The correlates of immune protection against SARS-CoV-2 are still being defined (1, 2). Antibodies likely prevent or lessen early infection (3–5), but have little capacity to ameliorate established severe infection (6–8). The mRNA vaccines afford protection from disease after a single dose before detectable neutralizing antibodies, indicating importance of cellular immunity (2, 9). Likely both antibodies and T cells have important roles, separately or in concert.

A global Th1 T cell profile correlates with positive outcome after infection (10), including bias of virus-specific T cells (11–14). Development of virus-specific cellular immunity correlates to recovery from infection (12, 15). Given the protective role of CD8⁺ T cells in many viral infections, SARS-CoV-2-specific CD8⁺ T cells (16–19) may be particularly important for preventing severe disease. Antiviral CD8⁺ T cell frequency and breadth have not been clearly defined in most studies, most of which have used single pools of predicted epitopes from across the proteome and/or unseparated PBMC for qualitative evaluations.

SARS-CoV-2 vaccines protect against serious illness and death (20, 21). There have been increasing observations of vaccinated persons becoming infected, associated with viral spike mutations mediating antibody resistance (22–25). Despite these breakthrough infections, vaccination still protects against severe illness or death, further underscoring the importance of cellular immunity (9, 26). To date, however, CD8⁺ T cell responses against the vaccine have not been compared in detail to natural infection.

Here we confirm the overall Th1 profile of SARS-CoV-2-specific T cells and examine the targeting of spike, nucleocapsid, matrix, and envelope proteins by the CD8⁺ T cell subset in persons recovered from recent SARS-CoV-2 infection. The stability of these responses is evaluated, as well as boosting after vaccination. In SARS-CoV-2-naïve individuals, vaccine-elicited antiviral CD8⁺ T cell targeting is compared to that from natural infection.

The protective contribution of SARS-CoV-2-specific CD8⁺ T cells is increasingly apparent (16–19), consistent with their known importance for clearing infected cells in other viral infections. Most studies have focused on the phenotypic characteristics of cellular immunity in bulk, such as cytokine production in response to pooled predicted epitope peptides, without detail on targeting of responses. We confirm Th1 bias in both CD4⁺ and CD8⁺ virus-specific T cells (11–14), using intracellular staining for IFN-γ, IL-2, IL-4, IL-17, and IL-10.

For both T cell subsets IFN-γ production dominated, followed by IL-2, with more CD4⁺ T cells than CD8⁺ T cells producing IL-2. Most IFN-γ-producing cells did not produce IL-2, particularly the CD8⁺ subset, supporting prior findings (35–37). Minimal production of IL-10 or IL-17 were observed; IL-10 (38) and IL-17 (39) production have been linked to disease progression, but our participants were mostly limited to those who recovered from mild illness.

A few individuals had significant populations of IL-4-producing CD8⁺ T cells of unclear significance, not previously reported in SARS-CoV-2 infection to our knowledge. IL-4-producing CD8⁺ T cells have been suggested to be noncytolytic helper cells that do not produce IFN-γ (40), associated with humoral immunity in old age (41), asthma in children (42), and autoimmune arthritis (43). Whether they might play a protective (anti-inflammatory), pathogenic (immunosuppressive), or mixed role in COVID-19 is unclear.

Our observations agree observations that most infected persons develop SARS-CoV-2-targeted cellular immune responses (36, 44–47). By intracellular cytokine staining flow cytometry with spike/CD4⁺ epitope megapools (27), 72% and 80% of persons had IFN-γ and/or IL-2 responses in the CD4⁺ and CD8⁺ T cell subsets, respectively. Considering all tested cytokines and both tested peptide megapools, all persons had both CD4⁺ and CD8⁺ T cell responses (not shown). By IFN-γ ELISpot for spike, nucleocapsid, matrix, and envelope, 93% of persons had detectable CD8⁺ T cell responses against at least one protein.

Others have observed preferential targeting of structural proteins (12, 14, 27, 35, 36, 44, 45, 48–50), using either predicted epitopes (12, 27, 45, 48, 50) or overlapping peptides (14, 35, 36, 44, 49) in pools. Two studies from Le Bert et al. (14) and Peng et al. (36) used multiple smaller pools in IFN-γ ELISpot assays to assess targeting breadth, but did not separate CD4⁺ from CD8⁺ T cell populations. Le Bert et al. found that responses against matrix were most targeted, followed by those against spike or nucleocapsid (envelope was not tested) (14), but Peng et al. found that spike was most highly targeted, while nucleocapsid and matrix were similar, and many persons had envelope targeting (36). However, the contributions of CD4⁺ versus CD8⁺ T cells to these patterns were not defined.

We follow up in greater detail using isolated CD8⁺ T cells with small pools of overlapping peptides spanning spike, nucleocapsid, matrix, and envelope in IFN-γ ELISpot assays of infected participants. In contrast to Le Bert et al. and Peng et al. using unseparated PBMC, we clearly identify nucleocapsid as the dominant target of CD8⁺ T cells. Matrix targeting is quantitatively similar to spike, but more densely targeted, and we observe no targeting of envelope. Similar to Le Bert et al. using unseparated PBMC, we find that targeting distributed across spike, and most persons target at least one spike epitope. Moreover, we observe modestly higher frequency of receptor binding domain (RBD) targeting, accounting for about a third of responses despite being about a sixth of spike. Even so, the few mutations mostly in the RBD defining various spike variants seem unlikely to affect recognition by CD8⁺ T cells (51). Overall, differences from the findings of Le Bert et al. and Peng et al. may result primarily from evaluation of isolated CD8⁺ T cells versus bulk PBMC (in which CD4⁺ T cells typically predominate); other studies grossly comparing CD4⁺ T cells to CD8⁺ T cells have shown that the former tend to predominate (12, 46, 49, 52–54), and that the two do not correlate (55). Finally, our results suggest that nucleocapsid might be a useful target for vaccine inclusion to elicit broader and more durable CD8⁺ T cell responses, which is also supported by a recent study suggesting that immunodominant targeting of nucleocapsid is associated with better outcome after SARS-CoV-2 infection (56).

Various studies reported differing results on longevity of virus-specific T cell responses (CD4⁺, CD8⁺ or combined) after SARS-CoV-2 infection, with observations of both persistence (37, 45, 47, 53–55, 57, 58) and decay (14, 47, 59, 60). We observed decay of CD8⁺ T cell responses similar to our observations of anti-RBD antibodies (29, 61), and most rapid waning of matrix responses, in agreement with Le Bert et al. (14). The reasons for discrepant findings between studies are unclear, but may relate to methodologies (intracellular cytokine staining versus ELISpot), CD4⁺ versus CD8⁺ versus unseparated subsets, targeting of responses measured, and differences in illness severity. As opposed to antibodies (and B cell memory) being required for immediate viral neutralization during exposure or early infection, T cells probably have a more prolonged effector role in containing and clearing infection, so it is unclear whether our observed peripheral blood decay of the CD8⁺ T cell response is functionally relevant. Data showing ongoing vaccine protection from severe illness or death despite waning protection from infection (26, 62) are further evidence that the falling level of circulating antiviral CD8⁺ T cells we observe do not preclude an effective recall response.

Some groups have observed either correlation (36, 63) or lack of correlation (47) of T cell responses to antibodies. We saw no correlation of either CD4⁺ or CD8⁺ T cell responses by intracellular cytokine staining, or CD8⁺ T cell responses by IFN-γ ELISpot to anti-RBD antibodies. These discrepancies again may be related to methodologies or clinical characteristics of the participants. Of note, our cohort included mostly persons who had mild infection (not requiring supplemental oxygen or hospitalization). Because antibody levels vary greatly by disease severity (29, 64–66), our dynamic range might have been too limited to see a correlation.

Our data address spike targeting induced by vaccination. While initial pilot studies of the two mRNA vaccines demonstrated cellular immunity (67, 68), these were measurements of whole PBMC by intracellular cytokine staining using a single peptide pool. We provide greater detail, showing vaccination elicits an average of over four targeted epitopes with summed frequency over 1,000 per million CD8⁺ T cells. While this breadth and frequency was higher than we observed for natural infection, our measurement after vaccination was at the peak, and infected persons were assessed past the peak during infection. A prior study provided results consistent with our suggestion that vaccination boosts prior memory responses against spike; in persons receiving the BNT162b2 vaccine, those with prior infection reached a level of spike targeting after a single dose that was attained after two doses in persons without prior infection (69). Qualitatively, however, we observed that vaccination and infection generated similar spike responses.

Our study has caveats. Evaluation of cytokine production and T cell “polyfunctionality” was limited to few cytokines and performed with too few PBMC for accurate quantitation or sensitivity below ~0.01%. Our evaluation of CD8⁺ T cells utilized a cell-sparing expansion method, although results with expanded cells have been shown to correlate well to bulk unexpanded CD8⁺ T cells (30, 34). Responses were evaluated only against spike, nucleocapsid, matrix, and envelope, and could miss dominant responses against other proteins. We did not map to the level of individual peptides/epitopes; thus, we likely underestimated the breadth and depth of targeting. The time points after infection were not frequent enough for precise estimation of decay rates, but only provide minimal boundaries for decay. Most of the COVID-19 participants had had mild illness, and there were too few severely ill subjects for comparisons. Finally, the numbers of participants were too small to compare responses between different vaccines.

In summary, we find a Th1-biased IFN-γ dominant cellular immune response after SARS-CoV-2 infection in both CD4⁺ and CD8⁺ subsets, although some persons have an unusual IL-4-producing CD8⁺ T cell population of unclear significance. CD8⁺ T cells predominately target nucleocapsid and those responses appear to be more durable compared to targeting of spike or matrix; no responses were seen against envelope. Vaccination of previously infected persons specifically boosts memory responses against spike, and generates new responses in previously uninfected persons that resemble those from infection. These results provide greater clarity on CD8⁺ T cell targeting, breadth, and persistence. Inclusion of nucleocapsid in vaccines may allow even broader and longer-lived cellular immune protection against COVID-19 to combat ongoing viral evolution in the pandemic.

The original contributions presented in the study are included in the article/ Supplementary Material , further inquiries can be directed to the corresponding author.

The studies involving human participants were reviewed and approved by UCLA Institutional Review Board. The patients/participants provided their written informed consent to participate in this study.

Overall study conceptualization: OY. Study design: ET, CH, FI, JF, PK, KF, NT, AR, GA, and OY. Conducting experiments: ET, CH, FI, and MH. Data analysis: ET, CH, FI, and OY. Providing reagents: JF, KF, NT, AR, and GA. Primary writing of the manuscript: ET and OY. Reviewing and revising the manuscript: ET, CH, FI, MH, JF, PK, KF, NT, AR, GA, and OY. All authors contributed to the article and approved the submitted version.

Funding was provided by AIDS Healthcare Foundation and private philanthropic donors (including William Moses, Mari Edelman, Beth Friedman, Dana and Matt Walden, Kathleen Poncher, Scott Z. Burns, and Gwyneth Paltrow and Brad Falchuk), with additional infrastructure support from the UCLA AIDS Institute Center for AIDS Research (NIH grant AI028697), James B. Pendleton Trust, and McCarthy Foundation. Ancillary support was provided by Thermo Fisher (represented by Russ Pong), who provided access to the Attune Flow Cytometer and gifted fluorescent tagged antibodies. We appreciate the collaboration of Lisa Kelly and Irene Trovato with HyClone products from Cytiva (Logan, UT, www.Cytiva.com). The funders were not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.

The 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.