Adaptive immunity to human coronaviruses is widespread but low in magnitude

Abstract Objectives Endemic human coronaviruses (hCoVs) circulate worldwide but cause minimal mortality. Although seroconversion to hCoV is near ubiquitous during childhood, little is known about hCoV‐specific T‐cell memory in adults. Methods We quantified CD4 T‐cell and antibody responses to hCoV spike antigens in 42 SARS‐CoV‐2‐uninfected individuals. Antigen‐specific memory T cells and circulating T follicular helper (cTFH) cells were identified using an activation‐induced marker assay and characterised for memory phenotype and chemokine receptor expression. Results T‐cell responses were widespread within conventional memory and cTFH compartments but did not correlate with IgG titres. SARS‐CoV‐2 cross‐reactive T cells were observed in 48% of participants and correlated with HKU1 memory. hCoV‐specific T cells exhibited a CCR6+ central memory phenotype in the blood, but were enriched for frequency and CXCR3 expression in human lung‐draining lymph nodes. Conclusion Overall, hCoV‐specific humoral and cellular memory are independently maintained, with a shared phenotype existing among coronavirus‐specific CD4 T cells. This understanding of endemic coronavirus immunity provides insight into the homeostatic maintenance of immune responses that are likely to be critical components of protection against SARS‐CoV‐2.


INTRODUCTION
In contrast to the high pathogenicity of MERS-CoV, SARS-CoV and SARS-CoV-2 coronaviruses, endemic human coronaviruses (hCoVs) circulate worldwide but typically cause common colds with only limited morbidity and mortality. 1 Endemic hCoVs encompass two alpha-coronaviruses (aCoV), NL63 and 229E, and two beta-coronaviruses (bCoV), HKU1 and OC43. 1 Sero-epidemiological studies suggest that infection and seroconversion to hCoVs occur during early childhood (typically by 4 years of age), [2][3][4] although there are discrepant reports on the prevalence of each virus within distinct geographical cohorts. 2,5 Despite the early development of immunity against multiple hCoVs, most adults remain susceptible to periodic reinfection, [6][7][8] with increased susceptibility among immunocompromised individuals. [9][10][11] This suggests the magnitude and/ or quality of hCoV-targeted immunity in adults is insufficient for sterilising protection but instead may limit the burden of disease to asymptomatic or mild infection. 8 Defining the extent of serological and/or cellular immunity required to protect individuals from reinfection or severe disease remains a key question in the SARS-CoV-2 pandemic. As neutralising responses wane after CoV infection, it is likely that a combination of serum antibody and B-cell/T-cell memory provides longer-term protection from the recurrence of disease. 12,13 The study of hCoV-specific T-and Bcell memory can therefore provide a key preview into the development of durable, protective SARS-CoV-2 immunity.
Characterisation of population-level immunity to hCoV can also inform our understanding of cross-reactive immune responses between high pathogenicity and endemic CoV. Studies of SARS-CoV-2-specific immunity in uninfected individuals clearly demonstrate pre-existing cross-reactive antibody, [14][15][16] B-cell 16 and T-cell responses. [17][18][19][20][21][22] Nevertheless, it is currently unclear what contribution, if any, cross-reactive immunity plays in modulating the response to SARS-CoV-2 infection or vaccination. 23 Detailed analyses of cross-reactive T cells suggest the majority of such responses are dominated by CD4 T cells and directed towards non-RBD epitopes of the spike (S) protein. 18,21,24 To date, however, consensus regarding the origin of these cross-reactive responses is lacking, with evidence both for 18 and against 25 a major contribution from hCoV-specific memory T cells.
Deconvolution of cross-reactive SARS-CoV-2 responses and de novo SARS-CoV-2 immunity requires a more detailed understanding of hCoVspecific serological and cellular memory. Relatively little is known about population-level T-or B-cell memory to hCoV in adults, despite evidence suggesting an impact of recent hCoV infection on COVID -19 severity. 26 Several groups find widespread but modest CD4 T-cell responses to hCoV proteins, with estimates for the prevalence of memory responses ranging from 70% to 100% of study participants. 24,25,27 Detection of hCoV-specific CD8 T-cell responses has been less reported, 27 and the prevalence of cross-reactive SARS-CoV-2-specific responses in these cohorts varies substantially. 24,25 Furthermore, data comparing hCoV-specific T-or Bcell responses in the circulation with the presence or absence of such responses in the respiratory tract or secondary lymphoid organs (SLO) are lacking. Studies in animal models suggest that respiratory infections can generate long-lived T-cell memory in lung-draining lymph nodes (LDLN), 28 raising the possibility of analogous responses following hCoV infection.
To address these knowledge gaps, we assessed the prevalence and phenotypic characteristics of hCoV spike-specific antibody, memory T-cell and memory B-cell responses in a cohort of SARS-CoV-2uninfected adults. We find that the magnitude of hCoV immunity is independent of age and is characterised by robust antibody titres, widespread CD4 T-cell memory within both Tmem and cTFH populations, and an enrichment of T-cell memory in LDLN. In contrast, neutralising antibody activity is relatively low and memory B cells are infrequently detected in either the circulation or LDLN. Overall, our data detail a consistent pattern of hCoV-specific immune memory in the circulation and SLO, which likely co-ordinate to provide longterm protection from hCoV infection.

RESULTS
hCoV-specific antibody and CD4 T-cell memory are common among adults We recruited a cohort of 42 SARS-CoV-2uninfected adults (n = 21 male, n = 21 female), ranging in age from 18 to 67 years with no recent cold or COVID-19 symptoms (Figure 1a).
hCoV-specific CD4 Tmem cells are predominately T CM cells with a CCR6 + phenotype Given divergent host receptor specificity and possible differences in tissue tropism among hCoV, 1 we assessed whether memory or chemokine receptor phenotypes differed among S-specific CD4 T-cell populations (gating in Supplementary figure 1b). Similar to the parental Tmem population, hCoV S-specific and CoV-2 cross-reactive CD4 T cells were predominately CD27 + CCR7 + , classically defined as central memory T cells (T CM ; Figure 2a and b). In contrast to the bulk Tmem population, however, hCoV S-specific cells were substantially enriched for CCR6 expression (with or without co-expression of CXCR3; Figure 2c and d). When comparing intraindividual responses, hCoV S-specific Tmem phenotypes were generally similar across all S antigens (Figure 2e). Prior studies have also described a dominant CCR6 phenotype of CoV-2 S-specific Tmem among convalescent COVID-19 subjects, 33 and here, we find that CoV-2 crossreactive responses are similarly highly CCR6-biased (Figure 2d and e).
Interestingly, hCoV responses comprised a greater proportion of the cTFH population than the Tmem compartment in a paired analysis (P < 0.002 for all hCoV antigens), with some donors exhibiting a greater than ninefold enrichment of hCoV-specific cells in the cTFH gate (data for HKU1 shown in Figure 3b). Similar to Tmem, antigen-specific cTFH were highly enriched for a CCR6 + CXCR3 À phenotype (Figure 3c   CoV-2-specific cTFH in SARS-CoV-2-uninfected donors are consistent with phenotypes previously described in convalescent COVID-19 subjects. 29 Comparison of antigen-specific cTFH and Tmem cells revealed a significant enrichment of the CCR6 + CXCR3 À phenotype among cTFH, including the CoV-2 cross-reactive population (Figure 3e and f). These data suggest that while hCoV memory is broadly observed among both Tmem and cTFH subsets, the frequency and phenotype of these responses are, to a degree, subsetspecific.

CoV-2 cross-reactive T cells correlate with HKU1 T-cell memory
It is currently unclear whether CoV-2 cross-reactive T-cell responses arise primarily from hCoV memory or reflect cross-reactivity from a broad array of antigen specificities. 18,25 Among the cohort, subjects with CoV-2 cross-reactive CD4 T-cell responses frequently exhibited memory responses to multiple hCoVs ( Figure 4a). We assessed the relationship between the frequency of CoV-2 and hCoV memory responses and found significant correlations only between bCoV and CoV-2 crossreactivity (P = 0.006 for HKU1 and P = 0.018 for OC43; Figure 4b). This association is consistent with a greater sequence homology among bCoV strains (CoV-2, HKU1 and OC43) than the aCoV 229E and NL63. 42 Among the subset of donors with cross-reactive responses, only HKU1 memory correlated with CoV-2 cross-reactivity (P = 0.030; Figure 4b).
Interestingly, while almost all individuals with CoV-2 cross-reactive responses exhibited HKU1 and OC43 memory, the converse was not observed. Indeed, individuals with relatively similar patterns of hCoV reactivity could exhibit notably different CoV-2 reactivity (Figure 4c). There was no significant association of demographic characteristics among individuals with or without CoV-2 cross-reactive responses, although the cross-reactive group did exhibit a greater representation of women than those without cross-reactivity (P = 0.06; Supplementary figure 3c and d). Given the association between HKU1 and CoV-2 T-cell frequencies, we assessed whether a particular phenotype of HKU-specific Tmem was related to the presence or absence of cross-reactive responses, but found no such distinctions (Supplementary figure 3e).

Cellular memory and humoral immune memory to hCoV are maintained independently
Studies of convalescent COVID-19 cohorts have demonstrated a strong correlation between SARS-CoV-2-specific cTFH, memory B cells (MBC), binding IgG and serum neutralisation, 19,29,32 as expected from a coordinated acute immune response. To assess whether such associations are maintained in long-term hCoV immunity, we explored correlations between antibody and Tcell responses across the cohort. Surprisingly, there was no relationship for any antigen between plasma IgG endpoint titre and the frequency of either S-specific CD4 Tmem or cTFH (P > 0.05 for all; data for HKU1 and NL63 shown in Figure 5a). To gain greater insight into the coordination of cellular and humoral hCoV memory, we undertook an in-depth interrogation of immunity to NL63, which shares use of the cellular entry receptor ACE2 with SARS-CoV and CoV-2, and therefore likely has similar tissue tropisms in vivo.
To assess S-specific MBC and quantify plasmaneutralising activity, NL63 S-specific MBC probes were generated as described previously, 29 and a novel NL63 pseudovirus-based neutralisation assay was performed with 293T cells stably expressing hACE2 as targets. MBC specific for NL63 and CoV-2 S were detected infrequently among the cohort, particularly in comparison with the frequency of CoV-2 S-specific MBC previously reported among convalescent COVID-19 donors 29 ( Figure 5b). Accordingly, the frequency of NL63 S-specific MBC did not correlate with plasma NL63 binding IgG titres (Figure 5c). Plasma-neutralising activity against NL63 pseudovirus was detected among all donors tested, with a median IC 50 of 100.7 (n = 12, IQR 56.6-234.6). Neutralising activity strongly correlated with NL63 S-specific antibody titres (P = 0.006) but was not associated with NL63 S-specific MBC frequencies (Figure 5d). We did, however, observe a trend towards a positive correlation of neutralisation with NL63 S-specific cTFH responses (P = 0.081; Figure 5e). Given our prior observation that CCR6 + CoV-2 S-specific cTFH responses were inversely associated with neutralising antibodies after COVID-19, 29 we assessed the correlation between NL63neutralising activity and NL63 S-specific cTFH phenotype (for donors with NL63 S-specific cTFH responses, n = 7). Interestingly, the frequency of CCR6 + cTFH again negatively correlated with plasma-neutralising activity (P = 0.048; Figure 5e), although the small sample size is a caveat of this analysis.

Enrichment of HKU1 and NL63 S-specific T cells in lung-draining lymph nodes
Although assessment of hCoV immunity has been primarily limited to peripheral blood, studies suggest that repeated infections with respiratory  viruses can seed long-lived memory T-cell responses in lung and lung-draining lymph nodes (LDLN). 28,43 We therefore assessed the frequency of HKU1, NL63 and CoV-2 S-specific CD4 T-cell responses in matched LDLN (n = 5) and lung samples (n = 6) from a human tissue biobank (gating in Supplementary figure 5). We detected robust HKU1 and NL63 responses within the CD45RA À CXCR5 À CD4 T-cell population of LDLN (Figure 6a). Given the higher levels of background T-cell activation in SLO than peripheral blood, we validated the specificity of the hCoV responses by confirming that antigen stimulation also drove expression of CD154 on OX-40 + cells (Figure 6a). Among the five donors studied, the median frequency of HKU1 and NL63 S-specific Tmem was 1.2% (range 0.12-2.19) and 1.12% (range 0.31-4.04), respectively (Figure 6b). Reactivity to CoV-2 S was substantially lower, with a median of 0.07% (range 0.01-0.99). Similar antigen-specific responses were observed within the CD4 + CD45RA À CXCR5 + population (Figure 6b). There was limited to no evidence of ongoing hCoV S-specific GC TFH activity among the samples (data not shown).
In contrast to the high frequencies of hCoVspecific CD4 T cells in LDLN, we found only modest hCoV reactivity among lung-derived CD4 T cells (Supplementary figure 6a and b). These data are consistent with reports that tissue-resident T cells (T RM ) in the lung are relatively short-lived compared with other tissues. 44 Furthermore, the majority of AIM + cells did not exhibit a CD69 + CD103 + phenotype, suggesting they are unlikely to represent bona fide lung-resident T cells 45 (Supplementary figure 6c). Similarly, we found little evidence for the presence of NL63 or CoV-2 cross-reactive MBC in either the LDLN or the lung tissues (Supplementary figure 6d and e).
It has been speculated that the dominant CCR6 + phenotype of CoV-2-specific CD4 T cells may reflect preferential homing of these cells to the lung. 32 We therefore compared the CCR6/CXCR3 phenotypes of hCoV-specific T cells in LDLN with the peripheral blood obtained from the unmatched healthy adult cohort presented in Figures 2 and 3. After adjustment for baseline activation, we found that LDLN-derived hCoV Sspecific Tmem exhibited a predominately CXCR3 + phenotype, with a substantial population of CCR6 À CXCR3 + cells (median 37.7% for HKU1 and 37.5% for NL63; Figure 6c and d). In contrast, only 10.8% and 11.2% of circulating HKU1 and NL63 S-specific Tmem among the blood donor cohort were CCR6 À CXCR3 + (Figure 2d). As we previously observed for cTFH in the periphery, CXCR5 + hCoVspecific T cells in the LDLN remained more likely to express CCR6 than their Tmem counterparts (Figure 6e and f). Nevertheless, LDLN-derived hCoV-specific CXCR5 + cells were enriched for CXCR3 expression (median 30.7% CCR6 À CXCR3 + for HKU1 and 19.4% for NL63) compared with the phenotypes observed among peripheral cTFH (median 6.6% CCR6 À CXCR3 + for HKU1 and 9.8% for NL63; Figures 3d and 6e and f). Collectively, these data suggest either differential retention or formation of CXCR3 + hCoV S-specific CD4 T cells in LDLN compared with peripheral blood.

DISCUSSION
Despite periodic reinfection, most adults experience only mild or asymptomatic hCoV infection, suggesting the presence of at least partially protective immune memory. We find that, in addition to near-universal plasma antibody reactivity to hCoV, memory T-cell responses to both aCoV and bCoV are widespread. In contrast, the relatively modest neutralising activity against NL63 and low frequencies of Sspecific MBC suggest that sterilising humoral immunity is likely absent. Instead, additive contributions of multiple arms of adaptive immunity, in particular antiviral T-cell responses, may underpin protection.
While several studies have quantified hCoVspecific T-cell responses in adult cohorts, 24,25,27 it was unclear whether the different viruses would elicit phenotypically distinct Tmem or cTFH responses. Together with studies of convalescent SARS-CoV-2-specific T-cell responses, 29,32 our data suggest that the CCR6 + phenotype of circulating hCoV-specific CD4 memory cells may be broadly reflective of coronavirus infection in humans. Indeed, AIM-based assays have consistently identified a high proportion of CCR6 + CXCR3 À cells among SARS-CoV-2 S-specific CD4 T cells, 29,32 in spite of low IL-17 production following antigen stimulation. 22,29,32 Interestingly, longitudinal follow-up of convalescent COVID-19 subjects indicated a time-dependent increase in the proportion of CCR6 + S-specific cTFH, 33 suggesting a convergence of phenotypes between CoV-2specific and hCoV-specific cTFH memory over time.
Chemokine receptor expression on CD4 T cells is often used as a surrogate of cytokine expression  12). (e) Spearman correlation between NL63-neutralising activity and either total NL63 S-specific cTFH or the frequency of CCR6 + antigen-specific cTFH. and Th1/Th2/Th17 function, but these receptors also regulate lymphocyte trafficking to SLO and tissues. While there was little evidence for hCoV Sspecific T-cell memory in the lung, both HKU1 and NL63 responses were robustly detected in LDLN. The enrichment of CXCR3 + hCoV T-cell responses in LDLN compared with peripheral blood suggests a potential involvement of CXCR3 expression in recruitment or retention of these cells out of the circulation. These data are consistent with observations in other respiratory infections, where CXCR3 mediates lung trafficking of antigenspecific CD4 T cells. 46,47 Future studies will be required to address the role, if any, for these cells in contributing to protection from re-exposure to CoV infection.
Consistent with other cohorts, 17,19 we find evidence for CoV-2 cross-reactive CD4 T cells in uninfected donors. In vitro expansion of CoV-2 cross-reactive T-cell clones has demonstrated the potential for shared specificity with all hCoVs. 18,24 However, at a cohort-wide level, we find the frequency of CoV-2 cross-reactive cells correlates most strongly with HKU1 memory, although no immediate immunological or demographic features distinguish HKU1-reactive individuals with or without cross-reactive CoV-2 responses. Future studies incorporating in vitro expansion of hCoV-specific or SARS-CoV-2 cross-reactive CD4 T cells may yield additional insights into the nature of the relationship between HKU immunity and cross-reactivity with other coronaviruses. Larger population-based studies will be required to determine any associations between particular HLA class II alleles and cross-reactive CD4 responses. Although it has been speculated that pre-existing cross-reactive T-cell immunity could be beneficial in the context of SARS-CoV-2 vaccines, 23 it should be noted that only CXCR3 + , but not CCR6 + , cTFH responses appear to correlate with neutralising antibody titres during COVID-19 convalescence. 29,31,48 While recall of the CCR6 + cTFH could induce expression of CXCR3, currently available evidence suggests the highly CCR6biased responses to hCoV may not be beneficial in the context of vaccination or re-exposure.
Overall, these data clarify the characteristics of long-term immunity to endemic coronaviruses, which have comparable magnitudes and share phenotypic features of S-specific antibody and T-cell memory across all four hCoVs. Insight into the homeostatic maintenance of hCoV immunity is likely to provide a preview of long-term CoV-2-specific immunity established in the population after vaccination or widespread infection.

Subject recruitment and sample collection
SARS-CoV-2-uninfected controls were recruited as part of a previous COVID-19 study, 29 and relevant demographic characteristics are indicated in Figure 1a. For all participants, whole blood was collected with sodium heparin anticoagulant. Plasma was collected and stored at À80°C, and PBMCs were isolated via Ficoll-Paque separation, cryopreserved in 10% DMSO/FCS and stored in liquid nitrogen. The study protocols were approved by the University of Melbourne Human Research Ethics Committee (#2056689), and all associated procedures were carried out in accordance with the approved guidelines. All participants provided written informed consent in accordance with the Declaration of Helsinki.
The use of tissue samples from human donors was approved by the University of Melbourne Human Research Ethics Committee (#1954691), and all associated procedures were carried out in accordance with the approved guidelines. Tissues were collected from six donors: male, age 41, brain death; female, 33, brain death; male, 34, circulatory death; male, 52, brain death; female, 60, brain death; and female, 59, brain death. Tissues were passed through 70-µM filters and homogenised into single-cell suspensions, which were subsequently cryopreserved in 10% DMSO/FCS.

Expression of coronavirus antigens
A set of trimeric, pre-fusion stabilised coronavirus S proteins (HKU1, 229E, NL63, OC43 and SARS-CoV-2) were generated for serological and flow cytometric assays using techniques previously described. 29 Genes encoding the ectodomain of SARS-CoV-2 S (NC_045512; AA1-1209) with six proline stabilisation mutations and furin site removal (Hexapro 49 ), the HKU1 S (NC_006577; AA1-1291) and NL63 S (DQ445911.1; AA1-1291) with two proline stabilisation mutations (S-2P), were cloned into mammalian expression vectors. Plasmids encoding S-2P versions of the ectodomains of OC43 and 229E were kindly provided by Dr Barney Graham, NIH. S proteins were expressed in Expi293 or ExpiCHO cells (Thermo Fisher, MA, USA) using the manufacturer's instructions and purified using Ni-NTA and size-exclusion chromatography. Protein integrity was confirmed using SDS-PAGE.

ELISA
Antibody binding to recombinant S proteins was determined by ELISA as previously described. 29 Briefly, 96well Maxisorp plates (Thermo Fisher) were coated overnight at 4°C with 2 µg mL À1 recombinant S, blocked with 1% FCS in PBS and incubated with plasma dilutions for two hours at room temperature. Plates were washed, incubated with 1:20 000 dilution of HRP-anti-human IgG (Sigma-Aldrich, MO, USA) and developed using TMB substrate (Sigma). Endpoint titres were calculated as the reciprocal serum dilution giving signal 2 9 background using a fitted curve (4 parameter log regression).

Flow cytometric detection of hCoV-reactive B cells
Probes for delineating NL63 or SARS-CoV-2 S-specific B cells within cryopreserved human PBMC were generated by sequential addition of streptavidin-PE (Thermo Fisher) or streptavidin-BV421 (BD Biosciences, San Jose, CA, USA), respectively, to trimeric S protein biotinylated using recombinant Bir-A (Avidity). Cells were stained with Aqua Viability Dye (Thermo Fisher). PBMC, lung and lymph node cells were surface-stained with the following monoclonal formaldehyde and acquired on a BD LSR Fortessa using BD FACSDiva.

Statistics
Statistical analysis was performed in GraphPad Prism v9. All T-cell stimulation data are presented after background subtraction using the unstimulated control. Two group comparisons were performed using the Mann-Whitney Utest, or the Wilcoxon test for paired data. Correlations were performed using the Spearman test. P-values were considered significant if P < 0.05.