The study cohort consisted of 72 COVID-19 patients (27 female, 45 male; age range, 21-80 years), including 52 subjects hospitalized with severe COVID-19 and 20 individuals with mild COVID-19 who tested positive for SARS-CoV-2 RNA by reverse transcriptase quantitative polymerase chain reaction from nasopharyngeal swabs. Blood samples were obtained between March 2020 and February 2021 at 1 month and 6 months post infection ( Supplementary Table 1 , Figure 1A ). Blood samples from 44 individuals were obtained before and after the first and second COVID-19 mRNA vaccine dose. Blood samples from 18 healthy controls were collected between 2013 and 2019 through routine screening of blood donations performed by the Austrian Red Cross for prior unrelated studies. The donors were considered unexposed controls, given that the early phase of the first COVID-19 wave in Austria occurred in the spring of 2020 (34). The analysis of cross-reactive T cells was conducted in a subcohort of participants (“HLA matched cohort”, Supplementary Table 1 ) who carried specific HLA-DP or HLA-DR alleles that were previously shown to bind CoV-cross-reactive SARS-CoV-2 S_813-829 and S_1002-1018 peptides (35). Follow-up samples were obtained 1 and 3 years after the 2^nd vaccine dose ( Supplementary Table 1 , Figure 1A ). Details of the follow-up cohort (vaccine types, SARS-CoV-2 omicron variants) are provided in Supplementary Table 7 .

SARS-CoV-2 RNA was extracted from respiratory specimens using NucliSENS easyMAG extractor (BioMérieux, Marcy l’Etoile, France). SARS-CoV-2 real-time TaqMan PCR was performed as described previously (36).

Nasopharyngeal swabs from routine diagnostics were PCR-tested for SARS-CoV-2. Variants were identified using Vir-SNiP SARS-CoV-2 assays (TIB MOLBIOL, Berlin, Germany) to distinguish between BA.1, BA.2, BA.5 and XBB.1.5. In some cases, variants were additionally confirmed by next-generation sequencing using amplification with ARTIC network tiled amplicon primers followed by Nextera XT library preparation and sequencing on an Illumina MiSeq device.

Plasma was separated from whole blood samples by centrifugation for 10 min at 3000g and stored at -20°C until further use. Peripheral blood mononuclear cells (PBMCs) were isolated by density gradient centrifugation and stored in liquid nitrogen until further use. HLA typing of PBMCs was carried out by next generation sequencing, as described previously (37, 38).

Predictions were performed using NetMHCIIpan4.1EL from the IEDB database (39), and percentile rank was selected as an output. Predictions were done for experimentally validated S_813-829 (SKRSFIEDLLFNKVTLA) and S_1002-1018 (QSLQTYVTQQLIRAAEI) peptides with the HLA II alleles from all individuals ( Supplementary Table 2 ). Peptide:MHCII combinations with a percentile rank score >2% that were present in ≥10% of the study cohort were selected for further analysis.

For T cell stimulation, Peptides S_813-829 (SKRSFIEDLLFNKVTLA) and S_1002-1018 (QSLQTYVTQQLIRAAEI) as well as pools containing non-conserved SARS-CoV-2-specific peptides were obtained through BEI Resources, NIAID, NIH: Peptide Array, SARS-Related Coronavirus 2 Spike (S) Glycoprotein, NR-52402. PepMix™ SARS-CoV-2 peptide pools covering the entire sequences of the SARS-CoV-2 spike proteins of the ancestral strain as well as omicron BA.4/BA.5, and XBB.1.5 variants (product codes: PM-WCPV-S-1, PM-SARS2-SMUT15-1 and PM-SARS2-RBDMUT10-1) were purchased from JPT (Berlin, Germany). The peptide libraries comprise 15-mer peptides overlapping by 11 amino acids (including a few 17-mers). Spike responses were tested in two separate pools, SI and SII, to distinguish responses to the N-terminal spike (covered by the SI peptide pool) from the more conserved C-terminal part (covered by the SII peptide pool), which includes peptides S_813-829 and S_1002-1018.

ELISpot assays were performed as previously described (38, 40, 41). In brief, MultiScreen HTS IP Sterile Plates (Millipore) were coated with 1µg anti-human IL-2 antibody (3445-3-250, Mabtech) overnight and blocked with 5% BSA. PBMCs (200,000 cells/well) were incubated at 37°C and 5% CO2 for 20 hours with SARS-CoV-2 peptides (2 µg/ml; duplicates), AIM-V medium (negative control; 3-5 wells) or leucoagglutinin (PHA-L; L4144, Sigma; 0.5 µg/ml; positive control). Plates were washed and spots developed with biotinylated anti-IL-2 antibody (3445-6-250; Mabtech), streptavidin-coupled alkaline phosphatase (ALP; 3310-10, Mabtech; 1:1000) and 5-bromo-4-chloro-3-indolyl phosphate/nitro blue tetrazolium (BCIP/NBT; B5655, Sigma). Spots were counted using the Bio-Sys Bioreader 5000 Pro-S/BR177 and Bioreader software generation 10. The cut-off for positivity was >15 SFU/10⁶ PBMCs after background subtraction (mean spot count from 3-5 unstimulated wells).

Flow cytometry and propagation of PBMCs was conducted as described previously (42). Briefly, PBMCs were stimulated with S_813-829 (SKRSFIEDLLFNKVTLA) or the homologous OC43 S_911-927 (RSAIEDLLFDKVKLSDV) peptide and cultured for 10 days adding 100 IU IL-2 on day 5. In vitro expanded cells were analyzed by intracellular cytokine and cell surface marker staining. PBMCs were incubated with 2 μg/ml of S_813-829 (SKRSFIEDLLFNKVTLA) or the homologous OC43 peptide S_911-927 (RSAIEDLLFDKVKLSDV) and 1 μg/ml anti-CD28/49d antibodies (L293 and L25, Becton Dickinson) or without antigen (negative control) for 6h. After 2h, 0.01 μg/ml brefeldin A (Sigma) was added. Tubes were stored in the freezer overnight. Cells were stained with APC/H7 anti-human CD3 (SK7, Becton Dickinson), Pacific Blue anti-human CD4 (RPA-T4, Becton Dickinson), and PE anti-human CD8 (HIT8a, Becton Dickinson). After surface staining, cells were fixed and permeabilized using a Caltag Laboratories Fix&Perm^® cell permeabilization kit (Invitrogen) as recommended by the manufacturer. Following permeabilization, cells were stained with APC anti-human IL-2 (5.34411e+006, Becton Dickinson), FITC-anti-human IFN-γ (25723.11, Becton Dickinson) and PE-CY7-anti-human TNF (MAB11, Becton Dickinson). Cell viability was assessed using ViViD Aqua Dead (Invitrogen). Boolean gating function of FlowJo was used to assess each cytokine combination. Responses obtained by Boolean analysis were considered positive if they were ≥ twofold above samples with no antigen stimulation. Frequencies below the cut-off for positivity were set to 10^-3. Analysis was carried out using FACS Canto II (Becton Dickinson) and data processed using the FlowJo software v. 10.7.2 (Becton Dickinson).

Neutralization tests were performed as described previously (10). In short, duplicates of 2-fold diluted heat-inactivated serum were incubated with 50–100 TCID₅₀ SARS-CoV-2 for 1h at 37°C and then added to a monolayer of Vero E6 cells. After incubation for 3-5 days, the cytopathic effect of non-neutralized infectious virus was assessed, and neutralization titers were assessed as the reciprocal of the serum dilution for protection against induction of cytopathic effect. The threshold for positivity was set at a neutralization titer of ≥10.

Statistical analysis was performed using Stata 17 (StataSoft, College Station, TX, USA) and GraphPad Prism 9 (GraphPad Software, Boston Massachusetts USA). For medians and post-vaccination neutralization titers, values <10 were set to 5. Unpaired t-test and one way ANOVA was performed for group comparison, the Spearman rank test was used for correlations, and the Wilcoxon matched-pairs test and Dunn´s multiple comparisons test for paired samples. Univariate analysis examined demographic and immunological parameters as predictors of viral load and disease severity. A multivariate linear or logistic regression model was used to examine the adjusted effect of T cells (independent variable), viral load and hospitalization (dependent variable), respectively. Generalized Structural Equation Models were applied with hospitalization as a Bernoulli variable, and age, log transformed viral RNA load and T cell counts as normal variates. For all analyses, a p-value <0.05 was considered significant.

Studies were conducted in accordance with the Declaration of Helsinki in terms of informed consent and approval by the institutional review board of the Medical University of Vienna, Austria. All patients provided written informed consent. The use of biobanked plasma and cell samples from healthy pre-pandemic blood donors, patients and vaccinees, as well as virological data received by the Center for Virology has been approved by the Ethics committee of the Medical University of Vienna, Austria (EK 2156/2019, EK 2283/2019 and 1291/2021). Anonymized leftover samples were stored in the biobank of the Center for Virology, following established protocols (EK 1513/2016) in accordance with national legislation that required no additional written informed consent.

We first determined the overall extent of SARS-CoV-2 T cell responses in ex vivo IL-2 ELISpot assays using peptide libraries composed of overlapping 15-mer peptides covering the full-length of spike protein in two pools SI and SII. The rationale for using an IL-2 ELISpot assay was based on previous studies, which demonstrated that IL-2 is a highly sensitive read-out to measure SARS-CoV-2 T cells associated with protection against infection (1), and on initial experiments indicating that IL-2 and IFN-γ assays detected similar quantities of SARS-CoV-2 T cell responses. We analyzed peripheral blood mononuclear cells (PBMCs) from 44 SARS-CoV-2-naïve individuals who received BNT162b2-mRNA vaccination, and 72 COVID-19 patients as well as 18 pre-pandemic controls (for characteristics of study participants, see Supplementary Table 1 ). The analysis revealed a significant increase in the levels of IL-2-secreting T cells (herein referred to as IL-2⁺ T cells) against SI and SII in both SARS-CoV-2-infected and vaccinated cohorts ( Figures 1B, C ). Virus-neutralizing antibodies, determined in live-virus neutralization assays, were detected in sera from 93% COVID-19 patients, and neutralizing titers correlated with IL-2⁺ T cell reactivity against spike SI and SII ( Figures 1D, E ). In vaccinated individuals, neutralizing antibodies were detected in 43% after the 1^st and 100% after the 2^nd vaccination ( Figure 1D ). IL-2⁺ T cell levels present after the 1^st vaccine dose positively correlated with neutralizing titers after the first dose ( Figure 1F ), while no correlation was found between neutralizing titers and IL-2⁺ T cell frequencies present after the second vaccine dose ( Figure 1G ). These findings are consistent with previous studies (41, 43), suggesting a role of antigen-specific T cells in facilitating the development of SARS-CoV-2-specific antibody responses.

To assess cross-reactivity in SARS-CoV-2 T cell responses, we focused on two previously described, experimentally validated, cross-reactive HLA class II epitopes in the spike S2 domain (44): S_813-829, presented by HLA-DP4 (HLA-DPA1*01:03/DPB1*04:01; 50% frequency, Europe; 15-80%, world-wide (45–47)), and S_1002-1018, presented by HLA-DR15 (HLA-DRB*15:01; 13% frequency, Europe; 11% world-wide (46)). The epitopes are conserved between SARS-CoV-2 and CoVs ( Figure 2A ), in particular OC43, with homologous peptides presented by the same HLA-II molecules (35). The HLA-DP4 and HLA-DR15 core sequences differ by one (S_813-829) or three residues (S_1002-1018) between SARS-CoV-2 and OC43, respectively ( Figure 2A ). We performed HLA-II genotyping and in silico prediction of peptide binding strength via NetMHCpan4.1EL (39). From 222 distinct HLA-II alleles present in our study cohort, we selected a subcohort of participants with HLA-DP4, HLA-DR15 or three additional HLA-DP alleles that met two criteria: (i) predicted with high binding scores to S_813-829 or S_1002-1018, with percentile ranks <2 and (ii) detected at frequencies ≥10% in the study population ( Figure 2B , Supplementary Table 2 ). Consistent with previous findings, we observed T cell reactivity to S_813-829 in pre-pandemic samples from 3 of 16 HLA-matched blood donors, whereas no significant T cell reactivity was detected against S_1002-1018 ( Supplementary Figure 1A ). To confirm the S_813-829 peptide as a CoV-cross-reactive epitope, we analyzed responses against SARS-CoV-2 S_813-829 or the homologous OC43 S_911-927 peptide in PBMCs from the three pre-pandemic donors. After 10-day in vitro stimulation with SARS-CoV-2 S_813-829 or the OC43 S_911-927 peptide, we assessed cytokine production (interferon gamma (IFN-γ), interleukin (IL)-2, and tumor necrosis factor alpha (TNF-α)) by intracellular cytokine staining (ICS) after restimulation with S_813-829 or the homologous OC43 S_911-927 peptide. Paired analysis revealed statistically significant differences in IFN-γ and IL-2 responses of SARS-CoV-2 S_813-829 or OC43 S_911-927-stimulated samples compared with unstimulated controls in several PBMC lines generated from 2 out of 3 donors ( Supplementary Table 3 ). Results from two representative cell lines are illustrated in Supplementary Figure 1B . PBMC lines generated with SARS-CoV-2 S_813-829 showed cross-reactivity with the OC43 S_911-927 peptide ( Supplementary Figure 1C ). Furthermore, we generated six cell lines following 10-day stimulation with OC43 S_911-927, and confirmed that these cells were cross-reactive with the SARS-CoV-2 S_813-829 peptide ( Supplementary Figure 1C ). These data indicate that cross-reactive T cells recognizing the CoV OC43 and SARS-CoV-2 spike epitopes were present in unexposed individuals.

We next examined ex vivo ELISpot responses against S_813-829 and spike SI and SII peptide pools in a sub-cohort of 55 individuals with SARS-CoV-2 infection and 31 individuals who received BNT162b2 mRNA vaccination and carried the relevant HLA-DP or HLA-DR alleles. While only 15% (5/34) HLA-DR15⁺ individuals displayed a response to the S_1002-1018 peptide, S_813-829-reactive T cells were detected in 51% COVID-19 patients and in 55% individuals after the 2^nd vaccine dose, with significantly higher frequencies than in non-exposed donors ( Figures 2C, D ). Of note, samples from individuals who possessed S_813-829-specific T cells (herein referred to as S_813-829 responders, S_813-829 R⁺) displayed significantly enhanced SII-specific responses compared with those in whom S_813-829-specific T cells were not detected (S_813-829 non-responders, S_813-829 R^-) ( Figures 2E, F ). In contrast, less pronounced or no differences were observed between S_813-829 responders and non-responders when we analyzed T cell reactivity to a pool of 22 SARS-CoV-2 peptides in non-conserved regions of the spike with no homolog in CoVs (SARS-CoV-2* peptide pool; Supplementary Table 4 ). As illustrated in Figure 2G , individuals mounted robust SARS-CoV-2* peptide pool responses that were similar or higher in S_813-829 R⁺ than in S_813-829 R^-, suggesting that cross-reactive T cells did not abrogate the formation of de novo SARS-CoV-2 T cell responses.

To further examine the impact of cross-reactive T cells, we performed viral RNA (vRNA) load analysis from nasopharyngeal secretions, available from 44 COVID-19 patients (mild infection, n=15; hospitalized with severe disease, n=29). At one month after symptom onset, 40.9% (18/44) of the patients displayed detectable vRNA (range: 4x10²-1.6x10⁷ copies/ml). Univariate analyses revealed that lower vRNA loads correlated with higher T cell levels against S_813-829 (p<0.001; R² = 0.26) and SII peptides (p=0.022; R² = 0.12). By contrast, higher vRNA load, older age and increased reactivity to SI peptides correlated with severe disease (hospitalization) ( Supplementary Tables 5 and 6 ). Linear regression analysis revealed an age-dependent decline in S_813-829 T cell levels (R² = 0.1187, p=0.01), consistent with previous studies (3). Therefore, we performed multiple regression analyses, including age as a covariate, to assess the effect of cross-reactive T cells on vRNA load. This analysis indicated that S_813-829-specific T cell reactivity correlated with significantly reduced viral loads (p=0.004; regression coefficient -0.877) ( Figure 2H ). When we analyzed the impact of S_813-829-specific T cells and viral load on disease severity using structural equation modeling with age as a covariate, we observed that while the direct effect of S_813-829-specific T cells on disease severity was not significant (p=0.808; standardized regression coefficient 0.024), the indirect effect, mediated by viral load reduction was significant (p=0.021; standardized regression coefficient 0.12) ( Figure 2H ).

A key unresolved question is whether CoV-cross-reactive T cells contribute to immune responses beyond the first SARS-CoV-2 encounter. To measure T cell maintenance, we compared IL-2⁺ T cell frequencies to S_813-829 and spike SI and SII peptide pools in paired samples 6-12 months apart ( Figure 1A ). Follow-up samples from patients were taken 6 months post infection (T2, n=13), and from vaccinated individuals one year after 2^nd vaccination (T2, n=21); all vaccinated individuals had received a booster dose at a median of 3 months before sample collection ( Supplementary Table 1 ). A significant reduction in IL-2⁺ T cell frequencies to S_813-829 and SI and SII peptide pools was observed between acute SARS-CoV-2 infection and 6 months post-infection, while IL-2⁺ T cell levels in vaccinated individuals who had received a booster dose were not different to those post 2^nd vaccination ( Supplementary Figure 2 ). Notably, S_813-829-specific T cells could still be detected in a significant proportion of individuals both after infection (46%) or vaccination (38%) ( Figure 3A ). To assess long-term T cell persistence, we compared IL-2⁺ T cell frequencies to S_813-829 and spike SI and SII peptide pools after booster vaccination with follow-up samples obtained approximately 3 years post 2^nd vaccination (T3, median: 2.5 years after booster vaccination) ( Supplementary Table 1 ). Between time points T2 and T3, 76% (16/21) of the individuals received one to three additional booster doses, including bivalent wild-type/BA.5 vaccine (n = 7) and/or monovalent XBB.1.5 vaccine (n = 9), and 71% (15/21) had a confirmed omicron breakthrough infection ( Supplementary Table 7 ). Evaluations of IL-2⁺ T cell frequencies showed no significant differences in the response against spike SI or SII and S_813-829 between T2 and T3 ( Figure 3A ). Since T cell responses to S_813-829 were maintained in three-year-follow-up samples, we analyzed the frequency of these cells targeting the CoV OC43 homologous peptide. PBMCs were propagated with S_813-829 for 10 days and restimulated with S_813-829 or the OC43 S_911-927 peptide. Responsive cells were identified by expression of IFN-γ, IL-2 or TNF-α production by flow cytometry. Despite substantial individual variability in the frequencies of responses, we detected CD4 T cells targeting SARS-CoV-2 S_813-829 and OC43 S_911-927 peptides in all individuals who mounted S_813-829-specific responses, while the proportions of CD8 T cells targeting these peptides were considerably lower ( Figure 3B ). Analysis of cytokine-positive subsets, including polyfunctional cells (defined as cells producing ≥ 2 cytokines) and single cytokine positive cells revealed that SARS-CoV-2 S_813-829 and OC43 S_911-927 peptides induced mainly triple- and dual-positive cells with positive correlations between individual IL-2 and IFN-γ subsets (Spearman R=-0.81, p<0.0001), and IL-2 and TNF-α subsets (Spearman R=-0.98, p<0.0001) ( Supplementary Figure 3 ). In summary, these data indicate that SARS-CoV-2 S_813-829 peptide induced polyfunctional cross-reactive CD4 T cell populations, and that a substantial proportion of these cells was maintained in the memory T cell pool.

Furthermore, we compared the frequency of T cells targeting ancestral and omicron BA.5 and XBB.1.5 spike peptides. The data revealed similar frequencies of spike SI and SII T cell responses across these variants ( Figure 3C ). In contrast, serum samples from these individuals showed significantly reduced neutralizing antibody titers against omicron BA.5 and XBB.1.5 variants compared to the ancestral strain, demonstrating a significant escape from antibody-mediated neutralization ( Figure 3D ).

To assess whether cross-reactive T cells were associated with enhanced T cell responses to Omicron variants, we compared spike-specific responses between S_813-829 R+ and S_813-829 R-. The analysis revealed that S_813-829 responders displayed significantly enhanced SII-specific T cell responses to ancestral and omicron BA.5 spike compared with those in whom S_813-829-specific T cells were not detected (ancestral, p=0.0227; BA.5, p=0.0148) ( Figures 3E, F ). For omicron XBB.1.5, a similar trend for enhanced SII-specific T cell levels in S_813-829 R⁺ vs S_813-829 R^- was observed, although the difference was not statistically significant (p=0.0718) ( Figure 3G ). Together, the data suggest that S_813-829 contributes to long-lasting memory T cells that recognize CoV OC43 and SARS-CoV-2 variants, such as omicron BA.5 and XBB.1.5.