SARS-Cov-2 is a single-stranded RNA virus composed of four main structural and 16 non-structural proteins encoded by at least 6 open reading frames contained within genomic and subgenomic RNA regions (1–3). Among the spike (S), membrane (M), envelope (E), and nucleocapsid (N) structural proteins, a key biological and pathogenetic role is played by the trimeric spike glycoprotein (the main component of the available vaccines), which is responsible for the attachment of the virus to host receptors and is the main target of neutralizing antibodies that can block virus entry into target cells (4–6). The spike protein protrudes as a homotrimer from the viral surface in a metastable ‘prefusion’ state (7–9). Each monomer is composed of two subunits: S1, which is primarily involved in host cell recognition via the RBD with structural support by the N-terminal domain (NTD), and the S2 subunit, which is mainly comprised of α-helixes, such as the heptad repeat 1 and 2 helixes (HR1 and HR2) and the central helix (CH) (see below for further details) (7). Additional S1 subunit modules are the C-terminal (CTD-1 and CTD-2) domains, which are involved in key intermolecular contacts with the S2 subunit within the trimeric S structure. In fact, upon interaction with the host cell surface, the primed S1 subunit is shed and the S2 subunit undergoes a dramatic conformational change that promotes the transition to the ‘postfusion’ state, ultimately leading to viral fusion and cell entry, mediated by the S2 fusion peptide (10).

As all RNA viruses, SARS-Cov-2 is prone to mutations, but its mutation rate is restrained by the proof-reading activity of an exoribonuclease (non-structural protein 14) that substantially reduces mistaken nucleotide incorporation into nascent RNA molecules (11, 12). Despite this proof-reading activity, SARS-Cov-2 remains capable of accumulating mutations, that upon selection and subsequent fixation can interfere with virus recognition by the immune system, thus compromising immune-protection. New mutations tend to be fixed either because of the enhanced infection and transmission capacity they confer and/or because they allow variant viruses to evade control by neutralizing antibodies and cytotoxic CD8 T cells, with preferential elimination of the parental virus and selection of the mutated strain. Both mechanisms are thought to be causally involved in the selection of SARS-Cov-2 variants (the so-called Variants of Concern – VoC - and Variants of Interest - VoI). Indeed, some of these variants become more infectious and can spread more quickly due to a mutationally acquired enhancement of the binding affinity between the spike protein Receptor Binding Domain (RBD) and Angiotensin-Converting Enzyme 2 (ACE2), the main virus receptor exposed on the surface of target host cells. The same mutations can also allow the virus to escape neutralization by circulating anti-spike antibodies, produced in the context of a humoral immune response.

The possibility of a complete virus escape, however, is theoretically limited by the polyclonality and multispecificity of the antibody response (13). This would obviously imply that different spike regions are simultaneously recognized by polyclonal anti-spike antibodies of different specificity, elicited by the virus (or by a vaccine) in each patient. Moreover, a variable proportion of the elicited antibodies may be directed against largely invariable spike regions, i.e., protein regions whose mutational change would severely impair the fitness of the virus (14–18). Thus, loss of immune-recognition of a certain spike region caused by a mutational event, should, in principle, be compensated by the persistent recognition of the virus by antibody molecules directed against spike regions that have not changed or that are intrinsically not permissive to amino acid substitutions because of structural or functional constraints. This concept is only partially confirmed by recent evidence suggesting that wild type spike-induced antibodies in vaccinees and convalescent patients have a diminished neutralizing capacity in vitro against some of the most recent spike variants (19–21), an immune-reactivity reduction that, however, does not appear to be strong enough to cause a complete loss of anti-viral protection in vivo (22–25). To this persistence of protection in vivo may also contribute a process of antibody maturation, which has been reported to increase the neutralizing potency and breadth of protection of the antibody response (26–29). As we will discuss in detail in this review, a persisting, vaccine-stimulated CD8 T cell effector function is also potentially capable of compensating the negative effect of escape spike mutations on antibody responses.

In addition to circulating antibodies, an important role in antiviral protection is played by cytotoxic CD8 T cells. Although not capable of preventing infection by neutralizing free viral particles and blocking their entry into host cells, CD8 T cells can avoid the spread of infection by eliminating infected cells through their cytotoxic activity and by purging intracellular virus through non-cytolytic mechanisms mediated by antiviral cytokine secretion at the site of infection (30–32). The importance of CD8 T cells is further indicated by the recent observation that SARS-Cov-2 spike protein can induce cell fusion and can exploit this mechanism for cell to cell spread, escaping antibody neutralization (33, 34). Data derived from infected patients indicate that CD8-mediated control of infection should be considerably less affected by mutational loss of immune reactivity, because CD8 T cell responses are broadly multi-specific (35–38). This makes mutation-mediated virus escape from CD8 T cell control much more unlikely than virus escape from neutralizing antibodies, meaning that even in case of a decline or loss of antibody neutralizing activity against SARS-Cov-2 variants, the persisting activity of cytotoxic CD8 T cells can likely limit or prevent a severe evolution of infection. Such a role of CD8 T cells in SARS-Cov-2 control is further corroborated by evidence indicating that T cell memory responses have an intrinsic capacity to persist for a very long time (up to 17 years after SARS-Cov-1 infection) (36), remaining detectable well after antibody waning. Additional support to a prominent CD8 T cell-mediated protective activity is provided by available sequencing and immunological data (39–43).

Thirty-one amino acid substitutions and three amino acid deletions arising from multiple selected mutations have been identified by comparing the in silico translated spike polypeptide sequences derived from the UK (Alpha), South-African (Beta), Brazilian (Gamma), Californian (Epsilon), Indian (Delta) VoCs, as well as Nigerian (Eta) and New Yorker (Iota) VoIs with the sequence of the original Wuhan SARS-Cov-2 strain (GISAID data base - https://cov.lanl.gov/content/index) (44–55). Twenty-three of these amino acid changes are located in spike regions that encompass recently described CD8 epitopes and account for a total of 35 CD8 T cell epitopes with mutated sequences (56–62). The remaining 11 variant mutations, instead, do not affect any known CD8 epitope. The linear distribution of CD8 epitopes-containing variant mutations along the spike polypeptide sequence and the spatial location of each mutation within the 3D structure of a spike monomer are shown in Figure 1 in the context of a single spike monomer.

A number of CD8 T cell epitopes (n=133), presented to CD8 T cells by different HLA class I alleles, has been reported to be located within unmutated spike regions (56–74). They are characterized by different levels of immunodominance and endowed with different CD8 T cell stimulatory capacities, as deduced from the intensity of CD8 responses measured by functional assays [e.g., TCR-dependent Activation Induced Marker (AIM) and Elispot assays] and through the assessment of the frequency of circulating, multimer-positive SARS-Cov-2 specific CD8 T cells (56–74). In particular, 85 of these epitopes are recognized by CD8 T cells in association with the most highly represented HLA-class I alleles (frequency > 5%, as calculated by their median expression value in the world population), namely HLA A*01, A*02, A*03, A*11, A*24, A*26, A*31, A*68, B*07, B*08, B*15, B*35, B*40, B*44, B*51, whereas 13 additional epitopes are presented to CD8 T cells by less represented HLA-class I alleles, such as HLA A*23, A*29, A*30, A*32, B*53, B*57 (Figure 2A). Importantly, the remaining 35 epitopes described so far share a degenerate HLA class I presentation, and can be recognized by CD8 T cells in the context of different HLA-class I alleles (Figure 2B). Finally, available data indicate that different epitopes can be simultaneously recognized by individual patients confirming the multi-specificity of CD8 T cell responses (56).

To gain a deeper insight into the effect of SARS-Cov-2 mutations on CD8 T cells responses, we analyzed 1,400,000 million spike protein sequences retrieved (as of June 1^st 2021) from the daily updated GISAID database. Considering the Shannon entropy (SE; a measure of variability of genetic mutations (75) for each amino acid position, calculated on a multiple alignment of spike protein sequences), we identified eight additional and potentially significant amino acid changes (SE values > 0.05, corresponding to a frequency of variation of more than 1% with respect to the Wuhan spike sequence). Four of these changes are located in regions spanning previously described CD8 T cell epitopes (S98F, A222V, A262S, T732A), while the other four (N439K, T478K, Q675I/R, K1191N) map to spike regions where no CD8 T cell epitopes have been described so far.

We then assessed if some of the epitopes that span spike sequences which have thus far been spared from mutation are actually located within spike regions unable to tolerate any sequence variation because of their essential role for proper structure/function of the virus. By referring to the two subunits of the spike protein, a significantly lower sequence variation was found to be associated to the S2 fusion subunit (Avg SE= 6.2x10e-3) compared to the S1-RBD-containing subunit (Avg SE=1.3x10e-2). This fits with the key role played by the latter subunit in the recognition of host cell receptors (ACE2, but also neuropilin and some newly identified membrane lectins) (76, 77) through the RBD and the NTD domains as well as with its role as the main target of neutralizing antibodies. Mutations that increase the binding affinity of the S1 subunit for its receptors and/or enable escape from neutralizing antibodies are, in fact, instrumental to virus propagation and highly prone to selection and fixation. Conversely, the less exposed S2 subunit, which upon S1 dissociation undergoes a large conformational change ultimately leading to viral-host cell membrane fusion, is a less likely candidate for viral evolution. This differential sequence conservation is also in keeping with the amino acid sequence identity (S2>S1) shared by the spike subunits of different SARS-CoV-2 variants and of more distantly related SARS-CoV viruses. Also in the case of this broader-range comparison, the S2 subunit appears to be considerably less variable (89% identity) than the S1 subunit (65% identity).

Using SE values of 0.0025, 0.001 and 0.0005 as decreasing entropy thresholds, we then arbitrarily defined three levels of amino acid (AA) residue conservation, which allowed us to tag different spike regions as ‘conserved’, ‘highly conserved’ and ‘hyper conserved’, respectively (see Figures 3, 4A, where epitopes classified as above are color-coded in yellow, red and purple, respectively).

Interestingly, 37 previously reported CD8 epitopes, varying in length from 8 to 12 AA, plus four additional putative epitopes identified with NetMHC-4.0 (79) and characterized by probability scores comparable to or higher than those of experimentally validated epitopes, all map to structurally constrained regions of the spike protein (Figure 4B). Although some of the 37 immunogenic sequences are extensively overlapping (e.g., the FERDISTEI and FERDISTEIY sequences both starting at AA 464; Figure 3), and thus likely correspond to the same core epitope, a sizeable number of non-redundant CD8 epitopes appears to be located within conserved spike structural elements, whose mutation would compromise SARS-CoV-2 functionality (Figures 3, 4). The distribution of CD8 T cell epitopes within spike regions of different sequence conservation with indication of their main features is reported in Table 1.

Although the CD8 epitopes discussed in this section have originally been identified in infected and convalescent patients, available studies in vaccinees (43, 80–83) confirm a substantial enrichment of CD8 T cell epitopes within the S2 domain, which contrasts with the lower CD8 epitope representation found in the S1 N-terminal domain (83).

We then took advantage of multiple spike structures (6, 10, 84, 85) to precisely map specific CD8 T cell epitopes within functionally and structurally distinct regions of the spike protein. In fact, a number of predictably invariant CD8 T cell epitopes map to specific spike regions with different degrees of sequence and structural conservation such as the RBD core, the carboxy-terminal 1 (CTD1) and 2 (CTD2) domains within the S1 subunit and throughout the S2 subunit (Figure 4B).

Even the S1 subunit N-terminal domain (NTD), which holds approximately 50% of all spike VoC mutations and represents the only site of amino acid deletions so far (Figure 1), contains a few peptide segments that are significantly conserved (e.g., AA 34-48 and AA 119-137) (Figures 3, 5A). The conserved AA 34-48 sequence, in particular, is close to the CTD1 of the adjacent protomer. This proximity region, which modulates spike conformational dynamics and is positionally shifted upon RBD binding to the ACE2 receptor, contains three epitopes (starting AA positions: 35, 36 and 41), two of which can be presented to CD8 T cells by more than one allele (AA 35-44 and 41-49).

A region of more marked structural/functional constraining within the S1 subunit, is the RBD (AA 328-533), a key domain for ACE2 receptor engagement, which is specifically mediated by the receptor-binding motif (RBM; AA 437-508) and only takes place when the spike protein is in the ‘up conformation’. Despite the sustained average mutation rate of this domain, there is a subset of conserved peptides located within the RBD core β-structure, part of the flanking regions, and the hinge loops that connect the RBD to the CTD1 (Figure 5B). These regions are key to the ‘up’ to ‘down’ transition. In fact, they control the dynamic interaction between the RBD and the NTD domains of adjacent protomers and are involved in the sensing and transduction of specific conformational inputs from the S2 subunit (see Figure 5B, where the RBD is shown in the most interaction-competent, ‘down’ conformation). Eleven CD8 epitopes map to these conserved RBD regions and some of them can be presented to CD8 T cells by the most highly represented HLA-class I alleles (see Figures 2, 3 for specific epitope ranking). Interestingly, the AA 456-474 RBM sub-segment that interacts with the NTD of the adjacent protomer is characterized by a very low mutation rate and harbors at least three CD8 epitopes (one of which degenerate and with high predicted MHC-I binding affinities) comprised between amino acid residues R454 and Y473 (Figure 5B). In contrast to the high mutability of the upper portion of the RBM, this RBM sub-segment, which is laterally positioned with respect to the main body of the RBD, as well as other RBD flanking regions are highly conserved and are specifically targeted by some class 3 and 4 neutralizing antibodies (86). Of note, S309, a class 3 cross-neutralizing antibody initially isolated from a SARS-CoV patient, is directed against an N343 glycan that maps to a conserved RBD flanking region (AA 331-345) and its RBD footprint partially overlaps two CD8 epitopes (AA 328-338 and 339-347) (87). Likely due to the structural/functional constraints (and presumed sequence invariance) of its RBD target, the neutralizing activity of S309 has proven to be resilient to a number of SARS-Cov-2 mutations, so far, and is considered a very promising candidate for a broadly protective monoclonal antibody therapy (88). Because of the co-existence of the B-cell epitopes targeted by the most potent neutralizing antibodies and of multiple CD8 epitopes within a conserved region, the RBD lends itself as a key site for the development of artificially multimerized, multi-functional (humoral and cellular immunity) pan-coronavirus vaccines.

A subset of conserved CD8 epitopes are located within the CTD2 domain, which includes the hyper conserved AA 592-602 sequence (Avg SE=1.5 x10e-4) and the slightly more variable AA 589-591 sequence (Avg SE=8.8x10e-4). A CD8 T cell epitope (AA 588-597) is close to and partially overlaps these highly conserved sequences, while another one (AA 660-670) is located in the smaller β-sheet of CTD2 (Figure 5C). The structurally dynamic and highly flexible 630-loop (AA 610-630), which is also located in this region, is crucial for the crosstalk between the S1 and the S2 subunits. When conformationally ordered, this loop interacts with the above mentioned highly conserved sequences through hydrophobic interactions and stabilizes the CTD2. This interaction, which enhances spike stability, thus preventing premature dissociation of the S1 subunit and the concomitant loss of spike functionality, plays a key role in the fine-tuning of the RBD up to down transition (84, 89). Indeed, the D614G mutation increases S1 stability and by stiffening the 630-loop, it strongly stabilizes the whole spike protein (84). Another CD8 epitope (AA 583-592) maps to the C-terminal portion of CTD1 (Figure 5C). This domain acts as a structural relay connecting the RBD to the S2 subunit through the so-called Fusion Peptide Proximal Region (FPPR), which, when conformationally ordered, can clamp down the entire RBD (10) (see Figure 5C).

By far the strongest sequence conservation entails the S2 subunit, where the central helix (CH) region and the helixes located at the N- and the C-terminal ends of heptad repeat 1 (HR1) appear to be the most conserved (Figure 5D). In keeping with its extremely low sequence variation, the S2 subunit holds the highest number (a total of 20) of putatively conserved CD8 epitopes, many of which belong to the top-scoring classes of predicted MHC-1 affinities (Figures 3, 5D). In particular, starting from the AA 989-1000 region, at least three CD8 T cell epitopes map to the nearly invariant AA 987-1005 sequence of the CH region (Avg SE=1.1x10e-4). This region is located in the innermost portion of the trimeric spike protein and serves as a pivot point during S2 transition from a prefusion to a postfusion conformation (Figure 5D). Critical for this transition is the flexibility of the hinge-loop located on the top of the HR1 and the CH helices. In fact, stiffening of this loop by Pro substitution of K986 and V987 (a modification introduced in many of the present spike-based vaccines) strongly reduces S2 flexibility and prevents the conformational changes associated with receptor interaction (7, 90, 91). Proline substitution of four additional S2 subunit residues (F817P, A892P, A899P, A942P), as in the HexaPro spike derivative (92), has similarly been shown to further impair the postfusion transition, thus significantly increasing spike protein stability and immunogenicity. One of these residues (A899) is part of the best-scoring CD8 epitope (AA 898-906) retrieved by NetMHC-4.0 prediction (Figure 5D) (92). On the other hand, the extremely low mutation rate (Avg SE=2.5x10e-4) of the AA 910-921 HR1 region, which contains one CD8 epitope (AA 917-927), likely reflects its interaction, after the postfusion structural rearrangement, with helix HR2. This interaction is involved in the formation of a six-helix bundle structure in the apical postfusion region 6HB-2 (close to the transmembrane fusion peptide), which, in turn, is critical for viral-host cell membrane fusion and virus internalization (10) (see Figure 5D).

Upon conversion to the postfusion form, some of the remaining CD8 epitopes located in the S2 subunit, either become part of another six-helix bundle (6HB-1; CD8 epitope comprised between AA 747-756), or of the connector β-sheet (four conserved epitopes comprised between AA 1051 and 1063) (Figure 5D) (10). A summary of the CD8 T cell epitopes that are located within conserved spike sequences which cannot tolerate the emergence of mutations because of their critical functional or structural roles is reported in Table 2.

Our literature search and in silico prediction analysis confirms, and actually underlines, the fact that the cytotoxic CD8 T cell activity induced by SARS-CoV-2 infection is widely multispecific in the context of individual HLA class I alleles and that some of the relevant degenerate and immunodominant epitopes are located within conserved spike regions that cannot tolerate mutational changes (Tables 1, 2). Thus, even in the event of abrogation of specific CD8 T cell responses directed against epitopes that have been inactivated as a result of SARS-CoV-2 variant mutations, the evidence that CD8-mediated responses are not only widely multispecific but can also target highly conserved (and potentially invariable) spike sequences makes mutation-mediated evasion from the overall protection afforded by CD8 T cells most unlikely. The ability of genetic vaccines to stimulate the endogenous synthesis of the spike protein should closely mimic spike antigen presentation to CD8 T cells during natural infection. Thus, CD8 T cell activation is expected to play a key and arguably more durable role also in antiviral protection induced by last-generation vaccines.

The identification of CD8 epitopes mapping on spike regions which are not prone to variation provides an additional, highly stringent criterion that may enable to select potent and potentially VOC-resistant epitopes. CD8 epitopes mapping on low entropy spike segments (listed in Figure 3) represent promising candidates for a CD8-potentiated and broadly protective vaccine prototype. As highlighted by the present analysis, epitope sorting can be comprehensively achieved based on a combination of different criteria including: i) epitope sequence position within highly- or hyper- conserved spike SARS-CoV-2 regions, ii) HLA-class I allele frequency and affinity, iii) immunodominance (Figure 3 and Table 1) and iv) predicted structural/functional invariance (Table 2). The innermost S2 central helix region, which harbors a number of CD8 epitopes (Figures 3, 5D and Table 2), is an almost invariable, hyper-conserved portion of the spike protein that is essential to drive the prefusion to postfusion conformational transition. Within this region is located the AA 1000-1008 epitope which displays both highly significant prediction scores and immunodominance properties. Similar features are shared by another promising epitope (AA 1052-1062) which is located in the conserved, S2 connector β-sheet (Figures 3, 5D and Table 2). While S2-located epitopes may exhibit cross-reactivity with other sarbecoviruses, CD8 epitopes mapping to the RBD, which is involved in ACE2 receptor recognition and is thus subjected to a high evolutionary pressure, are likely to be SARS-CoV-2 specific. Despite an overall high mutability, however, several CD8 epitopes are associated to conserved RBD segments (Table 1 and Figures 3, 5B). In particular, five epitopes (AA 454-473) are specifically located in the RBM subsegment and two additional epitopes map to the RBD flank (AA 328-347), a region that is targeted by neutralizing antibodies of high potential therapeutic interest (86–88).

In conclusion, it is likely that a subset of SARS-CoV-2 mutations can actually influence some CD8-mediated responses. The remaining unaffected responses, however, especially those stimulated by highly conserved epitopes, should largely compensate for this loss. Even without direct experimental data, these observations strongly support the notion that present vaccines will likely maintain a significant protection capacity also against VoCs. In other words, memory CD8 T cell responses are strongly expected to be capable of slowing-down the spread of infection and to prevent evolution to severe forms of disease, even under conditions in which protection mediated by neutralizing antibodies is significantly diminished or even lost. This also suggests that enrichment of presently available SARS-CoV-2 vaccines with highly conserved and degenerate CD8 T cell epitopes, such as the ones we have prioritized in this work, may represent a valuable strategy for amplifying CD8 T cell induction, thereby increasing vaccine robustness and long-term efficacy.

Considering that the spike protein is by far the most prevalent, if not the only target of current vaccines, we restricted our analysis to this particular target. It is known, however, that additional SARS-CoV-2 proteins are strongly targeted by CD8+ T-cell responses. In particular, the N protein has been reported to account for more than 35% of the overall CD8-mediated responses to other coronaviruses, 50% of which are induced by the spike protein (35). In human SARS-Cov-2 infection, the scenario is quite different, and the number of CD8 epitopes described so far within the N protein is about one third of those identified within the spike protein (41 N vs 168 S), but the S is about 3 times longer than the N protein (56, 57, 61, 67). In addition, N is less variable than the S protein and only 7 VoC/VoI mutations (within the variants analyzed in our study) have emerged so far within N compared to 34 mutations in the case of the S protein (https://cov-lineages.org/index.html). Finally, available studies point to a hierarchy of CD8 T cell immunodominance, with a prevalent reactivity against S (accounting for approximately 26% of all the identified CD8 T cell epitopes) associated with lower but still frequent recognition of other structural and non-structural proteins, including nsp3 (20%), nsp12 (9%) as well as, N (7%), M (6%) nsp4 (5%), ORF3a (5%), nsp6 (3%) and ORF8 (1%) (78). Thus, available genetic vaccines should benefit even more substantially from the inclusion of immunodominant and possibly degenerate CD8 epitopes located outside of the spike protein that are particularly relevant to the overall potency of naturally-induced antiviral protective responses and that, obviously, cannot be stimulated by immunization with a spike-only vaccine.

CB, DC, and AB: design and writing of the manuscript. MR, AV, and CT: contribution to figure drawing, analysis and interpretation of data, statistical analysis. VB and PF: discussion of the concepts to be introduced in the text, contribution to the selection of the references to quote; SO and CF: critical revision of text and figures. All authors contributed to the article and approved the submitted version.

This work was supported by private donations to the Unit of Infectious Diseases and Hepatology (CB, MR, AV, CT, VB, PF, and CF) of the Azienda Ospedaliero-Universitaria of Parma and by a grant from the University of Parma (DC, AB, and SO) supporting research activities in the field of Covid-19 infection.

CF: Grant: Gilead, Abbvie. Consultant: Gilead, Abbvie, Vir Biotechnology Inc, Arrowhead, Transgene, BMS.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.