Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) causes coronavirus disease 2019 (COVID-19), a disease that has now become a global pandemic. The steps following infection include a first phase during which native immunity mechanisms and human leukocyte antigen (HLA) class I molecules (of the A, B, C genes) combat the virus by recruiting cells entered by the virus which is then fragmented via proteasomal cleavage to 9–13 amino acid (AA) fragments. HLA class I genes code for cell-surface glycoproteins are expressed on nucleated cells and present on the cell surface those antigen peptides to CD8⁺ cytotoxic T cells to signal cell destruction, thus eliminating infected cells. Thus, HLA class I restricted processing and presentation alerts the immune system to any infectious processes unfolding intracellularly and provides potential targets for a cytotoxic T cell response. During the second phase of response to the viral infection, HLA class II molecules are involved to initiate the production of specific antibodies against the virus. HLA class II molecules (of the HLA-DR, -DQ and -DP genes) are expressed on professional antigen-presenting cells (e.g., macrophages, dendritic cells) and present endocytosed extracellular antigen peptides to CD4⁺ T cells to promote B-cell mediated antibody production and immune memory. The HLA region in chromosome 6 is the most highly polymorphic in the human genome resulting in considerable individual and population variability in HLA composition, reflecting the long evolutionary history of exposure to and dealing with elimination and ultimate protection from various pathogens [1, 2].

In a previous study [3], we reported on the relations between SARS-CoV-2 virus spike glycoprotein (S protein) and 66 common HLA class II alleles, investigated using an in silico approach [4] by assessing the binding affinity of epitopes of the S protein to these most commonly occurring alleles (frequency ≥ 0.01). In the current study, we employed the same sliding epitope window methodology to exhaustively scan the entire S protein and open reading frame 1ab (ORF1ab) protein and determine the binding affinity of each n-mers (n = 10) AA epitopes to 61 common HLA class I alleles.

The main objective of this study was to exhaustively assess the binding affinities of HLA class I molecules to the SARS-CoV-2 S protein and the ORF1ab polyprotein. For that purpose, we assessed the binding affinities of 61 common class I alleles as described below.

In the current study, we investigated the relations between two important proteins of the SARS-CoV-2 proteome (the S protein, which is the main antigenic molecule of the virus [9] and ORF1ab polyprotein), and 61 common HLA class I alleles. There were three major aims in this in silico study, as follows. First, we sought to quantify the binding affinity of suitable fragments (linear epitopes) of the S protein and ORF1ab polyprotein to the 61 HLA class I alleles in an exhaustive manner using the sliding epitope window approach [3, 4]; second, we aimed to assess differences in allele binding affinities between the two proteins, among the three HLA class I genes, and their interaction; and third, we sought to assess the distribution of binding affinities across the sequences of the two proteins. The results obtained were clear cut. First, all 61 alleles exhibited good binding affinities to epitopes of both the spike and ORF1ab proteins; second, the percentage of good binding affinities (LPR < 1) did not differ significantly between the two proteins (Figure 2) but did so among the genes, with gene A having the best binding performance, followed by gene B and gene C (A > B > C) (Figure 3); third, there was no significant interaction between proteins and genes (ANOVA), indicating a similar effect of the genes in both proteins (Figure 4); and fourth, good binding affinities were distributed throughout the two protein sequences (Figure 5 and 6). These findings indicate that the HLA class I system is well suited to contribute effectively to the elimination of SARS-CoV-2 at the early phase of its entry to the body in otherwise immunocompetent individuals. The fact that good binding affinities are distributed throughout both protein sequences provides for robustness in the fight against the virus, since it almost guarantees a good possibility that linear epitopes of the proteins can bind to class I molecules irrespective of the location of glycoprotein cleavage by cell proteases.

With respect to the ORF1ab polyprotein, it should be mentioned that CD8⁺ T cells recognize peptides presented by class I HLA surface receptors, and such peptides most commonly originate from cytosolic viral proteins following proteasomal degradation by host intracellular antigen processing pathways. Since T cells do not recognize surface antigens, in contrast to B cells and the antibodies that are produced and secreted by them, ORF1ab provides intracellular peptides that originate from essential enzymes and are thus more conserved. Proteins encoded by SARS-CoV-2 ORF1ab are more conserved among coronaviruses relative to the S protein, which has lower homology [10, 11]. ORF1ab was included in this study because it is the largest viral gene in the Coronaviridae family and encodes both non-structural and accessory proteins that are less susceptible to evolutionary pressure and hence mutational changes, in contrast to the S protein.

Nguyen and colleagues [12] have published a similar in silico analysis of viral peptide binding affinities to HLA-A, -B and -C loci. In their study, the entire SARS-CoV-2 proteome was investigated, whereas our study focused on the ORF1ab polyprotein and S protein, which together comprise a considerably smaller number of potential n-mers epitopes. Furthermore, sequence homology was used to identify highly conserved sequences in SARS-CoV-2 from other common coronaviruses that could function as epitopes. Here, we did not specifically investigate for sequence homology of the S protein or ORF1ab polyprotein with other coronaviruses as this has been documented in previous studies [10, 11]. Interestingly, Nguyen and colleagues [12] report that the HLA-A and HLA-C alleles exhibit the highest and lowest capacities to present SARS-CoV-2 antigens, respectively. This ranking is corroborated in our study (Figure 3) focusing on multiple proteomic elements (the S protein and the ensemble of intracellular proteins encoded in ORF1ab) as opposed to the entire viral proteome. With respect to specific alleles with the highest and lowest binding affinities to SARS-CoV-2 antigens, our results differ from those of Nguyen et al. [12] who report that B*15:03 is a better binder than B*46:01. This difference could be due to the smaller number of n-mers that we used (8,354 10-mers peptides that linearly partitioned the ORF1ab polyprotein and S protein vs. 32,257 peptide samples from the full SARS-CoV-2 proteome used by Nguyen and colleagues [12]).

Nelde and colleagues [13] recently published a study in which they identified HLA class I and HLA-DR binding peptides using experimental and computational methods. Their prediction workflow incorporates the NetMHCpan algorithm which was used in our study and reports immunogenic SARS-CoV-2-derived class I T cell epitopes. Among these are two 9-mers, LTDEMIAQY (HLA-A*01) and QYIKWPWYI (HLA-A*24), both of which have very high binding affinity scores in our analysis as well. The authors [13] used ELISA-type assays with in vitro amplified T cells from patients convalescing from SARS-CoV-2 infection and donors who were never exposed to the virus. Nelde et al. [13] reported that 29% of SARS-CoV-2-derived HLA class I binding peptides were validated as naturally occurring T cell epitopes, showing that NetMHCpan is useful as a predictive algorithm for class I immunogenicity.

To the best of our knowledge, there have been only a limited number of published studies (by Nguyen et al. [12] and Nelde et al. [13]) that investigated viral protein binding affinity across a wide range of HLA alleles on a per-allele basis using predictive algorithms. Our present study builds on previous work [3, 4] and examines the relationship between two viral proteins of the SARS-CoV-2 proteome and HLA class I antigen presentation, with the results discussed above. Finally, we devised metric (Q^P) to calculate the overall binding affinity of a gene in a specific population by taking into account both the goodness of binding affinities of alleles of a gene and their frequency of occurrence in the population. Using the global allele frequencies of HLA class I alleles (Table 1), we found that the HLA class I A gene had the best overall binding performance, followed by genes B and C. It is possible that this result could differ for different populations (e.g., ethnicities, locations on Earth, etc.), thus providing a population-specific measure of how well the HLA class I makeup of that population could contribute to SARS-CoV-2 elimination at the early stage of fighting the viral infection.

We acknowledge, however, an important limitation to our work. This study was performed exclusively in silico and thus the data presented are subject to any constraints that may affect HLA binding affinity predictive tools (NetMHCpan EL in this case). Although important to consider, this limitation is mitigated by the fact that NetMHCpan and other binding affinity prediction algorithms included in the IEDB suite are trained using experimental datasets. In absence of clinical and patient data regarding actual Covid-19 cases, we are unable to assess the impact of disease-modifying risk factors such as age and clinical comorbidities [14, 15] on HLA class I-conferred neutralization of the SARS-CoV-2 virus. In spite of this limitation, we believe that computational studies centered around binding affinity prediction can offer important insights as to how HLA genotype affects viral susceptibility, which can hopefully guide vaccination strategies.