Immune Toxicities Elicted by CTLA-4 Blockade in Cancer Patients Are Associated with Early Diversification of the T-cell Repertoire

Among significant recent advances in cancer immunotherapy, antibodies blocking immunologic checkpoints such as CTLA-4 and PD-1 can potentiate immune responses to cancer leading to durable clinical responses. Ipilimumab (Bristol-Myers Squibb) is a monoclonal, fully human IgG1 antibody against the T-cell coinhibitory receptor CTLA-4 that is approved for the treatment of melanoma (1, 2), but may have activity in a subset of patients in other malignancies including metastatic castrate-resistant prostate cancer (mCRPC; ref. 3). While some patients respond to ipilimumab, some also experience organ-specific toxicities, which are the result of increased immune activation, termed immune-related adverse events (IRAE). These include diarrhea and colitis, transaminitis, rash and pruritus, panhypopituitarism, adrenal insufficiency, thyroiditis, and pneumonitis (4, 5), and are often serious, requiring cessation of therapy and initiation of immunosuppression. For instance, in the context of mCRPC, 13% of mCRPC patients treated with ipilimumab at the 10 mg/kg dose demonstrate >50% prostate-specific antigen (PSA) responses, while 26% of patients develop grade 3–4 IRAEs (3). Biomarkers associated with clinical response to ipilimumab have been described (6–12), but those predicting IRAEs have not.

We and others have previously shown that CTLA-4 blockade can remodel the pool of circulating T cells. T cells recognize antigens through their T-cell receptors (TCR), which are primarily comprised of a α and β chain. The antigenic diversity of T cells is the result of VDJ gene recombination, which can be used to identify individual T-cell clonotypes. In prior work, we used next-generation sequencing of TCRβ chains from peripheral blood mononuclear cells (PBMC) from patients with cancer to show that CTLA-4 blockade promotes active remodeling of the T-cell repertoire, with both gain and loss of clonotypes but net bias toward gains leading to increased repertoire diversity overall. However, overall survival was not correlated with the generation of new clones, but rather with the maintenance of preexisting, high frequency (greater than 1 in 1,000) clonotypes (13). Here we examined whether development of IRAEs is associated with changes in the T-cell repertoire. Because CTLA-4 blockade is associated with inflammation in specific tissues, we postulated that the treatment would expand tissue-reactive T-cell clones in patients that developed the corresponding toxicity, as has been observed in other tissue-specific autoimmune pathologies such as celiac sprue and gastrointestinal graft-versus-host-disease. Instead, we found that an early increase in diversity and the generation of new clones after administration of the combination of ipilimumab and granulocyte-monocyte colony-stimulating factor (GM-CSF) is specifically correlated with the development of IRAEs. These results indicate that increased T-cell diversity can be deleterious to patients via promotion of autoreactivity.

Cryopreserved PBMCs were sequenced from mCRPC patients treated with anti–CTLA-4 (ipilimumab; Bristol-Myers Squibb) and GM-CSF (sargramostim; Sanofi) in a phase I/II clinical trial (ClinicalTrials.gov identifier: NCT00064129) as described previously (14, 15). Up to 4 doses of ipilimumab were given ranging from 1.5 to 10 mg/kg every 4 weeks, and GM-CSF was given at 250 μg/m² per day on the first 14 days of every cycle. Patient characteristics for the entire 42-patient cohort were previously reported (15). Ipilimumab was administered every 4 weeks for 4 planned doses. Serial PBMCs obtained at baseline and six posttreatment timepoints from 35 of these patients were sequenced; of these, 21 patients had available sequence at both pretreatment (week 0) and immediate posttreatment (week 2) timepoints and were utilized for the analyses at these early timepoints unless otherwise stated. Patients were not restricted by HLA alleles. Informed consent was obtained for all investigations. We restricted our initial analysis to consensus IRAEs listed in the manufacturer's prescribing information for ipilimumab (http://packageinserts.bms.com/pi/pi_yervoy.pdf).

PBMCs were thawed into FACS wash (PBS 2% BSA) and washed twice with FACS wash. Samples were stained with designated panels for 20 minutes at 4°C and washed twice with FACS wash. Cells requiring intracellular staining were fixed and permeabilized with BD Cytofix/Cytoperm buffer (catalog no. 554722) according to the manufacturer's protocol. Intracellular staining with antibodies was carried out for 30 minutes at 4°C and washed twice with FACS wash. Cells were acquired on a LSRII flow cytometer (BD Biosciences) and data were analyzed with FlowJo analysis software version 9.7.5 (FlowJo, LLC). Absolute counts (per microliter of blood) for each immune subset is calculated by multiplying the percentage of each subset with the preceding parent subset and with the absolute lymphocyte count quantitated on the day of blood drawn.

The amplification and sequencing of TCRβ repertoire from RNA, read mapping to clonotypes via identification of V and J segments, and counting of the number of unique clonotypes have been previously described in detail (13). Of note, after filtering for read quality, reads were mapped to a clonotype if at least 2 identical reads were found in a given sample. Clonotype frequencies were calculated as the number of sequencing reads for each clonotype divided by the total number of passed reads in each sample.

Demographic and clinical characteristics were summarized by descriptive statistics. In general, frequency distribution and percentages were used to summarize categorical variables, and median with interquartile range was used to describe continuous variables. Comparison of continuous variables between two groups was performed using the Wilcoxon t test. χ² test was applied to determine statistical association between two categorical variables. Statistical significance was declared at α < 0.05 and no multiple testing adjustment was done. All statistical analysis was done with the statistical computing software R (https://www.r-project.org/).

Clonality was used to measure the diversity of the clonotype population for each patient at each time point. The clonality was calculated using the formula |$1 - \mathop \sum_{i{\rm{\ }} = {\rm{\ }}1}^n {p_i}{\rm{\ }}lo{g_e}({p_i})/lo{g_e}( {\rm{n}} )$|⁠, where |${p_i}{\rm{\ }}$|is the frequency of clonotype i for a sample with n unique clonotypes. Of note, this metric is normalized to the number of unique clonotypes, and in our dataset, clonality was found to be a robust metric and was not significantly correlated with the number of unique clonotypes found in each sample (P values for correlation with clonality at week 0 = 0.263 for molecules, 0.852 for counts; P values for correlation with clonality at week 2 = 0.048 for molecules but this is a positive correlation, 0.309 for counts). Clonality was compared between week 0 and 2 by paired Wilcoxon test, and clonality comparisons between patients with IRAEs versus patients without IRAEs, or between responders versus nonresponders at each timepoint were performed using two-sample Wilcoxon test. To determine the relative change in diversity over time, relative clonality was calculated as the ratio of the clonality at two consecutive timepoints; comparisons of this metric between patients with versus without IRAEs were done by two-sample Wilcoxon test. To explore the effect of other adverse events (AE) on the change of clonality from week 0 to 2 for each type of AE, the relative clonality (week 2/week 0) between patients with that AE versus those without that AE was compared by two-sample Wilcoxon test.

To measure the commonality between TCR sequences in week 0 (pretreatment) and week 2 (posttreatment) for each subject, the proportions of clones only present at week 2, only present at week 0, and present in both week 0 and week 2 were calculated. The read depth as far as RNA molecules was largely similar between IRAE and non_IRAE groups as well as week 0 and week 2 samples (P = 0.412 for week 0 vs. week 2; 0.9712 for IRAE vs. non_IRAE at week 0) and did not vary in a way that would account for the presence of new clones in the AE group at week 2. In addition the amount of RNA input was not significantly correlated either with IRAE status, or with clonality of the entire cohort or IRAE/non_IRAE groups (data not shown). To examine the change in TCR sequence frequency from week 0 to week 2, the fold change (FC) was defined as the sequence frequency at the week 2 divided by the sequence frequency at week 0. Each sequence was categorized as “increased” if fold change (FC) was ≥ 4, as “decreased” if FC was ≤ 0.25, and as “unchanged” if 0.25 < FC < 4. All TCR sequences detectable at week 0 or 2 were included in the analysis. For clones with nonmeasurable frequency counts at only one timepoint (and measurable at the other timepoint), the number of reads at the nonmeasurable timepoint was arbitrarily set to 1, and then FC was calculated as above. For each subject, the percentage of TCR sequences falling into each change category was computed. The comparison of the proportions between IRAE versus non_IRAE patients was done by two-sample Wilcoxon test.

PBMCs from weeks 2 and 6 were FACS sorted (FACSAria, BD Biosciences) into four populations: Treg (CD4⁺ CD25^hi CD127^lo), helper T (CD4⁺ CD25^lo CD127⁺), naïve CTL (CD8⁺ CD27⁺ CD45RA⁺), and nonnaïve CTL (CD8⁺ CD27⁻ or CD27⁺ but CD45RA^−/lo), and then the clonotypes present in these subpopulations were identified as above. Each of these clonotypes from sorted cells was used to mark this clonotype as arising from a particular T-cell subset when found in bulk PBMCs from the same patient at all timepoints for further analysis.

To characterize the functional effects of ipilimumab and GM-CSF on T-cell phenotype, we first looked at changes in T-cell activation markers in the peripheral blood of mCRPC patients who received both of these agents. Treatment with ipilimumab and GM-CSF induced proliferation in both circulating CD4⁺ FOXP3⁻ and CD3⁺ CD4⁻ T cells (which contain CD8⁺ T cells as well as a mixture of other populations) as measured by increases in either absolute numbers or percentages of Ki67⁺ lymphocytes (relative percentages shown for illustration in Fig. 1A). Proliferation was maximal at 2 weeks after the first dose (Supplementary Fig. S1A and S1C; for week 0 versus week 2 comparisons of Ki67⁺ CD4⁺ FOXP3⁻ cells, P < 0.001 for absolute numbers or percentages; for Ki67⁺ CD3⁺ CD4⁻ cells, P = 0.001 for absolute numbers and P < 0.001 for percentages; n = 12 patients with available Ki67 data from weeks 0 and 2). These changes, however, were not significantly associated with IRAEs at any timepoint (Supplementary Fig. S1B and S1D; n = 21 patients across all timepoints).

Given this finding of robust activation across the study cohort, we assessed how this is reflected in changes in the TCR repertoire. Using next-generation sequencing of TCRs from peripheral blood across time, we calculated the clonality index for each available sample (n = 35 patients with available sequence at any timepoint), which is inversely proportional to diversity (i.e., a lower clonality denotes a more diverse TCR repertoire) and is normalized to the number of unique clonotypes. A majority of treated patients experienced an immediate decline in clonality within 2 weeks of starting ipilimumab with GM-CSF (Supplementary Fig. S2A); similar results are found with other diversity indices such as Hill numbers (16). Declining clonality continued through the first 6 weeks of treatment (Supplementary Fig. S2B), which indicates that there is successive TCR repertoire diversification with repetitive anti-CTLA-4 administration, which in our cohort of mCRPC patients, occurs within the first several doses.

Given that we initially observed robust T-cell activation following ipilimumab and GM-CSF, our initial hypothesis was that IRAEs would be correlated with the emergence of an oligoclonal response (i.e., characterized by a narrowing of the repertoire toward fewer TCR clones) following checkpoint inhibition, consistent with clonal expansion driven by a dominant autoantigen-driven process. To evaluate this, we examined the frequency distribution of unique clonotypes in ipilimumab-treated patients who either experienced an IRAE or did not (non_IRAE). Contrary to our initial expectation of an induced oligoclonal response, treatment triggered a broad expansion in lower-frequency clonotypes in IRAE patients (representative patient shown in Fig. 1B).

To more directly characterize the diversity of the TCR repertoire at each timepoint, we assessed both the number of unique TCR clonotypes and the clonality index in patients. Total number of unique clonotypes increased in the overall population (Fig. 1C), with a concomitant decrease in clonality during the first 2 weeks of treatment (Fig. 1D, P = 0.0016) indicating significant diversification of the T-cell repertoire. Comparing the patients with IRAE (Fig. 1E and F) versus non_IRAE (Fig. 1G and H), we found that the IRAE patients had the significant declines in clonality (P = 0.023), while non_IRAE patients did not (P = 0.057). Similar results supporting a greater increase in diversity in IRAE patients are found using other diversity indices such as Hill numbers (16). Hence, early T-cell diversification after ipilimumab plus GM-CSF is related to the development of IRAEs.

To assess the nature of these changes in the repertoire, we classified each unique clone as found only after checkpoint blockade at week 2, only at week 0, or at both timepoints. Most clones were newly detected at week 2 (>50%), while slightly less than half of clones were only present at baseline; and a small fraction of clones (<10%) were present at both timepoints (Fig. 2A; new clones from week 2 at left; clones present only at baseline at week 0 at right). Although both IRAE and non_IRAE patients have a large fraction of clones newly detected at week 2, IRAE patients had a significantly greater fraction of newly detected clones compared with non_IRAE patients (P = 0.042), as well as a smaller fraction of clones that were present only at baseline (P = 0.049; Fig. 2B). Hence, the development of IRAEs is associated with an increased number of TCR clones newly detected in the periphery following ipilimumab plus GM-CSF.

To assess the dynamics of T-cell clones, we classified clonotypes into those whose frequency increased >4-fold, decreased >4-fold, or remained the same over time (change in frequency <4-fold) from week 0 to week 2 (data for all patients are shown in Fig. 2C). IRAE patients had a higher proportion of clones that increased with treatment compared with non_IRAE patients (P = 0.028; Fig. 2D). Thus, IRAEs are also associated with increases in preexisting clonotypes in addition to newly detected TCR clones in the blood.

To assess the kinetics of changes in clonality in IRAE and non_IRAE patients, we calculated the relative clonality for each group at each timepoint relative to the immediately preceding timepoint, and found significantly greater declines in clonality at week 2 relative to week 0 in the IRAE versus non_IRAE population (P = 0.045; Fig. 3A), but not at later timepoints. This confirms that changes in diversity in the IRAE group are concentrated at early timepoints after ipilimumab and GM-CSF. This early change was evident in one representative patient (Fig. 3B), who had an initial drop in clonality at the 2-week timepoint, preceding the development of an IRAE, panhypopituitarism, and clinical response. This patient also appeared to demonstrate a declining clonality in the several timepoints preceding onset of the actual IRAE. However, when we assessed all 8 IRAE patients with data at weeks 0 and 2 for changes in clonality from the timepoint immediately preceding IRAE onset to the timepoint immediately after onset, any changes in clonality at the time of IRAE onset were not significant (P = 0.07813 by Wilcoxon test). In our 42-patient cohort, there was correlation between toxicity and response in a subset of patients: 12 patients had IRAEs and 5 patients had clinical responses (≥50% PSA decline), and 4 of 5 responders also developed IRAEs. As with IRAEs, clinical responses were also significantly associated with an early decline in clonality at 2 weeks after starting ipilimumab and GM-CSF (Fig. 3C, left), while nonresponders lack this correlation (Fig. 3C, right, P = 0.01 and 0.055, respectively). However, when we dichotomized the population into those patients who lived longer or shorter than the median overall survival on study of 23.6 months, there was no specific correlation between decline in clonality and longer overall survival (data not shown).

We examined whether other observed AEs might be similarly classified as IRAEs based on their effect on clonality. Given the small number of patients with any given AE, we added patients with various AEs to the IRAE category to see whether this addition would increase the significance of the IRAE versus non_IRAE comparison. Including temporal arteritis with IRAEs as a revised AE classifier increased the significance of the comparison between patients with and without AE (Table 1, P = 0.029 compared with P = 0.045 for consensus IRAEs alone); this is one example of an AE that may be plausibly reclassified as an IRAE. Other AEs did not have this effect: in particular, other thrombotic AEs (including deep venous thrombosis/pulmonary embolism), cardiac AEs (including troponin leak, arrhythmias), all grade 3+ AEs, and fatigue (a common AE with checkpoint inhibition) did not result in a decreased P value when added to the IRAE group (Table 1 and data not shown).

Using FACS to isolate CD4⁺ and CD8⁺ T cells that were then sequenced, we followed the evolution of the repertoire after ipilimumab and GM-CSF in these distinct T-cell subsets. We sorted these subsets from 2 IRAE patients and 1 non_IRAE patient at 2 posttreatment timepoints (weeks 2 and 6), and used the identified clonotypes coming from CD4⁺ or CD8⁺ T cells to track these populations in all available bulk PBMC populations over time. Compared with one non_IRAE patient, the two available IRAE patients appeared to have lower CD4⁺ T-cell clonality (Fig. 4A, top plot), and lower CD8⁺ clonality (Fig. 4A, bottom plot), across all timepoints. Despite these trends, we did not see a significant difference in the numbers of Tregs between IRAE and non_IRAE patients after ipilimumab treatment (Fig. 4B–D). Using a similar clonotype analysis as before, we found that IRAE patients also demonstrated a trend in CD8⁺ T cells toward higher frequencies of newly detected clones (Supplementary Fig. S3A), as well as clones with increased frequencies (Supplementary Fig. S3B), compared with CD4⁺ T cells.

Despite assessing circulating T cells serially over the first 12 weeks of treatment, we found that the most relevant change in T-cell repertoire occurred at 2 weeks. Early T-cell diversification after checkpoint inhibitors may therefore prove useful as an indicator of patients at risk for later IRAE development, and could serve as a new adjunct to current clinical management algorithms for IRAEs, which rely upon early recognition and intervention (17). This is especially pertinent as CTLA-4 blockade is being combined with anti-PD1 therapy (18) with significantly improved clinical outcomes, but where the vast majority of patients develop IRAEs with combination therapy (19). Also this may be relevant to the timing or sequencing of CTLA-4 blockade when given in combination with radiotherapy and how this may affect the types of repertoire changes leading to either toxicity or response (20). The directionality of changes in diversity we observe are likely related to the specific modality of immunotherapy used (e.g., checkpoint inhibition in this study); in other work, we have recently shown that treatment of mCRPC patients with the immunotherapy sipuleucel-T, which is thought to be a cancer vaccine, results in a decline in circulating T-cell repertoire diversity, which is the opposite of what we see here (21).

The specific changes in the TCR repertoire we observe in IRAE patients place some constraints on possible mechanisms by which IRAEs develop. First of all, although we demonstrate an early increase in diversity in IRAE patients, the actual toxicities do not typically manifest until later, with variable kinetics depending on the affected organ system: the skin is typically involved earliest, followed by colitis (after 1–3 doses), then hepatitis and endocrinopathies last (5). Particularly for later-onset IRAEs, this lag between early repertoire changes and toxicity implies that if repertoire diversification is part of the pathogenesis of IRAEs, this may represent a necessary step but is clearly not sufficient by itself—other steps may occur at subsequent times prior to the clinical manifestation of toxicity. We cannot exclude the possibility of additional repertoire changes that are more proximal in time to the actual onset of toxicity; however, our findings speak to early repertoire diversity immediately following initiation of ipilimumab as a possible means of predicting IRAE development, and also a possible first step in the pathogenic mechanism underlying toxicity.

Our findings of increased diversity would support a mechanism whereby autoreactivity to multiple antigens is induced by ipilimumab, in part by mobilization of newly detected clones. The specific clones mediating IRAE may be enriched within the population of activated T cells we see after ipilimumab, although for technical reasons it would not be possible to look specifically at clones enriched in Ki67⁺ T cells as Ki67 staining requires fixation. It remains unclear whether additional ipilimumab doses result in further enrichment of these clones, although at a population level, there does not appear to be further diversification after the first dose in IRAE patients based on relative clonality (e.g., Fig. 3B). This would represent a distinct mechanism of pathogenesis from other tissue-specific immunopathologies, which are thought to be driven by an oligoclonal response; for instance, celiac sprue is thought to be driven by distinct gluten-derived antigens presented by HLA-DQ (22), and in acute gastrointestinal graft-versus-host disease, patients with steroid-refractory disease exhibit similar TCRβ repertoires at different sites in the gastrointestinal tract as well as oligoclonal expansion postdiagnosis of the TCR clones found most frequently in the gut, suggesting a pathogenic mechanism driven by relatively few antigens (23). Formal testing of this hypothesis would require additional extensive efforts including confirmation of increased TCR diversity in IRAE-affected tissue, further FACS profiling of the surface phenotype of tissue-infiltrating lymphocytes including checkpoint and adhesion receptor expression, pairing of TCRα with β chains using single-cell approaches to provide a more complete picture of repertoire specificity, and mapping of specific antigens that may be recognized by identified TCRα/β heterodimers. These efforts may be hampered by the difficulty in obtaining tissue involved by IRAEs as well as the potential difficulty of identifying specific autoantigen-reactive clones given that numerous low-frequency clonotypes are generated by ipilimumab. However, pending confirmation, our results may suggest an alternate scenario where the emergence of TCR diversity in IRAE patients reflects the polyclonal expansion of numerous T-cell clones, which contribute to the development of end-organ toxicity, but the actual recognition of specific autoantigens may in fact be less critical in the pathogenesis of IRAEs. This is reminiscent of the phenotype of CTLA-4 knockout mice, which develop a fulminant multi-organ lymphoproliferative disorder characterized by tissue infiltration by polyclonal-activated T-cell blasts, which appears to be more autoinflammatory rather than autoimmune in nature (24–26). Of note, we did not find evidence for a quantitative difference in Tregs correlating with IRAE development.

Our findings raise a larger question of whether IRAEs and clinical response after treatment with ipilimumab are driven by shared versus unique processes. TCR diversity can be associated with both beneficial and detrimental outcomes, consistent with clinical associations that have been previously reported (27, 28). The observation that almost all of our clinical responders had IRAEs, and that increased diversity is correlated with PSA response as well as IRAE development, suggests that in a subset of patients who experience both toxicity and response, there may be some common underlying mechanisms that may involve diversification of the TCR repertoire. At the same time, our observation that most of our patients with IRAEs did not develop PSA responses, and that longer overall survival was not correlated with increased diversity, also indicates that for most patients, toxicity and clinical benefit are likely mediated by divergent pathways. If initial diversification of the T-cell repertoire represents a common initial pathway in the subset of patients experiencing both IRAEs and response, it is unlikely that these distinct outcomes are driven by shared or common antigens. Whether these initial changes are stochastic or reflect other factors (including tumor-intrinsic features such as neoantigen load, or host-intrinsic polymorphisms) will require further investigation.

L. Fong reports receiving a research grant from Bristol Myers Squibb, Dendreon, Genentech, Amgen, Merck, and Abbvie. M. Faham has ownership interest (including patents) from Adaptive Biotechnologies. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Fong, J. Cham, M. Klinger

Development of methodology: L. Fong, J. Cham, L. Zhang, M. Klinger, M. Faham

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Cham, S.S. Kwek

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Fong, D.Y. Oh, J. Cham, L. Zhang, G. Fong, M. Klinger

Writing, review, and/or revision of the manuscript: L. Fong, D.Y. Oh, J. Cham, L. Zhang

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

Study supervision: L. Fong

This work was supported by the NIH 1R01 CA163012 grant to J. Cham, L. Zhang, and L. Fong. L. Fong also received the NIH 1R01 CA136753 grant and a grant from the Prostate Cancer Foundation. S.S. Kwek received a grant from the Peter Michael Foundation. D.Y. Oh received the NIH 4T32 CA177555 grant.

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.