Immune-awakening revealed by peripheral T cell dynamics after one cycle of immunotherapy.

Our understanding of how checkpoint inhibitors (CPI) affect T cell evolution is incomplete, limiting our ability to achieve full clinical benefit from these drugs. Here we analyzed peripheral T cell populations after one cycle of CPI and identified a dynamic awakening of the immune system revealed by T cell evolution in response to treatment. We sequenced T cell receptors (TCR) in plasma cell-free DNA (cfDNA) and peripheral blood mononuclear cells (PBMC) and performed phenotypic analysis of peripheral T cell subsets from metastatic melanoma patients treated with CPI. We found that early peripheral T cell turnover and TCR repertoire dynamics identified which patients would respond to treatment. Additionally, the expansion of a subset of immune-effector peripheral T cells we call TIE cells correlated with response. These events are prognostic and occur within 3 weeks of starting immunotherapy, raising the potential for monitoring patients responses using minimally invasive liquid biopsies."

Introduction

T cell maturation begins when pro-T cells enter the thymus and attempt to generate a functional T cell receptor (TCR). Cells that fail to do so are eliminated through β-selection, but those that succeed must still pass positive and negative (+/-) selection for HLA binding and absence of reactivity to self-antigen in order to survive (Extended Data Fig. 1a,b). Successful naïve T cells enter the circulation as early thymic emigrants (ETE), and if stimulated by antigen presenting cells in the lymphatic system, they expand and migrate to sites of inflammation to combat harmful agents, a process that resolves through clonal contraction and more T cell death[1] (Extended Data Fig. 1c-e).

Extended Data Fig. 1

Schematic summarizing T cell maturation and life-cycle.

a Pro-T cells undergo sequential somatic recombination of their T cell Receptor β (TCR) loci in attempts to generate functional TCR with unique CDR3 antigen binding regions. Cells that fail to generate a functional TCRβ at the first attempt can recombine their second TCR allele, but cells which fail to produce a functional TCR at the end of the process (crossed red box) are eliminated (β-selection) and their DNA, which encodes the CDR3 unique regions, enters the blood as circulating cell-free DNA (cfDNA). Surviving cells retain the T cell receptor excision circle (TREC) generated during TCR locus rearrangement as an episome in the nucleus. The TREC does not replicate so is diluted during subsequent cell divisions. b T cells with a functional TCR undergo positive and negative selection (+/- selection) for HLA and self-antigen recognition. The CDR3 DNA from T cells eliminated during this step is released into the blood. c Naive T cells enter the circulation as early thymic emigrants (ETE). d T cells primed by antigen presenting cells (APC) in the lymphatic system undergo clonal expansion, which dilutes the TREC amongst the daughter cells. e T cell homeostasis is maintained by subsequent contraction (turnover cycles), releasing further CDR3 DNA into the blood.

Checkpoint inhibitor (CPI) drugs awaken the immune system so that they attack tumors. CPI have revolutionized cancer care and, over the last decade, have contributed to a 4-fold improvement in the survival of metastatic melanoma patients[2]. Despite these remarkable advances, our understanding of T cell evolution under the selective pressure of CPI is still incomplete and that limits our ability to derive full clinical benefit from these drugs. Consequently, most patients with advanced stage melanoma still die of their disease. Sharing features with responses to infectious diseases[3-5], tumor control by the immune system requires coordination between systemic and intra-tumoral immunity[6], and although several studies have investigated intra-tumoral responses to immunotherapy[7-9], few have focused on how CPIs affect the peripheral immune system, or whether changes in peripheral T cells are associated with patient responses[7,8,10].

We hypothesized that because immune responses to tumors mirror normal defensive responses to pathogens, it would be possible to study patient responses to CPI by monitoring peripheral T cell evolution during treatment. T cell receptors (TCR) are generated by error-prone recombination of the TCR locus[11,12], creating the enormous diversity needed for effective immune function. This process is ~80% efficient, so most peripheral T cells carry only productive TCR sequences. However, in ~20% of peripheral T cells the first attempts at TCR locus rearrangement failed due to acquisition of stop codons or because the protein coding region was out of frame, so these T cells carry both productive and non-productive TCR sequences[13]. The complementarity determining region-3 (CDR3) of the TCR in particular is highly variable and the sequences are unique to individual T cell clones, so both the productive and non-productive TCR sequences serve as a “fingerprints” for individual T cell clones.

We posited that by sequencing peripheral T cell CDR3 regions, we could track T cell responses to CPI and, because dying cells release their DNA into the circulation, we could also sequence CDR3 regions in cell free DNA (cfDNA) in the blood to monitor T cell turnover in patients receiving CPI. We found an increase in productive TCR sequences in plasma cfDNA of patients who responded to CPI, and this correlated with response. These events were accompanied by evolution of the peripheral T cell repertoire in a manner that mimicked changes induced by anti-viral vaccines. The dynamics of T cell turnover revealed by the cfDNA analysis also correlated with expansion of a specific subset of cytotoxic memory effector peripheral T cells we call immune-effector or TIE cells. Importantly, TIE cell expansion after one cycle of CPI anticipated which patients would go on to respond to treatment. Our data reveal an awakening of the immune system that occurs within 3 weeks of initiating CPI and which anticipates clinical response to first-line therapy. These changes are dynamic and quantifiable and can be monitored with minimally invasive liquid biopsies, features that could be used to identify which patients will benefit from CPI early during their treatment, allowing delivery of more precise treatment planning.

Results

Immunotherapy does not alter thymic output

First, we examined the effects of CPI on thymic function. We used fluorescent-activated cell sorting (FACS, Extended Data Fig. 2) to quantify the ETE (CD3+/CD45RA+/CD45RO-/CCR7+/CD27+/CD31+ T cells[14]) in peripheral blood mononuclear cells (PBMC) from 50 metastatic melanoma patients (#1-50) receiving first-line anti-PD1 or anti-PD1/anti-CTLA4 treatment (Extended Data Fig. 2i). As expected[15], we observed an age-related decrease in ETE levels in pre-treatment (T0) patient blood (Fig. 1a), but we also found that one cycle of CPI did not affect ETE levels, measured at week 3 (W3)(P=0.274; Fig. 1b). Next, we examined the TCR excision circle (TREC) in the peripheral T cells of 16 of our patients (#1,10-13,22,24-27,30,42,51-54). The TREC, a by-product of TCR locus rearrangements, is a non-replicating episome that is diluted when T cells divide[16] (Extended Data Fig. 1a-d). We found that the TREC:genome ratio in T cells was not affected by CPI (P=0.129, Fig. 1c).

Extended Data Fig. 2

Gating strategy for the identification of T cell subsets in peripheral blood of melanoma patients.

Multiparametric fluorescence activated cell sorting analysis using the indicated gates. a Lymphocyte gate on side scatter/forward scatter; b single cell gate to exclude doublets; c live gating to exclude dead cells from subsequent gates; d CD3+ gate for T cells; e,f CD4+ and CD8+ gates for “helper” and “killer” T cell subsets, CD8 was detected with a PE-Cy7 labelled antibody for the Treg panel (e) and with a FITC labelled antibody for the T maturation panel (f); g CD4+/CD25+/CD127-/low regulatory T cells (Treg); h naive (top left) and memory (bottom right) gates total T cells; i ETE (top) and CD31-naive (bottom) gates for naive T cells; j naive (top left) and memory (bottom right) gates for CD8+ T cells; k CD8+ memory T cell subsets, the left bottom subset (CCR7-/CD27-) represents the TIE cells.

Figure 1

CPI induced peripheral TCR repertoire divergence.

a Graph showing early thymic emigrants in pre-treated patients’ blood (% ETET0 relative to total naive T cells; determined by FACS) relative to age (P=0.002, linear regression R2=-0.17; n=50). b Levels of ETE in pre-treatment (T0) and week 3 (W3) of CPI in paired patient samples (P=0.274, two-sided Wilcoxon test, n=50). c TREC (T cell receptor excision circle) concentration relative to genomic DNA was measured by droplet digital PCR in sorted CD3+ peripheral T cells at T0 (median 0.5x10-3) and W3 (median 0.1x10-2)(P=0.129, two-sided Wilcoxon test, n=17). d Tumor infiltrating T lymphocyte (TIL) CDR3 sequences also present in peripheral PBMC and cfDNA for one patient at T0 and W3. See Table 1 for specific DNA sequence; tot = total. e Venn diagram showing unique predicted productive CDR3 sequences in PBMC and TIL for patient #01 at T0 (Supplementary Table 1)[18]. Numbers show unique nucleotide sequence counts for PBMC-private (pink), TIL-private (brown) and tePBMC (tumor emigrant PBMC; intersection, orange) pools. f Clonal relatedness (the proportion of amino acids sequences that are related by maximum edit distance=3) for CDR3 in the PBMC-private pool, tePBMC and TIL-private pools at T0. Horizontal lines: comparison of clonal relatedness between PBMC-private and TIL-private TCR sequences at T0; ***: P=0.003; n=18, two-sided Wilcoxon test; median=0.4x10-6 and 0.4x10-3, respectively; ****: P<0.0001; n=18, two-sided Wilcoxon test; median=0.4x10-6 and 0.2x10-2, respectively. g Clonal relatedness (maximum edit distance=3 amino acids) for CDR3 sequence in PBMC TCR pools at T0 and W3. Comparison between the clonal relatedness of PBMC-private TCR of patients with progressive disease (PD, orange, n=11, median=4.3x10-5 and 5.6x10-5, respectively) or disease control (DC, green, n=7, median=4.0x10-5 and 8.0x10-5, respectively) after 12 weeks of treatment; ns: not significant (P=0.413 and P=0.999, two-sided Wilcoxon test) and between the clonal relatedness of tePBMC TCR of patients with PD (n=11) or DC (n=7); ns: not significant (P=0.638; two-sided Wilcoxon test; median=0.002 and 0.0008, respectively); *: P=0.031; n=7, two-sided Wilcoxon test; median=0.0017 and 0.0007, respectively). Dot is one patient; line is median; error bar is standard deviation; connecting line is paired samples; ns indicates not significant P values, n represents patients.

CPI induce TCR repertoire divergence in peripheral T cells

The observations above indicate that CPI did not affect thymic output in melanoma patients, so to monitor how CPI affected post-thymic T cell evolution, we analyzed the TCR in peripheral PBMC and melanoma metastases. For patient #12 we obtained a fresh tumor biopsy at T0, and whole blood at T0 and after the first cycle of CPI at W3. Using ImReP[17] we identified 16 unique CDR3 DNA sequences from the biopsy and found that 6 of these were also present in the PBMC and cfDNA (Fig. 1d, Table 1). Thus, about one third of the sequences in the tumor were also in the periphery, four in pre-treatment PBMC (sequences CDR3, CDR3, CDR3, CDR3), three in the W3 PBMC (CDR3, CDR3, CDR3), and one in the W3 cfDNA (CDR3)(Table 1). Intriguingly, CDR3 and CDR3 both encoded TCR CDR3prot1 (Table 1), and CDR3 and CDR3 both encoded CDR3prot2 (Table 1), suggesting convergence by these TCR on dominant tumor antigens. We also analyzed CDR3 sequences in 18 paired PBMC and TIL from a published melanoma cohort[18]. As an example, at T0 patient #01 presented 123,981 unique CDR3 sequences in the bulk PBMC, 21,052 in the TIL and 3,741 shared sequences (Fig. 1e), comparable patterns were seen in the other patients (Supplementary Table 1).

Table 1

Tumor infiltrating T lymphocyte (TIL) CDR3 sequences also present in peripheral PBMC and cfDNA for patient #12 at T0 and W3.

The table shows the DNA (CDR3) and their paired predicted protein (CDR3prot1-4) sequences for the TCR that were identified in the pre-treatment tumor biopsy and also in the periphery, either in the T0 or W3 PBMC, or in the W3 cfDNA. For the proteins, the different CDR3 are color-coded, with CDR3prot1 red, CDR3prot2 purple, CDR3prot3 green and CDR3prot4 blue; the black text is the flanking TCR protein sequence. The red underlined bases in CDR3 highlights that CDR3 and CDR3 encode the same protein. Similarly, the red underlined base in CDR3 highlight that CDR3 and CDR3 encode the same protein. *CDR3 sequence count in the biopsy; #CDR3 sequence frequency in the sample.

Peripheral sample	Sequence																																		Count*	Frequency#	V gene name	D gene name	J gene name
T0 PBMC	CDR3prot1	L	E	S	P	S	P	N	Q	T	S	L	Y	F	C	A	S	S	L	Q	G	A	N	Y	E	Q	Y	F	G	P	G	T	R	L
	CDR3DNA1.1	CTGGAGTCGCCCAGCCCCAACCAGACCTCTCTGTACTTCTGTGCCAGCAGCCTACAGGGGGCGAACTACGAGCAGTACTTCGGGCCGGGCACCAGGCTC																																	7	0.000599	TCRBV27-01	TCRBD01-01	TCRBJ02-07
	CDR3DNA1.2	CTGGAGTCGCCCAGCCCCAACCAGACCTCTCTGTACTTCTGTGCCAGCAGCCTACAGGGGGCGAACTACGAGCAGTACTTCGGGCCAGGCACGCGGCTC																																	1	8.56E-05	TCRBV27-01	TCRBD01-01	TCRBJ02-05
	CDR3prot2	L	T	V	T	S	A	H	P	E	D	S	S	F	Y	I	C	S	A	R	G	P	E	E	T	Q	Y	F	G	P	G	T	R	L
	CDR3DNA2.1	CTGACAGTGACCAGTGCCCATCCTGAAGACAGCAGCTTCTACATCTGCAGTGCCCGGGGTCCGGAAGAGACCCAGTACTTCGGGCCAGGCACGCGGCTC																																	1	8.56E-05	TCRBV20	TCRBD02-01	TCRBJ02-05
	CDR3prot3	S	D	L	E	L	G	D	S	A	L	Y	F	C	A	S	S	V	G	R	S	S	Y	N	E	Q	F	F	G	P	G	T	R	L
	CDR3DNA3	AGCTCTCTGGAGCTGGGGGACTCAGCTTTGTATTTCTGTGCCAGCAGCGTAGGGAGAAGCTCCTACAATGAGCAGTTCTTCGGGCCAGGGACACGGCTC																																	1	8.56E-05	TCRBV09-01	TCRBD02-01	TCRBJ02-01
W3 cfDNA	CDR3prot4	S	A	H	P	E	D	S	S	F	Y	I	C	S	A	R	D	W	L	A	G	D	T	G	E	L	F	F	G	E	G	S	R	L
	CDR3DNA4	AGTGCCCATCCTGAAGACAGCAGCTTCTACATCTGCAGTGCTAGAGATTGGCTAGCGGGAGACACCGGGGAGCTGTTTTTTGGAGAAGGCTCTAGGCTG																																	2	0.0069444	TCRBV20	TCRBD02-01	TCRBJ02-02
W3 PBMC	CDR3prot1	L	E	S	P	S	P	N	Q	T	S	L	Y	F	C	A	S	S	L	Q	G	A	N	Y	E	Q	Y	F	G	P	G	T	R	L
	CDR3DNA1.1	CTGGAGTCGCCCAGCCCCAACCAGACCTCTCTGTACTTCTGTGCCAGCAGCCTACAGGGGGCGAACTACGAGCAGTACTTCGGGCCGGGCACCAGGCTC																																	10	0.0016176	TCRBV27-01	TCRBD01-01	TCRBJ02-07
	CDR3prot2	L	T	E	T	S	A	H	P	E	D	S	S	F	Y	I	C	S	A	R	G	P	E	E	T	Q	Y	F	G	P	G	T	R	L
	CDR3DNA2.1	CTGACAGTGACCAGTGCCCATCCTGAAGACAGCAGCTTCTACATCTGCAGTGCCCGGGGTCCGGAAGAGACCCAGTACTTCGGGCCAGGCACGCGGCTC																																	1	0.000162		TCRBD02-01	TCRBJ02-05
	CDR3DNA2.2	CTGACAGTGACCAGTGCCCATCCTGAAGACAGCAGCTTCTACATCTGCAGTGCCCGGGGTCCGGAAGAGACTCAGTACTTCGGGCCAGGCACGCGGCTC																																	1	0.000162	TCRBV20	TCRBD02-01	TCRBJ02-05

These data established that tumor resident T cell clones were also present in the periphery, so we called these cells tumor-emigrant PBMC (tePBMC). We compared the clonal relatedness[19,20] of the CDR3 regions from the tePBMC to the PBMC-private and TIL-private pools. At T0, the tePBMC and TIL-private TCR pools displayed more clonal relatedness than the PBMC-private CDR3 regions (Fig. 1f), suggesting more TCR convergence in tumor-associated T cells than the bulk PBMC population. At week 3, we did not observe differences in clonal relatedness in the PBMC-private or TIL-private TCR when we compared patients who achieved disease control (DC) at week 12 with those who developed progressive disease (PD) (Extended Data Fig. 3a). Similarly, when we compared clonal relatedness in the PBMC-private TCR at T0 and W3, we did not observe differences between DC or PD patients, whereas in the tePBMC pool, there was a significant reduction in TCR clonal relatedness in patients who achieved PD, but not in patients who had PD (Fig. 1g). This suggests recruitment of T cells with a broader TCR repertoire from the periphery to the tumors of patients who responded, but not to the tumors of patients who did not respond.

Extended Data Fig. 3

Clonal relatedness in tumor infiltrating T cells and PBMC.

a Clonal relatedness changes in PBMC-private and TIL-private TCR pools; comparison of week 3 (W3) CDR3 clonal relatedness in patients with progressive disease (PD, n=11 patients) and disease control at week 12 (DC, n=7 patients) in the PBMC-private (P=0.724, median=0.6x10-6 and 0.6x10-6, respectively; two-sided Mann-Whitney U test) and TIL-private pools (P=0.246, median= 0.5x10-4 and 0.8x10-5, respectively; two-sided Mann-Whitney U test). Dot represents one patient; green indicates DC; orange indicates PD; error bar is standard deviation.

CPI induce peripheral T cell turnover

Next, we compared the CDR3 clonal relatedness in PBMC and cfDNA of 28 CPI-treated metastatic melanoma patients (#11-27,#29-39). In T0 blood from patient #27 we observed 14,112 unique CDR3 sequences in the bulk PBMC, 844 in the cfDNA and 193 shared sequences (Fig. 2a). Comparable patterns were seen in the other 27 patients (Supplementary Table 2). Intriguingly, the number of unique PBMC/cfDNA-shared CDR3 sequences increased after one cycle of CPI (Fig. 2b), so we investigated how CPI affected the immune-recognition landscape in these pools. At T0 the PBMC-private CDR3 clonal relatedness was ~0, but was significantly higher in both the cfDNA-private and PBMC/cfDNA-shared pools (Fig. 2c). Critically, clonal relatedness in the PBMC/cfDNA-shared pool decreased significantly in patients who achieved DC at week 12, but not in patients with PD (Fig. 2d), suggesting repertoire divergence in the T cells that turnover in the responding patients.

Figure 2

CPI induced peripheral T cell turnover.

a Venn diagram showing unique predicted productive CDR3 sequences in PBMC (left, pink), PBMC/cfDNA-shared pool (intersection, purple) and cfDNA (right, blue) for patient #27 at T0 (Supplementary Table 2). b Total number of CDR3 clones at T0 (pink) and W3 (purple) in the PBMC/cfDNA-shared pool (P=0.010, two-sided Wilcoxon test). c Clonal relatedness (maximum edit distance=3 amino acids) for CDR3 in the PBMC-private pool, PBMC/cfDNA-shared pool and cfDNA-private pools at T0. Horizontal lines: comparison of clonal relatedness between PBMC-private and cfDNA-private TCR sequences T0 (P<0.0001; two-sided Wilcoxon test; median was 0.3x10-3 and 0.01, respectively); comparison of clonal relatedness between PBMC-private and PBMC/cfDNA shared TCR sequences T0 (P<0.0001, two-sided Wilcoxon test; median was 0.3x10-3 and 0.06). d Clonal relatedness of CDR3 sequence in PBMC/cfDNA-shared pool at T0 and W3 for patients with progressive disease (the number of patients is 12, PD, orange) or disease control (the number of patients was 16, DC, green) at week 12. Comparison (horizontal lines) of clonal relatedness between: T0 PBMC/cfDNA-shared pool TCR sequences for patients with PD or DC (P=0.623, two-sided Mann-Whitney U test; median is 0.04 and 0.08); W3 PBMC/cfDNA-shared pool TCR sequences for patients with PD or DC (P=0.026; two-sided Mann-Whitney U test; median=0.06 and 0.03); T0 and W3 for the PBMC/cfDNA-shared pool TCR sequences for patients with PD (P=0.733; the number of patients was 12, two-sided Wilcoxon test; median=0.04 and 0.06) or DC (P=0.039; n=16, Wilcoxon test; median=0.08 and 0.03). e Pre-treatment TCR rearrangement efficiency score (REST0) of rearranged CDR3 in healthy donors (HD) and patients on CPI in PBMC (P=0.445; median 0.83 and 0.81; the number of HD was 77 batches and the number of patients was 29; two-sided Mann-Whitney U test) and cfDNA (P=0.09, median 0.44 and 0.62; n=3 and 28; two-sided Mann-Whitney U test). Comparisons (horizontal lines) between: HD PBMC and cfDNA REST0 (P<0.0001, two-sided Mann-Whitney U test); matched samples of patients’ PBMC and cfDNA RES T0 (P<0.0001, Wilcoxon test). f ΔW3RES (change in RES from T0 to W3) according to response group at week 12; **: P=0.008, two-sided Wilcoxon test, median was 0.001 and 0.08; *: P=0.037, two-sided Mann-Whitney U test, median=0.02 and 0.08. Total number of melanoma patients equal to 28; dot is one patient; error bar is standard deviation; connecting line is paired samples; horizontal line is median; T0 indicates pre-treatment; W3 indicates week 3; ns is not significant; * indicates P=0.05-0.01; **** indicates P<0.0001.

The cfDNA CDR3 sequences come from T cell turnover in the thymus and periphery and contain both productive and non-productive sequences (Extended Data Fig. 1). We used ImmunoSeq to quantify productive (reading frame intact) and non-productive (out of frame, stop codon) CDR3 sequences and calculate a Rearrangement Efficiency Score (RES; productive/[productive+non-productive]). In healthy donors (HD) the PBMC RES was ~0.8 as expected[13], but in cfDNA the RES was 0.44 (P<0.001; Fig. 2e), presumably from non-productive TCR sequences released by T cells failing β-selection in the thymus. The T0 PBMC and cfDNA RES (REST0) values were similar in HD and patients (Fig. 2e), suggesting that melanoma does not overtly affect the efficiency of T cell rearrangements in the thymus and also that it does not affect bulk T cell turnover in the periphery. We therefore compared the RES in PBMC and cfDNA at T0 and W3 to generate difference scores (ΔW3RES) and measure how CPI affects T cell rearrangement and turnover during the first cycle of immunotherapy. The PBMC ΔW3RES measured changes in TCR rearrangement efficiency and was ~0 irrespective of whether the patients responded or not (Fig. 2f). Thus, CPI did not affect TCR rearrangement efficiency in the thymus, meaning that the cfDNA ΔW3RES measured the changes in the peripheral T cell turnover alone. Notably, the cfDNA ΔW3RES rose from ~0 in patients with PD to 0.09 (P=0.037) in patients who achieved DC at 12 weeks (Fig. 2f). Thus, CPI increased peripheral T cell turnover in responding patients, but not in non-responding patients.

CPI stimulate expansion of specific peripheral T cell subsets

Our data reveals dynamic TCR repertoire reorganization during T cell expansion/contraction in responding patients, so we used high dimensional FACS to characterize peripheral T cell subsets and monitor T cell evolution under CPI (Extended Data Fig. 2). We found that a CD8+ memory effector cytotoxic T cell subset that we called immune-effector T cells (TIE) expanded proportionally to the cfDNA ΔW3RES (Fig. 3a). TIE cells are characterized by the surface phenotype CD3+/CD4-/CD8+/CD45RA-/CD45ROhigh/CD27-/CCR7- (Extended Data Fig. 1k) and have been shown to be associated with response to infections[21,22].

Figure 3

Identification of TIE cells.

a Correlation between TIE cell abundance (ΔW3TIE) and changes in cfDNA RES (ΔW3RES: RESW3-REST0,) at W3 relative to T0 (P=0.001; linear regression R2=0.34, n=28 patients); dotted line is the linear regression line. b Similarity matrix of TCR sequences in cfDNA and peripheral CD8+ T cell subsets. TCM, TIE Tnaive and ETE similarity with cfDNA in 6 patients at T0 (median=0.026, 0.045, 0.004 and 0.003, respectively; P=0.0013; Friedman analysis of variance, Friedman statistics=15.64; patient #16 naive subset not assessed) and W3 (median=0.043, 0.136, 0 and 0.003, respectively; P<0.0001; Friedman analysis of variance, Friedman statistics=23.05; patient #16 naive subset not assessed). c Clonality (Gini coefficient) in peripheral CD8+ T cell subsets. TIE subset clonality relative to other subsets at baseline (T0, TIE median clonality=0.46) and after the first cycle of CPI (W3, TIE median clonality=0.61) in 6 matched patient samples (P=0.0006 and 0.0002, Friedman analysis of variance Friedman analysis of variance, Friedman statistics=12.6 and 13.08; patient #16 naive subset not assessed); error bar is standard deviation. The small sample size did not allow the comparison between responders (#16-18) and patients who progressed (#12,19,29); horizontal line is median; error bar is standard deviation. d Graph showing the frequency of pre-treatment TIL CDR3 sequences in patient #12 sorted peripheral CD8+ T cell subsets at T0 and W3. Dot in a and c is a single patient.

To study cytotoxic T cell turnover, we sequenced the TCR in T0 and W3 purified CD8+ peripheral memory and naive T cells from 3 PD and 3 DC patients (#12,16-19,29). More than other peripheral T cell subsets, the TIE cells had the highest similarity to cfDNA CDR3 sequences (Fig. 3b) and presented the highest clonality (Fig. 3c). This suggests TIE cells are actively turning over, but with convergence towards dominant clones. Moreover, although it is not yet possible to establish if the TIE cell CDR3 regions recognized neo-epitopes, largely these cells did not express TCR known to recognize public epitopes such as Melan-A or viral proteins (Supplementary Table 3). In patient #12, from whom we obtained a biopsy, we identified identical CDR3 sequences in the tumor and the peripheral TIE cells (Fig. 3d), demonstrating that individual TIE clones coexisted in the tumor and periphery. Moreover, expansion of intratumoral cells with the TIE phenotype is reported to be associated with responses to CPI[23], and we used published CyTOF data[24] to confirm that TIE cells were resident in melanoma and renal cell carcinoma (immune-responsive tumors), but were negligible in glioblastoma (immune-refractory tumor) and tonsils (Fig. 4a-d).

Figure 4

TIE cells infiltrate tumors that respond to immunotherapy.

T-SNE (t distributed stochastic neighbor embedding) plots of biopsy cell clusters according to T cell surface markers in (a) melanoma (n=18 patient samples), (b) renal cell carcinoma (RCC, n=3 patient samples), (c) glioblastoma (GMB, n=3 patient samples), and (d) non cancerous tonsils (n=4 patient samples). Samples for this analysis were from reference #24. Black arrows highlight the clusters with the TIE phenotype.

Next, we analyzed PBMC from 30 CPI-treated metastatic melanoma patients from our cohort (#1-30) and show that the TIE cells expanded at W3 in patients who achieved DC, including late responders, but not patients with PD (P=0.0007; Fig. 5a), irrespective of therapy protocol (P=0.200, Fig. 5b). Notably, despite a W3 TIE expansion of ~20%, the W12 CT scan revealed that patient #20 was progressing (Fig. 5a). Unfortunately, the patient died of complications so a late response could not be measured, but from day 40 we observed a steady decline in this patient’s NRASQ61R circulating tumor DNA (ctDNA)[25] (Fig. 5c), revealing that consistent with the observed expansion in TIE cells, the patient achieved a biochemical response (Fig. 5c). An increase of >0.8% in the TIE ratio relative to all CD8+ memory T cells at W3 was associated with increased overall survival and segregated DC (including late-responders) from PD patients with sensitivity = 0.94 and specificity = 0.79 (accuracy = 0.87 and area under the curve [AUC] = 0.85) (Fig. 5d). The Hazard Ratio for patients without W3 TIE expansion was 3.7 (95% CI 1.12-11.9, P=0.032) (Fig. 5e). We confirmed these observations in an independent cohort of 20 CPI-treated patients (#31-50, Fig. 5f,g), with sensitivity = 0.82 and specificity = 1 (accuracy = 0.90, AUC = 0.92). By week 9, TIE cells no longer discriminated DC from PD patients (Extended Data Fig. 4a). Accordingly, when we analyzed published CyTOF data[3], we confirmed TIE to be a distinct T cell subset in PBMC of patients with metastatic melanoma (Extended Data Fig. 4b,c), but consistent with our findings, the TIE levels at week 12 did not distinguish DC from PD patients (Extended Data Fig. 4d). Thus, changes at W3 were prognostic for melanoma responses to CPI, but were no longer prognostic by W9 or W12, demonstrating the dynamic nature of these responses and consistent with a previous study showing that the peak of immune activation is at W3[7,10].

Figure 5

Peripheral TIE cell expansion in response to first-line CPI.

a ΔW3TIE in patients with best response PD (orange, n=14, median=-0.58%) or best response DC (green, n=16, median=10.04%; ***: P=0.0007, two-sided Mann-Whitney U test) in the training set. Arrow: patient #20. b ΔW3TIE in patients receiving anti-PD1 monotherapy (αPD1, n=18, median=1.35%) or combination of ipilimumab plus nivolumab (I+N, n=12, median=10.84%; P=0.2, two-sided Mann-Whitney U test). c Variations over time (days from treatment initiation) of variant allele frequency (VAF) of mutant NRAS measured in circulating tumor DNA by droplet digital PCR over the course of CPI treatment for patient #20 (arrow in a). Blood collection ceased after day 90 due to patient complications leading to patient death. d Receiver operating curve showing the sensitivity and false positive rate (1-specificity) of ΔW3TIE values in identifying the patients that will achieve disease control; *: maximum accuracy (cut-off=+0.8%). e Kaplan-Meier survival curves for patients with TIE expansion ≥0.8% (pink, n=12; median survival not reached) at W3 compared to patients without TIE expansion <0.8% (blue, n=18; median survival=9.6 months; P=0.013; log rank test), in the training set. f ΔW3TIE in patients with best response PD (orange, n=3, median=-1.3%) or best response DC (green, n=17, median=3.3%; *: P=0.019, two-sided Mann-Whitney U test) in the validation set. g Kaplan-Meier survival curves for patients with TIE expansion≥0.8% (pink, n=15; median survival not reached) at W3 compared to patients without TIE expansion≥0.8% (blue, n=5; median survival=4.2 months; two-sided log rank test, P=0.003), in the validation set. OS is overall survival; dot represents one patient; n represents patients; horizontal line is median; error bar is standard deviation; dotted vertical line is landmark at week 3.

Extended Data Fig. 4

Identification of TIE in CPI-treated patient PBMC.

a Comparison of differential abundance of TIE in CD8+ memory T cells in the PBMC of The Christie NHS Foundation Trust patients with best response progressive disease (PD, orange, n=14) and disease control (DC, green, n=16) at T0 (n=30, light shade) and week 9 (W9; n=10, dark shade; PD, n=4, DC, n=6). Differences over time were not significant for PD (median=15.2 and 35.5; P=0.375; two-sided Wilcoxon test) or DC (median=7.9 and 24; P=0.219; two-sided Wilcoxon test); PD vs DC patient values did not differ at T0 (P=0.275; two-sided Mann-Whitney U test) or W9 (P=0.762; two-sided Mann-Whitney U test). b Distributions of marker intensities of the T cell surface markers in the 20 cell populations (clusters) for PBMC from a published cohort3 (n=20 patients). Cluster 5 was identified as the TIE subset. Blue densities are calculated over all the cells and serve as a reference and red densities represent marker expression for cells in a given cluster. Arrows highlight the TIE subset. c T-stochastic neighbor embedding of single cell profiles (dots) performed in an external cohort3 using the T cell surface markers CD3, CD4, CD8, CD45RA, CD45RO, CCR7 and CD27; different colors are attributed by clustering. Arrow highlights the TIE subset. d Comparison of the differential abundance of the TIE cluster in the PBMC from a published cohort3 of patients with PD (orange, n=9) or DC (green, n=11) at pre-treatment (light shade, n=20; PD, n=9; DC, n=11) and at week 12 (W12, dark shade, n=20) on treatment with pembrolizumab or nivolumab in the external cohort. Horizontal bars indicate the differences over time for the PD (median at T0=5.9 and W12=9.1; P=0.164; two-sided Wilcoxon test) or DC patients (median at T0=3.8 and W12=3.3; P=0.831; two-sided Wilcoxon test), and difference in the two response groups at T0 or W12 (P=0.37 and P=0.201, respectively; two-sided Mann-Whitney U test). Light and dark orange indicate PD for T0 and W9-W12, respectively, light and dark green indicate DC for T0 and W9-W12, respectively; n represents patients; ns means not significant P values; error bars are standard deviation.

We note that the W3 TIE expansion identified patients who achieved DC early during treatment with superior accuracy to the W3 peripheral T lymphocyte invigoration to tumor burden ratio (Ki67/TB), where the accuracy = 0.64 (16/24 patients with Ki67/TB>1.94 had an objective response compared to 3/17 patients with Ki67/TB<1.94)[7]. We note also that the W3 TIE expansion also had greater accuracy than PD-L1 staining in pre-treatment melanoma biopsies where the accuracy in a phase III clinical trial = 0.67 (78/148 patients with PD-L1 positive biopsy had an objective response compared to 89/270 patients with PD-L1 negative biopsy)[26].

Next, we analyzed published single cell RNA expression and protein sequencing (REAP-Seq) data from healthy donors[27] and found that TIE cells have the additional surface phenotype CD69+/PD1low/dim/TIGIT+/CD25+/-/CD155+/CD40med/high/CD154med/high/CD357med/high and a distinct transcriptome signature including immune-activation genes (cluster 9 in Extended Data Fig. 5a,b and Supplementary Table 4). Our analysis of this data also showed that TIE cells expanded from healthy donor CD8+ naive PBMC following in vitro stimulation, and that they expressed genes associated with immune-effector function (Extended Data Fig. 5c-d). Using FACS analysis of 5 patients’ PBMC (#1,24,29,42,54) we observed a trend for TIE reinvigoration (increased Ki67+ expression) after 1 cycle of CPI (Extended Data Fig. 6a), although the limited sample size could not support robust conclusions. Note that the W3 TIE expansion was not associated with toxicity, but expansion of a separate T regulatory (Treg) subset characterized by surface phenotype CD3+/CD4+/CD8-/CD25+/CD127-/low28 correlated with toxicity grade (Fig. 6a,b).

Extended Data Fig. 5

Characterization of TIE in PBMC.

Analysis of published cohort of PBMC single cell data from reference #27. a Violin plots of the expression level of selected phenotypic and transcriptomic features of the clusters identifying peripheral T cell subsets (n=7488 single cells), the cluster with TIE phenotype is indicated in red; the plots represent the density probability, the area shapes reflect the data distribution; horizontal lines represent the minima and maxima values; central dots represent the medians. Overall minima, mean and maxima values: surface CD3=0, 0.3785, 4.1396; surface CD8a=0, 0.96327, 6.21476; surface CD45RA=0, 0.8161, 4.8508; surface CD45RO=0, 0.6628, 4.6468; surface CD197/CCR7=0, 0.8961, 5.7975; surface CD69=0, 0.5219, 4.2200; surface CD279=0, 0.09787, 3.84886; surface CD25=0, 0.08653, 4.00428; surface TIGIT=0, 0.4663, 4.2381; surface CD155=0, 0.4850, 4.6679; surface CD40=0, 0.6003, 5.5083; surface CD154=0, 0.4062, 3.8159; surface CD357=0, 0.1193, 4.0316; LGALS2=0, 0.561, 6.089; TYROBP=0, 1.337, 6.662; FCN1=0, 1.290, 6.789; CST3=0, 1.404, 6.504; LST1=0, 1.042, 6.097; LYZ=0, 1.775, 6.859. b T-SNE plot showing the clusters identified by means of the antibody derived tags (ADT) targeted to surface markers (n=7488 single cells); the black arrow indicates the cluster with TIE phenotype. c Plot showing the proportion of cells with the TIE phenotype from the same published cohort after standard in vitro culture (CTRL, n=3 sorted healthy donor peripheral blood CD8+ naïve T cell samples in standard culture) or following stimulation with anti-CD3/anti-CD27 Dynabeads[23] (STIM, n=3 sorted healthy donor peripheral blood CD8+ naïve T cell samples after stimulation) (P=0.0267, two-sided paired t test, two degrees of freedom) and d Volcano plot representing the transcriptomic differential expression of the cells with the TIE phenotype in PBMC presented in a (n=7488 single cells) or expanded from naive CD8+ T cells from the experiment presented in c[22] (n=12217 single cells; two-sided Wilcoxon test with Bonferroni correction for multiple comparisons).

Extended Data Fig. 6

Expression of Ki-67 and PD-1 in peripheral TIE cells before and after 1 cycle of CPI.

a Expression of Ki67 and PD1 in the TIE subset as measured by FACS in n=5 frozen samples of PBMC from The Christie NHS Foundation Trust metastatic melanoma patients treated with CPI, at pre-treatment (T0) and after 1 cycle of CPI (W3); horizontal line indicates median; error bar indicates standard deviation. The small sample size did not allow statistical comparison of the outcome groups.

Figure 6

Expansion of a peripheral regulatory T cell subset associated with toxicity.

a Graph showing expansion of TIE at W3 (ΔW3TIE) in patients with no ≥grade 3 toxicity (median=1.6; n=33) or ≥grade 3 toxicity (median=3.72; P=0.347; n=17; two-sided Mann-Whitney U test). Line, median; error bar, standard deviation. Orange=PD, green=DC, dot=single patient, triangles=single agent anti-PD1, squares=combination ipilimumab+nivolumab. b Expansion of CD3+/CD4+/CD8-/CD25+/CD127-/low Treg cells (Extended Data Figure 2g) at W3 (ΔW3Treg) according to toxicity at any time between 2 weeks and 6 months (P<0.0001; n=50, two-sided linear regression analysis, R2=0.29). Horizontal line is the median; error bar is standard deviation; orange represents PD; green represents DC; triangle represents one patient treated with single agent anti-PD1 drug; square represents one patient treated with combination ipilimumab+nivolumab; n represents patients; dotted line is linear regression line; ns is not significant; T0 is pre-treatment; W3 is week 3.

CPI induce peripheral T cell repertoire rearrangements

Our findings revealed intriguing parallels between immune responses to infection[21,22] and CPI, and we hypothesized that immune responses to CPI mirror the defense against pathogens. To test this, we compared T cell repertoire rearrangements in people receiving vaccination or CPI. Using published data[29,30], we calculated T cell clonality (measures clone dominance) and diversity (indicates heterogeneity) and note that 1-2 weeks after anti-viral vaccines were administered, healthy donor TCR repertoires presented bifurcated reorganization with either increased clonality or increased diversity (Fig. 7a). We next compared clonality (ΔW3clonality) and diversity (ΔW3diversity) in T0 and W3 PBMC from 17 CPI-treated metastatic melanoma patients (#11-27). In patients who went on to achieve DC at week 12 we observed bifurcated reorganization of the TCR repertoire, with substantially increased clonality or diversity, whereas no such response occurred in patients with PD (Fig. 7b). Using our training cohort, we developed a linear discriminant analysis (LDA) algorithm that at W3 classified patients according to response (assessed at week 12) with an accuracy of 0.9 (specificity=1 and sensitivity=0.8). We validated these findings in an independent cohort of 27 patients with advanced melanoma[7,31], and again found a bifurcated TCR repertoire reorganization in DC, but not PD patients (Fig. 7c). Our LDA accuracy for response prediction was 0.77 in this validation cohort.

Figure 7

TCR repertoire evolution after immune-stimulation.

a Changes in CDR3 clonality (Δclonality, Gini coefficient) and diversity (Δdiversity, Renyi index, alpha=1) in peripheral T cells from T0 to W1-2 in healthy donors who received anti-viral vaccination (n=25 healthy donor samples). b Changes in CDR3 clonality (Δclonality, Gini coefficient) and diversity (Δdiversity, Renyi index, alpha=1) in peripheral T cells from T0 to W3 in the training cohort (The Christie NHS Foundation Trust) of advanced melanoma patients receiving first line anti-PD1 based immunotherapy (n=17 patients) who progressed (n=9 patients) or responded (n=8 patients) at week 12. c Δclonality (Gini coefficient) and Δdiversity (Renyi index, alpha=1) in peripheral T cells from T0 to W3 in the validation cohort (The Christie NHS Foundation Trust, Huang et al.[2] and Amaria et al.[26]cohort n=12 patients, n=4 patients and n=11 patients, respectively) of advanced melanoma patients who progressed (n=11 patients) or responded (n=16 patients) at week 12 of anti-PD1 based treatment. Dot is one healthy donor or patient; maroon represents healthy donors in the vaccination cohort; orange represents patients who progressed after 12 weeks of immunotherapy; green represents patients who achieved disease control after 12 weeks of immunotherapy; dotted line is the linear regression line.

Discussion

We examined how the selective pressure of a single cycle of CPI affects peripheral T cell evolution in previously untreated metastatic melanoma patients. We found that CPI induced immune-awakening that was revealed by increased levels of productive CDR3 sequences released into the blood and dynamic changes in the TCR repertoire. CPI did not affect thymic output but did induce peripheral T cell turnover and this correlated with the expansion of a CD8+ cytotoxic memory effector subset that was CCR7-/CD27-. This subset of lymphocytes is involved in cytotoxic response to infections[32,33] and we have now established that they are also associated with CPI responses. Immune effector cells are the cells of the immune system that support anti-cancer immune surveillance[34], and since our data has identified a specific T cell subset involved in this network, we called them immune-effector T cells, or TIE cells.

It was recently shown that following CPI treatment, the expansion of tumor infiltrating T cell clones did not come from pre-existing TIL[35], but rather from novel clonotypes, most probably in the peripheral compartment. Those observations are consistent with a model that the tumor is an open compartment with active cross-talk with the peripheral immune-system and accordingly, in responding patients we observed a significant early T cell repertoire rearrangement in the fraction of TIL circulating in the blood, and which we call tumor emigrant T cells or tePBMC.

Clonotype modulation by checkpoint blockade has been described previously, largely in the tumor microenvironment[7,36-38], but we determined that the pattern of peripheral turnover and overall repertoire rearrangement of T cells in blood identify the patients with an effective immune-awakening and who will go on to respond to CPI. It has also been shown in animal models that anti-CTLA4 and anti-PD1 drugs induce distinct cellular reactions[39]. That we did not observe significant differences between single-agent and anti-PD1/CTLA4 combined therapy supports that our observations reflect the final effects needed for tumor elimination, and importantly we showed that these changes could be detected in the periphery. Although we could not discriminate if the T cell clones driving these changes were melanoma-specific, or if this reflected a general, off-target immune-activation, our results nonetheless contribute to our understanding of the dynamics of immune-system evolution after one cycle of CPI. That these responses also occur with infection could limit specificity in the CPI setting, necessitating further kinetic analysis and clinical validation, but our results have established that these responses could provide tractable tools for the delivery of precision immunotherapy. Moreover, our hypothesis-generating results contribute to improved understanding of immune system biology and they could have broader implications beyond the oncoimmunology field.

Note that CCR7 and CD27 were downregulated in TIE cells, and that phenotype has been associated with differentiated effector T cell release from lymph nodes to the periphery[40-Cytometry A. 2014 ">43]. Also in line with previous observations that in vitro stimulation of CD8+ T cells induced downregulation of CD27 and CCR7[44,45], our analysis of previous data showed that TIE cells expanded following in vitro stimulation. Our data also showed that the expansion of these cells after one cycle of CPI identified the patients for whom therapy overcame melanoma-induced immune-suppression. Within the scope of this study, we have not analyzed the anti-tumor reactivity of TIE cells, but our data showed that peripheral TIE clones infiltrated melanoma and represent an abundant fraction of tumor infiltrating lymphocytes with high repertoire clonality. The TIE were also in active turn-over. The relatively small size of our sample could limit the generalization of our results, but both TIE and TCR peripheral repertoire reorganization could identify which patients will benefit from CPI with greater accuracy than standard biopsy PD-L1 staining or Ki67/TB. Future research will investigate the anti-tumor cytotoxicity and specificity of these cells, and their potential for clinical development.

In summary, here we identified a peripheral blood early immune-signature characterized by significant rearrangements of the peripheral T cell repertoire and by turnover of specific T cell subsets. Critically, the magnitude of the immune-signature changes after the first cycle of therapy anticipated which patients would go on to respond and were detected in the blood, providing the advantages inherent in minimally invasive liquid biopsies. Although further studies are required to determine the mechanisms underpinning our observations and their specificity for CPI-induced responses, our research provides a strategy to analyze immune cell evolution under the selective pressure of CPI. Our findings advance our knowledge of immune system responses to immunotherapy, and critically they provide a potentially tractable tool to identify which patients will benefit from CPI early during treatment. This could help clinicians to stratify their patients more effectively to thereby improve personalization of therapeutic planning.

Materials and Methods

Patient samples

Blood samples from patients and healthy donors were collected under the Manchester Cancer Research Centre (MCRC) Biobank ethics application #07/H1003/161+5 with written informed consent from the patients at The Christie NHS Foundation Trust. The study was approved by MCRC Biobank Access Committee application 13_RIMA_01. All clinical investigations were conducted according to the principles expressed in the Declaration of Helsinki and good clinical practice guidelines. A total of 54 patients with metastatic melanoma, treated with either pembrolizumab or nivolumab plus ipilimumab as first-line therapy as per standard of care were included in the study. Inclusion criteria: the diagnosis of metastatic melanoma; exclusion criteria: previous systemic oncological treatment in the neoadjuvant, adjuvant or metastatic setting for melanoma or other cancers, concomitant therapy with immunosuppressant drugs at enrolment and synchronous other active malignancies. With the exception of two patients who had rapid, severe unequivocal clinical disease progression (#11, who then died, and #3, who was switched to BRAF targeted therapy) before the scheduled re-evaluation, response to treatment was assessed at 12 weeks after the first cycle infusion by radiographic imaging using RECIST 1.1 (week 12 response); for late response evaluation, progression was confirmed or excluded after an additional 12 weeks of treatment (best response). Disease control was defined as complete response, partial response or stable disease. Toxicity was measured according to the Common Terminology Criteria for Adverse Events v.4.0.

PBMC and plasma extraction

PBMC were isolated from blood samples using Lymphoprep (STEMCELL Technologies, Cambridge, UK) and SepMate tubes (STEMCELL Technologies, Cambridge, UK) as per manufacturer’s instructions. Red cell lysis was performed with RBC Lysis Buffer (BioLegend, San Diego, CA, USA) as per manufacturer’s instruction. PBMC’ and sorted CD3+ T cell subsets’ DNA was extracted using QIAamp DNA Blood Mini kits (Qiagen, Manchester, UK) as per manufacturer's instructions.

cfDNA analyses

Extraction and quantification of cfDNA was carried out as described previously[46] for patients #11 to #27 and #29 to #39.

FACS analysis

Following isolation, PBMC for patients #1 to #50 were kept at 4°C in phosphate buffered saline plus 2% fetal bovine serum and analyzed within 24 hours. PBMC were suspended in FACS buffer (PBS containing 2% FBS, 2mM EDTA and 0.02% sodium azide) plus 50ul of Brilliant Stain Buffer (BD Biosciences, San Diego, CA, USA) and Human TruStain FcX (BioLegend, San Diego, CA, USA) as per manufacturer’s instructions, and incubated at room temperature for forty minutes with Treg and Tmaturation panels of fluorochrome labelled antibodies from BioLegend (San Diego, CA, USA); Treg panel: CD3 (1:100, cat 317336), CD4 (1:100, cat 317438), CD8a (1:40, cat 300914), CD25 (1:10, 302610), CD127 (1:40 351304); Tmaturation panel: CD3 (1:100, cat 317337), CD4 (1:100, cat 317438), CD8a (1:40, cat 300906), CD45RA (1:100, 304130), CD45RO (1:200, 304228), CD31 (1:40, cat 303118), CD27 (1:200, 356410), CCR7 (1:20, cat 560765; BD Pharmingen, Franklin Lakes, NJ, USA). LIVE/DEAD Fixable Blue Dead Cell Stain (Thermo Fisher Scientific, Waltham, MA, US) was added to the final suspension to exclude dead cells. Stained PBMC were washed once at 300g for 7 minutes in FACS buffer and analyzed using LSR II, LSR Fortessa, Aria II or Aria III (Special Order Research Product) (BD Biosciences, Franklin Lakes, NJ, USA) cytometers and FlowJo software (v.10, Tree Star Inc., Ashland, OR, USA). CD8+ T cell subsets were live-sorted with Aria III and frozen before DNA extraction.

Treinvigoration staining (performed for patients #1, #24, #29, #42, #54): PBMC previously frozen in FBS+10% DMSO were thawed in cold RPMI and washed twice. Then, PBMC were suspended in FACS buffer and Human TruStain FcX (BioLegend, San Diego, CA, USA) as per manufacturer’s instructions, and incubated at room temperature for forty minutes with antibodies targeting CD3 (1:100, cat 317337), CD4 (1:100, cat 317438), CD8a (1:40, cat 300906), CD45RA (1:100, 304130), CD45RO (1:200, 304228), CD27 (1:200, 356410), CCR7 (1:20, cat 560765; BD Pharmingen), PD1 (1:40, cat 329939; BioLegend) and LIVE/DEAD Fixable Blue Dead Cell Stain (Thermo Fisher Scientific). Stained PBMC were then fixed and permeabilised with Perm/Fix kit according to manufacturer instructions, and stained for Ki67 (1:20, cat 350507; BioLegend) for 30 minutes at room temperature. Stained cells were resuspended in FACS buffer and analyzed using LSR Fortessa (BD Biosciences) cytometer and FlowJo software (v.10, Tree Star Inc., Ashland, OR, USA).

In patients #1, #10, #11, #12, #13, #22, #24, #25, #26, #27, #30, #42, #51-54 an aliquot of PBMC was frozen in FBS+10 % DMSO immediately after separation and then thawed and stained with CD3 fluorescent antibody as per above; CD3+ cells were sorted with Aria III (BD Biosciences) and used for the TREC quantification.

TIE cells were quantified as the percentage of CD27-/CCR7- cells in the CD3+/CD8+/CD45RO+/CD45- gate (Extended Data Extended Data Fig. 1). Gating strategy is shown in Extended Data Extended Data Fig. 1.

TREC

TREC analysis was performed using frozen PBMC for patients #1, #10, #11, #12, #13, #22, #24, #25, #26, #27, #30, #42, #51-54. TREC quantification was performed with droplet digital polymerase chain reaction (ddPCR) using custom TREC assay (TREC forward primer 5’-CACATCCCTTTCAACCATGCT-3’ at final concentration 450nM, TREC reverse primer 5’-GCCAGCTGCAGGGTTTAGG-3’ at final concentration 450nM and HEX-Black Hole probe 5’-ACACCTCTGGTTTTTGTAAAGGTGCCCACT-3’ at final concentration 250nM, as per C Falci et al.[47], Sigma-Aldrich, Dorset, UK). For ddPCR, the TREC assay was added to 20ng DNA from sorted CD3+ peripheral T cells, 11μL ddPCR Supermix for Probes (NodUTP) (Bio-rad, Hercules, CA) and 1.1μL of TERT TaqMan Copy Number Reference Assay (Thermo Fisher Scientific, Waltham, MA) in a total volume of 22μL. Droplets were generated and analyzed using the QX200 AutoDG ddPCR system according to the manufacturer’s instructions (Bio-Rad, Hercules, CA, USA). Cycling conditions were 95°C for 10 minutes, followed by 40 cycles of 95°C for 15 seconds; 55°C for 1 minute; 50°C for 2 minute. To set up and optimise the assay, we used a TREC-plasmid positive control, designed as follows: TREC-plasmid forward primer 5’-AAAGAGGGCAGCCCTCTCCAAGGCAAA-3’ and TREC-plasmid reverse primer 5’-AGGCTGATCTTGTCTGACATTTGCTCCG-3’ as per Richardson et al.[48] (Sigma-Aldrich, Dorset, UK) were used to amplify the 376 bp TREC junction from healthy donor PBMC DNA. PCR amplification was performed using Quick-Load Taq 2X Master Mix (NEB, Hitchin, UK) on a Mastercycler Nexus Gradient thermal cycler (Eppendorf, Stevenage, UK). Cycling conditions were 95°C for 30 seconds, followed by 30 cycles of 95°C for 15 seconds, 60°C for 30 seconds, 68°C for 30 seconds, and 68°C for 5 minutes. The resulting amplicon was purified using the QIAquick PCR Purification Kit (Qiagen, Manchester, UK), TA cloned into the pGEM-T Easy plasmid (Promega, Southampton, UK) and transformed into competent E. coli strain JM109 cells prepared using the Mix & Go E. coli Transformation Kit (Zymo, Irvine, USA). Plasmid DNA was purified using the QIAprep Spin Miniprep Kit (Qiagen, Manchester, UK) and Sanger sequencing using the T7 5’-TAATACGACTCACTCTAGGG-3’ and SP6 5-ATTTAGGTGACACTATAG-3’ primers was used to confirm correct insert identity. The final plasmid was designated TREC-plasmid.

RNA sequencing

RNA was extracted from a pre-treatment human fresh frozen tumor sample for one patient with available tissue (patient #12) using AllPrep DNA/RNA kit (Qiagen, Manchester, UK) according to manufacturer’s instructions. Indexed PolyA libraries were prepared using 200ng of total RNA and 14 cycles of amplification with the Agilent SureSelect Strand Specific RNA Library Prep Kit for Illumina Sequencing (Agilent, G9691B, Santa Clara, CA, US). Libraries were quantified by qPCR using the KAPA Library Quantification Kit for Illumina platforms (Kapa Biosystems Inc., KK4873, Wilmington, MA, US). Paired-end 100bp sequencing was carried out by clustering 15pM of pooled libraries on the cBot and sequenced on the Illumina HiSeq 2500 in high output mode using TruSeq SBS V3 chemistry (Illumina Inc., San Diego, CA, US). After removing adapters using Cutadapt (v1.14) and trimming poor quality base calls using Trimmomatic (v0.36)[49], the human reads were aligned to GRCh37 (release 75) using STAR (v2.5.1) aligner[50], respectively.

TCR analysis

TCR sequences were inferred from RNA Seq data from one patient for which we had a frozen pre-treatment metastasis biopsy sample using ImReP.[17]

ImmunoSEQ® TCRb Assay kit (Adaptive Biotechnologies, Seattle, WA, USA) was used to amplify and sequence TCR sequences in cfDNA and PBMC’s DNA as per manufacturer’s instructions. We loaded the same DNA input for all PBMC (350ng, patients #11 to #39) and cfDNA (40ng, patients #11 to #27 and #29 to #39) samples, while for the sorted T cell subsets (for patients #12, #16, #17, #18, #19, #29) we loaded all the DNA extracted from the sorted cells. A metafile will be available with each single sample and anonymous patient information. Pooled libraries were quantified by qPCR using the KAPA Library Quantification Kit for Illumina platforms (Kapa Biosystems Inc., KK4873, Wilmington, MA, US). Sequencing was carried out by clustering 0.6-1.1pM of pooled libraries on the Illumina NextSeq 500 according to Adaptive Biotechnologies’ instructions. Healthy donor PBMC TCR control data were downloaded from ImmunoSEQ® Immune ACCESS (Adaptive Biotechnologies, Seattle, WA, USA)(http://adaptivebiotech.com/pub/3d047774-29d9-441f-a71e-7725b5891b4d). TCR sequencing data were analyzed with ImmunoSEQ® ANALYZER (Adaptive Biotechnologies, Seattle, WA, USA) and R LymphoSeq package (v. 3.4.1, The R Foundation for Statistical Computing, Vienna, Austria). The clonal relatedness was calculated setting an edit distance=3 using the function clonalRelatedness from R package LymphoSeq. Matched paired pre-treatment and week 3 melanoma biopsy and PBMC samples of locally-advanced melanoma patients were downloaded from referenced accession Amaria et al.[18] (TCR sequencing data were downloaded from EGAS00001003178 EGA study accession dataset EGAD00010001608, patient clinical history metadata file from EGAD00001004352). Consecutive patients #11 to #27 from The Christie NHS Foundation Trust constituted the training cohort. The external validation data were pooled from the cohorts of patients from an independent cohort of patients from The Christie NHS Foundation Trust (patients #28 to #39), a cohort of metastatic melanoma patients from AC Huang et al.[7] (PBMC CD38+ plus PBMC CD38- merged populations from patient #12288, #13471, #14746 and #14835; TCR sequencing data were kindly made available by the Authors) and the cohort of locally-advanced treatment naive melanoma patients from referenced accession Amaria et al.[18] (patients #01, #02, #04, #05, #06, #07, #08, #010, #011, #013, #015; TCR sequencing data were downloaded from EGAS00001003178 EGA study accession dataset EGAD00010001608, patient clinical history metadata file from EGAD00001004352). Gini coefficient was used as a measure of clonality[51] and calculated with the function clonality from LymphoSeq R package. Clonal relatedness was calculated with clonalRelatedness LymphoSeq function, and similarity was assessed by means of Bhattacharyya coefficient using bhattacharyyaMatrix LymphoSeq function. The diversity was calculated using Renyi index (α=1) as per Spreafico et al.[52], with time point pairwise analysis for each single patient. Linear discriminant analysis (coefficient of linear discriminants LD1: [ΔW3Renyi index]2=5.2 and LD2:[ΔW3Gini coefficient]2=261.3) and validation to calculate balanced accuracy were performed using R MASS and caret packages (v. 3.4.1, The R Foundation for Statistical Computing, Vienna, Austria). For further details refer to the Extended R scripts that are available on GitLab (https://gitlab.com/cruk-mi/tcell-immune-awakening).

CyTOF surface phenotype analysis

Differential marker expression analysis was performed on CyTOF (cytometry by time-of-flight mass spectrometry) data from Krieg et al.[8] and Greenplate et al.[24] downloaded from a publicly available repository (referenced accession https://flowrepository.org/experiments/1124 and http://flowrepository.org/id/FR-FCM-ZZMC), using the custom workflow described in Nowicka et al.[53]. All analyses on CyTOF data were performed after arcsinh (with cofactor=5) transformation of marker expression and correction for batch effect (function removeBatchEffect from limma R package). The TIE subset was identified by differential expression of T lymphocytic markers CD3, CD8a, CD45RA CD45RO, CCR7 and CD27. Cell populations clustering was obtained with R package FlowSOM after the metaclustering step (ConsensusClusterPlus R package).

REAP-Seq single cell analysis

Differential proteomic and RNA expression analysis was performed on REAP-Seq data from Peterson et al.[27] downloaded from a publicly available repository (referenced accession https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE100501), using R package Seurat v.2.4.0[54]. The single cells with TIE phenotype were identified based on the expression level of antibody derived tags (ADT) CD8a, CD45RA, CD45RO, CD27 and CD197 (CCR7) (FetchData command with set conditions based on the expression level, cut-offs for positive/negative and high/low have been set based on the normalized data), and the cell identities have been used to define the TIE cluster (SetIdent command). The cell differential expression analysis in the TIE from PBMC vs after in vitro stimulation was performed with the function FindMarkers in the combined Seurat object (RunCCA). The ADT data Seurat matrices were imported in Cytobank[55] to analyze the differential representation of the CD8+ subsets under different experimental conditions.

Statistics and Reproducibility

Unless otherwise stated, all statistical tests were two-tailed. The statistical differences between two groups for numerical variables were assessed using two-tailed Mann-Whitney U test (unpaired comparisons) or Wilcoxon test (paired comparisons). The statistical differences between multiple, paired measures was assessed using Friedman test. Deltas were calculated as the difference between W3 and T0 values. The statistical differences of categorical variables between groups were assessed using two-tailed Chi-square or Fisher’s exact test, according to group dimension. Correlation between continuous variables was assessed with Spearman test (independent variables) or with linear regression (dependent variables). Kaplan-Meier plots with the log-rank test (3 week landmark analysis) were used to analyze survival data. Univariate Cox regression was used to calculate the hazard of death. P values<0.05 were retained as significant, Cox-Snell residuals were used to verify the proportional hazard hypothesis (P=0.141, with a P value>0.05 confirming the hypothesis). Sample size calculation was performed using G*Power software (Erdfelder, Faul, & Buchner, 1996), using the effect size and standard deviation. For the comparison of ΔRES in peripheral blood mononuclear cells vs cfDNA in patients with disease response sample size was 14 for alpha=0.05 and 1-beta=0.8. For the linear discriminant analysis we used the power and sample size calculation for linear regression with 2 covariates and effect size f2=0.55, and total sample size was calculated=17 for alpha=0.05 and 1-beta=0.8. For the T cell subset analyses for the immune-effector T cells in patients with progression vs disease control the total sample size was 32 for alpha=0.05 and 1-b=0.8. No data were excluded from the analyses. The Investigators were blinded during experiments; outcome assessment was performed after experiments. Analyses were performed with GraphPad Prism version 7 (GraphPad Software, La Jolla California USA) or R (v. 3.4.1, The R Foundation for Statistical Computing, Vienna, Austria). Further information on research design is available in the Nature Research Reporting Summary linked to this article.