T cell epitope mapping reveals immunodominance of evolutionarily conserved regions within SARS-CoV-2 proteome

Abstract

As SARS-CoV-2 variants continue to emerge capable of evading neutralizing antibodies, it has become increasingly important to fully understand the breadth and functional profile of T cell responses to determine their impact on the immune surveillance of variant strains. Here, sampling healthy individuals, we profiled the kinetics and polyfunctionality of T cell immunity elicited by mRNA vaccination. Modeling of anti-spike T cell responses against ancestral and variant strains of SARS-CoV-2 suggested that epitope immunodominance and cross-reactivity are major predictive determinants of T cell immunity. To identify immunodominant epitopes across the viral proteome, we generated a comprehensive map of CD4⁺ and CD8⁺ T cell epitopes within non-spike proteins that induced polyfunctional T cell responses in convalescent patients. We found that immunodominant epitopes mainly resided within regions that were minimally disrupted by mutations in emerging variants. Conservation analysis across historical human coronaviruses combined with in silico alanine scanning mutagenesis of non-spike proteins underscored the functional importance of mutationally-constrained immunodominant regions. Collectively, these findings identify immunodominant T cell epitopes across the mutationally-constrained SARS-CoV-2 proteome, potentially providing immune surveillance against emerging variants, and inform the design of next-generation vaccines targeting antigens throughout SARS-CoV-2 proteome for broader and more durable protection.

Fig. 1.. SARS-CoV-2 mRNA vaccine-induced adaptive immunity.

(A) Peripheral blood samples were collected from individuals receiving COVID-19 mRNA vaccines, mRNA-1273 (by Moderna) or BNT162b2 (by Pfizer/BioNTech) longitudinally: before vaccination (V0), 14 days after 1st dose (V1D14), 7 days after 2nd dose (V2D7) and 14 days after 2nd dose (V2D14), and from convalescent COVID-19 patients, who were not vaccinated. (B) Serum neutralization capacity was assessed by a pseudotype particle (pp) infection system, VSVΔG-Rluc bearing the SARS-CoV-2 D614G spike glycoprotein targeting 293T cells stably expressing ACE2 and TMPRSS2. 4-point nonlinear regression curves were used to calculate 50% pseudovirus neutralization titers (pVNT50) for vaccine recipients (n=12) and convalescent patients (n=20). Horizontal lines denote median pVNT50 values. Peripheral blood mononuclear cells (PBMCs, 2x10⁵ cells/well) were stimulated with pools of overlapping peptides, 15mers with 5 amino acid offsets, spanning the N-terminal (Spike_N or S_N) or C-terminal half (Spike_C or S_C) of Spike protein for 24 h and IFN-γ secretion was measured by ELISPOT. (C) Representative ELISPOT wells from a vaccinated donor (top left) and summary of ELISPOT data (n=11 vaccinated donors, BNT162b2 recipients in shown red, mRNA-1273 recipients in blue, and n=5 convalescent patients, shown in black), where Spike_Total (top right) is the sum of spots acquired by Spike_N (bottom left) and Spike_C (bottom right) peptide pools. T cells were expanded following stimulation with Spike_N and Spike_C peptide pools. Antigen-specific cytokine production by expanded T cell subsets, CD4+ or CD8+, were measured by intracellular staining by flow cytometry. IFN-γ production by Spike-specific T cells (D) in vaccinated donors and (E) convalescent patients. (F) Polyfunctionality of Spike-specific T cells at V2D14 as demonstrated by % of T cells (y axis) co-expressing effector cytokines: IFN-γ, TNF-α, and IL-2 (x axis). (G) Distribution of Spike-specific (Spike_Total) T cell responses in each vaccinated donor at V2D14. Spot numbers and cytokine+ cell frequencies were demonstrated after background subtraction. Statistical significance (p < 0.05) was evaluated by Wilcoxon matched-pairs test by comparing vaccination timepoints and Welch’s t-test was used for comparing T cell responses elicited by mRNA-1273 vs BNT162b2.

Fig. 2.. Adaptive immune recognition of SARS-CoV-2 variants.

(A) Antibody reactivity from vaccinated (before vaccination (V0), 14 days after 1st dose (V1D14), 7 days after 2nd dose (V2D7) and 14 days after 2nd dose (V2D14)), convalescent (conv), or pre-COVID-19 (control) serum to RBD were assessed by Luminex antibody binding assay where MagPlex-C Microspheres Regions were conjugated to recombinant wild-type (WT, Wuhan-1) and mutant RBD constructs (Alpha [N501Y], Beta [N501Y/K417N/E484K], and Gamma [N501Y/K417T/E484K] with mean fluorescence intensity (MFI) used as a readout for binding affinity. (B) Summary of ELISPOT data: peripheral blood mononuclear cells (PBMCs, 2x10⁵ cells/well) from vaccinated donors (V2D14) or from convalescent patients were stimulated with pooled peptides covering the mutations found in Alpha, Beta and Gamma variants (listed in (E)) or the corresponding WT sequences for 24 h and IFN-γ secretion was measured by ELISPOT. (C) V2D14 T cells from vaccinated donors were stimulated with variant or WT peptide pools and expanded prior to being re-stimulated with either the initial stimulation peptide pool (WT→WT, Mut→Mut) or the variant pool to measure cross-reactivity (WT→Mut). Antigen-specific cytokine production by expanded T cell subsets, CD4+ or CD8+, was measured by intracellular staining by flow cytometry. (D) Bars show Bayesian Information Criterion (BIC) values for different models of T cell reactivity shown in C, where lower BIC is better. The basic model includes the effect of peptide pool stimulations and WT-Mutant peptide cross-reactivity. Different peptide dominance models are shown on the y axis, which correspond to the 9mer aggregation function used. Performance of other models was also measured, blue: accounting for sequence similarity with IEDB epitopes, brown: including vaccination effect as the initial stimulation event, pink: excluding the effect of patient-specific amplitudes, green: excluding the effect of cross-reactivity between peptides from 1^st and 2^nd stimulation events. (E) Deconvolution of IFN-γ production by T cells in response to individual mutations within the peptide pools tested in C (WT→WT vs Mut→Mut). Statistical significance (p < 0.05) was evaluated by Wilcoxon matched-pairs test.

Fig. 3.. T cell responses against non-spike SARS-CoV-2 proteins.

(A) Schema demonstrating peptide selection for each protein. The y axis denotes amino acid residue number. Tested regions are shown in red. (B) Experimental design summary. Peripheral blood mononuclear cells (PBMCs, 2x10⁵ cells/well) from convalescent, unvaccinated patients (n=10) were stimulated for 24 h with pools of overlapping peptides, 15mers with 5 amino acid offsets, spanning each protein as indicated on the x axis. Pools contained no more than 25 peptides. IFN-γ secretion was measured by ELISPOT. Summary of ELISPOT data showing total T cell responses (C) per peptide pool and (D) per patient. T cells from the same convalescent, unvaccinated patient cohort (n=15) were expanded following stimulation with non-spike peptide pools. Antigen-specific cytokine production (IFN-γ+) by expanded T cell subsets, (E) CD4+ or (F) CD8+, was measured by intracellular staining by flow cytometry. Each dot corresponds to a patient. Normalized values were shown. Statistical significance (p < 0.05) was evaluated by the Wilcoxon matched-pairs test comparing DMSO vs peptide stimulation.

Fig. 4.. Deconvolution of T cell responses against non-spike SARS-CoV-2 proteins.

T cells from convalescent, unvaccinated patients were expanded following stimulation with non-spike peptide pools. Then, expanded T cells were re-stimulated by individual 15mers constituting the peptide pools. Antigen-specific cytokine production by expanded T cell subsets, CD4+ (in blue) or CD8+ (in red), was measured by intracellular staining by flow cytometry. Heat maps demonstrate the percentage of reactive, polyfunctional T cells (secreting both IFN-γ and TNF-α) after normalization (subtraction of background, DMSO stimulation). y and x axes indicate the patients and peptides tested, respectively. Peptide sequences are reported in table S2. X indicates that the data was not collected. Statistical significance (p < 0.05) for peptides inducing T cell responses across the tested population was evaluated by Wilcoxon matched-pairs test comparing DMSO vs peptide stimulation and denoted in red stars if significant (immunodominant peptides). A peptide was considered immunogenic, a “hit”, if in at least one patient, the % of reactive cells was greater than the paired DMSO % plus 3 times the standard deviation (>DMSO + 3SD) of all DMSO values across the population, denoted by green dots.

Fig. 5.. Conservation of T cell epitopes.

(A) Conservation of the “hit” T cell peptides, identified in fig. 4, was measured as percent of previously deposited GISAID sequences with an exact match to the reactive peptide. The median conservation percentages for the CD4 and CD8 “hit peptides” were 99.54% and 99.47%, respectively. (B) Entropy values for the residues found in “hit” peptides (green) or “hit” peptides that were significantly enriched across the test population (orange) were compared to the entropy values for all other residues (purple) in the ORFs tested in our study. Statistical significance was calculated by Welch’s t-test. (C) Topography of immunogenic nucleocapsid residues is displayed. NTD: N-terminal domain, RBD: RNA binding domain, DD: dimerization domain, CTD: C-terminal domain. The black bars represent the difference between per residue immunogenicity (experimental, shaded in pink) and per residue binding value (predicted, shaded in lilac). Per residue immunogenicity values were generated by summing for each residue the normalized percentage of CD8 or CD4 T cells expressing both TNFα and IFNγ (>DMSO + 3SD) directed at each peptide for which the residue belongs. Per residue binding value corresponds to the frequency that a given residue appeared in a peptide predicted to bind to a patient MHC. The per residue values were scaled from 0-1 with 1 representing the residue with the highest immunogenicity value or the most frequently included in a predicted binding peptide and 0 representing the residue with the lowest immunogenicity or least frequent. The per residue values for binding predictions were then multiplied by −1 to reverse the sign. The Y axis was then transformed to represent the log odds ratio of the probability of being immunogenic vs predicted binders by dividing by the background probability (1/number of residues in the nucleocapsid protein). The normalized entropy per amino acid (aa) codon was also aligned with nucleocapsid, black bars denoting the degree of entropy and heatmaps showcasing the intensity of sharing of immunogenic residues in our cohort. (D) Summary of conservation degree for nucleocapsid CD8+ hit sequences analyzed by CLUSTAL O (1.2.4) multiple sequence alignment for H-COV Nucleocapsid Protein: 229E (UniProt Accession: A0A127AU35), NL63 (UniProt Accession: Q06XQ2), HKU1 (UniProt Accession: Q5MQC6), OC43 (UniProt Accession: P33469), SARS_2 (UniProt Accession: P0DTC9). Hit peptides are marked in green and other tested sequences are in purple. Sharing was denoted as the following: “*” Residue shared among all coronaviruses in sequence alignment, “:” Conservation between groups of strongly similar properties, “.” Conservation between groups of weakly similar properties. (E) Mapping of immunogenic regions within RNA binding domain (RBD) on the 6YVO crystal structure. (F) The average change in protein stability (DDG) of RBD upon mutating each residue to alanine for the top ten most immunogenic peptides. Positive values indicate destabilizing mutations. The dashed line indicates the sliding window average (length 9 for MHC-I and 15 for MHC-II) across the nucleocapsid RBD. (G) Mapping of immunogenic regions within the dimerization domain onto the 6ZWO crystal structure. (H) Number of immunogenic peptides that contained at least one residue that could destabilize the interface when mutated to the alanine position (DDG>0 when mutated) was calculated for each patient.