Calreticulin mutant myeloproliferative neoplasms induce MHC-I skewing, which can be overcome by an optimized peptide cancer vaccine

Abstract

The majority of JAK2^V617F-negative myeloproliferative neoplasms (MPNs) have disease-initiating frameshift mutations in calreticulin (CALR), resulting in a common carboxyl-terminal mutant fragment (CALR^MUT), representing an attractive source of neoantigens for cancer vaccines. However, studies have shown that CALR^MUT-specific T cells are rare in patients with CALR^MUT MPN for unknown reasons. We examined class I major histocompatibility complex (MHC-I) allele frequencies in patients with CALR^MUT MPN from two independent cohorts. We observed that MHC-I alleles that present CALR^MUT neoepitopes with high affinity are underrepresented in patients with CALR^MUT MPN. We speculated that this was due to an increased chance of immune-mediated tumor rejection by individuals expressing one of these MHC-I alleles such that the disease never clinically manifested. As a consequence of this MHC-I allele restriction, we reasoned that patients with CALR^MUT MPN would not efficiently respond to a CALR^MUT fragment cancer vaccine but would when immunized with a modified CALR^MUT heteroclitic peptide vaccine approach. We found that heteroclitic CALR^MUT peptides specifically designed for the MHC-I alleles of patients with CALR^MUT MPN efficiently elicited a CALR^MUT cross-reactive CD8⁺ T cell response in human peripheral blood samples but not to the matched weakly immunogenic CALR^MUT native peptides. We corroborated this effect in vivo in mice and observed that C57BL/6J mice can mount a CD8⁺ T cell response to the CALR^MUT fragment upon immunization with a CALR^MUT heteroclitic, but not native, peptide. Together, our data emphasize the therapeutic potential of heteroclitic peptide–based cancer vaccines in patients with CALR^MUT MPN.

INTRODUCTION

Philadelphia chromosome-negative myeloproliferative neoplasms (MPNs) are myeloid blood cancers arising from hematopoietic stem cells (1, 2) and are characterized by hyperactivated Janus kinase–signal transducer and activator of transcription (JAK-STAT) signaling (3). The majority of JAK2^V617F-negative MPN tumors have an insertion or deletion (INDEL) mutation in the C-terminal region of calreticulin (CALR), creating a +1–base pair (bp) frameshift (4, 5). Whereas multiple unique INDELs are found, nearly all generate a 44–amino acid common peptide. A few rare cases generate a shorter 36–amino acid fragment (4, 5). Mutant CALR (CALR^MUT) develops a pathogenic binding interaction with the extracellular portion of the thrombopoietin receptor (myeloproliferative leukemia protein; MPL), inducing ligand-independent constitutive JAK-STAT signaling pathway activation and oncogenesis (6–8). Consequently, the oncogenic CALR^MUT fragment is an attractive source of frameshift-derived neoantigens for cancer vaccines in patients with CALR^MUT MPN (9). However, in the few studies examining CALR^MUT fragment immunogenicity, T cells from patients with CALR^MUT MPN had less immunoreactivity to CALR^MUT-derived peptides compared with healthy individuals (10–14), even though many immunogenic peptides are predicted (14, 15). On the other hand, T cells from healthy donors display a stronger and more frequent response to CALR^MUT peptides compared with T cells from patients with CALR^MUT MPN (13). In addition, several of the healthy donor T cell responses were elicited by memory T cells (13). This indicates that CALR^MUT peptides are immunogenic in normal donors and suggest that CALR^MUT-specific immune responses may be a mechanism of immunosurveillance eliminating the early tumor before its clinical manifestation. However, it is not clear at which extent tumors could escape this control in patients with clinical disease and why T cells from these patients did not respond to the CALR^MUT fragment.

For protein antigens to be recognized by CD8⁺ T cells, they must first be processed into smaller peptides, translocate to the endoplasmic reticulum (ER), and, if conditions are met, bind to the class I major histocompatibility complex (MHC-I) to form a peptide:MHC-I complex (pMHC-I) capable of reaching the cell surface and being recognized by the T cell receptor. However, not all peptides fit the stoichiometric requirements for MHC-I binding. Successful pMHC-I binding requires peptides to be the correct length and to have the appropriate anchor residues at specific locations to stabilize binding to pockets of the MHC-I molecules (16). MHC genes have evolved to be extremely polymorphic across individuals. Human cells encode two copies of three different MHC-I genes, human leukocyte antigen A (HLA-A), HLA-B, and HLA-C, with each having hundreds or thousands of polymorphisms (17). Different polymorphic residues alter the MHC-I pocket affinity for peptide anchor residues such that peptides that bind to MHC-I molecules encoded by some alleles may not bind to others (18, 19). As a result, the sum of presented peptides can vary greatly across individuals. We therefore hypothesized that the suboptimal immune reactivity to the CALR^MUT fragment in patients with CALR^MUT MPN compared to healthy individuals could be due to differential expression of MHC-I alleles able to efficiently present CALR^MUT-derived neopeptides. More specifically, individuals harboring MHC-I presenting CALR^MUT-derived peptides would be able to eliminate early CALR^MUT-positive MPNs, in contrast to those harboring other MHC-I alleles, who would be more likely to develop the disease.

Peptide binding affinities to MHC-I can range from strong to weak with many intermediate affinities possible. This affinity affects the number of surface-bound pMHC-I available for recognition by CD8⁺ T cells (20) and, ultimately, their activation. For productive stimulation, naïve T cells require interaction with a greater number of pMHC-I complexes than antigen-experienced memory T cells (21–26), and some antigens may be unable to activate naïve T cells. However, peptides can be synthetically optimized by altering one or two residues to increase MHC binding affinity. These resulting strong MHC binding peptides, called “heteroclitic” peptides (or anchor optimized or anchor improved), robustly activate naïve T cells and are similar enough to the original peptides that the activated T cells cross-react with the original antigen (27–30), assuming that the original antigen has an adequate intermediate binding affinity.

In this study, we find that six MHC-I alleles predicted to efficiently bind to multiple CALR^MUT-derived peptides are less frequently observed in patients with CALR^MUT MPN in two independent cohorts. This points to a higher risk of patients developing MPN with CALR^MUT when lacking these MHC-I alleles. It also suggests that individuals with these MHC-I alleles can potentially control primordial CALR^MUT-expressing tumors as part of the immunoediting process. In addition, this suggests that patients with CALR^MUT MPN may generate suboptimal responses to cancer vaccines composed of the CALR^MUT fragment. This supports the investigation of heteroclitic CALR^MUT peptide–based vaccines, eventually in combination with immune checkpoint blockade, as a new therapeutic approach for patients with CALR^MUT MPN.

RESULTS

Skewing of MHC-I allele frequencies in patients with CALR^MUT MPN is associated with peptide binding affinity

To test for MHC-I allele skewing in patients with CALR^MUT MPN, MHC-I allele haplotypes from two medical centers in the northeastern United States (NEUS) and from eight medical centers in Denmark (Table 1 and tables S1 and S2) were investigated. MHC-I allele frequencies were initially calculated by group, such as HLA-A*02, for under- or overrepresentation Boschloo’s test (fig. S1A). Representative MHC-I alleles from significant (P < 0.05) or trending (P < 0.25) MHC-I groups were then assessed bioinformatically for predicted binding to peptides derived from the CALR^MUT sequence (fig. S1B) using the NetMHCpan (v3) algorithm. We observed that MHC-I alleles that were less likely to be found in patients with CALR^MUT MPN had good predicted binding to CALR^MUT peptides, whereas overrepresented alleles had poor predicted binding.

Table 1. Distribution of patients with MPN according to mutation.

Total patient numbers (and relative frequency) associated with each cohort and center, and type of MPN, for which MHC-I haplotypes were collected.

Cohort/institution	CALRMUT (N = 54)	JAK2V617F (N = 254)	Total (N = 308)
Northeastern U.S. cohort	25 (19.2)	105 (80.8)	130
Memorial Sloan Kettering Cancer Center	11 (16.7)	55 (83.3)	66
Dana-Farber Harvard Cancer Center	14 (21.9)	50 (78.1)	64
Danish cohort	29 (16.3)	149 (83.7)	178

To refine our analysis, we examined MHC-I allele frequencies at the protein level, such as HLA-A*02:01 (tables S3 and S4). Furthermore, because MHC allele frequencies vary greatly by geographic location (31), we analyzed each cohort separately as well as together. Last, because the NEUS cohort is 88% Caucasian (table S5), we also compared MHC-I frequencies to those found in the U.S. Caucasian population from the National Marrow Donor Program (17). Unfortunately, MHC-I allele frequencies from the general Danish population were not available for comparison (31). To test for differences in MHC-I allele frequencies, we performed a principal components analysis (PCA) comparing frequencies from both NEUS MPN groups and the general U.S. Caucasian population. We observed that the JAK2^V617F MPN group clusters in proximity to the U.S. Caucasian group, whereas the CALR^MUT MPN group is isolated (Fig. 1A), again suggesting distinct MHC-I allele representation specifically in patients with CALR^MUT MPN. Unfortunately, because of the large number of possible alleles and the modest number of total patients from each cohort, statistical power was further reduced (tables S4 and S5). We therefore, opted to analyze alleles over a minimal frequency of 5% (in either MPN groups) with a fold change above or below a set threshold of ±0.2-fold frequency change in the NEUS cohort and ±0.125-fold frequency change in the Danish cohort (Fig. 1, B and C).

Fig. 1. MHC-I alleles with predicted binding to CALR^MUT-derived peptides are less frequent in CALR^MUT MPNs.

(A) Principal components analysis of MHC-I allele frequencies from patients with CALR^MUT MPN, patients with JAK2^V617F MPN, and U.S. Caucasian population. (B) Comparison of MHC-I allele frequencies from patients with CALR^MUT MPN, patients with JAK2^V617F MPN, and U.S. Caucasian population compared to each other in the NEUS cohort and (C) the Danish cohort. Frequencies are expressed as percentages. MHC-I alleles that are overrepresented in patients with CALR^MUT are in blue (NEUS cohort) and green (Danish cohort), whereas MHC-I alleles that are underrepresented in patients with CALR^MUT are in red (NEUS cohort) and pink (Danish cohort). For the NEUS cohort, only MHC-I alleles differentially expressed between patients with CALR^MUT MPN compared to both patients with JAK2^V617F MPN and U.S. Caucasian population were considered for further analysis. (D) Heatmap of predicted binding of each CALR^MUT-derived peptide to each MHC-I allele from (B) (left). MHC-I alleles that are underrepresented or overrepresented in patients with CALR^MUT MPN in both cohorts are noted with white or black circles, respectively. Actual MHC-I allele frequencies in CALR^MUT and JAK2^V617F MPN patient populations are also noted (right). (E) Cohort breakdown of CALR^MUT MPN MHC-I allele frequencies of individual institution consisting of the NEUS cohort for the six less frequent MHC-I alleles.

We observed that HLA-B*51:01 is the only allele overrepresented in patients with CALR^MUT MPN from both cohorts, whereas six MHC-I proteins are similarly underrepresented: HLA-A*11:01, HLA-B*08:01, HLA-B*44:02, HLA-C*07:01, HLA-C*07:02, and HLA-C*06:02. This trend is also reflected in the fraction of patients who are positive for these alleles (tables S6 and S7). Five of the six MHC-I alleles that met these requirements for underrepresentation had a moderate predicted affinity (<10,000 nM) to about a quarter of all 9-mer (Fig. 1D) or, to a lesser extent, 10-mer (fig. S2) peptides, of which many had <500 nM predicted affinity. On the other hand, all MHC-I alleles that were overrepresented in patients with CALR^MUT MPN had poor binding to almost all CALR^MUT-derived peptides with the exception of HLA-B*15:01 and HLA-C*12:03, which only had a high frequency in the Danish cohort (Fig. 1D). The HLA-B*07:02 allele, which has good predicted binding to CALR^MUT peptides, was underrepresented in both cohorts but only met the designated threshold for underrepresentation in the Danish cohort. When the NEUS cohort was separated into the two original cohorts [Memorial Sloan Kettering (MSK) and Dana-Farber Harvard Cancer Center], the same MHC-I skewing was confirmed except for HLA-C*06:02, which was slightly elevated in patients from MSK (Fig. 1E). Considering that five of the six underrepresented MHC-I alleles happened to have a good peptide binding suggests a trend worthy of further examination.

MHC-I skewing in CALR^MUT MPN is specific for CALR^MUT-derived peptides

Prediction algorithms for pMHC-I binding based on neural networks like NetMHC are generally accepted to be useful yet imperfect tools, and their biases are typically hard to capture. To control for the prediction algorithm, we hypothesized that MHC-I allele frequency bias should not be observed for proteins or protein fragments that are not under selective immune pressure. To test this, we scored the predicted binding affinity of each CALR^MUT-derived peptide or other irrelevant proteins in each individual patient with MPN to generate what we have termed the patient:peptide score (PPS). Briefly, the PPS of a peptide in a patient is equal to the binding score (nM) of that peptide against the MHC-I allele with the strongest predicted binding affinity of the six possible MHC-I alleles for that peptide (Fig. 2A). Furthermore, because statistically significant differences for any single MHC-I allele frequency was difficult to achieve, taking each patient’s six MHC-I proteins into consideration would offer a more robust statistical analysis of MHC-I protein skewing. As expected based on our initial findings, the average PPS of CALR^MUT-derived peptides was elevated in patients with CALR^MUT MPN from both cohorts compared to control groups, where a higher score is associated with worse predicted binding (Fig. 2, B and C). However, the average PPS of peptides derived from the wild-type (WT) portion of CALR was barely changed in the NEUS cohort and completely unchanged in the Danish cohort for the patients with CALR^MUT MPN compared to control groups (Fig. 2, B and C). Similarly, the capacity to bind peptides derived from the irrelevant foreign protein neuraminidase (NA) from the influenza virus did not substantially change when comparing patients with CALR^MUT or JAK2^V617F MPN and the general U.S. Caucasian populations (Fig. 2, B and C). Furthermore, when CALR^MUT-derived peptides were subdivided based on predicted binding (<10,000 nM) and nonbinding (>10,000 nM) scores, the greatest shift in average PPS occurred in predicted binding peptides (<10,000 nM), suggesting selection pressure against the capacity to efficiently bind those peptides in patients with CALR^MUT MPN (Fig. 2, B and C). A generous cutoff of 10,000 nM, as opposed to a much stricter 500 nM threshold, was used here to account for any bias in the prediction algorithm. Together, our findings suggest that patients with CALR^MUT MPN have a skewed MHC-I allele repertoire that is less immunologically responsive to the CALR^MUT protein fragment. The implication of this observation also suggests that these patients would be less likely to respond to cancer vaccines consisting of the CALR^MUT fragment.

Fig. 2. MHC-I bias is selective against the CALR^MUT fragment.

(A) Schema depicting calculation of patient:peptide score (PPS). Briefly, single peptides derived from longer sequences were applied to NetMHCpan v3 to predict class-I HLA binding against any given patient’s six possible class-I HLAs. In each patient-peptide combination, the strongest affinity was registered as the PPS. (B) Mean PPS of individual peptides derived from indicated protein sequence (9- to 10-mers) from patients with CALR^MUT and JAK2^V617F MPN and pseudopopulation created from expected frequencies of U.S. Caucasian population (NEUS only). Protein sequences are the 44–amino acid CALR^MUT sequence (CALR^MUT−44aa), the wild-type CALR sequence upstream of CALR^MUT−44aa (CALR^1–361), and the irrelevant foreign antigen neuraminidase (NA) from influenza. Also shown are peptides broadly subdivided from predicted possible binding (<10⁴ nM) and predicted nonbinding (>10⁴ nM) peptides. (C) Difference in mean PPS of each group from (B). The Student’s t test was performed to calculate significance. The horizontal bars represent the combination where the statistical test resulted in the noted P value.

Underrepresented MHC-I alleles potentiate response to CALR^MUT peptides

To test whether underrepresented MHC-I alleles were more likely to potentiate an immune response against the CALR^MUT fragment, peripheral blood mononuclear cells (PBMCs) from seven healthy donors positive for at least one of the underrepresented MHC-I alleles, and four healthy donors that were negative for these MHC-I alleles (table S8) were stimulated in vitro with peptides covering the entire CALR^MUT fragment and examined for reactivity by interferon-γ (IFN-γ) enzyme-linked immunospot (ELISpot) after a final round of peptide restimulation. In the donors positive for the underrepresented MHC-I alleles, seven of seven (100%) responded, whereas a response was observed in only one of four (25%) of donors negative for the underrepresented MHC-I alleles (Fig. 3A). Separately, to confirm MHC-I specificity, we generated CD8⁺ T cell cultures specific for the CALRLong36 peptide from a patient with HLA-B*07:02–positive CALR^MUT post–essential thrombocythemia primary myelofibrosis who was enrolled in the previously reported CALR^MUT vaccine trial (13). Reactivity was validated by IFN-γ ELISpot (Fig. 3B) and flow cytometry (Fig. 3C). After this, CALRLong36-primed– specific CD8⁺ T cells were analyzed in a standard Cr51 cytotoxicity assay with peptide-pulsed normal T2 cells or HLA-B*07:02– transduced T2 cells as target cells. In this context, only the peptide-pulsed T2-HLA-B*07:02 but not the peptide-pulsed T2 cell or the unpulsed T2-HLA-B*07:02 was killed (Fig. 3D), demonstrating that CD8⁺ T cells were restricted to HLA-B*07:02, one of the underrepresented alleles from the Danish cohort. Together, we show that patients with CALR^MUT-positive MPNs were less likely to have an MHC-I allele predicted to bind with high affinity to peptides derived from the CALR^MUT fragment. This may be due to the fact that individuals with MHC-I alleles that can efficiently present CALR^MUT-derived peptides are less likely to develop CALR^MUT MPN.

Fig. 3. Underrepresented MHC-I alleles potentiate response to CALR^MUT peptides.

(A) IFN-γ ELISpot of PBMCs from healthy donors that had either at least one (black circles) or zero (white circles) underrepresented MHC-I allele expanded with a peptide pool (15-mers) derived from the entire CALR^MUT fragment and restimulated after 10 days with either the irrelevant peptide (MOG) or the same CALR^MUT fragment peptide pool. (B) Reactivity of CALRLong36-specific CD8⁺ T cells derived from an HLA-B*07:02⁺ patient with post–essential thrombocythemia myelofibrosis. Cells restimulated with different CALR^MUT peptides in an IFN-γ ELISpot, (C) intracellular IFN-γ and TNFα cytokine staining on the CALRLong36-specific CD8⁺ T cells stimulated with CALRLong36 (D) target cell lysis of wild-type or HLA-B*07:02–transduced T2 cells pulsed with different CALR^MUT peptides.

Full-length CALR^MUT antigen is processed and presented by human primary myeloid cells

A necessary requirement for immune-mediated targeting of CALR^MUT-positive MPN tumor cells by CALR^MUT-specific CD8⁺ T cells is the endogenous processing and presentation of CALR^MUT peptide by MHC-I. Whereas many in the field presumed this to be happening, it has not yet been proven directly. To test whether endogenous processing and presentation of CALR^MUT peptide by MHC-I is occurring in human primary myeloid cells, we used two different strategies. In the first set of experiments, the aforementioned patient-derived CALRLong36 specific CD8⁺ T cells were enriched and cloned using a limiting dilution. The CD8⁺ T cell clones’ ability to recognize CALR^MUT target cells was assessed by IFN-γ ELISpot after restimulation with autologous CALR^MUT-positive CD14⁺ monocytes that were subjected to CALR-specific siRNA or control knockdown (fig. S3A). Reactivity to the CALR siRNA–transfected target cells was diminished in all clones. In a separate experiment, primary human myeloid cells from healthy donors were infected by adenovirus (AdV) expressing the full-length 44–amino acid antigen, and these cells were used to activate autologous PBMCs for 11 days in vitro, whereupon CALR^MUT-specific reactivity was assessed by secondary peptide restimulation followed by flow cytometry. In 7 out of 20 donors, there was a clear priming of CALR^MUT-specific CD8⁺ T cells (fig. S3, B and C). Furthermore, almost all of the donors that had reactivity had at least one of the previously discussed underrepresented MHC-I allele (fig. S3D). We therefore conclude that the CALR^MUT antigen can be processed and presented endogenously by human primary myeloid cells, further supporting the model that CALR^MUT MPN can be subject to immune-mediate eradication, either endogenously or vaccine mediated, presuming a favorable MHC-I allele haplotype.

Patients with CALR^MUT MPN exhibit skewed MHC-II allele frequencies

We also examined MHC-II allele frequencies for skewing using haplotype data from the NEUS cohort. Unfortunately, MHC-II haplotypes were unavailable for the Danish cohort. Similar to MHC-I alleles, MHC-II alleles also exhibited distinct frequencies by PCA (fig. S4A). No single allele had a statistically significant decrease in frequency in patients with CALR^MUT MPN (table S9), and the same fold change threshold approach as for HLA-I alleles was therefore applied. We observed that HLA-DRB1*03:01, HLA-DRB1*04:01, HLA-DRB1*07:01, HLA-DRB1*13:01, HLA-DQB1*02:01, and HLA-DQB1*06:03 were less frequent in patients with CALR^MUT MPN compared to patients with JAK2^V617F MPN and U.S. individuals of European descent, whereas HLA-DRB1*11:01, HLA-DRB1*11:04, and HLA-DQB1*03:01 were more frequent (fig. S4B). Both of the HLA-DRB1 MHC-II alleles found at higher frequency in patients with CALR^MUT MPN that could be assessed for predicted binding were predicted to bind strongly to more than half of 15-mer peptides, whereas only one of the four HLA-DRB1 MHC-II alleles found at lower frequency appears to do so (fig. S4C). When we examined the MHC-II alleles of the healthy donors from which PBMCs were used to generate memory CD4⁺ T cell lines against long CALR^MUT peptides in a previous study (13), both donors were positive for HLA-DRB1*13:01, and one of these was also positive for HLA-DRB1*04:01 (table S10).

Nonresponding human PBMCs can cross-react with CALR9p2 if first primed with heteroclitic peptides

We hypothesized that one approach to elicit an immune response against CALR^MUT MHC-I skewed patients is to use a vaccine consisting of MHC-I binding-optimized heteroclitic peptides to increase the initial activation of CALR^MUT-specific CD8⁺ T cells. To test this hypothesis, we examined the peptides with the lowest PPS in the patients with CALR^MUT MPN as possible candidates. We observed that the top peptide in both cohorts was the 9mer CALR^MUT peptide starting at position 2 (CALR9p2) RMMRTKMRM (fig. S5A), which is predominantly a function of its predicted binding to frequently observed MHC-I alleles such as HLA-A*02:01 and HLA-A*03:01 (fig. S5, B and C), which are not underrepresented in patients with CALR^MUT MPN (table S3). Therefore, based on the mean PPS of CALR9p2 in both cohorts, we reasoned that this peptide was less likely to directly activate naïve CD8⁺ T cells, but could be recognized and targeted by CD8⁺ T cells preprimed with peptides of similar specificities and improved/optimized MHC-I binding, such as heteroclitic peptides. We therefore investigated whether heteroclitic peptides could be used to induce cross-reactivity against CALR^MUT-derived peptides in human samples.

To identify the best candidate CALR9p2 heteroclitic peptide, we examined the mean predicted PPS score of the NEUS CALR^MUT MPN cohort to every possible CALR9p2 peptide variant containing a single amino acid substitution (Fig. 4A). Seven of the top heteroclitic CALR9p2 were selected for testing based on their predicted binding to HLA-A*02:01 (Fig. 4B). Five peptides were selected with a substitution at position 6 (K6), and two were selected with a substitution at position 1 (R1). Accordingly, alterations of the residues at positions 1 and 6 were predicted to affect HLA-A*02:01 binding through its minor anchor sites instead of the main anchor sites at positions 2 and 9, where the existing native residues have predicted weakly favorable binding conditions (fig. S6) (32, 33). A closer examination of binding predictions of each heteroclitic candidate to the top 10 most frequent MHC-I alleles in patients with CALR^MUT MPN revealed that the K6 heteroclitic peptides could target multiple MHC-I alleles with predicted binding of the CALR9p2 ranging from 500 to 5000 nM, whereas the R1 heteroclitic peptides mostly affected HLA-A*02:01 binding affinity (Fig. 4C). We thus tested the seven candidate heteroclitic CALR9p2 peptides for their ability to bind to and stabilize HLA-A*02:01 on TAP (transporter associated with antigen processing)2-deficient HLA-A*02:01–positive T2 cells and confirmed that they have greater binding capacity than native CALR9p2 peptide, which had weak binding signal compared with dimethyl sulfoxide (DMSO) control (fig. S7). However, all of the heteroclitic CALR9p2 peptides had weaker binding potential than the MART1-A2 peptide positive control.

Fig. 4. Human CD8⁺ T cells activated with heteroclitic CALR9p2 peptides cross-react with CALR9p2 peptide.

(A) Predicted mean PPS in patients with CALR^MUT MPN of all single amino acid substitution of CALR9p2. (B) Predicted affinity of all single amino acid substitution of CALR9p2 to HLA-A*02:01. Seven heteroclitic peptides were chosen for further testing and are identified here only by their amino acid substitution in the CALR9p2 peptide. (C) Binding affinity of CALR9p2 and indicated heteroclitic peptide to the most common MHC-I in the CALR^MUT MPN patient cohorts. Shadowed area indicates predicted binding affinity range of 5000 to 500 nM. (D) Percent IFN-γ⁺ CD8⁺ T cells after primary in vitro stimulation of PBMCs with CALR9p2 or CALR9p2 heteroclitic peptides (noted on x axis) followed by secondary restimulation with either control (MOG, median represented by dotted blue line) peptide, CALR9p2, or initial CALR9p2 heteroclitic peptide (noted by color of bar). (E) Summary of responding donor PBMCs to CALR9p2 or CALR9p2 heteroclitic peptides.

To test the ability of these peptides to elicit a T cell response, we stimulated PBMCs from six healthy HLA-A*02:01 individuals with known MHC-I haplotypes (table S11) and PPS (table S12) in the presence of native CALR9p2, using each heteroclitic peptide individually, as well as using a pool of all heteroclitic peptides. Cells were then exposed to a secondary restimulation with either control peptides, the same peptides used in the initial stimulation, or the native CALR9p2 peptide and tested for IFN-γ production (Fig. 4, D and E). As controls, some cells were left unstimulated (DMSO) or were stimulated with a positive control peptide pool of T cell epitopes from cytomegalovirus (CMV), Epstein-Barr virus (EBV), influenza, and Clostridium tetani (CEFT), or with phorbol myristate acetate (PMA) and ionomycin (fig. S8). We found that four of six donors responded to at least one heteroclitic peptide (Fig. 4, D and E). Three of the six healthy donors had significant (P < 0.05) or trending (P < 0.30) CD8⁺ T cell responses to CALR9p2 alone (donors 2, 3, and 4), and two of them also showed significant CD8⁺ T cell cross-reactivity against CALR9p2 if primed with a heteroclitic peptide (donors 3 and 4; P < 0.05). In two of the samples that did not respond to CALR9p2 alone (donors 1 and 6), there was significantly detectable CD8⁺ T cell cross-reactivity with CALR9p2 when the PBMCs were primed with heteroclitic peptides (P < 0.05). To test whether the heteroclitic peptides were promoting cross-reactivity through HLA-A*02:01, we again activated healthy donor HLA-A*02:01– positive PBMCs using pooled heteroclitic peptides, but the final restimulation was provided using the peptide-pulsed K562-HLA-A*02:01 cell line. In the four additional donors tested, two had a cross-reactive response against the CALR9p2 peptides, demonstrating that HLA-A*02:01 was at least one of the HLA alleles through which the heteroclitic peptides were providing a cross-reactive response (fig. S9). Together, our results suggest that cross-reactive immunity to CALR^MUT can be achieved in human cells, especially if multiple different heteroclitic peptides are used.

Modeling CALR^MUT MPN patient MHC-I allele skewing in mice

To determine whether a heteroclitic peptide cancer vaccine is a viable strategy against the CALR^MUT fragment, we tested this approach in a preclinical mouse model mimicking CALR^MUT MPN MHC-I allele skewing. We analyzed the predicted binding of all CALR^MUT-derived peptides against all murine MHC-I alleles for which predictions were possible. We found no strong binding peptide (<500 nM) to any murine MHC-I alleles (Fig. 5A) but did observe that H-2K^b had weak predicted binding affinity to CALR9p2. When tested for its ability to stabilize MHC-I in the H-2K^b–expressing TAP-deficient RMA/S cell line, CALR9p2 did not elicit detectable H-2K^b stabilization compared to the control strong binding chicken ovalbumin (OVA)–derived peptide SIINFEKL (Fig. 5B). However, when serum was omitted from the assay, H-2K^b was stabilized only with the highest concentration of peptide (100 μg/ml), suggesting poor but still detectable binding (Fig. 5B). To investigate whether the CALR^MUT protein fragment is immunogenic in vivo, we immunized H2-K^b–expressing C57BL/6J (B6) mice with a DNA vaccine (Fig. 5C) encoding the full-length 52-bp deletion variant of the CALR^MUT sequence (4–6) and analyzed CALR9p2 peptide– specific CD8⁺ T cells from draining lymph nodes. Compared to the immunogenic CD8⁺ T cell response against the SIINFEKL peptide in OVA-immunized mice, mice immunized with the fulllength CALR^MUT sequence did not elicit any CD8⁺ T cell response against CALR9p2 (Fig. 5D). Likewise, mice immunized in the footpad (Fig. 5E) with the CALR9p2 peptide emulsified in the TiterMax adjuvant did not elicit any CALR9p2-s pecific CD8⁺ T cell response compared to SIINFEKL-immunized mice either (Fig. 5F). Because we observed poor but detectable binding of CALR9p2 to H-2K^b but no vaccine-induced CALR9p2-specific CD8⁺ T cell responses in B6 mice, this mouse model constitutes a relevant preclinical setting resembling patients with CALR^MUT MPN with an MHC-I skewed haplotype in which to test the in vivo efficacy of CALR^MUT heteroclitic peptide-based vaccines.

Fig. 5. CALR^MUT sequence is not immunogenic in C57BL/6 J mice.

(A) The predicted binding affinity of CALR^MUT-derived peptides (8- to 10-mers shown) against all available murine MHC-I alleles. The strongest binding peptide CALR9p2 is identified in red. (B) H-2K^b stabilization assay using TAP-deficient RMA/S cells was performed for CALR9p2 in the presence and absence of serum. The chicken ovalbumin (OVA)–derived peptide SIINFEKL was used as a positive control. (C) Timeline of DNA immunization schedule and CD8⁺ T cell collection for the experiment in (D). (D) IFN-γ ELISpot depicting secondary reactivity of CD8⁺ T cells isolated from draining lymph nodes of mice DNA immunized with full-length CALR^WT, CALR^MUT, and OVA. (E) Timeline of peptide immunization and CD8⁺ T cell collection for the experiment in (F). (F) IFN-γ ELISpot depicting secondary reactivity of CD8⁺ T cells isolated from draining lymph nodes of mice peptide immunized with adjuvant and DMSO, CALR9p2, or SIINFEKL. Data shown represent results from one repeat of experiments performed at least three times.

Full-length CALR^MUT variant does not have dominant-negative activity

One possible issue with a CALR^MUT vaccine is the uncertainty as to whether the CALR^MUT fragment itself can inhibit antigen presentation. Wild-type CALR is required for antigen presentation in healthy cells (34, 35), and the full-length CALR^MUT is reported to be non-functional with respect to peptide loading (35). However, nearly all CALR^MUT-positive MPN tumors are heterozygous (4, 5) and therefore have one WT copy of CALR, yet it is unknown whether CALR^MUT acts as a dominant-negative with respect to its role in antigen presentation. To exclude this possibility, we cotransfected the murine B16F10 cells with the DNA sequences encoding OVA and either CALR^WT-mCherry, CALR^MUT-mCherry, or the mCherry constructs, and measured surface expression of H-2K^b–presented SIINFEKL peptide. We observed that cells transfected with the CALR^MUT variant had an equal percentage of H-2K^b-SIINFEKL peptide–expressing cells (fig. S10, A and B), suggesting that the CALR^MUT variant did not inhibit antigen presentation. We observed a marginal decrease in total surface MHC-I expression (fig. S10C), suggesting some effect of the CALR^MUT variant, although it is unclear whether such a small reduction in MHC-I would have any functional effect. As a result, we conclude that CALR^MUT does not prevent tumor cells from presenting antigens and suggest that CALR^MUT tumors could respond to cancer vaccines if the correct antigen is selected and is expressed by the tumor.

C57BL/6J-optimized heteroclitic CALR9p2 peptide elicits cross-reactive immunity in mice

To identify the best candidate CALR9p2 heteroclitic peptide in B6 mice, we examined the predicted binding affinity of every possible CALR9p2 peptide variant containing a single amino acid substitution to H-2K^b. We observed that the variant with the strongest predicted affinity had a threonine (T)–to–phenylalanine (F) substitution at position 5 (T5F) of the CALR9p2 peptide (Fig. 6A). This is consistent with previous studies showing that this site is an important anchor residue for H-2K^b (Fig. 6B and fig. S11) (19, 36, 37). Specifically, for 9-mers, in addition to the major anchor sites at positions 2 and 9 of H-2K^b, which bind to the nonoptimal but weakly favorable M2 and M9 residues of CALR9p2, the T residue is predicted to have a net negative binding effect at the other important position 5 anchor site, where the F residue instead favors stronger binding. When we investigated the ability of CALR9p2(T5F) to stabilize H-2K^b in RMA/S cells, this heteroclitic peptide demonstrated about a 10-fold greater H-2K^b stabilization compared to CALR9p2 (Fig. 6C). We then tested whether a CALR9p2(T5F) peptide vaccine could elicit a cross-reactive CALR9p2 immune response. Mice immunized with a single dose of CALR9p2(T5F) peptide vaccine elicited a CD8⁺ T cell response capable of cross-reacting to CALR9p2 in vitro (Fig. 6D) and killing tumor cells pulsed with the CALR9p2 peptide (Fig. 6E). Likewise, mice immunized with a DNA vaccine against the full-length CALR^MUT sequence encoding the CALR9p2(T5F) variant (fig. S12A) also developed a cross-reactive response against the native CALR9p2 peptide (fig. S12B). The ability of the full-length antigen to induce a specific response demonstrates again that the full-length CALR^MUT sequence can be endogenously processed and presented. As expected, immunizing mice with the heteroclitic CALR9p2(T5F) variant, by peptide or DNA vaccine, elicited an even stronger IFN-γ and cytotoxic T cell response against the same CALR9p2(T5F) heteroclitic peptide, which served as positive control in these experiments (Fig. 6, D and E, and fig. S12B). When we monitored CALR9p2(T5F)-specific CD8⁺ T cell responses in mice by tetramer staining upon vaccination with either the CALR^MUT DNA encoding the CALR9p2(T5F) heteroclitic peptide (fig. S12, C and D) or directly with the CALR9p2(T5F) heteroclitic peptide (fig. S13, A and B), we found that these cells were expanded only by the heteroclitic sequence, and not by WT CALR. Furthermore, CD8⁺ T cells from heteroclitic-immunized mice displayed higher surface expression of the activation markers CD44, TIM3 (T cell immunoglobulin domain and mucin domain 3), and PD-1 (programmed cell death protein 1) compared to tetramer-negative CD8⁺ T cells from the same animals (fig. S12, E and F). Only the CALR9p2(T5F) tetramer-positive CD8⁺ T cell fraction responded to the CALR9p2 native peptide upon in vitro secondary restimulation, as measured by up-regulating IFN-γ and tumor necrosis factor–α (TNFα) production (fig. S13, C and D), indicating that the same T cell clones could recognize both CALR9p2(T5F) and CALR9p2. In addition, a second restimulation with either CALR9p2(T5F) or CALR9p2 peptides of in vivo CALR9p2(T5F)-primed CD8⁺ T cells induced similar up-regulation of PD-1 and TIM3 in CALR9p2(T5F) tetramer-positive CD8⁺ T cells (fig. S13, E and F).

Fig. 6. Heteroclitic CALR9p2 peptide vaccine elicits cross-reactive CD8⁺ T cell response against CALR9p2 and controls tumor growth in mice.

(A) Predicted binding affinity to H-2K^b of all single amino acid substitution variants of CALR9p2. Top predicted peptide CALR9p2(T5F) is shown. (B) Cartoon of the expected effect of T5F substitution in CALR9p2 peptide conformational binding into H-2K^b. Known dominant anchor sites and minor anchor sites are depicted in red and blue, respectively. (C) MHC-I stabilization assay using TAP-deficient RMA/S cells was performed for CALR9p2 and CALR9p2(T5F) in the absence of serum. SIINFEKL was used as a positive control. Representative results from one repeat of an experiment performed at least three times. (D) IFN-γ ELISpot depicting secondary reactivity of CD8⁺ T cells isolated from draining lymph nodes of mice peptide immunized with adjuvant and DMSO, CALR9p2, or CALR9p2(T5F). Representative results from one repeat of an experiment performed at least three times. The Student’s t test was performed to calculate significance. (E) Killing assay of peptide-pulsed B16 cells by CD8⁺ T cells isolated from peptide-immunized mice. Representative results from one repeat of an experiment performed at least three times. The Student’s t test was performed to calculate statistical significance. (F) Timeline of peptide immunization prior to tumor implantation for prophylactic vaccine. (G) RMA/S^pER-CALR9p2 tumor growth over time after prophylactic peptide immunization for individual tumors or (H) average growth until the second mouse reached the end point of a tumor size of 400 mm². (I) Survival of mice after the prophylactic vaccine. (J) Timeline of peptide immunization and tumor implantation for therapeutic vaccine and in combination with anti–PD-1 therapy. ip, intraperitoneally. (K) Tumor growth over time after therapeutic vaccine for individual tumors or (L) averaged tumor growth until the second mouse reached the end point. (M) Survival of mice after the therapeutic vaccine, with or without anti–PD-1 therapy. Data from tumor growth experiments represent results from one repeat of experiments performed twice. Significance for tumor growth experiments was calculated by performing a Student’s t test on the area under the curve of each tumor. Significance for survival was calculated by performing a log-rank test.

We then tested the ability of these cross-reacting immune responses induced by CALR^MUT heteroclitic vaccines to induce antitumor effects in vivo. To generate a suitable tumor model for this, we used the newly developed PresentER antigen minigene system (38). Here, the nucleotide sequence of the CALR9p2 peptide is cloned downstream of an ER signal sequence (SS) and virally transduced into TAP-deficient RMA/S lymphoma cell line (RMA/S^pER-CALR9p2). Once expressed, the peptide-SS is shuttled into the ER and the peptide is cleaved from the SS, releasing CALR9p2 into the ER, where it can be loaded into MHC-I, assuming binding is possible. When mice were given three doses of the heteroclitic peptide vaccine prior to tumor implantation, RMA/S^pER-CALR9p2 tumors grew significantly slower than those injected into mice given adjuvant alone in the same vehicle (DMSO) or preimmunized with the native CALR9p2 peptide, which both grew at the same rate (P < 0.0001; Fig. 6, F and G), although differences in survival were not significant (Fig. 5H). However, when RMA/S^pER-CALR9p2 tumors were allowed to grow before mice received multiple therapeutic doses of the heteroclitic peptide vaccine (Fig. 6I), tumors grew significantly slower (P < 0.0001; Fig. 6, J and K) and mice had improved survival (Fig. 6L) compared to the adjuvant alone condition. The effect of the vaccine was even more prominent when the immunization was administered with PD-1 immune checkpoint blockading antibody (Fig. 6, I to K). The therapeutic vaccine had greater efficacy than the prophylactic vaccine. As this was unexpected, we hypothesized that CALR9p2-specific cross-reactive CD8⁺ T cells were diminishing in efficacy over time and that the available CALR9p2 antigen present in the tumor cells was not generating a strong memory response. To investigate this further, we immunized mice with three doses of the heteroclitic CALR9p2(T5F) peptide vaccine and compared cross-reactive potential in conditions where mice instead received CALR9p2 peptide boosts after an initial CALR9p2(T5F) priming dose. Consistent with the prophylactic vaccine results, mice that received an initial CALR9p2(T5F) followed by two CALR9p2 boosts had no detectable cross-reactive CD8⁺ T cell responses to CALR9p2 in vitro and a very small response to the CALR9p2(T5F) peptide (fig. S14). On the other hand, mice that received two doses of CALR9p2(T5F) and a final CALR9p2 boost had no detectable CALR9p2 cross-reactive CD8⁺ T cells but an intermediate response to CALR9p2(T5F). Therefore, it appears that cross-reactive immunity to the CALR9p2 peptide in the context of B6 mice is a function of time-from-last CALR9p2(T5F) dose. Together, this provides a proof of principle that a heteroclitic peptide cancer vaccine strategy in a murine model is a viable option for tumor cells expressing weakly binding tumor epitopes, such as those predicted to be presented in CALR^MUT MPN tumors.

DISCUSSION

It has long been postulated that the immune system shapes tumors by eliminating clones expressing immunogenic antigens in a process called immunoediting (39–41). We have also recently shown that there is a trade-off between tumorigenicity of a mutation and its immunogenicity (42). Clones that evade elimination would either enter a state of equilibrium with the immune system or manage to evade it completely, such that it manifests clinically. However, by its very nature, the elimination of primordial tumor clones has been difficult to visualize experimentally and in patients. Patients with MPNs with the CALR^MUT frameshift fragment offer a simple yet elegant example of the elimination phase in vivo. In this tumor type, the CALR^MUT frameshift is a driver truncal mutation and may constitute a neoantigen (15). Although this indicates that an effective immune response against CALR^MUT can be potentially generated, not everyone can develop immunity against CALR^MUT MPN. To unravel the molecular mechanisms explaining the lack of immunogenicity of CALR^MUT in patients developing MPN, we investigated potential differences in MHC-I and MHC-II allele frequencies in this patient population.

In this study, we observe that patients with CALR^MUT MPN from two independent cohorts are less likely to carry the MHC-I alleles HLA-A*11:01, HLA-B*08:01, HLA-B*44:02, HLA-C*06:02, HLA-C*07:01, and HLA-C*07:02. Of these six MHC-I alleles, all except HLA-B*44:02 are predicted to bind to multiple CALR^MUT-derived peptides. This contrasts with MHC-I alleles that are more frequent in patients with CALR^MUT MPN, which are predicted to not bind efficiently to CALR^MUT-derived peptides. We therefore posit that individuals with a negatively skewed MHC-I allele, which is predicted to bind to CALR^MUT, are more likely to undergo immune-mediated tumor rejection such that the tumors never manifest clinically.

In a study examining CALR^MUT immunoreactivity in healthy donors, HLA-C*07:01 was expressed in four of five healthy donors that responded to multiple CALR^MUT-derived peptides (14). In another study, the MHC-I allele HLA-B*07:02, which is underrepresented in patients with CALR^MUT MPN in the Danish but not the NEUS cohort, was found to elicit an immune response to multiple CALR^MUT-derived peptides (12). In another example, in a rare case where CD8⁺ T cells from a CALR^MUT MPN patient did respond to a long CALR^MUT peptide in vitro, this patient was identified to be positive for the HLA-C*06 allele (10, 11). Therefore, many of the MHC-I alleles that we find to be negatively skewed in patients with CALR^MUT MPN have been reported to be immunogenic against some CALR^MUT-derived peptides, which supports our main hypothesis.

Our findings suggest that oncogenic CALR mutations are more frequent than it is currently observed, because we hypothesize that individuals with advantageous MHC-I haplotypes can eliminate the disease before it manifests and develop specific immunological memory. This would imply that many healthy individuals can mount an immune response against the CALR^MUT fragment. Evidence for such MHC-I allele–specific immunity against recurring mutations in blood cancers has been observed elsewhere. Specifically, it was observed that MHC-I alleles that have shown to bind to the BCR-ABL fusion protein are less frequent in patients with BCR-ABL⁺ chronic myeloid leukemia (43), and likewise, MHC-I alleles predicted to bind to nucleophosmin 1 frameshift mutations are less frequently observed in patients with acute myeloid leukemia (44).

As we still observe patients with CALR^MUT MPN with the negatively skewed predicted immunogenic MHC-I alleles, it is clear that MHC-I haplotype alone is insufficient for the elimination of the disease. It is currently unclear why negatively skewed predicted immunogenic MHC-I–expressing patients with CALR^MUT MPN still develop the disease, but many explanations are plausible: (i) Many tumors evade immune-mediated rejection by inhibiting the antigen processing machinery itself. This includes the epigenetic down-regulation of MHC-I, or by acquiring mutations in TAP or β2M, which are frequently observed in tumors under immune pressure (45, 46). (ii) It is possible that the tumor establishes an immunosuppressive microenvironment such that priming of the immune system or the elimination of the tumor is inhibited. A recent study found that T cells from patients with CALR^MUT MPN had elevated surface expression of PD-1 and cytotoxic T lymphocyte-associated antigen-CTLA-4 compared to healthy donors, as well as increased numbers of myeloid-derived suppressor cells and regulatory T cells in the blood (14), all of which are associated with immunosuppression (47). It is possible that negatively skewed MHC-I–expressing patients with CALR^MUT MPN may have defects in the antigen presentation machinery or a higher than normal immunosuppressive phenotype preventing tumor control. The same study showed that CALR^MUT MPN PBMCs, which were unresponsive to multiple CALR^MUT-derived peptides in vitro, could elicit a response when a PD-L1 antibody was added to the coculture (14). Therefore, negatively skewed MHC-I–expressing patients with CALR^MUT MPN might be promising candidates for immune checkpoint blockade therapy as well as a heteroclitic peptide cancer vaccine, because it could favor immune-mediated recognition of CALR^MUT peptides correctly presented in proper MHC-I.

Our study had three main limitations. First, the limited number of patients with CALR^MUT MPN resulted in trending but nonsignificant statistical power when calculating whether any individuals MHC-I alleles were underrepresented. We tried to address this limitation by examining two different independent patient cohorts for similar patterns of MHC-I allele frequencies, as well as showing that patients with CALR^MUT MPN have significantly lower global MHC-I binding affinities to all CALR^MUT-derived peptides when all six MHC-I alleles are considered. Second, we were unable provide direct evidence that the full-length CALR^MUT antigen could be endogenously processed and presented in CALR^MUT MPN tumor cells in vivo. However, we showed that antigen-presenting cells (APCs) expressing the full-length CALR^MUT antigen can activate CD8⁺ T cells in an experimental setting, strongly suggesting that endogenous processing and presenting in CALR^MUT MPN tumor cells in vivo in patients are possible. Last, whereas we have shown that a heteroclitic peptide vaccine is effective in our mouse model of a tumor presenting a CALR^MUT antigen, it is important to note that heteroclitic peptide-based cancer vaccines have had limited success in past human clinical trials (48). However, all of these trials were targeting nonmutated tumor-associated antigens or cancer germline antigens where the cross-reactive pool of CD8⁺ T cells may have been subject to some immunological tolerance, which is not the case in the context of the CALR^MUT fragment.

Should a vaccine prove effective in patients with CALR^MUT MPN, it would provide a valuable nonredundant benefit in these patients, because there are currently no rationally designed treatments specifically targeting the CALR mutation. Although the JAK1/JAK2 inhibitor ruxolitinib is approved by the U.S. Food and Drug Administration for the treatment of patients with MPN, this approval was granted primarily based on symptomatic benefits (49, 50). Although patients with CALR^MUT MPN demonstrate clinical responses to ruxolitinib, there is no reduction in CALR^MUT allele burden after JAK2 inhibition, and as a result, ruxolitinib does not have substantial disease-modifying activity in MPN and is not curative (51). A longer-term goal of peptide vaccination should be to treat patients with CALR^MUT MPN early in the course of their disease before genetic and clonal evolution has occurred. By directing autologous immune responses specifically against CALR^MUT, peptide vaccination offers the potential to preferentially target the disease-initiating MPN stem cell in patients, which is a deficiency of current MPN drugs, including JAK inhibitors.

In the first human clinical phase 1 vaccination trial with a CALR^MUT peptide fragment with the adjuvant Montanide, it was shown that although the majority of patients displayed a marked CALR^MUT-specific CD4⁺ T cell response, only 2 of 10 patients mounted a CD8⁺ T cell response to the vaccine and none had a clinical response (52). Although limited patient responses suggest that CALR^MUT presentation by peptide vaccination is suboptimally immunogenic, the possibility exists that native CALR^MUT immunization combined with strong immunostimulatory adjuvancy could result in successful priming in vivo. In addition, our findings indicate that these patients could further benefit from a vaccine approach incorporating optimized heteroclitic CALR^MUT peptides. Last, as we show that this strategy can be enhanced by PD-1 blockade in our mouse model, we propose that an off-the-shelf vaccine composed of one or more CALR^MUT heteroclitic peptides in combination with anti–PD-1 or other immune checkpoint inhibitor therapies can be a viable treatment for patients with CALR^MUT MPN.

METHODS

Study design

The primary objective of this study was to investigate associations between the frequency of MHC-I alleles in patients with CALR^MUT MPN compared to control groups (patients with JAK^V617F MPN and, if possible, the general population) and the likelihood that these specific MHC-I alleles were contributing to anti-CALR^MUT MPN immunity. The only criteria for admission into this study were for a patient to be diagnosed with MPN and having a known MHC-I haplotype. All patients analyzed in this study were diagnosed with MPN and tested positive for the CALR^MUT or JAK^V617F mutations. Approval was obtained for the use of patient-derived specimens and access to clinical data extracted from patient charts by the Institutional Review Boards at MSK Cancer Center, the Dana-Farber Cancer Institute, and the Massachusetts General Hospital, as well as by the Danish Regional Science Ethics Committee. Our analysis supported the hypothesis that MHC-I alleles that were predicted to induce anti-CALR^MUT MPN immunity were underrepresented in patients with CALR^MUT MPN. Therefore, the secondary objective of this study was to develop a strategy to counteract the skewing of MHC-I haplotypes in patients with CALR^MUT MPN using heteroclitic peptides. Eight- to 10-week-old female C57BL/6J mice (strain #000664) were purchased from the Jackson Laboratory. Mouse experiments were performed in accordance with institutional guidelines under a protocol approved by the MSK Cancer Center Institutional Animal Care and Use Committee. All mice were maintained in a pathogen-free facility according to the National Institutes of Health (NIH) Animal Care guidelines. For prophylactic cancer vaccines, animals were randomized for the tumor inoculation. For therapeutic cancer vaccines, animals were randomized after tumor inoculation. No mice were excluded from the experiment. All in vivo experiments consisted of at least 10 mice per group.

MHC allele frequencies

For the NEUS cohort, MHC genotypes were manually extracted from patient charts. For the Danish cohort, HLA class I genotypes were determined by next-generation sequencing performed on biobanked samples collected for a previous study (53) using recently described methods (54). In cases where National Marrow Donor Program (NMDP) allele codes were used instead of the World Health Organization nomenclature, the conversion was done according to https://bioinformatics.bethematchclinical.org/hla-resources/allele-codes/allele-code-lists/. If multiple alleles were plausible for a given NMDP allele code, we selected the most likely allele based on ethnicity [typically around 99% confidence, based on known frequencies in the general population (17)]. MHC allele frequency for each HLA gene (A, B, C, DQ, and DR) was broadly calculated as the number of each specific allele divided by the number of the total allele in that cohort (2n per individual). In rare samples, certain patients had incomplete haplotype information where one or more alleles were unknown or incomplete. If the allele was missing, it was censored from the number of total alleles. If locus and group were known but the exact protein was unknown, this allele was censored from the frequency calculation only for alleles from the same group. For an MHC allele to be considered positively skewed, it was required to have an allele frequency of >0.05 in CALR^MUT MPN cases and have >0.2-fold frequency increase compared to both the JAK2^V617F and U.S. Caucasian population group allele frequency in the NEUS cohort, or a >0.125-fold frequency increase compared to the JAK2^V617F group allele frequency in the Danish cohort. Likewise, for an MHC allele to be negatively skewed, it was required to have a frequency of at least 0.05 in both the JAK2^V617F and U.S. Caucasian population groups for the NEUS cohort, or just the JAK2^V617F group for the Danish cohort, and have a >0.2-fold decrease in the CALR^MUT MPN compared to both the JAK2^V617F and U.S. Caucasian population group allele frequency in the NEUS cohort, or a >0.125-fold frequency decrease compared to the JAK2^V617F group allele frequency in the Danish cohort. PCA was calculated in R and plotted in GraphPad Prism v7. All data processing and analysis were performed using the R version 3.3.2 Sincere Pumpkin Patch and GraphPad Prism v7.

Binding affinity predictions and PPS

pMHC-I binding predictions and binding motifs Seq2Logo graphs were collected using NetMHCpan v3 (32) for human MHC-I alleles and NetMHC v4 (36) for murine MHC-I alleles. pMHC-II binding predictions were collected using NetMHCIIpan v3.2 (55). To calculate the PPS, peptide affinities for all six possible MHC-I alleles were identified, and only the lowest affinity value was retained. The protein fragment RRMMRTKMRMRRMRRTRRKMRRKMSPARPRTSCREACLQGWTEA was used for the CALR^MUT. For the CALR^WT sequence, amino acids 1 to 361 of the UniProt sequence P27797 were used. For the influenza NA sequence, the full UniProt sequence D7ED91 was used. All subsequent data processing and analyses were performed using the R version 3.3.2 Sincere Pumpkin Patch and GraphPad Prism v7.

MHC-I stabilization assay

To test peptide binding to murine H-2K^b or human HLA-A*02:01, TAP-deficient RMA/S and T2 cells were used, respectively, as previously reported (56, 57). Briefly, RMA/S cells (58) were serum starved in serum-free RPMI 1640 media overnight at 31°C, and peptides were added at indicated final concentrations, followed by 30 min at 31°C and another 3 hours at 37°C before measuring H-2K^b by flow cytometry (BD Biosciences, clone AF6–88.5). T2 cells (59) were serum starved in serum-free RPMI 1640 media overnight, and cells were added at indicated final concentrations for 16 hours before measuring HLA-A*02:01 surface expression by flow cytometry (BD Biosciences, clone BB7.2). The SIINFEKL peptide was acquired as a custom order from GenScript. The MART1-A2 peptide (ELAGIGILTV) was purchased from JPT Peptide Technologies.

Murine immunizations and subsequent antigen-specific analysis

All peptides were purchased as custom peptide synthesis orders from GenScript at a purity of >98% and resuspended at 10 mg/ml in DMSO (Sigma-Aldrich). For peptide vaccines, peptides were diluted in phosphate-buffered saline (PBS) and emulsified with TiterMax (TiterMax USA Inc.) at a 1:1 ratio immediately prior to immunization, such that each dose was composed of 10 μg of peptide in a total volume of 25 μl. In control (DMSO) immunization, an equivalent volume of DMSO is substituted for the diluted peptide in the TiterMax emulsion. For subsequent in vitro assays, draining inguinal and popliteal lymph nodes were collected at indicated time points, and CD8⁺ T cells were isolated using positive magnetic sorting using mouse CD8a (Ly-2) MicroBeads (Miltenyi Biotec). DNA vaccines were performed using a gene gun as previously described (27, 60) according to indicated time points. Briefly, mice received four injections (400 lb/inch²) of DNA-coated gold particles into the abdominal region of the skin for a total 4 μg of DNA per dose. DNA plasmids encoding the WT or 52-bp deletion CALR^MUT sequences fused with the flag sequence are as previously described (61) and were cloned into the pING plasmid (62). For pING-OVA, the full-length chicken OVA sequence was also cloned into the pING plasmid (62). For pING-CALR^{MUT-CALR9p2(T5F)}, the sequence was directly modified from pING-CALR^MUT by site-directed mutagenesis using Q5 Site-Directed Mutagenesis Kit using the following primers: GGCGAATGATGCGGTTCAAAATGCGCA and TTGTTCTCTGCTCTTCGTCT. CD8⁺ T cells were collected as before, but only from draining inguinal lymph nodes. To test for antigen specificity, the mouse IFN-γ ELISpot set (BD Biosciences) was used according to the manufacturer’s instructions. Briefly, CD8⁺ T cells were frozen immediately after purification in fetal bovine serum (FBS) containing 10% DMSO, thawed 1 day prior to restimulation, and allowed to recover overnight in interleukin-1 (IL-2; 20 U/ml) (PeproTech) RPMI 1640 medium containing 10% FBS, Na-pyruvate, l-glutamine, and penicillin/streptomycin (pen/strep). As a source of APCs, splenocytes from naïve mice were depleted of T cells using magnetic microbeads for CD8a (Ly2) and CD4 (L3T4) (Miltenyi Biotec), pulsed for 1 hour with peptide (100 μg/ml) in serum-free media at 37°C followed by a wash. For each well, 10⁵ CD8⁺ T cells were cocultured with 1 × 10⁵ to 3 × 10⁵ peptide-pulsed APCs and incubated for about 18 hours. Spots were counted using the ImmunoSpot analyzer (Cellular Technology Limited). For tetramer-specific immune characterization, cells were stained with the CALR9p2(T5F) tetramers (NIH Tetramer Core Facility), CD8a (BioLegend, clone 53–6.7), TIM3 (BioLegend, clone RMT3–2.3), CD3e (eBioscience, clone 17A2), CD4 (BD Biosciences, clone RM4–5), CD279/PD-1 (eBioscience, clone J43; Invitrogen, clone RMP1–30), CD44 (Invitrogen, clone IM7), IFN-γ (eBioscience, clone XMG1.2), and TNFα (BD Biosciences, clone MP6-XT22), and the fixable viability dye (Thermo Fisher Scientific), assessed using the BD LSR II Flow cytometer and analyzed by FlowJo v10.

SIINFEKL:H-2Kb expression

B16F10 cells were cotransfected with equal parts pING-OVA and pCMV-Sport6-CALR constructs fused to mCherry, which are previously described (61), using the Megatran 1.0 transfection reagent (Origene). Each construct was mixed with the transfection reagent separately such that all cells received the same amount of pING-OVA construct. B16F10 cells were originally obtained from I. Fidler (M. D. Anderson Cancer Center) and cultured in RPMI 1640 medium supplemented with 7.5% inactivated FBS, 1× nonessential amino acids, and 2 mM l-glutamine. One day after transfection, cells were stained by flow cytometry with H-2K^b (BD Biosciences, clone AF6–88.5) and H-2K^b-SIINFEKL (BioLegend; Clone 25-D1.16).

B16 killing assay

A total of 10⁴ B16F10 cells were seeded in 24-well plates 1 day before coculture. On the day of the killing assay, B16F10 cells were washed and peptide pulsed with peptide (100 μg/ml) for 2 hours at 37°C before a final wash. CD8⁺ T cells from peptide-immunized mice were thawed and allowed to recover as before, and 10⁶ CD8⁺ T cells were added to wells containing the peptide-pulsed B16F10 cells in the presence of recombinant murine interleukin(IL)-2 (rmIL-2) (20 U/ml) (PeproTech) and cocultured for 72 hours at 37°C. After the coculture, supernatants and floating cells were carefully removed, and the remaining B16F10 cells were washed and collected after trypsinization. Countbright counting beads (Invitrogen) were added to the samples, and the cell population was stained with LIVE/DEAD Fixable Green Dead cell stain kit (Invitrogen) and CD45-APC-Cy7 (BD Biosciences, clone 30-F11) before being analyzed by Guava easyCyte 8HT (Millipore). Live CD45-negative cells were normalized to bead counts and normalized by sample and target cells.

Tumor growth experiments

RMA/S cells were maintained in RPMI 1640 medium supplemented with 7.5% inactivated FBS, 1× nonessential amino acids, and 2 mM l-glutamine. The DNA sequence encoding the CALR9p2 peptide (bold) was cloned into the PresentER-IRES-GFP (38) construct using the following oligo 5′-GGCCGTATTGGCCCCGCCACCTGTGAGCGGGAGGATGATGAGGACAAAGATGA.

GGATGTAAGGCCAAACAGGCC-3′ after Sfi I digestion and T4 ligation (New England Biolabs). The resulting construct was used to generate retrovirus by cotransfection with pCL-Ampho into ecotropic Phoenix cells (American Type Culture Collection). PresentER-IRES-GFP and pCL-Ampho were generously provided by D. Scheinberg’s group. Viral supernatants were collected at 48 and 72 hours, and pooled and Retro-X Concentrator (Takara Bio)– concentrated retrovirus was used to transduce RMA/S cells by spinoculation using polybrene (Sigma-Aldrich). GFP-positive cells were FACS (fluorescence-activated cell sorting)–sorted (BD FACSAria III) and cultured in puromycin (4 μg/ml) (Gibco) media. A total of 5 × 10⁶ cells were injected subcutaneously in the flank of mice. For anti–PD-1 treatment, 250 μg of RMP1–14 was injected intraperitoneally in PBS at indicated time points, starting 10 days after tumor inoculation.

Human PBMC in vitro restimulation

For the experiment described in Figs. 3A and 4 (D and E), freshly isolated or thawed cryopreserved healthy donor PBMCs were restimulated with cytokines and peptides as previously described (14). Briefly, on day 0, PBMCs were resuspended in X-VIVO15 media (Lonza) and seeded at 10⁵ per well of a 96 U-bottom plate with granulocyte-macrophage colony-stimulating factor (GM-CSF) (1000 IU/ml) (Sanofi), IL-4 (500 IU/ml) (R&D Systems), and Flt3L (50 ng/ml) (R&D Systems). On day 1, media were refreshed with lipopolysaccharide (0.1 μg/ml) (InvivoGen), 10 μM R848 (InvivoGen), IL-1β (5 μg/ml) (R&D Systems), and indicated peptides (1 μg/ml), and incubated for 24 hours. The CMV, EBV, flu, tetanus (CEFT), and the myelin oligodendrocyte glycoprotein (MOG) peptide pools (JPT Technologies) were used as positive and negative controls, respectively. On days 2 and 5, half the media was refreshed with RPMI 1640 (Gibco) containing 10% human serum (Gemini Bio-Products), gentamycin (10 μg/ml) (Gibco), Hepes (Gibco), GlutaMAX (Gibco), and human(h) IL2 (hIL-2) and hIL-7 to a final concentration of 10 IU/ml and 10 ng/ml, respectively (R&D Systems). On day 8, the media was refreshed without cytokines. On day 10, PBMCs were restimulated with corresponding peptides in the presence of anti-hCD28 and anti-hCD49d (1 μg/ml) (BD Biosciences). As controls, some cells were stimulated with PMA (50 ng/ml; Sigma-Aldrich) and ionomycin (1 μg/ml; Sigma-Aldrich). All antibodies were purchased from BioLegend. LIVE/DEAD Fixable Blue Dead Cell Stain Kit by Thermo Fisher Scientific was used for live and dead cell discrimination. Data were acquired using the BD LSRFortessa, and the data were analyzed on FlowJo V10 (TreeStar). For the ELISpot, methods are as previously described (14) using the follower 15-mer peptides that cover the entire CALR^MUT fragment: RTRRMMRTKMRMRRM, MMRTKMRMRRMRRTR, TKMRMRRMRRTRRK, RMRRMRRTRRKMRRK, MRRTRRKMRRKMSPA, RRKMRRKMSPARPRT, RRKMSPARPRTSCRE, SPARPRTSCREACLQ, PRTSCREACLQGWTE, TSCREACLQGWTEA.

For the experiments described in Fig. 3 (B to D) and fig. S3A, generation of specific T cell cultures and subsequent ELISpot and flow cytometry is as described previously (13), whereas Chromium-51 release cytotoxic assay, autologous CD14⁺ purification, culture, and siRNA-mediated CALR silencing are as described previously (11).

For experiments described in fig. S3 (B and C), frozen dendritic cells purified by magnetic bead positive selection specific to CD1c surface antigen were thawed and cultured in 5% human serum Iscove’s modified Dulbecco’s medium (IMDM) (Gibco) with pen/strep, l-glutamine, and Hepes (henceforth I5 media) with IL-4 (80 ng/ml) (PeproTech) and GM-CSF (Gibco) for 1 day, and subsequently infected by empty, CMV, EBV and influenza (CEF) encoding or the 44–amino acid long CALR^MUT antigen-encoding AdV5 in human lactoferrin (1 mg/ml) (Sigma-Aldrich) I5 media at a multiplicity of infection of 5000 for 24 hours, followed by a PBS wash. Autologous pan T cells were thawed and cultured with the transduced dendritic cells at 10:1 T:DC ratio in I5 media with IL-15 (10 ng/ml) (PeproTech) and IL-21 (30 ng/ml) (R&D Systems) for 2 days. Media was refreshed every 2 to 3 days with I5 media containing IL-15 (10 ng/ml) and IL-2 (10 IU/ml). For peptide restimulation, T cells were cultured overnight in fresh IL-15/IL-2 I5 media containing a CALR^MUT peptide pool (95% purity Vivitide) at 2 μg/ml and eBioscience transport inhibitor cocktail (Thermo Fisher Scientific). As controls, human PepTivator AdV5 Hexon Peptide Pool (Miltenyi) or CEF-MHC Class I Control Peptide Pool “Plus” (Cellular Technologies) was used. Cells were stained with LIVE/DEAD Fixable Aqua Dead Cell Stain (V500) (Thermo Fisher Scientific), CD3 PerCP Cy5.5 (BioLegend), CD4 BV421 (BioLegend), CD8 APC-Cy7 (BioLegend), TNFα fluorescein isothiocyanate (FITC) (R&D Systems), and IFN-γ BV785 (BioLegend). The eBioscience Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific) was used to fix and permeabilize the cells. Samples were assessed by BD FACSCelesta flow cytometer and analyzed by FlowJo v10.

For experiments involving K562 cell restimulation from fig. S9, PBMCs were restimulated in complete IMDM media + 10% human serum + 4 mM l-glutamine + 55 μM β-mercaptoethanol + rhIL-2 (100 U/ml) (Chiron), rhIL-7 (25 ng/ml), and rhIL-15 (25 ng/ml) (PeproTech) with DMSO or 10 μg/mL per peptide for 14 days. Cytokines were refreshed twice a week, and peptides were refreshed at day 7. For the secondary restimulation, HLA-null K562 cells were transduced to express HLA-A*02:01 using the retroviral construct SFG-HLA-A*02:01 (provided by Dr. R. O’Reilly, MSKCC), were peptide pulsed with indicated peptides (100 μg/ml in serum-free RPMI 1640 for 2 hours at 37C), and used as target cells in the presence of anti-hCD28 and anti-hCD49d (1 μg/ml). For intracellular staining, monensin and brefeldin A (BD Biosciences) were added 1 hour after restimulation and culture were left to incubate for another 12 hours. Cells were then stained for CD3 (clone: OKT3, FITC), CD4 (clone: RPA-T8, APC), and CD8a (clone: RPA-T4, BV785), permeabilized, and fixed with BD Cytofix/Cytoperm reagents according to the manufacturer’s protocol and subsequently stained for IFN-γ [clone: B27, phycoerythrin (PE)] and TNFα (clone: Mab11, PE/Cy7). Samples were assessed by BD LSR II Flow cytometer and analyzed by FlowJo v10.

Statistical analysis

Details of the study outline, sample size, and statistical analysis are shown in the main text, figures, tables, and figure legends. To calculate significance in distribution of MHC frequencies, Barnard’s unconditional test and the chi-square test were used as indicated in R using the barnard.test() function (two-tail) from the Barnard package and as well as the base chisq.test() function. The R version 3.3.2 Sincere Pumpkin Patch was used. For unpaired Student’s t tests, area under the curve calculations, log-rank survival test, and GraphPad Prism v7 were used.