TECH2P.930
High-throughput pairing of T cell receptor
alpha and beta sequences
AAI Annual Meeting
May 8 – 12, 2015
New Orleans, LA USA
Anna Sherwood1†, Bryan Howie1†, Ashley Berkebile1, Jan Berka1, Ryan Emerson1, David Williamson1,
Michael McCormick1, Ilan Kirsch1, Marissa Vignali1, Mark Rieder1, Christopher Carlson1,2, Harlan Robins1,2*
1Adaptive
Biotechnologies Corporation, Seattle, WA; 2Fred Hutchinson Cancer Research Center, Seattle, WA
INTRODUCTION
The  T cell receptor (TCR) protein, which determines the
antigenic specificity of an  T cell, is a heterodimer
composed of two peptides: a longer  chain (TCRB) and a
shorter  chain (TCRA).1 Conventional high-throughput
immunosequencing methods profile TCR alpha (TCRA) and/or
TCR beta (TCRB) repertoires, but do not enable pairing of the
cognate subunits that comprise functional TCRs.2 In order to
reconstitute T cell receptors for functional analysis, therapeutic
use, or modeling of receptor-antigen binding, the  and 
chain from a complete TCR must be identified as a pair. While
published methods using single-cell approaches exist, we
present technically simpler method that uses combinatorics
rather than physical isolation. We call this method, pairSEQ.
The method leverages the diversity of TCR sequences to
accurately pair hundreds of thousands of TCRA and TCRB
segments in our single experiment while using standard
laboratory consumables and equipment. To validate and show
application are aims are:
• Test the pairing algorithm’s False Discovery Rate (FDR)
by comparing calculated false pairs to empirically
measured false pairs.
• Test if the method can be tuned to detect clones of
interest by setting up three experiments to detect: the
most abundant clones, clones with a frequency of
1:10,000, and clones with a frequency <1:10,000.
• Detect the alpha and beta cognate pairs of Tumor
Infiltrating Lymphocytes (TILs) from solid tumors and
matching blood.
MATERIALS AND METHODS
PairSEQ process (Fig 1): Cells are randomly distributed
across a 96 well plate, number of cells/well varies by
experiment (Table 1). From each well, RNA is extracted and
reverse transcribed into cDNA, then the TCRA and TCRB
repertoires are amplified using a multiplex PCR. A wellspecific barcode and sequencing adapters are added to each
PCR amplicon. The TCR repertoire from each well is
sequenced using an illumina sequencer, and cognate pairs
are identified based on alpha and beta sequences sharing
more wells than expected by chance (Equation 1).
MATERIALS AND METHODS
RESULTS
Samples:
False discovery rate
• Experiments 1-3: PBMCs were isolated from ~100 mL of
blood collected from two healthy consented adults.
• Experiments 4 and 5: Blood and matched tumors from 9
consented donors were collected by Conversant Bio
(Huntsville, AL). PBMCs were isolated from the blood and
tumors were dissociated with a miltenyi gentleMACs.
• Concordance between estimated and observed FDR is
extremely high.
• For Experiment 1 (2000 cells/well/subject):
Experimental set-up:
• Cells are normalized to 5,000 T cells/μL in RNAlater, and
distributed across a 96-well plate according to experimental
design (Table 1).
• Experiment 1 and 2: To directly measure the FDR, PBMCs from
sample X and Y are mixed and distributed across a 96-well
plate at 2,000 cells/well and 80,000 cells/well (Fig. 2).
• Experiment 3: To identify the cognate pairs of rare clones,
PBMCs from sample X were distributed across 96 wells at
160,000 cells/well.
• Experiment 4: To identify the cognate pairs of TILs, dissociated
cells from 9 tumors were mixed and distributed across a
96 well plate.
• Experiment 5: PBMCs from matched blood were distributed
across a 96-well plate.
• At 1% FDR (Fig 3):
• Detected 1621 X/X and 1616 YY pairs.
• Expect 32 ((1621+1616)*.01) false pairs, 16 of which are X/Y
or Y/X and 16 of which are un-observable XX or YY.
• Observed 16 X/Y pairs.
• Model matches the observed number of false pairs.
• Across range (0.1% to 10%) of FDR (Fig 3B):
• Observed and estimated number of false pairs (X/Y) is
concordant across a large range.
Detecting cognate pairs from clones across a large
frequency spectrum
• Combined, paired 362,528 TCRB sequences with at least 1
TCRA sequence in experiments 1, 2, and 3, (Fig 5A).
• Together, experiments 1, 2, and 3 detected, as measured by
TCRB immunosequencing (Fig 5B) most of the 10,000 most
abundant clones.
• The number of T cells/well and the number of wells
determines which frequency of clones are “pair-able” and the
upper bound of detectable clones (Fig 5C).
Fig. 5. High-throughput pairSEQ experiment results
A
B
Fig. 3. Validation of pairSEQ FDR
Table 1. Summary of experiments and pairing results
C
Experiment
Subjects
Input T cells
per well
Pairs called at
FDR=1%
1
X and Y
2,000 per subject
4,143
2
X and Y
80,000 per subject
155,805
3
X
160,000
212,651
4
9 tumors
tailored to each
sample*
6,172
5
9 matched
PBMC
tailored to each
sample*
14,123
*For each tumor or blood sample, TCRB repertoire frequencies from the relevant tissue were used
to choose a number of input T cells for each of the 9 samples such that common clones were
likely to be paired.
Fig. 2: FDR validation experiment
Fig 3. Validation of pairSEQ FDR using mixed
PBMCs from two subjects: False discovery rate
curves for an experiment in which PBMCs from
two subjects (‘X’ and ‘Y’) were mixed. Separately,
~6 million cells from each subject were sequenced
to identify TCRB and TCRA sequences unique to
one subject or the other. Pairs are split into groups
named ‘X/X’ (blue), ‘Y/Y’ (orange), and ‘X/Y’ grey).
The red dotted line shows the cutoff for an
estimated FDR of 1%.
Fig. 4: Comparison of estimated and observed FDR
Fig. 1: pairSEQ approach
(A) A total of 362,528 pairs of TCR
sequences were called in Subject X
and Subject Y. (B) Percentage of
paired TCRB sequences among the
N most frequent, between 10^2 and
10^4, in Subject X. (C) Repertoire
frequency distributions of paired
TCRB sequences from Subject X in
Experiments 1-3. Clone frequencies
were estimated by the immunoSEQ
assay using in-frame, expressed
gDNA sequences.
Pairing tumor infiltrating lymphocyte (TILs) TCRs
(Table 2)
• Paired 3,284 TCRs from TILs from 9 tumor samples.
• Number of pairs per sample ranged from 6 to 1,166.
• Paired 7,492 TCR pairs from 9 blood samples.
• Pairs in tumor and blood with same TCRB were paired
with the same TCRA regardless of source.
Table 2. Pairing results from nine tumors and matched
PBMC samples
Fig.1 Schematic of the pairSEQ approach: A fixed number of T cells is
randomly allocated to each well on a 96-well plate, where their mRNA is
extracted, converted to cDNA, and amplified by TCR-specific primers.
Well-specific bar codes are attached and the TCR molecules are pooled
for sequencing, followed by computational de-multiplexing to map each
TCR sequence back to the wells in which it originated.
Fig 4 Predicted versus empirical log10 FDR in
Experiment 1. Predicted FDR values were
provided by pairSEQ statistics, whereas empirical
FDR values were computed as twice the number
of X/Y pairs divided by the total number of called
pairs, under the assumption that X/Y pairs
represent half the total number of errors. As in (A),
the red dotted line shows the cutoff for an
estimated FDR of 1%.
Equation 1: pairSEQ approach
Fig 2. Schematic of FDR validation experiment (Experiment 1):
If each well contains the same number of T cells, the probability of
seeing this amount of well-sharing by chance is
Peripheral blood was collected from two subjects, X and Y, and deep
immunosequencing was used to characterize the TCRA and TCRB
repertoire of each subject. PBMCs from the two subjects were then
mixed, and the resulting mix was used to perform a pairSEQ experiment.
True-positive pairs must include a TCRA and a TCRB from the same
subject, while approximately half of false-positive results will be crosssubject TCRA/TCRB pairs.
References
1. N. R. Gascoigne, Y.-H. Chien, D. M. Becker, J. Kavaler, M. M. Davis, Genomic
organization and sequence of T-cell receptor β-chain constant-and joining-region
genes. Nature 310, 387-391 (1984).
2. H. S. Robins, P. V. Campregher, S. K. Srivastava, A. Wacher, C. J. Turtle, O. Kahsai, S.
R. Riddell, E. H. Warren, C. S. Carlson, Comprehensive assessment of T-cell receptor
beta-chain diversity in alphabeta T cells. Blood 114, 4099-4107 (2009); published
online EpubNov 5 (10.1182/blood-2009-04-217604).
Tumor
sample
Tumor
pairs
PBMC
pairs
Pairs with
TCRB
Pairs with same
TRCB and TCRA
Breast1
6
13
0
0
Breast2
337
95
19
19
Breast3
188
1,782
67
66
Breast4
185
4,906
66
62
Kidney1
189
186
24
23
Kidney2
364
261
77
77
Kidney3
509
53
19
19
Kidney4
1,166
33
19
19
Lung1
340
163
35
35
Total
3,284*
7,492**
326
320
*2,888 pairs called at FDR=1% are not included in this count because
they could not be unambiguously assigned to samples of origin.
**6,631 pairs called at FDR=1% are not included in this count because
they could not be unambiguously assigned to samples of origin.
CONCLUSION
The pairSEQ process enables rapid reconstruction of native
TCRs using standard laboratory equipment and techniques.
• Accurate and tunable:
• Predicted FDR matches empirical data.
• Experiments can be designed to target clones of interest.
• High throughput:
• Identified >212 K unique TCR pairs in a single 3-day
experiment.
• Able to target TILs.
†These
authors contributed equally.
*For more information contact [email protected]
1551 Eastlake Ave E, Ste 200 • Seattle, WA 98102 • adaptivebiotech.com
© 2015 Adaptive Biotechnologies Corporation. All Rights Reserved.