Proteogenomic approach to discover cancer aberrant peptides and antibody peptides using large ...

184 Pages·2017·6.06 MB·English

UC San Diego
UC San Diego Electronic Theses and Dissertations
Title
Proteogenomic approach to discover cancer aberrant peptides and antibody peptides using
large-scale next-generation sequencing data
Permalink
https://escholarship.org/uc/item/5kz1z434
Author
Cha, Seong Won
Publication Date
2017

Peer reviewed|Thesis/dissertation
eScholarship.org Powered by the California Digital Library
University of California
UNIVERSITYOFCALIFORNIA,SANDIEGO
Proteogenomicapproachtodiscovercanceraberrantpeptidesandantibodypeptidesusing
large-scalenext-generationsequencingdata
Adissertationsubmittedinpartialsatisfactionofthe
requirementsforthedegree
DoctorofPhilosophy
in
ElectricalEngineering(CommunicationTheoryandSystems)
by
SeongWonCha
Committeeincharge:
ProfessorVineetBafna,Chair
ProfessorDrewHall,Co-Chair
ProfessorCKCheng
ProfessorWilliamHodgkiss
ProfessorPavelAPevzner
2017
Copyright
SeongWonCha,2017
Allrightsreserved.
The dissertation of Seong Won Cha is approved, and it is
acceptable in qualityand form for publication onmicrofilm
andelectronically:
Co-Chair
Chair
UniversityofCalifornia,SanDiego
2017
iii
DEDICATION
I dedicate this dissertation to my beloved parents, Myungsik Cha and Jinjune Kim,
for all their love, patience, kindness, and support.
iv
TABLEOFCONTENTS
SignaturePage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
TableofContents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
ListofFigures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
ListofTables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviii
AbstractoftheDissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix
Chapter1 ProteogenomicdatabaseconstructiondrivenfromlargescaleRNA-seqdata 1
1.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 METHOD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 ConvertingsplicegraphstructuretoaFASTAfileformat . . 6
1.2.2 Datasetsandexperimentalprocedure . . . . . . . . . . . . 10
1.3 RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Chapter2 Proteogenomicstrategiesforidentificationofaberrantcancerpeptidesusing
large-scaleNextGenerationSequencingdata . . . . . . . . . . . . . . . 21
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.1 DatabasecreationfromRNA-seqdata . . . . . . . . . . . . 25
2.2.2 DatabaseSearchDetails . . . . . . . . . . . . . . . . . . . 27
2.2.3 FDRbasederrorcontrolstrategies . . . . . . . . . . . . . . 28
2.2.4 SamplepreparationandLC-MS/MSanalysis . . . . . . . . 30
2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Chapter3 Integrativeproteogenomicpipelineforidentificationofmutatedpeptidesand
immunoglobulingenerearrangements,anditsapplicationtocoloncancer 45
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
v
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Chapter4 Theantibodyrepertoireofcolorectalcancer . . . . . . . . . . . . . . . . 62
4.1 Abbreviationspage . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.3 ExperimentalProcedures . . . . . . . . . . . . . . . . . . . . . . . 65
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.5 Discussionandfuturestudy . . . . . . . . . . . . . . . . . . . . . . 81
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
AppendixA Appendix: Proteogenomic strategies for identification of aberrant cancer
peptidesusinglarge-scaleNextGenerationSequencingdata . . . . . . . 89
A.1 Comparisonwithothergenepredictionmethods . . . . . . . . . . . 90
A.2 CalculationofsplitmappedcoordinatesfromCIGAR
stringinSAMfileformat . . . . . . . . . . . . . . . . . . . . . . . 91
A.3 DetailedRNA-seqmethods . . . . . . . . . . . . . . . . . . . . . . 91
A.4 Comparison of spectra dataset used in this study with Merrihew et al.
(2008)[73] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
A.5 Proof of correctness and completeness in applying Rule1, Rule2, and
Rule3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
A.6 ProofofcorrectnessandcompletenessinDFSalgorithmimplementation
ofRule1,Rule2,andRule3 . . . . . . . . . . . . . . . . . . . . . . 101
A.6.1 Rule1: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
A.6.2 Rule2: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
A.6.3 Rule3: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
AppendixB Appendix: Proteogenomic strategies for identification of aberrant cancer
peptidesusinglarge-scaleNextGenerationSequencingdata . . . . . . . 104
AppendixC Appendix: Integrativeproteogenomicpipelineforidentificationofmutated
peptides and immunoglobulin gene rearrangements, and its application to
coloncancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
AppendixD Appendix: Theantibodyrepertoireofcolorectalcancer . . . . . . . . . . 126
D.1 Supplementalmethod . . . . . . . . . . . . . . . . . . . . . . . . . 126
D.1.1 AntibodystructureandIMGTreference . . . . . . . . . . . 126
D.1.2 NodegroupingmethodforSdBgraph . . . . . . . . . . . . 127
D.1.3 MathematicalcomparisonbetweentheSdBanddBgraph . 128
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
vi
LISTOFFIGURES
Figure1.1: (a)GivenRNA-seqread,findoverlappingregionswiththeexistingsplicegraph.
(b) Split and add nodes. (r , node s is split into nodes u and u , and node
1 1 1 2
u is added.) (c) Assign edges for each spliced-read. (d) Revisit each pair of
3
contiguousnodes. The nodesare mergedif thereis noedge attheboundaries.
(Nodesu andu aremerged,whilee isaddedbetweenu andu .) . . . . 7
1 2 5 2 3
Figure1.2: (a) By traversing the graph using a depth first search(DFS), we generate a
sequence from the first visited start to end node path. (b) While traversing
in DFS, when we encounter an outgoing edge that is already visited, only
maintainalengthL−1suffix. (c)WhiletraversinginDFS,whenweencounter
an incoming edge that is already visited, only maintain a length L−1 prefix.
(d) For a pair of sequences(paths) with a prefix-suffix match, combine two
sequences. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Figure1.3: (a)Growthof thedatabase file size(Bytes)whileincorporating moreRNA-seq
data. (b) Increase in the percentage of covered splice junctions compared to
RefSeq. (c)Increaseinthenumberofsplicejunctionsexpressedinsplicegraph
databasewhichdoesnotexistinRefSeq. . . . . . . . . . . . . . . . . . . 14
Figure1.4: (a) Shows a novel gene area where two peptides are identified in a non-
genomic region. (b) Two peptides with alternative splice junctions. Peptide
T.LNVNGQE:IVYSMENEK.Lissupportedby13splitmappedRNA-seqreads,
and R.EIKK:QHTSFQVSGPKEEIVYSMENEK.L is supported by 40 reads.
(c) Peptide ‘TIVFTVPLSQCQMVSPMISK.E’ matches in a different frame
compared to the gene eef-2. Two neighboring peptides, ‘R.FIEPIEDIPSG
NIAGLVGVDQYL:S.R’,and‘G.HVFEESQVTGTPMFVV:R.L’areidentified
with1bpdeletion,thatallowfortheframe-shifttooccur. . . . . . . . . . . 17
Figure2.1: Number of peptide identifications in 439,858 spectra collected from a single
sample(sampleid: TCGA-24-1467)usingdifferentFDRbasederrorcontrol
strategies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Figure2.2: Overlapbetween novel identifications fromunified andsingle sample database.
41
Figure2.3: Alignmentofidentifiedspectraofmutatedpeptides. . . . . . . . . . . . . 42
Figure2.4: Alignmentofidentifiedspectraofnoveljunctionpeptides. . . . . . . . . . 43
Figure2.5: Alignmentofidentifiedspectraofmutatedpeptides. . . . . . . . . . . . . 44
Figure2.6: Insertionsandsubstitutionsarerepresentedasadditionalnodeandedgeshaving
negativecoordinatevalues. Deletionsarerepresentedsameassplicejunctions
withactualDNAcoordinates. . . . . . . . . . . . . . . . . . . . . . . . . 44
Figure3.1: Illustrationofproteogenomicdatabaseconstructionforimmunoglobulinpeptide
identifications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
vii
Figure3.2: (a) Comparison of aberrant peptide identifications against previous findings
using multi-stage FDR (b)Comparison of overlapping aberrant peptide iden-
tifications using combined FDR. Our proteogenomic database was created
from raw RNA-seq alignments from TCGA repository and database used in
Zhangetal.[138]iscreatedfromSNVinformationsreportedbydbSNP[105],
COSMIC[35],andTCGAsomaticmutationcalls[75]. . . . . . . . . . . . 59
Figure3.3: (a)GenescontainingmostfrequentsomaticmutationsreportedbytheTCGA
study. (b) RefSeqidentified spectraper genein 10based logscale. Mostfre-
quently mutated genes in DNA level are under expressed in protein level.
COL6A3 had 35463 spectra counts, TTN (188), KRAS (71), DMD (76),
SYNE1 (43), LRP1B (37), ANK2 (59), and rest of the DNA level highly
mutatedgenes hadless than25 spectracounts. (c)Percentageof samplescon-
taining identified protein mutations in TCGA reported most frequently genes.
While most of the DNA level top frequently mutated genes were under ex-
pressed in protein level, we observed that some genes showed even higher
mutationfrequenciesacrosssamplesinproteinlevel. . . . . . . . . . . . . 60
Figure3.4: Percentage of IG gene peptide identifications in each sample normalized by
the number of known peptide identifications across sample subtypes. This
percentileratioiscalculatedbydividingthenumberofknownpeptideidentifi-
cationsfromthetotalnumberofIGpeptideidentificationswithineachsample.
(ratio= (#ofIG peptides)/(# ofknownpeptides)* 100)Differentkinds ofIG
gene segmentsare colored. Subtype C(sample groups showingboth hypermu-
tationandCIMPcharacteristics)showedcomparablyhighnumberofIGgene
peptideidentificationcomparedtoothersamplesubtypes. Chi-squaredtestof
thisplotshowed p-value<0.0001,χ2 =2927.71. . . . . . . . . . . . . . 61
Figure4.1: Relativelocationsofidentifiedantibodypeptides. Eachhorizontalblackline
representsadistinctpeptidesequence. Trypsinwasappliedforthecolorectal
tumor MS/MS spectra assessment, and four different enzymes were applied
for polyclonal antibody MS/MS spectra assessment. Both spectra sets were
searchedagainstthesameantibodydatabaseconstructedusingtumorRNA-seq
readsdrivenbyTCGA.(a)AntibodyPSMsfromcolorectaltumorMS/MSdata.
(b)AntibodyPSMsfrompolyclonalantibodyMS/MSdata. . . . . . . . . 85
viii
Figure4.2: Comparison of identified antibody PSMs per experiment and sample (a)
The source of antibody peptides in different samples. PSMs that match non-
referencepeptidesareeithermutationsorantibodypeptides. Antibodypeptides
should not be observed in cell-lines. However, floating antibodies could be
observed in normal colorectal samples. Antibodies from Tumor infiltrating
lymphocytes should only be observed in tumor samples. (b) Occurrence of
antibody peptides in tumor, normal, and tumor derived cell-lines are signifi-
cantlydifferentforMS/MSspectraoftumor,normal,andcell-linecolorectal
samples. EachspectrasetweresearchedagainsttheEnsemblGRCh38protein
database[20]and acustomantibody database. The numberofPSMs identified
as antibody peptides were 54K (colorectal tumor), 711 (colorectal normal),
and 0 (Cell-lines). The PSM counts were normalized against the number of
PSMstoknownpeptides. 5.5M incolorectaltumor,1.7M incolorectalnormal,
and0.1M inCell-lines. Thenormalizedratiossuggestthatasignificantlylarger
fractionof thecolorectal tumorPSMsare antibodypeptides, comparedto the
other two data-sets (Pearson’s χ2 p-val <10−4). (c) The distribution of the
numberofsamplescarryinganormalizedfractionofantibodypeptides. COAD
samplescarryahigherfractionofantibodypeptides. . . . . . . . . . . . . 86
Figure4.3: Peptide correlation test. We tested the correlation between the antibody
peptides and mutated peptides. For every pair of peptides, we counted the
number of samples co-occurring with these peptides and then we applied
Fisher exact test to calculate the p-value. For example, the peptide pairs of
NTLYLQMDSLR(antibody)andAAQAQGQSCEYSLMVGYQCGQVF(Q→
R)(SAAVpeptide) co-occurredin 26samples, andtherewasaco-absence in
42samples. Itwasrevealedthat68ofthe90samplessharedtheco-occurrence
ofthispairwithap-valueof2.60×10−6. Wedrewthehistogramofp-valuesof
allpairsinSupplementalTable4. Wealsodrewthehistogramofthep-values
from the decoy table generated by the random permutation of values. A 5%
FDRthresholdwasappliedtocollectthehighcorrelatedpairs. . . . . . . . 87
Figure4.4: Kaplan-Meiersurvivalestimator. Foranysubsetofpeptides,webi-partioned
peptides based on co-expression in samples. Next, we scored each sample
based on the homogeneity of peptides from a single partition in that sample
(Methods). The highest and lowest scoring samples (one-third each) were
grouped,andweretestedtodeterminetheclinicaloutcome. TheKaplan-Meier
survival estimator and log-rank test were applied to test the difference of the
clinical outcome of two groups. When testing with co-occurring mutated
peptide/antibody peptide pairs, we observed a significant correlation with
survival (Plot (a): p-value =0.032). In contrast, the correlation was reduced
when testing with only antibody peptides (Plot (b): p-value = 0.040), and
therewas no-correlationwhen testingwith mutated peptides. (Plot(c): p-value
=0.522). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
FigureA.1: Infilteringstage,RNA-seqreadsthathaveidenticalsplicejunctionsaremerged,
andextendedinbothends . . . . . . . . . . . . . . . . . . . . . . . . . . 90
ix