Peptides were analyzed for potential MHC-II binding using immune epitope database (IEDB) (http://www.iedb.org/) (28–30). Peptides ( Table 1 ) were synthesized at >90% purity (Genscript), stored lyophilized at −80°C and then reconstituted in PBS on day of use.

Animal experiments were carried out with ethical approval under a Home Office approved project license. HLA-DR4 mice (Model #4149, Taconic), HLA-A2/DR1 (HLA-HHDII/DR1, Pasteur Institute), HLA-A2.1+/+ HLADP4+/+ hCD4+/+ (HLA-HHDII/DP4; EM:02221, European Mouse Mutant Archive) transgenic mice and C57Bl/6J mice or Balb/c mice (Charles River) aged 8-12 weeks were used.

The murine melanoma B16F1 cell line (ATCC-CRL-6323) obtained from ATCC was engineered with mouse MHC K/O and transfected with HHDII (A2) alone or in combination with either constitutive HLA-DP4, HLA-DR1, HLA-DR4 or IFNγ-inducible HLA-DR4 (iDR4) using plasmids as previously described (17, 26). Cells were cultured in RPMI medium 1640 with L-glutamine (2 mmol/L), 10% FCS and appropriate antibiotics to maintain plasmids. Cell lines utilized were mycoplasma free, authenticated by suppliers (STR profiling), and used within ten passages.

Mice were injected subcutaneously (s.c.) with 25 μg of each peptide either as a pool or individually in combination with 5μg of CpG ODN 1826 (CpG) and Monophosphoryl Lipid A (MPLA) (InvivoGen) on days 1, 8 and 15 and responses analyzed on day 21.

For anti-tumor experiments, mice were implanted s.c. with tumor cells three days prior to immunization regime above. Implant doses were 5.0 x 10⁵ cells/mouse for B16 HHDII/DR1, 2.5x10⁴ cells/mouse for B16F1DR4 and 5x10⁴ cells/mouse for B16F1 iDR4 and B16F1 h2Ab1 B2M dKO cell lines (17, 31, 32). Tumor growth was monitored twice weekly and mice were humanely euthanized once tumors approach the license limit of 15 mm in diameter.

Murine ELISpot kits (Mabtech) were used with 5 × 10⁵ splenocytes/well and 10μg/ml synthetic peptide added to triplicate wells. 20μg/ml anti-CD8, or anti-CD4 antibodies were added to appropriate wells for blocking studies. Plates were incubated at 37°C for 40 hours in an atmosphere of 5% CO2. Example plate images are shown in Supplementary Figure 1 . Spots were analyzed using an automated plate reader (Cellular Technologies Ltd).

PBMCs were isolated from healthy donors and cancer patients ( Table 2 ) using histopaque-1077 then depleted using anti-CD25 microbeads and MACS Cell Separation Columns (Miltenyi). Isolated CD25-depleted PBMCs were loaded with 5µM carboxyfluorescein succinimidyl ester (CFSE) (Thermofisher) (32). Cells were resuspended at 1.5×10⁶/ml in RPMI with autologous serum as previously described in full (32), plated and stimulated with vehicle, PHA (10 µg/ml), soluble anti-CD3 (positive control, 30ng/ml) or peptide (10µg/ml). Peptide concentration was selected based on reports in the literature using similar post-translationally modified epitopes (16, 18, 33). On day 10, 500µl of cells were stained with anti-CD4 (Thermofisher clone RPA-T4, 1/50 dilution) and anti-CD8 (Miltenyi Biotec clone REA734, 1/50 dilution) antibodies for 30 minutes and then samples were washed, fixed and analyzed immediately. For restimulation assays, cells were cultured in media with IL-2 IL-7 and IL-15 cytokines (Peprotech) and 10ug/ml of homocitrullinated peptides. On day 14, cells were harvested. Human IFNγ ELISpot kits (Mabtech) were used with 1 x 10⁵ cells/well with 10ug/ml peptide or PHA (10 µg/ml) in quadruplicate wells. Where appropriate, 7.6 µg/ml blocking antibodies (Isotype control clone MOPC-173 IgG2a kappa Biolegend, Anti-DR,DP,DQ clone Tu39 Biolegend, Anti-DP clone B7/21 Leinco, Anti-A,B,C cloneW6/32 Biolegend, Anti-DR clone G46-6 BD) were added for 30 minutes prior to peptides. ELISpot plates were incubated at 37°C for 24 hours in an atmosphere of 5% CO2 and then developed following the manufacturer’s instructions. Spots were analyzed using an automated plate reader (Cellular Technologies Ltd). All samples showed responses to PHA (data not shown).

Stained CD4+/CFSE^high and CD4+/CFSE^low cell populations were bulk sorted into RNA protect (Qiagen) using a MoFlo XDP Sorter (Beckman Coulter). RNA was purified from sorted cells, RT-PCR performed, the cDNA was subjected to Amplicon rescued multiplex PCR (ARM-PCR) using human TCR α and β 250 PER primers (iRepertoire Inc., Huntsville, AL, USA) (26). Ten assessed sample libraries were pooled and sequenced using the Illumin//a MiSeq platform (Illumina, USA). Tree maps and D50 diversity values generated using iRweb software. Tree maps show each unique CDR3 as a colored rectangle, the size of each rectangle corresponds to each CDR3s abundance within the repertoire and the positioning is determined by the V region usage. Diversity was measured using D50 immune repertoire diversity index. The D50 index is a quantitative measure of the degree of diversity of T cells within a sample. The D50 is the percentage of T-cell clones that account for the cumulative 50% of the total CDR3s counted in the sample. The more diverse a library, the closer the value will be to 50. Low diversity values are associated with decreased diversity. Data are presented as non-normalized (which takes into account the frequency of each unique CDR3).

Statistical analysis was performed using GraphPad Prism software version 9. Comparative analysis of the ELISpot results was performed by applying paired or unpaired ANOVA or Student t test as appropriate with P values calculated accordingly. Comparison of tumor survival was assessed by log-rank test. Human responses were analyzed by Wilcoxon matched-paired test. P<0.05 values were considered statistically significant.

We have previously shown that homocitrullinated peptides can stimulate responses restricted through the HLA-DP4 allele (22). In this study we first aimed to determine whether peptide immunization can induce homocitrulline-specific responses via HLA types other than the HLA-DP4 allele. The vimentin (Vim), cytokeratin 8 (Cyk8) and aldolase A (Aldo) sequences were analyzed by IEDB for sequences containing lysine within the predicted core binding region and since homocitrulline modification may influence the binding affinity, peptides with variable predicted binding scores for HLA-DR4 or HLA-DR1 alleles were selected. Predicted binding scores are shown in Table 1 . Responses to five homocitrullinated peptides were assessed in HLA-transgenic mice expressing human HLA-DR1 along with chimeric human HLA-A2 (HHDII) or HLA-DR4 ( Figures 1A, B ). This screening identified significant responses to Vim116-135^Hcit in both strains. The homocitrullinated peptide Vim116-135^Hcit was synthesized with the lysine residues at positions 120 and 129 both converted to homocitrulline ( Table 1 ).

Responses were also screened for five Cyk8 peptides ( Figure 1C ) and two Aldo peptides ( Figure 1D ) in HLA-transgenic mice expressing human HLA-DR1 and chimeric HLA-A2 (HLA-HHDII/DR1) mice. This screening identified Cyk8 371-388^Hcit, Cyk8 112-131^Hcit and Aldo140-157^Hcit as inducing immune responses in the HLA-HHDII/DR1 mice. Cyk8 371-388^Hcit was synthesized with the lysine converted to homocitrulline at position 381 and Aldo140-157^Hcit contained three homocitrulline residues at positions 140, 147 and 153 ( Table 1 ).

We have previously observed that both Aldo140-157^Hcit and Cyk8 371-388^Hcit can induce immune responses in HLA-HHDII/DP4 mice (22) suggesting an additional HLA restriction. The peptides were therefore tested in additional mouse strains. In HLA-HHDII/DP4 mice, the previously described responses were seen for Cyk8 371-388^Hcit, Aldo74-93^Hcit and Aldo140-157^Hcit ( Figure 1E ). No immune responses to Vim116-135^Hcit were observed in HLA-HHDII/DP4 ( Figure 1E ) nor C57Bl/6 mice ( Figure 1F ). However, significant Vim116-135^Hcit (p=0.0018), Cyk8 371-388^Hcit (p=0.0005) as well as Aldo 74-93^Hcit (p=0.0001) responses were seen in Balb/c mice ( Figure 1G ).

This data shows that immunization with homocitrullinated peptides can induce responses in mice expressing the shared epitope human HLA alleles DR4 and DR1 as well as human HLA-DP4 and murine H-2d haplotype. However, cross reactivity between HLA alleles varies between peptides.

Immunization with long peptides has the potential to stimulate both CD4 and CD8 responses, therefore we further characterized the immune responses to three peptides to determine if responses were CD4 mediated and homocitrulline specific. Mice were immunized with the Vim116-135^Hcit peptide and immune responses determined by IFNγ ELISpots. In HLA-HHDII/DR1 ( Figure 2A ), HLA-DR4 ( Figure 2B ) and Balb/c ( Figure 2C ) mice responses to the modified homocitrulline peptide were significantly greater than the responses to the wild type (wt) peptide (p<0.0001 for all strains). Anti-CD4 blocking antibodies significantly reduced the IFNγ response to Vim116-135^Hcit (p<0.0001, p=0.0054 and p=0.0143 for HLA-HHDII/DR1, HLA-DR4 and Balb/c respectively) but anti-CD8 antibody had no significant effect on the immune response. Example ELISpot images are shown in Supplementary Figure 1 .

The Cyk8 371-388^Hcit response in HLA-HHDII/DR1 ( Figure 2D ) and Balb/c ( Figure 2E ) mice was also assessed. IFNγ responses to Cyk8 371-388^Hcit showed minimal cross reactivity to the wt peptide with the Hcit responses significantly greater than the wt response in both strains (p=0.0045 and p=0.0073 for HLA-HHDII/DR1 and Balb/c respectively). In both strains the IFNγ response to Cyk8 371-388^Hcit was partially blocked by anti-CD8 antibodies but fully blocked by anti-CD4 antibodies. In HLA-HHDII/DR1 mice responses were reduced from a mean of 503 spots/million splenocytes to 73 spots/million splenocytes (p<0.0001) after blocking with the anti-CD4 antibody and only slightly reduced to a mean of 423 spots/million splenocytes in responses to anti-CD8 blocking antibodies (p=0.0148). In Balb/c mice the mean response of 433 spots/million splenocytes was reduced to 53 spots/million splenocytes after CD4 blocking (p=0.0117) and 289 spot/million splenocytes after CD8 blocking (p=0.0048), suggesting that the Cyk8 371-388^Hcit responses were also CD4 and CD8 mediated.

Finally, the Aldo74-93^Hcit response was characterized in Balb/c mice ( Figure 2F ). IFNγ responses to Aldo74-93^Hcit were significantly greater than wild type responses (p<0.0001). The Aldo74-93^Hcit response was partially blocked by anti-CD8 antibodies but fully blocked by anti-CD4 antibodies. Responses were reduced from a mean of 583 spots/million splenocytes to 24 spots/million splenocytes (p=0.0020) after blocking with the anti-CD4 antibody and partially reduced to a mean of 255 spots/million splenocytes in responses to anti-CD8 blocking antibodies (p=0.0059).

To further confirm that the Vim116-135^Hcit response is modification specific immunizations with the wild type Vim116-135wt peptide were performed ( Figure 2G ). No immune responses to Vim116-135^Hcit were observed in either HLA-HHDII/DR1 or HLA-DR4 mice. This confirms that only the modified peptide can induce an immune response.

Together these results confirm that the immune responses to Vim 116-135^Hcit, Cyk8 371-388^Hcit and Aldo 74-93^Hcit were CD4 and possibly CD8 mediated and homocitrulline specific in mice with the appropriate MHC alleles.

Having demonstrated that immunized transgenic mice expressing human HLA types show homocitrulline-specific immune responses, we next determined whether healthy human donors exhibit a repertoire for homocitrullinated peptides. For this we concentrated on the Vim116-135^Hcit peptide.

PBMCs were isolated from 17 healthy donors and assessed for proliferative responses to homocitrullinated peptides. Donor details including HLA-types are shown in Table 2 . Flow cytometry staining was used to assess CFSE proliferation in the CD4+ or CD8+ cell populations. Proliferation was assessed at day 10 post-stimulation as this timepoint showed clear peptide-induced proliferative responses. An example staining profile is shown in Figure 3A and additional example CFSE profiles for day 7 and day 10 are shown in Supplementary Figure 2 . CFSE proliferation to the positive control PHA was seen in both the CD4 and CD8 populations, however, Vim116-135^Hcit proliferation was predominantly seen in the CD4 population. Collated data from healthy donors is shown in Figure 3B whilst CD4 proliferative responses in individual healthy donors are shown in Figure 3Ci . In PBMCs from healthy donors, 15/17 showed CD4 T-cell proliferation in response to Vim116-135^Hcit peptide. Across all donors, Vim116^Hcit induced a significant increase in CD4 proliferation compared to media only (p<0.0001) ( Figure 3Cii ).

To determine whether the Hcit responses seen in healthy donors were modification-specific restimulation assays were performed on 5 healthy donor PMBCs that had been cultured with modified peptide for 14 days ( Figure 3D ). IFNγ ELISpot showed restimulation responses to modified but not wt peptide. All responding donors showed a significantly higher response to Vim116-135^Hcit compared to Vim116-135wt peptide. For two donors restimulation assays were also performed in the presence of blocking antibodies ( Figure 3E ). In both donors’ responses to Vim116-135^Hcit were significantly reduced by addition of pan-MHC-II blocking antibodies or anti-HLA-DR blocking antibodies indicating the involvement of the HLA-DR allele. Together this data confirms that healthy human PBMCs have a modification specific repertoire for this peptide that is likely restricted through HLA-DR alleles.

To determine the clonality, TCR Vα and Vβ sequencing was performed on proliferating and non-proliferating CD4 T-cells isolated from healthy donor that had been stimulated ex vivo. TCRα and β CDR3 tree maps were used to display clonality ( Figure 4 ). Stimulation with vehicle showed minimal proliferative responses and no consistent differences in the clonality of the non-proliferative population ( Figure 4A ). Polyclonal CD3 stimulation also led to no clear changes in the diversity of clonotypes, whereas responses to the HLA-DP4 restricted Hepatitis B peptide showed decreased diversity after proliferation ( Figure 4B ). Vim116-135^Hcit stimulated samples also showed decreased CDR3 diversity in the proliferating CD4+CFSE^low cells with a dramatic increase in the relative frequency of subsets of CDR3 sequences suggesting a focused repertoire of cells responding to Vim116-135^Hcit ( Figure 4C ). The oligoclonal nature of the proliferating CD4 T cells was confirmed by the lower diversity index (D50) of the CDR3 sequences from the proliferating (CD4⁺CFSE^low) compared to non-proliferating (CD4⁺CFSE^high) populations ( Figure 4D ).

We have identified HLA DR-allele associated homocitrulline-specific immune responses. Our previous studies have shown that MDSC-produced MPO can mediate homocitrullination in the tumor microenvironment making homocitrullinated epitopes good targets for tumor therapy (22). However, not all homocitrulline-specific immune responses show anti-tumor responses suggesting that not all lysines are naturally homocitrullinated or presented. Therefore, we next determined if the immune response to Vim116-135^Hcit was sufficient to provide tumor therapy in vivo. Transgenic mice were implanted with HLA-matched B16 F1 cells and immunized with Vim116-135^Hcit peptide. Immunization schedule is presented in Supplementary Figure 3A . In mice containing the HLA-HHDII/DR1 allele ( Figure 5A ), Vim116-135^Hcit immunization provided significantly better survival than mice from the unimmunized control (p<0.0001), Vim116-135wt peptide immunized (p=0.0030) or CpG/MPLA adjuvant only (p-=0.0266) groups. Tumor growth curves are presented in Supplementary Figure 3B . In HLA-DR4 mice ( Figure 5B ) Vim116-135^Hcit immunization was also associated with a significant increase in survival when compared to unimmunized control mice (p=0.0042) or CPG/MPLA control (p=0.0235).

Since, both these cell lines were engineered to express MHC-II constitutively and expression of MHC-II in non-APCs, including most cancer cells, is not constitutive but driven by the IFNγ-induced CIITA promoter pIV (34) we engineered B16 cells to express transgenic MHC-II under an IFNγ-inducible promoter (iDR4. Mice implanted with B16 iDR4 ( Figure 5C ) and immunized with Vim116-135^Hcit showed a significant survival advantage over mice immunized with adjuvant only (p=0.0097) or unimmunized mice (p=0.0005). Tumor growth curves are shown in Supplementary Figure 3C . Together this data shows that immunization with Vim116-135^Hcit confers a potent anti-tumor response.

To confirm whether other Homocitrullinated peptides can also induce an anti-tumor response the Cyk8 371-388^Hcit and Aldo 140-157^Hcit peptides were also tested in vivo. In HLA-HHDII/DR1 mice implanted with B16 HHDII/DR1 tumor cells immunization with Cyk8 371-388^Hcit induced a significant survival advantage when compared to unimmunized control (p=0.0072) ( Figure 5D ). In this same model, immunization with Aldo 140-157^Hcit also resulted in a significant anti-tumor effect (p<0.0001) when compared to unimmunized control ( Figure 5E ).

CD4 T cells recognize homocitrullinated epitopes but the anti-tumor effect may be mediated by a bystander effect of CD4 activation of other effector cells such as CD8, NK cells or macrophages or, CD4s may act directly via induction of MHC-II on tumor cells. We therefore investigated if the expression of MHC-II on tumor cells was required for the efficacy of the vaccine. HLA-HHDII/DR1 mice were implanted with B16 cells expressing matched MHC-I but no MHC-II (B16 HHDII/MHC-I/II KO). In this model APCs express MHC-II and CD8 T-cells can recognize the tumor via MHC-I. The anti-tumor effect of Vim116-135^Hcit peptide immunization was completely lost ( Figure 6A ) in this model, suggesting therapy was not mediated by other effector cells. In contrast, HLA-DR4 mice implanted with B16 MHC-II K/O cells and immunized with Vim116-135^Hcit ( Figure 6B ). did demonstrate a significant survival advantage over control unimmunized mice (p=0.0010). HOWEVER, this advantage was still reduced compared to the model expressing MHC-II shown Figure 5B an suggesting that, in this case, both bystander and direct CD4 recognition of tumor may play a role in the vaccine efficacy(p=0.0010).

Given that Vim116-135^Hcit shows a strong anti-tumor response in vivo in mouse models, we next determined if cancer patients had an immune repertoire capable of responding to this peptide. PBMCs were isolated from seven lung cancer patients, two ovarian cancer patients and one breast cancer patient (n=11, Table 3 ) and assessed for proliferative responses to Vim116-135^Hcit. PBMCs from 4/11 cancer patients showed responses to Vim116-135^Hcit ( Figure 7A ). Across all donors, Vim116-135^Hcit induced a significant increase in CD4 proliferation compared to media only (p=0.0420) ( Figure 7B ). This suggests that some cancer patients may be able to respond to immunization making this a potential anti-cancer therapy. Overall, this study has identified three homocitrullinated peptides that have different but overlapping HLA-restrictions. The responses are modification specific and CD4-mediated. The results suggest that these responses can be utilized to target tumors and that they can induce responses in healthy donors and cancer patients.