PD-1 Blockade Expands Intratumoral Memory T Cells

The programmed cell death protein 1 (PD-1) is an immune checkpoint protein expressed in T cells. PD-1 inhibits T-cell responses to cancer after binding to one of its ligands, PD-1 ligand 1 (PD-L1 or B7-H1) or PD-L2 (also called B2-DC; refs. 1–4). PD-1 limits the activity of T cells by inducing a phosphatase that inhibits T-cell receptor downstream signaling (2, 4, 5). In addition, effects on other lymphocyte subsets have been described, including an enhancement of regulatory T cell (Treg) proliferation and suppressive activity (6), and a decrease in the activity of both B cells and natural killer (NK) cells (7).

Therapeutic blockade of PD-1 or PD-L1 with monoclonal antibodies leads to durable tumor regressions in patients with several cancer types (8–12). These emerging clinical data have led to the approval by the FDA of two antibodies to PD-1 for the treatment of metastatic melanoma and lung cancer, the humanized IgG4 antibody pembrolizumab (MK-3475) and nivolumab (BMS-936558).

Clinical responses to PD-1 blockade are associated with increased PD-L1 expression on tumor-resident cells, induced by preexisting tumor-infiltrating lymphocytes (TIL), in what is termed “adaptive immune resistance” (1, 10, 13). Patient biopsies obtained from patients before and on therapy with pembrolizumab showed that intratumoral CD8⁺ T cells proliferated only in patients with an objective response to therapy, as assessed by quantitative immunohistochemistry (IHC). However, it is currently unclear which T-cell functional subsets are involved in this response, and other tumor microenvironment hematopoietic lineage cells have not been well characterized in patient samples.

In the current study, we undertook a comprehensive analysis using multicolor flow cytometry and single-cell multiparametric data interpretation of immune cell infiltrates, particularly in T cells, in biopsies of patients with metastatic melanoma treated with PD-1 blockade.

We collected baseline and at least one tumor biopsy while patients were on treatment with pembrolizumab, taken at a mean time of 74 days (range, 15 to 230 days) from 53 patients with metastatic melanoma (stage M1a to M1c; Table 1) treated with pembrolizumab within a phase I clinical trial at UCLA (UCLA IRB 11-003066; NCT01295827) between January of 2012 and May 2013. Patients received single-agent pembrolizumab intravenously at one of three dosing regimens, 0 mg/kg every 2 weeks, or 2 mg/kg or 10 mg/kg every 3 weeks. Three patients had two baseline biopsies, and one patient had three baseline biopsies in different metastatic lesions, for a total of 62 baseline biopsies. Tumor response was assessed at 3 months, with scans performed at 3-month intervals thereafter.

Pieces of tumors from dermatologic, surgical, or image-guided biopsies were mechanically dissociated, centrifuged at 500 g for 5 minutes, and the pellets collected. Cells were either stained immediately or, in a few cases, cryopreserved as described (14). Peripheral blood mononuclear cells (PBMC) were isolated as described in Ibarrondo and colleagues (14).

TILs and PBMCs were stained and acquired in an LSR II Flow Cytometer (BD Biosciences) as described by Ibarrondo and colleagues (14). A description of flow antibody reagents used is on Supplementary Table S1. Panel 1 in that table defines white blood cell (WBC) subpopulations, such as B cells, NK cells, T cells, Tregs, monocytic myeloid-derived suppressor cells (moMDSC), and monocytes. Panel 2 characterizes different T-cell subsets. Panels 3 and 4 designate putative memory T stem cells (T_MSC; refs. 15, 16). A healthy donor sample of PBMCs (IRB# 10-001598) or PBMCs from patients was stained and run in parallel with the TIL sample as an internal quality control for staining and gating strategies. Of note, although we analyzed the PD-1 marker by flow cytometry (clone MIH4; Affimetrix), it was discarded from the final analysis after noting that MIH4 and pembrolizumab cross-reacted on the same binding site.

All flow data analyses were done with either FlowJo (Tree Star Inc.) or cyt software for visualizing viSNE maps (17). The gating strategy is described in Supplementary Figs. S1 to S3. Biexponential display was used in the analyses. The CytoAnalytics program Vasco (version 1.1.3) is based on exhaustive expansion software (18). One of its functions allows the analysis of all of the possible combinations (positive, negative, and unspecified) from the results of the FlowJo analysis. The statistical parameters of the filter utilized were a delta minimum of 2 and baseline readout ≥1, baseline versus treated, or responder versus nonresponders P value of ≤ 0.05, excluding null P values. Delta was defined as day of treatment minus baseline, and it serves to prevent large fold changes when the baseline is small (18). We also used the viSNE software program (17), where we gated for live lymphocytes and then removed all of the events found to be negative for all phenotypical markers. Then we used the viSNE algorithm with the cyt software package on the remaining cells.

Descriptive statistical analyses were done with GraphPad Prism (GraphPad) and/or the Vasco software program. The Pearson χ² test was used for testing difference in the percentage of responders in two dosage groups. Mann–Whitney (unpaired samples) and Wilcoxon matched-pairs signed rank (paired samples) tests were used to compare the pretreatment and on-treatment effect, and/or the Vasco software program. Confidence intervals (CI) were calculated by the Clopper–Pearson method.

Fifty-three patients receiving pembrolizumab underwent biopsies for intratumoral cell analyses from February 2012 to May 2013. Table 1 displays the patient characteristics, treatment administered, and clinical outcome. Seven patients (13%) had stage M1a, 15 (28%) had stage M1b, and 31 (58%) had stage M1c metastatic melanoma. Fourteen patients (26%) had prior immunotherapy only, 27 (51%) had previously received other treatments, and seven (13%) were treatment naïve. There was no correlation between the two different doses of pembrolizumab and patient response (P = 0.18). One patient was treated under Keynote 002, and his/her dose still remains blinded. Three (4%) patients had grade 3 or 4 toxicities on pembrolizumab (one with grade 3 elevation of liver function test, one with grade 3 colitis, and the other with grade 4 acute kidney injury). The rest of the toxicities were grade 1 or 2 in 14 (28%) patients, including vitiligo, myalgia, diverticulitis, fatigue, colitis, and pneumoniti. Nineteen (36%) patients had an objective tumor response, whereas 34 (64%) were nonresponders by the RECIST 1.1 criteria (19).

Twenty-seven baseline and 24 on-therapy tumor biopsies were analyzed to study changes in TILs (WBC) subsets (Supplementary Fig. S1). The percentage of cells expressing leukocyte common antigen (CD45⁺) in tumor biopsies increased, independent of clinical response, on PD-1 blockade (Fig. 1A). Of these CD45⁺ cells, the percentage of T cells (CD3⁺; P = 0.01) and B cells (CD19⁺CD3⁻ and CD20⁺CD3⁻; P = 0.04) increased in biopsies taken on treatment. Tumors from responding patients on therapy contained a higher percentage of T cells. The percentage of monocytes (CD14⁺) and CD56⁺CD3⁻ NK cells showed no significant change on treatment (Fig. 1B). Among T cells, there was a nonsignificant increase in the ratio of CD8⁺/CD4⁺ T cells when examining 22 pairs of tumors pretreatment and on treatment (P = 0.054; Fig. 1C). The frequency of the late activation marker HLA-DR, but not the CD25 early activation marker (refs. 20, 21; gating strategy described in Supplementary Fig. S2A and S2C), was slightly increased in both CD4⁺ and CD8 (CD4⁻) T-cell subsets (CD4⁺: P = 0.024; CD4⁻: P > 0.05; Supplementary Fig. S2B). A marginal increase was evident in B cells expressing the activation marker HLA-DR in tumors from patients who were treated (Supplementary Fig. S2D).

Two types of immune-suppressor cells were studied. Tregs were defined by the phenotype of CD45⁺CD3⁺CD4⁺CD25^HighCD127^Low (22); the proportion of Tregs changed little on treatment (Fig. 1D). In addition, we assessed the percentage of moMDSC, based on the phenotype of CD45⁺CD14⁺HLA-DR^Low (refs. 23–25; Supplementary Fig. S1), and found them significantly increased on treatment (P = 0.04; Fig. 1E). No differences between responders and nonresponders were detected for either population of immune-suppressor cells within the TILs, at baseline or while on PD-1 blockade.

To analyze the status of T cells at baseline, we assessed the CD3⁺ T-cell population for relevant immune phenotype markers (Fig. 2). Most T cells had a phenotype of previous exposure to cognate antigen, revealed by expression of CD45RO. Some T cells had a naïve phenotype expressing CD45RA, IL7 receptor α (CD127), and CD62L, but few cells expressed CCR7. The costimulatory marker CD28 was expressed more frequently than CD27. We also analyzed three markers (CD95, CD57, and PD-1) that are usually expressed by effector T cells, terminally differentiated T cells, or exhausted T cells (26); we observed a high expression of CD95, but lower CD57 and PD-1 at baseline.

In order to further characterize T-cell subsets, we analyzed the 14-parameter flow cytometry data (Supplementary Fig. S3), using multiparametric single-cell resolution in 46 baseline and 37 on-treatment biopsies. Of the 19,683 possible phenotypes analyzed, we discarded all phenotypes with either P values higher than 0.05 and less than 0.05, but with upper or lower 95% CI close to zero. We found a significant increase in the frequency of CD8⁺ T cells expressing the memory marker CD45RO in combination with the absence of CD57, CCR7, and CD27, which corresponds to an effector memory T-cell (T_EM) phenotype (Fig. 3A) in biopsies obtained while on anti–PD-1. Figure 3B and C show the phenotype with the least number of immune markers (CD8⁺CD4⁻CD45RO⁺), the phenotype with the highest number of immune markers (CD8⁺CD4⁻CD45RO⁺CCR7⁻CD27⁻CD57⁻), and T_EM phenotypes comparing baseline and on-therapy biopsies. Biopsies from patients who responded had an enrichment of memory T cells on treatment (Fig. 3B and C, right graph; P = 0.002 for CD8⁺CD4⁻CD45RO⁺; P = 0.006 for CD8⁺CD45RO⁺CCCR7⁻CD27⁻CD57⁻). Expression of other phenotypes did not change over time or was lower than 1%. These data demonstrate that CD8⁺ T_EM is the main characterized functional phenotype present in biopsies from patients on anti–PD-1 therapy, in particular in tumors of patients who responded to therapy.

Within the CD4⁺ T-cell subset (CD3⁺CD8⁻CD4⁺), we found a dominant population of CD4⁺ T cells expressing the memory marker CD45RO that were CD57 negative, in combination with the absence of CCR7, CD62L, and CD45RA (Fig. 4A). The CD4⁺ T_EM phenotypes significantly decreased on anti–PD-1 therapy. On the contrary, the CD4⁺ effector T-cell–like phenotypes CD8⁻CD4⁺CD45RO⁺CD57⁺ (P = 0.047), and two variations of this phenotype, CD8⁻CD4⁺CD45RO⁺CCR7⁻CD57⁺ (P = 0.03) and CD8⁻CD4⁺CD45RO⁺CCR7⁻CD62L⁻CD57⁺ (P = 0.05), significantly increased on therapy (Fig. 4B, left graph). Tumors from nonresponding patients had greater percentages of the following phenotypes: CD8⁻CD4⁺CD45RO⁺CD57⁺ (P = 0.024); CD8⁻CD4⁺CD45RO⁺CCR7⁻CD57⁺ (P = 0.022); and CD8⁻CD4⁺CD45RO⁺CCR7⁻CD62L⁻CD57⁺ (P > 0.05; Fig. 4B, right plots; refs. 27, 28). Taken together, these results show a decrease in the percentage of CD4⁺ T_EM in tumors on pembrolizumab treatment in both responders and nonresponders, and an augmented percentage of CD4⁺ effector T cells on anti–PD-1 treatment only in tumors of nonresponding patients.

We then focused the analysis on the nine paired biopsies obtained from patients who responded to therapy. The analysis of changes in CD8⁺CD45RO⁺ T cells was suggestive of a T-memory phenotype (Fig. 5A). The starting lesions were of different sizes, and the postdose tumors were obtained at different time points on-therapy (range, 19 to 186 days; median, 39 days). Five of the biopsies showed an apparent increase in the percentage of CD8⁺CD45RO⁺ cells, whereas three showed a clear decrease. Only one of those patients (patient #47) had both biopsies (before treatment and during treatment) performed from the same lesion, a left suprarenal metastasis. We applied viSNE analysis to better visualize the changes in T cells in this biopsy (Fig. 5B), which provided a global view of all cell subsets simultaneously. This analysis allowed us to conclude that the decrease in the frequency of CD45RO observed in Fig. 5A could be explained by a strong decrease in the total number of T cells in the postdose tumor as observed in Fig. 5B (lower row), likely due to taking the biopsy from a residual regressed tumor.

We also explored the presence and potential of on-treatment changes in the frequency of naïve-like T-cell (29) and T_MSC phenotypes described originally by Gattinoni and colleagues (29) and modified by Cieri and colleagues (16). Both of these populations were present at very low frequencies among TILs at baseline and did not change with treatment (Supplementary Fig. S4).

We performed similar analyses in peripheral blood samples from nine baseline and 14 on-therapy blood draws. The percentage of leukocytes slightly decreased on anti–PD-1 treatment, mainly due to a decrease in CD3⁺ T cells. Monocytes, B cells, and NK cells did not change. The ratio of CD8/CD4 increased slightly (Fig. 6A). We did not observe any change in cells with the Treg or moMDSC (Fig. 6B). There was no statistically significant change in any particular T-cell phenotype comparing baseline and on-therapy blood draws (Fig. 6C).

To better analyze any potential changes in leukocyte populations upon PD-1 blockade that were not evident by the combinatorial gating analysis, we applied the viSNE program. viSNE is not limited to high-low thresholds and can discern mid-level phenotypes, as well as nonlinear relations between markers, and hence is capable of searching for more complex phenotypes. We mapped the T cells of two responders and nonresponders facilitating a comprehensive comparison of all the cell phenotypes (Fig. 6D and E). Responders were patients #4 and #45, with blood taken at 59 and 41 days from study start, respectively. Nonresponders were patients #12 and #14 with the blood collected 42 and 60 days from study start, respectively. A distinct population of leukocytes had high expression of HLA-DR and CD56, and mid-level expression of CD14 and CD4. This population of cells increased by 9-fold in both on-treatment peripheral blood samples from the patients with a response to therapy.

In this study, we performed a thorough characterization of TILs in tumors from patients treated with PD-1 blockade therapy. The major finding was the increase in the frequency of memory T cells in patients with a tumor response, where the principal immune marker was CD8⁺CD4⁻CD45RO⁺ and combinations of it, giving a final most common phenotype of T_EM cells. This finding is consistent with the anticipated mechanism of action of this therapy. However, we would anticipate an increase in the frequency of CD8⁺ T cells with an effector phenotype, which we were not able to document. As opposed to the changes in CD8⁺ T cells, we did find CD4⁺ T cells expressing CD57, an effector T-cell phenotype also called proliferation-incompetent CD4 T cells, as they have been reported to produce IFNγ but have limited proliferation capacity (30). CD4⁺CD57⁺ T cells increased in tumors from patients with disease progression as opposed to responding lesions. Tumor-infiltrating T_EM cells have been found in other studies and may have good prognostic implications. Pagès and colleagues reported a T_EM phenotype in primary tumors of patients with colorectal cancer analyzed by flow cytometry (31). Furthermore, Koelzer and colleagues reported the presence of CD8 and CD45RO-positive TILs, which correlated with improved survival in patients, based on the IHC analysis of colorectal carcinoma samples (32).

Our findings of an enhanced proportion of cells with a phenotype of moMDSC (24) in patients' on-therapy biopsies from both responders and nonresponders could be interpreted as a by-product of inflammation in the tumor microenvironment on PD-1 blockade (33, 34). In fact, IHC and transcriptome analyses of on-therapy tumor biopsies suggest that several inflammatory cells and markers increase with anti–PD-1 or anti–PD-L1 treatment (10, 13). The same process may attract B cells into tumors obtained from patients on treatment. The significant enrichment of B cells in the tumor microenvironment may have a role in local antigen presentation (35). However, we could not link these B cells to tertiary lymphoid structures (36, 37) as our approach, based on single-cell flow cytometry analysis of tumor lysates, does not preserve tissue morphology (38) as does IHC (39).

Polychromatic flow cytometry has flourished over the past decade (40, 41), and it has been one of the primary factors contributing to the ability to document the heterogeneity of immune cells at the single-cell level (42). However, until recently, the ability to plot multidimensional flow data has remained a challenge despite efforts to simplify it with different computational methods (17,43–45). We chose to work with the Vasco (18) and the viSNE programs (17). The former has the advantage of the simplicity of the plots, the flexibility of the filters used to scan results across the sets, and the ability to study the weight of each immune marker from one at a time to all simultaneously (18). The latter has the ability to visualize the multidimensional data in simple two-dimensional plots maintaining the nonlinearity and geometry of the sample at the single-cell level. Thus, the strengths of the approach are the visualization of the samples and the simultaneous integration of all of the dimensions (17). In our studies, the application of both programs provided complementary data and improved our understanding of the immune cells in biopsies from patients treated with anti–PD-1.

The effects of PD-1 blockade in cells circulating in peripheral blood were minimal compared with the changes in tumors. No major characterized blood cell subtype changed significantly. However, a population of cells with both monocyte and T-cell markers was detected in on-treatment blood samples from 2 of 4 patients. Monocytes expressing CD56 have been previously described in peripheral blood by Gruenbacher and colleagues (46). Moreover, we cannot rule out the possibility that these cells may be the human counterpart of the mice iKDC or premature NK cells. In a study of patients with refractory gastrointestinal stromal tumors, HLA-DR⁺CD56⁺ cells correlated with overall survival and progression-free survival after treatment with imatinib mesylate and IL2 (47).

In conclusion, anti–PD-1 therapy results in an intratumoral increase in the frequency of T cells, B cells, and MDSCs. Memory T-cell phenotypes were found specifically increased in patients whose tumors responded to therapy. In contrast, changes were minor in peripheral blood cells, although we could detect an unusual population of what may be a subset of dendritic cells or HLA-DR NK cells in some responding patients. Despite being subjected to challenges derived from tumor heterogeneity that can drive different human immune signatures (48, 49), the different anatomical locations of the biopsies, the relatively small sample size, and the different collection time points, we have found some common characteristics in the intratumoral T-cell populations in melanoma biopsies from the responding patients. Our results help advance the knowledge of the effects of PD-1 blockade on the human T cells of the immune system.

A. Ribas is a consultant/advisory board member for Merck. J. Siebert has ownership interest (including patents) in CytoAnalytics. B. Chmielowski has received honoraria from the speakers bureau of Bristol-Myers Squibb, Genentech, and Prometheus; and is a consultant/advisory board member for Amgen, Astellas, Bristol-Myers Squibb, Genetech, Lilly, and Merck. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Ribas, B. Chmielowski, B. Comin-Anduix

Development of methodology: T. Chodon, D. Pe'er, B. Comin-Anduix

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Avramis, C. Kivork, B. Chmielowski, P.C. Tumeh, T. Chodon, B. Comin-Anduix

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Ribas, D.S. Shin, J. Zaretsky, J. Frederiksen, A. Cornish, E. Avramis, J. Siebert, X. Wang, B. Chmielowski, D. Pe'er, B. Comin-Anduix

Writing, review, and/or revision of the manuscript: A. Ribas, D.S. Shin, E. Avramis, P. Kaplan-Lefko, X. Wang, B. Chmielowski, J.A. Glaspy, P.C. Tumeh, T. Chodon, B. Comin-Anduix

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Ribas, E. Seja, C. Kivork, P. Kaplan-Lefko, B. Chmielowski, P.C. Tumeh, B. Comin-Anduix

Study supervision: A. Ribas, B. Chmielowski, B. Comin-Anduix

Flow cytometry was performed in the UCLA Jonsson Comprehensive Cancer Center (JCCC) and Center for AIDS Research. The authors thank Rongqing Guo and Li Wang for technical support.

A. Ribas has received NIH grants R35 CA197633, P01 CA168585, and Stand Up To Cancer – Cancer Research Institute (SU2C-CRI) Cancer Immunology Dream Team Translational Research Grant (SU2C-AACR-DT1012). A. Ribas and D. Pe'er were supported by a Stand Up To Cancer Phillip A. Sharp Innovation in Collaboration Award (SU2C-AACR-PS04). D.S. Shin was supported by Oncology (5T32CA009297-30), Dermatology (5T32AR058921-05), and Tumor Immunology (5T32CA009120-39) training grants and a Tower Cancer Research Foundation Grant. The Flow Cytometry Core Facility is supported by NIH grants CA16042 and AI 28697, and by the JCCC, the UCLA AIDS Institute, and the David Geffen School of Medicine at UCLA.

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.