Data-driven Approaches to Identify the Regulators of the ... · Association between CTL and...
61
Data-driven Approaches to Identify the Regulators of the Anticancer Immune Response Peng Jiang, Ph.D. July 9th, 2020 Bioinformatics Training and Education Program
Transcript of Data-driven Approaches to Identify the Regulators of the ... · Association between CTL and...
Data-driven Approaches to Identify the Regulators of the Anticancer Immune Response
Peng Jiang, Ph.D.July 9th, 2020
Bioinformatics Training and Education Program
Is my result supported by other datasets?
2
3
T lymphocyte
Cancer cell T lymphocyteCancer cell
Tumor Immune Evasion
Knowledge database
• Focused Presentation• iATLAS (Thorsson et al., Immunity 2018)• TCIA (Charoentong et al., Cell Report 2017)
• Customized Analysis• TIDE (Jiang et al., Nature Medicine 2018)• TISIDB (Ru et al., Bioinformatics 2019)
4
iATLAS: Flagship website of TCGAStudy the immune ecosystem from bulk tumor profiles
5
iATLAS: Flagship website of TCGA
Major result to browser, the six clusters they defined
6
Select subtypes to browse TCGA association
7
Knowledge database
• Focused Presentation• iATLAS (Thorsson et al., Immunity 2018)• TCIA (Charoentong et al., Cell Report 2017)
• Customized Analysis• TIDE (Jiang et al., Nature Medicine 2018)• TISIDB (Ru et al., Bioinformatics 2019)
8
Cyotoxic T Lymphocyte
T cell Dysfunction T cell ExclusionCTL high tumor CTL low tumor
TIDE: Tumor Immune Dysfunction and Exclusion
9
http://tide.dfci.harvard.edu
(Gajewski, Schreiber, Fu et al., 2013)
TIDE: Tumor Immune Dysfunction and Exclusion
10
http://tide.dfci.harvard.edu
Gene Query
11
Gene Query: Example TGFB1
12
T-cell Dysfunction in CTL-high Tumors
13
Cyotoxic T Lymphocyte
T cell Dysfunction T cell ExclusionCTL high tumor CTL low tumor
High level of cytotoxic T lymphocyte indicates better survival
0.2
0.6
1.0
Surv
ival
Fra
ctio
n
0 100 200 300Month
Low CTLn=229
High CTLn=88
p = 0.0028
14
Cytotoxic T Lymphocyte (CTL)Mark genes:CD8A, CD8B,GZMA, GZMB,PRF1
TCGA melanoma
Association between CTL and survival depends on TGFB1 level
0.2
0.6
1.0
Surv
ival
Fra
ctio
n
0 100 200 300Month
Low CTLn=229
High CTLn=88
p = 0.0028
0 100 200
0.2
0.6
1.0
TGFB1 high
Month
Surv
ival
Fra
ctio
n
Low CTLn=19
High CTLn=17
p = 0.78
(1SD above mean )
0 100 200 300Month
Low CTLn=210
High CTLn=71
TGFB1 low
p = 0.0016
(all others)
0.2
0.6
1.0
Surv
ival
Fra
ctio
n
15
T-cell dysfunction test in a linear model
Hazard = a * CTL + b * P + d * CTL * P
CTL (Cytotoxic T Lymphocyte)
Death Hazard
TGFB1 expression
T cell dysfunction score = d / StdErr(d)16
Antagonistic Interaction between CTL and TGFB1
Coef Stderr Z Pr(>|z|)
Age 0.02 0.01 3.55 3.78E-04
Gender 0.02 0.17 0.12 9.07E-01
Stage 0.29 0.09 3.31 9.34E-04
CTL -0.50 0.15 -3.32 9.05E-04
TGFB1 -0.10 0.10 -1.04 3.00E-01
CTL * TGFB1 0.11 0.03 3.47 5.18E-04
Hazard = c1*Age + c2*Gender + c3*Stage + a*CTL + b*TGFB1 + d * CTL * TGFB1
Background Interaction17
Interaction between variables matters!
18
Coef StdErr Z-score Pr(>|z|)
Age 0.02 0.01 3.95 7.85E-05
Gender 0.00 0.17 0.00 9.97E-01
Stage 0.30 0.09 3.36 7.74E-04
TGFB1 -0.13 0.09 -1.42 1.55E-01
Coef StdErr Z-score Pr(>|z|)
Age 0.02 0.01 3.55 3.79E-04
Gender 0.02 0.17 0.12 9.08E-01
Stage 0.29 0.09 3.31 9.34E-04
CTL -0.54 0.17 -3.29 9.95E-04
TGFB1 -0.10 0.10 -0.97 3.31E-01
CTL*TGFB1 0.13 0.04 3.47 5.28E-04
TGFB1 encodes an immuno-suppressive cytokine (Sharma et al., 2017)
TGFB1 alone associate with lower death risk
Correlation(TGFB1, CTL) = 0.39 TGFB1 antagonizes the effect of cytotoxic T lymphocyte
Synergistic Interaction between CTL and SOX10
Coef Stderr Z Pr(>|z|)
Age 0.02 0.01 3.26 1.11E-03
Gender 0.03 0.17 0.15 8.80E-01
Stage 0.29 0.09 3.33 8.63E-04
CTL -0.79 0.21 -3.79 1.51E-04
SOX10 -0.01 0.10 -0.11 9.10E-01
CTL * SOX10 -0.59 0.16 -3.69 2.23E-04
Hazard = c1*Age + c2*Gender + c3*Stage + a*CTL + b*SOX10 + d * CTL * SOX10
Background InteractionSOX10 level in cancer cells can promote the T-cell mediated tumor killing (Patel et al., 2017; Khong et al., 2002). 19
The p-value histogram should have a significant peak
20
p−value
Perc
enta
ge (%
)
0.0 0.4 0.8
0.2
0.6
1.0
Melanoma (SKCM)
p−value0.0 0.4 0.8
Glioblastoma (GBM)
Non-significant
0.2
0.6
1.0
Large-scale Cancer Genomics Data Cohorts
PRECOGMETABRIC
135 cohorts 18K samples
49 cohorts, 13K samples21
Five datasets with significant p-value peaks
p−value
Perc
enta
ge (%
)
0.0 0.4 0.8
0.2
0.6
1.0
Melanoma (SKCM)
Endometrial (UCEC) Leukemia (AML)
Breast (TNBC) Neuroblastoma (NB)
p−value0.0 0.4 0.8
0.2
0.6
1.0
p−value0.0 0.4 0.8
0.2
0.6
1.0
p−value0.0 0.4 0.8
0.5
1.0
1.5
2.0
p−value0.0 0.4 0.8
0.5
1.0
1.5
2.0
Perc
enta
ge (%
)
22
Genes with Significant T Dysfunction Scores-4
04
Z-score ( d / SE )Endometrial (UCEC)
Breast (TNBC)Neuroblastoma (NB)
Leukemia (AML)Melanoma (SKCM)
Average Profile TMEM
130LIM
E1NUB1DIDO
1W
FS1M
ALKHNYNXCL1W
NT10AADAM
TS17NAV2SFNCO
L11A2VSNL1G
UCA1AUSH1GTM
PRSS3DNAJC30NDST4SERPINB9CO
Q10A
FAM65B
LRMP
KCNA5ARID3BSTARD3PNM
TANKRD29TG
FB1NFATC3SPNELM
O1
ADGRG
5PO
U2F2NLRC5ZNFX1PD-L1TREM
L1ICO
SBTN3A1G
PR155RO
RAIL2RBCD5VO
PP1NDST3ADAM
19KCNM
A1
AMPD3
LRRC15ZBTB7AIL21RZBTB32RHO
FPIM
2FCRL5LY9PVRIGCD6SPO
CK2KLRB1KCNA3G
RK2PDE1ABM
P4CACNA2D2AXIN2O
STCSEC23ASNRNP48PPP3CADDX59SESTD1ST3G
AL6TIPRLCO
PG1
NDUFS2PEX13HSDL2DXOLM
BRD2PHG
DHCO
PB2PG
M3
PAPSS1ATP5C1HDAC2TRIT1CCDC43SOX10
Endometrial (UCEC)Breast (TNBC)
Neuroblastoma (NB)Leukemia (AML)
Melanoma (SKCM)Average Profile
23
Correlation with Cytotoxic T Lymphocyte Level
24
●
● ●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●●●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●●●
●●
●
●●
●
●
●
●
●
●●●
●
●●●●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●●●●
● ●
●
●
● ●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
● ●
●●
●●
●
●●
●
●
●
●●
●
●●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
●●
●●●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
● ● ●
●
●
●●
●
●
●
●
●●
●
●●
●
●●
●
●
●
●
●●●
●
●
●●●
●
●
●
●●
●
●●
●
●
●
●
●
●
●●
●●●
●
●●
●
●●
●
●●
●●
●
●●
●
●
●
●●
●
●●●
●
●
●
●●
●
●●
●
●●●●
●
●
●
●
●●
●
●
●
●
●●
●●●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
0 1 2 3 4 5 6
02
46
8
r= 0.385 , p= 1.18e−12
CTLTG
FB1
Correlation with Death Risk
25
0 50 100 200 300
0.0
0.2
0.4
0.6
0.8
1.0
Continuous z= −1.42 , p= 0.155
OS (month)
Surv
ival F
ract
ion
TGFB1 Top (n=268)TGFB1 Bottom (n=49)
Association with T cell Dysfunction : Core
26
0 50 100 200 300
0.0
0.2
0.4
0.6
0.8
1.0
TGFB1 High
OS (month)
Surv
ival F
ract
ion
CTL Top (n=32)CTL Bottom (n=32)
0 50 100 150 200 250
0.0
0.2
0.4
0.6
0.8
1.0
TGFB1 Low
OS (month)
Surv
ival F
ract
ion
CTL Top (n=127)CTL Bottom (n=126)
Continuous z= 3.47 , p= 0.000518
Association with T cell Dysfunction : More
27
0 10 20 30 40
0.0
0.2
0.4
0.6
0.8
1.0
TGFB1 High
OS
Surv
ival F
ract
ion
CTL Top (n=7)CTL Bottom (n=7)
0 50 100 150
0.0
0.2
0.4
0.6
0.8
1.0
TGFB1 Low
OS
Surv
ival F
ract
ion
CTL Top (n=45)CTL Bottom (n=45)
Continuous z= 3 , p= 0.00272
Exclusion : Example PTGS2
28
Tumor Immune Evasion through T-cell Exclusion
29
Cyotoxic T Lymphocyte
T cell Dysfunction T cell ExclusionCTL high tumor CTL low tumor
Immunosuppressive Cell Types in T-cell Exclusion
(Gajewski, Schreiber, Fu et al., 2013)
T cell exclusion factors
30
Cell signature
Infiltration
Tumor expression
High Low
Correlation
geno
me-
wid
e
(Dougan et al., 2017)
Immunosuppressive cell signatures predict T-cell exclusion in tumors
Corr(Tumor, Immune suppressive cell signature)
Myeloid derived suppressor Tumor assoc macrophage Cancer assoc fibroblast Average signature
●
●
●
●●
●●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
● ●
●
● ●
●
●
●
●●
●●
●
●●●●●
●
●
●
●
●
●
●●
●● ●
●
●
●
●●
●
●●●
●
●
●●●
●●
●
●●●
●
●
●
● ●
●
●●
● ● ●
● ●
●●
●
●
●
●
● ●●●
●
●● ●
●
●
●
●
●●
●●●● ●● ●
●
●
●
● ●●● ●
●●●
●●
●
●
●
●
●
●
●
●
●
●
●● ●●●●
●
●
●
●●
●
●
●
●
●
●●
●
●
●●
●
● ●
●
●
●●
●
● ●●
●
●● ●
●● ●
●●●
●
●
●
●
●
●
●
●●
●
●●
●●
●
●
●
●●●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●●
●
● ● ●
●
●
●
●●
●●
● ●●
●
●●
●
● ● ●
●
●
●
●●● ●
●
●
●
● ●●●●●●
●●
●
●●●
●
●
●● ●●●●●
●
●
● ●
●●
●●
●●● ●●
●
●●
●
●●●
●
●
●● ●
●
● ●●
●
●
●●●
●●
●
−0.3 −0.1 0.1
R= −0.722
●
●
●
●●
● ●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
● ●
●
● ●
●
●
●
●●
●●
●
●●●● ●
●
●
●
●
●
●
●●
●● ●
●
●
●
●●
●
●●●
●
●
●●●
●●
●
●● ●
●
●
●
● ●
●
●●
● ● ●
● ●
●●
●
●
●
●
● ●●●
●
●● ●
●
●
●
●
●●
● ●●● ●● ●
●
●
●
● ●●
● ●
●●●
●●
●
●
●
●
●
●
●
●
●
●
●● ●●●●
●
●
●
●●
●
●
●
●
●
●●
●
●
●●
●
● ●
●
●
●●
●
● ●●
●
●● ●
●● ●
●●●
●
●
●
●
●
●
●
●●
●
●●
●●
●
●
●
● ●●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●●
●
● ● ●
●
●
●
●●
●●
● ●●
●
●●
●
● ● ●
●
●
●
●●● ●
●
●
●
●●● ●●●●
●●
●
●●●
●
●
●● ●● ●●●
●
●
● ●
●●
●●
● ●● ●
●
●
●●
●
●●●
●
●
●●●
●
●● ●
●
●
●●●
●●
●
−0.3 −0.1 0.1
01
23
45
6 R= −0.634
CTL
●
●
●
●●
●●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●●
●
● ●
●
●
●
●●
●●
●
●● ●● ●
●
●
●
●
●
●
●●
●● ●●
●
●
●●
●
●●●
●
●
● ●●
● ●
●
● ● ●
●
●
●
●●
●
●●
● ● ●
● ●
●●
●
●
●
●
● ● ●●
●
●● ●
●
●
●
●
●●
● ●●●●●●●
●
●
●●●
● ●
● ●●
●●
●
●
●
●
●
●
●
●
●
●
●●● ●●●●
●
●
●●
●
●
●
●
●
●●
●
●
●●
●
● ●
●
●
●●
●
● ●●
●
●● ●
●● ●
●●
●
●
●
●
●
●
●
●
● ●
●
●●● ●
●
●
●
●●●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●●
●
●● ●
●
●
●
●●
●●● ●●
●
●●
●
● ●●
●
●
●
●●● ●
●
●
●
● ●● ●●● ●
●●
●
●●●
●
●
●● ● ●●●●
●
●
● ●
●●
●●
● ●●●
●
●
●●
●
●● ●
●
●
●● ●
●
● ●●
●
●
●●●
●●
●
−0.15 −0.05 0.05
R= −0.718
●
●
●
●●
●●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
● ●
●
●●
●
●
●
●●
●●
●
● ●● ●●
●
●
●
●
●
●
●●
●●●●
●
●
●●
●
● ●●
●
●
●●●
● ●
●
●●●
●
●
●
● ●
●
●●
●●●
●●
●●
●
●
●
●
● ●●●
●
●● ●
●
●
●
●
●●
●● ●● ●●●
●
●
●
● ●●
● ●
● ●●
●●
●
●
●
●
●
●
●
●
●
●
● ● ●●●●
●
●
●
● ●
●
●
●
●
●
●●
●
●
●●
●
●●
●
●
●●
●
● ●●
●
●●●
●●●
●●●
●
●
●
●
●
●
●
●●
●
●●
● ●
●
●
●
●●●
●
●
●
●●
● ●
●
●
●
●
●
●
●
●
●●
●
● ●●
●
●
●
●●
●●
●● ●
●
●●
●
●● ●
●
●
●
● ●●●
●
●
●
● ●●●● ● ●
●●
●
●●●
●
●
●●●●●● ●
●
●
●●
●●
●●
●●●●
●
●
●●
●
●●●
●
●
●● ●
●
● ●●
●
●
●●●
●●
●
−0.2 0.0 0.2
R= −0.387
31
CTL : expression average of CD8A, CD8B, GZMA, GZMB, PRF1
43 solid tumor types
Corr(CTL, Exclusion score)
Perc
enta
ge (%
)
−1.0 0.0 1.0
010
2030
32
Immunosuppressive cell signatures predict T-cell exclusion in all solid tumors
Exclusion : Example PTGS2
33
Two Categories of Tumor Immune Evasion
(Gajewski, Schreiber, Fu et al., 2013) 34
Cyotoxic T Lymphocyte
T cell Dysfunction T cell ExclusionCTL high tumor CTL low tumor
TIDE: Tumor Immune Dysfunction and Exclusion
35
Prediction with a Two-Part Procedure
CTL high
All positive
CTL low
Others
Corr
Dysfunctionsignature
Corr
Exclusionsignature
CTL
36
CD8A, CD8B, GZMA, GZMB, PRF1
TIDE Scores of Pre-Treatment Tumors
PD1 RNASeq
TID
E pr
edic
tion
scor
e−1
01
23
high lowCTL
Long−survival
CTLA4 RNASeq
high lowCTL−1
01
23
Responder Non-responder
PD1 NanoString
high lowCTL−1
01
23
Hugo et al., 2016 n = 25 VanAllen et al., 2015 n = 42 Prat et al., 2017 n = 33
37
Receiver Operating Characteristic (ROC) Curve
False positive rate
True
pos
itive
rate A
B
Biomarker A is better than Biomarker B
38
False positive rate
AUC : Area Under ROC Curve
True
pos
itive
rate
From 0 to 1 with random performance as 0.5
TIDE outperforms other biomarkers
False positive rate
True
pos
itive
rate
0.2 0.6 1.00.0
0.4
0.8
TIDEPDL1IFNGMutation
PD1 RNASeq CTLA4 RNASeq PD1 NanoString
False positive rate0.2 0.6 1.00.
00.
40.
8
TIDEPDL1IFNGMutation
False positive rate0.2 0.6 1.00.
00.
40.
8
TIDEPDL1IFNG
39
PD1 RNASeq CTLA4 RNASeq PD1 NanoString
AUC
0.5
0.6
0.7
0.8
TID
EM
utat
ion
PDL1
CD
8IF
NG
T.C
lona
lity IP
SSC
NA
TID
EM
utat
ion
PDL1
CD
8IF
NG
T.C
lona
lity
IPS
SCN
A
TID
EPD
L1C
D8
IFN
GIP
STI
DE.
Mel
anom
aTI
DE.
Aden
oTI
DE.
Squa
mou
s0.5
0.6
0.7
0.8
0.5
0.6
0.7
0.8
40
TIDE outperforms other biomarkers
Mutation: nonsynonymous mutation load; PDL1: PD-L1 expression; CD8: CD8A + CD8B expression;IFNG: interferon gamma response expression signature (Ayers et al. 2017); T.Clonality: Clonality of T cell receptor;IPS: ImmunePhenoScore (Charoentong et al. 2017); SCNA: variation of copy number alterations (Davoli et al. 2017).
TIDE can predict patient overall survival
OS (day)200 600 1000
0.0
0.4
0.8
p = 0.011
Surv
ival F
ract
ion
Top
Bottom
OS (day)500 1000 1500
0.0
0.4
0.8
p = 0.001
Top
Bottom
PFS (month)5 15 25 35
0.0
0.4
0.8
p = 0.042
Top
Bottom
PD1 RNASeq CTLA4 RNASeq PD1 NanoString
41
Independent validation of TIDE
42
Eva Pérez-Guijarro et al., Nature Medicine 2020 @ Glenn Merlino Lab
Double-blinded test @ BMS064
Immunotherapy Clinical CohortsExample : STAT1
43
CRISPR ScreensExample: STAT1
44
CRISPR screen for T-cell mediated tumor killing
45Pan*, Kobayashi*, Jiang* et al., Science, 2018
Gene Prioritization Putting everything together
46
Gene Prioritization
47
Biomarker Evaluation
48
Biomarker Comparison
49
Biomarker Comparison : Survival Plot
50
Some online demo
51
2: Cell Signaling Research Platform
52
https://cellsig-dev.ccr.cancer.gov (temporary domain, only within NIH firewall)
Data Collection
53
AEENASRA GEO
MetaData
Expression matrix
Automatic ProcessingTreatment, Condition, Dose, Duration
Controlled Vocabulary
Curate Proofread
+ logFCGenes
Response within signaling molecules
54
−4
−2
0
2
4logFC
Ligand
Receptor
IL4
Activ
inA
BMP4
BMP6
IFNG IL27
IFN1
IFNL IL12 IL21
IL10
GM
CSF
IL15 LIF
MCS
FBD
NFBM
P2PG
E2TG
FB1
TGFB
3IL
2IL
17A
CD40
LIL
36TN
FA IL1A
IL1B NO
FGF2
VEG
FA
GDF15GDF3BMP2
FGF17IL11
IFNGLIF
OSMIL32
IL36GCCL5
PTGS2 (PGE2)CXCL8CXCL2CXCL3
IL1AIL1B
CSF2 (GMCSF)CCL20CXCL1CCL2
WNT5AIL12B (IL12,IL23)
CCL23EBI3 (IL35)
INHBA (Activin A/AB, Inhibin A)CCL7
CXCL5CXCL6
IL6CCL18
IL1RN (IL1RA)CCL24CCL17CCL22GDF5
ANGPTL1LGALS9 (Galectin 9)
CD274 (PDL1)TNFSF10 (TRAIL)
CXCL10CXCL11
CCL8CXCL9
CXCR4 (MIF,CXCL12)FGFR3 (FGFs)
TNFRSF12A (TWEAK)IL7R (IL7,TSLP)
FPR2 (CCL23)
Gene Query
55
Signal Prediction
56
Signal Prediction
57
signaling molecules
input = a1 s1 + ... + an sn
Application on Single Cell RNA-Seq
58
Application on Spatial Transcriptomics
59
Some online demo
60
X. Shirley Liu
• Shengqing Gu
• Jingxin Fu
Kai Wucherpfennig
• Deng Pan
• Aya Kobayashi
Current Lab MembersBeibei Ru Yu Zhang
Jie Yuan (expected on June, 2020)
Postdoc mentors
Eytan RuppinBranch Chief
S. Cenk Sahinalp Sridhar Hannenhalli
CDSL PIs
