`
`{MD I
`
`
`
`Volume 24 Number 5 2014
`
`Executive Editor Hillary E. Sussman
`
`Assistant Editor Laura E. DeMare
`
`Aravinda Chakravarti
`
`Editors
`
`Evan E. Eichler
`
`Richard A. Gibbs
`
`Tara Kulesa
`
`Eric Green
`
`Richard M. Myers
`
`William]. Pavan
`
`Administrative Assistant
`Peggy Calicchia
`
`Production Manager
`Linda Sussman
`
`Production Editor
`Marie Cotter
`
`Production Assistant
`
`Editorial Board
`
`I. Akey (Univ. of Washington)
`B. Andersson (Karolinska Institute)
`S.£. Antonarakis (Univ. of Geneva Medical School)
`B.E. Bernstein (Broad Institute)
`W. Blckmore (Medical Research Council)
`M. Boehnke (Univ. of Michigan)
`M. Brudno (Univ. of Toronto)
`M. Buiyk (Brigham 5t Women's Hospital and Harvard
`Medical School)
`L. Carrel (Penn State College of Medicine)
`N.P. Carter (Wellcome Trust Sanger Institute)
`G.A. Churchill (The lackson Laboratory)
`B. Cohen (Washington Univ. in St. Louis School of Medicine)
`G.M. Cooper (HudsonAlpha Institute for Biotechnology)
`G.E. Crawford (Duke Univ.)
`A. DI Rlenzo (Univ. of Chicago)
`M. Dunham (Univ. of Washington)
`P.|. Farnham (Univ. of California, Davis)
`5. Gabriel (Broad Institute)
`M.B. Cersteln (Yale Univ.)
`M. Hahn (lndiana Univ.)
`I.M. Hall (Univ. of Virginia)
`D.B. Iaffe (Broad Institute)
`L. Iin (Fudan University)
`S. Iones (BC Cancer Agency)
`1. Korbei (European Molecular Biology Laboratory)
`].D. Lleb (Univ. of North Carolina in Chapel Hill)
`E. Liu (Genome Institute of Singapore)
`|.R. Lupski (Baylor College of Medicine)
`T.F.C. Mackay (North Carolina State Univ.)
`P. Maiurnder (Indian Statistical Institute)
`
`K. Makova (Pennsylvania State Univ.)
`E. Mardls (Washington Univ. in St. Louis School of Medicine)
`E.H. Marguiles (National Human Genome Research Institute)
`G.T. Marti-I (Boston College)
`A.S. McCalllon (Johns Hopkins Univ. School of Medicine)
`M.L. Meyerson (Dana-Farber Cancer Institute and Harvard
`Medical School)
`A. Milosavlievic (Baylor College of Medicine)
`ll. MItra (Washington Univ. in St. Louis School of Medicine)
`|.V. Moran (Univ. of Michigan Medical School)
`M.A. Nobrega (Univ. of Chicago)
`|.P. Noonan (Yale Univ. School of Medicine)
`J. Parkhili (The Wellcome Trust Sanger Institute)
`W.R. Pearson (Univ. of Virginia)
`].H. Postlethwait (Univ. of Oregon)
`0. Rando (Univ. of Massachusetts Medical School)
`A. Regev (Broad Institute)
`D.A. Relrnan (Stanford Univ.)
`I. Rogers (Baylor College of Medicine)
`S.L. Salzberg (Johns Hopkins Univ. School of Medicine)
`P.C. Scacheri (Case Western Reserve Univ.)
`E. Segal (Weizmann Institute)
`1. Shendure (Univ. of Washington)
`].A. Stamatoyannopoulos (Univ. of Washington)
`M.R. Stratton (Wellcome Trust Sanger Institute)
`S. Tishkoff (Univ. of Pennsylvania)
`A.].M. Walhout (Univ. of Massachusetts Medical School)
`S.T. Warren (Emory Univ. School of Medicine)
`D.A. Wheeler (Baylor College of Medicine)
`R. Wolfe (University College, Dublin)
`K. Zhao (National Heart, Lung, and Blood Institute)
`
`
`
`Genome Research (ISSN 1088-9051) is published monthly by Cold Spring Harbor Laboratory Press, 500 Sunnyside Blvd, Woodbury, NY 11797-2924.
`
`
`Periodicals paid at Woodbury, NY and additional mailing offices. Canada Post International Publications Mail Product (Canadian distribution) Sales Agreement
`No. 1321846. POSTMASTER: Send address changes to Cold Spring Harbor Laboratory Press, 500 Sunnyside Boulevard, Woodbury, NY 11 797-2924.
`Subscriptions: Kathleen Cirone, Subscription Manager. Individual subscribers have a choice of "online only" or "print + online" subscriptions for this journal. Oniine
`
`
`only: 385; Print + Online: U.S., 5135; Canada and Mexico, $220; R.O.W., 5250 (includes airlift). For 2014 institutional pricing, visit http:f/genome.org/site/
`
`
`subscriptions/costdtl. Orders may be sent to Cold Spring Harbor Laboratory Press, Fulfillment Department, 500 Sunnyside Boulevard, Woodbury, New York
`
`
`117974924. Telephone: Continental U.S. and Canada 1-800-843—4388; all other locations 516-422-4100. Fax 516-422-4097. Personal subscriptions must be
`
`
`prepaid by personal check, credit card, or money order. Claims for missing issues must be received within four months of issue date.
`
`
`Advertising: Marcie Siconolfi, Advertising Manager, Cold Spring Harbor Laboratory Press,
`I Bungtown Rd, Cold Spring Harbor, New York 11724-2203. Phone:
`
`
`516-422—4010: fax: 516-422-4092.
`
`
`Information for Contributors: Author instructions are available at our website, http:/fwwwgenomeorg.
`
`
`Dnllne Manuscript Submission: http://submiLgenomeorg.
`
`
`Copyright Information: Authors of articles published in Genome Research retain copyright of the articles (except US Government employees) but grant Cold
`
`
`Spring Harbor Laboratory Press exclusive right to publish the articles for a period of six months following full-issue publication. This includes the rights to publish,
`
`
`reproduce, distribute, display, and store the article in all formats; to translate the article into other languages; to create adaptations, summaries, extracts, or
`
`
`derivations of the article; and to license others to do any or all of the above. Authors can reuse their articles in their work as long as Genome Research is credited as
`
`
`the place of original publication. Six months following publication, these articles are made available under a CC—BY-NC 4.0 Creative Commons license.
`
`
`Articles that carry the Open Access designation are available immediately under a CC-BY 4.0 0r CC-BY-NC 4.0 Creative Commons license. Detailed copyright terms
`
`
`are online at http:/lgenomecshlp.org/site/misc/terms.xhtm|. Permission information is online at http:/rgenomecshlp.org[site/miscfperrnissionsjanding.xhtml.
`
`
`Copyright (9) 2014 by Cold Spring Harbor Laboratory Press
`
`
`
`
` Volume 24 Issue 5 May 2014
`
`Perspecfive
`Can I be sued for that? Liability risk and the disclosure of clinically significant genetic
`research findings
`Amy L. McGuire, Bartha Maria Knoppers, Ma'n H. Zawati, and Ellen Wright Clayton
`
`Research
`
`PRDM? binding organizes hotspot nucleosomes and limits Holliday Junction migration
`Christopher L. Baker, Michael Walker, Shimpei Kaiita, Petko M. Petkov, and Kenneth Paigen
`
`Somatic mutations found in the healthy blood compartment of a llS-yr—oid woman
`demonstrate oligoclonal hematopoiesis
`Henne Holstege, Wayne Pfeiffer, Daoud Sie, Marc Hulsman, Thomas J. Nicholas,
`Clarence C. Lee, Tristen Ross, Jue Lin, Mark A. Miller, Bauke Ylstra, Hanne Meijers-Heijboer,
`Martijn H. Brugman, Frank J.T. Staal, Gert Holstege, Marcel |.T. Reinders, Timothy T. Harkins,
`Samuel Levy, and Erik A. Sisterrnans
`
`Neo—antigens predicted by tumor genome meta-analysis correlate with increased patient
`survival
`
`Scott D. Brown, Rene L. Warren, Ewan A. Gibb, Spencer D. Martin, John J. Spinelli,
`Brad H. Nelson, and Robert A. Holt
`
`Enhancer-targeted genome editing selectively blocks innate resistance to oncokinase
`inhibition
`
`Dan E. Webster, Brook Barajas, Rose T. Bussat, Karen J. Yan, Poornima H. Neela, Ross I. Flockhart,
`Joanna Kovalski, Ashley Zehnder, and Paul A. Khavari
`
`Recurrent epimutations activate gene body promoters in primary glioblastoma
`Raman P. Nagaraian, 30 Zhang, Robert J.A. Bell, Brett E. Johnson, Adam B. Olshen,
`Vasavi Sundaram, Daofeng Li, Ashley E. Graham, Aaron Diaz, Shaun D. Fouse, Ivan Smirnov,
`Jun Song, Pamela L. Paris, Ting Wang, and Joseph F. Costello
`
`Identifying mRNA sequence elements for target recognition by human Argonaute proteins
`Jingiing Li, TaeHyung Kim, Razvan Nutiu, Debashish Ray, Timothy R. Hughes,
`and Zhaolei Zhang
`
`Evolution of splicing regulatory networks in Drosophila
`C. Joel McManus, Joseph D. Coolon, Jodi Eipper—Mains, Patricia J. Wittkopp,
`and Brenton R. Graveley
`
`Tempo and mode of regulatory evolution in Drosophila
`Joseph D. Coolon, C. Joel McManus, Kraig R. Stevenson, Brenton R. Graveley,
`and Patricia J. Wittkopp
`
`719
`
`724
`
`733°A
`
`7430A
`
`75]
`
`761
`
`775
`
`786
`
`7970"
`
`(continued)
`
`
`
`High-resolution mapping definES the cooperative architecture of Polycomb response elements
`Guillermo A. Orsi, Sivakanthan Kasinathan, Kelly T. Hughes, Sarah Saminadin—Peter,
`Steven Henikoff, and Kami Ahmad
`
`Genome methylation in D. melanogasrer is found at specific short motifs and is independent
`of DNMT2 activity
`Sachiko Takayama, Joseph Dhahbi, Adam Roberts, Guanxiong Mao, Seok—Jin Heo, Lior Pachter,
`David l.l<. Martin, and Dario Boffelli
`
`Widespread and frequent horizontal transfers of transposable elements in plants
`Moaine E| Baidouri, Marie-Christine Carpentier, Richard Cooke, Dongying Ciao, Eric Lasserre,
`Christel Llauro, Marie Mirouze, Nathalie Picault, Scott A. Jackson, and Olivier Panaud
`
`Predicting the virulence of MRSA from its genome sequence
`Maisem Laabei, Mario Recker, Justine K. Rudkin, Mona Aldeljawi, Zeynep Gulay,
`Tim J. Sloan, Paul Williams, Jennifer L. Endres, Kenneth W. Bayles, Paul D. Fey,
`Viiaya Kumar Yajjala, Todd Widhelm, Erica Hawkins, Katie Lewis, Sara Pan‘ett, Lucy Scowen,
`Sharon J. Peacock, Matthew Holden, Daniel Wilson, Timothy D. Read, Jean van den Elsen,
`Nicholas K. Priest, Edward J. Feil, Laurence D. Hurst, Elisabet Josefsson, and Ruth C. Massey
`
`809
`
`82]
`
`3310A
`
`8390A
`
`A genomic portrait of the genetic architecture and regulatory impact of microRNA
`expression in response to infection
`
`850
`
`Katherine J. Siddle, Matthieu Deschamps, Ludovic Tailleux, Yohann Nédélec, Julien Pothiichet,
`Geanncarlo Lugo-Villarino, Valentina Libri, Brigitte Gicquel, Olivier Neyrolles, Guillaume Laval,
`Etienne Patin, Luis B. Barreiro, and Lluis Quintana-Murci
`
`Methods
`
`General approach for in vivo recovery of cell type-specific effector gene sets
`Julius C. Barsi, Qiang Tu, and Eric H. Davidson
`
`ISMARA: automated modeling of genomic signals as a democracy of regulatory motifs
`Piotr J. Balwierz, Mikhail Pachkov, Phil Arnold, Andreas J. Gruber, Mihaela Zavolan,
`and Erik van Nimwegen
`
`860
`
`8690A
`
`0AOpen Access pa per
`
`Cover Horizontal transfer of genetic material has been demonstrated on a small scale
`for several organisms. In this issue, this phenomenon is demonstrated to be widespread
`in the plant kingdom. On the cover is a Pointillist-Iike view of the horizontal transfer of
`transposable elements between palm (top left) and grape (bottom right). The two circles
`depict the respective DNA molecules and leaf colors indicate the chimeric nature of
`plants. Horizontal branches on the "tree of life“ at the bottom left illustrate frequent
`transfer during the evolution of plants. (Cover illustration by Abdelkebir El Baidouri.
`[For details, see El Baidouri et al., pp. 831—838.])
`
`FSCmiscw MIX
`
`Paper from
`responsible sources
`FSC“ 001294?
`
`
`
`This material may be protected by Copyright law (Title 17 U.S. Code)
`
`Research ,,
`
`Neo—antigens predicted by tumor genome meta-analysis
`correlate with increased patient survival
`
`Scott D. Brown}2 Rene L. Warren,1 Ewan A. Gibb,1'3 Spencer D. Martin,1'3'4
`John J. Spinelli,5'6 Brad H. Nelson,3'4'7 and Robert A. Holt1'3'8'9
`'Canada’s Michael Smith Genome Sciences Centre, BC Cancer Agency, Vancouver, British Columbia V52 1L3, Canada; ZGenome
`Science and Technology Program, University of British Columbia, Vancouver, British Columbia V6 T 124, Canada; 3Department of
`Medical Genetics, University of British Columbia, Vancouver, British Columbia V6T 124, Canada; 4Deeley Research Centre, BC Cancer
`Agency, Victoria, British Columbia V8R 6V5, Canada,- 5Cancer Control Research Program, BC Cancer Agency, Vancouver, British
`Columbia V52 1L3, Canada; “School of Population and Public Health, University of British Columbia, Vancouver, British Columbia
`V6 T 124, Canada; 7Department of Biochemistry and Microbiology, University of Victoria, Victoria, British Columbia VBP 5 C2,
`Canada,- SDepartment of Molecular Biology and Biochemistry, Simon Fraser University, Burnaby, British Columbia V5A 156, Canada
`
`Somatic missense mutations can initiate tumorogenesis and, conversely, anti‘tumor cytotoxic T cell (CTL) responses.
`Tumor genome analysis has revealed extreme heterogeneity among tumor missense mutation profiles, but their relevance
`to tumor immunology and patient outcomes has awaited comprehensive evaluation. Here, for 5l5 patients from six tumor
`sites, we used RNA-seq data from The Cancer Genome Atlas to identify mutations that are predicted to be immunogenlc
`in that they yielded mutational epitopes presented by the MHC proteins encoded by each patient’s autologous HLA—A
`alleles. Mutational epitopes were associated with increased patient survival. Moreover, the corresponding tumors had
`higher CTL content, inferred from CDBA gene expression, and elevated expression of the CTL exhaustion markers PDCDl
`and CTL“. Mutational epitopes Were very scarce in tumors without evidence of CT1. infiltration. These findings suggest
`that the abundance of predicted immunogenic mutations may be useful for identifying patients likely to benefit from
`checkpoint blockade and related immunotherapies.
`
`{Supplemental material is available for this article]
`
`The accumulation of somatic mutations underlies the initiation
`
`and progression of most cancers by conferring upon tumor cells
`unrestricted proliferative capacity (Hanahan and Weinberg 2011).
`The analysis of cancer genomes has revealed that tumor muta-
`tional landscapes (Vogelstein et al. 2013) are extremely variable
`among patients, among different tumors from the same patient,
`and even among the different regions of a single tumor (Gerlinger
`et al. 2012). There is a need for personalized strategies for cancer
`therapy that are compatible with mutational heterogeneity, and in
`this regard, immune interventions that aim to initiate or enhance
`anti-tumor immune responses hold much promise. Therapeutic
`antibodies and chimeric antigen receptor (CAR) technologies have
`shown anti-cancer efficacy (Fox et al. 2011), but such antibody-
`based approaches are limited to cell surface target antigens (Slamon
`et a1. 2001: Coiffier et a1. 2002: Yang et a1. 2003; Cunningham
`et al. 2004: Kalos et al. 2011). In contrast, most tumor mutations
`are point mutations in genes encoding intracellular proteins. Short
`peptide fragments of these proteins, after intracellular processing
`and presentation at the cell surface as MHC ligands, can elicit Tcell
`immunoreactivity. Further,
`the presence of tumor infiltrating
`lymphocytes (TIL). in particular, CD8’ T cells, has been associated
`with increased survival (Sato et al. 2005: Nelson 2008; Obie et al.
`2009; Yamada et al. 2010; Gooden et al. 2011: Hwang et al. 2012),
`suggesting that the adaptive immune system can mount protective
`anti-tumor responses in many cancer patients (Kim et al. 2007; Fox
`
`at al. 2011). The antigen specificities of tumor~infiltrating T cells
`remain almost completely undefined (Andersen el al. 2012), but
`there are numerous examples of cytotoxic T cells recognizing sin-
`gle amino acid coding changes originating from somatic tumor
`mutations (Lennerz et al. 2005: Matsushita et al. 2012; Heemsketk
`et a1. 2013; Lu et al. 2013; Robbins et al. 2013; van Rooii et a1. 2013:
`Wick et al. 2014). Thus, the notion that tumor mutations are reser-
`voirs of exploitable hen-antigens remains compelling (Heemskerk
`et a1. 2013). For a mutation to be recognized by CD8' T cells, the
`mutant peptide must be presented by MHC I molecules on the
`surface of the tumor cell. The ability of a peptide to bind a given
`MHC l molecule with sufficient affinity for the peptide-MHC
`complex to be stabilized at the cell surface is the single most lim-
`iting step in antigen presentation and T cell activation (Yewdell
`and Bennink 1999]. Recently, several algorithms have been de—
`veloped that can predict which peptides will bind to given MHC
`molecules (Nielsen et al. 2003,- Bui et a1. 2005; Peters and Sette
`2005; Vita et al. 2010,- Lundegaard ct al. 201 1}, thereby providing
`guidance into which mutations are immunogenic.
`The Cancer Genome Atlas (TCGA) (http:/‘lcancergenomenih.
`govl) is an initiative of the National Institutes of Health that has
`created a comprehensive catalog of somatic tumor mutations
`identified using deep sequencing. As a member of The Cancer
`Genome Atlas Research Network, our center has generated exten-
`sive tumor RNA-seq data. Here, we have used public TCGA RNA-seq
`data to explore the T cell immunoreactivity of somatic missense
`
`’Comspondlng author
`E-rnall rholtflbcgse.“
`© 2014 Brown et al. This article, published in Genome Research, is available
`Article published online before print. Article, supplemental material, and publi-
`under a Creative Commons License (Attribution 4.0 international), as described
`cation date are athttp://www.genome.orglcgi/doi/'l0.110119r.155985.113.
`Freely available online through the Genome Research Open Access option,
`at httpzllcreativecornmons.otgl|icensesibyi4.0.
`
`
`24:743—750 Published by Cold Spring Harbor Laboratory Press;
`
`lSSN 1088-90‘ill14: wwwgennmeorg
`
`Genome Research
`wwwgeoomeorg
`
`743
`
`
`
`Brown et al.
`
`mutations across six tumor sites. This type of analysis is challenged
`not only by large numbers of mutations unique to individual pa-
`tients, but also by the complexity of personalized antigen pre-
`sentation by MHC arising from the extreme HLA allelic diversity in
`the outbred human population. Previous studies have explored the
`potential immunogenicity of tumor mutations (Segal et al. 2008;
`Warren and Holt 2010; Khalili et al. 2012}, but these have been
`hampered by small sample size and the inability to specify autoi-
`ogous HLA restriction. Recently, we described a method of HLA
`calling from RNA-seq data that shows high sensitivity and speci-
`ficity (Warren et al. 2012). Here, we have obtained matched tumor
`mutational profiles and HIA-A genotypes from TCGA subjects and
`used these data to predict patient-specific mutational epitope
`profiles. The evaluation of these data together with RNA-seq-de-
`rived markers of T cell infiltration and overall patient survival
`provides the first comprehensive view of the landscape of poten-
`tially immunogenic mutations in solid tumors.
`
`Results
`
`Summary of available data
`
`Raw TCGA RNA-seq data plus clinical metadata and complete
`profiles of sequence-verified missense mutations were obtained
`with permission from the Cancer Genomics Hub (https:I/cghub.
`ucsc.edu). Our analysis covers six tumor sites, including colon and
`rectum (combined as colorectal), ovary, breast, brain, kidney, and
`lung. These were the only tumor sites with complete and non-
`embargoed data at the time of this study. The RNA-seq data were
`first processed using HLAminer (Warren et al. 2012) to predict, at
`four-digit resolution, the two HLA-A alleles carried by each subject.
`Data from 515 patients with unambiguous HLA-A calls were pro-
`cessed further. The distribution of missense mutation counts
`
`across patients with different tumor types is shown in Figure 1. For
`each of the 22,758 total missense mutations, we evaluated binding
`
`
`
`Numberofmutationsperpatient
`
`100150200250
`
`50
`
`
`
`of all possible 8— to Il-mer mutant peptide variants to autologous
`HLA-A encoded MHC proteins using the Immune Epitope Database
`(IEDB) T Cell Epitope-MHC Binding Prediction Tool (Vita et al.
`2010) (http:/lwwaedborgl). We EOCused our analysis on HlA-A
`alleles because (1) MHC I proteins (encoded by HLA.A, -B, and -C
`genes) present antigens to CD8+ cytotoxic T cells, which are the
`subset of T cells most strongly linked to patient survival, and (2)
`HLA-A alleles of MHC I yield the most accurate peptide binding
`affinity predictions by IEDB and most other algorithms due to the
`abundance of HLA-A-specific training data (Hoof et al. 2009). All
`mutational data, RNA-seq derived HlA—A calls, IEDB epitope pre-
`dictions, RNA-seq-derived' gene expression values, and clinical
`metadata were compiled in a MySQL database for analysis.
`
`CDEA expression is associated with survival
`
`We first asked if we could reproduce the known association be-
`tween increased numbers of tumor-infiltrating CD8‘ T cells and
`increased overall survival (Sato et al. 2005; Nelson 2008; Obie et al.
`2009; Yamada et al. 2010; Gooden et al. 2011; Hwang et al. 2012).
`CD8‘ TIL levels are usually measured by immunohistologica]
`staining. To interrogate RNA—seq data, we used the expression of
`CDSA (one component of the C08 dimer) as a surrogate for CD8*
`TIL levels. We observed significantly higher overall survival for pa-
`tients with high CD8A expression than for those patients with low
`CD8A expression (HR: 0.71, 95% CI = 0.53 to 0.94.1”: 1.7 x 10—2)
`(Fig. 2A). Likewise, the data recapitulated the known association
`between high HLA-A expression and improved overall survival
`(HR = 0.59, 95% CI = 0.44 to 0.81, P = 8.6 X 10’ 4‘) (Fig. 28; Concha
`et al. 1991; Ogino et al. 2006: Kitamura et al. 2007; Han et al. 2008;
`Biien et al. 2010). Based on these positive findings with established
`T cell and MHC markers, we proceeded to evaluate candidate
`peptide epitopes, which represent the third molecular compo-
`nent required forTcell recognition and destruction of target cells.
`
`The abundance of tumor missense mutations is not associated
`with suthrai
`
`Initially, we asked if there is a relationship between overall muta-
`tion count and CBS“ TIL. Ranking patients by decreasing CDSA
`expression and displaying the mutation count for each patient’s
`tumor revealed a skewed distribution whereby tumors with low
`CDSA expression had sparse mutations and tumors with high mu-
`tation counts were among those with elevated CDBA expression
`(Fig. 3A). Tumors with above median CDBA expression contained
`73.6% of the total mutations (P = 2.0 X 10"6 by iterative randomi-
`zation and resampling as described in Methods). However, there was
`no association between total mutation count and overall survival
`
`(HR = 0.91, 95% CI: 0.68 to 1.23, P: 5.5 x 10") (Fig. 33).
`
`Tumor missense mutations that have predicted
`immunoreactivity are associated with increased survival
`
`Lung
`I1 = 34
`
`Ovary
`n = 218
`
`Breast
`n = 24
`
`Colorectal
`n = 170
`
`Brain
`n = 16
`
`ifldney
`n = 53
`
`We reasoned that missense mutations yielding peptides with poor
`MHC l binding would be immunologically silent and hence likely
`to obscure any association between missense mutations, anti-tu-
`Flgure 1. Boxpiots showing the number of mutations per patient for
`mor immunoreactivity, and survival. To address this, we repeated
`each cancer type. The y—axis is cut off at 250 mutations for better visual.
`the above analysis focusing on those mutations that were most
`ization of the majority of the data. The dark horizontal bar shows the
`median, whereas the box encompasses the interquartile range (middle
`likely to be immunogenic by several criteria, including (1) the
`50% of the data). Whiskers reach the farthest data point that is within
`expression of the gene in the tumor bearing the mutation was
`1.5x the interquartiie range from the nearest box edge (quartile). Box
`above the median expression level of that same gene in all tumors,
`width is proportional to the sample size (lung: 34, ovary: 218, breast: 24,
`colorectal: 170, brain: 16, kidney: 53).
`(2) HlA-A expression in the tumor bearing the mutation was above
`
`
`744 Genome Research
`wwwgenomeorg
`
`
`
`lmmunogenic tumor mutations correlate with survival
`
`
`
`
`
`SurvivalProbability
`
`
`
`
`
`
`
`
`
`SurvivalProbability
`
`0
`
`1000
`
`2000
`
`3000
`
`Days
`
`4000
`
`5000
`
`0
`
`"300
`
`2000
`
`3000
`
`4000
`
`5000
`
`Figure 2. Overall survival for patients based on CDEA or HLA-A expression. Kaplan-Meier curves were constructed to look at the difference in survival of patients
`(n = 512) with low and high expression levels of (IDEA (A) or HLA-A (3). Patients were split into two groups based on the median expression value. Patients with
`high expression showed increased survival compared to those with low expression of either (A) CDBA (HR 2 0.71, 95% CI = 0.53 to 0.94, P = 1.7 X 104) or
`(B) HLA-A (HR = 0.59, 95% CI = 0.4-4 to 0.81, P = 8.6 X 10"). Tick marks on the gmph denote the last time survival status was known for living patients.
`
`the HiA—A expression requirement from the definition of a predicted
`immunogenic mutation, we fit a model including all prognostic
`factors to the subset of patients with high (above median) tumor
`HLA-A expression. Within this subset of patients, we observed that
`patients with at least one predicted immunogenic mutation had
`a significantly lower relative risk of death than those without (HR =
`0.44, 95% CI = 0.22 to 0.88, P: 2.0 X 10’2). Evaluating the reciprocal
`group of patients with low (below median) HLA-A expression, where
`the potential of immunogenic mutations to elicit bona fide anti-
`tumor responses is expected to be curtailed, there was no significant
`association between the presence of predicted immunogenic muta-
`tions and survival (HR = 1.30, 95% CI = 0.83 to 2.04, P: 2.6 X 10").
`The results from all survival analyses are summarized in Table 1.
`
`the median expression of HLA-A in all tumors, and (3) the pre—
`dicted autologous HLA-A binding affinity of the best scoring pep-
`tide containing a given mutation had an leo value of 500 nM or
`less. This value has been estimated, experimentally, to be the af-
`finity necessary for an epitope to elicit an immune response (Sette
`et al. 1994). Applying these filters, the predicted immunogenic
`mutation count was zero in 334 patients. The remaining 181 pa-
`tients had predicted immunogenic mutation counts ranging from
`1 to 147, with a median of 3. The predicted immunogenic mutation
`count showed a strong relationship with tumor CDSA expression,
`where tumors with higher numbers of such mutations had higher
`CDSA expression (Fig. 3C). Of all predicted immunogenic mutations,
`84. 7% were in tumors with above median CDBA expression (P = 1.0 X
`10*). We did not see any relationship between predicted immuno-
`genic mutation count and CD4 expression by tumors (P = 6.9 x 10’ 1)
`(Supplemental Fig. 1), consistent with the fact that we had assessed
`epitopes presented by MHC class I, which is recognized exclusively
`by CD8+ T cells. Interestingly, patients with tumors containing at
`least one predicted immunogenic mutation showed markedly in-
`anti-tumor T cell responses (Schneider et at. 2006; Blank and
`creased overall survival compared to those without predicted im-
`Mackensen 2007). Blockade of these inhibitory receptors by tar-
`munogenic mutations (HR = 0.53, 95% CI = 0.36 to 0.80, P = 2.1 X
`geted monoclonal antibodies can disinhibit anti-tumor immunity
`10—3) (Fig. 3D). To further examine this association, we fit a model
`and improve clinical outcomes (Hodi et al. 2003, 2008, 2010,-
`including all available prognostic factors (age, gender, cancer type,
`Hamanishi et al. 2007; Mansh 201 1; Brahmer et al. 2012; Topalian
`and tumor stage), as well as predicted immunogenic mutations. This
`et al. 2012; ). Given that many patients in the current study had
`model also showed significantly improved overall survival for patients
`clinically significant cancer despite having predicted immuno-
`with predicted inununogenic mutations relative to those without
`genic mutations and CD8+ TIL, we asked if there was an association
`(HR = 0.50, 95% CI = 0.31 to 0.80, P: 3.9 x 10“), indicating that the
`between immunogenic mutation load and expression of PDCDI or
`effect of predicted immunogenic mutations was independent of the
`CTLA4. We found that patients with higher numbers of predicted
`other prognostic factors. Fitting a model which contained an in-
`immunogenic mutations had increased expression of not only CD821
`teraction between cancer type and predicted immunogenic mutations
`but also PDCDi and C1144. Displaying these values in a three-way
`did not yield a significant result (P = 9.2 X 10”), indicating that the
`hive plot (Krzywinski et al. 2012) highlights the association between
`prognostic effect is not limited to a specific cancer diagnosis.
`these T cell markers and immunogenic mutation load (Fig. 4).
`Given that tumor HLA-A expression alone is a known in-
`Significance was assessed by iterative randomization and resam-
`dicator of favorable patient survival
`(Fig. 23), we asked if the
`pling (as described in Methods). Of all tumors with predicted im-
`number of predicted immunogenic mutations provides additional
`munogenic mutations, 45.9% had above median expression of all
`three of PDCDI, CTLA4, and CD8A (P = 1.0 x 10-6).
`predictive value independent of HLA-A expression. After removing
`
`
`Predicted immunogenic mutation counts correlate
`with the expression of T cell exhaustion markers
`PDCDl and CTLA4 are T cell surface molecules that can inhibit
`
`Genome Research
`www.genome.org
`
`745
`
`
`
`Brown et al.
`
`
`
`
`
`
`
`
`
`100—i
`
`> C
`
`
`
`DflAFPKM
`
`
`
`TotalMutationCounl
`
`O C
`
`
`
`UBAFPKM
`
`
`
`
`
`lmmunogenicMutationCount
`
`Survival
`
`Probability — High Mutation Count
`
`0.0
`
`
`
`. Low Mutation Count
`.
`—.—-——1—‘—'—T—'—'—‘——l——
`0
`1000
`2000
`3000
`4000
`5000
`
`DIYS
`
`
`
`
`
`._
`H.
`i
`
`.
`
`‘i.
`
`gE
`.EE
`Em
`.2
`
`.3
`
`—-.
`V‘.
`° ‘
`
`9| co
`
`a
`CS
`
`\
`I
`
`D
`
`77*
`
`1
`ztimmunogenkMutatlons
`fllmmunngenchulabom .
`
`r
`
`T000
`
`2000
`
`3000
`
`4000
`
`5000
`
`ISO-J
`Flgure 3. The total number of mutations in tumors is not associated with survival, while the number of predicted immunogenic mutations is associated
`with survival. (AC) A "skew plot" was made for all patients (n = 515), ordering patients along the x—axis according to their CDSA expression. Each patient's
`C081! expression was plotted above the x»axis, and total mutation count (A) or predicted immunogenic mutation count (C) was plotted below the x—axis.
`73.5% of the total mutation count belonged topatients with above median CDSA expression (P = 2.0 x 10“), and 84.7% of the total predicted
`immunogenic mutation count belonged to patients with above median (DEA expression (P: l .0 x 10"). (3,0) Kapian-Meier curves were constructed to
`look at the difference in survival between patients with low versus high numbers of mutations. Patients (n = 468) were split into two groups based on the
`median mutation count. There was no difference in survival between the two groups when stratifying on total mutation count (3) (HR = 0.91, 95% CI =
`0.68 to 1.23, P = 5.5 x 10"), but there was a statistically significant difference between the two groups when stratifying on predicted immunogenic
`mutation count (D) (HR = 0.53, 95% CI = 0.36 to 0.80, P = 2.1 x 10’3). Tick marks on the Kaplan-Meier graphs denote the last time survival status was
`known for living patients.
`
`Discussion
`
`the most frequent type of oncogenic mutation, which raises the
`question of whether missense mutations also underlie tumor im—
`munoreactivity. Exome analysis in mice has revealed specific
`The adaptive immune system opposes tumor development, and
`missense mutations that encode MHC class I presented mutational
`the elicitation of immunogenic cell death is a key component of
`epitopes that are capable of eliciting T cell-mediated tumor re-
`both targeted immunotherapies and conventional treatment mo-
`jection (Castle et a1. 2012,- Matsushita et a]. 2012). Moreover, hu-
`dalities including radiation and chemotherapy (Kroerner et a1.
`man tumor exome sequencing studies have identified mutational
`2013). There is a robust association between T ceil infiltration of
`epitopes recognized by autologous CD8+ TIL (Heemskerk et al.
`solid tumors and favorable patient outcomes. Missense variants are
`
`
`746 Genome Research
`www.genome.org
`
`
`
`lmmunogenic tumor mutations correlate with survival
`
`Table 1. Summary of survival analysis
` Predictor HR 95% Cl P—value
`
`
`
`
`CDEA expression
`HUI-A expression
`Total mutations
`Predicted immunogenic
`mutationsa
`Predicted immunogenic
`mutations, low HlJi-J‘ia
`2.0 x 1072'
`0.22—0.88
`0.44
`Predicted immunogenic,
`high HLA-A‘
`
`0.5 3—0.94
`0.44—0.81
`0.68—1.23
`0.31 —0.80
`
`1.7 X 10 ' 2*
`8.6 X 10' 4“
`5.5 x 10"
`3.9 X 10’3“
`
`0.83—2.04
`
`2.6 x 10"
`
`0.71
`0.59
`0.91
`0.50
`
`1.30
`
`(‘) P-values < 0.05. (") P—values < 0.005.
`“Analysis that accounted for variation from known prognostic factors,
`
`infiltration by activated CD8” T cells (Dolcetti et al. 1999}: thus,
`a mutator tumor phenotype may, in general, enhance immuno—
`reactivity. Other classes of potentially immunogenic mutations
`require exploration, such as gene fusions resulting from genomic
`rearrangements. Instances of tumors with high CD8+ TIL but few
`immunogenic mutations may also be due to immune editing
`(Matsushita et al. 2012; Vesely and Schreiber 2013). Specifically,
`tumor cells bearing highly immunogenic mutations may have been
`selectively eliminated by T cells, resulting in accumulation of CD8+
`TIL but fewer irnmunogenic mutations remaining to be detected.
`The results of the present study have clinical implications. We
`have shown that patients with tumors bearing missense mutations
`predicted to be immunogenic have a survival advantage (Fig. 3D).
`These tumors also show evidence of higher CD8+ TIL, which sug-
`gests that a number of these mutations might be immunoreactive.
`The existence of these mutations is encouraging because, in prin-
`ciple. they could be leveraged by persona