Immune Checkpoint Blockade Enhances Shared Neoantigen-Induced T-cell Immunity Directed against Mutated Calreticulin in Myeloproliferative Neoplasms

Immune-based therapies have revolutionized the treatment of cancer, resulting in unprecedented response rates and even complete remission (1, 2). Inhibition of immune checkpoint receptors, such as PD-1 and CTLA4, in particular, have proved successful in the treatment of several different tumor types, and their combination with other modalities is also gaining approval. Blockade of PD-1 and CTLA4 signaling, either alone or in combination, has been reported to reinvigorate tumor antigen-specific T cells, thereby ameliorating antitumor activity (3, 4). The clinical efficacy of these immunotherapies is correlated in part with mutation load and is likely based on restoring T-cell recognition of neoantigens, non-self antigens that arise from somatic mutations in tumors, in the context of MHC molecules (3, 5–8). Neoantigen-specific T cells are not subject to immune tolerance, and hence they have the potential to exhibit strong effector responses specifically against malignant cells. However, due to intertumoral heterogeneity of mutations, neoantigen-based immunotherapies for the most part remain personalized and limited to individual patients (9, 10).

BCR–ABL1-negative myeloproliferative neoplasms (MPN), which include polycythemia vera (PV), essential thrombocythemia (ET), and primary myelofibrosis (MF), are chronic hematologic malignancies that are characterized by hyperproliferation of blood cells. The initiation and progression of MPN are associated with recurrent somatic driver mutations in Janus kinase 2 (JAK2), the thrombopoietin receptor gene MPL and calreticulin (CALR). Sixty-seven percent of patients with ET and 88% of patients with MF, who lack JAK2 and MPL mutations, carry mutations in the exon 9 of CALR gene (11, 12). The mutated CALR protein (mut-CALR) mediates transformation by activating JAK–STAT signaling through its binding to MPL (13, 14). Unlike JAK2 and MPL mutations, CALR mutations are frameshift mutations. To date, more than 50 types of insertions or deletions in the exon 9 of CALR gene have been reported (15). All of these mutations lead to a +1 shift in the open reading frame and result in the formation of an altered C-terminus with an identical 36-amino acid (aa) sequence that is shared by all patients with CALR⁺ MPN (11). The two most frequent mutation types, L367fs*46 (type I) and K385fs*47 (type II), found in 80% of patients with CALR⁺ MPN, share a 44-aa sequence (16).

Except for hematopoietic stem cell transplantation, curative treatments for patients with MPN are not available (17). Therefore, it is imperative to identify more effective treatment options for these patients. We hypothesized that the uniformity of mut-CALR marks it as an attractive MPN-specific, shared neoantigen candidate that could be targeted for the development of immunotherapy regimens for CALR⁺ patients. The shared mut-CALR C-terminus is at least 36 aa long and exhibits limited similarity to wild-type (WT) CALR (16). Hence, mut-CALR could incorporate multiple epitopes, which if presented on MHC could elicit antitumor T-cell responses with minimal cross-reactivity to WT protein expressed on nonmalignant cells. We therefore evaluated the immunogenicity of mut-CALR and found that a subset of patients with CALR⁺ MPN indeed develops specific T-cell responses against mut-CALR that can be detected in vitro.

We also considered that an exhausted state, driven by chronic antigen exposure, might blunt the detection of mut-CALR responses in some patients. Therefore, we determined whether the expression of checkpoint molecules, namely PD-1 and CTLA4, regulated mut-CALR–specific T-cell immunity. We report here that blockade of PD-1 and CTLA4 signaling in patients with CALR⁺ MPN led to both in vivo and ex vivo clonal expansion of T cells recognizing mut-CALR. Together, our results support the development of immunotherapy approaches targeting mut-CALR, either in the form of neoantigen-specific vaccines or adoptive T-cell therapies for elimination of malignant clones, and also provide a rationale for testing immune checkpoint blockade in patients with CALR⁺ MPN.

Patients carrying frameshift mutations in the CALR gene share an identical 36-aa sequence in the C-terminus of the mut-CALR protein (11). This novel sequence exhibits limited similarity to WT CALR (16), is not expressed elsewhere, and hence is not subject to immune tolerance. Therefore, we hypothesized that the mut-CALR as a neoantigen would elicit specific T-cell responses that exhibit minimal cross-reactivity to the WT CALR. To assess whether this neoantigen could potentially elicit a cellular immune response, we investigated T-cell responses against mut-CALR neopeptide in peripheral blood mononuclear cells (PBMC) from a cohort of patients with MPN carrying CALR mutations. Eighteen patients with MPN with CALR⁺ ET (n = 7), MF arising from ET (ET-MF; n = 7), or primary MF (n = 4; Supplementary Table S1) were assessed for underlying mut-CALR–specific immunity. To this end, PBMCs from patients with CALR⁺ MPN were stimulated in vitro with pooled overlapping long peptides (OLP; 14–15 aa) spanning the mutated (Supplementary Fig. S1A) or WT (Supplementary Fig. S1B) CALR C-terminus. After expansion, the cells were restimulated with the OLP pools, and IFNγ production was measured by ELISPOT (Fig. 1A; ref. 18). A significant increase in IFNγ production was observed when the cells were stimulated with mut-CALR OLPs as compared with WT OLPs that spanned the C-terminus tail of the protein (Fig. 1B and C). Of note, mut-CALR–induced IFNγ production was observed with a greater frequency in ET, but not in patients with primary MF (Supplementary Fig. S2A–S2C). Next, we characterized the mut-CALR–specific T-cell subsets in patients with MPN by intracellular staining. Because of limitations in cell numbers, we were able to analyze 11 of the 18 patients in our cohort. Mut-CALR–induced IFNγ production, identified by comparing stimulation with mut-CALR OLPs and control peptides derived from myelin oligodendrocyte glycoprotein (MOG), was observed primarily in CD4⁺ T cells (Fig. 1D and E). These results demonstrated that T cells from patients with CALR⁺ MPN could react to mut-CALR. Yet, as T cells in these assays were stimulated with mut-CALR OLPs prior to expansion and immunogenicity evaluation, it remains possible that T cells from patients with CALR⁺ MPN were primed in vitro. To assess directly whether in vivo T-cell priming has occurred in patients with CALR⁺ MPN, we performed ex vivo T-cell ELISPOT assays using PBMCs from a total of 19 patients with JAK2^V617F+ MPN and 22 patients with CALR⁺ MPN. PBMCs from patients with MPN were stimulated with mut-CALR OLPs or control peptides and mut-CALR–specific T-cell responses were monitored after 48 hours. No mut-CALR–specific T-cell responses were detected ex vivo (Supplementary Fig. S2D), suggesting either a lack of spontaneous mut-CALR–specific T-cell immunity in patients with CALR⁺ MPN or limitations in the detection of responses in nonexpanded cells due to a low frequency of mut-CALR–specific T cells.

CALR protein is critical for the folding and assembly of MHC class I molecules, and CALR-deficient cells are reported to have reduced levels of cell-surface MHC class I expression as well as reduced efficiency in antigen presentation (19, 20). Therefore, we examined whether the presence of CALR mutations in patients with MPN was associated with a decrease in cell-surface expression of MHC molecules. Both MHC class I and II molecules were expressed by patient PBMCs (Supplementary Fig. S3A and S3B), and the expression of MHC class I molecules was significantly greater in CALR⁺ MPN PBMCs than in healthy donor (HD) PBMCs (Supplementary Fig. S3A). The proinflammatory milieu that characterizes the MPNs (21), as well as certain treatment modalities, such as IFNα, among patients with MPN (22, 23) may contribute to the increased MHC class I expression. Because some of the patients in our cohort have been treated with IFNα, we assessed the role of IFNα treatment in modulating surface MHC class I expression levels. To this end, we treated cells from a CALR⁺ MPN patient–derived cell line (24) with IFNα and observed a dose-dependent increase in MHC class I expression (Supplementary Fig. S3C). Importantly, CD34⁺ cells of patients with MPN, which may harbor CALR mutations (25), also expressed MHC molecules (Supplementary Fig. S3D–S3F). These observations suggest that the paucity of anti–mut-CALR responses in patients with CALR⁺ MPN was not due to a reduction of MHC class I expression by PBMCs or CD34⁺ cells. Altogether, our results indicate that a subset of patients with CALR⁺ MPN develop T-cell responses that are specific to mut-CALR, which are primarily of the CD4⁺ T-cell phenotype.

Although we observed mut-CALR–specific T-cell responses in patients with CALR⁺ MPN, the responses were found only in a subset of study subjects and were of low frequency in peripheral blood. Therefore, we considered the possibility that antigen-specific T cells were undergoing exhaustion due to chronic antigen exposure. We therefore evaluated the surface expression levels of a number of immune checkpoint receptors (PD-1, CTLA4, LAG3, TIM3, TIM4, TIGIT, B7-H3, B7-H4, and VISTA) on PBMCs of patients with CALR⁺ MPN. We found that MPN T cells exhibited greater degrees of expression of multiple cell-surface inhibitory molecules compared with HD T cells (Fig. 2A and B). Among these checkpoint receptors, PD-1 displayed the highest proportional increase (19.2% ± 10% in MPN vs. 3.62% ± 1.69% in HD), with expression detected in both CD8⁺ and CD4⁺ T-cell subsets (Fig. 2C). Similarly, increased CTLA4 expression was also detected in both T-cell subsets (Fig. 2D). On the basis of this, we hypothesized that the elevated PD-1 and CTLA4 expression could play a role in suppressing mut-CALR–specific T-cell responses. Accordingly, we reexamined mut-CALR–specific T-cell responses in CALR⁺ MPN PBMCs in the context of PD-1 or CTLA4 blockade. PD-1 or CTLA4 signaling on T cells was inhibited by monoclonal blocking antibodies, and mut-CALR–specific T-cell immunity was evaluated by ELISPOT, as described in Fig. 1A–C. T-cell responses against mut-CALR OLPs were recovered in 3 CALR⁺ MPNs. These cells now produced IFNγ when PD-1 or CTLA4 signaling was blocked (Fig. 2E and F). In addition to IFNγ, cells produced multiple other cytokines including TNF, suggesting that mut-CALR–specific T cells are polyfunctional (Supplementary Fig. S4). When pre- and post-checkpoint inhibition responses were combined, 11 of the 18 patients in the MPN cohort exhibited mut-CALR–specific T-cell immunity.

Apart from the increased checkpoint receptor expression, suppression of T cells in patients with cancer can be attributed to the expansion of other immunoregulatory cell populations (26). Therefore, we investigated the presence of myeloid-derived suppressor cells (MDSC) and regulatory T cells (Treg) in MPN PBMCs from the same cohort. We found that MDSCs, defined as Lin⁻CD33⁺CD11b⁺HLA-DR⁻CD14⁻ cells (27), were significantly expanded in patients with CALR⁺ MPN when compared with HDs (Supplementary Fig. S5A and S5B). However, the frequencies of Treg populations, CD4⁺CD25⁺FOXP3⁺, were comparable between patients with CALR⁺ MPN and HDs (Supplementary Fig. S5C and S5D). We also found that the average percentage of CD3⁺ T cells in the peripheral blood of patients with MPN was lower than in HDs (Supplementary Fig. S5E–S5G). These observations may account for why blockade of PD-1 and CTLA4 failed to rescue mut-CALR–specific T-cell responses in some patients. Together, these data indicate that peripheral blood T cells in patients with CALR⁺ MPN exhibit an exhausted phenotype, and checkpoint blockade might be an effective strategy to restore mut-CALR–specific T-cell responses in a subset of patients.

To gain further insights into checkpoint receptor–mediated T-cell suppression in patients with CALR⁺ MPN, we took advantage of an ongoing phase I/II clinical trial (NCT03065400), in which patients with advanced MPN are undergoing treatment with pembrolizumab (Fig. 3A). One of the patients enrolled in the study expressed type I CALR mutation (c1092_1143del52; PT#15 in Supplementary Table S1). Using longitudinal blood samples available from this patient, we monitored the changes in the frequencies of T cells before and after pembrolizumab treatment by both flow cytometry and T-cell receptor (TCR) sequencing. Prior to therapy, only 4% of the patient's PBMCs were T cells, and of these 32% expressed PD-1 (Supplementary Fig. S6A). Importantly, the proportion of T cells in peripheral blood greatly increased after pembrolizumab treatment, reaching up to 26% and 18% of PBMCs after 2 (T1) and 6 (T2) cycles of treatment, respectively (Fig. 3B–D). Pembrolizumab treatment was accompanied by increases in both peripheral blood CD4⁺ and CD8⁺ T cells (Supplementary Fig. S6A). After pembrolizumab administration, PD-1–expressing T cells in peripheral blood were undetectable by commercial anti–PD-1 antibodies (Supplementary Fig. S6A), probably due to the persistent binding of pembrolizumab to PD-1 on T cells in vivo, as described previously (28, 29). In addition, TCR Vβ sequencing was undertaken to evaluate the extent of clonal expansion. TCR Vβ chains of peripheral blood T cells before and after pembrolizumab treatment were sequenced, and the diversity of TCR repertoire was calculated by Pielou's evenness metric. The results indicated that T cells not only expanded but also showed greater clonality after pembrolizumab treatment (Fig. 3E). After the initial treatment cycles, 8 T-cell clones were significantly expanded compared with baseline (pretreatment; Fig. 3F). Each of these clones remained significantly expanded after additional cycles of treatment and there were 15 more T-cell clones that were also significantly expanded compared with baseline (Fig. 3G). These observations suggested that in vivo PD-1 blockade may help reinvigorate T cells in patients with MPN.

To address this possibility, we evaluated changes in antigen-specific, namely mut-CALR–specific, T-cell responses in this patient with CALR⁺ MPN prior to and during pembrolizumab administration. At baseline, peripheral blood T cells from this patient did not have detectable mut-CALR–specific responses, even after in vitro inhibition of PD-1 with antibody blockade. After pembrolizumab treatment, however, mut-CALR–specific T-cell responses were evident, as shown by increased IFNγ production (Fig. 3H). Interestingly, in vitro addition of PD-1–blocking antibody was necessary to observe such responses. We speculate that this was due to the upregulation of PD-1 expression on the T cells during the in vitro expansion (Supplementary Fig. S6B). Hence the blockade of PD-1 was required to uncover mut-CALR–specific responses. Next, we investigated the changes in clonal T-cell populations in these in vitro cultures in response to stimulation with mut-CALR OLPs. Peripheral blood T cells, collected after the initial treatment (T1), were stimulated and expanded in vitro with WT or mutant CALR peptides, in the absence or presence of PD-1–blocking antibodies. Subsequently, TCR Vβ chains of the expanded T cells were sequenced. Multiple T-cell clones uniquely expanded in response to stimulation with mut-CALR OLPs in the presence of PD-1–blocking antibodies (Supplementary Fig. S7A–S7D). Among these, 2 clones that also significantly expanded in peripheral blood after pembrolizumab treatment were identified (Fig. 3I). Interestingly, the frequency of 1 of these 2 clones reached above 30% in all of the in vitro groups. The cells utilized in these in vitro cultures were collected after pembrolizumab administration, and clonal expansion of T cells was evident at this point compared with baseline. We speculate that this specific clone of T cells was invigorated in vivo after pembrolizumab treatment and had a proliferative advantage, thereby populating the in vitro cultures. Although these cells might recognize mut-CALR epitopes, it remains possible that these are bystander T cells recognizing epitopes unrelated to mut-CALR (30). Together, these observations, although limited to one patient, indicate that PD-1 blockade in patients with CALR⁺ MPN may lead to expansions of clonal T-cell populations that could potentially recognize neoantigens derived from mut-CALR.

Our data suggest that systemic T-cell exhaustion may limit the ability to observe mut-CALR–specific T-cell responses in patients with MPN. Schumacher and colleagues have previously demonstrated that neoantigen-specific T cells can be primed from T cells in HD blood (31). Similarly, we hypothesized that mut-CALR–specific T cells could be readily primed from HD blood. To test this hypothesis, PBMCs from 16 HDs were stimulated with OLPs derived from mut-CALR, and the responses were measured by ELISPOT and ICS. Responses to mut-CALR were observed in both CD4⁺ and CD8⁺ T cells and were characterized by the production of effector cytokines, such as IFNγ, as well as an increase in the surface expression of the activation marker CD137 (4-1BB; Fig. 4A–E). Importantly, the observed T-cell responses were specific to mut-CALR, as responses to stimulation with WT CALR OLPs were not higher than the background (Fig. 4A–C). Consistent with the hypothesis that T-cell exhaustion limits mut-CALR–specific T-cell responses in patients with MPN, we observed mut-CALR–specific T-cell responses in a greater fraction of HD PBMCs than MPN PBMCs [P = 0.001 for HD (Fig. 4B) and P = 0.037 for MPN (Fig. 1C)]. In addition, T-cell responses observed in HD PBMCs were of a higher magnitude than those observed in MPN PBMCs.

Healthy blood donors presumably have had no prior exposure to mut-CALR. Therefore, any observed mut-CALR–specific T-cell response should result from effective priming of naïve mut-CALR–specific T-cell precursors. To test this hypothesis, naïve (CD3⁺CD45RO⁻CD45RA⁺CCR7⁺) and memory (CD3⁺CD45RO⁺) T cells and non–T-cell (CD3⁻) antigen-presenting cells (APC) were sorted from PBMCs of reactive HD. APCs, pulsed with mut-CALR OLPs, were then separately cocultured with either naïve or memory T cells. Consistent with the hypothesis that mut-CALR–induced responses were derived from de novo T-cell priming, we observed mut-CALR–specific T cells in cocultures with sorted naïve T cells, but not with memory T cells (Fig. 4F). To confirm these findings, umbilical cord blood mononuclear cells, which are unlikely to have memory T cells (32), were stimulated with mut-CALR OLPs. The induction of mut-CALR–specific T-cell responses was also evident in these cultures (Supplementary Fig. S8). These data confirm the immunogenicity of the mut-CALR neoantigen and demonstrate the feasibility of priming antigen-specific T-cell responses from naïve T cells.

The mut-CALR C-terminus is >36 aa long (16) and may give rise to multiple immunogenic epitopes. In addition, our tested cohorts of patients with MPN and HDs will exhibit diverse human leukocyte antigen (HLA) haplotypes. An individual's HLA alleles will determine whether they are able to present mut-CALR epitopes, which will depend upon the relative binding affinities. To this end, we interrogated the binding affinities of epitopes derived from the altered C-terminus using in silico peptide binding prediction algorithms through the immune epitope database (IEDB). For maximal population coverage, we used IEDB's HLA reference set that covers >97% and >99% of the population for class I and class II alleles, respectively. Our analysis showed that multiple epitopes could theoretically bind to several HLA class I and II alleles (Supplementary Fig. S9A and S9B; Supplementary Tables S2 and S3). Furthermore, HLA-I alleles with predicted high binding affinities for mut-CALR–derived epitopes (Supplementary Table S2) were represented in approximately 60% of the population of the United States, according to recorded allele frequencies of 2.9 million typed donors covering 16 races (33, 34). Hence, a significant proportion of patients with MPN with the CALR frameshift mutation may have the potential to develop antigen-specific T cells.

Next, we aimed to identify which, if any, of the predicted HLA-I/II-restricted epitopes could induce T-cell responses. We selected 3 HDs that displayed both CD8⁺ and CD4⁺ T-cell effector responses upon stimulation with the mut-CALR OLP pool. We deconvoluted the mut-CALR OLP pool by priming HD precursor T cells with individual OLPs within the pool (Supplementary Fig. S1A). We observed CD8⁺ and CD4⁺ T-cell responses against multiple individual OLPs (Fig. 5). We also performed sequence-based HLA-I/II genotyping (SBT) for these subjects and determined the predicted HLA-I/II binding affinities of the epitopes that induced IFNγ production by CD8⁺ or CD4⁺ T cells using the IEDB's recommended algorithm. Each stimulating OLP yielded epitopes that were predicted to bind strongly to the HLA alleles of the donors tested (Fig. 5), suggesting that the predicted epitope may have been responsible for the observed mut-CALR–specific T-cell response. In addition, we evaluated whether the mut-CALR epitopes predicted to be strong binders to donor HLA (percentile rank < 2) were enriched within the OLPs that induced CD8⁺ T-cell responses. Although such enrichment was observed for each donor tested, the correlation was not significant as there were a few OLPs that did not induce T-cell responses, yet were enriched for predicted epitopes (Supplementary Fig. S9C). This observation is expected because many studies have reported that the majority of predicted epitopes fail to elicit T-cell responses (35) and that high HLA-peptide binding affinity does not equate to immunogenicity. In addition, when we perform T-cell immunogenicity assays, we start with a limited number of PBMCs, typically around 1 million. Therefore, it is possible that we may not capture all reactive T cells, especially the ones that are low in precursor frequency.

Because mut-CALR OLPs induced CD8⁺ T-cell responses, we aimed to identify the minimal epitopes that were recognized by CD8⁺ T cells. To perform an unbiased screening of all potential epitopes, a complete library of peptides with 9-aa length was synthesized. The 9mers overlapped with an offset of 1 aa and spanned the last 46 aa of the C-terminus of mut-CALR, formed by the 52-bp deletion. PBMCs from 3 additional HDs were stimulated with pools of these overlapping 9mers (6–7 peptides/pool) to screen for CD8 responses. CD8 responses were detected in one of the donors tested (Fig. 6A). As expected, short peptides did not induce CD4 responses (Fig. 6B). Next, we aimed to identify which epitopes within the stimulating pool (pool 1) were inducing the response. First, we investigated the predicted binding affinities of each epitope to the donor's HLA alleles. Out of the 6 epitopes constituting the pool, only “RMMRTKMRM” was predicted to have a percentile rank <2 by NetMHCpan 3.0 (Fig. 6C; Supplementary Fig. S10A). We then restimulated cells expanded with pool 1 with individual epitopes constituting the pool. In accordance with the predictions, only one epitope, “RMMRTKMRM”, was recognized by CD8⁺ T cells (Fig. 6D and E). To further confirm the immunogenicity of this epitope, we predicted the binding affinities of “RMMRTKMRM” and other related epitopes (Supplementary Fig. S10B) to common HLA class I alleles (Supplementary Fig. S10C). Next, we identified HDs with at least one HLA allele that was predicted to bind to the epitopes tested (Supplementary Fig. S10D). As an example, we observed CD8⁺ T-cell responses in donor 5 (Fig. 6F and G), further demonstrating the immunogenicity of these epitopes. Notably, this donor carried 3 of the predicted alleles (Supplementary Fig. S10D). CD8⁺ T cells from donor 5 produced effector cytokines both when stimulated with minimal epitopes and with a long peptide containing these epitopes (Fig. 6G). These data suggest that the identified epitopes can be processed and loaded onto MHC molecules within the cells. To further test this hypothesis, we expanded T cells from donor 5 by stimulating with the minimal epitope p10.4, and cocultured the expanded T cells with autologous APCs that were pulsed with a long peptide containing p10.4 sequence. We observed that long peptide–loaded APCs could induce p10.4-specific CD8⁺ T cells (Fig. 6H), suggesting this epitope is endogenously processed and presented to T cells by APCs.

Together, these data show that mut-CALR can induce T-cell responses, even in HDs without prior exposure to mut-CALR. Moreover, it provides evidence that healthy donor T cells can be utilized as an alternative approach to evaluate the breadth and specificity of mut-CALR–specific T-cell responses and their HLA restriction.

In this study, we demonstrate that mut-CALR is a shared MPN-specific neoantigen that can induce T-cell responses. Our results reveal that mut-CALR elicits specific T-cell responses in patients with CALR⁺ MPN, delineate the immunogenic properties of mut-CALR and identify a role for immune checkpoint blockade in enhancing mut-CALR–specific T-cell immunity in patients with CALR⁺ MPN.

Recent studies from Holmstörm and colleagues and Tubb and colleagues also evaluated mut-CALR–specific T-cell immunity (36–38). Holmstörm and colleagues reported that CALR mutations could be recognized in vitro by T cells from patients with CALR⁺ MPN. However, the responses were restricted to CD4⁺ T cells (36, 37). Tubb and colleagues focused on identifying mut-CALR epitopes recognized by CD8⁺ T cells based on predicted epitope-MHC information. However, mut-CALR reactive CD8⁺ T cells were elicited only in healthy donors but not in patients with MPN (38). In either case, the majority of predicted peptides failed to elicit T-cell responses and the observed mut-CALR–specific T-cell responses were weak. In the current study, we systematically evaluated the mut-CALR C-terminus for peptides binding to MHC molecules by using in silico MHC-I and MHC-II peptide binding prediction algorithms. We identified 103 peptides predicted to bind to representative MHC-I and MHC-II molecules with high affinity (Supplementary Tables S1 and S2). To account for MHC-restricted peptides that may have been missed or otherwise incorrectly assigned by in silico predictors (35), we utilized an unbiased screening approach based on overlapping peptides spanning the full length of the shared mut-CALR C-terminus. Screening was performed in parallel in patients with MPN and healthy donors to control for factors associated with disease burden that may inhibit mut-CALR–specific T cells. We demonstrated that several MHC-I and MHC-II restricted peptides from mut-CALR were recognized by CD8⁺ and CD4⁺ T cells.

Significantly, mut-CALR–specific T-cell responses were lower in magnitude and less prevalent in patients with MPN when compared with healthy donors. Several putative explanations for these observations were explored. We stratified the patients in our cohort according to their type of MPN and explored associations with the mut-CALR–specific T-cell responses. We observed that the majority of patients with mut-CALR–specific T-cell responses had ET. Although mut-CALR–specific T-cell responses were occasionally observed in patients with MF who had a prior history of ET, none of the patients with primary MF had detectable T-cell responses. ET can evolve to MF, which is associated with greater symptom burden, inferior prognosis, and higher number of somatic mutations (39). Consistent with Holmstörm and colleagues’ report (37), these observations suggest that mut-CALR–specific T-cell responses may occur more frequently in patients with CALR⁺ MPN with low symptom burden. In addition, some of the patients with MF in our cohort were being treated with ruxolitinib at the time their samples were used in this study. Ruxolitinib has been previously shown to inhibit T-cell responses (40). Therefore, it is possible that ruxolitinib treatment impaired the reactivity of mut-CALR–specific T cells in some patients in our cohort, a possibility being examined in separate experiments.

Systemic T-cell dysfunction is frequently observed in patients with cancer. The severity of T-cell exhaustion is associated with disease progression due to factors such as prolonged exposure to tumor antigens, which results in elevated expression of checkpoint receptors that directly inhibit T-cell function (41). On the basis of the observed association with disease burden, we hypothesized that systemic T-cell exhaustion may limit the ability to observe spontaneous mut-CALR–specific T-cell responses in some patients with CALR⁺ MPN. In support of this hypothesis, we observed that T cells from patients with CALR⁺ MPN exhibited elevated expression of immune checkpoint receptors PD-1 and CTLA4, which have been associated with T-cell exhaustion. Second, we found that, in patients with MPN, antibody-mediated blockade of PD-1 or CTLA4 signaling ex vivo rescued mut-CALR–specific T-cell responses. Third, we observed that mut-CALR reactive T cells can be produced from naïve T cells from healthy donors in greater frequencies, again suggesting a role for suppression of mut-CALR–specific T cells in patients with MPN. Finally, administration of pembrolizumab was shown to rescue mut-CALR–specific T-cell responses in a patient with CALR⁺ MPN. Collectively, these data show that spontaneous mut-CALR–specific T-cell responses that occur in patients with CALR⁺ MPN may be actively suppressed by immune checkpoint receptor signaling and that blockade of said inhibitors augments the mut-CALR–specific T-cell responses in vitro and in vivo.

However, mut-CALR–specific T-cell responses were restored upon blockade of PD-1 and CTLA4 signaling only in a subset of patients. Several mechanisms might underlie these observations. The patients in our MPN cohort were not selected based on their HLA type and exhibit diverse HLA alleles. Hence, the lack of mut-CALR–specific T-cell responses could be the absence of the HLA alleles binding to mut-CALR epitopes with high affinity. Also, checkpoint receptors other than PD-1 and CTLA4 might contribute to suppression of mut-CALR–specific T-cell responses. Phenotypic analysis showed that several checkpoint receptors, including LAG3 and TIM4, were also highly expressed in peripheral blood T cells of patients with CALR⁺ MPN. Therefore, inhibition of such molecules could be beneficial in restoring mut-CALR–specific T-cell responses. Chronic antigen exposure can lead to terminal exhaustion with limited reinvigoration potential in later stages of disease (41); thus, earlier intervention would be a more favorable approach.

A key biomarker of response to PD-1 blockade is tumor mutational burden (TMB; refs. 3, 7). TMB for MPN has previously been reported to be significantly low, 1 to 32 mutations per patient (12). Hence, it can be argued that the activity of anti–PD-1 treatment in patients with MPN would be low. However, even tumors with low TMB have been shown to provide high-quality neoantigens that elicit antitumor T-cell responses (42). Recent work by Cristescu and colleagues corroborate such observations by demonstrating that factors, such as a T cell–inflamed microenvironment, can independently predict response to PD-1 blockade even when the TMB is low (43). We observed that pembrolizumab treatment led to an overall increase in the proportion of peripheral blood T cells and significant expansion of several T-cell clones. It is possible that some of the expanded clones target high-quality neoantigens, mut-CALR and potentially others. Although limited to one patient, these observations suggest that in vivo PD-1 blockade in patients with CALR⁺ MPN may reinvigorate neoantigen-specific T cells.

Although we demonstrated that mut-CALR epitopes could be recognized by both CD8⁺ and CD4⁺ T cells from healthy donors, patient-derived mut-CALR–specific T cells were primarily CD4⁺. We hypothesized that the lack of CD8⁺ mut-CALR–specific T cells in patients with CALR⁺ MPN was the result of inefficient antigen presentation by MHC class I. CALR is a chaperone protein critical for the folding and assembly of MHC class I molecules (20). CALR-deficient cells are reported to have reduced levels of cell-surface MHC class I expression as well as reduced efficiency in antigen presentation to cytotoxic T lymphocytes (19). Even a heterozygous expression of the mut-CALR protein, as is the case with majority of patients with CALR⁺ MPN, has been shown to lead to a significant, albeit small, reduction in surface expression of MHC class I molecules (44). However, we did not observe a reduction in the cell-surface expression of MHC class I molecules in either PBMCs or CD34⁺ cells from patients with CALR⁺ MPN as compared with HDs. Similarly, MHC class II expression was also intact in patients with CALR⁺ MPN. Hence the paucity of anti–mut-CALR responses in patients with CALR⁺ MPN cannot be explained by a reduction of MHC expression. Yet it remains possible that the presentation of mut-CALR on MHC class I may be compromised, which warrants further investigation to evaluate the impact of mut-CALR on peptide loading and presentation.

Among the current standard treatment options for patients with MPN, only hematopoietic stem cell transplantation (HSCT) is potentially curative (17). However, the application of HSCT is limited due to lack of appropriate donor options, the advanced age of patients, comorbidities, and poor functional status (45). Other treatment modalities, although they improve symptom burden, show only minimal effects in eliminating malignant clones and providing molecular remissions (21, 46, 47). Therefore, it is essential to identify new approaches that advance the treatment of patients with MPN. We anticipate that our findings will form the basis for three avenues of therapy. A neoantigen vaccine targeting the C-terminal of mut-CALR can be administered, perhaps as a prevention of progression strategy, given that the disease is long-standing in nature and affects longevity considerably later. Studies by several groups demonstrated that administration of neoantigen vaccines could induce neoantigen-specific T-cell priming and also boost spontaneous neoantigen-specific T-cell immunity in patients with melanoma (48–50). Alternatively, the vaccine may be given in combination with checkpoint inhibitors, as is being done in several clinical trials for solid tumors (NCT02897765). Finally, an adoptive T-cell therapy for the treatment of patients with MPN harboring mutated CALR may be feasible, in which the patients will receive autologous reinvigorated mut-CALR–specific T cells that are expanded ex vivo. Sequencing and cloning of patients’ TCRs could also be used to transduce healthy donor HLA-matched T cells for an alternative form of adoptive cell therapy, an approach we are pursuing. These immunotherapy regimens could potentially establish an effective mut-CALR–specific T-cell immunity, which would target and eliminate CALR⁺ malignant cells, thereby leading to improved clinical outcomes in this patient population. Because CALR mutations are the second most common MPN driver mutation and mut-CALR neopeptide is shared among patients carrying each type of CALR mutation (11), strategies targeting mut-CALR are anticipated to be applicable to a significant number of patients with MPN.

Detailed materials and methods are provided in the Supplementary Material.

The use of patient-derived specimens was approved by the Institutional Review Boards at Mount Sinai (HS#11-02054) and all patients provided written informed consent before the initiation of any study procedures. The specimens were provided by the Hematological Malignancies Tissue Bank (HMTB) of the Tisch Cancer Institute (New York, NY). Patient blood was collected by the clinical personnel and mononuclear cells (MNC) were isolated by HMTB personnel by Ficoll–Paque density gradient. All patients analyzed in this study were diagnosed with MPN and tested positive for a CALR or JAK^V617F mutation. Because of low viability of patient cells after thawing, only freshly isolated patient PBMCs were used in immunogenicity assays, unless noted otherwise. Therefore, assays were performed once for each patient. HD specimens were procured from the New York Blood Center as leukopak, and MNCs were isolated by density gradient centrifugation using Ficoll–Paque Plus (GE Healthcare). PBMCs were cryopreserved in human serum containing 10% DMSO. HD PBMCs were used after thawing and assays were repeated at least twice.

Peptide predictions were performed using algorithms available at the immune epitope database (IEDB) and analysis resource (www.iedb.org). Unless indicated otherwise, the IEDB recommended method was selected to perform prediction analyses for both MHC-I and MHC-II binding. Epitopes that were assigned an IC₅₀ value lower than 500 and/or percentile rank lower than 2 were considered high-affinity binders.

Custom peptide libraries for WT and mut-CALR peptides were chemically synthesized by JPT Peptide Technologies. Each peptide had >80% purity as determined by high-performance liquid chromatography. MOG and CEFT peptide pools were commercially available at JPT Peptide Technologies. Each peptide was resuspended in DMSO and used at a final concentration of 1 μg/mL.

For the induction of antigen-specific T cells, a previously published method (18) was used with modifications. Briefly, unfractionated PBMCs were cultured in X-VIVO15 media (LONZA) with cytokines and were stimulated with peptide(s) (1 μg/mL) or equal volume of DMSO in the presence of adjuvants. Cells were expanded with IL2 (R&D Systems, 10 IU/mL) and IL7 (R&D Systems, 10 ng/mL) in RPMI media containing 10% human serum (Gibco; R10 media). Cells were fed every 2 to 3 days. After 9 to 11 days of culture, cells were harvested and all wells were pooled within group. After washing with R10 media, cells were restimulated with either MOG to account for background signal or with the test peptide(s) they were initially stimulated with (1 μg/mL) in the presence of anti-CD28 (1 μg/mL) and anti-CD49d (1 μg/mL) antibodies (BD Biosciences). As controls, some cells were stimulated with PMA (Sigma-Aldrich, 50 ng/mL) and ionomycin (Sigma-Aldrich, 1 μg/mL).

The same immunogenicity assay was utilized to induce antigen-specific T-cell responses when checkpoint receptors were inhibited with the following modifications: On day 2, when PBMCs were stimulated with peptides and adjuvants, anti–PD-1 (clone EH12.1) or anti-CTLA4 (clone BNI3) antibodies (no azide/low endotoxin mouse anti-human antibodies by BD Biosciences, both used at 10 μg/mL) or the same concentration of isotype controls, mouse IgG1, κ or IgG2a, κ, respectively, were also added.

For ELISPOT analysis, plates with mixed cellular ester membrane (Millipore) were coated with anti-IFNγ antibody (clone 1-D1k by Mabtech, used at 4 μg/mL) overnight at 4°C. Plates were washed 3 times with plain RPMI and blocked by incubating with R10 media at 37°C for at least 1 hour prior to addition of cells expanded in immunogenicity assay. Cells were seeded at either 5 × 10⁴ or 10⁵ per well in duplicates and restimulated as detailed above for 48 hours, and plates were processed for IFNγ detection. Plates were first incubated with biotinylated anti-IFNγ antibody (clone 7-B6-1 by Mabtech, used at 0.2 μg/mL) for 2 hours at 37°C, then 1 hour at room temperature with streptavidin-AP conjugate (Roche, used at 0.75 U/mL) and lastly with the SigmaFast BCIP/NBT substrate for 10 minutes at room temperature. Plates were washed 6 times with PBS containing 0.05% Tween-20 and 3 times with water in between each step. Plates were scanned and analyzed by ImmunoSpot software.

For flow cytometry, cells expanded in the immunogenicity assay were seeded at 1–2 × 10⁵ per well. One hour after restimulation, BD GolgiStop, containing monensin, and BD GolgiPlug, containing brefeldin A, were added to cells according to the manufacturer's suggestion. Twelve hours after the addition of protein transport inhibitors, cells were processed for flow cytometry. Cells were first stained for surface molecules with the following antibodies: anti-CD3 (OKT3), anti-CD4 (RPA-T4), anti-CD8 (RPA-T8), and anti-CD137 (4B4-1) and then processed for intracellular staining using BD Cytofix/Cytoperm reagents according to the manufacturer's protocol. Cells were then stained with anti-IFNγ (B27) and anti-TNFα (MAb11) antibodies. All antibodies were purchased from BioLegend.

Longitudinal blood samples were available from a patient with CALR⁺ MF (PT#15 in Supplementary Table S1) undergoing pembrolizumab treatment. PBMCs from this patient were isolated before and after 2 and 6 cycles of pembrolizumab treatment. PBMCs isolated after 2 cycles were expanded in vitro in an immunogenicity assay as described above by stimulating with WT or mut-CALR OLP pools, with or without PD-1 blocking mAbs. After expansion, cells were harvested. Genomic DNA was isolated from these 6 samples using DNeasy Blood and Tissue Kit (Qiagen) according to the manufacturer's instructions. The quality and quantity of the extracted DNA samples were analyzed by NanoDrop. DNA samples were sent to Adaptive Biotechnologies for TCR Vβ sequencing. Briefly, TCRβ CDR3 regions for each sample were amplified by a multiplexed PCR method using a mix of forward and reverse primers specific to TCR Vβ and TCR Jβ, respectively. Amplified regions were sequenced using the Illumina HiSeq System. Data analyses were performed using the Adaptive Biotechnologies ImmunoSeq Analyzer 3.0. Clonality was calculated as (1 − normalized entropy). Normalized entropy was defined as (Shannon entropy/log₂R), where R = the total number of rearrangements. The frequency of T cells was calculated as (number of cells expressing TCR/number of total nucleated cells). Total nucleated cell numbers were obtained by converting the input DNA amount in the assay based on the assumption that each diploid cell has about 6.4 pg genomic DNA. In differential abundance analyses of TCRβ chain sequencing, only clones with a cumulative abundance of 10 or above were included. P values were calculated using the binomial method, and FDRs were controlled by the Benjamini–Hochberg method. P values ≤0.01 were considered statistically significant. The frequencies of VJ combinations for abundant clones were visualized using circular plots. To this end, the most abundant 50 clones at each time point were selected and the combination of these clones were mapped out. Circular plots were generated using circos software package (http://circos.ca/). Each arch represents a V or J allele. A joining ribbon indicates a unique VJ cassette combination. The arc length and the width of the ribbons indicate the frequency of V/J alleles and their cassette combination respectively.

MARIMO cells were provided by N. Arshad and P. Cresswell (Yale School of Medicine, New Haven, CT). MARIMO cells were maintained at a density of 1–2 × 10⁶ cells/mL in RPMI media (Gibco) containing 10% heat-inactivated FBS (Gibco) and penicillin–streptomycin (Gibco, used at 100 U/mL, 100 μg/mL, respectively). Cells were regularly tested for Mycoplasma contamination by PCR. The latest testing was performed in November 2018 and cells were negative for Mycoplasma.

Statistical analyses performed were detailed in the figure legends. Briefly, statistical differences were assessed by Wilcoxon matched-pairs signed rank test or t test for comparisons between two groups.P values <0.05 were considered statistically significant. Statistical analyses were performed using GraphPad Prism software version 7 and 8. Differential abundance analyses of TCRβ chain sequencing data were performed using the Adaptive Biotechnologies ImmunoSeq Analyzer 3.0. P values were calculated using the binomial method and FDRs were controlled by the Benjamini–Hochberg method. P values ≤0.01 were considered statistically significant.

J. Mascarenhas receives research support from Incyte, Novartis, CTI Biopharma, Janssen, and Roche, and is on the scientific advisory board for Celgene, Roche, and Incyte. R. Hoffman receives research support from Formation Biologics, Merus, Novartis, Roche/Genentech, Scholar Rock, and Elstar, and is on the scientific advisory board for Novartis. N. Bhardwaj reports receiving commercial research grants from Novocure, Celldex, Genentech, Oncovir, and Regeneron and other commercial research support from the Ludwig Institute, Cancer Research Institute, Leukemia & Lymphoma Society, and NYSTEM, and is a consultant/advisory board member for Neon, Tempest, Checkpoint Sciences, Curevac, PrimeVax, Array Biopharma, Roche, Avidea, and the Parker Institute for Cancer Immunotherapy. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Cimen Bozkus, J.P. Finnigan, R. Hoffman, C. Iancu-Rubin, N. Bhardwaj

Development of methodology: C. Cimen Bozkus, J.P. Finnigan, R. Hoffman, C. Iancu-Rubin, N. Bhardwaj

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Cimen Bozkus, J. Mascarenhas, R. Hoffman, C. Iancu-Rubin, N. Bhardwaj

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Cimen Bozkus, V. Roudko, C. Iancu-Rubin, N. Bhardwaj

Writing, review, and/or revision of the manuscript: C. Cimen Bozkus, V. Roudko, J.P. Finnigan, J. Mascarenhas, R. Hoffman, C. Iancu-Rubin, N. Bhardwaj

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Cimen Bozkus, R. Hoffman, C. Iancu-Rubin

Study supervision: C. Iancu-Rubin, N. Bhardwaj

We thank N. Arshad and P. Cresswell (Yale School of Medicine) for providing the MARIMO cells and C. McClain (Tisch Cancer Institute at the Icahn School of Medicine at Mount Sinai) for performing Mycoplasma testing of the MARIMO cells. We also thank J. Arandela at HMTB (Tisch Cancer Institute at the Icahn School of Medicine at Mount Sinai) for assistance in specimen processing. Research funds were provided by the Myeloproliferative Neoplasms Research Foundation and Leukemia & Lymphoma Society to C. Iancu-Rubin and N. Bhardwaj and by the Parker Institute for Cancer Immunotherapy to N. Bhardwaj. Human specimens were provided by the HMTB of the Tisch Cancer Institute supported by the NCI P30 Cancer Center Support Grant P30 CA196521.

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.