Research ArticleCORONAVIRUS

Clonal expansion and activation of tissue-resident memory-like Th17 cells expressing GM-CSF in the lungs of severe COVID-19 patients

See allHide authors and affiliations

Science Immunology  23 Feb 2021:
Vol. 6, Issue 56, eabf6692
DOI: 10.1126/sciimmunol.abf6692

Abstract

Hyperinflammation contributes to lung injury and subsequent acute respiratory distress syndrome (ARDS) with high mortality in patients with severe coronavirus disease 2019 (COVID-19). To understand the underlying mechanisms involved in lung pathology, we investigated the role of the lung-specific immune response. We profiled immune cells in bronchoalveolar lavage fluid and blood collected from COVID-19 patients with severe disease and bacterial pneumonia patients not associated with viral infection. By tracking T cell clones across tissues, we identified clonally expanded tissue-resident memory-like Th17 cells (Trm17 cells) in the lungs even after viral clearance. These Trm17 cells were characterized by a a potentially pathogenic cytokine expression profile of IL17A and CSF2 (GM-CSF). Interactome analysis suggests that Trm17 cells can interact with lung macrophages and cytotoxic CD8+ T cells, which have been associated with disease severity and lung damage. High IL-17A and GM-CSF protein levels in the serum of COVID-19 patients were associated with a more severe clinical course. Collectively, our study suggests that pulmonary Trm17 cells are one potential orchestrator of the hyperinflammation in severe COVID-19.

INTRODUCTION

On 11 March 2020, the World Health Organization communicated that the spread of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) had reached pandemic status. By the end of 2020, there were more than 80 million confirmed cases including 1.7 million deaths (1). These epidemiological data highlight the need to rapidly develop therapies for treating COVID-19 that reduce the high case fatality rate. The promising results of the clinical trial RECOVERY, in which Dexamethasone was administered to 2,104 COVID-19 patients (2), suggests that one of the causes of the acute respiratory distress syndrome (ARDS) and ultimately death of COVID-19 patients is the hyperactivation of the immune system. Supporting the pathogenic role of immune hyperactivation, the use of neutralizing antibodies, blocking for example GM-CSF and IL-1β, has shown encouraging clinical results (36). The efficacy of a therapy blocking IL-6 has not yet been broadly recognized (7), but one recent study showed that Toclizumab reduced disease progression in COVID-19 patients not receiving mechanical ventilation (8).

Considering that peripheral blood myeloid cells appear not to be able to produce high amounts of proinflammatory cytokines (9) and the numbers of blood T cells are reduced in COVID-19 patients (10, 11), the lungs may serve as a reservoir for cells producing these cytokines. However, additional investigation into the lung-specific cellular source of the proinflammatory cytokines typical of severe COVID-19, including IL-6, TNF-α, IL-1β and IL-17A, is needed. Using single cell RNA sequencing methods (scRNA-Seq), it has indeed been shown that pro-inflammatory macrophages expressing IL6, IL1B, and TNF and CD8+ T cells with a tissue-resident cytotoxic signature, are present in the bronchoalveolar lavage (BAL) and upper respiratory tract of COVID-19 patients (12, 13). The accumulation of IFN-γ producing CD4+ T cells in the BAL of COVID-19 patients has also recently been described (14). Still, the role of CD4+ T cells and of their different polarization states, especially at the site of infection, need to be further elucidated.

CD4+ T cells orchestrate the immune response for example, by impacting on macrophage function and activation of cytotoxic CD8+ T cells. In order to mediate these different functions, naïve CD4+ T cells differentiate into effector cells characterized by different polarization states such as Th1 and Th17. We have shown that Th17 cells can infiltrate the lungs and acquire a tissue-resident phenotype during bacterial infection (15). Upon infectious stimuli, these tissue-resident long-lived cells can reacquire the original cytokine profile, for example IL-17A/F, or switch toward the production of IFN-γ, the signature cytokine of the Th1 polarization state. While these long-lived tissue-resident Th17, referred to here as Trm17 cells, usually exert a protective function (15), we have also recently shown that these cells can contribute to immune-mediated inflammatory diseases (16). Whether Trm17 cells are present in the lungs of COVID-19 patients and how they interact with other potentially pathogenic immune cells of these patients, remains to be studied.

Here we identified two populations of Th17 cells in the BAL fluid of COVID-19 patients. One of these was mainly resident in the lung, characterized by the expression of GM-CSF and shared clones with Th1 cells. Moreover, we provide a lung-specific immune cell-cell interaction map showing the potential role of Trm17 cells in support of the already known pathological immune players, such as pro-inflammatory and pro-fibrotic macrophages, and cytotoxic CD8+ T cells (12, 13). These data provide support for continuing to test anti-cytokine therapies including those that have undergone preliminary clinical testing, for example GM-CSF neutralization (22, 23), or those currently under consideration, for example anti-IL-17A/F treatment (24).

RESULTS

Immune profile of T cells and myeloid cells in the blood and BAL fluid of COVID-19 patients

In order to provide a detailed analysis of the lung-specific and peripheral immune responses in COVID-19, bronchoalveolar lavage (BAL) fluid and peripheral blood was taken from 9 patients with severe COVID-19. The major goal of the study was to analyze the tissue-specific immune reponse in COVID-19 patients, with a particular focus on T cells. In addition, we also included BAL fluid and PBMCs of five patients with bacterial pneumonia, not associated with viral infection. All samples were analyzed by flow cytometry and T cells were FACS-sorted and subjected to single-cell RNA and TCR sequencing, as well as to sequencing-based epitope measurement (cellular indexing of transcriptomes and epitopes by sequencing, CITE-seq). From BAL samples, CD3- non-T cells, including mainly myeloid cells, were also analyzed by scRNA-seq and CITE-seq (56,735 cells from the BAL and 77,457 cells from the peripheral blood) (Fig. 1A and Fig. S1A,B).

Fig. 1

Immune landscape of severe COVID-19 and bacterial pneumonia. (A) Schematic representation of experimental setup. (B) Overview of baseline characteristics and clinical course of COVID-19 patients and patients with bacterial pneumonia. (C) Virus titers measured by qPCR from bronchoalveolar lavage fluid (BAL fluid), tracheal fluid and peripheral blood at time of sampling. (D) UMAP dimensionality reduction embedding of all cells from BAL fluid (n=56,735 cells, n=8 for COVID-19 and n=4 for bacterial pneumonia, samples of patients S6 and B1 were excluded for technical reasons) colored according to cell type assessed by gene expression and (E) epitope measurement using CITE-seq of key markers (scale bars indicate normalized expression). (F) Single cell analysis of CD3+ T cells from peripheral blood of all patients (n=77,457 cells, n=7 for COVID-19 and n=4 for bacterial pneumonia). (G) CITE-seq information of cluster-defining epitopes (scale bars indicate normalized expression). (H) Flow cytometry of peripheral blood and BAL fluid of COVID-19 (n=8) and bacterial pneumonia (n=5) patients. Per patient, an equal number of viable CD45+ cells were exported for analysis and concatenated together before calculating the UMAP (total cells in peripheral blood = 129,141; in BAL fluid = 114,927). Cell types were defined according to cell surface expression profiles by manual gating. Patient S9 was excluded from the statistical analysis due to low cell numbers. (I) Comparison of cell frequencies as measured by flow cytometry of cells from COVID-19 and bacterial pneumonia patients. (J) Hematoxylin and eosin staining (H&E) and (K) CD3 staining of lung autopsy tissue of one representative of 7 patients.

All patients were treated on the intensive care unit at the University Medical Center Hamburg-Eppendorf. Eight out of 9 COVID-19 patients and all patients with bacterial pneumonia were on mechanical ventilation at time of sampling, indicating the severity of disease, further reflected by high mortality in both groups (Fig. 1B). Detailed patient characteristics of SARS-CoV-2 positive and bacterial pneumonia patients, including comorbidities and relevant medications, are presented in Table S1. In 8 out of 9 COVID-19 patients, symptomatic SARS-CoV-2 infection was diagnosed >2 weeks before BAL sampling, while in the remaining patients, SARS-CoV-2 was detected 8 days before sampling (Fig. 1B). In line with this, viral titers in BAL fluid, tracheal secretion fluid and blood of the majority of patients were very low or negative at time of sampling (Fig. 1C).

In order to examine the immune profile in the lungs of COVID-19 patients and at the same time, achieve robust clustering, we integrated the scRNA-seq datasets of COVID-19 and bacterial pneumonia patients. The analysis revealed the presence of five main clusters in the BAL fluid based on key signature genes and standard surface markers (T cells, B cells, mast cells, myeloid cells, and epithelial cells) represented in the UMAP dimension reduction (Fig. 1D). The T cell cluster could be further subdivided into CD4+ T cells CD8+ T cells, and innate-like T cells / innate lymphocytes (NKG7, TRAV1-2). Also, the myeloid cluster consisted of several subsets that were identified as macrophages (CD68), neutrophils (FCGR3B) and DCs (HLA-DQA1) (Fig. 1D and E, Fig. S1C). Of note, all subsets identified were present in each sample of the two patient groups (Fig. S1D and E). We also combined RNA-, TCR-, and CITE-seq of peripheral blood T cells from COVID-19 and bacterial pneumonia patients and yielded a sufficient number of CD4+, CD8+, and innate-like T cells for further clonal and expression analyses, allowing for a detailed comparison of peripheral and lung-specific T cell responses (Fig. 1F and G, Fig. S2).

Next, we quantified the respective lymphoid and myeloid cell subsets in BAL fluid and PBMC of COVID-19 and bacterial pneumonia patients using flow cytometry data (Fig. 1H and Fig. S3). While in peripheral blood there was not any obvious difference in the frequencies of cells analyzed comparing COVID-19 and bacterial pneumonia patients, T cells subsets and NK cells showed significantly increased frequencies in the BAL fluid of COVID-19 patients compared to bacterial pneumonia (Fig. 1I). We confirmed the presence of T cells and mononuclear cells in the lung parenchyma of COVID-19 patients by hematoxylin and eosin staining (Fig. 1J) and by CD3 staining (Fig. 1K) on lung autopsy tissue. Interestingly, an accumulation of T cells was predominantly identified in the perivascular space (Fig. S4). Taken together, these data show an accumulation of T cells in the lungs of COVID-19 patients.

Tracking T cell clonality in blood and BAL identifies tissue-specific Trm17 cells

Considering the observation of the perivascular accumulation of T cells in the lungs of COVID-19 patients, we decided to further investigate the tissue-specific T cell response. We therefore integrated the blood and BAL fluid T cell data sets to examine tissue-specific clonal expansion and activation of T cells. We examined T cell activation by analyzing the concomitant expression of a chosen pool of pro-inflammatory cytokines, i.e., IL2, TNF, IL17A, IL17F, IFNG, and IL-22. Clonal expansion was analyzed by quantifying similar T cell clones based on the TCR-sequence information. We observed that both CD4+ and CD8+ T cells mainly underwent clonal expansion and activation predominantly in the BAL fluid in both groups of patients (Fig. 2A). Clonal expansion does not necessarily reflect Ag specificity, therefore we next investigated whether SARS-CoV-2-specific T cell clones are present in the BAL fluid by comparing the TCRs identified in our study with those of two publicly available datasets of SARS-CoV-2-specific TCR sequences obtained from peripheral blood (25, 26). The frequency of shared clones was higher in the COVID-19 cohort compared with the patients with bacterial pneumonia (Fig. S6A). In addition, we observed higher frequencies of shared clones in T cells from the BAL fluid as compared to peripheral blood.

Fig. 2

T cell clonality in pulmonary inflammation. (A) Blood-lung activation map of T cells from blood and BAL fluid of all patients: UMAP dimensionality reduction embedding of T cells (left); Clone size proportion (clone count divided by number of cells per sample) of T cells (middle) and the cytokine secretion score of T cells (right) from COVID-19 and bacterial pneumonia as indicated. (B) UMAP presentation of T cells from BAL fluid of all patients. Clusters were annotated according to gene expression and epitopes measurement of key markers. (C) Ratio of clonal expansion of bacterial pneumonia versus COVID-19 for the major expanded BAL fluid T cell clusters. (D) Sub-clustering analysis of clonally expanded CD4+ T cells of all patients. Clusters were annotated according to gene expression presented in the heat map. (E) Volcano plot showing differential gene expression between Th17 cluster1 and 2 of all patients. Genes were considered significant with adjust p < 0.05. Non-significant genes are shown in black. (F) Heat map of selected pathogenic gene markers of Th17 cells of all patients in comparison with other T cell clusters. (G) Clone size proportion of T cells in peripheral blood and BAL fluid of COVID-19 patients and presentation of high abundant clones (clone size > 5) that are shared between BAL fluid and blood and BAL-specific clones as indicated. (H) CD4 migration and tissue residency score of Th17 cluster1 (Trm17) and 2 (Tem17) from all patients. (I) Possible model of intraclonal diversification of CD4+ T cell subsets (left); Distribution of 2 representative BAL fluid clones from a patient with COVID-19 (patient S1 clone239 and clone218) on the UMAP (middle and right). (J) Barplot of top expanded BAL fluid clones containing Trm17 cells from COVID-19 patients. COVID-19: n=8 for BAL fluid and n= 7 for blood; bacterial pneumonia: n=4 for BAL fluid and n=4 for blood.

Considering this tissue-specific activation, we examined BAL fluid T cells with more granularity. By combining CITE-seq and differentially expressed genes (DEG) we identified five major CD4+ T cells clusters, including a population of Foxp3+ Treg cells and two CD8+ T cells clusters (Fig. 2B and Fig. S5). We also found three clusters composed of both CD4+ and CD8+ T cells and characterized by heat shock proteins, genes associated with proliferation (MKI67, STMN1) and expression of long non-coding RNAs (MALAT1, NEAT1). Finally, we identified a distinct cluster formed by MAIT cells (TCRVa7.2/TRAV1-2) and one by innate and innate like T cells (CD56, NKG7).

We then quantified the clonal expansion for each population (Fig. S6B and C). We observed a high clonal expansion in two CD4+ T cell clusters (i.e., CD4+ T effector memory (Tem) cells, cluster 1 and 2), in the two main CD8+ T cell populations, in MAIT cells and in the other innate like T cell cluster. We next wondered if these cells were also expanded in a different type of infection, by comparing the clone size proportion of the above indicated populations between the two patient groups (Fig. 2C). We found that the CD4+ Tem cell cluster 2 was most selectively expanded in COVID-19 patients compared to bacterial pneumonia. Therefore, we decided to further analyze this cluster and as controls, we used the other CD4+ Tem cell cluster (i.e., cluster 1), MAIT and Foxp3+ Treg cells (Fig. 2D and Fig. S7). We observed that the original cluster 2 was enriched for genes typical of Th17 polarization states, while the original cluster 1 primarily contained CD4+ Tem cells expressing genes associated with a Th1 polarization state. We then tested which of these clusters were selectively expanded in COVID-19 patients and found that while Th1 cells are expanded in all patients, both Th17 clusters were only expanded in COVID-19 patients (Fig. S7E and F).

Differential expression analysis revealed that while both Th17 clusters express similar levels of RORC and CCR6, Th17 cell cluster 1 is enriched for genes associated with cytotoxicity (SRGN, GZMB, GNLY) and for genes translating for pro-inflammatory cytokines (IL21, IL17F, IFNG, CSF2/GMCSF) and chemokines (CCL3, CCL4 and CCL5) (Fig. 2E, Fig. S7G). We also observed that this cluster has high expression of the transcriptional factor RBPJ which has been shown to be fundamental for the pathogenicity of Th17 cells in an experimental autoimmune encephalomyelitis mouse model (27). We next compared the expression of some of these DEG, in addition to other genes associated with Th17 cell pathogenicity, among all the sub clusters of the CD4+ T cells isolated from the BAL fluid. We found that among all, CSF2 (GMCSF) and IL21 were the most selective genes expressed by the cluster of the potentially pathogenic Th17 cells (Fig. 2F).

We have recently shown that Th17 cells can acquire a tissue-resident phenotype in the lungs (15). We therefore used TCR sequences as markers to test whether this population of Th17 cells are mainly found in the BAL fluid and not in the circulation, suggesting possible resident behavior. We found that virtually none of the highly expanded Th17 cluster 1 cells in the lungs shared clones with T cells in the blood, supporting that these cells are resident in the lung. In contrast, the other Th17 cell cluster (i.e., cluster 2) and the Th1 cell populations are composed of a mixture of resident and circulatory clones (Fig. 2G). Then we used a literature-based residency and migratory scores and found that the potentially pathogenic Th17 cell cluster 1 has on average a lower migratory and higher residency score compared to the other effector Th17 cell cluster 2 (Fig. 2H). These data suggest that the Th17 cell cluster 1 is enriched for pathogenic and resident cells compared to the Th17 cell cluster 2. To simplify the classification of these clusters and reflect their features, we named cluster 1 as tissue-resident memory-like Th17 cells (Trm17) and cluster 2 as effector memory Th17 cells (Tem17). Of note, we observed that the TCR-sequences shown to be specific for SARS-CoV-2 in other studies (25, 26) can also be identified in the cluster of Trm17 cells (Fig. S7H). Since it has been shown that Trm17 cells still retain a certain degree of plasticity, especially toward Th1 cells (28), we wondered if this was the case in COVID-19. We used the TCR sequences as natural lineage barcodes to follow the origin/fate of some of the Trm17 cell clones expanded in COVID-19 patients. We observed that sister clones of the Trm17 cells were also found to express other T cell phenotypes, such as the Th1 phenotype (Fig. 2I and J). This intraclonal diversification (29) suggests that some of the Trm17 cells have a dynamic developmental trajectory in common with other types of tissue-specific CD4+ T cell populations, in particular with Th1 cells which, as expected, are the dominant expanded clones in term of quantity. In summary, we identified lung-specific Trm17 cells in the BAL fluid of COVID-19 patients based on cytokine expression profiles and clonal expansion.

Different types of myeloid cells identified in the BAL fluid of COVID-19

Next, we set out to examine the different populations of myeloid cells in BAL fluid that were identified in our scRNA-seq analysis (Fig. 1D). As above, to achieve robust clustering, we included myeloid cells from all patients (COVID-19 and bacterial pneumonia) in this analysis. Subclustering of macrophages and neutrophils revealed heterogeneity of macrophage polarization status and stages in neutrophil maturation (Fig. 3A). In particular, alveolar macrophages were defined by the gene expression of class A scavenger receptor MARCO, the mannose receptor MRC1, the intracellular lipid transporter FABP5. They also express high levels of the pro-fibrotic gene SPP1 (12) (Fig. 3B and C). We also detected high levels of TREM2, a surface receptor able to prevent macrophage apoptosis upon viral replication (30). Next, we defined pro-inflammatory macrophages as cells expressing high levels of CCL2 and CCL3, chemokines involved in recruitment of adaptive and innate cells to sites of infection, and which were also characterized by the expression IL6, IL1B and TNF (Fig. 3A and C). Blood-derived macrophages have been defined on the basis of high expression of CD14. Their pro-inflammatory signature is mirrored by the high expression of FCN1, as previously described in COVID-19 patients (12), by the high expression of CD302, a C-Type lectin receptor induced in vitro upon LPS stimulation, and by the expression of alarmins such as S100A8 and S100A12, calcium binding proteins and DAMPs, whose expression is regulated by pro-inflammatory molecules such as IFN-γ and TNF-α, and that can lead to the secretion of IL-6 and IL-8 (Fig. 3B and Fig. S8A).

Fig. 3

Landscape of myeloid cells in the lung. (A) UMAP dimensionality reduction embedding of myeloid cells from BAL fluid of all patients from our study (COVID-19 n=8 and bacterial pneumonia n=4). (B) Heat map of key marker gene expression of the indicated clusters. (C) UMAP plots showing expression of genes mirroring key features of macrophage polarization and function (scale bars indicate normalized expression).

We defined a population of cells with a tissue-remodeling signature as pro-fibrotic macrophages, which under a persistant inflammatory trigger, might acquire a pro-fibrotic function (21). These pro-fibrotic macrophages expressed higher levels of APOE, TGFBI, TMEM176A and CD86 and were enriched in complement components (C1QB, C1QA, C1QC) (Fig. 3B and C and Fig. S8A), in line with what was previously described for pro-fibrotic macrophages in the context of SARS-CoV-2 infection (12). This macrophage cluster showed a pro-fibrotic signature mostly similar to the alveolar macrophage subcluster, potentially indicating that the two macrophage populations have a similar biological function. Interestingly, however, the pro-fibrotic macrophages were also characterized by the expression of high levels of FCGR3A (CD16) and intermediate/low levels of CD14, together with the expression of genes associated with antigen presentation (Fig. S8A), therefore suggesting them as a subcluster of cells potentially originating from a non-classical/intermediate monocyte population. Importantly, pro-fibrotic macrophages also expressed AXL, a receptor tyrosine kinase which is required for resolution of lung inflammatory disease upon viral infection, induced by GM-CSF and associated with development of tissue fibrosis in mouse models (31, 32). A population highly enriched in heat shock protein (HSPA6, HSPA1A, HSPH1, DNAJB1, HSPA1B) was also observed and named as HSP+ macrophages.

Finally, expression of CXCR2 and CXCR4, which define neutrophil maturation stages and regulate their trafficking from bone marrow, was detected in the two neutrophil clusters identified in COVID-19 patients (Aged and Non-aged neutrophils) (Fig. 3A and B). Of note, all subsets described were reproducibly found in both COVID-19 and bacterial pneumonia patients (Fig. S8B and C). The identification of myeloid cell populations in our dataset provided us with a foundation for investigating the interactions between immune cells in the lungs of COVID-19 patients.

Trm17 cell interactome with other pathological cell types in COVID-19

Once the landscape of the myeloid and lymphoid compartment was clarified, we investigated the cell-cell interactions of Trm17 cells with the other immune cells, in particular myeloid and cytotoxic CD8+ T cells, which are known to correlate with lung damage in COVID-19 patients (12, 13).

We constructed the interactome of all immune cells found in the BAL fluid of COVID-19 patients using the T cell and the myeloid cell subclusters defined in Fig. 2 and 3, respectively. Then, we performed a network analysis which is based on transcriptomic levels of ligand-receptor interactions between cell types. We identified different clusters of macrophages (pro-inflammatory, pro-fibrotic and alveolar macrophages) at the center of our communication network, having the highest number of different ligand-receptor pairs with other cell populations (Fig. 4A). We also observed that the cells of the main T cell populations (Th1, Trm17, Tem17, FOXP3 Treg, Tr1, MAIT) interact more with tissue macrophages than with other myeloid cell clusters, such as blood derived-macrophages, DCs or neutrophils (Fig. 4B). Among the T cells, Trm17, Tem17 and Th1 appear to have more ligand/receptor interactions with macrophages than the other T-cell clusters do (Fig. 4A and B). To further explore the interactions between lung Trm17 that we previously found to be clonally expanded and macrophages, we selected the 10 most specific ligand-receptor pairs of both populations based on rank calculated using CellPhoneDB (33). To calculate the connection strength of these interactions, we multiplied the average ligand expression with the proportion of cells expressing the receptor from the respective clusters. We found that CD40LG/CD40, LTA/LTBR and GM-CSF/GM-CSFR had the highest connection strength between Trm17 and pro-fibrotic macrophages (Fig. 4C). In addition, we identified CD40LG/CD40, LTA-LTBR and CSF2-CSF1R to demonstrate the most selective and strongest interactions of Trm17 with pro-inflammatory macrophages (Fig. S9A).

Fig. 4

Interactome of T cells and myeloid cells in the lungs of COVID-19 patients. (A) Interaction network of all BAL fluid clusters based on the number of ligand-receptor-interaction (>30 edges) based on Fruchterman-Reingold force-directed algorithm from COVID-19 patients (n=8). (B) Adjacency map of T cell-myeloid cell interactions. (C) Ligand and receptor interaction strength ([mean ligand expression] multiplied by [proportion of receptor expression per cluster]) of Trm17 cells (ligands) and pro-fibrotic macrophages (receptors). Interactions were filtered for cytokines and for specificity based on rank scoring. (D) Supervised interaction map of potential key players in sustaining lung inflammation in COVID-19 patients. Line width correlates with interaction strength. (E) Pathway-analysis of CD40L (CD40LG), LTA and GM-SCF (CSF2) signaling in Pro-inflammatory and Pro-fibrotic macrophages indicating the log2-fold change in COVID-19 versus bacterial pneumonia. (F) Cytotoxic module scores in all clusters which include CD8+ cells using pro-inflammatory and cytotoxic mediator genes in CTL from Chua et al., Nat Biotech 2020. (G) Module scores in the indicated clusters using the highest 50 differential expressed genes of CD8+ T cells receiving help or no help from CD4+ cells, respectively, according to Ahrends et al., Nat Comm 2019. (H-I) Ligands and receptor interaction strength (H) between Trm17 and CD8+ Tem CTL Cluster 2 and (I) between Trm17 and DC. Interactions were filtered according to their rank score. (J) Supervised interaction map of Trm17, CD8+ Tem CTL and DCs with annotated ligands. Line width correlates with interaction strength.

To understand how Trm17 cells directly interact with the lung epithelium, we analyzed the interactions of Trm17 cells with epithelial cells. Here, we identified the cytokines IL-26, IL-17A and IL-17F to be among the top ten interactions (Fig. S9B). IL-26 interacts with the IL-26R on epithelial cells (Fig. S9C), potentially playing a role in antiviral response (34). Due to the low number of epithelial cells in our dataset, we confirmed the expression of IL-26R, a heterodimer consisting of IL10RB and IL20RA, using the transcriptomic data on epithelial cells recently published in the context of SARS-CoV-2 infection (13) (Fig. S9D).

Then, focusing on the most relevant interactions based on rank and connection strength between Trm17 cells, pro-fibrotic macrophages, pro-inflammatory macrophages and epithelial cells, we constructed a smaller curated interaction map which depicts that Trm17 can act on both pro-fibrotic and pro-inflammatory macrophages as well as epithelial cells (Fig. 4D). In return, pro-inflammatory macrophages, by secreting IL-1β, may act on Trm17 cells (Fig. 4D, Fig. S9E) and additionally through secretion of various chemokines like CCL2, CCL3 and CCL20 target the respective chemokine receptors on Trm17 cells (Fig. S9E).

Next, we aimed to gain additional insight into intracellular signaling induced by CD40L, LTA, and GM-SCF in pro-inflammatory and pro-fibrotic macrophages. For this, we used KEGG pathways and annotated the DEG to the respective pathway by color coding genes up-regulated in macrophages from COVID-19 patients versus bacterial pneumonia. In pro-inflammatory macrophages, IL-1β, CXCL1, and CXCL8 might be produced because of signals transmitted by CD40, LTBR, and NFKB1. A similar signaling cascade could be induced in pro-fibrotic macrophages. Furthermore, GM-CSF was associated with a pathway capable of triggering survival and proliferation signals in pro-fibrotic macrophages (Fig. 4E).

Finally, since it is known that CD4+ T cells are necessary for regulating the magnitude and quality of the cytotoxic CD8+ T cell response, we investigated the molecular mechanisms by which Trm17 might regulate the CD8+ T cell cytotoxic response in COVID-19 patients. Of note, the cytotoxic CD8+ T cell response has been proposed to mediate lung tissue damage in these patients (12, 13). To identify highly cytotoxic CD8+ T cell clusters among the ones found (Fig. 2B), we created a cytotoxic scoring using pro-inflammatory genes which are expressed by CTL in critical COVID-19 patients as described in Chua et al. (13) and applied this dataset to the CD8+ clusters we identified in our analysis. The highest cytotoxic signature was observed in CD8+ Tem cluster 2 and CD8+ Tem cluster 1 (Fig. 4F). We then calculated a second scoring to investigate whether CD8+ T cell clusters from the BAL fluid of COVID-19 patients might receive help from CD4+ T cells. To this end, we used the top 50 DEG identified by Ahrends et al. as characteristic of Tem cells receiving help from CD4+ T cells or not (35) and applied this information to the different CD8+ T cell clusters from our dataset. We identified CD8+ Tem cluster 2 to display the highest help module score (Fig. 4G). Next, to gain insight on how Trm17 cells and CD8+ Tem affect each other, we determined the most specific and strongest interactions according to rank and connection strength. We identified CD70-CD27 and CCL5-CCR4 as the pathways highly engaged in this cell-to-cell interaction (Fig. 4H). Since CD4+-CD8+ T cell interaction occurs in a spatiotemporally organized interaction with DCs (36), we further dissected ligand-receptors interaction between Trm17 with DCs and CTL CD8+ T cells with DCs (Fig. 4I, Fig. S9F). Trm17 had the potential to affect DCs via CD40L, FTL3LG and GM-CSF (Fig. 4l). On the basis of all these data, we explored how the three cell populations could be connected, and observed potential connections among Trm17, CD8+ CTL and DCs (Fig. 4J).

In short, these data show the potential interaction of Trm17 cells and other tissue-specific immune cells, namely macrophages and CTL CD8+, which have been associated with disease severity of COVID-19.

The cellular map of GM-CSF-expressing cells

To test if GM-CSF and IL-17A correlate with the severity of the COVID-19, we first measured these two cytokines in the serum of patients with COVID-19 and of healthy blood donors and observed increased GM-CSF concentration in the patients (Fig. 5A). We excluded patient S9 from the Hamburg cohort, since this patient received intravenous cytokine treatment. Second, we analyzed a different cohort, obtained from the University of Halle, which included moderate and severe COVID-19 patients (37). We observed that GM-CSF and IL-17A appear to differentiate moderate vs. severe disease (Fig. 5A).

Fig. 5

Cytokine secretion profile and cellular source of GM-CSF. (A) GM-CSF and IL-17A protein in serum of patients with COVD-19 (n=8) and healthy controls (n=7) from Hamburg and of patients with moderate (n=8) or severe COVID-19 (n=11) from Halle as indicated. Cell map of (B) CSF2 (GM-CSF) expressing and (C) IL17A expressing cells (scale bars indicate normalized expression). Three different UMAPs with different cellular granularity showing the respective gene expression of in total cells of the BAL fluid (left panels), total T cells of blood and BAL fluid (middle panels), and total T cells in BAL fluid (right panels) from all patients. (D) Immunofluorescence of CD4+ (green) CCR6+ (red) Trm17 cells in the lungs of a deceased patient with COVID-19 infection (nuclear staining DAPI, blue) (two additional samples are presented in Fig. S10B). (E) Combined immunofluorescence (CCR6) and fluorescence in situ hypbridization (IL17A) of lung samples from one patient with COVID-19. (F) Concentrations of the indicated cytokines in the BAL fluid of COVID-19 and bacterial pneumonia patients.

In order to examine potential cellular sources of GM-CSF and IL-17A, we mapped CSF2 (GMCSF)- and IL17A-expressing cells on three UMAPs with different granularity and tissues (Fig. 5B and C): all cells in the BAL fluid (left), all T cells in BAL fluid and peripheral blood (center), and CD4+ T cells in the BAL fluid (right) and displayed gene expression in a violin/dot plot (overview, lung and blood; Fig. S10A). These data show that gene expression of both cytokines is mainly restricted to the Trm17 cell subset from BAL fluid.

Finally, to further support the presence of Trm17 cells in the lungs of patients with COVID-19 we first performed immunofluorescence staining showing the presence of CD4+ cells expressing CCR6 in the perivascular infiltrate of the lungs (Fig. 5D and Fig. S10B). Second, in a proof-of-principle analysis, the combination of immunofluorescence and fluorescence in situ hybridization of CCR6 and IL17A, respectively, shows the presence of CCR6 and IL-17A coexpressing cells in COVID-19 (Fig. 5E, Fig. S10C). Third, we found high concentrations of GM-CSF and IL-17A as well as IFN-γ and IL-6 in the BAL fluid of COVID-19 patients (Fig. 5F, Fig. S10D). These findings show that Trm17 cells are one potential source of the cytokines GM-CSF and IL-17A, which are prototypical of the hyperinflammation and are indeed present locally within the lungs and in circulation in severe COVID-19.

DISCUSSION

Here we report a comprehensive single-cell transcriptional and T cell receptor (TCR) landscape of CD4+ T cells collected from the BAL fluid and the peripheral blood of patients with severe COVID-19. We observed clonal expansion and characterized the activation profile of tissue-resident memory-like CD4+ T cells in the lungs of COVID-19 patients that persist even after clearance of the virus. These cells express high amounts of the genes encoding the pro-inflammatory cytokines IL-17A/F and GM-CSF. Cell-cell interactome analysis uncovered a pathogenic network in the lung, involving GM-CSF-expressing Trm17 cells, IL-1β-expressing macrophages with a pro-inflammatory phenotype, macrophages expressing the GM-CSF receptor and genes associated with fibrosis, and cytotoxic CD8+ T cells. The relevance of our findings is further supported by the fact that serum protein levels of GM-CSF and IL-17A were elevated in a cohort of COVID-19 patients with severe disease.

It has been speculated that reduction of T cells observed in the peripheral blood of COVID-19 patients might be due to the recruitment of T cells to inflamed tissues (40). In support of this hypothesis, our flow cytometry data revealed an increased frequency of T cells in the lungs of these patients compared to patients with bacterial pneumonia, while confirming reduced peripheral lymphocyte numbers. Moreover, activation status and cytokine expression were higher in lung T cells compared to peripheral T cells. By analyzing TCR clonality, we found a robust expansion of CD8+ T cells in the lungs of COVID-19 patients, as demonstrated by previous studies (12). However, the comparison to patients with bacterial pneumonia showed that this clonal expansion of CD8+ T cell subsets was a general hallmark of mild to severe lung inflammation. In contrast to this, CD4+ T cells and in particular, those displaying a Th17 polarization state mainly expanded in the BAL fluid of COVID-19 patients. One of these Th17 clusters expressed high levels of cytokines that have previously been associated with pathogenic activation of the immune system, such as GM-CSF (41, 42) and IL-17A (43), as well as other known markers of Th17 cell pathogenicity, such as the transcriptional factor RBPJ (27).

Moreover, using TCR clonality analysis across compartments, we showed that clonally expanded cells in this Th17 cell cluster were almost exclusively present in BAL fluid samples, but not in the peripheral blood. These data were supported by an enrichment of genes usually expressed by resident T cells and together provided evidence that these cells indeed represent tissue-resident lung Th17 cells which are probably responsive to SARS-CoV-2 related antigens. In a previous study, we identified and characterized Trm17 cells in the lungs of mice and showed that they play a critical role in protecting from experimental Klebsiella pneumonia infection (15). More recently, we found a correlation of kidney Trm17 cells and severity of immune-mediated kidney disease. Then, using mouse models, we demonstrated that Trm17 cells persist in the tissue after bacterial infection and can rapidly respond to inflammatory stimuli, such as IL-1β by producing IL-17A which ultimately aggravates immune-mediated tissue injury (16). These two studies suggested that Trm17 cells can orchestrate a protective function against extracellular pathogens or fungi, but they can also participate in tissue damage if overactivated by inflammatory stimuli, such as IL-1β. In particular, since the Th17/IL-17 axis has not in fact been linked to protective antiviral immunity, we propose that overproduction of IL-17A and GM-CSF by overactivated Trm17 cells is a feature of severe COVID-19 that might be involved in the immunopathology. However, our data do not rule out the possibility that Trm17 cells could provide a certain degree of protection at an early phase of the infection or in asymptomatic patients. Preclinical animal models of SARS-CoV-2 infection in which it would be possible block the activity of Trm17-derived cytokines are needed to fully address their role at different time points after the infection.

In order to provide a detailed view of how the different immune cell populations interact in COVID-19, we performed unbiased interactome analysis. Here, Trm17 cells were among the T cell subsets showing the strongest interaction with different myeloid cell subsets and CD8+ T cells. By ranking the interactions of Trm17 cells with myeloid cell subsets according to specificity and interaction strength, we identified GM-CSF, CD40L and Lymphotoxin-α as the most important effector pathways employed by these cells to induce pro-inflammatory cytokine and chemokine production such as IL-1β, CXCL1 and CXCL8 in macrophages. Release of IL-1β by pro-inflammatory macrophages, in turn, could signal back to Trm17 cells to increase their pathogenicity (16, 44). One of the most evident features of Trm17 cells was the expression of GM-CSF (see Figs. 2F and 5B) and the interaction with its receptor was among the top hits in the unsupervised interactome analysis of Trm17 cells with myeloid cell subsets. Indeed, T cell-derived GM-CSF can result in activation and differentiation of myeloid cells (45). GM-CSF has further been shown to promote inflammatory tissue damage in a mouse model of Kawasaki disease, which is characterized by hyperinflammation that may share some features with severe COVID-19 (46). Of note, enhanced frequencies of GM-CSF / IFN-γ co-producing T cells have been found in the blood of patients with COVID-19 and seemed to correlate with disease activity (47). Our data indicate that CSF2/GM-CSF-expressing cells are found in the lungs and co-express IL17A. These data, in addition to the presence of CD4+CCR6+ T cells in the lung tissue as well as GM-CSF and IL-17A in the BAL fluid, provide clinical evidence that Trm17 cell-associated cytokines are indeed present in severe COVID-19 patients.

The major conclusions of this study derive from the comparison between cells taken from the blood and BAL fluid of the COVID-19 patients. Nevertheless, in comparing COVID-19 patients with bacterial pneumonia patients, we observed that CD4+ T cells and in particular Trm17 cells were more clonally expanded in the BAL fluid of the virally infected group. This comparison, however, also poses a key limitation of our study since we were unable to conclude whether the clonal expansion of Trm17 cells is specific to COVID-19 patients or a common feature of severe viral infection. To address this point we would have benefitted from having the BAL fluids from other viral infections such as influenza, in which the type of immune cells engaged are overall similar to a SARS-CoV-2 infection. A secondary limitation is our use of the term Trm17 cell, which was used based on the expression profile and the reduced shared clonality between lung and blood of this population. However, the conclusion that these cells indeed reside in the lungs is not definitive, as determination of tissue residency in human tissues remains challenging. Another limitation is the limited sample size of our study and therefore the results of this study need to be further validated in a larger cohort of patients in which Trm17 cell associations with diseases severity are also examined. Indeed, almost all COVID-19 patients had severe COVID-19 according to the WHO classification in our study.

Finally, on the basis of our data, we propose a model in which Trm17 cells are activated or reactivated as part of an ongoing cytokine storm, during which they can start producing pro-inflammatory cytokines such as GM-CSF. This could lead to further activation of macrophages and CD8+ T cells, that others have linked to the severity of the disease (12), and finally mediate lethal lung damage (42, 45). Two small pilot studies have indicated that targeting GM-CSF in patients with severe COVID-19 lung diseases using anti-GM-CSF receptor monoclonal antibodies mavrilimumab or lenzilumab, respectively may be a strategy for improving clinical outcomes (3, 4), although larger controlled clinical trials would be needed to determine efficacy and biological impact of such approaches. This network of tissue-resident cells may persist in the lungs even after the initiating event, e.g., viral infection, has been cleared, contributing to chronic lung pathology. In conclusion, our study provides a snapshot analysis of CD4+ T cells in the lungs of patients with severe COVID-19 and identifies Trm17 cells as one of the components of the lung-specific immune response. In addition, our data provide a rationale for investigating therapeutic approaches targeting Trm17 cells and the GM-CSF network in the search for urgently needed therapies for treating COVID-19 pneumonia.

MATERIALS AND METHODS

Study design

Patients with SARS-CoV-2 infection can develop a severe COVID-19 course with pulmonary involvement and high mortality. Since the adaptive immune system may play a major role in COVID-19 pathogenesis, we sought to investigate the immune response in the lungs of these patients by scRNA-seq with a focus on T cells and their cytokines. To this end, we planned simultaneous gene expression, TCR repertoire sequencing and cell surface protein analyses. Since this is only possible from live cells, we obtained BAL fluid from the lungs of COVID-19 patients, and from patients with bacterial pneumonia which served as a control. We included 9 patients with COVID-19 and 5 patients with bacterial pneumonia in the comprehensive scRNA-seq analysis at the University Medical Center Hamburg (Tables S1 and S2). To compare blood cytokine levels in patients with moderate and severe COVID-19, we analyzed patients from the University of Halle, Germany (Table S3).

Cell isolation

Human BAL fluid and peripheral blood for flow cytometry and scRNA-seq were both obtained from patients undergoing BAL. The indication and performance of bronchoscopy was in accordance with the current guideline recommendations (49). These studies were approved by the Ethik-Kommission der Ärztekammer Hamburg, local ethics committee of the chamber of physicians in Hamburg and were conducted in accordance with the ethical principles stated by the Declaration of Helsinki. Informed consent was obtained from all participating patients or legal representatives. Single-cell suspensions were obtained from BAL fluid by washing with PBS followed by filtering through 100 μm, 70 μm, 40 μm (Greiner Bio-One, Kremsmünster, Austria) and 30 μm cell strainers (Partec, Görlitz, Germany). Leukocytes from blood samples were separated from red blood cells using BD Vacutainer CPT tubes with an integrated FICOLL gradient (BD Biosciences, San Jose, CA, USA). Samples were filtered through a 30 μm filter (Partec, Görlitz, Germany) before antibody staining and flow cytometry.

To minimize unspecific antibody binding, cells were incubated with Human BD FC Block (BD Biosciences) for 10 min. Next, cells were surface stained with fluorochrome conjugated antibodies (CD45 (clone HI30), CD3 (OKT3), CD4 (RPA-T4), CD8 (RPA-T8), CD56 (MEM-188), γδ-TCR 8(B1), CD31 (WM59), CD326 (9C4), CD14 (HCD 14), CD7 (CD7-6B7), CD16 (PC3G8), CD19 (HIB19), CD324 (DECM-1); Biolegend and BD Biosciences), barcode-labeled antibodies (Biolegend) for 15 min (see Table S4 for a complete list of antibodies and barcodes). Subsequently a fixable dead cell stain (live/dead fixable near-IR dead cell stain kit; Life Technologies, Carlsbad, CA) to exclude dead cells from analysis was used according do manufacturer´s instructions. Cells were analyzed and sorted on a BD Biosciences FACS AriaFusion.

Histology

For immunohistochemistry human paraffin-embedded lung sections (2 μm) from patients with SARS-CoV-2 infection were stained with an antibody directed against CD3 (polyclonal rabbit anti-human, Ref. A0452, DAKO, Glostrup, Denmark). Immunofluorescence microscopy was performed in 1 μm paraffin-embedded sections, following 15-min antigen retrieval with pH9 antigen retrieval solution (Agilent, Santa Clara, CA, USA) and incubation with polyclonal primary goat anti-CD4 antibody (AF-379, R&D Systems, Minneapolis, MN, USA), and rabbit anti-CCR6 antibody (ab140768, abcam, Cambridge, UK). Images were captured using a laser confocal microscope (LSM800, Zeiss, Jena, Germany).

For combined detection of IL17A mRNA and CCR6 fluorescence, in situ hybridization (FISH) was performed on FFPE human lung samples using RNAscope-technology as previously describe (50) in accordance with the directions from Advanced Cell Diagnostics. The RNAscope Hs-IL17A-C3 probe from Advanced Cell Diagnostics (Advanced Cell Diagnostics, 310931-C3) was used as the target probe to detect IL-17a mRNA. Fluorescent labeling of the target probe was performed using OPAL 690 dye (Akoya biosciences; FP1497001KT, dilution 1:1000). Subsequent immunofluorescence labeling was performed with an antibody against CCR6 (OriGene Technologies; TA316610, dilution: 1:200) in the same sections after completing the FISH protocol. Epifluorescence imaging was performed using the Thunder Imager 3D Live Cell & 3D Cell Culture (Leica Microsystems).

Multiplex

We used a bead-based immunoassay technology (LEGENDplex, Biolegend) to quantify the concentration of cytokines in the serum and BAL fluid for each sample. The pre-mixed Human Anti-Virus Response Panel (Cat# 740349) and the Human Essential Immune Response Panel (Cat# 740929) were applied to analyze the relevant cytokines following the manufacturer’s protocol. Values below the limit of detection were considered zero. Collection of the Halle cohort was performed under institutional review board approvals number 2020-039 and 11/17. This cohort is partially published (37).

Cell sorting, library preparation and next-generation-sequencing

To enrich for T cells from the BAL fluid, we FACS-sorted T cells, alveolar macrophages, monocytes, CD45highCD3neg cells (including innate lymphoid cells) and CD45neg cells (lung cells) according to the gating strategy presented in Fig. S1A. From peripheral blood, we FACS-sorted CD3pos T cells. Subsequent scRNA-seq using the 10x Chromium Controller (10x Genomics, Pleasanton, CA, USA) was loaded with the following proportions: 1st lane: 100% BAL fluid T cells; 2nd lane: BAL fluid 17% alveolar macrophages, 17% monocytes, 33% CD45neg cells, 33% CD45highCD3neg cells; 3rd lane: 100% blood T cells (cell numbers were based the FACS-sorting information).

Single-cell libraries were generated with the 10x Genomics Chromium Single Cell 5′v1.1 reagents kit according to the manufacture´s instructions. 50 nm cDNA were used for gene expression library construction. Quality control was performed with hsDNA Qubit (ThermoFischer, Waltham, MA, USA) and BioAnalyzer (Agilent). The libraries were sequenced on an Illumina NovaSeq 6000 system (S4 flow cell) with 150 base-pairs and paired end configurations.

Pre-processing of single-cell RNA-seq and CITE-seq data.

The Cell Ranger software pipeline (v3.1.0, 10x Genomics) was used to demultiplex cellular barcodes and map reads to the human reference genome (refdata-cellranger-GRCh38-3.0.0) (command cellranger count). The CITE-seq antibody and barcode information was included in a feature reference csv file and passed to the cellranger count command. As the output, we obtained the feature-barcode matrix that contains gene expression counts alongside CITE-seq counts for each cell barcode. The feature-barcode matrices for all the sample were further processed by the R package Seurat (v3.1.4) (51). As a quality-control (QC) step, we first filtered out the cells in which less than 200 genes were detected in the BALF samples and less than 500 genes were detected in the blood samples. To remove potential doublets, we excluded cells with total number of detected genes more than 5000. Following visual inspection of the distribution of cells by the percentage of mitochondrial genes expressed, we further removed low-quality cells with more than 5% mitochondrial genes of all detected genes. We used LogNormalize method in Seurat to normalize the scRNA-seq and CITE-seq counts for the cells passed the quality control.

Sample aggregation and integration

For the BALF cell analysis, we first aggregated the BALF CD3 positive sample and CD3 negative sample for each patient using the function merge in Seurat. (We excluded the CD3 negative samples of patient S2 and B3 and the CD45 negative sample of patient S9 due to low sequence quality. For patient S9, we merged the CD45 positive sample and EpCAM positive sample). After we obtained the merged BALF Seurat object for each patient, to remove the batch effects across different patients, we applied the integration method implemented in Seurat (function FindIntegrationAnchors and IntegrateData, dims = 1:30). For the blood CD3 positive cell analysis, we directly applied the integration to the samples of all patients. For the combined analysis of BALF and blood T cells, we selected T cell clusters identified in the BALF samples (as described below) and aggregated with corresponding blood CD3 positive samples for each patient using the merge function. Integration was then applied to the merged objects for all the patients.

Dimensionality reduction and clustering

For each integrated object, the integrated matrix was scaled by ScaleData function (default parameters) and highly variable genes were detected (function FindVariableFeatures, selection.method = “vst”, nfeatures = 2000). Principal component analysis was performed on the scaled data (function RunPCA, npcs = 30) in order to reduce dimensionality. 30 principal components were used to compute the KNN graph based on the euclidean distance (function FindNeighbors), which then generated cell clusters using function FindClusters. The resolution parameter of the FindClusters function for each dataset were also determined by exploration of top marker genes of each clusters. Uniform Manifold Approximation and Projection (UMAP) was used to visualize clustering results. The top DEG in each cluster were found using the FindAllMarkers function (min.pct = 0.25, logfc.threshold = 0.25) that ran Wilcoxon rank sum tests. Seurat functions AverageExpression and DoHeatmap were used to visualize the expression of the top marker genes or CITE-seq protein expression in each cell cluster. The top marker genes as well as the CITE-seq expression patterns were then used to determine the cell type of each cluster. The differential expression between selected clusters were calculated by the FindMarkers function (min.pct = 0.1), which also ran Wilcoxon rank sum tests.

BALF T cells, myeloid cells re-integration and sub-clustering

For the separate analysis of BALF T cells and BALF myeloid cells, we selected the clusters identified in the total BALF samples and re-integrated them by patients. Re-clustering was performed after integration as described above and a detailed cell type annotation was obtained after exploring the top marker genes and the CITE-seq expression profiles of clusters. For the sub-clustering analysis of CD4 positive cells, re-integration by patients and re-clustering were also performed respectively before the identification of the cell subtypes.

Processing of TCR-seq data and integration

TCR-seq data for each sample were assembled by the Cell Ranger software (v3.1.0, 10x Genomics) with the command cellranger vdj using the reference genome (refdata-cellranger-vdj-GRCh38-alts-ensembl-3.1.0). For each sample, Cell Ranger generated an output file, filtered_contig_annotations.csv, containing TCR α-chain and β-chain CDR3 nucleotide sequences for single cells that were identified by barcodes. The R package scRepertoire (v1.2.1) (52) was used to further combine the contig_annotation data of different samples to a single list object (function combineTCR). The combined TCR contig list file was then integrated with the corresponding Seurat object of the scRNA-seq data using the function combineExpression (cloneCall=”gene+nt”). Only the cells with both TCR and scRNA-seq data were kept for downstream clonotype analysis. The clonotype was defined according to the genes comprising the TCR and the nucleotide sequence of the CDR3 region. The frequency of the each clonotype in each patient was then calculated as clone count. To get a normalized clone count size for each clonotype, we also calculated the clone size proportion (clone count divided by number of cells per patient). The clone count and clone size proportion were added to metadata of the single cell matrices.

Calculation of gene signature scores

Signature scores of gene sets were calculated by Seurat function AddModuleScore (nbin = 24, ctrl = 100). The cytokine secretion gene set includes major pro-inflammatory cytokines produced by T cells. The residency and migration gene sets were obtained from a core list of up-regulated and down-regulated genes by CD4 tissue-resident T cells (Table S5) (16, 53).

Cell-cell interaction analysis

We applied CellphoneDB’s statistical analysis method (2.1.2) and receptor-ligand database (2.0.0) to calculate statistically enriched cell-cell-interactions (https://github.com/Teichlab/cellphonedb). We used the log-normalized RNA assay of our BAL fluid dataset containing all samples and selected either cells from COVID-19 or bacterial pneumonia patients to gain the count matrix. Because we have different levels of subclustering, we annotated each cell according to its cluster in its deepest level and used it as meta data input (Clustering Level: all BALF samples > T-cells > CD4+ T-cells; all BALF samples > myeloid cells). We ran CellphoneDB with the default parameters. In total CellphoneDB returned 13034 significant (p-value < 0.05) interactions. The rank of a ligand-receptor pair was calculated by CellphoneDB dividing its total number of significant p-values by the number of cluster-cluster comparisons. For downstream analysis we excluded integrin ligand-receptor pairs and interactions being annotated as not secreted. Moreover, we excluded CCL20-CXCR3 interactions (Id_cp_interaction “CPI-SS0F8C664D9”) because of a lack of evidence in the literature.

Connection Strength

The Connection Strength of a specific interaction between two clusters were calculated by multiplying the mean expression of the ligand in the ligand-cluster by the proportion of cells expressing the receptor in the receptor-cluster. The receptor was expressed if the log-normalized expression value was greater than 0. In case of a receptor complex, the receptor component expressed in the least cells was used.

Pathway analysis

The differential expression of SARS-CoV-2 infected patients was calculated using patients with bacterial pneumonia as control group. With the differential expression data, enriched pathways were determined using Gene Ontology Terms and KEGG pathways. This reveals commonly known pathways (e.g., JAK-STAT-signaling and P3K pathway). From these results, relevant parts of the pathways were curated and combined to the final pathway, using the KGML schema (https://www.kegg.jp/kegg/xml/docs/). Coherent components of these enriched pathways were combined to a single representative pathway for each subset of macrophages. The log-fold change of DEG was added as color code to the elements of the pathway.

Network plots

Network plots were created using the R package igraph (1.2.5) (https://github.com/igraph). The layout of the network in Fig. 4A was calculated using the Fruchterman-Rheingold-Algorithm (function layout_with_fr, niter=5000). The weight parameter was set to the number of interactions between two clusters. The clusters “CD8_Tcells”, “M1_HSP”, and “epithelial cells” were excluded. The vertex size was set by using graph strength, which sums up the edge weights for each vertex. The curated interaction map in Fig. 4D was made using the R package igraph (https://github.com/igraph) without applying the Fruchterman-Reingold-Algorithm. The thickness of the lines correlates with the respective connection strength ([average ligand expression] x [proportion of cells expressing the receptor]).

Statistics

Statistical analysis was performed using GraphPad Prism (La Jolla, CA). The results are shown as single data points with the mean ± SEM in a scatter dot plot. Differences between two individual groups were compared using a Mann-Whitney test. In the case of three or more groups, Wilcoxon test was used.

Supplementary Materials

immunology.sciencemag.org/cgi/content/full/6/56/eabf6692/DC1

Fig. S1. Sorting strategy and clustering information of BAL fluid immune cells from all patients.

Fig. S2. Clustering information of blood T cells from all patients.

Fig. S3. Flow cytometry of cells from peripheral blood and BAL fluid.

Fig. S4. Location of T cells in the lungs of COVID-19.

Fig. S5. Clustering information of BAL fluid T cells from all patients.

Fig. S6. Clonal expansion analysis of BAL fluid T cells from all patients.

Fig. S7. Subclustering of clonally expanded BAL fluid CD4 T cells from all patients.

Fig. S8. Clustering information of BAL fluid myeloid cells from all patients.

Fig. S9. Cell-cell interaction of T cells with myeloid cells and epithelial cells.

Fig. S10. Cytokine levels in BAL fluid.

Table S1. Baseline characteristics and disease-related parameters of COVID-19 patients and controls.

Table S2. Relevant medication of COVID-19 patients and controls

Table S3. Baseline characteristics of the Halle cohort

Table S4. CITE-seq antibodies and barcodes (Excel file).

Table S5. Gene sets (Excel file).

Table S6. Raw data file (Excel file).

https://creativecommons.org/licenses/by/4.0/

This is an open-access article distributed under the terms of the Creative Commons Attribution license, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

References and Notes

Acknowledgments: We sincerely thank healthy donors, patients and their families for participating. FACS sorting was performed at the UKE FACS sorting core facility. We thank Elaine Hussey for editing the paper. Funding: This study was supported by grants from the Bundesministerium für Bildung und Forschung, BMBF (IT-COVID-19 to N.G. and C.F.K.); Deutsche Forschungsgemeinschaft, DFG (SFB1192 to J.E.T., V.G.P., S.H., T.B.H., S.B., U.P., N.G., C.F.K.; SFB841 to S.H., N.G.; KFO296 to N.G.; SFB1286 to S.B.); European Research Council (Diet-namic, 715271) and Deutsche Zentrum für Infektionsforschung, DZIF to N.G.; grants from the Deutsche Nierenstiftung and Deutsche Gesellschaft für Nephrologie to C.F.K.; a grant from the Werner Otto Stiftung to L.U.B.E.; a grant from the Else Kröner-Fresenius-Stiftung (EKFS) (iPRIME). Author contributions: Study initiation, S.H., T.B.H., S.K., U.P., N.G. and C.F.K; Conceptualization, J.E.T., M.B., S.H., T.B.H., S.K., S.B., U.P., N.G. and C.F.K.; Methodology and analysis, Y.Z., C.Ki., L.B. K.R., P.B., A.C.G., F.C., C.S., M.H., L.U.B.E., F.H., A.B., H.J.P., D.J., M.L., S.P., S.S., J.P.S., S.B., N.G., C.F.K.; Writing, J.E.T., L.B., U.P., N.G., C.F.K.; Visualization, Y.Z., C.Ki., J.E.T., L.B., V.P., S.S., S.B., N.G., C.F.K.; Supervision, U.P., N.G., C.F.K.; Competing interests: M.L. has the following financial relation relationships: Roche Diagnostics: member of an advisory board, received speaking fees and research funding; Diasorin: received speaking fees; bioMérieux: received speaking fees. Data and materials availability: The raw sequencing data are available in the Sequence Read Archive under accession number PRJNA701725. All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. This work is licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) license, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/. This license does not apply to figures/photos/artwork or other content included in the article that is credited to a third party; obtain authorization from the rights holder before using such material.

Stay Connected to Science Immunology

Navigate This Article