Immunological memory to SARS-CoV-2 assessed for up to 8 months after infection

We recruited 188 individuals with COVID-19 for this study. Subjects (80 male, 108 female) represented a range of asymptomatic, mild, moderate, and severe COVID-19 cases (Table 1) and were recruited from multiple sites throughout the United States. The majority of subjects were from California or New York. Most subjects had a “mild” case of COVID-19, not requiring hospitalization. Ninety-three percent of subjects were never hospitalized for COVID-19; 7% of subjects were hospitalized, some of whom required intensive care unit (ICU) care (Table 1). This case severity distribution was consistent with the general distribution of symptomatic disease severity among COVID-19 cases in the United States. The study primarily consisted of symptomatic disease cases (97%, Table 1), owing to the nature of the study recruitment design. Subject ages ranged from 19 to 81 years old (Table 1). Most subjects provided a blood sample at a single time point, between 6 and 240 days PSO (Table 1), with 43 samples at ≥6 months PSO (178 days or longer). Additionally, 51 subjects in the study provided longitudinal blood samples over a duration of several months (two to four time points; Table 1), allowing for longitudinal assessment of immune memory in a subset of the cohort.

The vast majority of SARS-CoV-2–infected individuals seroconvert, at least for a duration of months (1, 2, 4, 43–45). Seroconversion rates range from 91 to 99% in large studies (44, 45). Durability assessments of circulating antibody titers in Fig. 1 were based on data ≥20 days PSO, with the plot of the best-fitting curve fit model shown in blue (see materials and methods). SARS-CoV-2 spike immunoglobulin G (IgG) endpoint enzyme-linked immunosorbent assay (ELISA) titers in plasma were measured for all subjects of this cohort (Fig. 1, A and B). Spike receptor binding domain (RBD) IgG was also measured (Fig. 1, C and D), as RBD is the target of most neutralizing antibodies against SARS-CoV-2 (4, 27, 46, 47). SARS-CoV-2 pseudovirus (PSV)–neutralizing antibody titers were measured in all subjects (Fig. 1, E and F). Nucleocapsid (N) IgG endpoint ELISA titers were also measured for all subjects (Fig. 1, G and H), as nucleocapsid is a common antigen in commercial SARS-CoV-2 serological test kits.

SARS-CoV-2 spike IgG titers were relatively stable from 20 to 240 days PSO, when assessing all COVID-19 subjects by cross-sectional analysis (half-life t_1/2 = 140 days, Fig. 1A). Spike IgG titers were heterogeneous among subjects (range 5 to 73,071; median 575), as has been widely observed (45, 47). This gave a wide confidence interval (CI) for the spike IgG t_1/2 (95% CI: 89 to 325 days). Although the antibody responses may have more complex underlying decay kinetics, the best fit curve was a continuous decay, likely related to heterogeneity between individuals. SARS-CoV-2 nucleocapsid IgG kinetics were similar to those of spike IgG over 8 months (t_1/2 68 days; 95% CI: 50 to 106 days, Fig. 1G). As a complementary approach, using paired samples from the subset of subjects who donated at two or more time points, the calculated spike IgG titer average t_1/2 was 103 days, (95% CI: 66 to 235 days; Fig. 1B) and the nucleocapsid IgG titer average t_1/2 was 68 days, (95% CI: 55 to 90 days; Fig. 1H). The percentage of subjects seropositive for spike IgG at 1 month PSO (20 to 50 days) was 98% (54 out of 55). The percentage of subjects seropositive for spike IgG at 6 to 8 months PSO (≥178 days) was 90% (36 out of 40).

Cross-sectional analysis of SARS-CoV-2 RBD IgG titers from 20 to 240 days PSO gave an estimated t_1/2 of 83 days (95% CI: 62 to 126 days; Fig. 1C). As a complementary approach, we again used paired samples, which gave an average t_1/2 of 69 days (95% CI: 58 to 87 days; Fig. 1D). The percentage of subjects seropositive for RBD IgG at 6 to 8 months PSO was 88% (35 out of 40). Thus, RBD IgG titer maintenance largely matched that of spike IgG. SARS-CoV-2 PSV neutralization titers in the full cohort largely matched the results of SARS-CoV-2 RBD IgG ELISA binding titers (Fig. 1, E and F). A one-phase decay model was the best fit (P = 0.015, F test; initial decay t_1/2 27 days, followed by an extended plateau phase, Fig. 1E), whereas a continuous decay fit gave an estimated t_1/2 of 114 days (Fig. 1E, black line). Paired time points analysis of the PSV neutralization titers gave an estimated t_1/2 of 90 days, (95% CI: 70 to 125 days; Fig. 1F). The percentage of subjects seropositive for SARS-CoV-2–neutralizing antibodies (titer ≥20) at 6 to 8 months PSO was 90% (36 out of 40). Notably, even low circulating neutralizing antibody titers (≥1:20) were associated with a substantial degree of protection against COVID-19 in nonhuman primates (24, 48). Thus, modest amounts of circulating SARS-CoV-2 neutralizing antibodies are of biological interest in humans.

SARS-CoV-2 spike IgA (Fig. 1, I and J) and RBD IgA (Fig. 1, K and L) titers were also assessed. Paired time points analysis of spike IgA titers yielded an estimated t_1/2 of 210 days (95% CI 126 to 703 days, Fig. 1J). Cross-sectional analysis of spike IgA fit a short one-phase decay model with an extended plateau phase (initial t_1/2 of 14 days, Fig. 1I). Circulating RBD IgA had an estimated initial t_1/2 of 27 days, decaying by ~90 days in most COVID-19 cases to levels indistinguishable from those of uninfected controls (Fig. 1K), consistent with observations 3 months PSO (44, 49). By paired sample analysis, long-lasting RBD IgA was made in some subjects, but often near the limit of sensitivity (LOS) (Fig. 1L).

To identify SARS-CoV-2–specific memory B cells, we used fluorescently labeled multimerized probes to detect B cells specific to spike, RBD, and nucleocapsid (Fig. 2A and fig. S1). Antigen-binding memory B cells (defined as IgD^– and/or CD27⁺) were further distinguished according to surface Ig isotypes: IgM, IgG, or IgA (Fig. 2B and fig. S1).

Cross-sectional analysis of COVID-19 subjects revealed that frequencies of SARS-CoV-2 spike-specific memory B cells increased over the first ~120 days PSO and then plateaued (pseudo–first- order model for best fit curve, R = 0.38; better fit than second-order polynomial model by Akaike’s information criterion; Fig. 2C and fig. S2A). Spike-specific memory B cell frequencies increased from the first time point (36 to 163 days) to the second time point (111 to 240 days) in paired samples from 24 of 36 longitudinally tracked donors (Fig. 2D). Spike-specific memory B cells in SARS-CoV-2–unexposed subjects were rare (median 0.0078%; Fig. 2, A and C).

RBD-specific memory B cells displayed similar kinetics to spike-specific memory B cells. RBD-specific memory B cells were undetectable in SARS-CoV-2–unexposed subjects (Fig. 2E and fig. S2C), as expected. RBD-specific memory B cells appeared as early as 16 days PSO, and the frequency steadily increased in the following 4 to 5 months (Fig. 2E and fig. S2, B and C). Twenty-nine of 36 longitudinally tracked individuals had higher frequencies of RBD-specific memory B cells at the later time point (Fig. 2F), again showing an increase in SARS-CoV-2–specific memory B cells several months after infection. About 10 to 30% of spike-specific memory B cells from SARS-CoV-2 convalescent donors were specific for the RBD domain (Fig. 2A and fig. S2B).

SARS-CoV-2 nucleocapsid-specific memory B cells were also detected after SARS-CoV-2 infection (Fig. 2A). Similar to spike- and RBD-specific memory B cells, nucleocapsid-specific memory B cell frequency steadily increased during the first ~4 to 5 months PSO (Fig. 2, G and H, and fig. S2D). Antibody affinity maturation could potentially explain the increased frequencies of SARS-CoV-2–specific memory B cells detected by the antigen probes. However, geometric mean fluorescent intensity (MFI) of probe binding was stable over time (fig. S2, I and J), not supporting an affinity maturation explanation for the increased memory B cell frequencies.

Representation of Ig isotypes among the SARS-CoV-2 spike-specific memory B cell population shifted with time (Fig. 2, I to O). During the earliest phase of memory (20 to 60 days PSO), IgM⁺ and IgG⁺ isotypes were similarly represented (Fig. 2O), but IgM⁺ memory B cells then declined (Fig. 2, M to O), and IgG⁺ spike-specific memory B cells then dominated by 6 months PSO (Fig. 2O). IgA⁺ spike-specific memory B cells were detected as a small fraction of the total spike-specific memory B cells (~5%, Fig. 2O). IgG⁺ spike-specific memory B cell frequency increased, whereas IgA⁺ frequency was low and stable over the 8-month period (Fig. 2, I to L). Similar patterns of increasing IgG⁺ memory, short-lived IgM⁺ memory, and stable IgA⁺ memory were observed for RBD- and nucleocapsid-specific memory B cells over the 8-month period (Fig. 2, O to Q, and fig. S2, E to H).

There is limited knowledge of memory B cell kinetics following primary acute viral infection in humans. A recently published SARS-CoV-2 study found RBD-specific memory B cells up to ~90 days PSO, with increasing frequencies (and a low frequency of IgA⁺ cells) (50), consistent with observations reported here. For other acute infectious diseases, we are not currently aware of other cross-sectional or longitudinal analyses of antigen-specific memory B cells by flow cytometry covering a 6+-month window after infection, except for four individuals with Ebola (51) and two individuals studied after yellow fever virus immunization (52) (we exclude influenza vaccines for comparison here, because people have numerous exposures and complex immune history to influenza). In the yellow fever study, short-lived IgM⁺ memory and longer-lasting isotype-switched memory B cells were observed in the two individuals. Overall, on the basis of the observations here, development of B cell memory to SARS-CoV-2 was robust and is likely long-lasting.

SARS-CoV-2 memory CD8⁺ T cells were measured in 169 COVID-19 subjects using a series of 23 peptide pools covering the entirety of the SARS-CoV-2 ORFeome (2, 5). The most commonly recognized open reading frames (ORFs) were spike, membrane (M), nucleocapsid, and ORF3a (CD69⁺ CD137⁺; Fig. 3A and fig. S3, A and B), consistent with our previous study (2). The percentage of subjects with detectable circulating SARS-CoV-2 memory CD8⁺ T cells at 1 month PSO (20 to 50 days) was 70% (40 out of 57, Fig. 3B). The proportion of subjects positive for SARS-CoV-2 memory CD8⁺ T cells at ≥6 months PSO was 50% (18 out of 36). This could potentially underestimate CD8⁺ T cell memory, as 15-mer peptides can be suboptimal for detection of some antigen-specific CD8⁺ T cells (53); however, pools of predicted SARS-CoV-2 class I epitope of optimal size also detected virus-specific CD8⁺ T cells in ~70% of individuals 1 to 2 months PSO, indicating consistency between the two experimental approaches (2).

SARS-CoV-2 memory CD8⁺ T cells declined with an apparent t_1/2 of 125 days in the full cohort (Fig. 3B) and t_1/2 190 days among 29 paired samples (Fig. 3C). Spike-specific memory CD8⁺ T cells exhibited similar kinetics to the overall SARS-CoV-2–specific memory CD8⁺ T cells (t_1/2 225 days for the full cohort and 185 days among paired samples; Fig. 3, D and E, respectively). Phenotypic markers indicated that the majority of SARS-CoV-2–specific memory CD8⁺ T cells were terminally differentiated effector memory cells (T_EMRA) (54), with small populations of central memory (T_CM) and effector memory (T_EM) (Fig. 3, F and G). In the context of influenza, CD8⁺ T_EMRA cells were associated with protection against severe disease in humans (55). The memory CD8⁺ T cell half-lives observed here were comparable to the 123 days t_1/2 observed for memory CD8⁺ T cells after yellow fever immunization (56). Thus, the kinetics of circulating SARS-CoV-2–specific CD8⁺ T cell were consistent with what has been reported for another virus that causes acute infections in humans.

SARS-CoV-2 memory CD4⁺ T cells were identified in 169 subjects using the same series of 23 peptide pools covering the SARS-CoV-2 ORFeome (2, 5). The most commonly recognized ORFs were spike, M, nucleocapsid, ORF3a, and nsp3 (CD137⁺ OX40⁺; Fig. 4A and fig. S4, A and B), consistent with our previous study (2). Circulating SARS-CoV-2 memory CD4⁺ T cell responses were quite robust (Fig. 4B); 42% (24 out of 57) of COVID-19 cases at 1 month PSO had >1.0% SARS-CoV-2–specific CD4⁺ T cells. SARS-CoV-2 memory CD4⁺ T cells declined with an apparent t_1/2 of 94 days in the full cohort (Fig. 4B) and 64 days among 36 paired samples (Fig. 4C). The percentage of subjects with detectable circulating SARS-CoV-2 memory CD4⁺ T cells at 1 month PSO (20 to 50 days) was 93% (53 out 57, Fig. 4B). The proportion of subjects positive for SARS-CoV-2 memory CD4⁺ T cells at ≥6 months PSO was 92% (33 out of 36).

Spike-specific and M-specific memory CD4⁺ T cells exhibited similar kinetics to the overall SARS-CoV-2–specific memory CD4⁺ T cells (whole cohort t_1/2 of 139 days and 153 days, respectively; Fig. 4, D and E, and fig. S4D). A plurality of the SARS-CoV-2 memory CD4⁺ T cells present at ≥6 months PSO had a T_CM phenotype (Fig. 4F).

T follicular helper (T_FH) cells are the specialized subset of CD4⁺ T cells required for B cell help (57) and are therefore critical for the generation of neutralizing antibodies and long-lived humoral immunity in most contexts. Thus, we examined circulating T_FH (cT_FH) memory CD4⁺ T cells, with particular interest in spike-specific memory cT_FH cells, owing to the importance of antibody responses against spike. Memory cT_FH cells specific for predicted epitopes across the remainder of the SARS-CoV-2 genome were also measured, using the MP_R megapool. Memory cT_FH cells specific for SARS-CoV-2 spike and MP_R were detected in the majority of COVID-19 cases at early time points (16 out of 17; Fig. 4, H and I, and fig. S5, A to D). cT_FH memory appeared to be stable, with almost all subjects positive for spike and MP_R memory cT_FH cells at 6 months PSO (11 out of 12 and 10 out of 12, respectively; Fig. 4, H and I). Recently activated cT_FH cells are PD-1^hi (57). Consistent with conversion to resting memory cT_FH cells, the percentage of PD-1^hi SARS-CoV-2–specific memory cT_FH dropped over time (Fig. 4J). CCR6⁺ SARS-CoV-2–specific cT_FH cells have been associated with reduced COVID-19 disease severity (5) and have been reported to be a major fraction of spike-specific cT_FH cells in some studies (5, 50, 58). Here we confirmed that a significant fraction of both spike-specific and MP_R memory cT_FH cells were CCR6⁺. We also observed increases in CCR6⁺ cT_FH memory over time (p = 0.001 and p = 0.014 at ≥6 months PSO compared to bulk cT_FH, Fig. 4K). Overall, substantial cT_FH memory was observed after SARS-CoV-2 infection, with a durability ≥6 months PSO.

Immune memory to SARS-CoV-2 was considered, including relationships between the compartments of immune memory. Males had higher spike IgG [analysis of covariance (ANCOVA) p = 0.00018, Fig. 5A] and RBD and nucleocapsid IgG (ANCOVA p = 0.00077 and p = 0.018; fig. S6, A and B), consistent with other studies (46, 47). Higher spike IgG was also observed in males when only nonhospitalized cases were considered (ANCOVA p = 0.00025, fig. S6C). By contrast, no differences were observed in IgA or PSV neutralization titers (fig. S6, D to F), and no differences were detected in SARS-CoV-2 memory B cell, memory CD8⁺ T cell, or memory CD4⁺ T cell frequencies between males and females (fig. S6, G to K).

Immune memory was examined for associations between magnitude of memory and COVID-19 disease severity. The number of previously hospitalized COVID-19 cases (n = 13) limited analysis options. However, the cases were well distributed between males and females (Table 1), data from large numbers of nonhospitalized cases were available for comparison, and the analyses in Figs. 1 to 4 demonstrated that immune memory was relatively stable over the time window analyzed. Therefore, we could simplify the disease severity analysis by grouping all samples from 120+ days PSO [also limiting data to a single sample per subject (figs. S7 to S9); most of the previously hospitalized subjects were sampled at two time points, fig. S7A] and then comparing nonhospitalized and hospitalized subjects. Spike and RBD IgG titers in hospitalized cases were higher than in nonhospitalized cases (Fig. 5B), consistent with other studies (46, 47). Spike and RBD-specific memory B cell frequencies were also higher in hospitalized cases (~1.7-fold and ~2.5-fold, respectively; Fig. 5C and fig. S8). By contrast, memory CD8⁺ T cell frequencies were not higher in hospitalized cases compared to nonhospitalized cases (Fig. 5D and fig. S9), and memory CD4⁺ T cell frequencies trended lower in hospitalized cases compared to nonhospitalized cases (Fig. 5E and fig. S9). Therefore, although our conclusions are limited by the number of hospitalized subjects, increased spike IgG titers were consistent across three independent studies, and increased memory B cells among hospitalized cases were observed here (not measured in other studies), indicating that both compartments of long-term humoral immunity to SARS-CoV-2 are higher in individuals who experienced a more severe COVID-19 disease course. T cell memory did not follow the same pattern, consistent with indications that hospitalized cases of COVID-19 can be associated with poorer T cell responses in the acute phase (5, 59). Additionally, these data show that, although gender and COVID-19 disease severity contribute to differences in immune memory to SARS-CoV-2, neither factor could account for the majority of the heterogeneity in immune memory to this virus.

Very few published data sets compare antigen-specific antibody, B cell, CD8⁺ T cell, and CD4⁺ T cell memory to an acute viral infection in the same individuals. We therefore made use of this combined data set to examine interrelationships between compartments of immune memory. We focused on RBD IgG, RBD memory B cells, spike IgA, total SARS-CoV-2–specific CD8⁺ T cells, and total SARS-CoV-2–specific CD4⁺ T cells, owing to their putative potential roles in protective immunity. The majority (64%) of COVID-19 cases were positive for all five of these immune memory compartments at 1 to 2 months PSO (Fig. 5, F and G), with the incomplete responses largely reflecting individuals with no detectable CD8⁺ T cell memory and/or poor IgA responses (Fig. 5G). At 5 to 8 months after COVID-19 infection, the proportion of individuals positive for all five of these immune memory compartments had dropped to 43%; nevertheless, 95% of individuals were still positive for at least three out of five SARS-CoV-2 immune memory responses (Fig. 5G). Immune memory at 5 to 8 months PSO represented contributions from different immune memory compartments in different individuals (Fig. 5G). Similar results were obtained if RBD IgG was replaced by neutralizing antibodies (fig. S10A). Overall, these findings again highlight heterogeneity of immune memory, with different patterns of immune memory in different individuals.

Interrelationships between the components of memory were next examined by assessing ratios between immune memory compartments over time. The ratio of SARS-CoV-2 CD4⁺ T cell memory to SARS-CoV-2 CD8⁺ T cell memory was largely stable over time (Fig. 5H and fig. S10B). Given that serological measurements are the simplest measurements of immune memory at a population scale, we examined how well such serological measurements may serve as surrogate markers of other components of SARS-CoV-2 immune memory over time. The relationship between circulating RBD IgG and RBD-specific memory B cells changed ~20-fold over the time range studied (R = 0.60, Fig. 5H and fig. S10C). The changing relationship between circulating spike IgA and RBD-specific memory B cells was even larger (R = 0.55, Fig. 5H and fig. S10D). The relationship between RBD IgG and SARS-CoV-2 CD4⁺ T cell memory was relatively flat over the time range studied (Fig. 5H); however, variation spanned a ~1000-fold range (fig. S10E). Thus, predictive power of circulating RBD IgG for assessing T cell memory was poor because of the heterogeneity between individuals (R = 0.046). In sum, although heterogeneity of immune responses is a defining feature of COVID-19, immune memory to SARS-CoV-2 develops in almost all subjects, with complex relationships between the individual immune memory compartments.

In this study, we aimed to fill gaps in our basic understanding of immune memory after COVID-19. This required simultaneous measurement of circulating antibodies, memory B cells, CD8⁺ T cells, and CD4⁺ T cells specific for SARS-CoV-2, in a group of subjects with a full range of disease, and distributed from short time points after infection to 8 months later. By studying these multiple compartments of adaptive immunity in an integrated manner, we observed that each component of SARS-CoV-2 immune memory exhibited distinct kinetics.

The spike IgG titers were durable, with modest declines in titers at 6 to 8 months PSO at the population level. RBD IgG and SARS-CoV-2 PSV–neutralizing antibody titers were potentially similarly stable, consistent with the RBD domain of spike being the dominant neutralizing antibody target. We collected data at two time points for most longitudinal individuals herein. It is well recognized that the magnitude of the antibody response against SARS-CoV-2 is highly heterogeneous between individuals. We observed that heterogeneous initial antibody responses did not collapse into a homogeneous circulating antibody memory; rather, heterogeneity is also a central feature of immune memory to this virus. For antibodies, the responses spanned a ~200-fold range. Additionally, this heterogeneity means that long-term longitudinal studies will be required to precisely define antibody kinetics to SARS-CoV-2. We are reporting the simplest statistical models that explain the data. These curve fits do not disprove more complex kinetics such as overlapping kinetics, but those models would require much denser longitudinal sampling in future studies. Biologically, IgG antibodies having a half-life of ~21 days, and the magnitude of the antibody response over time, reflect antibodies produced first by short-lived plasma cells and then long-lived plasma cells, with affinity maturation also affecting the apparent magnitude in conventional binding assays and neutralization assays. Overall, at 5 to 8 months PSO, almost all individuals were positive for SARS-CoV-2 spike and RBD IgG.

Notably, memory B cells specific for the spike protein or RBD were detected in almost all COVID-19 cases, with no apparent half-life at 5 to 8 months after infection. Other studies of RBD memory B cells report similar findings (50, 60). B cell memory to some other infections has been observed to be long-lived, including 60+ years after smallpox vaccination (61), or 90+ years after infection with influenza (62). The memory T cell half-lives observed over 6+ months PSO in this cohort (~125 to 225 days for CD8⁺ and ~94 to 153 days for CD4⁺ T cells) were comparable to the 123 days t_1/2 observed for memory CD8⁺ T cells after yellow fever immunization (56). SARS-CoV-2 T cell memory at 6 months has also now been reported in another study (63). Notably, the durability of a fraction of the yellow fever virus–specific memory CD8⁺ T cells possessed an estimated t_1/2 of 485 days by deuterium labeling (56). Using different approaches, studies determined the long-term durability of memory CD4⁺ T cells to smallpox, over a period of many years, to be an estimated t_1/2 of ~10 years (61, 64), which is also consistent with recent detection of SARS-CoV-specific T cells 17 years after the initial infection (65). These data suggest that T cell memory might reach a more stable plateau, or slower decay phase, beyond the first 8 months after infection.

Although immune memory is the source of long-term protective immunity, direct conclusions about protective immunity cannot be made on the basis of quantifying SARS-CoV-2 circulating antibodies, memory B cells, CD8⁺ T cells, and CD4⁺ T cells, because mechanisms of protective immunity against SARS-CoV-2 or COVID-19 are not defined in humans. Nevertheless, some reasonable interpretations can be made. Antibodies are the only component of immune memory that can provide truly sterilizing immunity. Immunization studies in nonhuman primates have indicated that circulating neutralization titers of ~200 may provide sterilizing immunity against a relatively high-dose URT challenge (66), and neutralizing titers of ~3400 may provide sterilizing immunity against a very high dose URT challenge (67), although direct comparisons are not possible because the neutralizing antibody assays have not been standardized (3). Conclusions are also constrained by the limited overall amount of data on protective immunity to SARS-CoV-2.

Beyond sterilizing immunity, immune responses that confine SARS-CoV-2 to the URT and oral cavity would minimize COVID-19 disease severity to that of a “common cold” or asymptomatic disease. This outcome is the primary goal of current COVID-19 vaccine clinical trials (3, 68). Such an outcome could potentially be mediated by a mixture of memory CD4⁺ T cells, memory CD8⁺ T cells, and memory B cells specific for RBD-producing anamnestic neutralizing antibodies, based on mechanisms of action in mouse models of other viral infections (69–71). In human COVID-19 infections, SARS-CoV-2–specific CD4⁺ T cells and CD8⁺ T cells are associated with less COVID-19 disease severity during an ongoing SARS-CoV-2 infection (5). Rapid seroconversion was associated with substantially reduced viral loads in acute disease over 14 days (29). Both of those associations are consistent with the hypothesis that SARS-CoV-2 memory T cells and B cells would be capable of substantially limiting SARS-CoV-2 dissemination and/or cumulative viral load, resulting in reduced COVID-19 disease severity. The likelihood of such outcomes is also closely tied to the kinetics of the infection, as memory B and T cell responses can take 3 to 5 days to successfully respond to an infection. As noted above, given the relatively slow course of severe COVID-19 in humans, resting immune memory compartments can potentially contribute in meaningful ways to protective immunity against pneumonia or severe secondary COVID-19. The presence of substerilizing neutralizing antibody titers at the time of SARS-CoV-2 exposure would blunt the size of the initial infection, and may provide an added contribution to limiting COVID-19 severity, on the basis of observations of protective immunity for other human respiratory viral infections (37, 72–74) and observations of SARS-CoV-2 vaccines in nonhuman primates (48, 67, 75).

The current study has some limitations. Longitudinal data for each subject, with at least three time points per subject, would be required for more precise understanding of the kinetics of durability of SARS-CoV-2–specific antibodies. Nevertheless, the current cross-sectional data describe well the dynamics of SARS-CoV-2 memory B cells, CD8⁺ T cell, and CD4⁺ T cell over 8 months PSO. This study was not sufficiently powered to control for many variables simultaneously. Additionally, circulating memory was assessed here; it is possible that local URT immune memory is a minimal, moderate, or large component of immune memory after a primary infection with SARS-CoV-2. This remains to be determined.

Individual case reports show that reinfections with SARS-CoV-2 are occurring (76, 77). However, a 2800-person study found no symptomatic re-infections over a ~118-day window (78), and a 1246-person study observed no symptomatic reinfections over 6 months (79). We observed heterogeneity in the magnitude of adaptive immune responses to SARS-CoV-2 persisting into the immune memory phase. It is therefore possible that a fraction of the SARS-CoV-2–infected population with low immune memory would become susceptible to reinfection relatively soon. Although gender and disease severity both contribute to the heterogeneity of immune memory reported here, the source of much of the heterogeneity in immune memory to SARS-CoV-2 is unknown and worth further examination. Perhaps heterogeneity derives from low cumulative viral load or a small initial inoculum in some individuals. Nevertheless, our data show immune memory in at least three immunological compartments was measurable in ~95% of subjects 5 to 8 months PSO, indicating that durable immunity against secondary COVID-19 disease is a possibility in most individuals.

We thank the LJI Clinical Core, specifically G. Levi and B. Schwan for healthy donor enrollment and blood sample procurement. We thank C. Moderbacher for input on data analysis. We are also grateful to the Mt. Sinai Personalized Virology Initiative for sharing banked samples from study participants with COVID-19. We are grateful to A. Wajnberg for study participant referrals and to the Personalized Virology Initiative (G. Kleiner, L.C.F. Mulder, M. Saksena, K. Srivastava, C. Gleason, C. M. Bermúdez-González, K. Beach, K. Russo, L. Sominsky, E. Ferreri, R. Chernet, L. Eaker, A. Salimbangon, D. Jurczyszak, H. Alshammary, W. Mendez, A. Amoako, S. Fabre, S. Suthakaran, M. Awawda, E. Hirsch, A. Shin) for sharing banked samples from study participants with COVID-19. Funding: This work was funded by the NIH NIAID under awards AI142742 (Cooperative Centers for Human Immunology) (A.S., S.C.), NIH contract no. 75N9301900065 (D.W., A.S.), U01 AI141995-03 (A.S., P.B.), and U01 CA260541-01 (D.W). This work was additionally supported in part by LJI Institutional Funds, the John and Mary Tu Foundation (D.S.), the NIAID under K08 award AI135078 (J.M.D.), UCSD T32s AI007036 and AI007384 Infectious Diseases Division (S.I.R., S.A.R.), and the Bill and Melinda Gates Foundation INV-006133 from the Therapeutics Accelerator, Mastercard, Wellcome, private philanthropic contributions (K.M.H., E.O.S., S.C.), and a FastGrant from Emergent Ventures in aid of COVID-19 research. This work was partially supported by the NIAID Centers of Excellence for Influenza Research and Surveillance (CEIRS) contract HHSN272201400008C (F.K., for reagent generation), the Collaborative Influenza Vaccine Innovation Centers (CIVIC) contract 75N93019C00051, the JPB foundation (F.K., V.S.), the Cohen Foundation (V.S., F.K.), and the Open Philanthropy Project (no. 2020-215611; F.K., V.S.), as well as by other philanthropic donations. We also thank all of the COVID-19 and healthy human subjects who made this research possible through their generous blood donations. Author contributions: Conceptualization, S.C., A.S., and D.W.; Investigation, J.M.D., J.M., Y.K., K.M.H., E.D.Y., C.E.F., A.G., S.H., and C.N.; Formal Analysis, J.M.D., J.M., Y.K., K.M.H., C.E.F., S.H., B.P., D.W., A.S., and S.C.; Patient Recruitment and Samples, S.I.R., A.F., S.A.R., F.K., V.S., D.M.S., and D.W.; Material Resources, F.K., V.S., V.R., E.O.S., D.W., A.S., and S.C.; Data Curation, Y.K., J.M.D., J.M., and S.H.; Writing, Y.K., J.M.D., J.M., S.I.R., D.W., A.S., and S.C.; Supervision, D.W., A.S., and S.C., Project Administration, A.F. Competing interests: A.S. is a consultant for Gritstone, Flow Pharma, Merck, Epitogenesis, Gilead, and Avalia. S.C. is a consultant for Avalia. L.J.I. has filed for patent protection for various aspects of T cell epitope and vaccine design work. Mount Sinai has licensed serological assays to commercial entities and has filed for patent protection for serological assays. D.S., F.A., V.S., and F.K. are listed as inventors on the pending patent application (F.K., V.S.), and Newcastle disease virus (NDV)–based SARS-CoV-2 vaccines that name F.K. as inventor. All other authors declare no conflict of interest. Data and materials availability: All data are provided in the supplementary materials. Epitope pools used in this paper will be made available to the scientific community upon request and execution of a material transfer agreement. This work is licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) license, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/. This license does not apply to figures/photos/artwork or other content included in the article that is credited to a third party; obtain authorization from the rights holder before using such material.