compare the survival curves of low-TMB patients with high versus low AS for each of the 10 cancer types individually. The initial cancer-type-specific analysis was performed by using the AS values provided by 6 (which calls chromosome-level CNAs using the cutoff of |log2 copy ratio| > 0.1; denoted as AS0.1). Unexpectedly, a Kaplan-Meier survival analysis of low-TMB patients identified a statistically significant improved survival following immunotherapy in a single individual cancer type, i.e., “cancer of unknown primary” (n = 70, hazard ratio HR = 2.27, p = 0.031; Fig. 1a ). Here, the HR denotes the relative risk of the AS0.1-high individuals compared to the AS0.1-low set as the reference. We further performed a cancer-type-specific multivariate Cox proportional-hazards regression of overall survival with AS0.1, TMB, and ICB drug class. None of the AS0.1 multivariate HRs were significant in any individual cancer types ( Fig. 1b ). Aiming to improve on these results, we observed that while the cutoff used to determine a CNA event in 6 was |log2 copy ratio| > 0.1, other studies, such as 7,8 , have used the cutoff of |log2 copy ratio| > 0.2 in calculating AS and/or FGA. A low cutoff of |log2 copy ratio| in calling CNA events might introduce noise (false positives), whereas a high cutoff might result in missing true events (false negatives). There are a number of parameters that may affect the optimal cutoff, e.g., tumor purity, tumor heterogeneity, and the platform used for CNA calling (e.g., whole exome sequencing, single nucleotide polymorphism arrays, and shallow whole genome sequencing). We hence reasoned that an arbitrary threshold could never be optimal for all datasets and searched for an unbiased approach for threshold calling. Aiming to identify an optimal CNA calling cutoff, we turned to the “elbow method”, which was developed to identify a cutoff point that optimally distinguishes between two qualitative, discrete states 9 . This method has been found to be effective in determining optimal parameter thresholds in a variety of data-driven optimization tasks including the determination of the number of clusters, determination of the number of principal components, and with relevance to our goal, determination of the threshold on a receiver operating characteristic (ROC) curve 10-12 . We calculated the elbow points of CNA calling cutoff |log2 copy ratio| for AS for all 10 individual cancer types (exemplified as in Fig. 2a ), which are in the range of 0.14-0.22 with 95% confidence interval (CI) in the range of 0.12-0.27 ( Supplementary Table 1 ). Therefore, the cutoff of 0.1 used in 6 is well-below the elbow points for all individual cancer types (and thus suboptimal). We thus re-evaluated the predictive power of AS by calculating AS using the elbow points as the CNA calling cutoff per cancer types (denoted as AS*). Following 6 , we determined the percentile to binarize AS* that optimally synergized with TMB to risk-stratify patients following immunotherapy by testing every tenth quantile within each cancer type, moving in increments from the 20th to 80th percentile, using a multivariate model with TMB (binned at 80th percentile) and ICB drug class. We identified the 30th percentile as the optimal binarization threshold to classify patients into AS*-high and AS*-low groups because it yielded both the lowest p-value and the highest multivariate HR; both pan-cancer metrics are better than that obtained with AS0.1 at its optimal binarization threshold (AS* HR = 1.36 vs AS0.1 HR = 1.22, AS* p-value = 2*10 -4 vs AS0.1 p-value = 5*10 -3 ; Fig. 2b ). Furthermore, the multivariate HRs of AS* in individual cancer types were, on average, greater than those obtained using AS0.1 (D mean HR = 0.21; Fig. 2c ); and the AS* HRs were now significant in one individual cancer type, i.e., melanoma (HR = 2.37, p-value = 0.001) and marginally significant in two other cancer types, i.e., non-small cell lung cancer (HR = 1.32, p-value = 0.075) and renal cell carcinoma (HR = 1.88, p-value = 0.078; Fig. 2c ). We hypothesized that CNA calling cutoff |log2 copy ratio| > 0.1 is a too low cutoff, which introduced noise in calculating patient AS, and thus dampened its predictive power of survival following immunotherapy. To test this hypothesis, we divided the low-TMB patients into four groups by their high/low AS 0.1/AS* scores and compared the Kaplan-Meier survival curves ( Fig. 2d ). We found that, among AS*-high or among AS*-low patients, there was no significant survival difference between patients that had high or low AS0.1 values. In contrast, among AS0.1-low patients, a subset of patients, i.e., the AS*-high patients, had significantly worse survival rates than AS*-low patients (HR = 1.32, p-value = 0.014); they actually had similar survival rates as the AS0.1-high / AS*-high patients (HR = 1.08, p-value = 0.41). On the other hand, among AS0.1-high patients, a subset of patients, i.e., the AS*-low patients, had much better survival rates than AS*-high patients (HR = 1/1.69 = 0.59, p-value = 0.29); they actually achieved similar survival rates as the AS0.1-low / AS*-low patients (HR = 0.86, p-value = 0.76).This result testifies that the AS0.1 indeed mis-classifies a number of patients as a result of the loose CNA calling cutoff used. We next turned to employ the elbow method to study the predictive power of FGA and compare it to that of AS. FGA combines both chromosomal and focal CNAs. If the association between CNAs and immunotherapy response is driven by the overall genomic instability, one would expect FGA to perform at least as well as AS in predicting immunotherapy response; therefore, the conclusion of 6 that AS is a better predictor than FGA in low-TMB patients is non-intuitive. We re-examined this question systematically, again employing the elbow method. The elbow method identified the elbow points of CNA calling cutoff |log2 copy ratio| for FGA among individual cancer types are in the range of 0.14-0.23 (95% CI, 0.12-0.27; Supplementary Table 1 ). Similar to the AS analysis, we identified the 20th percentile as the optimal binarization threshold to classify patients into FGA*-high and FGA*-low groups because it yielded the highest multivariate HR and significant Bonferroni-corrected p-value. As a result, the multivariate HRs of FGA* in individual cancer types were indeed very close to those obtained using AS* (D mean HR = -0.01; Supplementary Fig. 1 ), testifying that FGA performs at least as well as AS in predicting immunotherapy response. For completeness, we additionally explored the predictive power of AS and FGA with the pan-cancer cutoff of |log2 copy ratio| > 0.2 (denoted as AS0.2 and FGA0.2 respectively), which locates at the center of the 95% CI range across individual cancers (0.12-0.27; Supplementary Table 1 ) and has been previously used in the literature 7,8 . We identified the 60th percentile as the optimal binarization threshold for AS0.2 and 50th percentile as the optimal binarization threshold for FGA0.2 ( Fig. 3a ). Analogous to the FDA-approved 10 Mut/Mb of TMB for determining if patients are qualified to receive immunotherapy or not in solid tumors, we also aimed to study if we can identify such pan-cancer absolute binarization thresholds for FGA0.2 and AS0.2, in addition to the percentile binarization thresholds. To this end, we determined the AS0.2 or FGA0.2 absolute binarization threshold that combined with TMB could risk-stratify patients for immunotherapy. We tested absolute binarization thresholds from 0.05 to 0.8 (with step size of 0.01) in a multivariate model with TMB (binned at 10 Mut/Mb) and ICB drug class. We identified AS = 0.10 as the optimal threshold for AS0.2 and FGA = 0.16 as the optimal threshold for FGA0.2 for binarizing the patients into AS-high vs AS-low (or FGA-high vs FGA-low) groups, with significant Bonferroni-corrected p-values and the highest multivariate HRs ( Fig. 3b ). As evident, the absolute binarization thresholds performed very similarly to the relative percentile binarization thresholds in predicting survival following immunotherapy for both AS0.2 ( Fig. 3c ) and FGA0.2 ( Fig. 3d ). Moreover, FGA0.2 had significant univariate HRs in low-TMB patients in four individual cancer types in a Kaplan-Meier survival analysis ( Supplementary Fig. 2 ). In comparison with AS0.2, FGA0.2 yielded a significant multivariate HR in one more individual cancer type (melanoma) in the multivariate Cox model with adjustment for TMB and ICB drug class robustly with both percentile and absolute binarization thresholds ( Fig. 3d ). Overall, we conclude that FGA performs comparable to or better than AS in predicting immunotherapy response in individual cancers, suggesting that it is indeed the overall genome affected by CNAs (rather than the individual CNA length or mechanism of formation) that drives the observed CNA-immunotherapy response associations. To further test the robustness of FGA0.2 in predicting survival following immunotherapy in other datasets, we analyzed another MSK-IMPACT cohort published recently by Chowell et al. 8 (data from the supplementary table of the paper). In the Chowell et al. cohort, there are in total 15 cancer types, 8 of them are in common with the above used Samstein et al. cohort (we merged gastric and esophageal cancers in the Chowell et al. cohort into esophagogastric cancer to keep in line with the tumor type classification in the Samstein et al. cohort). We note that we could not use the Chowell et al. data to validate our AS analysis because these data do not include AS values and it is not possible to calculate the AS values based on the publicly available information. FGA0.2 had significant multivariate HRs in pan-cancer (HR = 1.26, p = 6*10 -4 ), renal cell carcinoma (HR = 2.28, p = 0.008), melanoma (HR = 1.63, p = 0.042), and non-small cell lung cancer (HR = 1.42, p = 0.002), with both percentile and absolute binarization thresholds ( Fig. 3e ). Merging two MSK-IMPACT cohorts together, FGA0.2 has significant multivariate HRs in the exact same analyses ( Fig. 3f ). In general, FGA0.2 had multivariate HRs greater than 1 for eight cancer types, i.e., renal cell carcinoma, hepatobiliary cancer, cancer of unknown primary, melanoma, colorectal cancer, non-small cell lung cancer, ovarian cancer, and bladder cancer; and it had close to 1 or less than 1 HRs for nine other cancer types, i.e., mesothelioma, head and neck cancer, esophagogastric cancer, sarcoma, glioma, pancreatic cancer, small cell lung cancer, breast cancer, and endometrial cancer ( Fig. 3f ). In summary, we have comparatively assessed the power of AS and FGA in predicting low-TMB patient survival following immunotherapy in pan-cancer and individual cancer types. Addressing our research questions, we first show that the AS measure defined in 6 (AS0.1) does not significantly predict survival benefit following immunotherapy in low-TMB patients in any single cancer type ( Fig. 1 ). Second, we show that the arbitrary cutoff of |log2 copy ratio| > 0.1 used in 6 is far below the elbow points, misclassifying many patients ( Fig. 2 ). Thirdly, FGA, re-calculated using a more appropriate pan-cancer CNA calling cutoff of |log2 copy ratio| > 0.2 identified by using elbow method in individual cancers, has considerably stronger predictive power of survival following immunotherapy ( Fig. 3 ). Finally, from a translational standpoint, the currently available data suggest that both AS and FGA can only significantly predict survival following immunotherapy in a few cancer types. Therefore, larger sample sizes are required in to evaluate, and ultimately use these measures within individual cancer types. Methods Data for the Samstein et al. 1 cohort (MSK-IMPACT) were downloaded and processed following the method described in 6 exactly. Kaplan–Meier survival analysis was performed using the R package “survfit” 13 , and log-rank P values are reported. Multivariate analysis was performed with Cox proportional hazard regression in individual cancer types using the R package “coxph” 13 , with inclusion of covariates including FGA (or AS), TMB and ICB drug class. For the elbow analysis, firstly, copy number alteration events, which were used to calculate AS and FGA, were called using |log2 copy ratio| cutoffs ranging from 0.01 to 0.5 with a step size of 0.01. Then, to calculate the elbow points of cutoff for pan-cancer and individual cancer types, mean values of AS/FGA across samples were calculated under each cutoff to generate the AS/FGA-cutoff curves. Finally, elbow point in each bootstrap replication was calculated using Python package “kneed” version 0.8.1; and 95% confidence interval of elbow point were determined from 1000-replicate bootstrapping. Declarations Data availability Data for the Samstein et al. cohort are available at https://www.cbioportal.org/study/summary?id=tmb_mskcc_2018 and the GENIE 14 v.7.1 release: https://www.synapse.org/#!Synapse:syn7222066/wiki/405659. Data for the Chowell et al. cohort are available from the supplementary table of 8 , where FGA, TMB, ICB drug class, and overall survival information are provided. Aneuploidy scores were called using ASCETS at https://github.com/beroukhim-lab/ascets and values for each sample are provided in the GitHub repository referenced below. Source data are provided with this paper. Code availability All code necessary to replicate these analyses are provided in the following GitHub repository: https://github.com/rootchang/Aneuploidy-FGA-ICB. Author contributions T.C. and E.R. conceived and designed the study. T.C. and Y.C. collected and managed the data. T.C., Y.C. and E.D.S. performed the statistical analyses. U.B.-D. and A.A.S. provided statistical advice. All authors critically revised the manuscript for important intellectual content. Competing interests E.R. is a co-founder of MedAware, Metabomed and Pangea Biomed (divested), and an unpaid member of Pangea Biomed’s scientific advisory board. U.B.-D. receives grant funding from Novocure. The other authors have no competing interests. Acknowledgments This research was supported in part by the NIH Intramural Research Program, National Cancer Institute. This work utilized the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov). The authors would like to acknowledge the American Association for Cancer Research and its financial and material support in the development of the AACR Project GENIE registry, as well as members of the consortium for their commitment to data sharing. Interpretations are the responsibility of study authors. References Samstein, R.M. et al. Tumor mutational load predicts survival after immunotherapy across multiple cancer types. Nature Genetics 51, 202-+ (2019). McGrail, D.J. et al. High tumor mutation burden fails to predict immune checkpoint blockade response across all cancer types. Annals of Oncology 32, 661–672 (2021). Ben-David, U. & Amon, A. Context is everything: aneuploidy in cancer. Nature Reviews Genetics 21, 44–62 (2020). Hieronymus, H. et al. Tumor copy number alteration burden is a pan-cancer prognostic factor associated with recurrence and death. Elife 7(2018). Sansregret, L. & Swanton, C. The Role of Aneuploidy in Cancer Evolution. Cold Spring Harbor Perspectives in Medicine 7(2017). Spurr, L.F., Weichselbaum, R.R. & Pitroda, S.P. Tumor aneuploidy predicts survival following immunotherapy across multiple cancers. Nat Genet 54, 1782–1785 (2022). Spurr, L.F. et al. Quantification of aneuploidy in targeted sequencing data using ASCETS. Bioinformatics 37, 2461–2463 (2021). Chowell, D. et al. Improved prediction of immune checkpoint blockade efficacy across multiple cancer types. Nature Biotechnology 40, 499-+ (2022). Satopaa, V., Albrecht, J., Irwin, D. & Raghavan, B. Finding a" kneedle" in a haystack: Detecting knee points in system behavior. in 2011 31st international conference on distributed computing systems workshops 166–171 (IEEE, 2011). Syakur, M.A., Khotimah, B.K., Rochman, E.M.S. & Satoto, B.D. Integration K-Means Clustering Method and Elbow Method For Identification of The Best Customer Profile Cluster. 2nd International Conference on Vocational Education and Electrical Engineering (Icvee) 336(2018). Linting, M., Meulman, J.J., Groenen, P.J.F. & van der Kooij, A.J. Nonlinear principal components analysis: Introduction and application. Psychological Methods 12, 336–358 (2007). Oh, J.H., Hong, J.Y. & Baek, J.G. Oversampling method using outlier detectable generative adversarial network. Expert Systems with Applications 133, 1–8 (2019). Therneau, T. A package for survival analysis in S. R package version 2(2015). Andre, F. et al. AACR Project GENIE: Powering Precision Medicine through an International Consortium. Cancer Discovery 7, 818–831 (2017). Additional Declarations There is a conflict of interest E.R. is a co-founder of MedAware, Metabomed and Pangea Biomed (divested), and an unpaid member of Pangea Biomed’s scientific advisory board. U.B.-D. receives grant funding from Novocure. The other authors have no competing interests. Supplementary Files Supplementaryinformation.docx Cite Share Download PDF Status: Published Journal Publication published 03 Jun, 2023 Read the published version in npj Precision Oncology → Version 1 posted Editorial decision: revise 29 Mar, 2023 Review # 3 received at journal 26 Mar, 2023 Review # 2 received at journal 24 Mar, 2023 Review # 1 received at journal 22 Mar, 2023 Reviewer # 3 agreed at journal 15 Mar, 2023 Reviewer # 2 agreed at journal 14 Mar, 2023 Reviewer # 1 agreed at journal 14 Mar, 2023 Reviewers invited by journal 13 Mar, 2023 Editor assigned by journal 07 Mar, 2023 Submission checks completed at journal 02 Mar, 2023 First submitted to journal 01 Mar, 2023 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-2644875","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Brief Communication","associatedPublications":[],"authors":[{"id":183195811,"identity":"8e29302b-f150-4b07-9c07-b348e45044b7","order_by":0,"name":"Tiangen Chang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4UlEQVRIie3RLwvCQBjH8UcOvPKA9Y5NX8PBYMPkWzG5YhPEYBgIGq2C4stYFg5msdgMCw5hWUFlSdz5D1ZuRsH7hodf+aQHwGT6xch7MHUGgKJMYJFsviGfpUhlDFBKWpSkyXkQQ20+SpPLMrY9oNsDagiSqufYmxRYHHlOPUyxGWDP0RNwLT6WIFg7H6FEscIOn2kJvVr8poifj8VXBF1+ChTpPkZOaMSOOiKxb0Ekke26zyEkEqEjdLoOeTaUjdrMf4yWWE+SfVtDVARf/yGvW/oaqGTFQfelxGQymf6qO7DaRiAyufnuAAAAAElFTkSuQmCC","orcid":"","institution":"National Cancer Institute","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Tiangen","middleName":"","lastName":"Chang","suffix":""},{"id":183195812,"identity":"d3a22a99-6d34-444e-81db-4b9f64a23ccc","order_by":1,"name":"Yingying Cao","email":"","orcid":"","institution":"National Cancer Institute","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Yingying","middleName":"","lastName":"Cao","suffix":""},{"id":183195813,"identity":"fd41ceca-c171-44cb-afbc-4d479a32043d","order_by":2,"name":"Eldad Shulman","email":"","orcid":"https://orcid.org/0000-0002-1859-8819","institution":"National Cancer Institute","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Eldad","middleName":"","lastName":"Shulman","suffix":""},{"id":183195814,"identity":"2268907f-cc5c-414d-aabc-61258275605d","order_by":3,"name":"Alejandro Schäffer","email":"","orcid":"https://orcid.org/0000-0002-2147-8033","institution":"NIH","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Alejandro","middleName":"","lastName":"Schäffer","suffix":""},{"id":183195815,"identity":"51ed3621-b8f8-45a3-922d-6fc015845c24","order_by":4,"name":"Uri Ben-David","email":"","orcid":"https://orcid.org/0000-0001-7098-2378","institution":"Tel Aviv University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Uri","middleName":"","lastName":"Ben-David","suffix":""},{"id":183195816,"identity":"2ea394d6-4a81-4fb4-bcaa-3de5a46a4c45","order_by":5,"name":"Eytan Ruppin","email":"","orcid":"","institution":"National Cancer Institute","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Eytan","middleName":"","lastName":"Ruppin","suffix":""}],"badges":[],"createdAt":"2023-03-01 22:11:05","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-2644875/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-2644875/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41698-023-00408-6","type":"published","date":"2023-06-03T04:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":34361311,"identity":"231cb475-8cd1-4cd6-b184-c48345c90a97","added_by":"auto","created_at":"2023-03-16 14:40:31","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":264766,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTumor aneuploidy score (AS0.1) is not significantly predictive of survival following immunotherapy in \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003eindividual \u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003ecancer types.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(a) \u003c/strong\u003eKaplan-Meier survival curves following immunotherapy are compared for low-TMB patients (\u0026lt;80th percentile) with high versus low AS0.1 (binned into AS-low and AS-high at the 50th percentile) across 10 individual cancer types. Two-sided log-rank p values are indicated, with univariate Cox regression HR with 95% confidence interval. \u003cstrong\u003e(b)\u003c/strong\u003e Multivariate survival analysis using Cox proportional hazards regression of overall survival with AS0.1, TMB, and ICB drug class (AS0.1 binned at the 50th percentile and TMB binned at the 80th percentile). HRs with 95% confidence intervals and Wald p values are displayed. AS0.1 is calculated using the same cutoff as in \u003csup\u003e6\u003c/sup\u003e (|log2 copy ratio| \u0026gt; 0.1). The HR denotes the relative risk of the AS0.1-high individuals compared to the AS0.1-low set as the reference.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-2644875/v1/63b943d8f3e77466c649d014.png"},{"id":34361309,"identity":"2231aff6-570b-420d-b61a-9b212741e809","added_by":"auto","created_at":"2023-03-16 14:40:31","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":484254,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOptimizing the CNA calling cutoff to improve AS prediction of survival following immunotherapy.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(a) \u003c/strong\u003eThe elbow method for determining the cutoff of |log2 copy ratio| was used in calling AS for individual cancer types (exemplified by esophagogastric cancer here). The AS for each patient with different calling cutoffs are shown in black curves. The mean value of all patients is shown in the red curve. The mean elbow point is shown with 95% confidence intervals, which are calculated using 1,000-replicate bootstrapping. \u003cstrong\u003e(b) \u003c/strong\u003eCandidate binarization percentiles to split patients into high versus low AS at each percentile. 1,660 multivariate Cox models as part of the leave-one-out cross validation analysis are constructed with AS (binned at the candidate binarization percentile), TMB (binned at the 80th percentile), and ICB drug class. The Wald p values and multivariate HRs with 95% confidence interval for AS0.1 and AS* are displayed respectively. Black arrows indicate Wald p-values and multivariable HRs at the optimal percentiles for AS0.1 (50\u003csup\u003eth\u003c/sup\u003e percentile) and AS* (30\u003csup\u003eth\u003c/sup\u003e percentile) respectively. Dashed line denotes the Bonferroni-corrected P = 0.05.\u003cstrong\u003e (c)\u003c/strong\u003e Comparison of HRs using AS0.1 or AS* in a multivariate Cox model with TMB (binned at the 80\u003csup\u003eth\u003c/sup\u003e percentile) and ICB drug class. Difference of mean HRs of AS0.1 and AS* and paired Wilcoxon test p value are displayed. Wald p values for HRs of AS* in individual cancer types are displayed at the right side of the plot.\u003cstrong\u003e (d)\u003c/strong\u003e Kaplan-Meier analysis of AS0.1 binned at the 50\u003csup\u003eth\u003c/sup\u003e percentile and AS* binned at the 30\u003csup\u003eth\u003c/sup\u003e percentile for pan-cancer low-TMB patients. HR and p values of pairwise comparisons between different groups are shown. The global p value is displayed in the bottom left of the plot. H, high; L, low.\u003c/p\u003e\n\u003cp\u003eThe data are from the Samstein et al. cohort \u003csup\u003e1\u003c/sup\u003e.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-2644875/v1/759c813d51461c5217d09baf.png"},{"id":34362349,"identity":"24524536-58ff-4823-bb66-65fe07b1129f","added_by":"auto","created_at":"2023-03-16 14:48:31","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":531451,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe fraction of copy number alterations (FGA0.2) robustly predicts survival following immunotherapy in certain cancers.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(a) \u003c/strong\u003eCandidate thresholds for splitting patients into high versus low AS or FGA at each \u003cem\u003epercentile\u003c/em\u003e threshold. 1,660 multivariate Cox models as part of the leave-one-out cross validation analysis are constructed with AS or FGA (binned at the candidate threshold), TMB (binned at the 80th percentile), and ICB drug class. The Wald p values and multivariate HRs with 95% confidence intervals for AS0.2 and FGA0.2 are displayed respectively. \u003cstrong\u003e(b) \u003c/strong\u003eCandidate thresholds for splitting patients into high versus low AS or FGA at each \u003cem\u003eabsolute\u003c/em\u003e threshold. Multivariate Cox model is constructed with AS or FGA (binned at the candidate absolute threshold), TMB (binned at 10 Mut/Mb), and ICB drug class. The Wald p values and multivariate HRs with 95% confidence intervals for AS0.2 and FGA0.2 are displayed respectively. \u003cstrong\u003e(c)\u003c/strong\u003e Multivariate survival analysis using Cox proportional hazards regression of overall survival with AS0.2, TMB, and ICB drug class in the Samstein et al. cohort (left: AS0.2 is binned at the 60th percentile and TMB is binned at the 80th percentile; right: AS0.2 is binned at 0.16 and TMB is binned at 10 Mut/Mb). HRs with 95% confidence intervals and Wald p values are displayed. \u003cstrong\u003e(d)\u003c/strong\u003e The same as (c), except that the AS0.2 is replaced by FGA0.2, FGA0.2 is binned at the 50th percentile in the left panel, and FGA0.2 is binned at the 0.16 in the right panel. \u003cstrong\u003e(e)\u003c/strong\u003e The same as (d), except that the Samstein et al. cohort is replaced by the Chowell et al. cohort. \u003cstrong\u003e(f)\u003c/strong\u003e The same as (d), except that the Samstein et al. cohort is replaced by the merged Samstein et al. and Chowell et al. cohorts.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-2644875/v1/cd0f694cb3f01cde6c39761f.png"},{"id":37970297,"identity":"f31e7f90-ce5c-4b8b-9110-c447466f76fb","added_by":"auto","created_at":"2023-06-04 07:08:07","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1365980,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2644875/v1/1a6e1197-b4ea-43c1-85e1-77201cbfb690.pdf"},{"id":34361310,"identity":"68d4a929-4434-4781-a929-911d317ae126","added_by":"auto","created_at":"2023-03-16 14:40:31","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":268909,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementaryinformation.docx","url":"https://assets-eu.researchsquare.com/files/rs-2644875/v1/d9296267feb74371655d9924.docx"}],"financialInterests":"There is a conflict of interest\nE.R. is a co-founder of MedAware, Metabomed and Pangea Biomed (divested), and an unpaid member of Pangea Biomed’s scientific advisory board. U.B.-D. receives grant funding from Novocure. The other authors have no competing interests.","formattedTitle":"Optimizing cancer immunotherapy response prediction by tumor aneuploidy score and fraction of copy number alterations","fulltext":[{"header":"Introduction","content":"\u003cp\u003eVarious studies have shown that high tumor mutation burden (TMB) may predict immunotherapy response, at least in some cancer types \u003csup\u003e1,2\u003c/sup\u003e. However, identifying patients with low TMB that are likely to respond to cancer immunotherapy is a challenging unmet clinical need. One promising approach to identify low-TMB responders of immunotherapy has been to study the predictive ability of other measures of genomic alterations in cancer in these patients. Two natural \u0026nbsp; candidates are scores based on copy number alterations (CNAs): (a) tumor aneuploidy, which measures chromosome-level CNAs, and (b) global genomic CNAs, which quantifies the extents of both chromosomal and focal copy-number events \u003csup\u003e3\u003c/sup\u003e. Both tumor aneuploidy and genomic CNAs have been shown to play a role in cancer progression and to be predictive for cancer prognosis \u003csup\u003e3-5\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eRecently, Spurr et al. reported that the tumor \u003cem\u003eaneuploidy score\u003c/em\u003e (AS), defined as the fraction of chromosome arms with arm-level copy number alterations in a sample, is significantly predictive of survival following immunotherapy in low-TMB patients in a pan-cancer analysis \u003csup\u003e6\u003c/sup\u003e. In addition, they reported that AS had stronger predictive power than another metric conceptually related to the AS, the \u003cem\u003efraction of genome encompassed by copy number alterations\u003c/em\u003e (FGA) which quantifies the extent of both chromosomal and focal copy-number events \u003csup\u003e6\u003c/sup\u003e. Intrigued by these potentially clinically impactful findings, we set out to explore several related fundamental questions: (1) Is AS predictive of survival of low-TMB patients following immunotherapy in \u003cem\u003eindividual\u003c/em\u003e cancer types? (2) Does the choice of cutoff during CNA calling influence the predictive power? (3) Under optimal calling cutoffs, is AS or FGA a better predictor?\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe first re-analyzed the same data used by Spurr et al. \u003csup\u003e6\u003c/sup\u003e, i.e., the Samstein et al. cohort \u003csup\u003e1\u003c/sup\u003e from MSK-IMPACT (data acquired from the AACR Project GENIE). This study has analyzed a published cohort of 1,660 advanced cancer patients from 10 different cancer types treated with immune checkpoint blockade (ICB). As a pan-cancer Kaplan-Meier survival analysis (as performed by Spurr et al. \u003csup\u003e6\u003c/sup\u003e) may be confounded by the cancer-type composition of the overall dataset, and as most clinical trials usually focus on individual cancer types, we first set out to compare the survival curves of low-TMB patients with high versus low AS for each of the 10 cancer types individually.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe initial cancer-type-specific analysis was performed by using the AS values provided by \u003csup\u003e6\u003c/sup\u003e (which calls chromosome-level CNAs using the cutoff of |log2 copy ratio| \u0026gt; 0.1; denoted as AS0.1). Unexpectedly, a Kaplan-Meier survival analysis of low-TMB patients identified a statistically significant improved survival following immunotherapy in a single individual cancer type, i.e., \u0026ldquo;cancer of unknown primary\u0026rdquo; (n = 70, hazard ratio HR = 2.27, p = 0.031; \u003cstrong\u003eFig. 1a\u003c/strong\u003e). Here, the HR denotes the relative risk of the AS0.1-high individuals compared to the AS0.1-low set as the reference. We further performed a cancer-type-specific multivariate Cox proportional-hazards regression of overall survival with AS0.1, TMB, and ICB drug class. None of the AS0.1 multivariate HRs were significant in any individual cancer types (\u003cstrong\u003eFig. 1b\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAiming to improve on these results, we observed that while the cutoff used to determine a CNA event in \u003csup\u003e6\u003c/sup\u003e was |log2 copy ratio| \u0026gt; 0.1, other studies, such as \u003csup\u003e7,8\u003c/sup\u003e, have used the cutoff of |log2 copy ratio| \u0026gt; 0.2 in calculating AS and/or FGA. A low cutoff of |log2 copy ratio| in calling CNA events might introduce noise (false positives), whereas a high cutoff might result in missing true events (false negatives). There are a number of parameters that may affect the optimal cutoff, e.g., tumor purity, tumor heterogeneity, and the platform used for CNA calling (e.g., whole exome sequencing, single nucleotide polymorphism arrays, and shallow whole genome sequencing). We hence reasoned that an arbitrary threshold could never be optimal for all datasets and searched for an unbiased approach for threshold calling. Aiming to identify an optimal CNA calling cutoff, we turned to the \u0026ldquo;elbow method\u0026rdquo;, which was developed to identify a cutoff point that optimally distinguishes between two qualitative, discrete states \u003csup\u003e9\u003c/sup\u003e. This method has been found to be effective in determining optimal parameter thresholds in a variety of data-driven optimization tasks including the determination of the number of clusters, determination of the number of principal components, and with relevance to our goal, determination of the threshold on a receiver operating characteristic (ROC) curve \u003csup\u003e10-12\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe calculated the elbow points of CNA calling cutoff |log2 copy ratio| for AS for all 10 individual cancer types (exemplified as in \u003cstrong\u003eFig. 2a\u003c/strong\u003e), which are in the range of 0.14-0.22 with 95% confidence interval (CI) in the range of 0.12-0.27 (\u003cstrong\u003eSupplementary Table 1\u003c/strong\u003e). Therefore, the cutoff of 0.1 used in \u003csup\u003e6\u003c/sup\u003e is well-below the elbow points for all individual cancer types (and thus suboptimal). We thus re-evaluated the predictive power of AS by calculating AS using the elbow points as the CNA calling cutoff per cancer types (denoted as AS*). Following \u003csup\u003e6\u003c/sup\u003e, we determined the percentile to binarize AS* that optimally synergized with TMB to risk-stratify patients following immunotherapy by testing every tenth quantile within each cancer type, moving in increments from the 20th to 80th percentile, using a multivariate model with TMB (binned at 80th percentile) and ICB drug class. We identified the 30th percentile as the optimal binarization threshold to classify patients into AS*-high and AS*-low groups because it yielded both the lowest p-value and the highest multivariate HR; both pan-cancer metrics are better than that obtained with AS0.1 at its optimal binarization threshold (AS* HR = 1.36 vs AS0.1 HR = 1.22, AS* p-value = 2*10\u003csup\u003e-4\u003c/sup\u003e vs AS0.1 p-value = 5*10\u003csup\u003e-3\u003c/sup\u003e; \u003cstrong\u003eFig. 2b\u003c/strong\u003e). Furthermore, the multivariate HRs of AS* in individual cancer types were, on average, greater than those obtained using AS0.1 (D mean HR = 0.21; \u003cstrong\u003eFig. 2c\u003c/strong\u003e); and the AS* HRs were now significant in one individual cancer type, i.e., melanoma (HR = 2.37, p-value = 0.001) and marginally significant in two other cancer types, i.e., non-small cell lung cancer (HR = 1.32, p-value = 0.075) and renal cell carcinoma (HR = 1.88, p-value = 0.078; \u003cstrong\u003eFig. 2c\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eWe hypothesized that CNA calling cutoff |log2 copy ratio| \u0026gt; 0.1 is a too low cutoff, which introduced noise in calculating patient AS, and thus dampened its predictive power of survival following immunotherapy. To test this hypothesis, we divided the low-TMB patients into four groups by their high/low AS 0.1/AS* scores and compared the Kaplan-Meier survival curves (\u003cstrong\u003eFig. 2d\u003c/strong\u003e). We found that, among AS*-high or among AS*-low patients, there was no significant survival difference between patients that had high or low AS0.1 values. In contrast, among AS0.1-low patients, a subset of patients, i.e., the AS*-high patients, had significantly worse survival rates than AS*-low patients (HR = 1.32, p-value = 0.014); they actually had similar survival rates as the AS0.1-high / AS*-high patients (HR = 1.08, p-value = 0.41). On the other hand, among AS0.1-high patients, a subset of patients, i.e., the AS*-low patients, had much better survival rates than AS*-high patients (HR = 1/1.69 = 0.59, p-value = 0.29); they actually achieved similar survival rates as the AS0.1-low / AS*-low patients (HR = 0.86, p-value = 0.76).This result testifies that the AS0.1 indeed mis-classifies a number of patients as a result of the loose CNA calling cutoff used.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe next turned to employ the elbow method to study the predictive power of FGA and compare it to that of AS. FGA combines both chromosomal and focal CNAs. If the association between CNAs and immunotherapy response is driven by the overall genomic instability, one would expect FGA to perform at least as well as AS in predicting immunotherapy response; therefore, the conclusion of \u003csup\u003e6\u003c/sup\u003e that AS is a better predictor than FGA in low-TMB patients is non-intuitive. We re-examined this question systematically, again employing the elbow method. The elbow method identified the elbow points of CNA calling cutoff |log2 copy ratio| for FGA among individual cancer types are in the range of 0.14-0.23 (95% CI, 0.12-0.27; \u003cstrong\u003eSupplementary Table 1\u003c/strong\u003e). Similar to the AS analysis, we identified the 20th percentile as the optimal binarization threshold to classify patients into FGA*-high and FGA*-low groups because it yielded the highest multivariate HR and significant Bonferroni-corrected p-value. As a result, the multivariate HRs of FGA* in individual cancer types were indeed very close to those obtained using AS* (D mean HR = -0.01; \u003cstrong\u003eSupplementary Fig. 1\u003c/strong\u003e), testifying that FGA performs at least as well as AS in predicting immunotherapy response.\u003c/p\u003e\n\u003cp\u003eFor completeness, we additionally explored the predictive power of AS and FGA with the pan-cancer cutoff of |log2 copy ratio| \u0026gt; 0.2 (denoted as AS0.2 and FGA0.2 respectively), which locates at the center of the 95% CI range across individual cancers (0.12-0.27; \u003cstrong\u003eSupplementary Table 1\u003c/strong\u003e) and has been previously used in the literature \u003csup\u003e7,8\u003c/sup\u003e. We identified the 60th percentile as the optimal binarization threshold for AS0.2 and 50th percentile as the optimal binarization threshold for FGA0.2 (\u003cstrong\u003eFig. 3a\u003c/strong\u003e). Analogous to the FDA-approved 10 Mut/Mb of TMB for determining if patients are qualified to receive immunotherapy or not in solid tumors, we also aimed to study if we can identify such pan-cancer absolute binarization thresholds for FGA0.2 and AS0.2, in addition to the percentile binarization thresholds. To this end, we determined the AS0.2 or FGA0.2 absolute binarization threshold that combined with TMB could risk-stratify patients for immunotherapy. We tested absolute binarization thresholds from 0.05 to 0.8 (with step size of 0.01) in a multivariate model with TMB (binned at 10 Mut/Mb) and ICB drug class. We identified AS = 0.10 as the optimal threshold for AS0.2 and FGA = 0.16 as the optimal threshold for FGA0.2 for binarizing the patients into AS-high vs AS-low (or FGA-high vs FGA-low) groups, with significant Bonferroni-corrected p-values and the highest multivariate HRs (\u003cstrong\u003eFig. 3b\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAs evident, the \u003cem\u003eabsolute\u003c/em\u003e binarization thresholds performed very similarly to the relative \u003cem\u003epercentile\u003c/em\u003e binarization thresholds in predicting survival following immunotherapy for both AS0.2 (\u003cstrong\u003eFig. 3c\u003c/strong\u003e) and FGA0.2 (\u003cstrong\u003eFig. 3d\u003c/strong\u003e). Moreover, FGA0.2 had significant univariate HRs in low-TMB patients in four individual cancer types in a Kaplan-Meier survival analysis (\u003cstrong\u003eSupplementary Fig. 2\u003c/strong\u003e). In comparison with AS0.2, FGA0.2 yielded a significant multivariate HR in one more individual cancer type (melanoma) in the multivariate Cox model with adjustment for TMB and ICB drug class robustly with both percentile and absolute binarization thresholds (\u003cstrong\u003eFig. 3d\u003c/strong\u003e). Overall, we conclude that FGA performs comparable to or better than AS in predicting immunotherapy response in individual cancers, suggesting that it is indeed the overall genome affected by CNAs (rather than the individual CNA length or mechanism of formation) that drives the observed CNA-immunotherapy response associations.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo further test the robustness of FGA0.2 in predicting survival following immunotherapy in other datasets, we analyzed another MSK-IMPACT cohort published recently by Chowell et al. \u003csup\u003e8\u003c/sup\u003e (data from the supplementary table of the paper). In the Chowell et al. cohort, there are in total 15 cancer types, 8 of them are in common with the above used Samstein et al. cohort (we merged gastric and esophageal cancers in the Chowell et al. cohort into esophagogastric cancer to keep in line with the tumor type classification in the Samstein et al. cohort). We note that we could not use the Chowell et al. data to validate our AS analysis because these data do not include AS values and it is not possible to calculate the AS values based on the publicly available information. FGA0.2 had significant multivariate HRs in pan-cancer (HR = 1.26, p = 6*10\u003csup\u003e-4\u003c/sup\u003e), renal cell carcinoma (HR = 2.28, p = 0.008), melanoma (HR = 1.63, p = 0.042), and non-small cell lung cancer (HR = 1.42, p = 0.002), with both percentile and absolute binarization thresholds (\u003cstrong\u003eFig. 3e\u003c/strong\u003e). Merging two MSK-IMPACT cohorts together, FGA0.2 has significant multivariate HRs in the exact same analyses (\u003cstrong\u003eFig. 3f\u003c/strong\u003e). In general, FGA0.2 had multivariate HRs greater than 1 for eight cancer types, i.e., renal cell carcinoma, hepatobiliary cancer, cancer of unknown primary, melanoma, colorectal cancer, non-small cell lung cancer, ovarian cancer, and bladder cancer; and it had close to 1 or less than 1 HRs for nine other cancer types, i.e., mesothelioma, head and neck cancer, esophagogastric cancer, sarcoma, glioma, pancreatic cancer, small cell lung cancer, breast cancer, and endometrial cancer (\u003cstrong\u003eFig. 3f\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn summary, we have comparatively assessed the power of AS and FGA in predicting low-TMB patient survival following immunotherapy in pan-cancer and individual cancer types. Addressing our research questions, we first show that the AS measure defined in \u003csup\u003e6\u003c/sup\u003e (AS0.1) does not significantly predict survival benefit following immunotherapy in low-TMB patients in any single cancer type (\u003cstrong\u003eFig. 1\u003c/strong\u003e). Second, we show that the arbitrary cutoff of |log2 copy ratio| \u0026gt; 0.1 used in \u003csup\u003e6\u003c/sup\u003e is far below the elbow points, misclassifying many patients (\u003cstrong\u003eFig. 2\u003c/strong\u003e). Thirdly, FGA, re-calculated using a more appropriate pan-cancer CNA calling cutoff of |log2 copy ratio| \u0026gt; 0.2 identified by using elbow method in individual cancers, has considerably stronger predictive power of survival following immunotherapy (\u003cstrong\u003eFig. 3\u003c/strong\u003e). Finally, from a translational standpoint, the currently available data suggest that both AS and FGA can only significantly predict survival following immunotherapy in a few cancer types. Therefore, larger sample sizes are required in to evaluate, and ultimately use these measures within individual cancer types.\u0026nbsp;\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eData for the Samstein et al. \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e cohort (MSK-IMPACT) were downloaded and processed following the method described in \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e exactly.\u003c/p\u003e \u003cp\u003eKaplan\u0026ndash;Meier survival analysis was performed using the R package \u0026ldquo;survfit\u0026rdquo; \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e, and log-rank P values are reported. Multivariate analysis was performed with Cox proportional hazard regression in individual cancer types using the R package \u0026ldquo;coxph\u0026rdquo; \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e, with inclusion of covariates including FGA (or AS), TMB and ICB drug class.\u003c/p\u003e \u003cp\u003eFor the elbow analysis, firstly, copy number alteration events, which were used to calculate AS and FGA, were called using |log2 copy ratio| cutoffs ranging from 0.01 to 0.5 with a step size of 0.01. Then, to calculate the elbow points of cutoff for pan-cancer and individual cancer types, mean values of AS/FGA across samples were calculated under each cutoff to generate the AS/FGA-cutoff curves. Finally, elbow point in each bootstrap replication was calculated using Python package \u0026ldquo;kneed\u0026rdquo; version 0.8.1; and 95% confidence interval of elbow point were determined from 1000-replicate bootstrapping.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData for the Samstein et al. cohort are available at https://www.cbioportal.org/study/summary?id=tmb_mskcc_2018 and the GENIE \u003csup\u003e14\u003c/sup\u003e v.7.1 release: https://www.synapse.org/#!Synapse:syn7222066/wiki/405659. Data for the Chowell et al. cohort are available from the supplementary table of \u003csup\u003e8\u003c/sup\u003e, where FGA, TMB, ICB drug class, and overall survival information are provided. Aneuploidy scores were called using ASCETS at https://github.com/beroukhim-lab/ascets and values for each sample are provided in the GitHub repository referenced below. Source data are provided with this paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll code necessary to replicate these analyses are provided in the following GitHub repository: https://github.com/rootchang/Aneuploidy-FGA-ICB.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eT.C. and E.R. conceived and designed the study. T.C. and Y.C. collected and managed the data. T.C., Y.C. and E.D.S. performed the statistical analyses. U.B.-D. and A.A.S. provided statistical advice. All authors critically revised the manuscript for important intellectual content.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eE.R. is a co-founder of MedAware, Metabomed and Pangea Biomed (divested), and an unpaid member of Pangea Biomed\u0026rsquo;s scientific advisory board. U.B.-D. receives grant funding from Novocure. The other authors have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported in part by the NIH Intramural Research Program, National Cancer Institute. This work utilized the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov). The authors would like to acknowledge the American Association for Cancer Research and its financial and material support in the development of the AACR Project GENIE registry, as well as members of the consortium for their commitment to data sharing. Interpretations are the responsibility of study authors.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSamstein, R.M. \u003cem\u003eet al.\u003c/em\u003e Tumor mutational load predicts survival after immunotherapy across multiple cancer types. Nature Genetics 51, 202-+ (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcGrail, D.J. \u003cem\u003eet al.\u003c/em\u003e High tumor mutation burden fails to predict immune checkpoint blockade response across all cancer types. Annals of Oncology 32, 661\u0026ndash;672 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBen-David, U. \u0026amp; Amon, A. Context is everything: aneuploidy in cancer. Nature Reviews Genetics 21, 44\u0026ndash;62 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHieronymus, H. \u003cem\u003eet al.\u003c/em\u003e Tumor copy number alteration burden is a pan-cancer prognostic factor associated with recurrence and death. Elife 7(2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSansregret, L. \u0026amp; Swanton, C. The Role of Aneuploidy in Cancer Evolution. Cold Spring Harbor Perspectives in Medicine 7(2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSpurr, L.F., Weichselbaum, R.R. \u0026amp; Pitroda, S.P. Tumor aneuploidy predicts survival following immunotherapy across multiple cancers. Nat Genet 54, 1782\u0026ndash;1785 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSpurr, L.F. \u003cem\u003eet al.\u003c/em\u003e Quantification of aneuploidy in targeted sequencing data using ASCETS. Bioinformatics 37, 2461\u0026ndash;2463 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChowell, D. \u003cem\u003eet al.\u003c/em\u003e Improved prediction of immune checkpoint blockade efficacy across multiple cancer types. Nature Biotechnology 40, 499-+ (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSatopaa, V., Albrecht, J., Irwin, D. \u0026amp; Raghavan, B. Finding a\" kneedle\" in a haystack: Detecting knee points in system behavior. in 2011 \u003cem\u003e31st international conference on distributed computing systems workshops\u003c/em\u003e 166\u0026ndash;171 (IEEE, 2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSyakur, M.A., Khotimah, B.K., Rochman, E.M.S. \u0026amp; Satoto, B.D. Integration K-Means Clustering Method and Elbow Method For Identification of The Best Customer Profile Cluster. \u003cem\u003e2nd International Conference on Vocational Education and Electrical Engineering (Icvee)\u003c/em\u003e 336(2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLinting, M., Meulman, J.J., Groenen, P.J.F. \u0026amp; van der Kooij, A.J. Nonlinear principal components analysis: Introduction and application. \u003cem\u003ePsychological Methods\u003c/em\u003e 12, 336\u0026ndash;358 (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOh, J.H., Hong, J.Y. \u0026amp; Baek, J.G. Oversampling method using outlier detectable generative adversarial network. Expert Systems with Applications 133, 1\u0026ndash;8 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTherneau, T. A package for survival analysis in S. R package version 2(2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAndre, F. \u003cem\u003eet al.\u003c/em\u003e AACR Project GENIE: Powering Precision Medicine through an International Consortium. Cancer Discovery 7, 818\u0026ndash;831 (2017).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"npj-precision-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjprecisiononcology","sideBox":"Learn more about [npj Precision Oncology](http://www.nature.com/npjprecisiononcology/)","snPcode":"41698","submissionUrl":"https://submission.springernature.com/new-submission/41698/3","title":"npj Precision Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Cancer, Fraction of copy number alterations, Immunotherapy response, Low-TMB, Tumor aneuploidy","lastPublishedDoi":"10.21203/rs.3.rs-2644875/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-2644875/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIdentifying patients with low tumor mutation burden (TMB) that are likely to respond to cancer immunotherapy is an important, yet highly challenging clinical need. Using 3,139 patients across 17 different cancer types, we comprehensively studied the ability of two common copy number alteration (CNA) scores \u0026ndash; the tumor aneuploidy score (AS) and the fraction of genome encompassed by copy number alterations (FGA) \u0026ndash; to predict survival following immunotherapy in both pan-cancer and individual cancer types. We propose an elbow-point based method to optimize the cutoff used for calling CNAs. The optimized AS and FGA scores show significantly improved predictive performance compared to the arbitrary cutoffs reported in the literature. However, our data suggests that the use of AS and FGA for predicting immunotherapy response is currently limited to only a few cancer types. Therefore, larger sample sizes are needed to evaluate the clinical utility of these measures for patient stratification in other cancer types.\u003c/p\u003e","manuscriptTitle":"Optimizing cancer immunotherapy response prediction by tumor aneuploidy score and fraction of copy number alterations","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-03-16 14:40:26","doi":"10.21203/rs.3.rs-2644875/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"revise","date":"2023-03-29T20:08:57+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"This content is not available.","date":"2023-03-27T03:47:37+00:00","index":3,"fulltext":"This content is not available."},{"type":"editorInvitedReview","content":"This content is not available.","date":"2023-03-25T02:56:31+00:00","index":2,"fulltext":"This content is not available."},{"type":"editorInvitedReview","content":"This content is not available.","date":"2023-03-22T23:13:08+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2023-03-15T06:50:59+00:00","index":3,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2023-03-14T15:05:09+00:00","index":2,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2023-03-14T07:29:18+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewersInvited","content":"","date":"2023-03-13T20:32:01+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2023-03-08T01:31:57+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2023-03-02T09:50:37+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Precision Oncology","date":"2023-03-01T22:08:16+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-precision-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjprecisiononcology","sideBox":"Learn more about [npj Precision Oncology](http://www.nature.com/npjprecisiononcology/)","snPcode":"41698","submissionUrl":"https://submission.springernature.com/new-submission/41698/3","title":"npj Precision Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bc72a512-2781-494d-905d-24e3095951ce","owner":[],"postedDate":"March 16th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":19890351,"name":"Biological sciences/Cancer/Tumour biomarkers"},{"id":19890352,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":19890353,"name":"Biological sciences/Cancer/Cancer genomics"}],"tags":[],"updatedAt":"2023-06-04T07:08:02+00:00","versionOfRecord":{"articleIdentity":"rs-2644875","link":"https://doi.org/10.1038/s41698-023-00408-6","journal":{"identity":"npj-precision-oncology","isVorOnly":false,"title":"npj Precision Oncology"},"publishedOn":"2023-06-03 04:00:00","publishedOnDateReadable":"June 3rd, 2023"},"versionCreatedAt":"2023-03-16 14:40:26","video":"","vorDoi":"10.1038/s41698-023-00408-6","vorDoiUrl":"https://doi.org/10.1038/s41698-023-00408-6","workflowStages":[]},"version":"v1","identity":"rs-2644875","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-2644875","identity":"rs-2644875","version":["v1"]},"buildId":"7rjqhiLT3MXkJMwkYKINL","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}