Adaptive MHC-E restricted tissue-resident NK cells are associated with persistent low antigen load in alveolar macrophages after SARS-CoV-2 infection

Natural killer (NK) cells are innate lymphocytes with potent activity against a wide range of viruses. In SARS-CoV-2 infection, NK cell activity might be of particular importance within lung tissues. Here, we investigated whether NK cells with activity against Spike+ cells are induced during SARS-CoV-2 infection and have a role in modulating viral persistence beyond primary clearance from nasopharyngeal and tracheal tissues. We performed an integrated analysis of NK cells and macrophages in blood and bronchoalveolar lavage fluids (BALF) of COVID-19 convalescent non-human primates in comparison to uninfected control animals. SARS-CoV-2 protein expression was detected for at least 9–18 months post-infection in alveolar macrophages. Convalescent animals segregated into two groups based on cellular phenotypes and viral persistence profiles in BALF. The animals with lower persistent antigen displayed macrophages with a regulatory phenotype and enhanced MHC-E restricted NK cell activity toward cells presenting peptides derived from the SARS-CoV-2 Spike protein leader sequence, while NK cell activity from the other convalescent animals, control animals and healthy humans were strongly inhibited by these Spike peptides. The adaptive NK cell activity was not detected in blood but in tissue-resident NK cells, and cross-reacted against MERS-CoV and SARS-CoV Spike-derived peptides.


Introduction
The alveolar membrane is the largest surface of the body in contact with the outside environment, making the lungs the most vulnerable organ of the whole body to microbial assault 1 . Protection of this barrier needs a rapid and effective immune response to prevent pathogen infections and spread.
Concomitantly, the duration and the amplitude of the resulting in ammation must be e ciently controlled to avoid tissue damage. From a global and historic perspective, the scope and scale of lower respiratory tract infections is greater than most other infections. Indeed, viral pneumonias are among the most lethal and pathologic human diseases. The current global outbreak of severe acute respiratory syndrome caused by the type 2 coronavirus (SARS-COV-2) is a dramatic contemporary example of the ability of viral pneumonias to rapidly disseminate and cause severe disease in human populations. SARS-CoV-2 replicates in both upper and lower airways 2 , and viral RNA is typically detected in patients 1-3 days before the onset of symptoms, with viral load in the upper respiratory tract peaking within the rst week of infection, followed by a gradual decline over time. SARS-CoV-2 infection triggers cellular and humoral immune responses as well as activation of neural pathways, which contribute to distant in ammatory effects. Although it is generally considered that SARS-CoV-2 is an acute infection that does not become chronic, a growing number of studies suggest that some SARS-CoV-2-infected individuals do not clear the virus over long periods of time [3][4][5][6] .
Macrophages are innate immune cells that sense and respond to microbial threats by producing in ammatory molecules, phagocytosing pathogens and promoting tissue repair. A dysregulated macrophage response can be damaging to the host, as for instance in infection-induced macrophage activation syndrome as observed during SARS-CoV-2 infection 7,8 . Macrophages are abundant in the lung, comprising about 70% of the total leukocyte population 9 . Monocytes and macrophages can both be infected by SARS-CoV-2 through ACE2-dependent and ACE2-independent pathways [10][11][12][13][14] .
Natural killer (NK) cells are innate immune responders critical for viral clearance and immunomodulation. Interferon-alpha (IFN-α) is known for its activating role on NK cells. Despite their vital role in viral infections, the contribution of NK cells to SARS-CoV-2 immunity is not clear 15,16 . NK cell frequencies decrease in the blood during the early phase of SARS-CoV-2 infection 17 − 19 , and display an exhaustion phenotype 20 − 22 . NK cells from healthy individuals can directly kill SARS-CoV-2-infected cells in vitro, but NK cells from individuals with a moderate or severe disease displayed impaired NK cell activity 20 . Recent reports showed a correlation between high levels of NK cells in the blood and a rapid decline in viral load 20 . The rapid decrease of blood NK cells might be due to their death and/or to recruitment to the lungs or other tissues 23 . Previous studies on NK cells in other viral infections have clearly shown that tissue NK cells display distinct characteristics than NK cells in the blood. During SARS-CoV-2 infection, it is therefore essential to get insights on NK cells not only present in the blood but also in tissues, particularly in respiratory mucosal tissues.
The lung seems to provide an environment propitious for the development of adaptive-like NK cells 24 . Adaptive NK cells have been predominantly described during chronic infections, such as CMV and SIV infections 18,25−28 . A common denominator for the majority of adaptive NK cells is the expression of the activating heterodimeric receptor CD94/NKG2C that binds to HLA-E. The latter is a non-classical HLA molecule that displays a comparably restricted expression pattern, very limited polymorphism and binds hydrophobic nonamers 29 . The strength of the HLA-E/NKG2C interaction in conjunction with variable conditions of co-stimulation and a favorable cytokine milieu is considered to be the molecular basis for the complex inter-and intra-individual heterogeneity of adaptive NK cells. In CMV infection, presentation of the viral peptides via HLA-E on monocyte/macrophages might be key for a secondary expansion of NK cells 29 . An increase of adaptive NK cells in blood from patients with severe COVID-19 has been reported, although the identi cation of adaptive NK cells was based solely on cell phenotype without any functional analysis nor data in tissues 18 .
Here we took advantage of a non-human primate model (cynomolgus macaques) to evaluate NK cells in tissues after SARS-CoV-2 infection. We raised the hypothesis that adaptive NK cells arise preferentially in the lung in response to SARS-CoV-2 infection. We searched for SARS-CoV-2 derived peptides capable to bind to HLA-E and modulate NK cell activity. We investigated the presence of SARS-CoV-2-trained NK cells in the blood and lung. We performed functional assays to identify the presence of adaptive NK cells and evaluated whether they can be detected in the blood and/or in the lung. We analyzed whether the presence of Spike-speci c adaptive NK cells is associated or not with viral persistence. Our study reveals the presence of viral antigen in alveolar macrophages up to at least 18 months after SARS-COV2 infection in macaques. Moreover, we uncover an association between lower level of persisting viral antigen in alveolar macrophages and the activity of Spike-speci c adaptive NK cells in the lung.

Persistent phenotypic alterations in bronchoalveolar macrophages from SARS-CoV-2 convalescent non-human primates
We aimed to investigate if viral antigens could persist over prolonged periods of time after SARS-CoV-2 infection in the lung and continue to stimulate immune cells. We used the nonhuman primate model to address this question in order to have access to tissues while still being as close as possible to the human infection and to human tissue parameters. Fifteen cynomolgus macaques were infected with SARS-CoV-2 (Wuhan strain) and followed between 6 and 18 months (Supplementary Table 1) as previously described 30 . The viral load (VL) in tracheal and nasal swabs reached median peak levels of 7.9x10 8 copies/ml by day 3 post-infection (p.i.) (Suppl. Figure 1). All animals subsequently became PCR negative by day 14 p.i., and remained negative for the presence of SARS-CoV-2 RNA in nasal and tracheal swabs during the entire follow-up (Suppl. Figure 1). We investigated whether the virus could be present in other body parts of the animals after convalescence. Alveolar macrophages reside on the epithelial surface of the alveoli and are thus in direct contact with the environment and foreign particles entering the lungs, where they have been described to be infected by some pathogens, including bacteria and viruses [31][32][33][34] . To characterize the alveolar macrophages from the convalescent monkeys, we collected bronchoalveolar lavage uids (BALF) from the 15 animals between 6 and 18 months after SARS-CoV-2 infection and compared them with BALF samples obtained from 7 healthy cynomolgus macaques that were never infected with SARS-CoV-2. We used multi-parameter ow cytometry and unsupervised analysis to investigate macrophage phenotypic pro les. The strategy for gating the macrophages is illustrated in Fig. 1a. Despite the long time after infection, we found that BALF macrophages segregated according to healthy and convalescent monkeys (all 15 animals except one) ( Fig. 1b and 1c).
Macrophages isolated from convalescent monkeys generally displayed higher levels of CD206, CD4, CD11c, CD226, MHC-E and IL-10 expression, and lower CD16 and CXCR4 expression when compared to healthy animals ( Fig. 1b-d). Thus, macrophages from convalescent monkeys were clearly distinct from those in control monkeys.
We next compared the macrophage phenotypes between the convalescent animals. Using principal components analysis (PCA) based on 12 phenotypic parameters, convalescent monkeys also clustered separately from healthy controls. Moreover, convalescent monkeys segregated into two groups (Fig. 1e).
Given the difference in the macrophage phenotype among convalescent monkeys, we analyzed whether there were differences in VL during primary infection between the two animal groups. We compared the area under the curve (AUC) of VL as quanti ed by PCR in the nasal and tracheal swabs during the acute phase of the infection between each group ( Supplementary Fig. 3). The AUC VL were higher in group 1 than group 2 in tracheal swabs (Fig. 1g). Group 1 also showed a trend for higher VL in tracheal swabs at day 1 p.i as compared to group 2 (p = 0.09). Collectively, our results show that BALF macrophages displayed profound alterations even after a long period after SARS-CoV-2 infection, suggesting an imprinting of the alveolar macrophages. Of note, monkeys with the strongest imprinted pro le toward regulatory M2-like macrophages (group 2) were those showing the lowest VL during acute infection.

Detection of SARS-CoV-2 protein in alveolar macrophages one year post-SARS-CoV-2 infection
We then investigated whether SARS-CoV-2 could be present in the lung despite the fact that the animals became PCR negative in nasal and tracheal swabs by probing viral antigens in alveolar macrophages of the 15 convalescent monkeys. To this end, we performed an antibody-based uorescent staining of SARS-CoV-2 Spike in alveolar macrophages in vitro, and detected Spike protein expression to variable extents in the animals (Fig. 2a). To quantify the frequency of macrophages harboring viral Spike protein and its expression level, we performed ex-vivo intracellular ow cytometry staining of Spike protein in BALF cells from the 15 convalescent and 7 control monkeys (Fig. 2b). The median frequency of Spike + macrophages was 21.3%, ranging from 6.1 to 72.7% depending on the animal (Fig. 2c). The seven monkeys from group 1 showed the highest frequencies of Spike + macrophages (median = 38.80%), and were among the 9 animals displaying frequencies around or above the median levels ( Fig. 2c). To con rm the presence of the virus, we searched for viral RNA in the BALF cells. A RT-qPCR was applied to all available samples (6 animals per group). SARS-CoV-2 RNA was detected in BALF cells to a variable extent among the animals (Fig. 2d). Viral RNA levels were higher in group 1 than group 2 animals (Fig. 2d). Thus, these analyses allowed identifying animals with high frequencies of alveolar macrophages positive for SARS-CoV-2 Spike protein and viral RNA even after 6-18 months postinfection, in particular in group 1.

Spike + alveolar macrophages display a distinct phenotype
We next aimed at determining the phenotype of BALF macrophages harboring Spike proteins. SARS-CoV-2 + macrophages were gated manually and projected onto a Uniform Manifold Approximation and Projection (UMAP) plot as shown in Fig. 1b. Most Spike + macrophages clustered together (Fig. 2e).
To identify subsets of Spike + macrophages and to quantify potential inter-group differences in their relative abundance, a PhenoGraph algorithm was applied to the BALF macrophages rendering 30 distinct clusters ( Fig. 2g and h). Among them, clusters LWHT_2, 5, 9, 12, 14,18, 23 and 24 were predominantly found in the control monkeys ( Fig. 2h-i and Supplementary Fig. 4a). Their BALF macrophages were characterized by a stronger expression of CD16 and CD64, CD163 and HLA-DR (Fig. 2j). Group 1 contained higher frequencies of clusters LWHT_3, 4, 6, 7, 8 and 10 compared to control animals ( Fig. 2h-i and Supplementary Fig. 4a). The clusters LWHT_3 and 7 expressed high levels of the Spike protein and corresponded to macrophages expressing higher levels of CD14 and low levels of CD4, CD206, IL-10 and MHC-E (Fig. 2j). Consistent with these ndings, the tracheal swab VL (AUC) in acute infection correlated positively with CD14 and CXCR4 expression (p = 0.05, r = 0.52 and p = 0.025, r = 0.58, respectively) ( Supplementary Fig. 4b). The other clusters with high level of Spike proteins were LWHT_21, 22, and 26, but their frequencies were below 0.43, 0.1 and 0.27% respectively ( Supplementary Fig. 4a). Group 2 displayed higher levels of clusters LWHT_1 compared to Group 1 and control animals. Cluster LWHT_1 corresponded to macrophages expressing high levels of CD4, CD206, IL-10 and MHC-E (Fig. 2i). When comparing the expression levels of all markers to each other, IL-10 and MHC-E expression were strongly correlated (p = 1.89e-7 , r = 0.95), further indicating that they are expressed by the same subpopulations ( Supplementary Fig. 4b). Altogether, most macrophages harboring Spike protein displayed a similar phenotype, corresponding to the predominant macrophage phenotype in group 1 and distinct from the regulatory-like macrophages in group 2.
Elevated NK cell levels in BALF of group 2 convalescent monkeys We next assessed whether NK cells differed between group 1 and group 2 convalescent monkeys. NK cells were gated as commonly described for macaque NK cells (i.e. CD45 + CD3 − CD20 − NKG2 A/C + ) 35,36 (Supplementary Fig. 5a). In Old world monkeys, such as macaques, most NK cells express NKG2 37 , but the antibodies available for these species do not differentiate between NKG2A and NKG2C and we therefore used the nomenclature NKG2A/C here. We rst analyzed if NK cell tissue distribution was altered 6-18 months post-infection. NK cells were quanti ed in the lungs (BALF), blood and bone marrow by ow cytometry (Supplementary Fig. 5a). NK cell frequencies in blood from group 1 convalescent monkeys were higher than in healthy and group 2 monkeys (Fig. 3a). Conversely, NK cell frequency in BALF was higher in group 2 than in healthy and group 1 monkeys (Fig. 3a). No difference was seen regarding bone marrow NK cell frequency between the groups (Fig. 3a).
Since we observed different NK cell frequencies in the blood and BALF depending on the animal group, we next determined by ow cytometry the frequency of major tra cking receptors for NK cells (CCR7, CD62L, CXCR3 and CXCR4) in the blood and BALF. CXCR3 is considered as the major homing receptor for human NK cells to the lung, also during SARS-CoV-2 infection 38, 39 . The frequencies of CXCR3 + and CCR7 + NK cells in BALF were higher in group 2 than group 1. There was no difference regarding CD62L between the groups ( Fig. 3b and 3c). This pattern is indicative of a higher capacity of NK cells from group 2 to home to the lung compared to group 1. However, CXCR3 + NK-cell frequencies were increased in blood of group 1 as compared to group 2 and control animals. Additional mechanisms could participate to the differences between animals in term of NK cell distribution pattern. We reasoned that the increases of NKcell frequency in lung of convalescent monkeys from group 2 might also be due to in-situ expansion of resident NK cells. Therefore, we next analyzed markers linked to tissue residency. By analogy with tissue resident human T lymphocytes and NK cells, we investigated the expression of CD69, the integrin CD49a and CD103 (Fig. 3d). NK cells in BALF expressing the tissue residency markers CD49a or CD69 were increased in all convalescent monkeys or in group 2, respectively. We also analyzed Ki-67 expression as a marker of recent proliferation. The NK cells from BALF showed higher expression levels of Ki-67 than those from the blood or bone marrow ( Fig. 3d and Fig. 4a-b). Frequencies of Ki-67 NK cells were higher in group 2 than group 1 in BALF. Together, these results demonstrate altered tissue distributions of NK cells in the convalescent monkeys. They indicate that NK-cell homing behavior differs between convalescent and healthy monkeys. NK cells from group 1 animals were increased in blood but not in BALF suggesting an impaired homing capacity to the lung. In contrast, NK-cell frequencies were increased in BALF of group 2 as compared to the other animals, and this was associated with a higher tissue homing receptor expression (CXCR3, CCR7), stronger proliferation (Ki-67) and lung tissue residency markers (CD49a, CD69) suggesting in-situ differentiation.

Convalescent animals with regulatory-like macrophages also harbor distinct NK cells in BALF
To determine if SARS-CoV-2 infection can have a long-term impact on the function and composition of the NK-cell pool, ow cytometric analysis of NK cells from blood, BALF and bone marrow was performed including 12 phenotypic markers of function and tissue residency via an unsupervised approach. We applied an UMAP analysis on tissue NK cells ( Fig. 4a-c). PCA analysis showed that monkeys belonging to group 2 clustered together and harbored phenotypic NK-cell populations clearly distinct from control and group 1 animals in BALF (Fig. 4d). Conversely, monkeys in group 1 displayed NK-cell populations in BALF similar to those in control monkeys (Fig. 4d). The same approach performed on NK cells from blood and bone marrow showed differences between control and convalescent animals but did not allow to cluster the animals into groups (Fig. 4d). Thus, NK cells in BALF clustered distinctly according to the tissues. In BALF, NK cells from group 2 were more distinct from control animals than group 1.
We further dissected the segregation pro les of NK-cell populations ( Fig. 4e and Supplemental Fig. 5a and 5b). In BALF, group 2 monkeys showed decreased frequencies of NK cells expressing NKp44 and IFNg, and increased levels of NKP80 + and CD107a + NK cells when compared to control and group 1 monkeys, as well as increased levels of cells expressing Ki-67 and GZMb when compared to control monkeys (Fig. 4f). Thus, NK cells from group 2 segregated from the other animals in BALF and displayed a highly differentiated, cytotoxic phenotype.
We next asked whether there is a relation between the frequency of Spike + macrophages in BALF and the NK cell phenotype. We analyzed the potential correlations between the levels of Spike + macrophages and NK cells expressing functional markers (IFN-g, CD107a, GzB) and/or markers of education (NKG2A/C) ( Fig. 4g). Both in the BALF and blood, NKG2A/C low NK cells, which are generally considered as cells with a more educated phenotype 40,41 , correlated negatively with the levels of Spike + macrophages. In addition, CD107a NK cells in BALF also correlated negatively with Spike + macrophage frequencies (Fig. 4g). In contrast, IFN-g + NK cells correlated positively with the frequency of Spike + macrophages in BALF ( Fig. 4g). Thus, NK cells with a phenotype frequently found in group 1 correlated positively with macrophage-associated persisting antigen load, while the cytotoxic NK cells from group 2 correlated negatively with the antigen load.
Peptides derived from the Spike protein leader sequence bind to HLA-E and inhibit NK cell activity Next, we aimed to analyse whether NK cells of group 2 harbor a stronger antiviral activity. Based on our results showing an up-regulation of MHC-E on macrophages in group 2, we rst investigated whether NK cells in group 2 were trained to speci cally recognize Spike + target cells in a MHC-E dependent manner.
We set up a functional assay to analyze MHC-E restricted viral suppressive activity of NK cells toward target cells presenting Spike antigens via MHC-E. To this end, we searched for SARS-CoV-2-derived peptides with a high probability of binding to MHC-E. We analyzed the whole Spike protein but focusing on the leader sequence (LS), assuming a higher probability in identifying peptides binding to MHC-E in vivo in analogy to the cellular peptides that generally bind to MHC-E and that derive from the LS of classical MHC class I molecules 42 (such as the VL9 peptide) or from the LS of the stress protein Hsp60 43 .
We analyzed the Spike LS of SARS-CoV-2 as well as those of MERS-CoV and SARS-CoV-1. We used the Immune Epitope Data Base (IEDB) tool to predict the capacity of Spike-derived peptides to bind to HLA-E*0101 and HLA-E*0103, the only two functional HLA-E alleles in humans. We found > 60 nonamer peptides (out of 1169, 1087 and 1097 analyzed Spike peptides from MERS-CoV, SARS-CoV-1 and SARS-CoV-2, respectively) showing a probability to be loaded by HLA-E (Fig. 5a). Among those, we selected peptides encoded by sequences similar to the canonical sequence motif of HLA-E binding peptides 44 . In this way, 2, 6 and 7 peptides were identi ed for MERS-CoV, SARS-COV-1 and SARS-CoV-2, respectively ( Fig. 5b and 5c). These peptides were all localized in similar regions of the LS from all three viruses (between amino acid positions 2-11, 10-24 or 40-56) (Fig. 5b). The 15 Spike LS-derived peptides were then analyzed for their capacity to stabilize HLA-E at the cell surface (HLA-E needing to bind peptides to be stably presented at the cell surface). The viral peptides increased in a concentration dependent manner the HLA-E expression on MHC-I devoid K562 cells stably transduced with HLA-E*0101 (K562-E*0101) (Fig. 5d), indicating that these peptides are able to bind HLA-E. HLA-E expression levels varied depending on the peptide, suggesting distinct HLA-E biding properties between the peptides.
We then analyzed NK cell activity against target cells presenting Spike LS-derived peptides via HLA-E. To validate the assay, we rst studied NK cells from human healthy donors. These donors were all naïve of SARS-CoV-2 infection or vaccination. NKG2A + NK cells from the donors were co-cultured with K562-E*0101 cells pre-loaded or not with the peptides, and NK-cell degranulation activity was measured by measuring CD107a expression ( Fig. 5e and 5f). As expected, the HLA-I-derived VL9 control peptide inhibited NK cell activity, while the control Hsp60 peptide did not, validating our assay (Fig. 5f). When pulsing cells with the viral peptides, NK cell activity was modulated in a peptide dependent manner. For instance, peptide 310 from MERS-CoV, peptide 315 from SARS-CoV-1 and peptides 319, 321 and 325 from SARS-CoV-2, blocked NK-cell degranulation, whereas the other peptides did not. Of note, peptides 310, 315 and 319 all mapped in the N-terminal portion of the LS for each virus (positions 2-11) (Fig. 5b), suggesting a key role of this region in inhibiting the MHC-E dependent NK-cell activity.
We next performed the assay using NK cells from the control monkeys, and focused on the peptides having shown the capacity to modulate human NK-cell activity (peptides 310, 315, 319 and 321). As expected, the VL9 peptide inhibited the NK cells isolated from the blood of control monkeys, while the HSP60 peptide did not (Fig. 5g). Two out of the four peptides (peptides 310 and 315) blocked degranulation similarly to VL9 (Fig. 5g). NK cells isolated from BALF of control monkeys were inhibited by SARS-CoV-2 peptides even more strongly than those from the blood as the four viral peptides inhibited their degranulation (Fig. 5h).
NK cells with enhanced MHC-E restricted Spike-speci c activity in BALF of group 2 convalescent animals Since we observed changes in the NK-cell phenotype in the convalescent monkeys as compared to healthy monkeys, and in particular signs of more educated and cytotoxic NK cells in group 2, we compared the MHC-E dependent suppressive activity of NK cells isolated from convalescent monkeys. The MHC-E dependent suppressive activity of NK cells in blood was similar between control and convalescent monkeys (Fig. 5i). Thus, the LS-Spike peptides also inhibited activity of NK cells from convalescent monkeys in blood. We then analyzed NK cells from BALF. MHC-E-dependent NK-cell activity was inhibited by the LS-spike peptides in healthy and group 1 convalescent monkeys (Fig. 5j). In contrast, NK cells isolated from group 2 showed a strong activity against all LS-Spike peptides tested (Fig. 5j). Altogether, we identi ed peptides derived from the LS of the SARS-CoV-2 Spike protein inhibiting NK cells from healthy monkeys and humans. In particular, the N-terminal part of the Spike LS encoded peptides were inhibitory for MHC-E dependent NK cell activity. Moreover, our data unraveled that SARS-CoV-2 infection can induce NK cells with enhanced MHC-E restricted activity against cells presenting Spikeantigens. This adaptive NK-cell activity was enhanced for half of the animals (group 2). It was observed in BALF but not in the blood. Moreover, it was cross-reactive against the LS-Spike peptides from other Coronaviruses.

Discussion
Innate immune responses have been shown to play a vital role against COVID-19, including IFN-a and TGF-b whose role consists, among others, to modulate NK cell responses 20,45,46 . However, more information is needed on the role of cellular innate immunity, particularly in tissues, during SARS-CoV-2 infection. The tissue-speci city of macrophages and NK cells, as well as their plasticity and function have been investigated here in tissues of a non-human primate model of SARS-CoV2 infection. We found profound changes in the NK cell repertoire and macrophage phenotype in convalescent monkeys that indicated imprinting of macrophages and training of NK cells. The convalescent animals separated into two groups based on macrophage phenotype and viral protein levels. SARS-CoV-2 protein was detected up to 18 months post-infection in alveolar macrophages to variable levels. The animals with lower viral proteins displayed strong MHC-E restricted NK cell activity toward cells presenting peptides derived from the LS of the SARS-CoV-2 Spike protein, while NK cell activity from the other convalescent animals and from healthy humans and control animals were strongly inhibited by these Spike LS-derived peptides. The adaptive NK cell activity was not detectable in blood but detected in tissue-resident NK cells. It was speci c for the Spike peptides, and remarkably, was cross-reactive against MERS-CoV and SARS-CoV spike-derived peptides.
One major nding in this study is the frequent detection of viral antigens in lung macrophages after a It is of note, that the MHC-E dependent adaptive activity was cross-reactive against MERS-CoV and SARS-CoV Spike-derived peptides, while the activity against the control peptides did not change. The Spike proteins from distinct Coronaviruses are only weakly homologous at the nucleotide or amino acid level, but they share helix structure and high hydrophobicity in common 86 . This is in agreement with structural analyses demonstrating a broad tolerance for hydrophobic and polar amino acids in the primary pockets of human (HLA-E) and simian (Mamu-E) MHC-E molecules 44 . Thus, one would anticipate that these adaptive NK cells are also reactive against SARS-CoV-2 variants other than the Wuhan strain, including against variants of concern.
In summary, our study revealed an unexpected high frequency of long-term persisting SARS-CoV-2 Spike antigen in macrophages of the lung. It unraveled that some animals were capable to escape MHC-E mediated inhibition of NK cells by SARS-derived peptides and become highly reactive against SARS-CoV-2 Spike LS peptides. This capacity was associated with the induction of regulatory macrophages and was speci cally present in the lung and not detectable in blood. Moreover, the adaptive NK cell activity was cross-reactive against MERS-CoV and SARS-CoV spike-LS derived peptides. Given the high potential of cross-reactivity, it will be interesting to study in the future, if these NK cells contribute to protection in SARS-CoV-2 infection against variants of concern, and if they are inducible by immunotherapies and/or The animals were healthy and seronegative for SIV, type D retrovirus, and simian T-cell lymphotropic virus type 1 at the time of infection and were housed in single cages within level 3 biosafety facilities after infection. At the inclusion in the study the average weight of the monkeys was between 3 and 6 kg. All monkeys were young adults with an average age of 3-5 years at inclusion. Both males and females were used. Sample collection was performed in random order. The investigators were not blinded while the animal handlers were blinded to group allocation.

Viruses
For the in vivo studies, SARS-CoV-2 virus (hCoV-19/France/lDF0372/2020 strain) was isolated by the National Reference Center for Respiratory Viruses (Institut Pasteur) as previously described 31 . Virus stocks used in vivo were produced by two passages on mycoplasma-free Vero E6 cells in Dulbecco's modi ed Eagle's medium (DMEM) without FBS, supplemented with 1% penicillin (10,000 U ml −1 ) and streptomycin (10,000 μg ml −1 ) and 1 μg ml −1 TPCK-trypsin at 37 °C in a humidi ed CO 2 incubator and titrated on Vero E6 cells.

Tissue collections and processing
Whole venous blood was collected in ethylenediaminetetraacetic acid (EDTA) tubes. Peripheral blood mononuclear cells (PBMCs) were isolated by Ficoll density-gradient centrifugation. Cells were either immediately stained for ow cytometry or cryopreserved in 90% foetal bovine serum (FBS) and 10% dimethyl sulfphoxide (DMSO) and stored in liquid nitrogen.
The BAL uids (BALF) were collected by immediate gentle aspiration after each aliquot and pooled in a sterile heparinate lithium container. BALF samples were centrifuged at 350 g for 10 min. The cells were washed with PBS and then mononuclear cells (MC) were separated by standard density-gradient centrifugation. The percentage of BAL uid recovered was approximately 75% of the instilled uid. The alveolar cellularity ranged from 50000 to 200000 cells per ml of uid recovered. No statistically signi cant difference in cellularity was noticed between baseline and post-infection broncho-alveolar lavage (data not shown). To rule out potential side-effects, we have veri ed that repeated bronchoalveolar lavages in uninfected animals do not affect the percentages of cells among BALMC. During the experiment, the majority of BALMC that were recovered were macrophages (more than 80%).

Sars-CoV-2 viral RNA quanti cation in nasopharyngeal and tracheal samples
Upper respiratory (nasopharyngeal and tracheal) specimens were collected with swabs (Universal transport medium, Copan; or Viral Transport Medium, CDC, DSR-052-01). Tracheal swabs were performed by insertion of the swab above the tip of the epiglottis into the upper trachea at approximately 1.5 cm of the epiglottis. All specimens were stored between 2 °C and 8 °C until analysis. Viral copy numbers were determined by quantitative RT-PCR with a plasmid standard concentration range containing an rdrp gene fragment including the RdRp-IP4 RT-PCR target sequence. The limit of detection was estimated to be 460 copies as described previously 30 . The protocol describing the procedure is also available on the WHO website (https://www.who.int/docs/default-source/coronaviruse/real-time-rt-pcr-assays-for-the-detectionof-sars-cov-2-institut-pasteur-paris.pdf?sfvrsn=3662fcb6_2).

Sars-CoV-2 viral RNA quanti cation in BALF cells
Sars-CoV-2 viral RNA quanti cation in BALF cells we performed as follow: Cryopreserved BALF cells were quick-thawed, centrifuged, and washed in 2% BSA solution in D-PBS. Cells were blocked for 5 min in 2% BSA and then incubated at room temperature for 30 min with anti-NKG2a/c Ab (clone z199) to isolate simian NK cells by positive selection using magnetic beads (Miltenyi biotec). The negative fraction, consisting predominantly of macrophages, was then centrifuged and total RNA was extracted using RNeasy mini KIT (Qiagen). Primer and probes for the real-time quantitative reverse transcriptase and polymerase chain reaction (RT-qPCR) were the same as described above and available on the WHO website (https://www.who.int/docs/default-source/coronaviruse/real-time-rt-pcr-assays-for-the-detectionof-sars-cov-2-institut-pasteur-paris.pdf?sfvrsn=3662fcb6_2). Relative quanti cation of the viral genome was performed by one-step RT-qPCR using the RNA to ct 1 step kit (applied biosystem, 4392938).

Thermal cycling was performed in a StepOnePlus Real-Time PCR System (Applied Biosystems) in
MicroAmp Fast Optical 96-well reaction plates. 2 Δ Δ Ct was calculated based on the mean of Ct measured for viral RNA using the nCoV-IP4 primers and 18s RNA Cts obtained for each monkey.

Polychromatic ow cytometry
PBMCs and BALF cells were stained as previously described 87 . The antibodies are listed in

Data availability
The authors declare that all other data supporting the ndings of this study are available within the article and its raw data les, or are available from the authors upon request. Figure 1 Immunophenotyping