SARS-CoV-2 infection generates tissue-localized immunological memory in humans

Description SARS-CoV-2–specific memory cells persist in multiple sites months after infection, particularly in lungs and associated lymph nodes. Adaptive immune responses to SARS-CoV-2 infection have been extensively characterized in blood; however, most functions of protective immunity must be accomplished in tissues. Here, we report from examination of SARS-CoV-2 seropositive organ donors (ages 10 to 74) that CD4+ T, CD8+ T, and B cell memory generated in response to infection is present in the bone marrow, spleen, lung, and multiple lymph nodes (LNs) for up to 6 months after infection. Lungs and lung-associated LNs were the most prevalent sites for SARS-CoV-2–specific memory T and B cells with significant correlations between circulating and tissue-resident memory T and B cells in all sites. We further identified SARS-CoV-2–specific germinal centers in the lung-associated LNs up to 6 months after infection. SARS-CoV-2–specific follicular helper T cells were also abundant in lung-associated LNs and lungs. Together, the results indicate local tissue coordination of cellular and humoral immune memory against SARS-CoV-2 for site-specific protection against future infectious challenges.


INTRODUCTION
Ending the global coronavirus disease 2019 (COVID-19) pandemic caused by the novel coronavirus SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2) depends on the establishment of immunological memory. SARS-CoV-2 infects the respiratory tract and induces adaptive immune responses, resulting in virus-specific T and B lymphocytes mediating viral clearance at the infection site and inhibiting viral dissemination through T cell effector functions and antibodies. It is now well documented that mild and severe infection generates circulating virus-specific T cells and antibodies detectable in peripheral blood for up to a year or more (1)(2)(3)(4)(5)(6)(7)(8). Moreover, the presence of neutralizing antibodies specific for the viral spike (S) protein correlates with protection for SARS-CoV-2 vaccines (9,10). However, the emergence of viral variants with potential for immune evasion (11)(12)(13)(14)(15) and variability in vaccination rates among populations enable ongoing SARS-CoV-2 spread. An understanding of the breadth and functional potential of virus-specific T and B cell memory is needed for developing improved strategies to protect against continually evolving strains.
A major limitation in studying human immune responses is that sampling is largely confined to peripheral blood, whereas adaptive immune responses are generated and carry out their protective functions in a range of tissues. Memory cells are also maintained in diverse tissues-including infection sites and lymphoid organs [reviewed in (16,17)]. Virus-specific memory CD4 + and CD8 + T cells comprise heterogeneous subsets of circulating subsets and noncirculating tissueresident memory (T RM ) cells in various sites (18,19). In mouse models, T RM cells in the lung mediate optimal protective responses to respiratory infection (20)(21)(22), and this localized protection also involves responses in lung-associated lymph nodes (LNs) (23,24). In humans, the majority of T cells in adults are memory cells. The subset composition of human T cell memory is specific to the tissue site; in mucosal, exocrine, and barrier sites, T RM cells predominate, whereas lymphoid sites contain circulating effector and central memory (T EM and T CM ) along with T RM subsets (19,25,26). All types of memory T cells in tissues express gene expression signatures distinct from memory T cells in peripheral blood (27), suggesting that they are differentially maintained compared with circulation. In SARS-CoV-2 infection, activated T EM and T RM were identified in the airways of severe COVID-19 (28,29), although the distribution and functional capacity of SARS-CoV-2-specific memory T cells across tissues remain uncharacterized.
The generation of memory B cells following an infection occurs in secondary lymphoid organs (LNs and spleen) and requires virusspecific follicular helper T (T FH ) cells, which promote B cell differentiation, survival, and somatic hypermutation in germinal centers (GCs) (30). Memory B cells can persist in multiple sites and exhibit tissue-resident phenotypes (31). In mouse respiratory infection models, resident memory B (B RM ) cells in lung and lung-associated LNs can be important for protection (32). In human lymphoid and mucosal sites, memory B cells are the predominant subset, whereas naïve B cells prevail in circulation (33). SARS-CoV-2 infection elicits generation of S-, receptor binding domain (RBD)-, and nucleocapsid (N)-specific memory B cells detectable in peripheral blood (1,3,7); however, the distribution and maintenance of SARS-CoV-2-specific memory B cells and GC B cells in tissues following natural infection have not been reported. Moreover, the relationship between human B and T cell memory in tissues is largely unexplored.
The use of physiologically healthy tissues from organ donors has enabled study of human immune cells across multiple sites (16,34,35). Investigating tissue immunity to SARS-CoV-2 is particularly challenging, as previously infected, but unvaccinated, donors are required. Here, we present an investigation of SARS-CoV-2specific memory T and B cell populations in lymphoid and mucosal sites of previously infected, seropositive organ donors, which we identified in the bone marrow (BM), spleen, lung, and LNs up to 6 months after infection. Lung and lung-associated LNs were the most prevalent sites for SARS-CoV-2-specific memory T and B cells, with a proportion exhibiting tissue-resident profiles. We also detected virus-specific GC B cells in lung-associated LNs along with T FH , suggesting ongoing generation of humoral immunity. Together, our results reveal local coordination of cellular and humoral memory immune responses for site-specific protective immunity.

Organ donor cohorts for analysis of SARS-CoV-2-specific immune responses in tissues
We have established a human tissue resource for obtaining multiple tissues from organ donors through collaborations with organ procurement organizations (OPOs) (16,34,35). Use of organ donor tissue allows for rapid isolation of live immune cells for functional analysis, thus enabling assessment of immune responses in multiple sites within an individual. We identified four organ donors ages 10 to 74 years with previous SARS-CoV-2 infection (Fig. 1A) who died of noninfectious-related causes and were SARS-CoV-2 polymerase chain reaction (PCR) negative at the time of organ procurement. Previous SARS-CoV-2 infection history was based on post-procurement testing for antibodies to N protein (see Materials and Methods) and/or a confirmed history of COVID-19 2 to 6 months previously (table S1). Controls were prepandemic seronegative organ donors procured before November 2019 who also died of noninfectious causes (table S1).
All seropositive donors had detectable serum immunoglobulin G (IgG) to N, S, and RBD, along with SARS-CoV-2-specific neutralizing antibodies, consistent with antibody responses generated from acute infection (36,37), whereas serum from prepandemic organ donors lacked antibodies to SARS-CoV-2 ( Fig. 1, B and C). Using this cohort, we examined SARS-CoV-2-specific T and B cell responses across the blood, BM, spleen, lung, lung-associated LNs, as well as gut-associated LNs of seropositive and seronegative donors.

SARS-CoV-2-specific T cells in lung and lymphoid tissues
SARS-CoV-2-specific T cells in different sites from seropositive and seronegative organ donors were measured based on expression of T cell receptor-dependent activation-induced markers (AIMs) after stimulation with SARS-CoV-2-specific peptide megapools (MPs), which enable simultaneous presentation of a large number of virusspecific epitopes (see fig. S1 and Materials and Methods) (2,38,39). Mononuclear cells from each site were stimulated for 24 hours in vitro with four different SARS-CoV-2-specific MPs: MP_S (containing overlapping peptides representing the entirety of S), MP_CD4_R [containing predicted human leukocyte antigen (HLA) class II viral epitopes minus S], and MP_CD8_A and MP_CD8_B (each containing predicted HLA class I epitopes from all viral proteins) (2). SARS-CoV-2-specific CD4 + T cells were identified on the basis of coexpression of two or more of the following three AIMs [OX40, 4-1BB, and/or CD40 ligand (CD40L)] ( Fig. 1D and fig. S2A), whereas virus-reactive CD8 + T cells were identified on the basis of coexpression of CD25 and 4-1BB ( Fig. 1E and fig. S2B). Quantification of SARS-CoV-2-specific CD4 + and CD8 + T cell responses was based on dimethyl sulfoxide (DMSO) background-subtracted frequencies of AIM + CD4 + and CD8 + T cells (Fig. 1, F and H).
For CD4 + T cells, significant responses to S protein were found in all sites examined (blood, BM, spleen, lung, lung-associated LNs, and gut-associated LNs) relative to prepandemic control samples (Fig. 1F, left). For non-S SARS-CoV-2 epitopes, there were significant CD4 + T cell frequencies in the BM, lung-associated LNs, and gutassociated LNs (Fig. 1F, right). Total SARS-CoV-2-specific CD4 + T cell responses largely reflected the pattern observed with S-specific responses (Fig. 1, F and G). SARS-CoV-2-specific CD8 + T cell frequencies were generally lower in magnitude than for CD4 + T cells and more variable between donors (Fig. 1, G to I). Significant SARS-CoV-2specific CD8 + T cell frequencies above controls were observed for lung-and gut-associated LNs for class I epitopes (Fig. 1, H and I). Comparing across all donors and sites, CD4 + T cells responding to S protein epitopes were the predominant SARS-CoV-2-specific T cells across tissue sites for all individuals (Fig. 1, F and G, and fig. S2C). Although the distribution patterns of SARS-CoV-2-specific T cell responses across tissue sites varied between donors ( Fig. 1 and fig. S3), the lung and lung-associated LNs were most consistently the dominant sites for virus-specific CD4 + and CD8 + T cells ( fig. S2D). These results indicate that SARS-CoV-2 infection generates virus-specific T cell responses across blood, multiple lymphoid sites, and lungs, with higher frequencies localized in lung tissue and lung-associated LNs.

SARS-CoV-2-specific T cells are maintained in tissues as circulating and resident memory subsets
We analyzed the subset distribution of SARS-CoV-2-specific T cells based on coordinate expression of CD45RA and CCR7, defining T CM cells (CD45RA − CCR7 + ), T EM cells (CD45RA − CCR7 − ), terminally differentiated effector T (T EMRA ) cells (CD45RA + CCR7 − ), and naïve or stem-like memory cells (CD45RA + CCR7 + ) (40,41). Each tissue had a distinct T cell subset composition that is conserved between individuals as we previously determined (26,35,42); T cell phenotypes for each site and for SARS-CoV-2-specific T cells are shown in representative flow cytometry plots ( Fig. 2A). The majority of SARS-CoV-2-specific CD4 + T cells were maintained as T EM (≥75%) in the blood and lung and as T EM or T CM (≥80%) in lymphoid sites (Fig. 2B). For SARS-CoV-2-specific CD8 + T cells, the majority were maintained as T EM and T EMRA cells (≥50%) for all sites; the proportion of T EMRA was higher than T EM for the BM, spleen, and lung, whereas T EM were more prevalent than T EMRA for LNs (Fig. 2C). Donors D495 and D498 had a particularly high proportion of SARS-CoV-2-specific CD8 + T EMRA cells in the lung, whereas the pediatric donor (HDL113) harbored more SARS-CoV-2-specific CD45RA + CCR7 + CD8 + T cells in the lung and lung-associated LN (Fig. 2C).
We also analyzed coexpression of residency markers CD69 and CD103 by SARS-CoV-2-specific T cells (see fig. S1 for gating), as assessment of CD69 alone as a T RM marker was confounded because of CD69 up-regulation by T cell receptor-stimulated T cells. Virusspecific CD69 + CD103 + memory CD4 + and CD8 + T cells (T RM ) were mostly confined to the lung, whereas lower frequencies of SARS-CoV-2-specific CD8 + T RM were also detected in LNs (Fig. 2, D and E). . n = 10 for seronegative donors (n = 4 for BM, LLN, and GLN; n = 3 for blood, spleen, and lung). Statistical analysis was performed using one-way analysis of variance (ANOVA), corrected for multiple comparisons by false discovery rate (FDR) using two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli. *q ≤ 0.05, **q ≤ 0.01, ***q ≤ 0.001, and ****q ≤ 0.0001. Datasets were log-transformed before statistical analysis.
Together, these results show that SARS-CoV-2-specific T cells are maintained across diverse tissue sites as memory T cells, with a portion of T cells persisting as tissue-resident populations, particularly within the lung.

Tissue specificity and heterogeneity of functional responses to SARS-CoV-2
The functional responses of SARS-CoV-2-responding cells from different sites were assessed by multiplex quantification of 50 immune mediators from culture supernatants of peptide-stimulated mononuclear cells as in Fig. 1. Blood and tissues exhibited distinct functional profiles, and the magnitude of responses varied between donors ( Fig. 3A and fig. S4). There was heterogeneity between individuals in the distribution of functional responses across sites. In particular, SARS-CoV-2-specific functional responses were primarily located in the lung for donor D492 and in the lung-associated LN and blood for donor D495, whereas D498 and HDL113 exhibited a broad functional response across multiple sites (Fig. 3A  Memory subset differentiation and residency marker analysis were conducted on tissue sites for which number of SARS-CoV-2-specific T cells was ≥5 based on AIM assays. SARS-CoV-2-specific CD8 + T cells in the blood were not detected above this threshold. n = 4 SARS-CoV-2 seropositive donors (n = 4 for blood, lung, and LLN; n = 3 for spleen and GLN; n = 2 for BM). n = 10 for seronegative donors (n = 4 for BM, LLN, and GLN; n = 3 for blood, spleen, and lung). Statistical analysis was performed using one-way ANOVA, corrected for multiple comparisons by FDR using twostage linear step-up procedure of Benjamini, Krieger, and Yekutieli. No comparisons were statistically significant. Datasets were log-transformed before statistical analysis. SARS-2, SARS-CoV-2; T EMRA , terminally differentiated effector T cell; Bld, blood; SP, spleen.
Concentration (normalized to DMSO; max absolute scaled per column) elderly donors (D492 and D498), lung-associated LN responses were much weaker compared with other tissue sites, even when the frequency AIM + CD4 + and CD8 + T cells in the lung-associated LN were high relative to other tissue sites ( Fig. 3A and fig. S3). Overall, these data show that virus-specific functional responses are differentially maintained across sites and individuals. Tissue-specific functional profiles were apparent in the profile of soluble mediators produced in response to stimulation with SARS-CoV-2 peptide pools (Fig. 3, B and C). The functional responses in the LN were the most diverse and included type 1 proinflammatory cytokines and cytolytic mediators [interferon- (IFN-), tumor necrosis factor- (TNF-), granzyme B, perforin, and granulocytemacrophage colony-stimulating factor (GM-CSF)], type 2 cytokines [interleukin-5 (IL-5), IL-9, and IL-13], and type 3 cytokines [IL-17A, IL-17F, and granulocyte colony-stimulating factor (G-CSF)]-most at increased levels compared with other sites (Fig. 3, B and C). The functional response in the lung was distinct from other sites and included proinflammatory profiles (TNF-, perforin, granzyme, IL-12, IL-17A, and G-CSF), IL-10 associated with regulation of inflammation during respiratory infections (43), and higher levels of IL-6 and the homeostatic cytokine IL-15 compared with other sites (Fig. 3, B and C). BM responses were mostly TNF-, perforin, granzyme B, and IL-10 and blood responses were similar to BM (Fig. 3, B and C). Comparing the immune mediator milieu between sites revealed that certain cytokines are produced across sites, whereas others are distinct to specific sites (Fig. 3C). Together, these results indicate that SARS-CoV-2-specific memory T cells in different sites exhibit distinct functional responses to viruses that are likely adapted to the site and contribute to a multifaceted protective response.

SARS-CoV-2-specific memory B cells and resident phenotypes in tissues
To characterize the nature of memory B cell responses to SARS-CoV-2 in tissues, we used fluorescently labeled, biotinylated, and multimerized probes of full-length S and RBD proteins to detect antigen-binding B cells among IgM + , IgG + , or IgA + memory B cells ( fig. S5A), as previously described (1). SARS-CoV-2-specific memory B cells were detected at frequencies substantially higher in seropositive donors than seronegative donors in all tissues examined, including the lungs, BM, spleen, lung-associated LNs, and gut-associated LNs (Fig. 4, A and B). IgG + was the dominant isotype of SARS-CoV-2-specific memory B cells in almost all samples, although IgM + and IgA + memory B cells were present (Fig. 4C). For D498, whereas few SARS-CoV-2-specific memory B cells were IgM + in the LNs, >40% were IgM + in the BM and spleen. For D495, whereas almost no IgA + SARS-CoV-2-specific memory B cells were found in the lung, ~25% were IgA + in gut-associated LNs (Fig. 4C). SARS-CoV-2-specific memory B cells were present at significantly higher frequencies in lung and lung-associated LNs than in the spleen or gut-associated LNs (Fig. 4, B and D). The highest proportion of SARS-CoV-2-specific IgG + memory B cells was found in the lung and lung-associated LNs (Fig. 4D).
Memory B cells can persist as tissue-resident cells (B RM ) in lymphoid or nonlymphoid tissues and are identified by CD69 expression (32,33,44). In tissues of seropositive and seronegative donors, we detected significant populations (>50 to 75%) of CD69 + B cells comprising >50% of total B cells in the lungs and LNs, whereas the BM and spleen contained much lower frequencies (0 to 10%) of CD69-expressing B cells ( fig. S5, B and C). Substantial frequencies of SARS-CoV-2-specific memory B cells exhibited CD69 expression indicative of tissue-resident profiles in lungs (50 to 80%) and LNs (20 to 40%) (Fig. 4, E and F). By contrast, negligible frequencies (<3%) of CD69 + SARS-CoV-2 S/RBD-specific memory B cells were detected in the BM of organ donors (Fig. 4, E and F), and in previously obtained peripheral blood samples from convalescent individuals with COVID-19 ( fig. S5, D to I), consistent with previous analysis of polyclonal B cells in these sites (33). Together, these results provide direct evidence for human antigen-specific B RM in lungs and LNs that are distinct from corresponding populations in the blood. Thus, SARS-CoV-2 infection leads to the preferential formation and/or retention of antigen-specific B cell memory within lungs and lung-associated LNs, with CD69 + tissue-resident cells representing the majority of the SARS-CoV-2 S-specific memory B cells in lungs.

SARS-CoV-2-specific GC B cells and T FH cells in LNs
GCs within lymphoid organs are important microanatomical sites in which activated B cells receive cognate help from T FH to undergo somatic hypermutation to evolve higher affinity antibody recognition of pathogens (45). While previous studies have demonstrated affinity-matured SARS-CoV-2-specific memory B cells in the blood (3), BM plasma cells (46), and circulating virus-specific T FH cells (47), which all indicate GC responses in COVID-19 (48), direct evidence of SARS-CoV-2 antigen-specific GCs (and S-specific GC B cells, in particular) is lacking. We identified GC B cells by assessing coexpression of Bcl6 [a transcription factor required for GC B cell differentiation (49)] and Ki67 (a marker of active cellular proliferation) among total CD19 + B cells (fig. S5A). The frequency of GC B cells (Bcl6 + Ki67 + CD19 + B cells) trended higher in lung-associated LNs of seropositive donors than seronegative donors (q = 0.053; Fig. 5, A and B). SARS-CoV-2-specific GC B cells identified based on binding to S and RBD proteins were identified in lung-associated LNs of three of four seropositive donors and in gut-associated LNs of one donor (Fig. 5, C and D, and fig. S5J). Virus-specific GC B cells were not detected in the BM, spleen, or lung of seropositive donors or in any tissues of seronegative donors (Fig. 5, C and D). These results provide direct evidence that SARS-CoV-2specific GC responses are induced by SARS-CoV-2 infection and are maintained in lung-associated LNs after resolution of infection. In addition, long-lasting GCs can even be generated in human gut-associated LNs.
T FH cells can be identified by coexpression of CXCR5 and PD-1 (50). T FH -phenotype cells (CXCR5 + PD-1 + CD4 + T cells) were found at low frequencies (5 to 10%) in lymphoid sites (spleen, lung-associated LNs, and gut-associated LNs) and in even lower frequencies in BM and most lungs of seronegative and seropositive donors ( fig. S6), demonstrating the relative rarity of this population among total T cells. However, SARS-CoV-2-specific T FH cells (identified by AIMs as in Fig. 1) were found in multiple sites of seropositive donors comprising 20 to 50% of SARS-CoV-2-reactive CD4 + T cells in LNs and lower but significant frequencies in the lung, spleen, and BM (Fig. 5, E to G). In one seropositive donor, more than 80% of virusspecific CD4 + T cells in the lung were T FH , representing ~4% of nonnaïve (NN) SARS-CoV-2-specific CD4 + T cells, consistent with the higher overall frequency of T FH cells in the lungs of that donor (Fig. 5,  F and G, and fig. S6). Together, the SARS-CoV-2-specific GC B cell and T FH data indicated robust GC responses to SARS-CoV-2 infection, distributed among lymphoid tissues and sites of viral infection, with some GCs being active for months after infection.

Coordinated adaptive immunity across tissues The identification of SARS-CoV-2-specific memory T and B cells in multiple tissues, along with T FH and GC B cells in LNs, suggested
site-directed coordination of cellular and humoral immunity. To identify potential associations between SARS-CoV-2-specific lymphocyte populations across sites, we performed an exploratory correlation analysis (Fig. 6 and fig. S7). The frequencies of S-specific CD4 + and CD8 + T cells (but not total SARS-CoV-2-specific T cells; fig. S7A) were positively correlated across tissue sites (P = 0.0116; Fig. 6A). Significant associations also emerged between SARS-CoV-2specific B and T cells (Fig. 6, B to F, and fig. S7, B to K). SARS-CoV-2specific CD4 + T cell frequencies correlated positively with S/RBD-specific memory B cells across all tissues (P = 0.0009), as well as IgG + (P = 0.0009) and IgA + (P = 0.0203) subpopulations (Fig. 6, B and C, and fig. S7, D and E). For tissue-resident SARS-CoV-2-specific lymphocytes, CD69 + CD103 + CD8 + T RM cells correlated with CD69 + B RMboth as frequencies of total lymphocytes (P = 0.0069) and as fractions within their respective antigen-specific populations (P = 0.01) (Fig. 6E  and fig. S7J). Positive correlations were also observed between CD4 + T RM and B RM across all tissues (P = 0.0198; fig. S7I). For follicular responses, S/RBD-specific memory B cells correlated with SARS-CoV-2-specific T FH cells across tissues (P = 0.0236; Fig. 6F and fig. S7K).
Given that lung and lung-associated LNs contained the highest frequencies of SARS-CoV-2-specific lymphocytes, we performed a targeted correlation matrix analysis to identify potential associations within and between these sites (Fig. 6G). SARS-CoV-2-specific T FH cells in the lung-associated LNs were significantly associated with lung memory B cells, whereas SARS-CoV-2-specific CD4 + T cell frequencies positively correlated with SARS-CoV-2-specific GC B cells lung-associated LNs. Conversely, certain LN and lung populations were inversely correlated. In particular, the amount of S/RBD-specific GC B cells in the lung-associated LNs negatively correlated with S-specific CD4 + T cells in the lung. Similarly, S/RBDspecific memory B cell frequencies in the lung-associated LNs negatively correlated with SARS-CoV-2-specific and S-specific CD8 + T cell frequencies in the lung. Similar correlations were observed when including blood and plasma antigen-specific lymphocyte populations in correlation matrix analyses ( fig. S8). Together, these results suggest opposing or compensatory effects of humoral and cellular immune responses in lung-associated LNs and lungs.

Immunological memory is maintained by heterogeneous subsets of virus-specific T and B cells in nonlymphoid tissue sites of infection
and multiple lymphoid organs. A comprehensive assessment of memory responses is therefore difficult to accomplish in humans. Here, we reveal the cellular complexity and functional diversity of SARS-CoV-2-specific memory T and B cells in lymphoid and mucosal tissues of previously infected organ donors up to 6 months after infection (see fig. S9 for summary schematic). SARS-CoV-2-specific CD4 + T, CD8 + T, and B cells predominantly localized in the lung and lungassociated LNs and were maintained as memory cell populations. Tissue-resident T and B cells, known to participate in protection against secondary viral infections, were found most abundantly in the lung and were correlated across multiple sites. Moreover, SARS-CoV-2-specific GC B cells and T FH cells were found in lung-associated LNs, suggesting persisting GC responses months after resolution of infection. Together, these results indicate that the maintenance of SARS-CoV-2-specific immune memory is characterized by localized, ongoing coordination of cellular and humoral immunity within tissues.
SARS-CoV-2-specific memory T and B cells were found throughout the body and localized preferentially to lung and lung-associated LNs, providing direct evidence that those sites are key locations for establishment of immune memory after SARS-CoV-2 infection. Gutdraining LNs in some donors were also significant sites for SARS-CoV-2-specific memory T and B cells (particularly T RM and B RM ), which could be because of the gut being a major site for SARS-CoV-2 replication in some cases (3,51). The low frequency of SARS-CoV-2specific memory T or B cells in the spleen further suggests that virus infection is generally limited to mucosal sites of entry. Our results show that a proportion of SARS-CoV-2-specific memory T cells in the lung were T RM , consistent with findings in airways of severe COVID-19 (28) and surgical lung samples from previously infected patients (52). Additionally, we show that the majority of the SARS-CoV-2-specific memory B cells were resident. In mouse models of influenza infection, localization and tissue residence of T and B cells to the lung and lung-associated LNs are correlated with optimal protective responses (20,23,32). Therefore, tissue-localized and resident memory T and B cells in the lung are likely important for site-specific protection and could be targets for site-specific boosting in vaccination.
SARS-CoV-2-specific memory CD4 + T cells were identified at significantly higher frequencies than CD8 + T cells across tissue sites, reflecting previous studies of peripheral blood showing that CD4 + responses are more robust than CD8 + responses months after resolution of infection (2,53). In addition, SARS-CoV-2-specific T cells exhibited tissue-specific functional profiles with cytotoxic proinflammatory, regulatory, and tissue repair functions variably manifested across different sites. In the lung-associated LNs, memory T cells exhibited broad proinflammatory, helper, and regulatory functional profiles. SARS-CoV-2-specific lung T cells produced higher levels of IL-10 compared with other sites consistent with a role for T cell-derived IL-10 in regulating lung inflammation in mice (43). We previously showed in paired airway and blood samples of patients with severe COVID-19 that the cytokine and chemokine profile in airway washes was distinct from that in plasma (28). Here, we further demonstrate that the functional responses of virus-specific T cells are tissue-specific-not only at the site of infection but also across numerous lymphoid tissues. Together, these results suggest that T cells in tissues mediate responses that are functionally adapted to the tissue site, resulting in heterogeneity of immune memory stored throughout the body.
SARS-CoV-2-specific memory B cells were distributed across multiple sites. While frequencies were highest in lung and LNs, there were also significant frequencies in BM. In all sites, virus-specific memory B cells exhibited a predominantly IgG + memory phenotype. The finding of S/RBD-specific B RM in lung and lymphoid sites was notable, as was the low frequency of IgA + SARS-CoV-2-specific memory B cells in mucosal tissue and associate LNs.
Our results directly demonstrate ongoing, persistent GC responses in LNs following resolution of SARS-CoV-2 infection-including at least one example 6 months after infection. Despite a report of potentially impaired GC responses in fatal COVID-19 (54) ongoing GC reactions after resolution of infection, consistent with reports of prolonged evolution of humoral responses in peripheral blood up to 6 months after SARS-CoV-2 infection (3,55). GC B cells were detected in donors spanning a broad age range-from 10 to 74 years, providing compelling evidence that the ability to establish robust GC responses to novel pathogens can be maintained with age.
These results also indicate ongoing interaction and coordination between T and B cells within LNs, which we also found related to memory populations in the lung. Significant correlations were also found between SARS-CoV-2-specific memory B and T cell populations across tissue sites, consistent with correlations between virusspecific T and B cell responses in peripheral blood of previously infected individuals (1, 7). We also identified potential inverse correlations between frequencies of virus-specific CD8 + T cells in the lung and memory B cells in the lung-associated LN, suggesting that lung responses in situ can affect the magnitude or requirement for humoral responses in the associated LN. Together, these findings suggest that dynamic coordination of adaptive immune responses across the body is a feature of antiviral immunity to SARS-CoV-2.
This work has certain limitations. Namely, our study focuses on four seropositive donors across seven decades of life to provide a representative profile of tissue-specific antiviral immune responses. In addition to the challenges of obtaining live cells for immunological studies from organ donors, the findings here also depended on SARS-CoV-2 seropositive donors who had not been vaccinated, thus limiting the size of the donor pool and the timeframe of collection (before December 2020). The consistency in cell type and sitespecific trends and correlations across all profiled donors, as well as corroboration of larger scale blood studies, demonstrates how this project provides new insights into tissue-specific immune memory maintenance and persistence of humoral and cellular responses after SARS-CoV-2 infection.
In conclusion, we reveal here that immunological memory from SARS-CoV-2 infection is maintained as heterogeneous subsets across multiple sites, with active and preferential maintenance in lung and associated LNs, as well as site-specific functional adaptations. These findings support the development of site-specific strategies for monitoring immune memory to infections and vaccines and for fortifying immune responses at the infection sites.

Study design
The objective of this study was to measure adaptive immune responses to SARS-CoV-2 in blood and tissues of seropositive individuals after resolution of infection. We measured the frequency of SARS-CoV-2-specific CD4 + T, CD8 + T, and B cells in seropositive organ donors compared with prepandemic seronegative donors to understand the maintenance of immunological memory to SARS-CoV-2 as T and B cell subsets across the body, the functional immune response in tissues, and the immune memory relationships across circulating and tissue-resident SARS-CoV-2-specific T and B cell populations.

Human samples
Human tissues were obtained from deceased organ donors at the time of organ acquisition for clinical transplantation through an approved protocol and material transfer agreement with LiveOnNY, the OPO for the New York metropolitan area, as previously described (34,35,(56)(57)(58)(59)(60). Human tissues from the pediatric donor (HDL113) were obtained through arrangements with multiple OPOs across the United States through the Human Atlas for Neonatal Development-Immunity program, which is an extension of the coordinating center set up for nPOD (network for Pancreatic Organ Donors with Diabetes) (61). A list of donors from which tissues were used in this study is presented in table S1. Organ donors were tested for SARS-CoV-2 infection (and confirmed as SARS-CoV-2 negative) by PCR testing of nasal swabs, tracheal aspirates, and/or bronchoalveolar lavage. A history of previous COVID-19 (D492 and D498) and positive serology (D495) was provided in the donor summary, and SARS-CoV-2 serology for all donors was measured in the Center for Advanced Laboratory Medicine at Columbia University Irving Medical Center. Tissues from all seropositive donors were obtained before December 2020, and all donors were free of cancer and seronegative for hepatitis B, hepatitis C, and HIV. Because tissues were obtained from brain-dead organ donors, this study does not qualify as "human subjects" research, as confirmed by the Columbia University Institutional Review Board.

Isolation of single-cell suspensions from tissue samples
Tissue samples were maintained in cold saline or media and transported to the laboratory within 2 to 4 hours of organ procurement for adult organs and shipped to the laboratory on ice within 24 hours of procurement for pediatric donors. Tissue processing protocols were adapted from protocols previously described (34,(56)(57)(58)(59)(60), with some recent optimizations. Briefly, mononuclear cells were isolated from the blood and BM samples by density centrifugation using Ficoll-Paque PLUS (GE Healthcare, catalog no. 17-1440-03). Spleen was processed using mechanical dissociation, followed by pushing through 100-m filters (Thermo Fisher Scientific, catalog no. 50-146-1428), and Ficoll-Paque density centrifugation as above. Lung and LN samples were first incubated with collagenase D (1 mg/ml) (Sigma-Aldrich, catalog no. 11088882001) and deoxyribonuclease (0.1 mg/ml) (Thermo Fisher Scientific, catalog no. NC9709009) in Iscove's modified Dulbecco's medium (Thermo Fisher Scientific, catalog no. 12-440-053) for 30 min at 37°C on a shaker followed by addition of 0.5 M EDTA (pH 8.0) (Thermo Fisher Scientific, catalog no. 150575-020), filtration, and density centrifugation as above, resulting in high yields of live leukocytes.

SARS-CoV-2 serology testing
Blood from deceased organ donors (D495, D498, and HDL113) was collected, and serum was obtained after centrifugation using serum separating clot activator tubes (Thermo Fisher Scientific, catalog no. 22040546). SARS-CoV-2 serology testing for N protein was then performed by the Center for Advanced Laboratory Medicine at Columbia University Irving Medical Center to determine previous exposure to SARS-CoV-2 for inclusion in the study.
SARS-CoV-2 enzyme-linked immunosorbent assay titers were determined as previously described (1). Briefly, Corning 96-well halfarea plates (Thermo Fisher Scientific, catalog no.3690) were coated with SARS-CoV-2 S protein (1 g/ml), RBD protein, or N protein (Sino Biological, catalog no. 40588-V07E) overnight at 4°C. The next day, plates were blocked with 3% milk (skim milk powder, Thermo Fisher Scientific, catalog no. LP0031) in phosphate-buffered saline (PBS) containing 0.05% Tween 20 (ThermoScientific, catalog no. J260605-AP) for 2 hours at room temperature. Heat-inactivated serum (30 min at 56°C) was then added to the plates and incubated for 1.5 hours at room temperature. Plates were washed five times with 0.05% PBS/Tween 20. Secondary antibodies were diluted in 1% milk containing 0.05% Tween 20 in PBS. IgG titers were determined using anti-human IgG peroxidase antibody (Hybridoma Reagent Laboratory, catalog no. HP6123-HRP) at 1:1000 dilution. End-point titers were plotted for each sample using background subtracted data. The limit of detection was defined as 1:3 for IgG.

Pseudovirus neutralization assay
The pseudovirus (PSV) neutralization assays were performed as previously described (1). Briefly, 2.5 × 10 4 Vero cells [American Type Culture Collection (ATCC), catalog no. CCL-81] were seeded in clear flat-bottom 96-well plates (Thermo Fisher Scientific, catalog no. 165305) to produce a monolayer at the time of infection. Recombinant SARS-CoV-2 S-D614G-pseudotyped VSV-G-GFP were generated by transfecting human embryonic kidney 293T cells (ATCC, catalog no. CRL-321) with plasmid phCMV3-SARS-CoV-2 S and then infecting with VSV-G-GFP. Pretitrated rVSV-SARS-CoV-2-S-D614G was incubated with serially diluted human heat-inactivated serum at 37°C for 1 to 1.5 hours before addition to confluent Vero cell monolayers. Cells were incubated for 16 hours at 37°C in 5% CO 2 , fixed in 4% paraformaldehyde in PBS (pH 7.4) (Santa Cruz, catalog no. sc-281692) with Hoechst (10 g/ml) (Thermo Fisher Scientific, catalog no. 62249), and imaged using a CellInsight CX5 imager to quantify the total number of cells and infected green fluorescent protein (GFP)-expressing cells to determine the percentage of infection. Neutralization titers or inhibition dose 50 (ID 50 ) were calculated using the One-Site Fit Log IC 50 model in Prism 8.0 (GraphPad). As internal quality control to define the interassay variation, three samples were included across the PSV neutralization assays. Samples that did not reach 50% inhibition at the lowest serum dilution of 1:20 were considered as non-neutralizing.

In vitro T cell stimulations with SARS-CoV-2 peptide MPs
Mononuclear cells from the blood, BM, spleen, lung, lung-associated LNs, and gut-associated LNs of SARS-CoV-2 seropositive donors were thawed, and dead cells were removed using the EasySep Dead Cell Removal (annexin V) Kit (STEMCELL Technologies, catalog no. 17899) containing 10% heat-inactivated human AB serum (Gemini, catalog no. 507533010) and penicillin-streptomycin-glutamine (Thermo Fisher Scientific, catalog no. 10378016) and incubated overnight at 37°C in 5% CO 2 . Cells were stimulated for 6 or 24 hours by the addition of SARS-CoV-2-specific CD4 and CD8 MPs (1 g/ml) (MP_S, MP_CD4_R, MP_CD8_A, and MP_CD8_B) designed and synthesized as previously described. Briefly, MP_S consists of 253 15-mer peptides overlapping by 10 residues and covering the entire S protein. MP_CD4_R consists of 221-predicted HLA class II CD4 + T cell epitopes covering all proteins apart from S protein. For CD8 epitopes, MPs were synthesized based on epitope predictions for 12 most common HLA class I A and B alleles; these resulted in 628 predicted CD8 + T cell epitopes, which were separated into MP_CD8_A and MP_CD8_B (1,2,62). Equimolar amount of DMSO was used as negative control. Before the addition of peptide MPs, cells were blocked for 15 min with anti-CD40 monoclonal antibody (0.5 g/ml) (Miltenyi Biotec, catalog no. 130-094-133), as previously described (63). After either 6 or 24 hours, the supernatant was collected for multiplex detection of cytokines, and cells were stained for AIMs and analyzed via flow cytometry.
AIM + antigen-specific CD4 + T cells were identified as positive following Boolean OR gating of the CD40L + OX40 + , 4-1BB + OX40 + , 4-1BB + CD40L + subsets (see fig. S1 for gating strategy). The resultant gate was used to quantify AIM + CD4 + T cell frequency. AIM + antigenspecific CD8 + T cells were identified as 4-1BB + CD25 + . Antigen-specific T FH calculated as a frequency of NN CD4 + T cells were gated on total CD4 + T cells excluding CD45RA + CCR7 + CD4 + T cells. Antigen-specific CD4 + and CD8 + T cells were measured as DMSO background-subtracted data. For quantification of the frequency of total SARS-CoV-2-specific CD4 + T cells, a weighted average was taken for percentage of AIM + CD4 + T cells identified for samples stimulated with MP_S or MP_CD4_R MPs. For quantification of the frequency of total SARS-CoV-2-specific CD8 + T cells, a weighted average was taken for percentage of AIM + CD8 + T cells identified for samples stimulated with MP_S, MP_CD8_A, or MP_CD8_B MPs.

Flow cytometry
For flow cytometry analysis of SARS-CoV-2 antigen-reactive T cells, cells were stained in 96-well U-bottom plates protected from light using fluorochrome-conjugated antibodies (see table S2 for antibodies in the T cell flow cytometry panel). Briefly, cells were washed with fluorescence-activated cell sorting (FACS) buffer (PBS with 2% heat-inactivated fetal bovine serum) and then resuspended with surface staining antibody cocktail for 20 min at room temperature. Surface-stained cells were fixed for 30 min at room temperature in fixation buffer (Tonbo, catalog no. TNB-0607-KIT), washed with permeabilization buffer (Tonbo, catalog no. TNB-0607-KIT), and washed again with FACS buffer. Flow cytometry data were collected using the five-laser Cytek Aurora flow cytometer (Cytek Bio) and analyzed using FlowJo V 10.7 software and Prism 9.0.1. software.
For flow cytometry analysis of SARS-CoV-2-specific B cells, biotinylated protein antigens multimerized on fluorescently labeled streptavidin were used as probes to detect antigen-specific B cells (see  table S3 for antibodies used in the B cell flow cytometry panel). Avitagged full-length SARS-CoV-2 S (2P-stabilized, double streptavidintagged) and RBD proteins were generated in-house. Biotinylation was performed using biotin protein ligase standard reaction kit (Avidity, catalog no. Bir500A) following the manufacturer's protocol and dialyzed against PBS. Biotinylated S was mixed with streptavidin BV421 (BioLegend, catalog no. 405225) and streptavidin BUV737 (BD Bioscience, catalog no. 612775) at 20:1 ratio (~6:1 molar ratio). Biotinylated RBD was mixed with streptavidin phycoerythrin (PE)-Cy7 (BioLegend, catalog no. 405206) and streptavidin BUV661 (BD Bioscience, catalog no. 612979) at 2.2:1 ratio (~4:1 molar ratio). Streptavidin PE-Cy5.5 (Thermo Fisher Scientific, catalog no. SA1018) was used as a decoy probe for nonspecific streptavidin-binding B cells. The probes were then mixed in Brilliant Stain Buffer (BD Bioscience, catalog no. 566349) containing 5 M free d-biotin (Avidity, catalog no. Bir500A). Cells (~10 7 ) were prepared in U-bottom 96-well plates and stained with 50 l of antigen cocktail containing 400 ng of S (200 ng per probe), 100-ng of RBD (50 ng per probe), and 20 ng of streptavidin PE-Cy5.5 at 4°C for 1 hour followed by staining for surface markers in Brilliant Stain Buffer at 4°C for 30 min. Dead cells were stained using the LIVE/DEAD Fixable Blue Stain Kit (Thermo Fisher Scientific, catalog no. L34962) in PBS at 4°C for 30 min. Cells were then fixed and permeabilized using eBioscience Intracellular Fixation and Permeabilization Buffer Set (Thermo Fisher Scientific, catalog no. 88-8824-00) before staining with antibodies against transcription factors in eBioscience Permeabilization Buffer (Thermo Fisher Scientific, catalog no. 00-8333-56). Samples were acquired on Cytek Aurora and analyzed using FlowJo10.7.1 (BD Biosciences). In each experiment, peripheral blood mononuclear cells from a known convalescent individual with COVID-19 and an unexposed individual were included to ensure consistent sensitivity and specificity of the assay.

Multiplex detection of cytokines
Cryopreserved supernatant from in vitro T cell stimulation experiments was sent to Eve Technologies Corp. (Calgary, Alberta) for quantification of 50 total human cytokines, chemokines, and growth factors. Luminex xMAP technology was used for multiplexed quantification of two human cytokines in one array (perforin and granzyme B) and 48 human cytokines, chemokines, and growth factors in a separate array [sCD40L, EGF (endothelial growth factor), eotaxin, FGF-2 (fibroblast growth factor 2), Flt Observed concentrations were calculated with the standard curve based on the fluorescence intensity of the bead population for a specific analyte. For analysis and visualization of cytokine/chemokine production by antigen-responding cells in multiple tissues sites within each individual donor, observed concentrations for each analyte were first subtracted from DMSO-negative control and then scaled across samples for each individual donor on a maximum absolute scale, with values ranging from −1 to 1 across all analytes using the MaxAbsScaler features of sklearn.preprocessing function of the Python scikit-learn library (36,63). Heatmap visualizations were generated using the Python data visualization library seaborn (64). For analysis comparing, the production of analytes across donors and tissue sites either absolute observed concentrations were used or observed concentrations for each analyte were normalized to DMSO-negative control.

Statistical analysis
Descriptive statistics of compiled flow cytometry data and statistical testing were performed using Prism (GraphPad). Graphs were generated using Prism (GraphPad), Python matplotlib and seaborn libraries (65,67), and RStudio corrplot package (66). Differences in means between two sample groups were compared using nonparametric test of null hypothesis Mann-Whitney U test. Pearson correlations were used to evaluate immune memory relationships. Multiple group comparisons were done using one-way analysis of variance (ANOVA, corrected for multiple comparisons by false discovery rate using two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli when comparing seropositive and seronegative donors. For comparing immune mediator profiles across tissue sites, statistical analyses were performed via one-way ANOVA corrected for multiple comparisons by Tukey's multiple comparison test. P < 0.05 was considered as statistically significant. ****P ≤ 0.0001, ***P ≤ 0.001, **P ≤ 0.01, and *P ≤ 0.05.

SUPPLEMENTARY MATERIALS
www.science.org/doi/10.1126/sciimmunol.abl9105 Figs. S1 to S9 Tables S1 to S3 Data files S1 and S2 View/request a protocol for this paper from Bio-protocol.