CoChat
Sign Up
Cited by 1,039Cancer Immunotherapy and BiomarkersCutaneous Melanoma Detection and ManagementImmunotherapy and Immune Responses

The DOI matches its Crossref record; 81 of the 81 references it cites were found in a registry or an index.

Integrative molecular and clinical modeling of clinical outcomes to PD1 blockade in patients with metastatic melanoma

Open access
Liu, David, Schilling, Bastian, Liu, Derek, Sucker, Antje, Livingstone, Elisabeth, Jerby-Arnon, Livnat, Zimmer, Lisa, Gutzmer, Ralf, Satzger, Imke, Loquai, Carmen, Grabbe, Stephan, Vokes, Natalie, Margolis, Claire A., Conway, Jake, He, Meng Xiao, Elmarakeby, Haitham, Dietlein, Felix, Miao, Diana, Tracy, Adam, Gogas, Helen, Goldinger, Simone M., Utikal, Jochen, Blank, Christian U., Rauschenberg, Ricarda, von Bubnoff, Dagmar, Krackhardt, Angela, Weide, Benjamin, Haferkamp, Sebastian, Kiecker, Felix, Izar, Ben, Garraway, Levi, Regev, Aviv, Flaherty, Keith, Paschen, Annette, Van Allen, Eliezer M., Schadendorf, Dirk·2019·Nature Medicine
DOI 10.1038/s41591-019-0654-5PMC PMC6898788
Open this paper in CoChat — free accountOpen PDF

Abstract

Immune-checkpoint blockade (ICB) has demonstrated efficacy in many tumor types, but predictors of responsiveness to anti-PD1 ICB are incompletely characterized. In this study, we analyzed a clinically annotated cohort of patients with melanoma (n = 144) treated with anti-PD1 ICB, with whole-exome and whole-transcriptome sequencing of pre-treatment tumors. We found that tumor mutational burden as a predictor of response was confounded by melanoma subtype, whereas multiple novel genomic and transcriptomic features predicted selective response, including features associated with MHC-I and MHC-II antigen presentation. Furthermore, previous anti-CTLA4 ICB exposure was associated with different predictors of response compared to tumors that were naive to ICB, suggesting selective immune effects of previous exposure to anti-CTLA4 ICB. Finally, we developed parsimonious models integrating clinical, genomic and transcriptomic features to predict intrinsic resistance to anti-PD1 ICB in individual tumors, with validation in smaller independent cohorts limited by the availability of comprehensive data. Broadly, we present a framework to discover predictive features and build models of ICB therapeutic response.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41591-019-0654-5

Cites

81 of 81 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Cohort genomic and clinical characteristics and association of TMB with response. a, CoMut plot showing association between clinical and genomic characteristics. Each column represents a tumor. Tumors are ordered by best RECIST criteria response (CR, PR, PD, SD or MR), and within each response subgroup by decreasing nonsynonymous (Nonsyn) mutational load (top row). Nonsynonymous mutational burden is further subdivided into clonal (purple) and subclonal (light purple) mutational load. Mutational signatures (sig) refer to the inferred relative contribution of UV-induced mutations, alkylating DNA damage process and other mutational signatures (aging+). The primary type of melanoma (skin, occult, acral or mucosal) is indicated. Tumor purity is the inferred proportion of the tumor sample that is from cancer cells compared to other cell types (Methods). The dominant mutational signature (that is, the mutational signature associated with the highest proportion of mutations) is indicated. Mutations in BRAF, NRAS and NF1 are shown for each tumor. b, Mutational load (mut load) in progressors (n = 65 patients), responders (n = 55 patients) and patients with SD and MR (n = 24 patients). Nonsynonymous mutational load is higher in responders (CR and PR) than in progressors (two-sided MWW, P = 0.026), but is not significantly different between responders and patients having SD or MR as the best RECIST response (two-sided MWW, P = 0.14). c, Mutational load by melanoma type. Different melanoma types have different mutational loads (Kruskal–Wallis, P = 2.4 × 10−5): mutational load is higher in cutaneous and occult melanomas (n = 124 patients) than in acral and mucosal melanomas (n = 20 patients; median 297.5 versus 58; two-sided MWW, P = 1.1 × 10−6). d, Response to anti-PD1 ICB by melanoma type. Cutaneous and occult melanomas (n = 124 patients) have higher response rates (~40% CR and PR) versus acral and mucosal melanomas (n = 20 patients, 20%; two-sided Fisher’s exact test, P = 0.06). e, Mutational load in responders versus progressors stratified by melanoma type. There was no significant difference between responder and progressor mutational loads when melanomas were stratified by type (two-sided MWW; progressors versus responders (PD/R): skin (n = 42/43), P = 0.27; occult (n = 10/8), P = 0.35; acral (n = 7/2), P = 0.19; mucosal (n = 6/2), P = 0.40). Mutational load was also not a significant predictor of response in combined logistic regression after adjusting for melanoma type (P = 0.24). f, TMB in responders versus nonresponders, stratified by skin or occult melanomas versus mucosal or acral melanomas. Within each subgroup, responders trended toward having higher TMB than nonresponders (cutaneous/occult (n = 52 progressors and 51 responders): MWW, P = 0.14; mucosal/acral (n = 13 progressors and 4 responders): MWW, P = 0.08). Notably, responders with mucosal or acral melanoma (n = 4) had a lower mutational load than progressors with cutaneous or occult melanoma (n = 52; MWW, P = 0.03). Boxplots: box limits indicate the IQR (25th to 75th percentiles), with a center line indicating the median. Whiskers show the value ranges up to 1.5 × IQR above the 75th or below the 25th percentiles, with outliers beyond those ranges shown as individual points. *P < 0.05, **P < 0.01, ***P < 0.001. NS, not significant.
Fig. 1 Cohort genomic and clinical characteristics and association of TMB with response. a, CoMut plot showing association between clinical and genomic characteristics. Each column represents a tumor. Tumors are ordered by best RECIST criteria response (CR, PR, PD, SD or MR), and within each response subgroup by decreasing nonsynonymous (Nonsyn) mutational load (top row). Nonsynonymous mutational burden is further subdivided into clonal (purple) and subclonal (light purple) mutational load. Mutational signatures (sig) refer to the inferred relative contribution of UV-induced mutations, alkylating DNA damage process and other mutational signatures (aging+). The primary type of melanoma (skin, occult, acral or mucosal) is indicated. Tumor purity is the inferred proportion of the tumor sample that is from cancer cells compared to other cell types (Methods). The dominant mutational signature (that is, the mutational signature associated with the highest proportion of mutations) is indicated. Mutations in BRAF, NRAS and NF1 are shown for each tumor. b, Mutational load (mut load) in progressors (n = 65 patients), responders (n = 55 patients) and patients with SD and MR (n = 24 patients). Nonsynonymous mutational load is higher in responders (CR and PR) than in progressors (two-sided MWW, P = 0.026), but is not significantly different between responders and patients having SD or MR as the best RECIST response (two-sided MWW, P = 0.14). c, Mutational load by melanoma type. Different melanoma types have different mutational loads (Kruskal–Wallis, P = 2.4 × 10−5): mutational load is higher in cutaneous and occult melanomas (n = 124 patients) than in acral and mucosal melanomas (n = 20 patients; median 297.5 versus 58; two-sided MWW, P = 1.1 × 10−6). d, Response to anti-PD1 ICB by melanoma type. Cutaneous and occult melanomas (n = 124 patients) have higher response rates (~40% CR and PR) versus acral and mucosal melanomas (n = 20 patients, 20%; two-sided Fisher’s exact test, P = 0.06). e, Mutational load in responders versus progressors stratified by melanoma type. There was no significant difference between responder and progressor mutational loads when melanomas were stratified by type (two-sided MWW; progressors versus responders (PD/R): skin (n = 42/43), P = 0.27; occult (n = 10/8), P = 0.35; acral (n = 7/2), P = 0.19; mucosal (n = 6/2), P = 0.40). Mutational load was also not a significant predictor of response in combined logistic regression after adjusting for melanoma type (P = 0.24). f, TMB in responders versus nonresponders, stratified by skin or occult melanomas versus mucosal or acral melanomas. Within each subgroup, responders trended toward having higher TMB than nonresponders (cutaneous/occult (n = 52 progressors and 51 responders): MWW, P = 0.14; mucosal/acral (n = 13 progressors and 4 responders): MWW, P = 0.08). Notably, responders with mucosal or acral melanoma (n = 4) had a lower mutational load than progressors with cutaneous or occult melanoma (n = 52; MWW, P = 0.03). Boxplots: box limits indicate the IQR (25th to 75th percentiles), with a center line indicating the median. Whiskers show the value ranges up to 1.5 × IQR above the 75th or below the 25th percentiles, with outliers beyond those ranges shown as individual points. *P < 0.05, **P < 0.01, ***P < 0.001. NS, not significant.CC BY · PubMed Central
Fig. 2 Genomic and transcriptomic features associated with response. All P values are unadjusted, unless otherwise indicated. a, Tumor heterogeneity, defined as the proportion of subclonal mutations in each tumor (Methods), in responders (CR or PR) versus progressors (PD). Progressors (n = 65 patients) had greater heterogeneity than responders (n = 55 patients; two-sided MWW, P = 0.02). b, Tumor ploidy, defined as the overall genomic copy number (a normal diploid cell has a copy number of 2; Methods), in responders versus progressors. Responders (n = 55 patients) had higher tumor ploidy than progressors (n = 65 patients; two-sided MWW, P = 0.04). c, Tumor purity, defined as the proportion of DNA from tumor versus other cells in the sample (Methods), in responders versus progressors. Progressors (n = 65 patients) had higher tumor purity than responders (n = 55 patients; two-sided MWW, P = 0.04). d, Response versus progression in TAP2-amplified tumors versus other tumors. TAP2 amplification (n = 6 patients) was associated with response (two-sided Fisher’s exact test, P = 0.008). e, Response versus progression in tumors with amplified MHC-I HLA genes (HLA-A, HLA-B or HLA-C) versus other tumors. MHC-I HLA amplification (n = 6 patients) was associated with response (two-sided Fisher’s exact test, P = 0.008). f, Venn diagram showing the overlap of TAP2-amplified tumors and tumors with amplification of MHC-I HLA genes. Four tumors had amplifications on chromosome 6, including the MHC-I genes HLA-A, HLA-B, HLA-C and TAP2 and two tumors each had amplifications in one but not the other region, for a total of eight tumors with amplifications in either. g, Difference in the median expression and two-sided MWW P value of association between 938 immune-related genes56 and features in responders versus progressors. Expression levels of MHC-II HLA genes (red), MHC-I HLA genes and antigen-presentation machinery (APM)-related genes (orange) are shown. h, Hierarchical clustering of the correlation matrix between genomic, clinical and transcriptomic features associated with response. Color indicates the Pearson correlation between features, from perfect negative correlation (Pearson, r = −1, blue) to perfect positive correlation (Pearson, r = 1, red). An immune-related cluster of MHC-I- and MHC-II-related gene expression is observed, with subclusters of MHC-I and MHC-II genes. Mutational and neoantigen load are highly correlated and form a cluster independently from the immune cluster (for example, Pearson correlation, r = 0.15, P = 0.11 between ssGSEA of MHC-II HLA genes and nonsynonymous mutational load). Purity is negatively correlated with the immune cluster and is independent of ploidy and heterogeneity. The sample size for each correlation depended on the number of available data points: correlations involving exclusively genomic or clinical data had n = 144 tumor samples, whereas correlations involving transcriptomic features had n = 121 tumor samples with data available. Tx, treatment. Boxplots: box limits indicate the IQR (25th to 75th percentile), with a center line indicating the median. Whiskers show the value ranges up to 1.5 × IQR above the 75th or below the 25th percentile, with outliers beyond those ranges shown as individual points. *P < 0.05, **P < 0.01, ***P < 0.001; NS, not significant.
Fig. 2 Genomic and transcriptomic features associated with response. All P values are unadjusted, unless otherwise indicated. a, Tumor heterogeneity, defined as the proportion of subclonal mutations in each tumor (Methods), in responders (CR or PR) versus progressors (PD). Progressors (n = 65 patients) had greater heterogeneity than responders (n = 55 patients; two-sided MWW, P = 0.02). b, Tumor ploidy, defined as the overall genomic copy number (a normal diploid cell has a copy number of 2; Methods), in responders versus progressors. Responders (n = 55 patients) had higher tumor ploidy than progressors (n = 65 patients; two-sided MWW, P = 0.04). c, Tumor purity, defined as the proportion of DNA from tumor versus other cells in the sample (Methods), in responders versus progressors. Progressors (n = 65 patients) had higher tumor purity than responders (n = 55 patients; two-sided MWW, P = 0.04). d, Response versus progression in TAP2-amplified tumors versus other tumors. TAP2 amplification (n = 6 patients) was associated with response (two-sided Fisher’s exact test, P = 0.008). e, Response versus progression in tumors with amplified MHC-I HLA genes (HLA-A, HLA-B or HLA-C) versus other tumors. MHC-I HLA amplification (n = 6 patients) was associated with response (two-sided Fisher’s exact test, P = 0.008). f, Venn diagram showing the overlap of TAP2-amplified tumors and tumors with amplification of MHC-I HLA genes. Four tumors had amplifications on chromosome 6, including the MHC-I genes HLA-A, HLA-B, HLA-C and TAP2 and two tumors each had amplifications in one but not the other region, for a total of eight tumors with amplifications in either. g, Difference in the median expression and two-sided MWW P value of association between 938 immune-related genes56 and features in responders versus progressors. Expression levels of MHC-II HLA genes (red), MHC-I HLA genes and antigen-presentation machinery (APM)-related genes (orange) are shown. h, Hierarchical clustering of the correlation matrix between genomic, clinical and transcriptomic features associated with response. Color indicates the Pearson correlation between features, from perfect negative correlation (Pearson, r = −1, blue) to perfect positive correlation (Pearson, r = 1, red). An immune-related cluster of MHC-I- and MHC-II-related gene expression is observed, with subclusters of MHC-I and MHC-II genes. Mutational and neoantigen load are highly correlated and form a cluster independently from the immune cluster (for example, Pearson correlation, r = 0.15, P = 0.11 between ssGSEA of MHC-II HLA genes and nonsynonymous mutational load). Purity is negatively correlated with the immune cluster and is independent of ploidy and heterogeneity. The sample size for each correlation depended on the number of available data points: correlations involving exclusively genomic or clinical data had n = 144 tumor samples, whereas correlations involving transcriptomic features had n = 121 tumor samples with data available. Tx, treatment. Boxplots: box limits indicate the IQR (25th to 75th percentile), with a center line indicating the median. Whiskers show the value ranges up to 1.5 × IQR above the 75th or below the 25th percentile, with outliers beyond those ranges shown as individual points. *P < 0.05, **P < 0.01, ***P < 0.001; NS, not significant.CC BY · PubMed Central
Fig. 3 Differential predictors of response and progression in ipilimumab-treated tumors versus ipilimumab-naive tumors. a, Timeline showing when sequenced biopsies were obtained from tumors that were treated with ipilimumab or naive to ipilimumab in the course of therapy. Subsequent analyses focused on comparing tumor biopsies obtained after ipilimumab treatment (n = 45 WES, n = 34 RNA-seq) to ipilimumab-naive tumor biopsies (n = 84 WES, n = 74 RNA-seq). b, Best RECIST response by ipilimumab pretreatment status. There was no difference between the distribution of responses in naive (n = 84) and pretreated (n = 60) patients (two-sided chi-squared test, P = 0.44; degrees of freedom (d.f.) = 3). c, ssGSEA of MHC-II HLA genes (Methods) in responders (n = 47 patients) versus progressors (n = 56 patients) in the overall cohort. MHC-II scores were higher in responders than in progressors (two-sided MWW, P = 0.03). d, ssGSEA of MHC-II HLA genes in responders versus progressors in the post-ipilimumab-treatment subgroup. MHC-II scores were higher in responders (n = 11 patients) than in progressors (n = 16 patients; two-sided MWW, P = 0.03). e, ssGSEA of MHC-II HLA genes in responders (n = 31 patients) versus progressors (n = 34 patients) in the ipilimumab-naive subgroup. There was no significant difference in MHC-II scores between responders and progressors (two-sided MWW, P = 0.31). f, MHC-II HLA gene set scores (ssGSEA) in responders (n = 10 patients) versus progressors (n = 22 patients) in a validation cohort (Methods; two-sided MWW, P = 0.34). g, MHC-II HLA gene set scores (ssGSEA) in responders (n = 4 patients) versus progressors (n = 11 patients) in the ipilimumab-treated subgroup of a validation cohort (two-sided MWW, P = 0.10). h, MHC-II HLA gene set scores (ssGSEA) in responders (n = 6 patients) versus progressors (n = 11 patients) in the ipilimumab-naive subgroup of a validation cohort (two-sided MWW, P = 0.80). i, Selected Cancer Hallmark gene sets (GSEA) enriched in responders versus progressors in the overall (n = 47 responders and 56 progressors), post-ipilimumab-treatment (n = 11 responders and 16 progressors), and ipilimumab-naive (n = 31 responders and 34 progressors) subgroups of our discovery cohort. IFN-γ and IFN-α response pathways were enriched in responders in the overall (FDR, q < 0.001 and q = 0.02, respectively) and ipilimumab-treated (q < 0.001, both) subgroups but not in the ipilimumab-naive subgroups (q = 0.13 and q = 0.997, respectively) in the discovery cohort (empiric, P = 0.183 and P = 0.18, respectively for the difference in q values between the subgroups; Methods). j, Selected Cancer Hallmark gene sets (GSEA) enriched in responders versus progressors in the overall, post-ipilimumab-treatment and ipilimumab-naive subgroups in an independent validation cohort. IFN-γ and IFN-α response pathways were enriched in responders in the overall (FDR, q < 0.0001 and q = 0.0034, respectively) and ipilimumab-treated (q < 0.0001, both) subgroups but not in the ipilimumab-naive subgroup (q = 0.87 and q = 0.03 (enriched in progressors), respectively) in the discovery cohort. Pathways favoring enrichment in progressors (as opposed to responders) are visualized here with an FDR q value of 1. All Hallmark pathways and their GSEA enrichment scores are shown in Supplementary Table 5. Boxplots: box limits indicate the IQR (25th to 75th percentile), with a center line indicating the median. Whiskers show the value ranges up to 1.5 × IQR above the 75th or below the 25th percentile, with outliers beyond those ranges shown as individual points. *P < 0.05, **P < 0.01, ***P < 0.001. NS, not significant.
Fig. 3 Differential predictors of response and progression in ipilimumab-treated tumors versus ipilimumab-naive tumors. a, Timeline showing when sequenced biopsies were obtained from tumors that were treated with ipilimumab or naive to ipilimumab in the course of therapy. Subsequent analyses focused on comparing tumor biopsies obtained after ipilimumab treatment (n = 45 WES, n = 34 RNA-seq) to ipilimumab-naive tumor biopsies (n = 84 WES, n = 74 RNA-seq). b, Best RECIST response by ipilimumab pretreatment status. There was no difference between the distribution of responses in naive (n = 84) and pretreated (n = 60) patients (two-sided chi-squared test, P = 0.44; degrees of freedom (d.f.) = 3). c, ssGSEA of MHC-II HLA genes (Methods) in responders (n = 47 patients) versus progressors (n = 56 patients) in the overall cohort. MHC-II scores were higher in responders than in progressors (two-sided MWW, P = 0.03). d, ssGSEA of MHC-II HLA genes in responders versus progressors in the post-ipilimumab-treatment subgroup. MHC-II scores were higher in responders (n = 11 patients) than in progressors (n = 16 patients; two-sided MWW, P = 0.03). e, ssGSEA of MHC-II HLA genes in responders (n = 31 patients) versus progressors (n = 34 patients) in the ipilimumab-naive subgroup. There was no significant difference in MHC-II scores between responders and progressors (two-sided MWW, P = 0.31). f, MHC-II HLA gene set scores (ssGSEA) in responders (n = 10 patients) versus progressors (n = 22 patients) in a validation cohort (Methods; two-sided MWW, P = 0.34). g, MHC-II HLA gene set scores (ssGSEA) in responders (n = 4 patients) versus progressors (n = 11 patients) in the ipilimumab-treated subgroup of a validation cohort (two-sided MWW, P = 0.10). h, MHC-II HLA gene set scores (ssGSEA) in responders (n = 6 patients) versus progressors (n = 11 patients) in the ipilimumab-naive subgroup of a validation cohort (two-sided MWW, P = 0.80). i, Selected Cancer Hallmark gene sets (GSEA) enriched in responders versus progressors in the overall (n = 47 responders and 56 progressors), post-ipilimumab-treatment (n = 11 responders and 16 progressors), and ipilimumab-naive (n = 31 responders and 34 progressors) subgroups of our discovery cohort. IFN-γ and IFN-α response pathways were enriched in responders in the overall (FDR, q < 0.001 and q = 0.02, respectively) and ipilimumab-treated (q < 0.001, both) subgroups but not in the ipilimumab-naive subgroups (q = 0.13 and q = 0.997, respectively) in the discovery cohort (empiric, P = 0.183 and P = 0.18, respectively for the difference in q values between the subgroups; Methods). j, Selected Cancer Hallmark gene sets (GSEA) enriched in responders versus progressors in the overall, post-ipilimumab-treatment and ipilimumab-naive subgroups in an independent validation cohort. IFN-γ and IFN-α response pathways were enriched in responders in the overall (FDR, q < 0.0001 and q = 0.0034, respectively) and ipilimumab-treated (q < 0.0001, both) subgroups but not in the ipilimumab-naive subgroup (q = 0.87 and q = 0.03 (enriched in progressors), respectively) in the discovery cohort. Pathways favoring enrichment in progressors (as opposed to responders) are visualized here with an FDR q value of 1. All Hallmark pathways and their GSEA enrichment scores are shown in Supplementary Table 5. Boxplots: box limits indicate the IQR (25th to 75th percentile), with a center line indicating the median. Whiskers show the value ranges up to 1.5 × IQR above the 75th or below the 25th percentile, with outliers beyond those ranges shown as individual points. *P < 0.05, **P < 0.01, ***P < 0.001. NS, not significant.CC BY · PubMed Central
Fig. 4 Progression versus response by immune infiltrate and MHC-II HLA expression stratified by ipilimumab treatment. a, Proportion of responders versus progressors in subgroups with high and low MHC-II HLA score (ssGSEA; divided by the median). Overall, a high MHC-II HLA score was associated with higher response (MHC-II high: n = 29 responders and 20 progressors; MHC-II low: n = 18 responders and 36 progressors; two-sided Fisher’s exact test, P = 0.01; OR = 2.9, 95% CI 1.3–6.5). b, As in a but in the ipilimumab-treated subgroup. Tumors with low MHC-II HLA score (ssGSEA) were associated with PD (MHC-II high: n = 9 responders and 5 progressors; MHC-II low: n = 2 responders and 11 progressors; two-sided Fisher’s exact test, P = 0.02; OR = 9.9, 95% CI 1.5–63.7). c, As in a but in the ipilimumab-naive subgroup. There was no significant difference between tumors with high and low MHC-II HLA scores (ssGSEA; MHC-II high: n = 18 responders and 14 progressors; MHC-II low: n = 13 responders and 19 progressors; two-sided Fisher’s exact test, P = 0.32; OR = 1.9, 95% CI 0.7–5.1). d, Proportion of responders versus progressors in subgroups with high and low immune infiltrate scores (divided by the median). Overall, there was no statistically significant difference in the proportion of responders versus progressors in tumors with high versus low immune infiltrate (infiltrate high: n = 27 responders and 24 progressors; infiltrate low: n = 20 responders and 32 progressors; two-sided Fisher’s exact test, P = 0.17; OR = 1.8, 95% CI 0.8–3.9). e, As in d but in the ipilimumab-treated subgroup. Tumors with low immune infiltrate scores were strongly associated with PD (infiltrate high: n = 10 responders and 7 progressors; infiltrate low: n = 1 responder and 9 progressors; two-sided Fisher’s exact test, P = 0.02; OR = 12.9, 95% CI 1.3–125.8). f, As in d but in the ipilimumab-naive subgroup. There was no statistically significant difference between tumors with high and low immune infiltrate scores (infiltrate high: n = 14 responders and 15 progressors; infiltrate low: n = 17 responders and 18 progressors; two-sided Fisher’s exact test, P = 1.0; OR = 0.99, 95% CI 0.4–2.6). g, Schematic of the hypothesized effect of ipilimumab treatment on immune response as a predictor of subsequent response to anti-PD1 ICB. Tumors that were treated with ipilimumab but failed to have an immune response or infiltrate in the tumor microenvironment were strongly predicted to have intrinsic resistance to anti-PD1 ICB. However, having an immune response did not guarantee a subsequent response to anti-PD1 ICB. In ipilimumab-naive tumors, neither the presence nor the absence of immune infiltrate was a good predictor of anti-PD1 ICB response.
Fig. 4 Progression versus response by immune infiltrate and MHC-II HLA expression stratified by ipilimumab treatment. a, Proportion of responders versus progressors in subgroups with high and low MHC-II HLA score (ssGSEA; divided by the median). Overall, a high MHC-II HLA score was associated with higher response (MHC-II high: n = 29 responders and 20 progressors; MHC-II low: n = 18 responders and 36 progressors; two-sided Fisher’s exact test, P = 0.01; OR = 2.9, 95% CI 1.3–6.5). b, As in a but in the ipilimumab-treated subgroup. Tumors with low MHC-II HLA score (ssGSEA) were associated with PD (MHC-II high: n = 9 responders and 5 progressors; MHC-II low: n = 2 responders and 11 progressors; two-sided Fisher’s exact test, P = 0.02; OR = 9.9, 95% CI 1.5–63.7). c, As in a but in the ipilimumab-naive subgroup. There was no significant difference between tumors with high and low MHC-II HLA scores (ssGSEA; MHC-II high: n = 18 responders and 14 progressors; MHC-II low: n = 13 responders and 19 progressors; two-sided Fisher’s exact test, P = 0.32; OR = 1.9, 95% CI 0.7–5.1). d, Proportion of responders versus progressors in subgroups with high and low immune infiltrate scores (divided by the median). Overall, there was no statistically significant difference in the proportion of responders versus progressors in tumors with high versus low immune infiltrate (infiltrate high: n = 27 responders and 24 progressors; infiltrate low: n = 20 responders and 32 progressors; two-sided Fisher’s exact test, P = 0.17; OR = 1.8, 95% CI 0.8–3.9). e, As in d but in the ipilimumab-treated subgroup. Tumors with low immune infiltrate scores were strongly associated with PD (infiltrate high: n = 10 responders and 7 progressors; infiltrate low: n = 1 responder and 9 progressors; two-sided Fisher’s exact test, P = 0.02; OR = 12.9, 95% CI 1.3–125.8). f, As in d but in the ipilimumab-naive subgroup. There was no statistically significant difference between tumors with high and low immune infiltrate scores (infiltrate high: n = 14 responders and 15 progressors; infiltrate low: n = 17 responders and 18 progressors; two-sided Fisher’s exact test, P = 1.0; OR = 0.99, 95% CI 0.4–2.6). g, Schematic of the hypothesized effect of ipilimumab treatment on immune response as a predictor of subsequent response to anti-PD1 ICB. Tumors that were treated with ipilimumab but failed to have an immune response or infiltrate in the tumor microenvironment were strongly predicted to have intrinsic resistance to anti-PD1 ICB. However, having an immune response did not guarantee a subsequent response to anti-PD1 ICB. In ipilimumab-naive tumors, neither the presence nor the absence of immune infiltrate was a good predictor of anti-PD1 ICB response.CC BY · PubMed Central
Fig. 5 Integrative predictive modeling of intrinsic resistance to anti-PD1 ICB. a, CoMut plot showing the relationship between response and predictive features in the ipilimumab-treated subgroup. Each column represents a patient, and the top row indicates whether the patient had PD (intrinsic resistance) or non-PD (CR, PR, SD or MR) as the best response. Patients are sorted by MHC-II HLA score, which was the most predictive feature and was correlated with MHC-I and IFN response pathway scores. MHC-II HLA, LDH at treatment initiation and the presence of lymph node metastases (LN met) were features used for our logistic regression model, chosen using forward selection (Methods). b, A receiver–operator characteristic (ROC) curve for our predictive model for ipilimumab-treated tumors (n = 34 patients) using MHC-II HLA, LDH and lymph node metastases as features. The AUC was 0.9, and the log-likelihood ratio was P = 0.0003. The fivefold cross-validation mean AUC was 0.83. c, CoMut plot showing the relationship between response and predictive features in the ipilimumab-naive subgroup. Each column represents a patient, and the top row indicates whether the patient had PD (intrinsic resistance) or non-PD as the best response. Patients are sorted by tumor heterogeneity (top), tumor ploidy (middle) and tumor purity (bottom), which were the three features chosen in our predictive model (Methods). Purity, ploidy and heterogeneity were independent predictors in the multivariate model (Supplementary Table 9). d, The ROC curve for our predictive model for ipilimumab-naive tumors (n = 84 tumors) using heterogeneity, purity and ploidy as features. The AUC was 0.77, and the log-likelihood ratio was P = 0.0003. The average tenfold cross-validation mean AUC was 0.73. e, Survival (PFS and OS as indicated) stratified by high versus low predictive model score (split by the median) in ipilimumab-treated tumors in our discovery cohort. Tumors with high scores had worse PFS and OS (two-sided KM log-rank test, P = 0.0001 and P = 0.001, respectively). f, Survival (PFS and OS as indicated) stratified by high versus low predictive model score (split by the median) in ipilimumab-naive tumors in our discovery cohort. Tumors with high scores had worse PFS and OS (two-sided KM log-rank test, P = 0.003 and P = 6.3 × 10−5, respectively).
Fig. 5 Integrative predictive modeling of intrinsic resistance to anti-PD1 ICB. a, CoMut plot showing the relationship between response and predictive features in the ipilimumab-treated subgroup. Each column represents a patient, and the top row indicates whether the patient had PD (intrinsic resistance) or non-PD (CR, PR, SD or MR) as the best response. Patients are sorted by MHC-II HLA score, which was the most predictive feature and was correlated with MHC-I and IFN response pathway scores. MHC-II HLA, LDH at treatment initiation and the presence of lymph node metastases (LN met) were features used for our logistic regression model, chosen using forward selection (Methods). b, A receiver–operator characteristic (ROC) curve for our predictive model for ipilimumab-treated tumors (n = 34 patients) using MHC-II HLA, LDH and lymph node metastases as features. The AUC was 0.9, and the log-likelihood ratio was P = 0.0003. The fivefold cross-validation mean AUC was 0.83. c, CoMut plot showing the relationship between response and predictive features in the ipilimumab-naive subgroup. Each column represents a patient, and the top row indicates whether the patient had PD (intrinsic resistance) or non-PD as the best response. Patients are sorted by tumor heterogeneity (top), tumor ploidy (middle) and tumor purity (bottom), which were the three features chosen in our predictive model (Methods). Purity, ploidy and heterogeneity were independent predictors in the multivariate model (Supplementary Table 9). d, The ROC curve for our predictive model for ipilimumab-naive tumors (n = 84 tumors) using heterogeneity, purity and ploidy as features. The AUC was 0.77, and the log-likelihood ratio was P = 0.0003. The average tenfold cross-validation mean AUC was 0.73. e, Survival (PFS and OS as indicated) stratified by high versus low predictive model score (split by the median) in ipilimumab-treated tumors in our discovery cohort. Tumors with high scores had worse PFS and OS (two-sided KM log-rank test, P = 0.0001 and P = 0.001, respectively). f, Survival (PFS and OS as indicated) stratified by high versus low predictive model score (split by the median) in ipilimumab-naive tumors in our discovery cohort. Tumors with high scores had worse PFS and OS (two-sided KM log-rank test, P = 0.003 and P = 6.3 × 10−5, respectively).CC BY · PubMed Central
Extended Data Fig. 1 Consort Diagram showing inclusion, exclusion, and quality control criteria for patients/tumors included in analysis.
Extended Data Fig. 1 Consort Diagram showing inclusion, exclusion, and quality control criteria for patients/tumors included in analysis.CC BY · PubMed Central
Extended Data Fig. 2 Survival and genomic copy number characteristics of responders (n = 55 patients) (defined as CR or PR as best response) vs. progressors (n = 65 patients) (defined as PD as best response). (a) PFS Kaplan Meier survival curves; two-sided log-rank p = 2.1e-28 (b) OS Kaplan Meier survival curves; two-sided log-rank p = 5.1e-18 (c) Proportion of tumor genome with copy number alterations (two-sided MWW p = 0.09). Boxplots: Box limits indicate the interquartile range (25th-75th percentile), with a center line indicating the median. Whiskers show the value ranges up to 1.5xIQR above the 75th or below the 25th percentile, with outliers beyond those ranges shown as individual points. PD = progressive disease. CR = complete response. PR = partial response.
Extended Data Fig. 2 Survival and genomic copy number characteristics of responders (n = 55 patients) (defined as CR or PR as best response) vs. progressors (n = 65 patients) (defined as PD as best response). (a) PFS Kaplan Meier survival curves; two-sided log-rank p = 2.1e-28 (b) OS Kaplan Meier survival curves; two-sided log-rank p = 5.1e-18 (c) Proportion of tumor genome with copy number alterations (two-sided MWW p = 0.09). Boxplots: Box limits indicate the interquartile range (25th-75th percentile), with a center line indicating the median. Whiskers show the value ranges up to 1.5xIQR above the 75th or below the 25th percentile, with outliers beyond those ranges shown as individual points. PD = progressive disease. CR = complete response. PR = partial response.CC BY · PubMed Central
Extended Data Fig. 3 Association of Genomic Alterations in Genes with Response vs. Progression. For each gene, the association of a genomic alteration with response was tested using a two-sided Fisher’s Exact test, and the odds ratio (OR) and p-value calculated. The adjusted OR is a conservative estimate generated by adding 1 to both mutant gene responders and progressors, thus moving the OR estimate closer to 1. The x- and y-axis are on log scales, but labeled in the original units. (a) Association of gene mutations with response (n = 55 patients) vs. progression (n = 65 patients). (b) Association of gene amplifications with response (n = 55 patients) vs. progression (n = 65 patients).
Extended Data Fig. 3 Association of Genomic Alterations in Genes with Response vs. Progression. For each gene, the association of a genomic alteration with response was tested using a two-sided Fisher’s Exact test, and the odds ratio (OR) and p-value calculated. The adjusted OR is a conservative estimate generated by adding 1 to both mutant gene responders and progressors, thus moving the OR estimate closer to 1. The x- and y-axis are on log scales, but labeled in the original units. (a) Association of gene mutations with response (n = 55 patients) vs. progression (n = 65 patients). (b) Association of gene amplifications with response (n = 55 patients) vs. progression (n = 65 patients).CC BY · PubMed Central
Extended Data Fig. 4 Amplification of Chromosome 6 Regions in TAP2, HLA-A/B/C amplified tumors. Patients with amplifications in TAP2 or HLA-A/B/C are shown, with inferred amplifications in the relevant section of Chromosome 6 shown. 6 patients had amplifications in TAP2, 6 patients had amplifications in HLA-A/B/C, and 4 patients had amplifications of a chromosomal region including all four genes.
Extended Data Fig. 4 Amplification of Chromosome 6 Regions in TAP2, HLA-A/B/C amplified tumors. Patients with amplifications in TAP2 or HLA-A/B/C are shown, with inferred amplifications in the relevant section of Chromosome 6 shown. 6 patients had amplifications in TAP2, 6 patients had amplifications in HLA-A/B/C, and 4 patients had amplifications of a chromosomal region including all four genes.CC BY · PubMed Central
Extended Data Fig. 5 Hierarchical clustering of the correlation matrix between transcriptional signatures previously associated with immunotherapy response and hallmark gene-sets, genomic, clinical, and transcriptomic features associated with response in our cohort. The color indicates the Pearson correlation between features, from perfect negative correlation (Pearson r = -1, blue) to perfect positive correlation (Pearson r = 1, red). Most previously hypothesized signatures cluster within an immune-activity related group with immune-related gene-sets. Sample size for each correlation depends on the number of available data points: correlations involving exclusively genomic or clinical data had n = 144 tumor samples, while correlations involving transcriptomic features had n = 121 tumor samples with data available.
Extended Data Fig. 5 Hierarchical clustering of the correlation matrix between transcriptional signatures previously associated with immunotherapy response and hallmark gene-sets, genomic, clinical, and transcriptomic features associated with response in our cohort. The color indicates the Pearson correlation between features, from perfect negative correlation (Pearson r = -1, blue) to perfect positive correlation (Pearson r = 1, red). Most previously hypothesized signatures cluster within an immune-activity related group with immune-related gene-sets. Sample size for each correlation depends on the number of available data points: correlations involving exclusively genomic or clinical data had n = 144 tumor samples, while correlations involving transcriptomic features had n = 121 tumor samples with data available.CC BY · PubMed Central
Extended Data Fig. 6 Expression of Genomic Features in Responders vs. Progressors in ipilimumab-treated and -naive subsets. Two-sided Mann-Whitney-Wilcoxon tests were used to compare responders to progressors (ipilimumab-treated: n = 16 responders/20 progressors; ipilimumab-naïve: n = 34 responders/37 progressors). All p-values are unadjusted. Log Nonsyn Mutload: Ipi-treated p = 0.15; Ipi-naïve p = 0.15. Log Clonal Mutload: Ipi-treated p = 0.17; Ipi-naïve p = 0.08. Heterogeneity: Ipi-treated p = 0.51; Ipi-naïve p = 0.057. Ploidy: Ipi-treated p = 0.45; Ipi-naïve p = 0.002. Boxplots: Box limits indicate the interquartile range (25th-75th percentile), with a center line indicating the median. Whiskers show the value ranges up to 1.5xIQR above the 75th or below the 25th percentile, with outliers beyond those ranges shown as individual points.
Extended Data Fig. 6 Expression of Genomic Features in Responders vs. Progressors in ipilimumab-treated and -naive subsets. Two-sided Mann-Whitney-Wilcoxon tests were used to compare responders to progressors (ipilimumab-treated: n = 16 responders/20 progressors; ipilimumab-naïve: n = 34 responders/37 progressors). All p-values are unadjusted. Log Nonsyn Mutload: Ipi-treated p = 0.15; Ipi-naïve p = 0.15. Log Clonal Mutload: Ipi-treated p = 0.17; Ipi-naïve p = 0.08. Heterogeneity: Ipi-treated p = 0.51; Ipi-naïve p = 0.057. Ploidy: Ipi-treated p = 0.45; Ipi-naïve p = 0.002. Boxplots: Box limits indicate the interquartile range (25th-75th percentile), with a center line indicating the median. Whiskers show the value ranges up to 1.5xIQR above the 75th or below the 25th percentile, with outliers beyond those ranges shown as individual points.CC BY · PubMed Central
Extended Data Fig. 7 Survival of patients stratified by progressive disease (PD) or non-PD (SD, MR, PR, or CR) best response (RECIST 1.1) to PD-1 ICB. (a) PFS (two-sided logrank p = 4.0e-38 (b) OS (two-sided logrank p = 1.2e-19).
Extended Data Fig. 7 Survival of patients stratified by progressive disease (PD) or non-PD (SD, MR, PR, or CR) best response (RECIST 1.1) to PD-1 ICB. (a) PFS (two-sided logrank p = 4.0e-38 (b) OS (two-sided logrank p = 1.2e-19).CC BY · PubMed Central
Extended Data Fig. 8 Model selection criteria with increasing number of features by forward selection; cross-validation AUC distribution of final models. (a) Model AUC for the ipi-treated subgroup. Each point is labeled with the additional feature, and the final model includes MHC-II, Lymph node metastasis, and LDH below the median as features. (b) Model AUC for the ipi-naive subgroup. Each point is labeled with the additional feature, and the final model includes ploidy, heterogeneity, and purity as features. (c) Model AIC/BIC for the ipi-experienced subgroup. (d) Model AIC/BIC for the ipi-naïve subgroup. (e) Using 10-fold crossvalidation in the ipilimumab-naïve subset, mean crossvalidation AUC is 0.73, with a SD of 0.25. Each fold split the cohort into a training (n = 74 patients) and test set (n = 10 patients) ((f) In the ipi-experienced subset, mean crossvalidation AUC is 0.85, with a SD of 0.18. Each fold split the cohort into a training (n = 27 patients) and test set (n = 7 patients).
Extended Data Fig. 8 Model selection criteria with increasing number of features by forward selection; cross-validation AUC distribution of final models. (a) Model AUC for the ipi-treated subgroup. Each point is labeled with the additional feature, and the final model includes MHC-II, Lymph node metastasis, and LDH below the median as features. (b) Model AUC for the ipi-naive subgroup. Each point is labeled with the additional feature, and the final model includes ploidy, heterogeneity, and purity as features. (c) Model AIC/BIC for the ipi-experienced subgroup. (d) Model AIC/BIC for the ipi-naïve subgroup. (e) Using 10-fold crossvalidation in the ipilimumab-naïve subset, mean crossvalidation AUC is 0.73, with a SD of 0.25. Each fold split the cohort into a training (n = 74 patients) and test set (n = 10 patients) ((f) In the ipi-experienced subset, mean crossvalidation AUC is 0.85, with a SD of 0.18. Each fold split the cohort into a training (n = 27 patients) and test set (n = 7 patients).CC BY · PubMed Central
Extended Data Fig. 9 Subgroup Model ROC curves applied to other subgroups. (a) Ipilimumab-treated subgroup model applied to ipilimumab-naive (n = 74 patients) subgroup; (b) Ipilimumab-naive model applied to ipilimumab-treated (n = 45 patients) subgroup; (c) Ipi-naïve model predicting PD as best response to PD-1 ICB applied to pre-ipilimumab-treated (n = 15 patients) tumors. (top) Comut plot overlaying best response and the ipi-naïve model score. Each column is one of 15 ipilimumab-treated tumors biopsied prior to ipilimumab therapy, ordered by the predictive model score. The 5 highest scoring tumors were all PD. (bottom) Receiver-operator curve for our ipilimumab-naive predictive model applied to pre-ipilimumab tumors (n = 15), with an AUC of 0.71.
Extended Data Fig. 9 Subgroup Model ROC curves applied to other subgroups. (a) Ipilimumab-treated subgroup model applied to ipilimumab-naive (n = 74 patients) subgroup; (b) Ipilimumab-naive model applied to ipilimumab-treated (n = 45 patients) subgroup; (c) Ipi-naïve model predicting PD as best response to PD-1 ICB applied to pre-ipilimumab-treated (n = 15 patients) tumors. (top) Comut plot overlaying best response and the ipi-naïve model score. Each column is one of 15 ipilimumab-treated tumors biopsied prior to ipilimumab therapy, ordered by the predictive model score. The 5 highest scoring tumors were all PD. (bottom) Receiver-operator curve for our ipilimumab-naive predictive model applied to pre-ipilimumab tumors (n = 15), with an AUC of 0.71.CC BY · PubMed Central
Extended Data Fig. 10 Limited validation testing single feature models in validation cohort. (a) Relationship between response and MHC-II HLA score in the ipilimumab-treated subgroup of a validation cohort. Each column is a patient, and patients are sorted by MHC-II HLA score. (b) Validation of Predictive Model for ipilimumab-treated tumors. In a validation cohort (n = 23), a model using the only common available feature (MHC-II HLA score) had an AUC of 0.65 (log likelihood ratio p = 0.09, empiric p = 0.21, Methods). (c) Relationship between response and MHC-II HLA score in the ipilimumab-treated subgroup of a validation cohort. Each column is a patient, and patients are sorted by MHC-II HLA score. (d) Validation of Predictive Model for ipilimumab-naive Tumors. In a validation cohort (n = 20), a model using the only common available feature (heterogeneity) had an AUC of 0.73 (log likelihood ratio p = 0.17, empiric p = 0.066, Methods).
Extended Data Fig. 10 Limited validation testing single feature models in validation cohort. (a) Relationship between response and MHC-II HLA score in the ipilimumab-treated subgroup of a validation cohort. Each column is a patient, and patients are sorted by MHC-II HLA score. (b) Validation of Predictive Model for ipilimumab-treated tumors. In a validation cohort (n = 23), a model using the only common available feature (MHC-II HLA score) had an AUC of 0.65 (log likelihood ratio p = 0.09, empiric p = 0.21, Methods). (c) Relationship between response and MHC-II HLA score in the ipilimumab-treated subgroup of a validation cohort. Each column is a patient, and patients are sorted by MHC-II HLA score. (d) Validation of Predictive Model for ipilimumab-naive Tumors. In a validation cohort (n = 20), a model using the only common available feature (heterogeneity) had an AUC of 0.73 (log likelihood ratio p = 0.17, empiric p = 0.066, Methods).CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 80 references

Cited by

Sorted by citations

See all 1,039

Related

Metadata

Authors
Liu, D., Schilling, B., Liu, D., Sucker, A., Livingstone, E., Jerby-Arnon, L., Zimmer, L., Gutzmer, R., Satzger, I., Loquai, C., Grabbe, S., Vokes, N., Margolis, C. A., Conway, J., He, M. X., Elmarakeby, H., Dietlein, F., Miao, D., Tracy, A., Gogas, H., Goldinger, S. M., Utikal, J., Blank, C. U., Rauschenberg, R., von Bubnoff, D., Krackhardt, A., Weide, B., Haferkamp, S., Kiecker, F., Izar, B., Garraway, L., Regev, A., Flaherty, K., Paschen, A., Van Allen, E. M., & Schadendorf, D.
Affiliations
  • Broad Institute (US)
  • German Cancer Research Center (DE)
  • Medizinische Hochschule Hannover (DE)
  • University Medical Center of the Johannes Gutenberg University Mainz (DE)
  • National and Kapodistrian University of Athens (GR)
  • University Hospital Zurich (CH)
  • The Netherlands Cancer Institute (NL)
  • Universitätsklinikum Tübingen (DE)
  • University Hospital Regensburg (DE)
  • Charité – Universitätsmedizin Berlin (DE)
  • Eli Lilly (United States) (US)
  • Massachusetts General Hospital (US)
Corresponding author
Schadendorf, Dirk
ORCID
Venue
Nature Medicine
Publisher
Springer Science and Business Media LLC
Published
2019-12
Volume
25
Issue
12
Pages
1916-1927
ISSN
  • 1078-8956 (ISSN-L)
  • 1546-170X
DOI
10.1038/s41591-019-0654-5
PMC
PMC6898788
OpenAlex
W2991341799
Cited by
1,039
References
80
Citations by year
2026: 118 · 2025: 156 · 2024: 203 · 2023: 210 · 2022: 157 · 2021: 129 · 2020: 64
FWCI
37.04
Type
journal article
Language
en
Update notices
correction (2020-06-19)
Open access
Licence
Locations
  • Nature Medicine (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by)
  • University of Regensburg Publication Server (University of Regensburg) (submittedVersion)
  • mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) (submittedVersion, other-oa)
  • mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) (submittedVersion)
  • Nature
Funders
  • National Science Foundation (1144152)
  • Conquer Cancer Foundation (T32 GM008313)
  • Damon Runyon Cancer Research Foundation
  • Dr. Miriam and Sheldon G. Adelson Medical Research Foundation
  • Deutsche Forschungsgemeinschaft (SCHA422/17-1, SCHA 422/17, PA 2376/1‐, HO 6389/2-1, KFO337, 2376/1-1, 6389/2-1, PA 2376/1‐1)
  • Universität des Saarlandes
  • Nationales Centrum für Tumorerkrankungen Heidelberg
  • National Institutes of Health (DGE1144152)
  • National Human Genome Research Institute (T32HG002295)
  • National Cancer Institute (R01CA227388, K08CA222663, K08CA234458, U01CA233100)
  • National Institute of General Medical Sciences (T32GM008313)
Topics
  • Health Sciences › Medicine › Oncology › Cancer Immunotherapy and Biomarkers
  • Health Sciences › Medicine › Oncology › Cutaneous Melanoma Detection and Management
  • Life Sciences › Immunology and Microbiology › Immunology › Immunotherapy and Immune Responses
Keywords
metastatic melanoma, predictive biomarkers, tumor mutational burden, MHC class I antigen presentation, MHC class II antigen presentation, immune checkpoint inhibitor resistance, transcriptomic features, clinical outcome prediction
MeSH terms
Exome Sequencing, Nivolumab, Female, Humans, Male, Melanoma, Mutation, Neoplasm Metastasis, Antigen Presentation, Drug Resistance, Neoplasm, Transcriptome, CTLA-4 Antigen, Programmed Cell Death 1 Receptor, Antibodies, Monoclonal, Humanized

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review