Item

Abstract

We analyze RNA and whole exome sequencing data of tumors from 8,705
donors spanning a range of 32 cancer types. Our study focuses on the
analysis of specific alternative splicing (AS) events involving a small number
of exons from RNA-seq data. We report results from comprehensive analyses
of a) the underlying genetic changes leading to splicing variability in tumors,
b) quantitative and qualitative changes of AS in tumors, and c) the extent
to which splicing aberrations can be exploited for immunotherapy. To our
knowledge, this study presents the first comprehensive analysis of known as
well as novel alternative splicing events across all suitable samples of The
Cancer Genome Atlas (TCGA). Depending on cancer type, we find an up to
40% increase in AS in tumor samples relative to normal and find on average
more than 900 tumor-specific exon-exon-junction.We uncover strong splicing
signatures for individual cancer types and subtypes, often overpowering the
respective tissue-specific effects. Further, we combine the splicing phenotypes
with variants obtained from exome sequencing data for a genome-wide splic-
ing association analysis. This is the largest reported splicing quantitative trait
loci (sQTL) study with respect to number of donors thus far. Here, we focus
on variants that have been shown to occur as somatic in some individuals
but may also occur in the germline genome in others. For the first time, the
available data provides sufficient statistical power to detect trans-sQTL that
were difficult to detect before. Aside from confirming known sQTL-variants in
splicing factors U2AF1 and SF3B1, we also detect novel trans-sQTL in IDH1,
TADA1 and PPP2R1A.Finally, our study is the first to comprehensively esti-
mate to which extent AS in tumors leads to new RNA transcripts that are trans-
lated into tumor-specific peptides, that are potential targets for immunotherapy.
Integrating data from TCGA and GTEx, we identify tumor-specific events. We
use CPTAC protein mass spectra of two tumor types to show that the resulting
mRNAs are indeed translated into tumor-specific peptides. From all peptides
that are predicted to be MHC-I binders, we are able to confirm on average 1.7
splicing derived neo-epitopes per sample – an almost 3-fold increase over the
number of neo-epitopes predicted with classic approaches taking into account
only SNVs (≈0.6). To our knowledge this is the first comprehensive analysis of
this type.