Nature | Vol 579 | 12 March 2020 | 277
The published analysis demonstrated that new clones appear after
treatment, with clonotypes distinct from those detected before treat-
ment. We extended the analysis by matching novel clones in tumours
after treatment with pre-existing clones in blood before treatment, as
assayed by bulk TCR sequencing (TCR-seq) (Extended Data Fig. 7a). We
found significant (P = 2.3 × 10−8 and P = 1.3 × 10−30) correlations in clone
size and transcript counts among novel CD8+ cells (Fig. 3a, Extended
Data Fig. 7b) but not novel CD4+ cells (Extended Data Fig. 7c), which
suggests that at least some of the new CD8+ clones originated periph-
erally. Moreover, although patients differed substantially in number
and expansion of new clones, these differences associated with the
degree of clonal diversity in peripheral blood (Fig. 3b, Extended Data
Fig. 7d), further supporting a relationship between peripheral and
intratumoural clonal expansion.
We further categorized new tumour clones based on whether they
shared TCR sequences with blood before treatment, with the caveat
that incomplete sampling of TCRs in blood (Extended Data Fig. 7e)
may misidentify some clones as being blood-independent. Because
bulk TCR-seq measures total TCR transcripts rather than cells with
a given TCR, we could not discriminate blood-expanded from non-
expanded clones, but we could distinguish blood-independent from
blood-associated clones by the absence or presence of a matching
bulk TCR-seq transcript. Notably, in the two patients with greatest
peripheral clonal diversity, we observed a statistically significant
(P = 3.8 × 10−16 and P = 2.2 × 10−11) interaction between blood association
and exhaustion phenotype (Fig. 3c): non-exhausted clones were more
likely to be blood-associated, whereas exhausted clones were more
likely to be blood-independent. Furthermore, non-exhausted clones
that associated with blood had significantly (P = 3.1 × 10−5, P = 5.2 × 10−12
and P = 2.0 × 10−14) greater clone sizes in tumours after treatment than
those independent of blood. Although patients available for study
were limited, our observations suggest two possible modes of clonal
expansion: (1) a local expansion independent of blood (or possibly
stemming from peripheral events in the remote past) that generates
exhausted T cells; and (2) an infiltrative process from contemporary or
relatively recent blood content that introduces novel, non-exhausted
clones into the tumour.
The prevalence of infiltrating T cells, which may be reactive not only
to the tumour but also to non-tumour antigens, could account for recent
observations of ‘bystander’ T cells that are reactive against non-tumour
antigens^16 and might help explain recent findings that only a small
fraction of intratumoral CD8+ T cells are reactive against tumour^17. We
compared our CDR3 amino acid sequences against databases of CDR3
sequences that are known or likely to react to common viral antigens.
Although TCRs often matched those corresponding to common viral
antigens, almost all of these TCRs belonged to singleton clones and there-
fore showed no evidence of clonal expansion. A few TCRs in each patient
did exhibit high levels of clonal expansion (Extended Data Fig. 8a),
which suggests that although bystander T cells may be prevalent at
the cellular level, relatively few clones may be responding to non-
tumour antigens. We observed over-representation of virally matching
TCRs among dual-expanded clonotypes (P = 3.7 × 10−5 and P = 0.0018)
(Extended Data Fig. 8b) and various CD8 clusters (Extended Data
Fig. 8c), which is consistent with the infiltration of clonally expanded
bystander T cells from the periphery.
As our analysis indicates, patients vary in their extent of peripheral
clonal expansion and resulting infiltration of T cells, and this variability
could potentially explain differing clinical responses to immune check-
point blockade. We evaluated bulk RNA-seq tumour samples from three
randomized phase II trials of the anti-PDL1 antibody atezolizumab^18 –^20
using gene signatures of tissue expansion patterns involving tumour tis-
sue: tumour singletons, tumour multiplets, and dual-expanded clones
(Extended Data Fig. 9a, b, Supplementary Table 5). Because clonal
expansion occurs primarily with CD8+ T cells, these gene signatures
correlated highly with CD8A expression (Extended Data Fig. 9c)—an
indicator of intratumoural prevalence of CD8+ T cells and a reported
predictor of response to immunotherapy^21 (Extended Data Fig. 9d).
Nevertheless, the potential exists for improving on CD8A expression
as a predictor, exemplified by the slightly stronger association of pro-
gression-free survival (PFS) with the expression of CCL5 (also known
as RANTES)—a marker of T cell activation^22 that ranked highly in both
tumour multiplet and dual-expansion signatures—in treatment arms
in which atezolizumab is active (with monotherapy in the IMmotion150
trial thought to have been mitigated by immunosuppressive myeloid
cells^20 ) (Extended Data Fig. 9e).
Similarly, gene signatures for tumour multiplets and dual expansion
were associated with greater PFS in relevant arms, both by the hazard
ratios of individual genes (Fig. 4a) and by the survival analysis of com-
bined signature scores (P ≤ 0.02) (Extended Data Fig. 10a). To handle
confounding expression of CD8A, we performed survival analyses
of CD8A expression and expansion signature scores as dual variates
(Fig. 4b, Extended Data Fig. 10b). Although most patients were concord-
ant in both measures, the relatively few discordant patients with high
CD8A expression and low expansion signature scores were anomalous
and therefore informative. In every atezolizumab-containing arm,
the discordant group had worse PFS than the concordant group, with
a hazard ratio > 1 in several cases. Furthermore, a similar dual-variate
analysis of tumour multiplet and dual expansion signatures showed
additive predictive power (Extended Data Fig. 10c), which suggests that
these two types of clonal expansion may contribute independently in
the clinical response of patients to cancer immunotherapy.
In summary, dual-expanded T cell clones in tumours, and their deri-
vation from peripherally expanded clones, may represent the overall
1550
Exhausted
Non-exh
Exhausted
Non-exh
bcc.su002
1550
bcc.su007
1550
bcc.su006
1550
bcc.su005
1550
bcc.su001
1550
bcc.su008
0
1,000
4.2 4.6 5.0 5.4
Novelclones
n = 6 P = 0.024
0
2,000
Novel 4.2 4.6 5.0 5.4
cells
Blood clonal diversity
n = 6 P = 0.029
bcc.su001 Novel CD8
0210 50
50
n = 176 r = 0.4 P = 4.6 × 10–8
bcc.su008 Novel CD8
0550 500
n = 474 r = 0.49 P = 6.2 × 10–30
Blood+
Blood–
CD8_eff CD8_mem CD8_act CD8_ex CD8_ex_act
Non-exhausted Exhausted
Clone size,post-rx tumour
No. of transcripts, pre-rx blood
post-rx tumourClone size,
Pn = 0.88 = 30 Pn = 0.068 = 25 Pn = 0.76 = 381 nP = 176 = 1.0 P = 3.8 × 10n = 633–16P = 2.2 × 10n = 1,320–11
P = 5.2×
10
–12
P = 2.0×
10
–14
P = 3.1×
10
–5
n = 560
672
n = 220
264
n = 135
245
a
c
b
5
1
50
5
1
Fig. 3 | Peripheral clonal expansion and novel intratumoural clones.
a, Correlated clone sizes in tumour and blood. Scatter plots of tumour clone
size after treatment (post-rx) and transcript counts in blood before treatment
(pre-rx) are shown for the two patients in Yost et al.^5 with bulk TCR-seq blood
samples and the largest numbers of novel clones. Vertical lines separate blood-
independent (left) and blood-associated (right) clones. Clones are coloured by
primary cluster from the original analysis, as in c. Two-sided P values are shown
from Pearson’s correlation coefficients r on log-transformed values from novel
blood-associated CD8 clones. b, Blood clonal diversity and novel clones. Data
from Extended Data Fig. 7b, d are summarized, showing the total numbers of
novel CD8 clones (top) and cells (bottom) after treatment across patients as a
function of blood clonal diversity (average of pre- and post-treatment
measurements). Two-sided P values are shown from a linear regression t-test
across patients, on the logarithm of y values to linearize the data. c, Non-
exhausted new clones and blood association. Novel clones in Yost et al.^5 are
separated by pre-treatment blood association (Blood+) or blood independence
(Blood−), and further divided by exhaustion or non-exhaustion of their post-
treatment primary cluster, as defined and coloured in the legend. Patients are
ordered by increasing blood clonal diversity. P values (top) are from a chi-
square test on counts of new CD8 clones. P values (side) are from a two-sided
t-test on log-transformed clone sizes (shown if P < 0.01).