Thermo Fisher Scientific • 5791 Van Allen Way • Carlsbad, CA 92008 • www.thermofisher.com For Research Use Only. Not for use in diagnostic procedures. The content provided herein may relate to products that have not been officially released and is subject to change without notice.

PD OCR

0.00

0.02

0.04

0.06

0.08

Pre-treatment TCR Convergence vs Response

Con

verg

ent T

CR

Fre

quen

cy

●

●●

●

●

●●●

3.5 4.0 4.5 5.0

3.5

4.0

4.5

5.0

cor= −0.2 , pvalue= 0.63

log10(Uniques from TF)

log1

0(U

niqu

es fr

om S

eque

nta)

●●●

●●

●●●

5.0 5.5 6.0 6.5

5.0

5.5

6.0

6.5

cor= 0.42 , pvalue= 0.304

log10(Read depth from TF)

log1

0(R

ead

dept

h fro

m S

eque

nta)

●

●●

●

●

●

●

●

4 6 8 10 12 14

46

810

1214

cor= 0.92 , pvalue= 0.001

Shannon from TF

Shan

non

from

Seq

uent

a ●

●●

●

●

●

●

●

0.4 0.5 0.6 0.7 0.8 0.9

0.4

0.5

0.6

0.7

0.8

0.9

cor= 0.92 , pvalue= 0.001

Eveness from TF

Even

ess

from

Seq

uent

a

L Zhang1, T Looney2, D Topacio-Hall2, G Lowman2, D Oh1, L Fong1 (1) University of California San Francisco (2) Thermo Fisher Scientific

Peripheral blood TCRB repertoire convergence and clonal expansion predict response to anti-CTLA-4 monotherapy for cancer

METHODS

Figure 4. Example of a convergent TCR group detected in an individual with melanoma. This group consists of three TCRβ clones that are identical in TCRβ amino acid space but have distinct CDR3 NT junctions owing to differences in non-templated bases at the V-D-J junction. Blue indicates bases contributed by the variable gene while yellow indicates bases contributed by the joining gene. Red arrows indicate positions where clones differ. Substitution sequencing errors and PCR errors can create artifacts that resemble convergent TCRs.

Figure 5: Comparative analysis of repertoire features in samples analyzed via Ion Torrent and Illumina-based assays. Eight peripheral blood leukocyte (PBL) samples derived from three donors were analyzed using the Oncomine TCRB-LR assay (Ion Torrent, X-axis) or Sequenta/Adaptive Biotechnologies TCRB assay (Illumina, Y-axis). Pearson's correlation coefficient was used to measure the consistency of two platforms with respect to clone diversity (Shannon entropy), overlap, evenness (normalized Shannon entropy) and TCR convergence. TCR convergence was calculated as the aggregate frequency of clones sharing an amino acid sequence with at least one other clone.

ABSTRACT Tumor antigen-driven selection may expand T cells having T cell receptors (TCRs) of shared antigen specificity but different amino acid or nucleotide sequence in a process known as TCR convergence. Efforts to evaluate the biomarker utility of TCR convergence through TCRB repertoire sequencing have been hampered by substitution sequencing errors, given that such errors may create artifacts resembling TCR convergence. Here we leveraged the low substitution error rate of the Ion Torrent platform to evaluate convergence as a predictive biomarker for response to anti-CTLA-4 monotherapy in a set of 22 individuals with cancer. For context, we compared convergence values obtained using this platform to those for the same samples interrogated with Illumina-based TCRB repertoire sequencing. Finally, we examined whether TCR convergence may be combined with measurements of clonal expansion to improve prediction of immunotherapy response. INTRODUCTION Chronic stimulation of T cells with tumor neoantigen may elicit convergent T cell responses. The frequency of convergent TCRs within a repertoire may provide an indication of the immunogenicity of a tumor and thus its sensitivity to checkpoint blockade therapy. Unlike biomarkers relying of the quantification of tumor genetic alterations, TCR convergence: (1) may detect T cell responses to tumor neoantigens beyond those arising from non-synonymous mutations; (2) avoids probabilistic models for prediction of immunogenicity; (3) is sequencing efficient, requiring ~1.5M reads per sample and (4) may be measured from the abundant genetic material within the buffy coat fraction of centrifuged peripheral blood to enable liquid biopsy applications.

•  We found measurements of TCR evenness, diversity and clonal overlap to be consistent between Ion Torrent and Illumina based TCR sequencing assays.

•  By contrast, TCR convergence values were not significantly correlated, with all 8 samples showing higher convergence in the Illumina-derived dataset. Given that substitution sequencing errors may give rise to artifacts resembling convergence, these findings support the notion that the Ion Torrent may be well suited to the measurement of TCR convergence.

•  Analysis of baseline PBL from 22 individuals receiving CTLA-4 blockade revealed that convergence and evenness values independently predicted response and could be combined to improve the accuracy of a logistic regression classifier (AUC = .89).

•  TCR convergence may serve as a predictive biomarker for a wide range of cancers, including those where tumor mutation burden is not predictive of response.

Figure 6. TCR evenness and convergence values for responders (N=11) and non-responders (N=11). TCR evenness is calculated as the normalized Shannon entropy of clone frequencies. Convergent TCR frequency was calculated as described above. All cancer types were included in the analysis.

Table 1. Cancer type and repertoire features for 22 individuals receiving CTLA-4 monotherapy. Each individual was profiled via the Oncomine TCRB-LR Assay at a single baseline timepoint using 25ng of PBL total RNA. Repertoire feature values indicate the average and range for responders and non-responders.

Figure 3. Substitution error rate for the Ion S5 530 chip and three common Illumina platforms. The substitution error rate for the Ion Torrent was obtained following sequencing of the ecoli dh1b genome and analyzed by Torrent Accuracy plugin as described in Looney et al (1). Illumina values are obtained from Schirmer et al (2). Substitution errors are a key challenge to the measurement of TCR convergence.

RESULTS RESULTS

Figure 7. Prediction of immunotherapy outcome via TCR evenness and convergence. A logistic regression classifier (R caret package) was trained using TCR evenness and convergence as features to predict response to immunotherapy. (A) Model scoring of responders and non-responders. (B) Receiver operator characteristic curves derived from the use of evenness, convergence, or the combination of evenness and convergence to predict immunotherapy response. The combination of TCR evenness and convergence improves prediction of response (AUC = .89).

Frequency of base substitution errors

0.000 0.001 0.002 0.003 0.004 0.005

GAII (0.51%)

HiSeq (0.26%)

MiSeq (0.41%)

Ion Torrent S5 530 (0.036%)

Frequency of substitution errors

CDR loops

T Cell Receptor

HLA presenting peptide

Variable Diversity Joining Constant

N1 N2

A. Adult IGH or TCRBeta chain rearrangement

In adult B and T cells, the process of VDJ rearrangement very often involves exonucleotide chewback of VDJ genes and the addition of non-templated bases, forming N1 and N2 regions in the B cell receptorheavy chain CDR3 and the T cell receptor Beta chain CDR3. These processes vastly increase IGH and TCRB CDR3 diversity.

Variable Diversity Joining Constant

B. Fetal IGH or TCRBeta chain rearrangement

In the fetus, the process of VDJ rearrangement often occurs withoutexonucleotide chewback of VDJ genes and addition of non-templated bases, resulting in a restricted IGH and TCRB CDR3 repertoire that is distinct from the adult repertoire.

These structural differences can be used to distinguish fetal B and T cellCDR3 receptors from maternal B and T cell CDR3 receptors in cell freeDNA present in maternal peripheral blood. In this way, fetal B and Tcell health and development may be monitored in a non-invasivemanner.

Figure 1. Structural differences between fetal and adult B and T cell receptors

~330bp amplicon

Sequencing in multiplex on S5 via 530 chip

FR1-C multiplex PCR ( AmpliSeq)

Ion Reporter analysis of repertoire features and

polymorphism

Workflow

Figure 2. The Oncomine TCRB-LR assay utilizes AmpliSeq multiplex PCR primers to target the TCRβ Framework 1 and Constant regions, enabling both clonotyping and detection of TRBV gene polymorphism linked to adverse events during immunotherapy.

Variable CDR3 AA CDR3 NT Freq

TRBV7-8 ASSLGQAYEQY GCCAGCAGCTTAGGTCAGGCATACGAGCAGTAC 1.8E-3

TRBV7-8 ASSLGQAYEQY GCCAGCAGCTTGGGACAGGCCTACGAGCAGTAC 4.8E-4

TRBV7-8 ASSLGQAYEQY GCCAGCAGCTTAGGGCAGGCCTACGAGCAGTAC 1E-4

●

●●

●

●

●●●

3.5 4.0 4.5 5.0

3.5

4.0

4.5

5.0

cor= −0.2 , pvalue= 0.63

log10(Uniques from TF)

log1

0(U

niqu

es fr

om S

eque

nta)

●●●

●●

●●●

5.0 5.5 6.0 6.5

5.0

5.5

6.0

6.5

cor= 0.42 , pvalue= 0.304

log10(Read depth from TF)

log1

0(R

ead

dept

h fro

m S

eque

nta)

●

●●

●

●

●

●

●

4 6 8 10 12 14

46

810

1214

cor= 0.92 , pvalue= 0.001

Shannon from TF

Shan

non

from

Seq

uent

a ●

●●

●

●

●

●

●

0.4 0.5 0.6 0.7 0.8 0.9

0.4

0.5

0.6

0.7

0.8

0.9

cor= 0.92 , pvalue= 0.001

Eveness from TF

Even

ess

from

Seq

uent

a

Variable CDR3

Polymorphic site

Category   Subdefini1on   Responder   Non-­‐responder  

Cancer  Type  

Prostate   2   4  Melanoma   7   6  

Adenocarcinoma   2   0  Not  Indicated   0   1  

Total   11   11  

Repertoire  Features  

Clones  Detected   32916  (5168-­‐56231)   30015  (5894-­‐58222)  TCR  Convergence   0.022  (.006-­‐  .092)   .008  (.002-­‐.019)  

Evenness   .760  (.624  -­‐  .945)   .867  (.673-­‐.945)  

Introduction

Abstract Methods Results Results

Conclusions

References

1.  Looney et al. Haplotype Analysis of the TRB Locus by TCRB Repertoire Sequencing (2018). bioRxiv 406157

2.  Schirmer et al. Illumina error profiles: resolving fine-scale variation in metagenomic sequencing data (2016). BMC Bioinformatics

Shannon Diversity cor =.92, pval =.001

Evenness cor =.92, pval =.001

Pre-treatment TCR convergence vs response

Pre-treatment TCR evenness vs response

Logistic regression scores for responders and non-responders

ROC for prediction of response by TCR convergence and evenness

Table 1. Types of antigens measured by tumor mutation burden and TCR convergence.

Antigen Source Tumor

Mutation Burden

TCR Convergence

Non-Synonymous Mutations ✔ ✔ Aberrant Post-Translational

Modifications ✖ ✔ Ectopic Gene Expression ✖ ✔

Splicing Defects ✖ ✔ Autoantigens ✖ ✔

Virus-Derived Antigens ✖ ✔

Seq

uent

a/A

dapt

ive

●

●●

●

●

●●

●●

0.0 0.2 0.4 0.6 0.8 1.00.

00.

20.

40.

60.

81.

0

cor= −0.29 , pvalue= 0.447

% of exisiting clones from TF

% o

f exi

sitin

g cl

ones

from

Seq

uent

a

●

●

●

●

●

●●

●

●

0.00 0.05 0.10 0.15 0.20

0.00

0.05

0.10

0.15

0.20

cor= 0.92 , pvalue= 0

% of overlap clones from TF

% o

f ove

rlap

clon

es fr

om S

eque

nta

●

●

●

●

●

●●

●

●

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0

cor= −0.17 , pvalue= 0.663

% of new clones from TF

% o

f new

clo

nes

from

Seq

uent

a

Clonal Overlap cor =.92, pval < .001 S

eque

nta/

Ada

ptiv

e

●

●

●

●●

●●

●

0.005 0.015 0.025

0.00

50.

015

0.02

5

cor= 0.57 , pvalue= 0.141

Convergent frequency from TF

Con

verg

ent f

requ

ency

from

Seq

uent

a

TCR Convergence cor =.57, pval =.141

Thermo Fisher Thermo Fisher

PD OCR

0.6

0.7

0.8

0.9

1.0

Pre-treatment TCR Evenness vs Response

Evenness

Eve

nnes

s

Con

verg

ence

PD OCR PD OCR

PD OCR

0.0

0.2

0.4

0.6

0.8

1.0

Logistic Regression Scores for Responders and non-Responders

Score

ROC for Prediction of Response by TCR Convergence and Evenness

Specificity

Sensitivity

1.0 0.8 0.6 0.4 0.2 0.0

0.0

0.2

0.4

0.6

0.8

1.0

Model

TCR Convergence, AUC= .77TCR Evenness, AUC = .74Convergence + Evenness, AUC = .89

Mod

el S

core

PD OCR

Sen

sitiv

ity

Specificity

p = .035, Wilcoxon p = .055, Wilcoxon

p = .001, Wilcoxon