Extended Data Fig. 8: HLA genetic variation and TMB correlations. | Nature Cancer

Extended Data Fig. 8: HLA genetic variation and TMB correlations.

From: Multimodal genomic features predict outcome of immune checkpoint blockade in non-small-cell lung cancer

Extended Data Fig. 8

(a) We investigated whether there is an enrichment for chromosome 6p which contains the HLA class I loci for 3,767 tumors from TCGA (BLCA N=89 tumors, BRCA N=737 tumors, COAD N=348 tumors, GBM N=483 tumors, HNSCC N=269 tumors, KIRC N=370 tumors, LUAD N=287 tumors, LUSC N=257 tumors, OV N=456 tumors, READ N=146 tumors, UCEC N=325 tumors). LOH events in NSCLC were compared to the background arm-level allelic imbalance of the same tumor type and across tumor types. Chromosome 6p losses were not more frequent compared to other chromosomal arm level deletions (on the contrary the degree of chromosome 6p LOH was lower compared to other chromosomal arms deletions in lung tumors, N=544 tumors, Fisher’s exact p=0.037). In contrast, when chromosome 6p LOH events were compared between lung tumors and 9 tumor types (BLCA, BRCA, COAD, GBM, HNSCC, KIRC, OV, READ and UCEC), we found that LOH events involving chromosome 6p that contains the HLA class I loci are more frequent in lung cancer (NSCLC N=533, other tumors N= 3223, 17.3% vs. 8.2%, Fisher’s exact p=6.8e-10), without any evidence for positive selection of these events in advanced stage disease. Analysis of variance (ANOVA) was applied to assess the correlation of germline homozygosity in HLA class I genes with tumor mutation burden in 6 tumor types (BLCA, BRCA, COAD, HNSCC, KIRC, LUAD and LUSC, total N=3,601). (b) Germline HLA zygosity was not correlated with TMB for the vast majority of tumors examined with the exception of bladder cancer (N=396 tumors, ANOVA p=0.02). (c) Germline and tumor HLA class I status was combined to determine the number of unique HLA class I alleles in each tumor. The number of unique HLA class I alleles appeared to correlate with TMB such that tumors with a higher number of unique HLA class I alleles in the tumor harbored a lower non-synonymous mutation load for BLCA (N=86 tumors, ANOVA p=0.02) and HNSCC (N=244, ANOVA p=0.07). When tumors heterozygous for all three HLA class I loci (6 HLA class I alleles) where compared to tumors that were homozygous in all three HLA class I loci (3 HLA class I alleles), a lower TMB was noted in the tumors with the more intact antigen presentation capacity (N=86 tumors, Wilcoxon p=0.05 for BLCA, N=693 tumors p=0.09 for BRCA, N=244 tumors p=0.01 for HNSCC, N=284 tumors p=0.01 for LUAD). (d-g) HLA class I distribution by supertype and association with TMB. Individual HLA-I alleles were classified into discrete supertypes, based upon similar peptide-anchor-binding specificities. (d) TMB did not differ among different HLA-A supertypes (N=89 patients, Kruskal-Wallis p=0.45). (e) Similarly, there were no differences in TMB among different HLA-B supertypes (N=89 patients, Kruskal-Wallis p=0.45. (f) HLA-A supertype distribution for cases in cohort 1. (g) HLA-B supertype distribution for cases in cohort 1. Center values of the box plots in (d) and (e) indicate median values and error bars denote 95% confidence intervals. P values are based on two-sided testing.

Source data

Back to article page