This document discusses applications of immunogenomics to cancer, including identifying tumor-specific antigens (TSAs) and developing personalized cancer vaccines and immunotherapies. It covers workflows for neoantigen discovery using exome sequencing, HLA typing, and predicting peptide-MHC binding. Tools discussed include Integrate-NEO, Polysolver, NetMHCpan, and pVAC-seq. The document also discusses the diversity of the immune system, passive and active immunotherapy approaches, and tools for immune repertoire profiling like TRUST and TIMER. It concludes by discussing applications of neoantigen discovery like vaccines and adoptive cell transfer, and perspectives on further improving neoantigen prediction and integrating repertoire analysis with functional assays
2. Outline
⢠Tumor Specific Antigen
⢠Diversity of the immune system
⢠Immunotherapy
⢠Workflow for neoantigen discovery -> personalized vaccine
⢠exome-based identification of neoantigen(Integrate-NEO)
⢠HLA typing (Polysolver)
⢠predict peptide-MHC interaction (NetMHC-pan4.0)
⢠Whole pipeline to predict tumor ag (pVAC-seq)
⢠Immune Repertoire Profiling (TRUST, TIMER)
⢠Applications of neoantigen discovery
⢠Future perspective
3. Tumor Specific Antigen (TSA)
⢠Inherently immunogenic, capable of eliciting strong
T-cell mediated immunity
⢠Expressed in a high % of tumors
⢠Uniform expression throughout tumor
â˘Restricted expression in normal tissues
â˘Functionally important for tumor cell survival
Ideal Tumor antigen?
4. Clonal selection theory
or, can the âtruthâ be learned?
⢠truth = the capacity to recognize self vs non-self
⢠Menoâs paradox: âWhat a man knows he cannot seek, since he knows
it; and what he does not know he cannot seek, since he does not even
know what to seek.â Philosophical fragments by Kierkegaard, 1844.
⢠Socrates resolves this difficulty by postulating that learning is nothing
but recollection:
âThe truth cannot be brought in, but was already inherent.â
Ălogical basis of the theory of the formation of antibodies:
antibodies specific for different antigens must have already existed even
before they encounter any specific pathogen.
Ăalso true for T cells
5. Diversity of the immune system
https://arxiv.org/pdf/1604.00487.pdf
⢠antibody repertoire > 1013
⢠1015-1020 TCR clonotypes
⢠Human genome has 3 x 109 b.p.
(genome alone cannot code for such
diversity)
Ă Diversity is created by:
1/ somatic recombination
2/ joining, N-nucleotide additions
3/ somatic hypermutation.
https://www.bsse.ethz.ch/lsi/research/systems-
immunology.html
10. Fig.1. Workflow for Neoantigen discovery and personalized
cancer vaccine design
1. cancer mutation/ neoantigen discovery
2. HLA typing
3. peptide + MHC binding prediction
Ă T cells neoantigen
Ă personalized vaccine
11. Exome-based identification of neoantigens
http://science.sciencemag.org/content/348/6230/69
depth of coverage
⢠300-500 fold exome coverage
⢠compensate for normal cellâs DNA reads
⢠founder clone vs sub clone
longer read length -> better
Algorithms
⢠Point mutation (FreeBayes, GATK...)
⢠INDEL:
Ă may produce stronger neoantigens
Ă challenging but can be improved with gapped
alignment or split-read algorithms
Ă Assembly-based realignment (ABRA)
⢠Structural Variant which creates protein fusion
Ă RNAseq data is better
Ă E.g. IntegrateNEO
12. Integrate-NEO
a pipeline for personalized gene fusion neoantigen discovery
INTEGRATE-neo: a pipeline for personalized gene fusion neoantigen discovery.
Bioinformatics 2017.
variant detection
translated peptide antigen
variant annotation
class I or class II predicted
binding affinity
13. HLA (human leukocyte antigen)
http://hla.alleles.org/nomenclature/stats.html
⢠encodes for cell surface proteins that bind
and present peptide to T cells
⢠class I is expressed in all nucleated cells
⢠class II is mostly expressed on professional
antigen presenting cells
⢠the most polymorphic region in the genome
⢠many duplicated segments + pseudogenes
Ă reference genome?
Ă de novo assembly?
Ă targeted sequencing?
Ă tools: Polysolver, HLAMiner and Optitype
14. Polysolver for high-precision HLA-typing
⢠realign unaligned reads to known HLA alleles in the IPD-IMGT/HLA database
⢠Bayesian classification approach
⢠provides full resolution of up to 8-digits.
15. Challenges to predict peptide-MHC interactions
⢠high rate of false positive epitope prediction
⢠variable peptide length : 8-15 aa for class I and 11-30 aa for class II
⢠Extreme polymorphism in HLA class I and Class II protein
⢠Many factors affect real peptide-MHC interactions:
1/ peptide-MHC binding
2/ biases in the antigen processing pathway
(TAP transport, ERAP trimming)
3/ stability of peptide-MHC complex
16. Fig.2. Idealized Selection of mutant containing peptides for
neoantigen prediction
(A) The localized peptides that tile across and
contain the mutated amino acid substitution are
identified and parsed into the neoantigen prediction
pipeline. Each peptide is considered for HLA
binding strength relative to its non-mutant (wildtype)
counterpart.
(B) Shown is the top scoring candidate peptide
that was selected across all specified k-mers and
between all HLA types that were input to the neoantigen
prediction pipeline.
18. NetMHC-pan 4.0:
Peptide-MHC class I interaction predictions
⢠Peptide inputs were represented as 9-mer pseudo peptide
(insert/delete, BLOSUM encoding)
⢠1st output neuron returns a score of affinity (NNAlign framework)
⢠2nd output neuron returns a score of eluted ligand
www.cbs.dtu.dk/services/NetMHCpan-4.0
⢠NN with two output neurons used for
combined training on binding affinity and
eluted ligand data.
⢠binding affinity data: (IEDB) +
IMGT/HLA (HLA sequence)
⢠ligand data: MS peptidome
⢠userâs peptide input = 8-15aa
19. Peptide-MHC class II predictions
⢠higher degree of diversity, encoded by 4 different loci (3 are extremely polymorphic)
⢠peptide binding are more âpromiscuousâ than class I
⢠class II binding groove is open on both ends (11-30aa peptide)
⢠difficult to set a cut off binding because of the wide range of binding affinity
20. pVAC-Seq: in silico prediction of tumor antigens
inputs:
1/ mutations from variant calling pipeline
2/ annotated mutations
3/ HLA haplotypes
⢠clinical assay
⢠in silico (HLAminer) using WGS
epitope prediction = netMHCv3.4
integrate expression data + filter
⢠retain neoantigens expressed as mRNA
⢠affinity < 500nM
⢠sequence depth (>=10x)
⢠fractions of read containing variant allele
(VAF>=40%)
21. Immune Repertoire Profiling
⢠profile the diverse repertoire of TCR and BCR
⢠TIL (tumor infiltrating lymphocyte) or PBMC by
IHC/flow
⢠multiplex PCR to amplify the V(D)J regions
-> paired-end sequencing
⢠immune signature reflects patient health and
predict response to therapy
e.g. 1/Potential Support Vector Machine-based
approach predicts individualâs age, health,
transplantation status, and development of lymphoid
cancer based on repertoire profile.
e.g. 2/increased Granzyme + perforin expression in
tumors with higher mutation load and lower
tumor stage.v sampling issues
22. TRUST
(TCR receptor Utilities for Solid Tissue)
Landscape of tumor-infiltrating T cell repertoire of human cancers.
Li et at. Nat Gen 2016
⢠analyze TCR/BCR sequences using unselected
RNAseq data from solid tumor/tissues.
⢠both single-end and paired end with any
read length
⢠de novo assembly of the CDR3 region
⢠report contigs containing CDR3 DNA/aa
⢠realign contigs to IMGT reference gene to
report the V and J genes
approach:
1/ map read to human genome
2/ search for read pairs with one mate mapped
to TCR and the other mate unmappable
3/ pairwise comparison between unmapped reads
to construct matrix -> undirected graph (node = read,
edge= overlap).
4/ graph is divided into disjoint cliques -> # CDR3
5/ assemble reads in each clique to obtain contigs
23. TIMER (Tumor Immune Estimation Resource)
Comprehensive analyses of tumor immunity:
implications for cancer immunotherapy.
Li et al. Genome Biol 2016https://cistrome.shinyapps.io/timer/
TIMER: web-based comprehensive resource to
analyze immune infiltrates in cancer
Cancer Res 2017
deconvolutional method
24. Applications of neoantigen discovery
1. Vaccines
Tumor antigen can be administered directly
or through Dendritic cells (DC)
- DNA (cheaper than peptide)
- recombinant proteins or peptides
- recombinant viruses
- autologous tumor protein extract
- total autologous tumor mRNA
Dendritic cell-the key sentinel who has
unparalleled capacity to activate naĂŻve T cells.
25. Applications of neoantigen discovery
2. Adoptive Cell Transfer (ACT)
CAR-T cell
Abbas et al : Cellular and Molecular Imm. 8th ed.
genetically engineered T cells to recognize
tumor-specific ag
26. Future perspectives
⢠High Throughput technology + computational innovation is instrumental in the field of
immunogenomics
⢠Lots of room for improvement with neoantigen discovery
( peptide-MHC and antibody-antigen)
⢠in silico finding needs validation
⢠demand the need for new high throughput screening assay of potential neoantigens
⢠Neoantigen discovery is challenging but eventually will create the next breakthrough in
targeted cancer immunotherapy
⢠repertoire analysis maybe useful for monitoring/ diagnosis/ predict response
⢠the need to integrate repertoire analysis with immune functional analysis (cytokine/ activation
/exhaustion marker etc)