T cell responses to SARS-CoV-2 spike cross-recognize Omicron

The SARS-CoV-2 Omicron variant (B.1.1.529) has multiple spike protein mutations1,2 that contribute to viral escape from antibody neutralization3–6 and reduce vaccine protection from infection7,8. The extent to which other components of the adaptive response such as T cells may still target Omicron and contribute to protection from severe outcomes is unknown. Here we assessed the ability of T cells to react to Omicron spike protein in participants who were vaccinated with Ad26.CoV2.S or BNT162b2, or unvaccinated convalescent COVID-19 patients (n = 70). Between 70% and 80% of the CD4+ and CD8+ T cell response to spike was maintained across study groups. Moreover, the magnitude of Omicron cross-reactive T cells was similar for Beta (B.1.351) and Delta (B.1.617.2) variants, despite Omicron harbouring considerably more mutations. In patients who were hospitalized with Omicron infections (n = 19), there were comparable T cell responses to ancestral spike, nucleocapsid and membrane proteins to those in patients hospitalized in previous waves dominated by the ancestral, Beta or Delta variants (n = 49). Thus, despite extensive mutations and reduced susceptibility to neutralizing antibodies of Omicron, the majority of T cell responses induced by vaccination or infection cross-recognize the variant. It remains to be determined whether well-preserved T cell immunity to Omicron contributes to protection from severe COVID-19 and is linked to early clinical observations from South Africa and elsewhere9–12.

response leads to lower disease severity 14 . A greater CD8 + T cell response in blood and highly clonally expanded CD8 + T cells in bronchoalveolar lavage were observed in convalescent patients who experienced mild or moderate disease compared with severe disease 15,16 , and CD8 + T cells provided partial protective immunity in the context of suboptimal antibody titres in macaques 17 . In this study, we include 138 participants grouped according to their vaccination and COVID-19 status, in order to determine whether T cells generated in response to vaccination or previous SARS-CoV-2 infection could cross-recognize Omicron and to define the profile of T cell responses in Omicron-infected patients compared with those infected with other variants of concern.

T cell cross-reactivity to Omicron
We examined T cell responses in participants who had received one or two doses of the Ad26.COV2.S vaccine ( Johnson and Johnson/Janssen, n = 20 per group), two doses of the BNT162b2 mRNA vaccine (Pfizer-BioNTech, n = 15 per group), or who had recovered from infection (n = 15 per group) (Fig 1a, Extended Data Table 1a, b). Convalescent donors were examined a median of 1.4 months (interquartile range (IQR): 1.3-6 months) after mild or asymptomatic infection. More than 85% of vaccinees generated a T cell response to vaccination, measured 22-32 days after the last dose (Fig. 1b). Both vaccination and infection induced spike-specific CD4 + T cell responses, whereas a CD8 response was less consistently detected (Fig. 1c). We measured cytokine production (IFN-γ, IL-2 and TNF) by intracellular cytokine staining in response to peptide pools covering the full Wuhan-1 spike protein (ancestral) and the Omicron spike (Fig. 1d, Extended Data Fig. 1a).
The levels of CD4 + T cell responses to Omicron spike were consistently and significantly lower than those responsive to ancestral spike in all groups tested (Fig. 1e). This translated to a median decrease of 14-30% of the CD4 response to Omicron, as demonstrated by fold change (Fig. 1f). Similar results were observed for the CD8 + T cell response (Fig. 1g, h): vaccinees who had received two doses of Ad26. COV2.S and convalescent donors exhibited a significantly lower frequency of Omicron spike-specific CD8 + T cells, although the other groups did not. There was a median reduction of 17-25% in the CD8 response to Omicron compared with the ancestral virus. Of note, a fraction of responders (5 out of 32, 15%) exhibited a loss of CD8 + T cell recognition of Omicron (Figure 1g, Extended Data Fig. 1b), probably reflecting specific human leukocyte antigen (HLA) molecules being adversely affected by mutations in particular CD8 epitopes 18 .

Article
In parallel, we measured the neutralizing activity against ancestral and Omicron spike from the plasma of the same participants who received BNT162b2 (n = 10) or two doses of Ad26.COV.S (n = 19) (Extended Data Fig. 2). As previously described 3,5,6 , Omicron escapes the SARS-CoV-2 neutralizing antibodies generated after BNT162b2 vaccination. Here we present neutralizing responses to Omicron after two doses of Ad26.COV2.S (Extended Data Fig. 2b), demonstrating diminished neutralization capacity compared with D614G ancestral virus and the Beta variant. Comparison of the fold change in T cell responses and neutralizing antibodies targeting ancestral or Omicron spike further emphasizes the preservation of the T cell response, even when neutralization is severely reduced.
Mutations in variant epitopes have the potential to affect the functional capacity of cells 19 . Thus, we compared the polyfunctional profiles of T cells in vaccinees and convalescent individuals and demonstrate similar capacities for cytokine co-expression across all groups for both ancestral and Omicron-specific T cells (Extended Data Figs. 3a, b, 4a, b). Notably, there were also no differences in the polyfunctional profiles between ancestral and Omicron spike for either CD4 + or CD8 + T cells (Extended Data Figs. 3c, 4c), indicating the absence of a functional deficit in cross-reactive Omicron T cell responses. We also compared Omicron spike responses to other variants of concern in Ad26.CoV2.S vaccinees, by testing spike peptide pools corresponding to the viral sequences of the Beta and Delta strains (Extended Data Fig. 5a). There were no significant differences in cross-reactive CD4 + and CD8 + T cell responses between Beta, Delta and Omicron (Extended Data Fig. 5b), with the exception of a greater decrease in the Omicron CD4 response compared with Beta in people who had received two doses of Ad26. COV2.S. Of note, whereas previous SARS-CoV-2 infection in vaccinees was associated with a higher frequency of spike-specific T cells (Extended Data Fig. 6a), it had no impact on Omicron cross-reactivity (Extended Data Fig. 6b). Overall, these results show that CD4 + and CD8 + T cell recognition of Omicron spike is largely preserved compared with the ancestral strain, and is similar to other variants of concern carrying fewer mutations.

T cell response to different variants
The SARS-CoV-2 epidemic in South Africa has been characterized by four virologically distinct infection waves (Fig. 2b). This enabled us to compare T cell responses in patients infected with SARS-CoV-2 during the current fourth epidemic wave, dominated by Omicron, with those infected in previous waves dominated by ancestral (wave 1, n = 17), Beta (wave 2, n = 16) and Delta (wave 3, n = 16) variants (Fig. 2a). In addition to extensive mutations in spike, Omicron has 20 additional mutations in other proteins which could also result in T cell escape. Therefore, we measured the frequency of CD4 + and CD8 + T cells reactive towards ancestral spike, nucleocapsid and membrane proteins, all major targets of the T cell response 20 . We studied SARS-CoV-2-infected patients who were hospitalized with COVID-19 (Fig. 2a). These recently hospitalized patients, recruited between 1 December 2021 and 15 December 2021 (n = 19), had no previous history of COVID-19 and were unvaccinated. Omicron infection was inferred by spike gene target failure (SGTF) 21 in nine of these patients. Although swabs were unavailable for the remainder, with Omicron accounting for more than 90% of sequences from South Africa at the time of recruitment and 98% in Tshwane from where the samples originated (Fig. 2b), there was a high probability of Omicron infection in all of these patients.
Despite differences in age, disease severity and co-morbidities across the infection waves (Fig. 2a, Extended Data Table 1c), T cell responses directed at spike, nucleocapsid and membrane proteins in wave 4 patients were of similar magnitude as those in patients infected with other SARS-CoV-2 variants in previous waves (Fig. 2c, d). The frequency of responders also did not differ markedly across the waves. Of note, we did not find any association between the absence of detectable CD4 + T cell responses and the time post COVID-19 diagnosis or disease severity. Furthermore, the magnitude of Omicron spike-specific CD4 responses mounted by those infected in wave 4 was highly comparable to those against ancestral spike (Fig. 2e), suggesting that the CD4 responses mostly target conserved epitopes in spike. Using data from the Immune Epitope Database (https://www.iedb.org), we assessed the frequency of T cell recognition of experimentally-confirmed epitopes spanning the entire spike protein. Data show that Omicron spike mutations occur in regions poorly targeted by CD4 + T cells, but are more common in regions frequently targeted by CD8 + T cells (Extended Data Fig. 7).
To gain deeper insight into the recognition of variable spike epitopes by CD8 + T cells, we also performed in silico analysis to define predicted HLA class I restriction for Omicron variable epitopes (Extended Data Table 2). Six confirmed spike epitopes containing Omicron mutations (A67V/Δ69-70, G142D/ Δ143-145, S373P, S375F, D614G, P681H and N764K) would be detrimentally affected for binding to specific class I alleles, four of which were located at positions that recorded a frequency of recognition greater than 10%. However, we also found another seven confirmed epitopes that contained Omicron mutations (T95I, S371L/S373P/S375F, K417N, G446S, Q493R, N764K and L981F) but had no effect on class I binding compared with the ancestral sequence, five of which were located at positions with a frequency of recognition greater than 10%. Overall, this suggests that although some Omicron mutations may mediate escape from specific HLA-restricted CD8 + T cells, not all mutations appear to have an impact on class I binding.

Discussion
Here we measured the ability of individuals to cross-recognize Omicron spike following vaccination, prior infection or both. We also studied unvaccinated individuals with no history of previous infection, whose first encounter with spike was with the Omicron variant. We demonstrate that vaccination and infection induce robust CD4 + and CD8 + T cell responses that largely cross-react with Omicron, consistent with recent work from our laboratory and others on limited T cell escape by Beta, Delta and other variants [22][23][24] . Despite extensive neutralization escape against Omicron 5 , 70-80% of the T cell response is cross-reactive. In contrast to neutralizing antibody epitopes, T cell epitopes are abundant and located across the entire spike protein 20 , suggesting that the majority of SARS-CoV-2 spike-specific T cell responses are directed against conserved epitopes and that SARS-CoV-2 viral evasion from T cells may be limited.
Of note, Omicron mutations appear to abolish CD8 + T cell recognition in 5 out of 32 participants (15%), in agreement with a recent report 25 . This loss of cross-reactive CD8 + T cell responses could have pathological consequences for some individuals. Further analyses are required to define specific HLA class I profiles and epitopes linked to loss of T cell responses.
T cells are crucial components of the antiviral immune response. Although they do not prevent infection, CD4 + T cells are indispensable for the generation of protective antibody responses and supporting the maturation of CD8 + T cells. Hence, given the ability of variants of concern to escape neutralization, the generation and maintenance of robust SARS-CoV-2-specific T cell responses could contribute to long-term vaccine efficacy against severe disease. Several studies have reported a waning of the neutralizing response after vaccination or infection [26][27][28] . However, humoral responses can be enhanced upon booster vaccination, improving Omicron neutralization 3,6,29,30 . Vaccine-and infection-induced T cell responses also decay after antigen clearance 31,32 , but SARS-CoV-2-specific CD8 + T cells exhibit the hallmarks of long-lived cells 33 , and T cell responses to SARS-CoV-1 infection were detectable 17 years later 34 . The longer-term durability of SARS-CoV-2-specific T cells and whether vaccine boosters can further enhance cellular immunity remain to be determined.
Despite the sharp increase in cases in South Africa in the current surge 35 , this has not translated into the expected increase in hospitalization or deaths, compared with previous waves 12 . This uncoupling of caseloads and severe outcomes could be attributed to population immunity, including maintenance of cross-reactive T cell responses observed in our study and/or intrinsic differences in Omicron severity. South Africa has high levels of SARS-CoV-2 seropositivity, driven mainly by previous infection (estimated at more than 60%) and a modest proportion of vaccinated people 36 (40%). Emerging data hint at reduced intrinsic severity of Omicron, including reduced infection of lower airway cells 37,38 . The relative contribution of high levels of immunity and potential changes in intrinsic virulence on clinical outcomes are difficult to disentangle. Moreover, it remains to be determined whether the apparently milder outcomes at a population level will be observed in other contexts with different exposure histories and vaccination coverage, or whether the higher transmissibility of Omicron and the expected massive increase in cases in a short period will offset any gains. So far, immune correlates of protection from disease are not clearly defined and large-scale prospective studies would be necessary to evaluate correlates of protection and define the role of T cell responses in disease.
Our study had several limitations. We studied Omicron crossreactivity of vaccine responses approximately one month after vaccination. Since T cell responses decline over time, the detection of continued cross-reactivity with variants over time will be related to the durability of the T cell response. Recall memory responses in vivo are likely to expand rapidly upon viral infection and contribute to limiting viral replication. We also focused on cytokine production by T helper 1 (T H 1) cells to quantify CD4 + and CD8 + T cell responses. Additional approaches such as the activation-induced marker assay may capture the cellular immune response in a more comprehensive manner 39 . The use of 15mer peptides will have underestimated SARS-CoV-2 specific CD8 + T cells, as 9mer or 10mer peptides are optimal for HLA class I binding, and it has been estimated that 15mer peptides capture 77% of the frequency of CD8 + T cells when compared with shorter peptides 40 . Moreover, the saturating concentration of peptides used in these studies may underestimate the effect of mutations on T cells. In addition, the use of peptides does not allow us to define the potential effect of mutations on antigen processing and presentation, thus underestimating the effect of Omicron mutations on T cell cross-recognition. Finally, confirmation of our results from cohorts in other geographical areas and exposure to other vaccines would offer further reassurance of
Overall, our data show that unlike neutralizing antibodies, the SARS-CoV-2 T cell responses generated upon vaccination or previous infection are highly cross-reactive with Omicron. Early reports emerging from South Africa, England and Scotland have reported a lower risk of hospitalization and severe disease compared with the previous Delta wave [9][10][11][12] . It remains to be defined whether cell-mediated immunity provides protection from severe disease and contributes to the apparent milder outcomes for Omicron. Moreover, the resilience of the T cell response demonstrated here also bodes well in the event that more highly mutated variants emerge in future.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04460-3.  Table 1a, with individual participant details presented in Extended Data Table 4.  Table 3.

Hospitalized COVID-19 patients.
Sixty-eight hospitalized COVID-19 patients were included in this study. These participants were grouped according to the time of their hospitalization, reflecting four distinct infection waves in South Africa, each dominated by a different SARS-CoV-2 strain (Fig. 2b). Wave 1, 2 and 3 participants were recruited from Groote Schuur Hospital in Cape Town (Western Cape, South Africa) and wave 4 patients were recruited from Groote Schuur Hospital and Tshwane District Hospital in Tshwane (Gauteng, South Africa). Wave 1 patients (n = 17) were enrolled between 11 June and 24 July 2020, at a time when ancestral (Wuhan-1 D614G)-related SARS-CoV-2 strains were circulating. No viral sequences are available for these patients, but we assumed that all were infected with a virus closely related to the ancestral virus, as sampling occurred almost three months before the emergence of the Beta variant in South Africa. Wave 2 patients (n = 16) were recruited between 31 December 2020 and 15 January 2021, when the Beta variant dominated. Viral sequences were available for 6 wave 2 participants, all of whom had confirmed Beta infection (GISAID accession numbers :   EPI_ISL_1040693, 1040658, 1040661, 1040685, 1040657 and 1040663). Wave 3 patients (n = 16) were recruited between 14 July and 21 July 2021. Wave 3 was dominated by the Delta variant. Viral sequences were available for 7 wave 3 participants, all of which were confirmed to be Delta infection (GISAID accession numbers: EPI_ISL_3506484, 3506367, 3957813, 3506504, 3506512 and 3506518). Wave 4 patients (n = 19) were recruited between 1 December and 15 December 2021. The SARS-CoV-2 Omicron variant was dominant during this current wave. Amongst those patients, nine had a Taqpath PCR test performed (Thermofisher), all of which were characterized by SGTF, highly suggestive of an Omicron infection. Although we did not have confirmation of Omicron for the remaining samples, they were recruited at a time when wave 4 was driven by Omicron infection (Fig. 2b; there was no concomitant Delta wave in South Africa as has occurred elsewhere), with the prevalence of Omicron in South Africa at the time of recruitment being over 90% by whole-genome sequencing (WGS). Moreover, in Tshwane, from where the remainder of the samples originated, Omicron was responsible for 98% of infections sequenced at the time of sampling (61 out of 62 samples sequenced).
All hospitalized patients from waves 1, 2 and 4 were unvaccinated at the time of sampling. Third wave participants with known vaccination status were all unvaccinated (n = 8), and the remainder (n = 8) had unknown vaccination status. Moreover, all hospitalized patients from wave 1, 2 and 4 had no clinical record of a previous symptomatic COVID-19 episode, apart from one Wave 4 participant with an unknown history. The majority of wave 3 patients had an unknown history of prior COVID-19. The demographic and clinical characteristics of hospitalized COVID-19 participants are summarized in Extended Data Table 1c, and individual patient clinical data are presented in Extended Data Table 3.

SARS-CoV-2 spike, WGS and phylogenetic analysis
WGS of SARS-CoV-2 was performed from nasopharyngeal swabs. Sequencing was performed as previously described 2 . In brief, RNA was extracted on an automated Chemagic 360 instrument, using the CMG-1049 kit (Perkin Elmer). Libraries for WGS were prepared using either the Oxford Nanopore Midnight protocol with Rapid Barcoding or the Illumina COVIDseq Assay. The quality control checks on raw sequence data and the genome assembly were performed using Genome Detective 1.133 (https://www.genomedetective.com) which was updated for the accurate assembly and variant calling of tiled primer amplicon Illumina or Oxford Nanopore reads, and the Coronavirus Typing Tool. Phylogenetic classification of the genomes was done using the widespread dynamic lineage classification method from the PANGOLIN software suite (v1.2.106) (https://github.com/hCoV-2019/pangolin).

Isolation of PBMC
Blood was collected in heparin tubes and processed within 4 h of collection. PBMC were isolated by density gradient sedimentation using Ficoll-Paque (Amersham Biosciences) as per the manufacturer's instructions and cryopreserved in freezing media consisting of heat-inactivated fetal bovine serum (FBS, Thermofisher Scientific) containing 10% DMSO and stored in liquid nitrogen until use.

SARS-CoV-2 antigens
For T cell assays on hospitalized patients, we used commercially available peptide pools (15mer sequences with 11 amino acids of overlap) covering the full length of the Wuhan-1 SARS-CoV-2 nucleocapsid, membrane and near full-length spike proteins (PepTivator, Miltenyi Biotech). For spike, we combined (1) a pool of peptides (15- Table 1. All peptides were individually resuspended in dimethyl sulfoxide (DMSO) at a concentration of 10-20 mg ml −1 . Megapools for each antigen were created by pooling aliquots of these individual peptides in the respective SARS-CoV-2 spike sequences, followed by sequential lyophilization steps, and resuspension in DMSO at 1 mg ml −1 . There were 253 peptides in the ancestral, Beta and Delta variant pool, and 254 peptides in the Omicron pool. Pools were used at a final concentration of 1 µg ml −1 with an equimolar DMSO concentration in the non-stimulated control.

Live virus neutralization assay
A live neutralization assay was performed on plasma obtained from 10 out of the 15 participants vaccinated with BNT162b2 included in this study. H1299-E3 cells were plated in a 96-well plate (Corning) at 30,000 cells per well 1 day pre-infection. Plasma was separated from EDTA-anticoagulated blood by centrifugation at 500g for 10 min and stored at −80 °C. Aliquots of plasma samples were heat-inactivated at 56 °C for 30 min and clarified by centrifugation at 10,000g for 5 min.

Statistical analysis
Statistical analyses were performed in Prism (v9; GraphPad Software). Non-parametric tests were used for all comparisons. The Kruskal-Wallis and Mann-Whitney tests were used for unmatched samples, and the Friedman and Wilcoxon tests for paired samples. P values less than 0.05 were considered statistically significant. Details of analysis performed for each experiment are described in the figure legends.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability
Complete genome sequences for the viral isolates were deposited in GISAID. Source data are provided with this paper.