Mild SARS-CoV-2 infection in rhesus macaques is associated with viral control prior to antigen-specific T cell responses in tissues

SARS-CoV-2 primarily replicates in mucosal sites, and more information is needed about immune responses in infected tissues. Here, we used rhesus macaques to model protective primary immune responses in tissues during mild COVID-19. Viral RNA levels were highest on days 1-2 post-infection and fell precipitously thereafter. 18F-fluorodeoxyglucose (FDG)-avid lung abnormalities and interferon (IFN)-activated monocytes and macrophages in the bronchoalveolar lavage (BAL) were found on days 3-4 post-infection. Virus-specific effector CD8+ and CD4+ T cells became detectable in the BAL and lung tissue on days 7-10, after viral RNA, radiologic evidence of lung inflammation, and IFN-activated myeloid cells had substantially declined. Notably, SARS-CoV-2-specific T cells were not detectable in the nasal turbinates, salivary glands, and tonsils on day 10 post-infection. Thus, SARS-CoV-2 replication wanes in the lungs of rhesus macaques prior to T cell responses, and in the nasal and oral mucosa despite the apparent lack of antigen-specific T cells, suggesting that innate immunity efficiently restricts viral replication during mild COVID-19.

Most studies of immune correlates of COVID-19 disease severity in humans have focused on sampling of peripheral blood. Nevertheless, some studies have observed infiltration of immune cells into the bronchoalveolar lavage (BAL) fluid, post-mortem lung tissue acquired from fatal COVID-19 cases, or from individuals undergoing medically necessary procedures (12,14,25,26). Notably, on autopsy, several reports CORONAVIRUS Mild SARS-CoV-2 infection in rhesus macaques is associated with viral control prior to antigen-specific T cell responses in tissues 1 T lymphocyte Biology Section, Laboratory of Parasitic Diseases, National Institutes of Allergy and Infectious Disease, National Institutes of Health, Bethesda, MD, USA. *Corresponding author. Email: barberd@niaid.nih.gov SARS-CoV-2 primarily replicates in mucosal sites, and more information is needed about immune responses in infected tissues. Here, we used rhesus macaques to model protective primary immune responses in tissues during mild COVID-19. Viral RNA levels were highest on days 1-2 post-infection and fell precipitously thereafter. 18 F-fluorodeoxyglucose (FDG)-avid lung abnormalities and interferon (IFN)activated monocytes and macrophages in the bronchoalveolar lavage (BAL) were found on days 3-4 postinfection. Virus-specific effector CD8 + and CD4 + T cells became detectable in the BAL and lung tissue on days 7-10, after viral RNA, radiologic evidence of lung inflammation, and IFN-activated myeloid cells had substantially declined. Notably, SARS-CoV-2-specific T cells were not detectable in the nasal turbinates, salivary glands, and tonsils on day 10 post-infection. Thus, SARS-CoV-2 replication wanes in the lungs of rhesus macaques prior to T cell responses, and in the nasal and oral mucosa despite the apparent lack of antigen-specific T cells, suggesting that innate immunity efficiently restricts viral replication during mild COVID-19. have observed a surprising lack of immune cells infiltrating into extrapulmonary tissues despite the presence of high levels of virus (27)(28)(29)(30)(31). These data highlight the importance of understanding the early host response in pulmonary and extrapulmonary tissues in the first few days after SARS-CoV-2 infection. Animal models can be employed to obtain a detailed understanding of the host response in infected tissues. Studies in SARS-CoV-2 susceptible species provide insights into COVID-19 disease pathogenesis. For example, transgenic mouse strains expressing human angiotensin converting enzyme (ACE)2 (K18-hACE2), and mice induced to express hACE2 with viral vectors, are highly susceptible to SARS-CoV-2 infection (32)(33)(34)(35)(36). Syrian hamsters and ferrets are also moderately susceptible and shed infectious virus (37)(38)(39)(40)(41). On the other hand, species more resistant to SARS-CoV-2 disease are useful tools in examining mechanisms of efficient control of viral replication. Several non-human primate (NHP) species can be experimentally infected with SARS-CoV-2. Rhesus macaques, cynomolgus macaques, and African green monkeys typically develop mild symptoms after SARS-CoV-2 infection (42)(43)(44)(45)(46)(47)(48)(49)(50)(51)(52). SARS-CoV-2 immune and vaccinated rhesus macaques are protected from reinfection primarily by neutralizing antibodies and to a lesser extent anamnestic T cell responses (53)(54)(55)(56)(57)(58)(59)(60)(61)(62)(63)(64). Thus, current NHP models are suitable for the study of protective host immune responses associated with mild SARS-CoV-2 infection, but not the mechanisms of pathogenesis during severe disease.
In this study, we use the rhesus macaque model of mild COVID-19 to examine the (1) kinetics of lung inflammation using 18 FDG positron emission tomography computed tomography (PET/CT) imaging, (2) innate immune responses using single cell RNA sequencing (scRNAseq), and (3) the tissue distribution of SARS-CoV-2-specific T cell responses by flow cytometry. Our findings suggest that mild SARS-CoV-2 disease and efficient control of the infection are temporally correlated with activation of myeloid cells by type I IFN, prior to the induction of Ag-specific T and B cell responses. Moreover, they reveal a strong propensity for Ag-specific T cell migration into the pulmonary compartment compared to other mucosal sites of infection.

Radiologic and virological outcomes of SARS-CoV-2 infection in rhesus macaques
Six, male rhesus macaques were infected with 1x10 6 tissue culture infectious dose (TCID) 50 intranasally (i.n.) and 1x10 6 TCID 50 intratracheally (i.t.), for a total dose of 2x10 6 TCID 50 of SARS-CoV-2/USA-WA-1 (Table S1). 18 FDG PET/CT imaging showed evidence of heterogeneous inflammatory foci with increased 18 FDG uptake (Fig. 1A, B) and lesion density (Fig. 1A, C) in the lungs of 5 of 6 animals at day 3 post-infection, which resolved by day 9. Total genomic Nucleocapsid (gN) and subgenomic Nucleocapsid (sgN) RNA levels from nasal and throat swabs peaked 1 to 2 days post-infection and decreased to undetectable levels by day 7 to 10 ( Fig. 1D). Viral RNA was also found in the BAL of all animals at day 4 post-infection and was mostly cleared by day 7-10. It should be noted that day 4 post-infection likely does not represent the peak of viremia in the BAL, and previous studies indicate peak viral loads are reached at day 1 post-infection in the BAL (42). Viral RNA was essentially absent from plasma at all timepoints, consistent with previous reports (42).
The animals were necropsied at day 10 post-infection for tissue analysis. A 3D reconstruction of the day 3 PET/CT images with conducting airways was used to locate and individually collect the previously PET hot lung regions and normal lung tissue separately. SARS-CoV-2 gN RNA was found on day 10 in all secondary lymphoid organs (SLO) and non-lymphoid tissues (NLT) tested, including the previously PET hot and normal lung tissue, nasal turbinates, salivary gland, and tonsils (Fig. 1E). sgN RNA was present at lower levels compared to gN RNA and was highest in lung tissue (Fig. 1F). The persistence of viral RNA at day 10, was confirmed with RNA scope immunohistochemical analysis (Fig. 1G). There was a correlation between genomic and subgenomic RNA levels in the mucosal swabs, BAL, and tissues with detectable RNA (Fig S1A-C). We did not observe a correlation between lung lesion severity at day 3 and viral RNA levels from nasal swabs and BAL at day 1 and 4, respectively (Fig S1D-G). Consistent with previous reports in macaques, various forms of CD62P + and fibrin + microthrombi were still detectable in the lungs on day 10 post-infection ( Fig. 1H) (44). Thus, in rhesus macaques SARS-CoV-2 viral loads peak in the upper airways ~1-2 days after exposure and this results in mild and transient radiographic evidence of lung inflammation at ~3 days post-infection, with residual viral RNA in tissues and microthrombi in the lungs at day 10.

Longitudinal scRNAseq analysis of BAL and PBMC
To compare cellular immune responses in circulation versus airways, scRNAseq was performed on cryopreserved peripheral blood mononuclear cells (PBMC) and BAL samples obtained prior to infection and at days 4, 7 and 10 post-infection. Uniform manifold approximation and projection (UMAP) and nearest neighbor clustering of PBMCs from all timepoints identified multiple myeloid and T/NK cell populations along with B cells, platelets, and a mixture of proliferating cells ( Fig. 2A). Due to PBMC isolation and cryopreservation, granulocyte populations were not accounted for in this study. Myeloid and T/NK cell populations were selected for subsequent clustering. We identified nine distinct T/NK cell subsets in PBMCs across all timepoints ( Fig  S2). Overall, we did not detect major alterations in the T/NK cell composition from PBMCs, but at day 4 after infection, we did observe a drop in naïve CD8 + T cells and an increase in An increase in certain subsets of dendritic cells (DC)2 have been associated with moderate/severe disease in COVID-19 patients (65,66). At day 4 post-infection we observed an increase in CD1c + conventional DC2s (cDC2 (67)) (PBMC myeloid subpopulation 4), which contracted by day 10. Conventional DC1 cells (XCR1, BATF3 -expressing PBMC myeloid population 6) were less abundant than cDC2s and changed relatively little in abundance during infection. The major alterations in the CD14 + monocytes substantially declined by day 7 post-infection and returned to baseline levels by day 10 (Fig. 2C-D). The PGTS2-expressing monocytes that were lost at day 4 returned by day 10 post-infection and did not show major changes in gene expression ( Fig. 2C-D, G).
In the BAL, multiple distinct T cell and myeloid populations were identified, along with proliferating cells, B cells, plasmacytoid DCs (pDC), MAST cells, and epithelial cells (Fig.  3A). Further clustering of BAL T cells identified 5 populations of T cells (Fig S2D). The largest change was the appearance on day 4 of a population with a mixture of CD8 + and CD4 + T cells that had a prominent IFN-stimulated gene signature (BAL T cell sub population 3) (Fig S2D-F). These IFNactivated T cells were no longer detectable by day 7 post-infection. Further clustering of BAL myeloid cells revealed 10 distinct populations of myeloid cells ( Fig. 3B-C). At baseline, BAL cells were mostly comprised of multiple MRC1 + MARCO + myeloid subsets (BAL myeloid subpopulations 0, 2, and 3) which are likely alveolar macrophages (Fig. 3C). At day 4 post-infection, there were major increases in populations of IFN-activated monocytes and macrophages in the BAL (BAL myeloid subpopulation 1 and 6), which declined by day 7 and returned to baseline levels by day 10 (Fig. 3B-D). At day 10 post-infection, the myeloid cells in the BAL were dominated by a population of CD1c + cDC2s (BAL myeloid sub population 4) (Fig. 3C). The cDC2 in the BAL had a pattern of differentially expressed genes that suggested that this population also responded to infection by upregulating type I IFN-responsive genes at day 4 post-infection (Fig. 3E). By day 10 the cDC2 had down regulated the type I IFN genes and up-regulated genes associated with responses to lipopolysaccharide (LPS), including additional chemokines and IL1B, as well as the macrophage markers MRC1 and MARCO.
Correlation analysis revealed strong positive correlations between viral RNA levels in the BAL, nasal swabs, and throat swabs (Fig. 3F). Viral RNA from BAL and nasal swabs was positively correlated with IFN-activated monocytes, macrophages, and T cells in the BAL. In contrast, PGTS2 + monocytes and naïve CD8 + T cells from PBMCs negatively correlated with viral RNA from nasal swabs and BAL. To ask if type I, II, or III IFN was the stimulus for the IFN gene signature observed in many cell subsets, we analyzed IFNB1, IFNG, and IFNL1 gene expression across all cell types (Fig.  3G). We found that IFNB1 was up-regulated at day 4 postinfection, the timepoint when IFN-activated immune cells were highest. Interestingly, IFNG and IFNFL1 showed a relative increase at day 10, when viral RNA had already decreased substantially. Across all cell types the most highly up-regulated, statistically significant IFN-stimulated genes were those downstream of type I IFN signaling and showed a pattern of up-regulation at day 4 post-infection. Together, these data indicated that SARS-CoV-2 infection induces a robust type I IFN-activated myeloid cell response in PBMC and BAL, which coincides with radiographic indications of inflammation and resolves along with viral RNA levels between day 7 to 10 post-infection.

Early B cell responses to SARS-CoV-2 infection
We measured multiple B cell subsets in PBMCs and BAL by flow cytometry, including resting naïve B cells (CD20 + IgD + CD95 -), activated naïve B cells (CD20 + IgD + CD95 + ), germinal center B cells (GC B cells: CD20 + IgD -BCL6 + Ki67 + ), plasmablasts (CD20 + IgD-BCL-6 -CD38 hi CD27 + ), and activated memory B cells (CD20 + IgD -BCL-6 -CD95 + ) (Figs S3A-B). Activated memory B cells were further subdivided into IgM + , IgG + , IgA + , and isotype undefined. After infection, we observed a decrease in total B cells in PBMCs (Fig S3C), an increase of 2-3% in the proportion of activated naïve B cells from PBMC at day 4 and 7 (Fig S3D), and a decrease in the overall proportion of activated B cells in PBMCs that are isotype undefined (Fig S3E). At necropsy, the frequency of B cells varied across tissues. While the spleen had the largest fraction of B cells, the BAL and lung had the highest proportion of activated memory B cells (Fig S3D-H). Anti-spike IgM and IgG were first detectable in the plasma and BAL at ~day 10 post-infection in most animals, although levels were low compared to positive controls (~2-5 fold above background) ( Fig S4). Low levels of neutralizing antibodies against SARS- CoV-2/USA-WA1 were also detected in the plasma on day 10. Overall, there were very few changes in B cell populations over the first 10 days of infection, and antibody responses were just developing on day 10 post-infection.

Kinetics of SARS-CoV-2-specific effector CD8 + and CD4 + T cell responses in the BAL and PBMC
We next performed a flow cytometric analysis of the Agspecific T cell response to SARS-CoV-2. We observed only minor changes in the activation of bulk T cell responses in the PBMC after infection, with more dynamic changes in the BAL after infection (Fig S5A-B). To examine SARS-CoV-2-specific T cell responses, we performed intracellular cytokine staining after ex vivo restimulation with peptide pools from the viral spike (S), nucleocapsid (N), and membrane (M) proteins, as well as peptide pools (megapools) derived from multiple SARS-CoV-2 antigens found to be immunogenic in humans (68,69). As expected, Ag-specific T cell responses were not detected at day 4 post-infection in PBMCs or BAL ( Fig. 4A-D). CD4 + T cell responses to S, N, and megapool, each reached ~4-6% on average by day 7, whereas Ag-specific CD8 + T cells were ~1% at this timepoint in the BAL (Fig. 4B,D). Consistent with a slightly delayed response, Ag-specific CD8 + T cells in the BAL continued to expand in frequency and maintained Ki67 expression between days 7 and 10 post-infection, while Ag-specific CD4 + T cells peaked in frequency at day 7 and decreased Ki67 expression between days 7 and 10 ( Fig. 4D-E). Of note, frequencies of Ag-specific T cells were ~10 to 20-fold higher in the BAL vs. PBMC. Moreover, CD8 + and CD4 + T cell responses against S and N were consistently immunodominant in comparison to M-specific T cells.
In addition to producing IFNγ and TNF after peptide stimulation, the majority of Ag-specific CD8 + T cells in the BAL also expressed granzyme B and degranulated after restimulation, as indicated by CD107a/b surface staining (Fig.  4F). Approximately 25-60% of Ag-specific CD4 + T cells in the BAL also made IL-2. Furthermore, both CD8 + and CD4 + Agspecific T cells in the BAL up-regulated markers of tissue residence, CD69 and CD103, between days 7 and 10 post-infection (Fig S5C-D). Thus, SARS-CoV-2-specific CD8 + and CD4 + T cells in the airways displayed typical effector functions associated with cytotoxic T lymphocyte (CTL) and Th1 cells, respectively.

Distribution of SARS-CoV-2-specific CD8 + and CD4 + T cell responses in mucosal tissues
At the day 10 necropsy, we examined SLO and NLT from the upper and lower respiratory tract for bulk and Ag-specific T cells. Tissue resident memory CD8 + and CD4 + T cells (CD95 + CD69 + CD103 +/− ) were detected in all nonlymphoid tissues measured, including the lung, nasal turbinates, salivary glands, and tonsils ( Fig. 5A-B). CD103 + Trm were more abundant among CD8 + compared to CD4 + T cells in the BAL, salivary glands, and lymph nodes, which has been shown in other model systems (70,71). Using intravenous (i.v.) antibody staining to distinguish between tissue parenchymal and intravascular cells (72)(73)(74), we confirmed that most cells in the BAL, nasal turbinates, salivary gland, tonsils, and lymph nodes were from the tissue parenchyma ( Fig S6). As expected for such a highly vascularized tissue, most cells from the lung were intravascular stain positive, but a small population of CD69 + ivcells were detectable in the lungs confirming that tissue resident cells were also detected in pulmonary tissue. We next quantified the magnitude of SARS-CoV-2-specific T cells in each of these tissues. S, N, and megapool-specific CD8 + and CD4 + T cells were detected in the BAL, previously PET hot lung lesions, pulmonary lymph nodes, peripheral lymph nodes, spleen, and PBMC (i.e., the frequency of IFNγ + and/or TNF + cells after peptide restimulation was statistically significantly higher than the unstimulated samples). Surprisingly, Ag-specific CD8 + and CD4 + T cell responses could not be detected in the majority of nasal turbinates, salivary gland, and tonsils (Fig. 5C). The absence of Ag-specific T cells cannot be accounted for by poor T cell isolation from tissues ( Fig. 5A-B) or lack of virus replication at these sites ( Fig. 1D-F). Thus, the early clonal burst of SARS-CoV-2-specific T cells is highly focused on the BAL and unexpectedly undetectable in the nasal and oral mucosa.
Overall, the kinetics of SARS-CoV-2 replication and innate/adaptive immune response in rhesus macaques appears typical of an acute viral infection (Fig. 5D). SARS-CoV-2 replication peaks within 1 to 2 days post-infection and rapidly decreases thereafter. IFN-responsive myeloid responses are rapidly detected in the PBMC and BAL at day 4 post-infection. IFN-driven innate immune responses and lung inflammation then decline by day 7 post-infection, followed by an increase in inflammatory DCs and Ag-specific T cells at day 10.

DISCUSSION
We show here that during mild COVID-19 in rhesus macaques, SARS-CoV-2 replication is largely suppressed prior to the induction of virus-specific T cell responses. PET/CT imaging showed regions of ground glass opacity and consolidation with elevated 18 FDG uptake in the lungs on day 3 after SARS-CoV-2 infection, which completely resolved by day 9. A longitudinal scRNAseq analysis identified early type I IFNresponsive monocyte, macrophage, and dendritic cells in PBMC and BAL that mostly dissipated prior to the arrival of virus-specific CD8 + and CD4 + T cells. SARS-CoV-2-specific effector T cells were abundant in the pulmonary compartment, but undetectable in nasal turbinates, tonsils, and salivary glands, highlighting major differences in the localization of antigen-specific T cells in pulmonary and extrapulmonary mucosal tissues during SARS-CoV-2 infection.
Type I IFN is emerging as a critical mediator of control of SARS-CoV-2 infection (4-6). Type I IFN likely acts directly on infected epithelial cells to suppress viral replication, but it may also be important in driving the activation of antiviral responses by myeloid cells. On day 4 after infection there was a striking increase in two populations of IFN-activated myeloid cells in the BAL, one having monocyte-like features and the other more closely resembling alveolar macrophages, indicating that recruited monocytes and lung resident macrophages are key early responders to type I IFN after SARS-CoV-2 infection. In our study, the abundance of type I IFNactivated myeloid cells in the BAL positively correlated with viral loads in the nasal swabs and BAL. These results are consistent with data from African green monkeys showing a strong type I IFN gene signature in macrophages from lung tissue three days after SARS-CoV-2 infection (48). However, several scRNAseq studies from patients with COVID-19 have found that inflammatory monocytes/macrophage populations are increased with disease severity (3,10,12,14), suggesting that early type I IFN responses are host-protective, but prolonged activation of this pathway may be detrimental. Consistent with this hypothesis, in this model of mild disease, we observed a rapid resolution of IFN-activated phagocytes in both the blood and BAL by day 7 post-infection. On day 10 post-infection we observed a major reduction in the frequency of alveolar macrophages, which has also been documented in individuals with severe COVID-19 (12,75). Our data indicate that loss of alveolar macrophages can also occur after mild SARS-CoV-2 infection, so their loss may not be directly related to disease outcome. We also observed a major increase in the proportion of CD1c + dendritic cells in the BAL on day 10 post-infection, a time when Ag-specific T cells were abundant in the airways. Interestingly, CD1c + DCs displayed a distinct gene expression profile on day 10 with a notable increase in IL-1β and multiple pro-inflammatory chemokines. These cells may correspond to a population of inflammatory cDC2 that have been shown to appear in the lung after viral infection in mice (76). Notably, these cells were found to have an enhanced ability to stimulate both virus-specific CD4 + and CD8 + T cells. Collectively, our data indicate there may be two distinct phases of cellular innate immune responses in the BAL after SARS-CoV-2 infection. The first is characterized by an early wave of type I IFNactivated phagocytes, which may play an important role in the initial suppression of viral replication. The second is characterized by an abundance of pro-inflammatory dendritic cells, which may be key stimulators of the Ag-specific T cells that simultaneously arrive in the airways.
Ag-specific T cell responses were substantially greater in the BAL versus PBMCs, with the average sum of spike, nucleocapsid, and membrane-specific T cells reaching ~12% of CD4 + T cells and ~7% of CD8 + T cells in the BAL compared to ~1% and ~0.2% in PBMC, respectively. In the BAL, virus-specific Th1 cell responses preceded CTL responses. Indeed, the CD8 + T cell clonal burst likely had not yet peaked, evidenced by their maintained expression of Ki67 at day 10 post-infection. The lack of virus-specific T cells in the nasal turbinates, salivary glands, and tonsils, despite virus infection and subsequent clearance from these tissues, was surprising. The mechanisms underlying the lack of antigen-specific effector T cells in the infected nasal and oral mucosa are not clear. It remains possible that SARS-CoV-2-specific T cell responses were not detected in these sites because they produce molecules other than IFNγ, TNF, IL-2, granzyme B, or degranulation markers CD107a/b, after ex vivo peptide stimulation (i.e., Th2 cytokines, which were not measured here). Alternative techniques for functionally agnostic detection of SARS-CoV-2-specific T cell responses, such as the activation induced marker (AIM) assay, should be tested in future studies (17,69,77). Nevertheless, it is unlikely that T cells in the nasal turbinates, salivary glands, and tonsils have a completely different functional profile compared to their counterparts in the rest of the host. There may also be T cells in these tissues specific to antigens other than the ones tested here, although this too seems unlikely, as the peptide pools used contain numerous immunogenic peptides from across the entire viral genome (68). Lastly, it is possible that T cells accumulate in these tissues after day 10, and further studies will be needed to determine the longevity and breadth of SARS-CoV-2-specific T cell responses in tissues at later time points. The paucity of SARS-CoV-2-specific T cells in the nasal and oral mucosa may at least partly explain the observation of individuals previously infected with SARS-CoV-2 becoming re-infected (78)(79)(80).
Our findings support the hypothesis that control of primary SARS-CoV-2 infection in these tissues is largely T cellindependent, which is consistent with a recent report showing that rhesus macaques depleted of CD4-and/or CD8αexpressing cells prior to SARS-CoV-2 infection controlled the virus in the upper and lower respiratory tract, albeit perhaps with a slight delay (64). Another study also found that CD8 depletion in cynomolgus macaques had no impact on control of SARS-CoV-2 infection (81). It is important to point out, however, that our data do not rule out a critical role for T cells in other settings of SARS-CoV-2 infection. For example, T cells likely play a significant role when SARS-CoV-2 infection does not resolve quickly, such as during moderate and severe COVID-19. T cells have been implicated in control of SARS-CoV-2 in other susceptible animal models, like the human ACE2-expressing mouse lines and Syrian hamsters (82-84). Furthermore, nucleocapsid-specific CD8 + T cells are correlated with less severe disease in patients (20). In addition, preclinical studies suggest that depletion of CD8 + T cells from vaccinated monkeys prior to SARS-CoV-2 challenge significantly impairs control of virus replication (59). T cells may also play a major role in vaccine-elicited protection, and T cell-targeted peptide vaccines are currently being developed (20, 85,86). Vaccine-elicited T cells may prove critical in protection against SARS-CoV-2 variants of concern that are able to evade neutralizing antibodies, as T cell epitopes are thought to be more conserved across isolates (87)(88)(89)(90)(91)(92). It should also be noted that we cannot rule out the possibility that T cells may have played a role in clearance of any residual virus-infected cells remaining in the lungs after the first week, when T cells arrived in the tissue. Altogether, these data show that mild SARS-CoV-2 infection is associated with effective innate immune-mediated control. Future studies are needed to determine the importance of individual innate and adaptive immune cell types in suppression of SARS-CoV-2 replication.

Study design
The study was designed with the goal of assessing the differential kinetics of lung inflammation, viral replication, innate, and adaptive cellular immune responses against SARS-CoV-2 infection in a model of mild disease. The study had a predetermined end point of day 10 post-infection. The number of animals included in the study was based on previous experience in detecting immune responses in non-human primate infection studies, as well as practical limitations. The first 4 animals to be infected were selected for subsequent single cell RNA sequencing analysis.

Animals and infection
Six, male rhesus macaques aged 2.5 to 6 years, weighing 3-10 kg were infected with SARS-CoV-2/USA/WA-1 (Table S1). For infection, animals were anesthetized as described below and administered 2x10 6 TCID 50 total of SARS-CoV-2/USA-WA-1: 1x10 6 TCID 50 in 3mL intratracheally with a plastic gavage tube attached to 5 mL syringe, and 5x10 5 TCID 50 in 0.5mL intranasally in each nostril. The animals were examined daily with a health scoring sheet, as previously described (42). Animals were anesthetized with ketamine and dexmedetomidine at pre-infection (day -4 to -55), day 0, 1, 2, 3, 4, 7, and 10 (necropsy) for exams, PET/CT scans, viral swabs, blood and BAL fluid draws, and analysis of complete blood count and C-reactive protein. During anesthesia, animals were weighed and monitored for heart rate, respiratory rate, body temperature, oxygen saturation. Glycopyrrolate and atipamezole were given for recovery from anesthesia. For each animal, the pre-infection/baseline PET/CT scans are as follows DGCX and DG3V: day -55, DHGF and DHKM: day -17, and DGRX and DG4i: day -26. The pre-infection timepoints for swabs, blood, and BAL are as follows DGCX and DG3V: day -4, DHGF and DHKM: day -14, and DGRX and DG4i: day -5. Graphs including data plotted longitudinally, any pre-infection timepoints were represented as day 0.
All animal experiments were approved by Animal Care and Use Committee (ACUC) and all methods were performed under animal safety protocol LPD-25E at the National Institute of Health. Experiments were conducted in an AAALAC accredited aBSL-3 vivarium facility. Animals were singly housed in vented air cages with a 12-hour light/dark cycle. The animals were monitored twice daily, with a detailed physical exam once per day during the study. The Institutional Biosafety Committee approved all work with SARS-CoV-2 in the BSL-3 level facility and approved any inactivation methods used.

FDG-PET/CT Acquisition and Data Analysis
Rhesus were sedated and imaged by PET/CT during mechanical ventilation (Hallowell Ventilator Model 2002) at baseline and on day 3, and 9 post-infection. To reveal metabolic hyperactivity consistent with inflammation, a [ 18 F]-FDG dose of 0.5 mCi/kg was given intravenously 1 hour prior to PET imaging. During the uptake time a high-resolution CT scan of the lungs was acquired with a breath hold on a LFER 150 PET/CT scanner (Mediso Inc, Budapest, Hungary) as previously described (93). The raw CT and PET data were reconstructed using the Nucline software (Mediso, Inc, Budapest, Hungary) to create individual DICOM files that were co-registered using MIM Maestro (v. 7.0, MIM Software Inc, Cleveland, Ohio).
By aligning baseline PET/CT fused images and those taken at day 3 and 9 in MIM Maestro, specific lung regions with abnormal density (> ~ -550 HU) or metabolic activity (> ~ 1.5 SUV) were identified as volumes of interest (VOI) or lesions similar to methods used previously, rather than using whole lung volume of interest (94). For each animal, the lesion VOIs (day 3 in this study) were transferred to the aligned PET/CT images acquired at baseline and the day 9 time point, adjusting for position variations but keeping the same volume. Disease volume was estimated by using two density thresholds: tissues harder than -550 Hounsfield Units (HU) or harder than -300 HU for evaluating change over time. Regarding metabolic activity, PET parameters were estimated using a threshold of > 2 standardized uptake value (SUV). Similar reference VOIs were used to identify metabolically activated tissues (SUV > 2) within peri-carinal lymph nodes (LN). LN [ 18 F]-FDG uptake was measured in activated regions of the hilar and subcarinal LNs of each animal. Our analysis also included calculations of total lesion glycolysis (TLG). Two readers independently performed image analysis for each animal using consistent lesion labeling determined by a third reviewer. Three-dimensional projections of FDG uptake in the lung regions were generated using Osirix v 5.9 software (Pixmeo, Geneva, Switzerland) as previously described (95).

Blood and BAL collection
Blood and BAL collection procedures followed ACUC approved standard operating procedures and limits. Blood was collected in EDTA tubes and centrifuged at 2,000rpm for 10 min at 22°C to isolate plasma. After plasma removal, remaining blood was diluted 1:1 with 1x PBS. 15mL of 90% Ficoll- Paque density gradient (Cytiva Cat#17144002), diluted with 10x PBS, was added to SepMate PBMC Isolation Tubes (Stem-Cell Cat#85450) and centrifuged at 1,000 g for 1 min at 22°C, to collect Ficoll below the separation filter. Blood and PBS mix was added to the SepMate tube with Ficoll-Paque and centrifuged at 1,200 g for 10 min at 22°C. The upper layer was poured into a 50mL conical and brought to 50mL with PBS +1% FBS, and then centrifuged at 1,600rpm for 5 min at 4°C. The cell pellet was resuspended at 2x10 7 cell/mL in X-VIVO 15 media + 10% FBS for subsequent analysis. BAL was collected after intubation by instillation of 50mL of warm pharmaceutical-grade PBS, 10mLs at a time. For cellular analysis, BAL was filtered through a 100um filter into a 50mL conical and centrifuged at 1,600 rpm for 15 min at 4°C. The cell pellet was resuspended at 2x10 7 cell/mL in X-VIVO 15 media + 10% FBS for subsequent analysis.

Necropsy
Intravenous antibody was administered prior to euthanasia, as previously described (73). Briefly, prior to necropsy, 10mL of blood was drawn as a negative control and 100ug/kg of αCD45-biotin (clone: ITS_rhCD45 developed by Roederer Lab) was infused. The infusion was circulated for 10 min prior to terminal exsanguination and necropsy. The BAL and another ~60mL of blood were collected during exsanguination. After prosection of the lung and airways, specific lung regions observed to have abnormal HU density or FDG uptake in the day 3 images were scan-matched and collected separately as follows. The PET/CT scan images from day 3 post-infection were used to locate the airway vascular bundle adjacent to the abnormality. At necropsy, the lung parenchyma was resected back to reveal the bronchus upstream of the abnormality, which was followed to the target region to be resected. We have used a similar technique for PET/CT scan matching at necropsy after Mycobacterium tuberculoisis infection (96). LNs identified as having regions of SUV > 2.5 were collected separately from those with lower SUV on day 3. Normal lung sections were taken from each lobe. The resected lung sections, nasal turbinates, salivary gland, tonsils, spleen, and lymph nodes along with were then divided for RNA isolation, histology, and flow cytometry analysis. Viral RNA quantification RNA from the nose and throat was collected by swabbing each nostril or back of the throat, respectively, with a sterile swab for 10 s. Swabs were placed in 1mL viral transport media (1x HBSS, 2% FBS, 100ug/mL Gentamicin, and 0.5ug/mL amphotericin B) and stored on ice until RNA extraction. Swabs were vortexed in swab media before removing the swab tip. For RNA extraction 140uL of sample (plasma, 1 st BAL wash, or swab media) was processed using a Viral RNA mini kit (Qiagen Cat# 52906) and eluted in 50uL RNase Free water. For RNA isolation from tissues, tissue pieces were weighed before placing in 1mL RNAlater® media (Sigma Cat# R0901) and stored at 4°C overnight and then stored at -80°C long term. Tissues were then thawed and processed in the RNeasy Plus Mini kit (Qiagen # 74136) and eluted in 50uL RNase Free water. Eluted RNA was stored at -80°C long-term.
Extracted RNA was used in a RT-qPCR reaction for detection of total or subgenomic RNA from the N gene of SARS-CoV-2. Total RNA reactions amplify both genomic viral RNA and subgenomic viral mRNAs and are labeled as genomic throughout the manuscript to differentiate from the subgenomic viral mRNA-specific amplifications. Each sample was prepared in a 12.5uL reaction, with 2.5uL of eluted RNA, 3.25uL Taqpath  Copies per/mL or copies/gram were calculated based on standard curves generated for each RT-qPCR run, with RNA standard of known quantity and 10, 5-fold dilutions, run in duplicate. The limit of detection was based on the CT limit of detection from the standard curve in each run. For genomic RNA, this was also limited to CT<35, based on manufacturer's instructions. For subgenomic RNA cutoff CT<37 was used.
Tissue digestion Tissues were processed for single cell suspension before flow cytometry or peptide stimulation as follows. Spleen (approximately 0.5 inch × 0.5 inch portion) and lymph nodes were placed in 5mL of PBS + 1% FBS in a gentleMACS C tube (Miltenyi Cat#130096334) and run on gentleMACS Octo Dissociator (Miltenyi), with m_spleen_02_01 setting, then filtered through a 100um filter into a 50mL conical and centrifuged at 1,600rpm for 5 min at 4°C. Salivary gland was gentleMACS dissociated as above, and after centrifugation cells were resuspended in 7mL 44% Percoll® (Sigma Cat# P1644) with 1xPBS and centrifuged at 2,000rpm for 20 min at 22°C without brake. The tonsil and lung were gentleMACS dissociated in 5mL digestion buffer (RPMI + 50U/mL DNase I + 1mg/mL hyaluronidase + 1mg/mL collagenase D (Roche)) and then agitated on a shaker at 220rpm for 45 min at 37°C. Digestion reaction was stopped with equal parts PBS + 20% FBS and centrifuged at 1,600rpm for 5 min at 22°C. The cell pellet was resuspended in Percoll gradient, as above for salivary gland. After processing, the spleen and lung were cleared of red blood cells by resuspending cell pellet in 2mL of ACK Lysing Buffer (Quality Biologicals Cat#118-156-101) for 2 min at room temperature, then stopping the reaction with 10-20mL of PBS + 1%FBS. Cells were resuspended at 2x10 7 cell/mL in X-VIVO 15 media + 10% FBS for further analysis.

Flow cytometry and antibody staining
Cells were resuspended in 50uL surface stain antibodies diluted in PBS + 1% FBS and incubated for 20 min at 4°C. Cells were washed 3 times with PBS + 1% FBS, before fixation with eBioscience Intracellular Fixation & Permeabilization Buffer Set (Thermo Cat# 88-8824-00) for 16 hours at 4°C.
After fixation cells were centrifuged at 2,200rpm for 5 min at 4°C without brake and washed once with eBioscience Permeabilization Buffer. Cells were resuspended in 50uL intracellular stains diluted in eBioscience Permeabilization Buffer, and stained for 30 min at 4°C. After staining cells were washed with eBioscience Permeabilization Buffer 2x and resuspended in PBS + 1% FBS + 0.05% Sodium Azide for running on the BD Symphony platform.
B cells were resuspended in 50uL Human Fc-Block (BD Cat#564220) diluted to1:500 in PBS + 1%FBS and incubated for 30 min at 4°C prior to washing and surface staining. Antibodies used can be found in Table S2. The biotinylated intravenous antibody was detected with Streptavidin conjugated to the indicated fluorophore included in the surface stain. The intravenous stain (i.v. stain) positive cells are localized to the tissue vasculature, while i.v. stain negative cells are localized to the tissue parenchyma. For phenotypic marker analysis of antigen-specific T populations, a cutoff of >35 peptide-specific T cells per samples was used for subsequent analysis to prevent the introduction of error from quantifying too few events.
Fluorescent ELISA High-binding polystyrene half-area 96 well plates (Greiner Bio-One Cat#675077) were coated with recombinant SARS-CoV-2 Spike Trimer (Miltenyi Cat#130-127-683) at 2ug/mL and 25uL per well for at least 16 hours at 4°C. Plates were washed with Wash Buffer (PBS + 0.05% Tween-20) using an automatic plate washer (5x with 180uL). Plates were blocked with 25uL per well of Blocking Buffer (PBS + 1% BSA + 0.01% Tween-20) for at least 16 hours at 4°C. Plates were washed again as above. Plasma and BAL fluid were initially diluted 1:10 with Blocking Buffer and 3 -fold serial diluted for a total of 12 dilutions per sample. The dilutions were plated at 25uL per well and incubated for 1 hour at 37°C. Plates were washed again as above. Goat anti-monkey IgG-HRP (Invitrogen Cat#PA1-84631) and goat anti-monkey IgM-HRP (Alpha Diagnostics Cat#70031) were diluted with Blocking Buffer 1:3000 and 1:2000, respectively. Secondary antibodies were plated at 25uL per well and incubated 30 min at 37°C. Plates were washed again as above. Plates were developed with QuantaRed Enhanced Chemifluorescent HRP Substrate Kit (Thermo Cat#15159), following manufacturer's instructions for reagent preparation, and incubated for 15 min for IgG and 30 min for IgM, and the reaction was stopped with the manufacturer provided Stop Solution. Plates were immediately read on a SpectraMax Gemini fluorescent microplate reader (Molecular Devices) with Excitation of 570nm and Emission of 585nm.

Live-virus neutralization assay
Vero-E6 cells stably expressing human TMPRSS2 (Vero-E6T2) (97), a gift from the laboratory of Jonathan Yewdell (NIAID), were maintained at 37°C and 5% CO 2 in D10+ medium (DMEM +10% FBS, 1X Glutamax, 1X Anti-Anti (Gibco) and 250ug/mL Hygromycin B (InVivoGen)). One day before the assay, cells were plated in 12-well plates (Falcon Cat#353043) at a density of 0.4 million cells per well in 2mL D10+ medium. The day of the assay, 12-well plates were washed twice with D2 medium (DMEM + 2%FBS, 1X Glutamax). Rhesus macaque plasma samples, collected at day 0 and day 10 post-infection with SARS-CoV-2, were heat inactivated at 56°C for 30 min. After an initial 1:5 dilution, plasma samples were further two-fold serially diluted with D2 medium and 100uL of each dilution was incubated for one hour at 37°C with an equivalent volume of D2 medium containing 40-50 PFU of SARS-CoV-2 USA-WA1/2020 virus. After the incubation, 100uL of plasma/virus mixture was added to Vero-E6T2 in duplicate wells containing 300uL of D2 medium. The virus was allowed to absorb onto the cells for one hour at 37°C and 5% CO 2 with occasional mixing. At the end of the incubation, 1.5mL of DMEM medium supplemented with 0.6% of methyl cellulose was added per well and the plates were maintained for another 66 to 72 hours at 37°C and 5% CO 2 . At the end of the incubation, the medium was removed and replaced with 1mL of Crystal Violet solution containing 5% ethanol and 3% neutral buffered formamide. The wells were stained for 20 min at room temperature, then washed briefly with deionized water and air dried in the biosafety cabinet. Once dry, the plates were scanned at 300dpi and plaques were counted from the images. Percent of inhibition was calculated based on the number of plaques observed for each dilution compared to controls. The IC50 of the percent inhibition for the dilution series was calculated based on nonlinear regression. Each experiment included the following controls: uninfected cells as negative control, virus in absence of rhesus macaque serum and virus incubated with a dilution of a known positive serum (kindly provided by SAB Biotherapeutics).
Immunohistochemistry and RNA scope Tissue for histology were collected in 10% neutral buffered formalin and stored at room temperature for 16 hours. Fixed tissues were transferred to 70% ethanol and stored at room temperature until processing. Slides were cut 10-microns thick using standard RNAse precautions. Immunohistochemical slides were deparaffinized and treated with AR6 Buffer (Akoya Biosciencecs, USA) for 20 min at 100°C. Tissues were then permeabilized using 0.2% TritonX 100 (Millipore Sigma, USA) for 10 min. After blocking, slides were incubated with primary antibodies against CD62P (clone EPR22850-190, Abcam, USA) and fibrin (clone 59D8, Millipore Sigma, USA) at a 1:500 and 1:200 concentration, respectively. Following washing, slides were stained according to the protocol for Im-mPRESS Duet Double Staining Polymer Kit (Vector Laboratories, USA) and counter-stained with hematoxylin. Slides used for in situ hybridization were deparaffinized and treated with RNAscope epitope retrieval buffer (ACD Biotechne, USA) for 20 min at 100°C. Endogenous peroxidases were then blocked with hydrogen peroxide and tissue permeabilized with a diluted RNAscope protease plus for 20 min at 40°C. Probes for SARS-CoV-2, containing 20 pairs of probes spanning S gene (Category # 848561, ACD Biotechne, USA), were incubated for 2 hours at 40°C. Slides were then processed according to RNAscope 2.5 HD Assay-RED (ACD Biotechne, USA) protocol and counterstained with hematoxylin. Slides stained immunohistochemically or by in situ hybridization were imaged using Aperio VERSA (Leica Microsystems, USA) and analyzed using quPath, an open-source software developed by the University of Edinburgh. SARS-CoV-2 puncta were confirmed using both positive and negative controls to ensure accurate staining.

Single cell RNA sequencing and data analyses
Cells from the BAL fluid and PBMCs from blood were obtained as described above and cryopreserved in 1ml of RPMI + 40% FBS + 15% DMSO. PBMC and BAL samples from days -7, 4, 7, 10 for monkeys DHGF, DG3V and DHKM and both sample types from days -7, 4 and 10 for monkey DGCX were processed for scRNAseq using the 10X Genomics Chromium Single Cell 3′ kit (v3.1). Briefly, cryopreserved samples were quickly thawed using a water bath set to 37°C and washed twice using 10% FBS in RPMI. Samples were then stained with unique TotalSeq-A hashtag antibodies (HTO) as per manufacturer's (Biolegend) protocol. Equal number of cells from each sample were pooled and super-loaded on a 10X Genomics Next GEM chip and single cell GEMs were generated on a 10X Chromium Controller as previously described (98). Subsequent steps to generate cDNA and HTO libraries were performed following 10X Genomics and Biolegend's protocol respectively. Libraries were pooled and sequenced on an Illumina NovaSeq S1 and S2 flow cells as per 10X sequencing recommendations.
The sequenced data was processed using cellranger (version 5.0) to demultiplex the libraries. The reads were aligned to Macaca mulatta mmul_10 genome to generate count tables. The count tables were then further processed and analyzed using the Seurat (version 4.0) in R (version 4.1.0). Samples from different PBMC and BAL libraries were integrated using IntegrateData function to account for possible batch effects and to generate one integrated dataset for each tissue type. Cells were then filtered for less than 15% mitochondrial contamination and only singlets as determined by the HTOs were included resulting in 16,769 PMBC and 7,274 BAL cells for downstream analysis. Data were normalized and scaled and FindVariableFeatures function was used to identify variable genes to subset and integrate the data to correct for animal bias. Principal component analysis was performed to find neighbors and clusters and UMAP reduction (Page numbers not final at time of first release) 10 was performed with 20 dimensions. FindAllMarkers with a filter of log fold change ≥ 0.25 and percent of cells expressing the marker ≥ 0.25 was used to identify gene markers that distinguish the cell clusters, and the clusters were manually assigned cell types based on identified canonical markers. Myeloid and T (and NK in case of PBMC) cell clusters were further subclustered and clusters were again manually annotated based on gene markers determined by the FindAllMarkers function. Differentially expressed genes between timepoints of a particular cluster were identified by running FindMarkers function with MAST and comparing one timepoint to all other timepoints or one timepoint to another in a pairwise manner. Genes with a log fold change ≥ 0.5, percent of cells expressing the marker ≥ 0.25 and adjusted p value ≤ 0.01 were considered significant and these genes were hierarchically clustered and displayed as a heatmap using the ComplexHeatmap function in R. Gene ontology enrichment analysis of genes up-regulated at a particular timepoint was performed using clusterProfiler to identify biological processes (adjusted p value ≤ 0.05). The AverageExpression function was used to calculate average gene expression of IFN and IFN-stimulated genes across all cells over time and was visualized using pheatmap.
Spearman's correlation test was performed between viral loads from various sites and fraction of cells in a particular cluster at all available timepoints and filtered for adjusted p value < 0.05. Correlations were visualized using a network diagram generated using igraph in R. Hashtag antibodies used can be found Table S2.
Statistical analysis Data were analyzed using a two-way ANOVA with a Sidak's multiple comparison test, a Turkey's multiple comparison test, or a Dunnett's multiple comparison test; a oneway ANOVA using a Dunnett's multiple comparison test; or a two-tailed paired t test. Tests used are indicated in figure legends. For all statistical analysis p<0.05 for the given test is considered significant: * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001 SUPPLEMENTARY MATERIALS www.science.org/doi/10.1126/sciimmunol.abo0535 Figs. S1 to S6 Tables S1 and S2 Data file S1 (Page numbers not final at time of first release) 18