PD-1⁺ Polyfunctional T Cells Dominate the Periphery after Tumor-Infiltrating Lymphocyte Therapy for Cancer

Ideal cancer treatment should induce durable tumor regression to give a meaningful life extension. In melanoma, work performed in the context of previous clinical trials based on transfer of autologous tumor-infiltrating lymphocytes (TIL) have identified a number of factors influencing the likelihood of tumor regression (1–5). However, despite the high degree of efficacy, over 50% of patients with significant tumor regression achieve only a temporary, partial remission (1). Sustained tumor responses may be dependent on the long-term fate of the infused tumor-specific T cells. However, the attributes of these persisting T-cell populations are largely unknown. Current TIL treatment regimens are based on a single infusion of expanded T cells derived from an autologous tumor metastasis. These T cells are not selected for any specific class or type of antigen, so this treatment takes advantage of a multitarget T-cell attack directed to multiple and mostly unknown antigenic specificities, including mutant neoantigens (6). Current methods reveal the existence of neoantigen-specific T cells in TILs from most patients with melanoma; however, whole tumor-cell recognition assays indicate that the frequency of naturally occurring tumor-specific T cells may largely exceed those observed for the neoantigens evaluated (7). Libraries of known shared antigens have been constructed and validated (8), but greatly underestimate the frequency of tumor-specific cells (9). Here, we evaluated the immunological dynamics (immunodynamics) of the total repertoire of tumor-reactive CD8⁺ T cells identified with whole-tumor cell recognition assays, regardless of the type or class of melanoma antigens recognized. This comprehensive analysis allowed the identification of several common features of CD8⁺ T-cell responses that persisted for up to 1 year after infusion.

All protocols were approved by the Scientific Ethics Committee of the Capital Region of Denmark. Written informed consent was obtained from patients before any procedure, according to the Declaration of Helsinki. Patients were treated with TIL immunotherapy in the clinical trial NCT00937625. This clinical trial consisted of two consecutive substudies: a pilot study (n = 6), with a low dose of IL2 administered after T-cell infusion (2 MIU/day subcutaneously for 14 days; ref. 10); an amended study (n = 25), with an intensified regimen of IL2 (decrescendo i.v. regimen or intermediate dose; ref. 5). Blood samples were collected at serial time points after infusion of TILs: at discharge (about 1–2 weeks after infusion) and at approximately 1 month, 2 months, 4 months, 6 months, 9 months (this sample was analyzed only in functional evaluation, see below), and 12 months, unless the patient was excluded from the protocol at a previous time point. Peripheral blood mononuclear cells (PBMC) were isolated with standard methods, and cryopreserved at −140°C until use. We included 16 selected patients where previous tumor recognition assays had identified sufficient numbers of tumor reactive T cells in the peripheral blood after infusion (described in ref. 5). Phenotypic assessment of T cells with defined antigen specificity and clonotypic analysis of tumor-reactive TILs were conducted each on three selected patients. All other analyses included at least 13 patients. We were unable to include the three remaining patients (total cohort of 16) in all analyses due to insufficient sample material.

Evaluation of T-cell responses was performed as previously described (5, 11). Briefly, TILs or PBMCs were cocultured with autologous short-term cultured melanoma cell lines (available for 10/16 patients), freshly cryopreserved autologous tumor single-cell suspensions (1/16 patients) or partially HLA-matched allogeneic melanoma cell lines (5/16 patients, in this case TILs and PBMCs were tested only for recognition of the cell line where TILs had previously shown the highest level of recognition, (see ref. 5 for details). For phenotypic assessment of polyfunctional tumor-reactive T cells, we combined assessment of tumor cell recognition with staining of surface phenotype markers. We also compared the phenotype of non–tumor-reactive CD8⁺ T cells in addition to polyfunctional tumor-reactive T cells.

For phenotypic assessment of T cells with defined antigen specificity, we combined staining with fluorochrome-conjugated peptide–MHC (p-MHC) multimers and staining of surface phenotype markers. Details on reagents and additional description of the methodologies are available as Supplementary Data.

The relative fractions of cells were analyzed with BD FACS Diva 6 (only for enumeration of CD107a⁺ cells within the TNF⁺/IFNγ⁺ gate) or FlowJo 9.7.1 (FlowJo LLC) subcategorizing flow cytometry events with parallel Boolean gating (gates drawn solely on the parameters of interest). For enumeration of CD107a⁺ cells within the TNF⁺/IFNγ⁺ gate, data were plotted in excel and displayed on exponential scale. Logarithmic r² values are calculated separately for each individual patient.

For functional characterization, the population of cells negative for all three functional markers was removed from the analysis. Thus, boolean combination gates of the three functional markers resulted in seven individual gates (eight minus one, which is the only group negative for all functional markers), each showing the percentage of CD8⁺ T cells expressing a unique combination of the three markers.

Data were exported into Pestle 1.7 (courtesy of Dr. Mario Roederer, ImmunoTechnology Section, VRC/NIAID/NIH) and formatted, according to the manufacturer's instructions. Analysis and presentation of distributions was performed using Simplified Presentation of Incredibly Complex Evaluations (SPICE) version 5.3, downloaded from http://exon.niaid.nih.gov (12). Background subtraction was performed with Pestle 1.7, according to the manufacturer's instructions. In SPICE, thresholds were set at 0.1. Comparison of distributions was performed using a Student's t test and a partial permutation test as described (12). To compare bar charts, a Wilcoxon signed rank-test assuming unequal variances was carried out, with P < 0.05 considered significant. Other statistical analyses were performed with GraphPad Prism 5 (GraphPad Software). For the analysis of single markers phenotype, TP cells were gated and data were exported in SPICE via FlowJo and Pestle. Bar charts were compared as described above.

For combinatorial analysis of phenotype, a “differentiation” scoring system was established according to the model proposed by Gattinoni and colleagues (13), which identifies “early” effector and “late” effector cells. Despite CCR7 is not included in the analysis, expression of CD45RO was used as marker of earlier differentiation because we and others have previously demonstrated that the vast majority of TILs consist of T cells with an effector memory (CD45RO⁺ CD45RA⁻ CCR7⁻) phenotype (14), and loss of CD45RO in this population is likely to represent a transition to TEMRA cells (15). The model of Speiser and colleagues (16), which identifies PD-1⁺ cells as terminally differentiated cells, was also incorporated into these analyses. Because LAG-3 was largely detected only in triple-positive (TP) cells present in TILs infusion products, this marker was removed from further analyses. The following groups were established: Group 0 (early effector cell) contained the CD27⁺, CD45RO⁺, CD57⁻, and PD-1⁻ while group 4 (late effector cell) contained the CD27⁻, CD45RO⁻, CD57⁺, and PD-1⁺ fraction. Group 1 contained one of the markers from group 4, group 2 two of the markers from group 4 and group 3 three of the markers from group 4. In brief, a score from 0 to 4 was assigned to TP cells, with 0 representing the less-differentiated cells to 4 the most differentiated. This is shown in Supplementary Fig. S6A, where intrasample analyses (i.e., the frequency of TP cells contained in each group, for each sample) are demonstrated. Furthermore, one single combined “Differentiation Score” was given to each sample (each patient, at each time point), by multiplying the arcsine transformed value of the frequency of cell subpopulations contained in the groups described above, with a predefined score for each group which was 0 for early effector cells (group 0) and 4 for late effector cells (group 4) as described above (intermediate values 1–3 were given to cells with 1, 2, or 3 phenotypic markers). The differentiation score of each sample is demonstrated in Fig. 3B. Supplementary Fig. S1 shows a graphical representation of the scoring system, and the unique T-cell subpopulations analyzed are reported in Supplementary Table S1. In addition, boolean combination gates of the four phenotype markers resulting in 16 individual gates were analyzed and processed as described above in functional characterization. These data are shown in Supplementary Fig. S7.

Autologous tumor cells were plated 1 day before sorting in RPMI medium supplemented with 100 U/mL penicillin, 100 μg/mL streptomycin, 2 mmol/L l-glutamine, and 5% heat-inactivated fetal calf serum (all from Life Technologies). TIL infusion products or PBMCs were stimulated with autologous tumor 1 day after defrosting, in order to avoid TCR repertoire bias due to culturing and expansion in vitro. TIL samples or PBMCs were coincubated with autologous tumor at a 1:1 ratio in the presence of 15 μL of anti-TNF PE-Cy7 (BD Biosciences), 15 μL CD107a PE (BD Biosciences) and 30 μmol/L of TAPI-0 (Calbiochem) for 5 hour at 37°C, 5% CO₂. Following incubation, cells were stained with LIVE/DEAD Fixable Violet Dead Cell Stain Kit (Life Technologies), anti-CD3 PerCP (Miltenyi Biotec), and anti-CD8 APC-Vio 770 (Miltenyi Biotec). Cells from each experimental condition were washed and the live CD3⁺TNF⁺CD107a⁺ fraction was sorted directly into 1.5 mL microcentrifuge tubes containing 350 μL of lysis buffer (Qiagen). RNA extraction was carried out using the RNEasy Micro kit (Qiagen). cDNA was synthesized using the 5′/3′ SMARTer kit (Clontech) according to the manufacturer's instructions. The SMARTer approach used a Murine Moloney Leukaemia Virus (MMLV) reverse transcriptase, a 3′ oligo-dT primer and a 5′ oligonucleotide to generate cDNA templates which were flanked by a known, universal anchor sequence. PCR was then set up using a single primer pair. A TCR-β constant region-specific reverse primer (Cβ-R1, 5′-GAGACCCTCAGGCGGCTGCTC-3′, Eurofins Genomics) and an anchor-specific forward primer (Clontech) were used in the following PCR reaction: 2.5 μL template cDNA, 0.25 μL High Fidelity Phusion Taq polymerase, 10 μL 5X Phusion buffer, 0.5 μL DMSO (all from Thermo Fisher Scientific), 1 μL dNTP (50 mmol/L each, Life Technologies), 1 μL of each primer (10 μmol/L), and nuclease-free water for a final reaction volume of 50 μL. Subsequently, 2.5 μL of the first PCR products were taken out to set up a nested PCR as above, using a nested primer pair (Cβ-R2, 5′-TGTGTGGCCAGGCACACCAGTGTG-3′, Eurofins Genomics and anchor-specific primer from Clontech). For both PCR reactions, cycling conditions were as follows: 5 minutes at 94°C, 30 cycles of 30 seconds at 94°C, 30 seconds at 63°C, 90 seconds at 72°C, and a final 10 minutes at 72°C. The final PCR products were loaded on a 1% agarose gel and purified with the QIAEX II gel extraction kit (Qiagen). Purified products were barcoded, pooled, and sequenced on an Illumina MiSeq instrument.

Patient characteristics are summarized in Table 1. Out of 16 TIL-treated patients selected on the basis of the availability of samples and measurable antitumor responses in the peripheral blood, 10 had response evaluation criteria in solid tumors (RECIST) version 1.0 confirmed responses and three additional patients had minor tumor regressions including unconfirmed partial responses. Two patients had stable disease and only one patient, M27, was classified as pure progressor after infusion of TILs. The predominance of responding patients in this cohort was largely expected, as we have previously shown that the detection of tumor-reactive CD8⁺ T cells in the peripheral blood lymphocyte (PBL) pool after TIL infusion is strongly associated with tumor regression (5).

It is believed that T cells expressing multiple functions may mediate superior immune responses (17) and polyfunctionality is known to correlate with antigen-sensitivity and TCR binding affinity (18). We have previously shown that over 95% of CD8⁺ T cells expressing at least one out of seven antitumor functions express either TNF, IFNγ, or CD107a (11). Thus, we limited our analyses to these three functional markers. A retrospective analysis of our previously published dataset on TIL recognition of autologous melanoma cells (11) indicates that the fraction of CD8⁺ T cells expressing CD107a was lower than the fraction producing cytokines (Supplementary Fig. S2A, S2B, and S2C), and the majority of CD107a⁺ cells (approximately 65%) also produced at least one cytokine (Supplementary Fig. S2D). This suggests that CD107a mobilization was mainly associated with cytokine production, while cytokine production could occur independently of CD107a upregulation. Thus, we initially focused on CD8⁺ T cells expressing cytokines, and to increase the sensitivity of the analysis, primarily on those expressing simultaneously TNF and IFNγ [double-positive (DP) cells]. Notably, among DP cells, the percentage of cells expressing CD107a increased over time in all (n = 13; median r² 0.63) but one patient (Fig. 1A). CD107a is a surrogate marker of lytic granule release and, therefore, the capacity of T cells to directly destroy tumor cells (19). In this context, it is interesting that the only patient where the CD107a⁺ fraction decreased was the one patient who, despite detection of tumor-reactive CD8⁺ T cells in post-infusion blood samples, had a significantly progressive disease at first evaluation after treatment, and was quickly excluded from the study (patient M27, Table 1). In light of this finding, we analyzed whether the frequency of CD8⁺ T cells expressing all three functions (polyfunctional T cells, triple-positive cells or TP cells) within the whole pool of tumor-reactive CD8⁺ T cells (expressing at least one function after recognition of whole tumor cells) showed a similar increasing trend. An initial drop in the frequency of TP cells was observed from TIL infusion products to the first blood sample, obtained 1 to 2 weeks after infusion (Fig. 1B; Supplementary Fig. S3A and S3B). However, TIL infusion products are cultured in vitro with high doses of IL2, which may potentially explain this phenomenon of hyperfunctionality of TIL infusion products. The fraction of TP cells in patient blood samples tended to gradually increase until ∼4 months of evaluation and then stabilized thereafter for up to 1 year (Fig. 1B; Supplementary Fig. S3A and S3B). These data indicate that polyfunctional T cells accumulate within the peripheral pool of tumor-reactive CD8⁺ T cells and may establish long-lived immune responses.

Next, we investigated whether any change in phenotype of polyfunctional tumor-reactive CD8⁺ T cells occurred in vivo after infusion. For these analyses, we focused on known markers of T-cell differentiation and immune checkpoints thought to constrain T-cell responses. The phenotypes of polyfunctional CD8⁺ T cells and of non–tumor-reactive CD8⁺ T cells were analyzed in parallel.

Programmed death-1 (PD-1) is a cell surface immune regulatory molecule (or immune checkpoint) which negatively regulates T-cell responses (20). PD-1 was expressed on nearly 35% of the TP cells in the infusion product, which increased to about 65% shortly after infusion and stabilized thereafter (Fig. 2A). No increase was observed in non–tumor-reactive CD8⁺ T cells (Fig. 3A; Supplementary Fig. S4A), indicating that PD-1 accumulation is a highly specific feature of tumor-reactive cells.

LAG-3 is another cell surface molecule functioning as an immune checkpoint (21), which may work synergistically with PD-1 to downregulate T-cell responses (22). Interestingly, while around 65% of the TP in the infusion product expressed LAG-3 very few cells expressed this marker in vivo even a week post infusion (Fig. 2B). This surprisingly dramatic change is consistent with previous data reported in a clinical trial with infusion of gene-engineered T cells (23). It is likely that some phenotypic changes, such as loss of LAG-3 expression post infusion, could simply be the result of having removed the high doses of IL2 used to culture cells in vivo. Indeed, a similar pattern was observed in non–tumor-reactive CD8⁺ T cells (Supplementary Fig. S4B). Alternatively, LAG-3⁺ cells might have reduced capacity to survive in vivo or might exhibit differential migration patterns so as not to be present in peripheral blood samples. In any case, it does not appear that LAG-3 is expressed at high levels on circulating tumor-reactive CD8⁺ T cells. However, LAG-3 may be upregulated on LAG-3–negative cells several hours after T-cell activation upon recognition of tumor cells (Borch TH, personal communication).

The majority of CD8⁺ T cells in heavily expanded TILs consist of antigen-experienced effector memory T cells (14). Thus, three additional surface markers were analyzed in order to characterize the differentiation status of TP cells (CD27 and CD45RO representing less differentiated cells – CD57 representing more differentiated cells). In vivo accumulation of CD27⁺ cells after T-cell infusion has been reported in other studies (24, 25) and the size of the CD27⁺CD8⁺ T-cell pool in bulk TILs was highly associated with their ability to mediate tumor regression (26). Thus, CD27 may be a key molecule for maintaining immunological memory. Our data show that the mean frequency of TP cells expressing CD27 increased significantly from around 5% in the TIL infusion product, with a progressive trend up to 30% to 35% at 4, 6, and 12 months after infusion (Fig. 2C). Very low expression of CD27 in infused TILs may be a consequence of the high doses of IL2 used to drive massive TIL expansion (26), and indeed a similar increase in CD27 was observed in non–tumor-reactive CD8⁺ T cells (Supplementary Fig. S4C and S5A). Therefore, it cannot be excluded that the observed phenomenon might be explained, at least in part, by IL2 withdrawal, as previously reported in vitro (26). Similarly, CD57 expression exhibited a progressive trend from around 2% to 5% in the infusion product to almost 37% 12 months after treatment (Fig. 2D). In contrast, around 75% of TP cells in the infusion product expressed CD45RO, which was mirrored 1 week later in the PBMCs before dropping to about 50% at over 1 month after infusion (Fig. 2E). Parallel increase in CD57⁺ and decrease in CD45RO⁺ were detected in non–tumor-reactive CD8⁺ T cells (Supplementary Figs. S4D, S4E, S5B, and S5C). Comparison of phenotype markers in TP and non–tumor-reactive CD8⁺ T cells revealed difference only for CD57 in the infusion products (Supplementary Fig. S5B), and major differences in the peripheral pools for PD-1 (Fig. 3A) and CD57 (Supplementary Fig. S5B), reflecting quantitatively different changes in the two subpopulations. However, in this context, it should be highlighted that changes in non–tumor-reactive CD8⁺ T-cell populations are likely to represent a replacement of the infused cells with de novo generated populations, including naïve CD8⁺ T cells (including CD45RO⁻ T cells, which for non–tumor-reactive CD8⁺ T cells may represent naïve CD45RO⁻ CCR7⁺ cells), during bone marrow recovery after lymhodepleting chemotherapy.

In summary, our data showed a characteristic change in the phenotype of bulk tumor-reactive TP CD8⁺ T cells, with rapid and highly specific accumulation of PD-1. Other phenotypic markers such as LAG-3, CD27, CD57, and CD45RO displayed similar early changes after infusion in both TP and non–tumor-reactive CD8⁺ T cells. This altered phenotype remained largely stable for up to 1 year after infusion. Thus, we next looked to dissect the complex picture emerging from analyses of individual phenotype markers using combinatorial analyses.

Given the complex picture obtained from the analysis of single markers, a differentiation scoring system was established. TP CD8⁺ T cells were categorized into five groups with a score from 0 to 4, with 0 representing early effector cells and 4 representing the most differentiated effector cells (described in Materials and Methods/Flow cytometry data processing and analysis; see also Supplementary Fig. S1 and Supplementary Table S1). LAG-3 was excluded from these analyses, because it could not be detected in all postinfusion samples. The frequency of cells in the five groups changed early postinfusion, with an increase in more differentiated cells up to 1 month after infusion that remained largely stable for up to 1 year (Supplementary Fig. S6). The differentiation score, which was given on a per sample basis (one single combined score for each sample), confirmed only early changes with progressive differentiation to up to at least 1 month after infusion and stabilization of the phenotype thereafter without further differentiation (Fig. 3B). Full combinatorial analyses of the unique combination of surface markers are shown in Supplementary Fig. S7. In summary, while the frequency of less-differentiated TP CD8⁺ T cells seemed to drop progressively during the period prior to 1 month after infusion, a similar phenotypic pattern of TP CD8⁺ T cells was observed from 1 month to 12 months after infusion.

Having established that several phenotypic markers on tumor-reactive CD8⁺ T cells were preserved up to 12 months after infusion, we next sought to determine whether TCR-β chain clonotypes from the TIL infusion product would persist in the periphery over time. We reasoned that the persistence of TCR-β chains found in the infusion product, in the periphery, would suggest the establishment of long-term memory, alternatively continuous antigen-specific stimulation. Three patients with prolonged tumor regression (M11, M15, and M24) were selected for clonotypic analyses of the tumor-reactive (CD107a⁺TNF⁺) T-cell populations in the infusion product and peripheral blood 6 months after infusion. The degree to which the TILs and peripheral blood repertoires overlapped varied across patients. Despite individual differences in overall repertoire diversity and composition, several tumor-reactive clonotypes found in the infusion product were detectable in peripheral blood 6 months after treatment (Fig. 4). While the relative frequency of certain TCR-β chain clonotypes was lower in the blood samples, compared with TIL infusion product, other clonotypes expanded over time. Interestingly, each patient had several tumor-reactive CD8⁺ T cell clones in the 6-month sample that were not detectable in the infusion product. In all 3 patients, TCR diversity appeared to rise posttreatment, as indicated by an increase in Shannon entropy. Thus, whereas many tumor-reactive T cell clones can persist for at least 6 months after infusion, other T-cell populations arise in the periphery over time from de novo recruitment or from expansion of rare clones present in the infusion product at frequencies below our limits of detection. The presence of tumor-reactive clonotypes in the blood that were not present in the infusion product might suggest epitope spreading.

In order to gain insights into whether the observed phenotype changes occur in the repertoires of infused CD8⁺ T cells targeting a defined antigen, and whether similar phenotype changes could be detected in resting cells (phenotype of TP cells might be influenced by T-cell activation upon recognition of naturally presented tumor antigens), we carried out similar phenotype analyses focusing on CD8⁺ T cells recognizing known tumor-associated antigens as detected by p-MHC multimer staining. Samples from 3 patients with high-frequencies of CD8⁺ T cells with defined tumor-antigen specificity were selected for these analyses (Supplementary Fig. S8A). We observed a similar trend as seen in TP CD8⁺ T cells, with patterns of strong LAG-3 downregulation (data not shown), increase of PD-1 (Fig. 5A), decrease of CD45RO (Fig. 5B), and increase of CD57 (Fig. 5C). Similar to TP CD8⁺ T cells, the increase of PD-1 appeared larger in tumor-antigen specific CD8⁺ T cells (Supplementary Fig. S8B). CD27 did not increase as much as observed in the whole pool of tumor-reactive CD8⁺ T cells (Fig. 5D). However, current data do not allow to establish whether the more pronounced changes observed in tumor-reactivity experiments were due to the activated state of TP CD8⁺ T cells or whether this discrepancy was simply due to the small sample size of p-MHC experiments. Combinatorial analysis of surface markers of these populations was similar to that seen in TP CD8⁺ T cells (compare pie charts in Supplementary Fig. S7 with Supplementary Fig. S8C).

Optimization of cellular immunotherapies requires a detailed understanding of the dynamic processes, leading to establishment of long-term immunological memory. We conducted a detailed, sequential, multidimensional analysis of bulk tumor-reactive T-cell populations derived from melanoma patients treated with TILs, dissecting dynamic functional and phenotypic changes of antitumor CD8⁺ T cells. Polyfunctionality is a highly desirable feature of potent CD8⁺ T cells, well known in infections (17), but only recently described in cancer immunity (11, 27). This feature is known to correlate with antigen-sensitivity and TCR affinity for cognate antigen (18, 28). We showed that polyfunctional tumor-reactive CD8⁺ T cells accumulate in the peripheral blood of patients treated with TILs and that individual clonotypes within the TILs persist in patient peripheral blood 6 months after treatment. The increase of polyfunctional (TP) cells over time may indicate that these populations have a very high capacity to survive in vivo and/or that monofunctional CD8⁺ T cells acquire further functions over time. Interestingly, the relative increase of the polyfunctional T-cell subset occurred mainly in the first few months after infusion, setting the premises for long-term memory. This expanding population of tumor-reactive T cells exhibited a partially differentiated cellular phenotype, which remained stable for up to 1 year after infusion and included the highly specific feature of early PD-1 upregulation.

After withdrawal of exogenous IL2, which is known to overcome PD-L1–mediated inhibition (29), this increase in PD-1 expression in tumor-reactive cells may render them sensitive to PD-L1. Thus, impaired T-cell activity via the PD-1 pathway may be responsible for relapses observed in some patients who initially responded to TIL treatment (1). The other markers we studied displayed similar early changes after infusion on both TP and non–tumor-reactive CD8⁺ T cells. For example, CD27 is an unstable marker that can be markedly affected by cytokines and exposure to (or even self-expression of) its ligand CD70 (26). Consequently, expression of CD27 can be sporadic and fluctuating during the life of clonal T cells. Thus, accumulation of CD27⁺ cells in vivo might represent a slow increase in CD27 expression following withdrawal of exogenously supplied IL2. The kinetic of increase in CD27 is seen within the first month after withdrawal of IL2 and thus consistent with in vitro studies from Huang and colleagues (26), and in parallel to similar increase in CD27 expression on non–tumor-reactive CD8⁺ T cells. High levels of CD27 were associated with clinical responses following ACT (1, 26); thus, overall, increasing or maintaining a sufficient level of CD27 expression highlights its role as a key marker of immunological memory. Upregulation of CD57 is normally interpreted as continued progression to a terminally differentiated phenotype (30). The population of CD27⁺CD57⁺ cells that we described here is unusual but has been described previously in patients with melanoma (31).

Importantly, the pattern of polyfunctionality and phenotypic markers that is established within the first month after TIL therapy is then maintained for at least a year. In this context, we proposed a differentiation scoring system based on the combinatorial expression of CD27, CD45RO, CD57, and PD-1. This system is imperfect due to the known instability of some markers such as CD27, and the absence of other potentially important markers such as CD62L and CCR7. Nevertheless, this system proved useful for visualizing the global phenotypic dynamics up to 1 year following infusion of TILs (Fig. 3B).

We also investigated whether individual T cells within the infusion product could persist in patient blood after treatment by examining the clonotypic architecture of the tumor-specific response in the infusion product and patient blood 6 months after treatment. These results confirmed that some tumor-reactive T cell clones from the TIL infusion product persisted and dominated the antitumor response in vivo. In addition, new tumor-reactive clonotypes were detected, suggesting that these might have been expanded postinfusion as the result of epitope spreading as the chosen patients successfully rejected their tumors. Ongoing work in our laboratories is aimed at determining the individual antigen specificities of some key clones in order to confirm whether they recognize new epitopes that were not recognized by the original TIL infusion product. Additional analyses indicated that the phenotypic profile observed in the bulk tumor-reactive CD8⁺ T-cell repertoire was similarly reflected within the responses to known tumor-antigens. It therefore seems reasonable to conclude that all non–de novo T-cell subpopulations that contributed to the bulk CD8⁺ T-cell response against tumor persisting over time had a similar phenotypic profile.

Our discovery of biomarkers correlating the fate of tumor-specific CD8⁺ T cells after adoptive transfer is of high interest, because these findings could have direct implications for the selection of CD8⁺ T cells for future adoptive immunotherapy studies or combination therapies. By using highly homogeneous T-cell clone populations specific for melanoma differentiation antigens, Chandran and colleagues have recently shown that tumor-specific CD8⁺ T cells expressing IL7 receptor and c-myc have superior capacity for persisting in vivo (32). Our present data do not allow determination of whether polyfunctionality and the observed partial differentiation phenotype can be used to predict the fate of preinfusion TILs, or whether this is simply a natural differentiation pathway of long-lived memory CD8⁺ T cells. Additional detailed studies will be required to definitively link cell phenotype to cell fate in vivo. Nonetheless, the specific and strong accumulation over time of PD-1 in polyfunctional T cells suggests that combined TIL/checkpoint inhibitor blockade or the use of PD-1 gene–edited TILs (33) should be explored in future clinical trials, especially to prevent disease relapses. With current adoptive cell therapy protocols, disease relapses occurred in up to 63% of patients with an initial objective response (analysis of clinical data presented in ref. 1).

In summary, we have begun to dissect the immunological dynamics of T-cell persistence and differentiation in vivo following adoptive transfer of TILs. The changes we observed mainly took place shortly after infusion, with early PD-1 upregulation as a highly specific feature of polyfunctional tumor-reactive CD8⁺ T cells, and remained stable for up to a year. These PD-1⁺ polyfunctional T cells may have a high capacity for persistence and generation of immunological memory in patients with metastatic melanoma treated with T-cell therapy and be crucial in keeping patients tumor free. Our results have important implications for the next-generation adoptive immunotherapy protocols as well as for immunological monitoring of patients treated with T-cell–based immunotherapy.

No potential conflicts of interest were disclosed.

Conception and design: M. Donia, A.K. Sewell, I.M. Svane

Development of methodology: M. Donia, S.R. Hadrup, A.K. Sewell

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Donia, R. Andersen, V. Bianchi, M. Legut, M. Attaf, G. Dolton, R. Lyngaa, S.R. Hadrup, A.K. Sewell

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Donia, J.W. Kjeldsen, R. Andersen, M.C.W. Westergaard, M. Legut, M. Attaf, B. Szomolay, S. Ott, G. Dolton, R. Lyngaa, A.K. Sewell

Writing, review, and/or revision of the manuscript: M. Donia, R. Andersen, M. Legut, M. Attaf, G. Dolton, A.K. Sewell, I.M. Svane

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Donia

Study supervision: M. Donia, A.K. Sewell, I.M. Svane

The Danish Cancer Society, The Capital Region of Denmark Research Foundation, and Danielsen Foundation are acknowledged for financial support. The authors thank all the staff of the Department of Oncology, Herlev Hospital, in particular Dr. Eva Ellebaek, for inclusion and treatment of patients in the clinical trial. The technicians of the CCIT lab are acknowledged for blood sample processing and storage. The authors thank the creators of SPICE for providing powerful data analysis tools. A.K. Sewell is a Wellcome Trust Senior Investigator.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.