Shared neoantigens’ atlas for off-the-shelf cancer vaccine development

Abstract

Background

We have recently described that the most prevalent 100 mutations identified in human cancers, both single nucleotide variations (SNVs) and InDels, generate a handful number of shared mutated neoantigens (SNV and InDel-NeoAgs) in association with 5 HLA-A and 7 B haplotypes.

Methods

In the present study, we expanded such analysis to 50 haplotypes in the three MHC class I loci (10 HLA-A, 27 HLA-B and 13 HLA-C), including all the mutated proteins identified in at least 5% of cancer patients.

Results

Overall, the extended analysis identified 15 SNV-NeoAgs and 55 InDel-NeoAgs with a significant affinity improvement over the corresponding wt (DAI > 10). These targetable shared NeoAgs are prevalently derived from PIK3CA_H1047R (6/15 SNV-NeoAgs) and LARP4B_T163Hfs (30/55 InDel-NeoAgs). From the HLA perspective, the HLA-A*33:03 is associated with the largest number of SNV-NeoAgs (4/15 NeoAgs) and the HLA-B*58:01 is associated with the largest number of InDel-NeoAgs (16/55 NeoAgs).

According to the distribution of each HLA haplotype in at least 10% of the regional populations, therapeutic cancer vaccines based on mutated shared SNV and InDel-NeoAgs, might be developed for COAD, STAD and UCEC cancers, with a global coverage, and for PAAD and UVM, with a regional coverage.

Conclusions

This represents the first in-depth analysis for the identification of a specific repertoire of shared mutated NeoAgs, most of which never reported before. Such shared SNV and InDel-NeoAgs are indispensable for the development of “off-the-shelf” cancer vaccines targeting a relevant percentage of cancers in a significant percentage of cancer patients worldwide.

Introduction

Driver mutations in specific proteins confer a growth advantage on the cellular survival or proliferation, possibly leading to transformation and cancer development [1,2,3,4,5]. Single nucleotide variants (SNVs), occurring at the first or second position of the codon, may result into a single amino acid substitution in the protein sequence (nonsynonymous change or a missense variant). Short insertions or deletions (InDels), if not a multiple of three bp, will result in a frameshift and a complete change in the reading frame of the downstream sequence of the gene. This will alter the product of translation, potentially leading to targeted decay of the alternative mRNA [6,7,8].

The resulting SNV or InDel mutation may fall in a peptide generated by the proteasome and loaded onto major histocompatibility complex (MHC) class I molecules by the transporter associated with antigen processing (TAP). When the peptide carrying the mutation is compatible with the cellular HLA allele, the stable peptide-MHC (pMHC) complex moves to the cell surface and cancer cells will present tumor-specific mutated antigens (“neoantigens”). They are scrutinized by the immune system and, if identified as non-self, become targets for immune-mediated destruction [9].

Consequently, tumor-specific mutated neoantigens represent the most specific and potent non-self immunogens to be used for developing cancer vaccines [10, 11]. Indeed, they would elicit a T cell immune response that can exclusively target the tumor while sparing healthy tissue [12]. To this aim, the identification of tumor-specific mutated neoantigens shared among patients with the same tumor or different type of tumors would represent the “holy grail” of cancer immunotherapy. Such neoantigens would be valuable tools for off-the-shelf vaccines, but only if they are shared among a substantial number of patients and presented by common HLA alleles. Unfortunately, this does not seem to be the case in the driver mutations found in all solid tumors at The Cancer Genome Atlas (TCGA) [13].

A list of about 20 putative shared mutated neoantigens, out of more than 1 million screened nonsynonymous missense mutations, have been previously predicted in at least 5% of patients in one or more cancer types [14, 15]. However, the selection parameters used to define such shared neoantigens raise few doubts. By definition, a “neoantigen” is a mutated epitope that, for a given HLA allele, is a strong binder and the wt nonmutated counterpart is not a binder. Alternatively, the ratio of MHC binding affinity between the mutant and normal peptide, namely differential agretopicity index (DAI), should be > ten [16]. Only neoantigens identified according to such a definition have been shown to correlate with intratumoral T-cell responses and predict patient survival [14]. Indeed, given the T cell receptor degeneration, if the efficiency of antigen presentation is very similar (e.g. DAI < 10), the mutated neoantigens would not be seen as non-self and suffer from the same immunological tolerance of the self nonmutated counterpart [17].

Along the same path of searching for shared mutated neoantigens, our group has recently performed a prediction analysis on the most frequent 100 mutations reported at the TCGA, which collectively occur in 56.65% of all cancer cases [18]. Moreover, these include the driver missense and frameshift mutations found in more than 5% of patients affected by a specific cancer or shared by more than 5% of cancers. Among others, were analyzed the BRAF_V600E (found in > 40% of melanoma and > 60% of thyroid ca), KRAS_G12D (found in > 30% of pancreas ca), IDH1_R132H (found in > 35% of brain ca) and GNA11_Q209L (found in > 40% of uveal melanoma). The neoantigen prediction was performed taking into consideration the most frequent 12 HLA-A (5) and B (7) alleles and selecting only the mutated peptides predicted to have very strong affinity (< 100 nM) while the corresponding non-mutated wt peptide show very low (DAI > 10) or no affinity to the same allele. Based on such stringent parameters, the results returned only 10 predicted neoantigens from 7 missense mutations (SNV-NeoAgs) and 9 predicted neoantigens from 6 frameshift mutations (InDel-NeoAgs). Of these, only the GNA11_Q209L FRMVDVGGL SNV-NeoAg may have a relevant application as off-the-shelf vaccine in > 40% of uveal melanoma (UVM) cases when positive for the HLA-B*27:05 or 39:01.

The aim of the present study was to expand such a prediction analysis to a much broader number of alleles, including HLA-A, B and C haplotypes. A total of 50 alleles (10 HLA-A, 27 HLA-B and 13 HLA-C) were considered to predict neoantigens from the SNV and InDel mutations found in more than 5% of patients affected by a specific cancer. The findings show that, based on the distribution of specific HLA haplotypes present in at least 10% of regional populations, therapeutic cancer vaccines targeting shared SNV and InDel-derived neoantigens could be developed with global coverage for colon adenocarcinoma (COAD), stomach adenocarcinoma (STAD), and uterine corpus endometrial carcinoma (UCEC), and with regional applicability for pancreatic adenocarcinoma (PAAD) and uveal melanoma (UVM).

Materials and methods

Selection of cancer mutations from TCGA

The TCGA was interrogated for the selection of cancer mutations. The top 100 most recurrent somatic mutations across all solid tumors in the TCGA database were selected for the analysis. Collectively, they are identified and reported in 51.8% of all cancer cases at the TCGA database. Each mutation was assessed to confirm that they are identified in more than 5% or shared by more than 5% of cancer cases.

Prediction of mutated neoantigens (NeoAgs)

Each of the wild-type (wt) proteins were downloaded from the UniProt database (https://www.uniprot.org). The amino acid sequences were manually modified, introducing the described mutation (substitution or insertion/deletion). The paired wt and mutated sequences from each protein were analyzed using the NetMHCpan4.1 algorithm (https://services.healthtech.dtu.dk/service.php?NetMHCpan-4.1) to predict the best epitopes. All the newly derived sequences from the premature stop codons were included in the analysis, even if very short. Peptides of 8–12 amino acidic residues with binding level defined as “SB”, “WB” to the 50 most frequent HLA-A, B and C alleles were selected. For subsequent analysis, epitopes with an affinity value < 100 nM were arbitrarily defined as strong binders (SB); < 200 nM as weak binders (WB) and > 200 nM as poor binders (PB). Differential agretopicity index (DAI) was used to evaluate the strength of NeoAgs derived from mutations and only epitopes with DAI > 10 (Affinity nonmutated/Affinity mutated) were selected.

Evaluation of novel identified NeoAgs

The mutated NeoAgs identified as SB with a DAI > 10 were submitted to the Immune Epitope Database & Tools (www.iedb.org) and literature search to verify whether the predicted epitopes have been already described and validated in literature. The setting parameters were for exact sequence matching search.

Statistical analysis

The statistical analyses were performed with GraphPad software (Version 6.01). Comparison between individual data points were performed with the two-sided Student’s t-test and ANOVA, as appropriate. Normally distributed data were represented as mean ± S.E.M. Two-way ANOVA and Bonferroni post-hoc analysis were used to examine the significance of differences among groups. All P values were two-tailed and considered significant if less than 0.05.

Results

Strategy for prediction of mutated antigens from shared mutations

Cancer mutations from the TCGA were selected if identified in more than 5% of patients affected by a specific cancer or shared by more than 5% of cancers. The most frequently identified 100 mutations have such characteristics. For each protein carrying such mutations, the corresponding wild-type sequence was manually curated to introduce either the missense point mutations from single nucleotide variations (SNV) or the new sequence from the alternative open reading frame resulting from InDel mutations.

In order to predict mutated antigens derived from SNVs, a 23mer peptide was extracted for each protein, centered on the mutated residue (from −11 to + 11), and overlapping 8–12aa peptides were designed with the mutated residue at each of the positions.

Alternatively, to predict mutated antigens from the InDel mutations, a peptide from each protein was based on a sequence starting at position −11 from the mutated aminoacid residue and including the entire new downstream sequence derived from the alternative open reading frame. The length of the latter mutated peptides ranged from 4 to 62 aa, according to the position of the newly generated stop codon along the shifted reading frame (Suppl. Figure 1).

For both types of mutations, wt and mutated peptides from each protein were subjected to the same prediction analysis, in order to assess the affinity to the selected 50 HLA-A, B and C alleles.

Selection of the HLA alleles

The NetMHCpan 4.1 algorithm was interrogated to predict antigens in the mutated peptides associated to the most prevalent MHC-I HLA alleles, with a global coverage in all Continents. To such aim, 50 alleles, 10 HLA-A, 27 HLA-B and 13 HLA-C alleles were selected. Among these, the A*02:01 and 24:02 are the most prevalent across the Continents; the A*11:01 is highly prevalent in South and South-East Asia as well as in Oceania; the A*31:01 is specifically represented only in South America; the A*33:03 is specifically represented only across Asia; the A*34:01 is specifically represented only in Oceania (Fig. 1A). The most prevalent B alleles across the Continents were selected, including those represented only in specific Continents, such as the B*51:01 which is the most frequent in South Asia and highly frequent in South-East Asia; the B*54:01 and the B*58:01 which are highly frequent or the most frequent in South-East Asia, respectively (Fig. 1B). Similarly, the most prevalent C alleles across the Continents were selected, including the C*04:03 which is the most frequent in South America and highly frequent in Oceania; the C*08:01 which is specifically frequent in North-East and South-East Asia as well as in Oceania; and the C*15:02 which is specific to South-Asia and Oceania (Fig. 1C).

Fig. 1

HLA haplotypes’ prevalence. The prevalence of the most common HLA-I haplotypes in sub continental regions for HLA-A (A), HLA-B (B) and HLA-C (C) alleles

Prediction of mutated antigens associated with HLA-A haplotypes

Predicted mutated antigens (mut-Ags) associated with HLA-A haplotypes are identified only in 11 and 7 proteins, characterized by SNVs (missense mutations) and InDel (frameshift mutations), respectively. Considering the total number from both SNV and InDel mutations, 78% are scored poor binders (> 200 nM affinity—PB), 10.5% are scored weak (100–200 nM—WB) and 11.7% strong (< 100 nM—SB) binders, on average (Suppl. Table 1A). Moreover, the majority of the strongest binder antigens (75.8%) are predicted to bind more than a single haplotype (shared) (Suppl. Table 1).

Regarding the SNVs, the number and the score of the predicted mut-Ags greatly varied among the 11 proteins. In particular, the largest number of WB and SB SNV mut-Ags are predicted in the five SNVs identified in KRAS (8 WBs and 3 SBs), however the four SNVs identified in PIK3CA provided the largest number of SBs (nr. 6). Overall, the individual SNVs generating the largest number of SBs (nr. 3) are the PIK3CA_H1047R and TP53_R175H. On the contrary, the individual SNVs identified in PTEN, IDH1, FGFR3 and the KRAS_G13D do not generate any WB nor SB (Fig. 2A and Suppl. Table 1A).

Fig. 2

Predicted shared mut-Ags for HLA-A. The diagrams show the number of putative mut-Ags derived from SNV (A) and InDels (B). The antigens are grouped by the affinity to the HLA molecules (PB poor binders, WB weak binders, SB strong binders)

Similar to SNVs, the number and the score of the predicted mut-Ags derive from the InDels (InDel mut-Ags) greatly varied among the 7 proteins. In particular, the largest number of WB and SB InDel mut-Ags are predicted in the LARP4B_{T163Hfs_47} (4 WBs and 4 SBs), of which 2 WBs and only 1 SB are shared. In parallel, the same LARP4B InDel mutation generates also the largest number of PBs (23 in total). Interestingly, the RNF43_{G659Vfs_41} and the JAK1_{K860Nfs_16} generate less WBs and SBs (6 and 3, respectively) but more shared SBs (2 each).On the contrary, the InDels identified in RPL22, UBR5 and ACVR2A do not generate any top scoring neo-antigen (Fig. 2B and Suppl. Table 1B).

HLA-A haplotypes presenting the predicted mutated neoantigens

The stratification of the predicted mut-Ags based on the HLA-A haplotypes, showed a significant variability for both SNVs and InDel mutations.

Considering the SNV mut-Ags, the largest number are predicted to be associated with the A*34:01 (nr. 27), the other haplotypes are associated with 14 mut-Ags, on average, except the A*02:01 which is associated to only 3 neoantigens. However, selecting only those with the highest predicted affinity (100–200 nM and < 100 nM) (nr. 35), none of these are associated with the A*34:01. Indeed, these are predicted in association only with A*01:01 (nr. 2), A*03:01 (nr. 6), A*11:01 (nr. 12), A*31:01 (nr. 7) and A*33:03 (nr. 8). Furthermore, taking into consideration the top scoring prediction (< 100 nM and SBs) (nr. 15) the associated haplotypes are A*03:01 (nr. 3), A*11:01 (nr. 3), A*31:01 (nr. 3) and A*33:03 (nr. 6) (Fig. 3A; Suppl. Figure 2).

Fig. 3

Stratification of shared mut-Ags based on HLA-A haplotypes. The plots indicate the number of SNV mut-Ags (A) and InDel mut-Ags (B) for each of the selected HLA-A haplotypes, grouped by the level of binding (WB; SB)

Considering the InDel mut-Ags, the largest number are predicted to be associated with the A*31:01 (nr. 21), the other haplotypes are associated with 9 mut-Ags, on average. In this case, even selecting those with the highest predicted affinity (100–200 nM and < 100 nM) (nr. 20), the A*31:01 remains the haplotype associated with the largest number of mut-Ags (nr. 9). On the contrary, the A*01:01, A*02:01, A*26:01 and A*34:01 do not show any predicted mut-Ags. Furthermore, taking into consideration the top scoring prediction (< 100 nM and SBs) (nr. 4) the associated haplotypes are A*03:01 (nr. 1), A*24:02 (nr. 1) and A*31:01 (nr. 2) (Fig. 3B; Suppl. Figure 3).

Affinity map of individual predicted mutated mut-Ags to HLA-A haplotypes

The affinity to HLA-A haplotypes of each predicted SNV mut-Ags and InDel mut-Ags is shown, including those with single and multiple affinities (Tables 1 and 2).

Table 1 Affinity map of SNV mut-Ags predicted to bind HLA-A haplotypes

Table 2 Affinity map of InDel mut-Ags predicted to bind HLA-A haplotypes

Considering the SNV mut-Ags, the average affinity to the individual haplotypes is 3502.3 nM, ranging from 8513.05 nM (A*01:01) to 330.19 nM (A*31:01). In detail, the average affinity to A*03:01, 11:01, 31:01 and 33:03 is significantly higher than to the remaining haplotypes (p < 0.01) and only mut-Ags associated with these four haplotypes are SBs. Indeed 4 mut-Ags for each of the A*03:01, 11:01, 31:01 and 8 for A*33:03 are predicted. Looking at the same results from the protein perspective, the average affinity of mut-Ags to all haplotypes is 2501.76 nM, ranging from 7166.87 (IDH1) to 233.07 (BRAF) (Table 1 and Fig. 4A and B). None of the proteins showed mut-Ags with an overall average affinity significantly different from all other proteins. However, SBs are predicted for all the proteins with the exception of FGF3R, IDH1, PTEN and NRAS. In particular, 2 mut-Ag for each of BRAF, GNA11, GNAQ, SF3B1, 3 for KRAS and TP53 as well as 6 for PIK3CA (Fig. 2 and Suppl. Table 1A).

Fig. 4

Affinity of shared SNV and InDel mut-Ags to HLA-A haplotypes. The affinity to HLA-A haplotypes of all predicted mut-Ags (PB, WB and SB) are plotted. Values are shown according to each haplotype (A and C) or mutated protein from which the mut-Ags are derived (B and D). nM nanomolarity

Considering the InDel mut-Ags, the average affinity to the individual haplotypes is 2510.3 nM, ranging from 6573.08 nM (A*01:01) to 541.58 nM (A*31:01). In detail, also for these antigens the average affinity to A*03:01, 11:01, 31:01 and 33:03 is higher than to the remaining haplotypes, although it did not reach the statistical significance. SBs are predicted only for these haplotypes, namely 1 each of the A*03:01, 11:01, 24:02 and 6 for A*31:01. Looking at the same results from the protein perspective, the average affinity of mut-Ags to all haplotypes is 2676.42 nM, ranging from 4807.62 (UBR5) to 1460.54 (LARP4B) (Table 2 and Fig. 4C and D). None of the proteins showed mut-Ags with an overall average affinity significantly different from all other proteins. However, SBs are predicted only for the proteins JAK1 (nr. 2), LARP4B (nr. 4) and RNF 43 (Nr. 3). None is predicted for the ACVR2A, NFKBIE, RPL22 and UBR5 proteins (Fig. 2B and Suppl. Table 1B).

Targetable predicted mutated neoantigens linked to HLA-A haplotypes

In order to verify whether the predicted SNV and InDel mut-Ags are real neoantigens and may be actual targets for shared off-the-shelf cancer immunotherapy strategies, the differential agretopicity index (DAI) for each WB and SB antigens was calculated. In particular, a ratio > 10 of MHC affinity of the mutant peptide:the nonmutated counterpart is considered as meaningful [16,17,18,19,20].

According to such analysis, only 5 predicted SNV-NeoAgs and 15 InDel-NeoAgs are targetable because they show either a DAI > 10 or they are de novo protein sequences which do not have a match in the wt corresponding sequence (Table 3). Regarding the SNV-NeoAgs, 4 of the 5 (80%) are SB; 3 are associated with the HLA-A*33:03 (PIK3CA_H1047R YFMKQMNDAR, EYFMKQMNDAR and FMKQMNDAR). The PIK3CA_E542K AISTRDPLSK peptide binds to both A*03:01 and 11:01 (Table 3).

Table 3 WB and SB targetable Neo-Ags binding to HLA-A haplotypes

Concerning the InDel-NeoAgs, 8 of 15 (53.3%) are SB and are associated to the HLA-A*03:01, 11:01, 24:02 and 31:01. While the first three are associated with one SB each, the 31:01 is associated with 6 of them (75%). 50% of them (4 out of 8) are derived from the LARP4B_T163Hfs protein and are all associated with the B*31:01 haplotype. Finally, the majority of the targetable SB InDel-NeoAgs are associated with a single haplotype ad only the JAK1_K860Nfs QLKWTPHILK peptide is associated with two haplotypes (HLA-A* 03:01 and 11:01) (Table 3).

Prediction of mut-Ags associated with HLA-B haplotypes

Predicted mut-Ags associated with HLA-B haplotypes are identified in the same 11 proteins, characterized by SNVs, and in 7 proteins characterized by InDel. Considering the total number of both types of mut-Ags, 86.1% are scored PB, 4.7% are scored WBs and 9.1% SBs, on average. Moreover, the majority of WB and SBs (61.9%) are predicted to bind more than a single haplotype (Suppl. Table 2A and B).

Also for the HLA-B, The number and the score of the predicted SNV mut-Ags greatly varied among the 11 proteins. The largest number of WB and SB SNV mut-Ags are predicted in the 4 SNVs identified in PIK3CA (9 WBs and 3 SBs); however the two SNVs identified in SF3B1 provided the largest number of SBs (nr. 5). Overall, the SNVs generating the largest number of such mut-Ags (nr. 3) are the SF3B1_R625C, GNA11_Q209L and FGFR3_S249C. On the contrary, the SNVs identified in TP53, NRAS, KRAS, IDH1 did not generate any SB (Fig. 5A, Suppl. Table 2A).

Fig. 5

Predicted shared mut-Ags for HLA-B. The diagrams show the number of putative mut-Ags derived from SNV (A) and InDels (B). The antigens are grouped by the affinity to the HLA molecules (PB = poor binders; WB = weak binders; SB = strong binders)

Similar to SNVs, the number and the score of the predicted InDel mut-Ags greatly varied among the 7 proteins. In particular, the largest number of WB and SB InDel mut-Ags are predicted in the LARP4B_{T163Hfs_47} (4 WBs and 13 SBs). On the contrary, the InDels identified in RPL22 and ACVR2A did not generate any WB and SBs (Fig. 5B, Suppl. Table 2B).

HLA-B haplotypes presenting the predicted shared mut-Ags

The stratification of the predicted mutated mut-Ags based on the HLA-B haplotypes, showed a significant variability for both SNVs and InDel mutations.

Considering the mutated mut-Ags derived from SNVs, the largest number are predicted to be associated with the B*18:01 (nr. 18), the other haplotypes are associated with 10.3 mut-Ags, on average, except the B*13:01 and B*51:01 which are associated to only 3 mut-Ags. However, selecting only WB and SBs (nr. 34), only 2 are associated with the B*18:01 and they are almost evenly distributed among several B haplotypes. Furthermore, taking into consideration the SBs only (nr. 19), the B*27:05 haplotype is associated with largest number (nr. 4) (Fig. 6A and Suppl. Figure 4).

Fig. 6

Stratification of shared mut-Ags based on HLA-B haplotypes. The plots indicate the number of SNV mut-Ags (A) and InDel mut-Ags (B) for each of the selected HLA-B haplotypes, grouped by the level of binding (WB; SB)

Considering the InDel mut-Ags, the largest number are predicted to be associated with the B*58:01 (nr. 18), the other haplotypes are associated with 7 mut-Ags, on average, except the B*15:02, B*35:01 and B*35:05 which are associated to only 1 neoantigen. In this case, even selecting the WB and SBs (nr. 32), the B*58:01 remains the haplotype associated with the largest number of mut-Ags (nr. 15). Such finding is further confirmed also taking into consideration the SBs only, indeed 11/23 are associated with the B*58:01 haplotypes. On the contrary, most of the B haplotypes (20/27) do not show any predicted SB (Fig. 6B and Suppl. Figure 5).

Affinity map of individual predicted mutated mut-Ags to HLA-B haplotypes

The affinity to HLA-B haplotypes of each predicted mut-Ags from SNVs and InDels is shown, including those with single and multiple affinities (Tables 4 and 5).

Table 4 Affinity map of SNV mut-Ags predicted to bind HLA-B haplotypes

Table 5 Affinity map of InDel mut-Ags predicted to bind HLA-B haplotypes

Considering the SNV mut-Ags, the average affinity to the individual haplotypes is 4890.5 nM, ranging from 15780.5 nM (B*46:01) to 618.5 nM (B*15:01). In detail, the average affinity to B*15:01 and 27:05 is significantly higher than to most of the remaining haplotypes (p < 0.01) and most of the SBs are found associated with these two haplotypes (4 each out of 16 total) (Table 4). Looking at the same results from the protein perspective, the average affinity of mut-Ags to all haplotypes is 4131.3 nM, ranging from 7088.05 (NRAS) to 935.67 (PTEN) (Table 4 and Fig. 7A and B). None of the proteins showed mut-Ags with an overall average affinity significantly different from all other proteins. However, SBs are predicted for all the proteins with the exception of IDH1, KRAS, NRAS and TP53. In particular, 1 mut-Ag for GNAQ and FGF3R, 2 for PTEN and GNA11, 3 each for BRAF and, PIK3CA, 4 for SF3B1 (Fig. 5A and Supplementary Table 2A).

Fig. 7

Affinity of shared SNV and InDel mut-Ags to HLA-B haplotypes. The affinity to HLA-B haplotypes of all predicted mut-Ags (PB, WB and SB) are plotted. Values are shown according to each haplotype (A and C) or mutated protein from which the mut-Ags are derived (B and D). nM nanomolarity

Considering the neoantigens from InDels, the average affinity to the individual haplotypes is 6027.8 nM, ranging from 18,499.7 nM (B*46:01) to 825.9 nM (B*15:01). In detail, also for the InDel mut-Ags the average affinity to B*15:01 and 58:01 is higher than to the remaining haplotypes, although it did not reach the statistical significance. In particular, 12 out 22 SBs are predicted for the 58:01 haplotype (Tables 5). Looking at the same results from the protein perspective, the average affinity of mut-Ags to all haplotypes is 4776.42 nM, ranging from 6901.62 (RNF43) to 2833.9 (NFKBIE) (Table 5 and Fig. 7C and D). None of the proteins showed mut-Ags with an overall average affinity significantly different from all other proteins. InDel-NeoAgs SBs are predicted for all the proteins, except ACVR2A and RPL22, and the largest number (nr. 13) is predicted for LARP4B (Fig. 5B and Supplementary Table 2B).

Targetable predicted mutated neoantigens in HLA-B haplotype

The differential agretopicity index (DAI) was calculated also for each WB and SB mut-Ags linked to B haplotypes. According to such analysis, 7 predicted SNV mut-Ags and 30 InDel mut-Ags can be considered as NeoAgs and are targetable because they show either a DAI > 10 or they are de novo protein sequences which do not have a match in the wt corresponding sequence (Table 6). Concerning the SNV-NeoAgs, 4 of the 7 are SB (57.1%) and are associated with HLA-B*15:01 (PIK3CA_R88Q RQLCDLRLF and PTEN_R130Q GQTGVMICAY), 15:02 (SF3B1_R625C YVCNTTARAF), 27:05 and 39:01 (GNA11_Q209L FRMVDVGGL) haplotypes. Therefore, only the latter is associated with > 1 hapotype (Table 6).

Table 6 WB and SB targetable Neo-Ags binding to HLA-B haplotypes

Regarding the InDel-NeoAgs, 22 of the 30 (73.3%) are SBs and are associated to 9 of 27 HLA-B haplotypes analyzed in the study. Most of such haplotypes are associated with one or two SB InDel-NeoAgs, while the 58:01 is associated with 12 of them (54.5%). Interestingly, most of the SB InDel-NeoAgs (13 out of 27) are derived from the LARP4B_T163Hfs protein and 8 of them are associated with the B*58:01 haplotype. Finally, the majority of the targetable SB InDel-NeoAgs are associated with a single haplotype ad only the JAK1_K860Nfs SEKNQQLKW and VSEKNQQLKW, NFKBIE_Y254Sfs LTSSPRTETRW, RTETRWSTW, TSSPRTETRW peptides are associated with two haplotypes (Table 6).

Prediction of mutated mut-Ags associated with HLA-C haplotypes

Predicted mut-Ags associated with HLA-C haplotypes are identified in the same 11 mutated proteins characterized by SNVs and in 5 mutated proteins characterized by InDels. Indeed, no mut-Ags are predicted in the RPL22_K15Rfs*5 and NFKBIE_Y254Sfs*13 proteins (Fig. 8A, B). Considering the total number of both types of mut-Ags, 91.9% are scored PB, 2.2% are scored WBs and 5.8% SB binders, on average. Moreover, the majority of WB + SB antigens (92%) are predicted to bind more than a single haplotype (Suppl. Table 3A and B).

Fig. 8

Predicted shared mut-Ags for HLA-C. The diagrams show the number of putative mut-Ags derived from SNV (A) and InDels (B). The antigens are grouped by the affinity to the HLA molecules (PB poor binders, WB weak binders, SB strong binders)

Regarding the SNVs, the number and the score of the predicted mut-Ags greatly varied among the 11 proteins. However, WB and SB SNV mut-Ags are only 2 in SF3B1 (1 from the R625H and 1 from the R625C mutations) and 1 in BRAF_V600M. 2 of them are predicted to bind more than a single haplotype. All other SNVs did not generate WB and SB (Fig. 8A and Suppl. Table 3A).

Similar to SNVs, the number and the score of the predicted mut-Ags derived from the InDels greatly varied among the 5 mutated proteins. In particular, considering the WB and SB InDel-NeoAgs, the largest number of mut-Ags are predicted in the LARP4B_{T163Hfs_47} (4 WBs and 10 SBs). On the contrary, InDels identified in JAK1_K860Nfs*16 and ACVR2A_K437Rfs*5 did not generate any top scoring mut-Ags (Fig. 8B and Suppl. Table 3B).

HLA-C haplotypes presenting the predicted mut-Ags

The stratification of the predicted mut-Ags based on the HLA-C haplotypes, showed a significant variability for both SNVs and InDel mutations.

Considering the mutated SNV mut-Ags, the largest number are predicted to be associated with the C*06:02 (nr. 18), the other haplotypes are associated with 10 mut-Ags, on average, except the C*08:01 and C*15:02 which are associated only with 5 and 4 mut-Ags. Restricting the analysis, no WB and only 2 SB with the selected haplotypes are identified, specifically 1 for the HLA-C*03:02 and one for the HLA-C*05:01 (Fig. 9A and Suppl. Figure 6).

Fig. 9

Stratification of shared mut-Ags based on HLA-C haplotypes. The plots indicate the number of SNV mut-Ags (A) and InDel mut-Ags (B) for each of the selected HLA-C haplotypes, grouped by the level of binding (WB; SB)

Considering the mut-Ags derived from InDels, the largest number are predicted to be associated with the C*01:02 (nr. 24), the other haplotypes are associated with 13 mut-Ags, on average. The evaluation of WB and SB shows that 7 WB are associated with HLA-C*03:02, 03:03, 03:04, 03:05, 15:02 haplotypes. Focusing on the 15 SB, these are associated with HLA-C*01:02 (nr.1), C*03:02 (nr.2), C*03:03 (nr.4), C*03:04 (nr. 4), C*03:05 (nr.2), C* 07:02 (nr.1) and C*15:02 (nr.1) (Fig. 9B and Suppl. Figure 7).

Affinity map of individual predicted mutated mut-Ags to HLA-C haplotypes

The affinity to HLA-C haplotypes of each predicted mut-Ags from SNVs and InDels is shown, including those with single and multiple affinities (Tables 7 and 8).

Table 7 Affinity map of SNV mut-Ags predicted to bind HLA-C haplotypes

Table 8 Affinity map of InDel mut-Ags predicted to bind HLA-C haplotypes

The evaluation of the SNV mut-Ags showed that the average affinity to the individual haplotypes is 5165.8 nM, ranging from 11,951.2 nM (C*04:01) to 1562.5 nM (C*08:01). In detail, the average affinity to C*03:02 is higher than to the other remaining haplotypes but does not reach the statistical significance, except for C*01:02, 04:01 and 08:01 which show an average affinity significantly lower. SB SNV mut-Ags are identified only in association with HLA- C*03:02 (Nr. 2) and C*05:01 (Nr. 1) haplotypes (Table 7 and Fig. 10A). Looking at the same results from the protein perspective, the average affinity of mut-Ags to all haplotypes is 3909,6 nM, ranging from 6447.7 nM (KRAS) to 1580,06 nM (IDH1) (Table 7 Fig. 10B). None of the proteins showed mut-Ags with an overall average affinity significantly different from all other proteins. However, SB are predicted only for BRAF (Nr. 1) and SF3B1 (Nr. 2) (Fig. 8A and Suppl Table 3A).

Fig. 10

Affinity of shared SNV and InDel mut-Ags to HLA-C haplotypes. The affinity to HLA-C haplotypes of all predicted mut-Ags (PB, Wb and SB) are plotted. Values are shown according to each haplotype (A and C) or mutated protein from which the mut-Ags are derived (B and D). nM nanomolarity

Considering the mut-Ags from InDels, the average affinity to the individual haplotypes is 5158.7 nM, ranging from 13,262,8 nM (C*04:01) to 1299.1 nM (C*03:02) (Table 8 and Fig. 10C). In detail, the average affinity to HLA-C*03:02 is higher than to the remaining haplotypes but does not reach the statistical significance, except for C*04:01, 04:03 and 08:01 which show an average affinity significantly lower (Fig. 8B and Suppl. Table 3B). SB InDel mut-Ags are predicted for 01:02, 03:02, 03:03, 03:04, 03:05, 07:02 and 15:02 haplotypes, specifically 1 each of the C*01:02, 07:02 and 15:02; 2 for the C*03:02, 03:05; 4 for the C*03:03, 03:04. No SB InDel mut-Ags are predicted for the remaining HLA-C haplotypes.

From the protein perspective, the average affinity of mut-Ags to all haplotypes is 4698.5 nM, ranging from 7243.1 nM (JAK1) to 779.4 nM (LARP4B). LARP4B protein showed a number of predicted mut-Ags with an overall average affinity significantly higher from all other proteins. In detail, SB InDel mut-Ags are predicted for this protein (nr. 5) and RNF43 (nr. 2). None are predicted for the ACVR2A, JAK1 and UBR5 proteins (Table 8 and Fig. 10D).

Targetable predicted mutated neoantigens linked to HLA-C haplotypes

In order to verify whether the predicted WB and SB mut-Ags derived from both SNVs and InDels may be effective targets for shared off-the-shelf cancer immunotherapy strategies, the differential agretopicity index (DAI) was calculated. According to such analysis, none of the SNV mut-Ags shows a DAI > 10. On the contrary, the 10 InDel mut-Ags identified in LARP4B_T163Hfs (6 NeoAgs), RNF43_G659Vfs (3 NeoAgs) and UBR5_E2121Kfs (1 NeoAg) are potentially targetable because they show either a DAI > 10 (RNF43_G659Vfs RGVPPSPPL) or they are de novo protein sequences which do not have a match in the wt corresponding sequence (Table 9). Of these, 3 are WB and 7 SB. The latter are predicted in LARP4B_T163Hfs (5 NeoAgs) and in RNF43_G659Vfs (2 NeoAgs). Moreover, while LARP4B_T163Hfs YHRWIVTSM and SAYLGRTLLV and RNF43_G659Vfs ITPPVWHIL are predicted to bind a single HLA-C haplotype, all others are promiscuous binding to more than one haplotype (Table 9).

Table 9 WB and SB targetable Neo-Ags binding to HLA-C haplotypes

Total targetable predicted mutated shared neoantigens

A recapitulation of the targetable mutated shared neoantigens derived from the SNV and InDel mutations most frequently identified in cancer and associated to the haplotypes of the three major HLA alleles is performed.

Considering the SNVs, PIK3CA_H1047R may generate 5 NeoAgs associated to three different haplotypes, namely HLA-A*31:01 and 33:03 as well as HLA-B*27:05. All other SNVs may generate 1 NeoAg associated to a single haplotype, except for GNA11_Q209L FRMVDVGGL, PIK3CA_E542K AISTRDPLSK and SF3B1_R625C YVCNTTARAF which are predicted to bind two haplotypes. From the HLA perspective, the HLA-B*27:05 is associated with the highest number of SNV-NeoAgs (2 PIK3CA_H1047R and 1 GNA11_Q209L). All others are associated to 1 or maximum 2 SNV-NeoAgs, and in the latter case, these are derived from different mutations (Fig. 11; Table 10).

Fig. 11

Targetable shared SNV-NeoAgs. The shared NeoAgs with DAI > 10 (targetable SNV-NeoAgs) derived from each protein characterized by SNV mutation are plotted according to the associated HLA-A, B or C haplotypes. Indicated are the numbers of NeoAgs from each of the mutations

Table 10 Novelty of targetable SNV-NeoAgs

Considering the InDels, the number of targetable NeoAgs is much larger. Indeed LARP4B_T163Hfs may generate 30 NeoAgs, RNF43_G659Vfs 9, NFKBIE_Y254Sfs 7, JAK1_K860Nfs 5 and UBR5_E2121Kfs 4. In particular, the 30 InDel-NeoAgs derived from LARP4B_T163Hfs are associated with 14 different haplotypes, of which, 10 with the HLA-B*58:01, 4 with HLA-A*31:01 and 3 with each HLA-B*27:05, C*03:03, C*03:04, respectively.

Remarkably, the HLA-B*58:01 is associated with 16 InDel-NeoAgs (10 LARP4B_T163Hfs, 4 NFKBIE_Y254Sfs and 2 JAK1_K860Nfs). Second in the ranking is the HLA-A*31:01 associated with 8 InDel-NeoAgs (5 LARP4B_T163Hfs, 2 RNF43_G659Vfs and 1 NFKBIE_Y254Sfs). All others are associated with 1 or maximum 6 InDels-NeoAgs, and in the latter case, these are derived from different mutations (Fig. 12; Table 11).

Fig. 12

Targetable shared InDel-NeoAgs. The shared NeoAgs with DAI > 10 or de novo (targetable InDel-NeoAgs) derived from each protein characterized by InDel mutation are plotted according to the associated HLA-A, B or C haplotypes. Indicated are the numbers of NeoAgs, starting from 2, from each of the mutations

Table 11 Novelty of targetable InDel-NeoAgs

In order to verify whether the targetable shared NeoAgs described in the present study are novel, a search in different databases was conducted Considering the 12 targetable SNV-NeoAgs, 10 have been already reported in literature or covered by a patent [21,22,23,24]. The remaining 2 NeoAgs are novel, namely PIK3CAE545K QAMESEITK (HLA-A*11:01) and SF3BIER625C YVCNTTARAF (HLA-B*15:02) (Table 10).

Considering the 53 targetable InDel-NeoAgs, 13 antigens, 4 derived from JAK1_K860Nfs and 9 RNF43_G659Vfs, are already covered by a patent [24, 25]. The remaining 40 NeoAgs are novel, namely all those derived from LARP4B_T163Hfs, NFKBIE_Y254Sfs, UBR5_E2121Kfs and 1 derived from the JAK1_K860Nfs mutation, covering a broad range of haplotypes in the HLA-A, B and C alleles (Table 11).

Potential clinical application of mutated shared NeoAgs

The subsequent step was to verify whether the targetable SNV and InDel-NeoAgs may have a potential clinical application in cancers. The mutated proteins from which such NeoAgs derive, are identified in a number of tumor types at a frequency > 5%. This ranges from 7% in Bladder&Prostate (BLCA) cancers to 50% in Uveal melanoma (UVM). Notably, tumor types with a 5-year overall survival < 10%, namely pancreatic (PAAD) and stomach (STAD) cancers, are characterized by mutated proteins generating targetable shared NeoAgs, which can be useful in 30 and 24% of patients, respectively (Fig. 13).

Fig. 13

Frequency of targetable shared NeoAgs in cancers. The distribution and percentage in cancers of the mutated proteins generating targetable shared SNVs and InDel-NeoAgs is shown. The percentage (blue) indicates the sum of the percentages for each mutated protein

Considering that the targetable shared SNV and InDel-NeoAgs are associated with haplotypes showing a uneven prevalence in different populations, vaccines based on such NeoAgs may have a global or a regional application. As example, an off-the-shelf cancer vaccine for colorectal (COAD), stomach (STAD) and uterus endometrial (UCEC) cancers may have a global coverage. Indeed, the SNV and InDel-NeoAgs derived from their tumor-specific mutated proteins are associated with haplotypes, including HLA-A*02:01, 03:01 and 11:01, which collectively cover > 10% of populations all over the globe. On the contrary, off-the-shelf cancer vaccines for pancreatic (PAAD) cancer or uveal melanoma (UVM) may be developed only for a strict regional coverage. Indeed, the SNV and InDel-NeoAgs derived from their tumor-specific mutated proteins are associated with HLA-A*03:01, which covers > 10% of populations only in Europe (LUAD and PAAD) or with HLA-A*33:03 and B*15:02, which covers > 10% of populations only in Oceania and South-East Asia (UVM) (Fig. 14).

Fig. 14

Coverage of off-the-shelf cancer vaccines based on targetable shared NeoAgs. The geographic applicability of off-the-shelf cancer vaccines based on targetable shared NeoAgs is shown according the HLA haplotypes frequency in distinct world populations. The size of the connecting ribbons correlates with the percentage of the specific SNV or InDel mutation in each tumor type (left). The height of the bar indicates the total percentage of cancer types targetable in each Continent, based on the HLA haplotype prevalence (right)

Discussion

The SNV and InDel mutations reported in the TCGA database as those identified in at least 5% of cancer patients are selected for predicting shared mutated neoantigens (SNV and InDel-NeoAgs). 50 haplotypes in the three MHC class I loci (10 HLA-A, 27 HLA-B and 13 HLA-C) were selected to cover > 80% of the global population.

Out of the 100 + mutated proteins falling in the selection parameter, only 18 proteins were identified to generate mutated antigens (mut-Ags) in association with one or more of the selected haplotypes. In particular, 11 proteins are characterized by single nucleotide variations (SNV), leading to a single aminoacid point mutation and 7 proteins are characterized by insertion/deletions (InDel), leading to a frameshift with an alternative de novo codon translation. Overall, 568 SNV and 502 InDel mut-Ags are predicted from all the 18 mutated proteins in association with the 50 haplotypes; however, only 72 (12.7%) SNV and 79 (15.7%) InDel show high affinity to the selected haplotypes (WB and SB). Furthermore, taking into consideration only the predicted mut-Ags with the highest affinity to the haplotypes (< 100 nM = SBs), only 42 SNV (7.4%) and 51 InDel (10.1%) are identified. Interestingly, SNVs in NRAS, IDH1 and PTEN as well as InDels in NFKBIE, RPL22, UBR5 and ACVR2 proteins do not generate SBs in any of the three HLA alleles. The biological explanation for this observation requires further experimental assessments. In the quest of targetable shared neoantigens (NeoAgs) from the WB and SBs, only 12/72 (16.7%) mut-Ags derived from SNV mutations and 55/79 (69.6%) mut-Ags derived from InDel mutations can be defined as neoantigens. Indeed, they either show an affinity to the same MHC > tenfold compared to the corresponding nonmutated counterpart (differential agretopicity index (DAI) > 10) or derive from de novo protein sequences which do not have a match in the wt corresponding sequence. In particular, the latter derive from “abnormal” mRNAs generated by the frameshift, contain premature termination codons (PTCs) which may be recognized and degraded by nonsense-mediated mRNA decay (NMD) or undergo a translational repression [26,27,28,29]. Nevertheless, InDel-NeoAgs have been identified by MS/MS on a set of tumor cells and their immunogenicity has been proven by ex vivo stimulation of PBMCs from both healthy donors (HD) as well as tumor patients [30]. Therefore, both SNV-NeoAgs and InDel-NeoAgs are presented by tumor cells and recognized by the immune system as “non-self” antigens representing potent immunogenic targets.

The PIK3CA_H1047R SNV mutation encodes the largest number of SNV-NeoAgs (nr. 5) providing a set of potential shared mutated neoantigens for developing off-the-shelf cancer vaccines targeting 12% of breast cancer (BRCA), 7% of uterine carcinosarcoma (UCS) as well as 5% of Uterine Corpus Endometrial Carcinoma (UCEC) patients.

Such SNV-NeoAgs are associated with three HLA haplotypes, namely HLA-A*31:01, 33:03, present in > 10% of the population in South America, North-East and South-East Asia, and HLA-B*27:05, present in < 5% only in Europe. All the other 7 SNV mutations encode a single NeoAg each.

The LARP4B_T163Hfs InDel mutation encodes for the largest number of InDel-NeoAgs (nr. 29) providing a set of potential shared mutated neoantigens for developing off-the-shelf cancer vaccines targeting 5.3% of stomach cancer (STAD) patients.

Such SNV-NeoAgs are associated with several HLA haplotypes, covering most of the Continents. The other 4 InDel mutations encode 3 to 9 NeoAgs each.

The HLA haplotypes more frequently associated with both SNV and InDel-NeoAgs are the A*31:01 (South America, North-East and South-East Asia), B*58:01 (South-East Asia) and C*03:03/03:04 (North-East and South-East Asia), considering the distribution in > 10% of the population.

The length of all predicted NeoAgs spans from 8 to 11 aa, with a predominance of 9mer (36/67) which is the most frequently identified in the peptides naturally presented by MHC class I molecules [30]. Of these, seven 9mer are among the SNV-NeoAgs and 29 are among the InDel-NeoAgs and they should be considered as the ones with the highest probability to be expressed by cancer cells. Only the shared NeoAg derived from PIK3CA_E542K (AISTRDPLSK) have been previously experimentally validated [22,23,24]. For the remaining predicted NeoAgs, although the highly stringent affinity values applied in the present study strongly suggest the natural presentation by cancer cells, a definitive experimental validation is required.

Overall, considering the mutations encoding the SNV and InDel-NeoAgs as well as the associated HLA alleles, our findings provide the most comprehensive set of immunologically relevant shared mutated neoantigens for development of cancer vaccines. In particular, they may have a global reach or be more appropriate for regional applications, depending on the frequency of the relevant haplotypes in different populations.

Specifically, colorectal cancer (COAD), stomach cancer (STAD), and uterine endometrial cancer (UCEC), are characterized by mutations encoding NeoAgs associated with a broad range of different haplotypes. Consequently, a universal "off-the-shelf" vaccine for these tumors can be developed. On the contrary, pancreatic cancer (PAAD), and uveal melanoma (UVM) are characterized by mutations encoding NeoAgs associated with haplotypes prevalent in certain ethnic or regional groups only. Therefore, a "regional" vaccine approach may be considered.

In the present study, the analysis has been focused only on MHC-class I neoantigens, which are the final effective target of CD8⁺ T cells cytotoxic effect. Nevertheless, both the SNV and InDel mutations might likely generate MHC-class II neoantigens which could elicit mutation-specific CD4⁺ T helper cells to potentiate the CD8⁺ CTLs. However, such analysis requires a subsequent follow up with integrated approaches, considering that prediction tools for MHC-class II epitopes are not as robust as those for the MHC-class I epitopes.

The identification and validation of shared mutated neoantigens, represent a key finding for the development of "off-the-shelf" cancer vaccines, which could completely change cancer treatment worldwide. The next step will be to develop a validation platform to experimentally prove that the predicted shared mutated neoantigens are identified in the ligandome of tumor cells and are truly immunogenic. Indeed, SNV-NeoAgs could be poorly immunogenic if the corresponding wt is tolerogenic and InDel-NeoAgs could be poorly represented in the cancer ligandome.

In conclusion if confirmed, such vaccines would provide a fast, cost-effective solution to the challenge of immunotherapy, enabling treatment to be quickly employed across various patient populations. This would help to address the global cancer burden by providing new options for patients who may not have access to personalized therapies or for cancers with low tumor mutational burden (TMB) and a low number of unique neoantigens.