key: cord-0716664-m9779vqe authors: Galván-Peña, Silvia; Leon, Juliette; Chowdhary, Kaitavjeet; Michelson, Daniel A.; Vijaykumar, Brinda; Yang, Liang; Magnuson, Angela; Manickas-Hill, Zachary; Piechocka-Trocha, Alicja; Worrall, Daniel P.; Hall, Kathryn E.; Ghebremichael, Musie; Walker, Bruce D.; Li, Jonathan Z.; Yu, Xu G.; Mathis, Diane; Benoist, Christophe title: Profound Treg perturbations correlate with COVID-19 severity date: 2020-12-15 journal: bioRxiv DOI: 10.1101/2020.12.11.416180 sha: 1b31752ecf9a0431e72740c05bbe31476cc1aa99 doc_id: 716664 cord_uid: m9779vqe The hallmark of severe COVID-19 disease has been an uncontrolled inflammatory response, resulting from poorly understood immunological dysfunction. We explored the hypothesis that perturbations in FoxP3+ T regulatory cells (Treg), key enforcers of immune homeostasis, contribute to COVID-19 pathology. Cytometric and transcriptomic profiling revealed a distinct Treg phenotype in severe COVID-19 patients, with an increase in both Treg proportions and intracellular levels of the lineage-defining transcription factor FoxP3, which correlated with poor outcomes. Accordingly, these Tregs over-expressed a range of suppressive effectors, but also pro-inflammatory molecules like IL32. Most strikingly, they acquired similarity to tumor-infiltrating Tregs, known to suppress local anti-tumor responses. These traits were most marked in acute patients with severe disease, but persisted somewhat in convalescent patients. These results suggest that Tregs may play nefarious roles in COVID-19, via suppressing anti-viral T cell responses during the severe phase of the disease, and/or via a direct pro-inflammatory role. The hallmark of severe COVID-19 disease has been an uncontrolled inflammatory response, resulting from poorly understood immunological dysfunction. We explored the hypothesis that perturbations in FoxP3+ T regulatory cells (Treg) , key enforcers of immune homeostasis, contribute to COVID-19 pathology. Cytometric and transcriptomic profiling revealed a distinct Treg phenotype in severe COVID-19 patients, with an increase in both Treg proportions and intracellular levels of the lineage-defining transcription factor FoxP3, which correlated with poor outcomes. Accordingly, these Tregs over-expressed a range of suppressive effectors, but also pro-inflammatory molecules like IL32. Most strikingly, they acquired similarity to tumorinfiltrating Tregs, known to suppress local anti-tumor responses. These traits were most marked in acute patients with severe disease, but persisted somewhat in convalescent patients. These results suggest that Tregs may play nefarious roles in COVID-19, via suppressing anti-viral T cell responses during the severe phase of the disease, and/or via a direct pro-inflammatory role. COVID-19 resulting from SARS-CoV2 infection is a major global health challenge. Many infected individuals remain asymptomatic or present with only mild flu-like symptomatology, appearing 5-10 days after exposure, and clearing over 1-2 weeks. But this is followed, in patients with more severe disease, by immunopathology and immune dysregulation of poorly understood origin, a major root of the fatality rate of 1-12 % in different locales. Broad immune-profiling studies (Mathew et al., 2020; Laing et al., 2020; Su et al., 2020) have documented several immune phenomena that track with disease severity: lymphopenia , myeloid cell abnormalities (Schulte-Schrepping et al., 2020) , impaired response to interferon (Hadjadj et al., 2020) , and high levels of inflammatory cytokines ("cytokine storm") (Del Valle et al., 2020) . Multiomic studies have shown that these manifestations are embedded within multi-trait immunotypes (Mathew et al., 2020; Laing et al., 2020; Su et al., 2020) , complicating mechanistic inference and the definition of potential therapies. Regulatory T cells (Tregs) expressing the transcription factor FoxP3 are essential to maintain immunologic homeostasis, self-tolerance, and to prevent runaway immune responses (Josefowicz et al., 2012) . Tregs regulate the activation of several lineages of the innate and adaptive immune systems, through several effector mechanisms (Vignali et al., 2008) . Furthermore, particular populations of "tissue Tregs" provide homeostatic regulation in several non-immunological tissues, controlling inflammation and promoting harmonious tissue repair (Panduro et al., 2016) . On the other hand, Tregs can also prove noxious, as evidenced most clearly by their suppression of effective cytotoxic responses in tumors, situations in which they adopt a distinctive phenotype (De Simone et al., 2016; Plitas et al., 2016; Magnuson et al., 2018) . They can also have paradoxical effects on antiviral responses (Lund et al., 2008; Almanan et al., 2017) In light of these contrasting influences, we hypothesized that Tregs might contribute to the balance of disease manifestations that distinguish mild from severe outcomes after SARS-CoV2 infection: for instance by insufficiently curtailing the inflammatory component, by over-curtailing the anti-viral response, or by phenotypic destabilization. We thus performed a deep immunologic and transcriptional analysis of circulating blood Tregs across a cohort of confirmed COVID-19 patients. We thus hypothesized that Tregs might contribute to COVID-19 pathology, and performed a deep immunologic and transcriptional analysis of circulating blood Tregs across a cohort of confirmed COVID-19 patients (n=57, Fig. 1A , Table S1 ; mild: outpatients; severe: hospitalized, 65% of which in intensive care (ICU), mostly sampled during the cytokine storm period; recovered: virus-negative convalescents). Flow cytometry, with a multiparameter panel that parsed CD25 + FoxP3 + Tregs and their different phenotypes (gating strategies in Fig. S1A ), revealed several perturbations (representative plots in Fig. 1B) . First, several severe patients showed increased Treg proportions among CD4+ T cells; for some as an increased proportion only (likely resulting from preferential resistance among CD4+ T cells during lymphopenia); for others, from true numerical increase relative to healthy donors (HD) ( Fig. 1C ; this conclusion differs from a recent analysis of CD4+ T cells that up-regulated CD69/4.1BB after 24hrs' culture with a multipeptide cocktail, hence very different from the present ex vivo study and possibly complicated by selection in culture (Meckiff et al., 2020) ). Recovered patients largely reverted to baseline. Second, expression of both FoxP3 and CD25 varied in severe patients, bidirectionally for CD25 ( Fig. S1B ), but as a reproducible increase of FoxP3 mean fluorescence intensity (MFI; Fig. 1D ). No hint of FoxP3 induction was observed in conventional CD4 + T cells (Tconv). Increased FoxP3 expression coincided with the increase in Treg percentage in most but not all patients (Fig. S1C) , and was observed within a broad window of 30-50 days after symptom onset, including some recently recovered patients (Fig. 1E) . It was not related to body/mass index (BMI), a risk factor for COVID-19 (Fig. S1D ), but coincided with disease severity, particularly marked in ICU-admitted patients (Fig. S1E) , and in lymphopenic cases (Fig. S1F ). No dependence of these phenotypes was found on any particular treatment, in particular glucocorticoids (Table S1 ). FoxP3 expression was not directly correlated with blood levels of the inflammatory marker C-reactive protein (CRP), but essentially all FoxP3 hi patients had elevated CRP (Fig. S1G) . Therefore, severe COVID-19 entails a striking induction of FoxP3 expression in Tregs, in a manner not been previously observed in any autoimmune or infectious context. We examined the expression of several markers and transcription factors (TF) to further assess Treg evolution during COVID-19. CD45RA, which marks naïve Tregs, was reduced in patients, especially in those with heightened Tregs and FoxP3 (Fig. S1H,I) . Expression of the activation markers KLRG1 and PD1 was increased (Fig. 1F ), although not necessarily coordinately (Fig. S1J ). Suppression of specific T cell effector functions is associated with the expression in Tregs of the same key TFs that drive those Tconv functions (Josefowicz et al., 2012) . No patient sample showed significant expression of Bcl6, which marks T follicular regulators (Linterman et al., 2011; Chung et al., 2011) . But Tbet, expressed in Tregs that preferentially control Th1 responses, was over-represented in severe patient Tregs (Fig. 1G) . Interestingly, severe patients with frequent Tbet+ Tregs were distinguished from those with highest Treg expansion and FoxP3 over-expression (Fig. 1H ). Ultimately deceased patients were mostly found in the latter group. Thus, severe COVID-19 seems to elicit divergent deviations among Treg cells. We surmise that the presence of Tbet + Tregs is related to the control of Th1 antiviral response among effector cells. To decipher the functional consequences of FoxP3 overexpression in Tregs from severe COVID-19 patients, we generated 86 RNAseq transcriptome profiles, passing quality thresholds, of blood Treg (CD4 + CD25 hi CD127 lo ) or Tconv (CD4 + CD25 -). Donors largely coincided with those analyzed above, and we opted to profile purified populations rather than single-cell (higher throughput, lower processing/computational costs) after magnetic purification (Fig. S2A) . The datasets matched profiles from flow-purified Tregs, with the usual differential expression of "Treg signature" genes ( Fig. S2B ). In line with the cytometry, Tregs from severe COVID-19 patients showed higher FOXP3 expression (Fig. S2C ). This over-expression had a consequence: Tregs from severe COVID-19 patients displayed a heightened expression of Treg signature transcripts ( Fig. 2A) , reflected by a high "TregSignature Index", most markedly biased in the severe group, but also in mild and recovered patients (Fig. 2B ), and correlating with FOXP3 expression (Fig. S2C ). This bias was not homogeneous across all Treg-up signature genes, a ranked plot of differential expression in each donor revealing a small subset of TregUp signature genes actually downregulated in severe COVID-19 patients (including TLR5, ID3, FCRL1; Fig. 2C ). At the other end of the spectrum, several Treg effector or activation transcripts were up-regulated in severe patients (ENTPD1, HPGD, IL12RB2). Most transcripts encoding known Treg effector molecules showed were up-regulated, albeit to various degrees (Fig. 2D) , with the marked exception of AREG, a dominant player in Treg promotion of tissue repair (Burzyn et al., 2013; Arpaia et al., 2015) . We also examined transcripts associated with effectiveness at suppressing different Tconv functions (Josefowicz et al., 2012) . Transcripts associated with T follicular regulators were largely unaffected in Tregs from severe patients (Fig. 2E) , with the exception of PDCD1 (encodes PD1), consistent with cytometry results. In contrast, they generally up-regulated markers related to Th1 suppression (CXCR3, GZMK, IL12RB1 or TBX21 (encodes Tbet)), especially marked in patients with low percentage of Tregs (Fig. 2E ), in line with the over-representation of Tbet+ Tregs noted above. Profiles from Tconv cells did not denote a particular bias towards any Th phenotype (Fig. S2D ,E). Thus, Treg traits observed in the flow cytometry data were confirmed by the transcriptomic signature of these Tregs, which tends towards a super-suppressive phenotype in severe COVID-19 patients. We then asked more generally what changes, beyond Treg signature and effector transcripts, characterized Tregs in severe COVID-19 patients, relative to HD (Fig. 3A , Table S2; hereafter 'Severe COVID19 Treg Signature', SCTS); only a minority of these transcripts belong to the classic Treg signature analyzed above. A SCTS Index computed from this geneset was high in all severe patients, but also persisted in muted fashion in many recovered patients (Fig. 3B ), correlating with FOXP3 expression (Fig. S3A ). The index was not directly related to patient age or BMI (Fig. S3B ). Grouping SCTS transcripts into a biclustered heatmap (Fig. 3C ) revealed several interesting features. Most donors clustered according to severity group, in relation to disease duration (more marked deviation at shorter times), but not to donor age (except inasmuch as severe patients were generally older). Transcripts could be parsed into 8 different modules of distinct composition. Among these, module M4 was almost exclusively composed of cell cyclerelated transcripts, and strongly correlated with FoxP3 MFI, indicating that Tregs in COVID-19 patients are highly proliferative, plausibly a compensation for the lymphopenia in many severe patients. M2 mostly included Interferon-stimulated genes (ISGs). The other co-regulated modules included transcripts related to immune cell crosstalk (Fig. 3C, right) . Expression of these modules evolved differently during the course of the disease (Fig. 3D ). The ISGs in M2 were mostly overrepresented at early times, consistent with sharp induction at the early phase of anti-viral responses. Changes in other modules, especially M3 and M6, were shared across all disease stages, indicating a broader and longer-lasting perturbation of the Treg pool, likely caused by secondary consequences of the disease rather than by the virus itself. Focusing on cytokines produced by Tregs, IL10 was only modestly induced, but CD70 (CD27L) and IL32 dominated (Fig. 3E ). The latter was intriguing in this context of the cytokine storm that occurs in these severe patients, since it is mainly a pro-inflammatory mediator, in positive feedback with IL6 and IL1 (Zhou and Zhu, 2015) . To answer the common question (whether transcriptional changes reflect a shift in subset balance or all shared by all Tregs), we re-analyzed a single-cell RNAseq PBMC dataset from COVID-19 patients (Wilk et al., 2020) , drilling down on identifiable Treg cells (Fig. S3C ). Tregs from severe patients were generally shifted in the UMAP projection relative to HD ( Fig. 3F) , and displayed an up-regulation of the SCTS (Fig. 3G, Fig. S3D ). IL32 was again one of the dominant up-regulated transcripts (Fig. S3E) . We then attempted to better understand the origin of the SCTS. Comparing changes in Treg and Tconv cells showed some induction in the latter (particularly for the inflammation-related components of M1), but overall much less than in Tregs (Fig. S3F) , indicating that the SCTS is largely Treg-specific. With regard to the cytokines, the upregulation of IL10 and IL32 were specific to Tregs, as was the down regulation of inflammatory cytokines (TNF, IL1B) ( Fig S2E) . Interestingly, the SCTS was associated with changes in several TFs previously associated with differential gene expression in activated Tregs (Fig. 3H ). PRDM1 (aka BLIMP1) and MAF were upregulated, while ID3, TCF7 and BACH2 were repressed, consistent with previous reports (Cretney et al., 2011; Xu et al., 2018; Maruyama et al., 2011; van Loosdregt et al., 2013; Grant et al., 2020) . More unexpected was the strong downregulation of NR4A1, normally an indicator of TCR signaling, which may indicate a decoupling of TCR-delivered signals. Gene enrichment analyses using a curated database of transcriptome variation in CD4+ T cells brought forth 229 datasets with significant overlap to the SCTS (Fig. S4A ). Besides the expected interferon-related genesets related to M2, many were related to Tconv and Treg activation, mostly overlapping with M4 and M5. Most intriguing were overlaps with datasets from tumor-infiltrating Tregs. Indeed, the expression of a signature that distinguishes colorectal tumorinfiltrating Tregs (TITR) from normal colon tissue (Magnuson et al., 2018) was strikingly biased in Tregs from severe COVID-19 patients (Fig. 4A ). The same bias was found with genesets that distinguish breast and lung TITRs from blood Tregs (De Simone et al., 2016; Plitas et al., 2016) ( Fig. 4B -C). Computing a "TITR index" from the colorectal tumor geneset showed that biased expression of tumor Treg transcripts was present in all severe patients, particularly in those eventually deceased (Fig. 4D ). This index remained slightly perturbed after recovery, and was highly correlated with the SCTS (Fig. S4B ). Conversely, with the exception of M1 and M6, all SCTS modules showed biased expression in tumor Tregs relative to normal tissue (Fig. S4C ), indicating widespread sharing not solely limited to activation-related transcripts. TITRs and tissue Tregs are transcriptionally similar, suggesting the possibility that the SCTS simply denoted Tregs circulating en masse from the lung tissue. This was not the case, however: when highlighted in a direct comparison of tissue and tumor Tregs (from (Magnuson et al., 2018) ), the SCTC was clearly biased towards the tumor angle (Fig. 4E ). Further, a set of transcripts shared by tissue Tregs showed little enrichment in Tregs from severe COVID-19 patients (Fig. S4D ). This bidirectional matching shows that COVID-19 disease appears to be turning blood Tregs into some equivalent of tumor Tregs. The discovery of this striking Treg phenotype in severe COVD-19 patients, with a unique up-regulation of FoxP3 expression and high transcriptional similarity to tumor Tregs, raises two key questions. First, how do they arise? They are not virally infected (no viral RNA reads to speak of in these cells), and none of the therapies administered to these patients correlates with the Treg traits. Rather, the phenotype must be induced by the immunologic milieu in these patients, and this uniquely in Tregs since Tconvs are far less branded. T cell receptor-mediated stimulation seems unlikely, given the widespread effect on Tregs in the single-cell data which likely transcends clonotypic specificity, and the strong loss of Nur77 (NR4A1). A non-specific trigger seems more likely, such as cytokines, interferon, DAMPs released by dying cells, or even more speculatively signals from non-structural viral proteins. Interferon might be involved, and ISGs are part of the signature. We note the strong representation of TNF Receptor family members among the induced SCTS transcripts. Also consistently upregulated is CXCR3, the receptor for CXCL10, one of the soluble mediators most induced in severe COVID-19 blood (Filbin et al., 2020) . Finally, one factor is common to the tumor microenvironment and to severe COVID-19, hypoxia, which is known to promote Treg suppressive function (Facciabene et al., 2011) . Thus, a combination of factors may conspire to achieve the Treg phenotype. Second, do these aberrant Tregs contribute to COVID-19 physiopathology? Patients with fewer Tregs, lower FoxP3 and less intense SCTS do better, raising the usual issue of inferring causality from correlation. On one hand, these Tregs might be beneficial, controlling a cytokine storm that would have been worse without their unusual contribution. On the other hand, their over-expression of FoxP3, of Treg effector molecules, and their similarity with dominantly suppressive tumor Tregs suggest that COVID-19 Tregs may overly dampen the anti-viral response during the cytokine storm phase (all severe patients profiled were from that period), contributing to the secondary re-expansion of disease. In addition, the intriguing up-regulation of the pro-inflammatory cytokine IL32 (Zhou and Zhu, 2015) might mark these Tregs as amplifying the cytokine storm instead of controlling it. There are precedents for rogue Tregs which acquire pro-inflammatory characteristics (Overacre-Delgoffe et al., 2017)although admittedly in opposite conditions of FoxP3 attrition. In summary, adding a key element to the multifaceted COVID-19 immune response, we identify a unique Treg deviation in COVID-19 patients, which might impact on COVID-19 pathogeny as it does in tumors. We thank Dr C. Blish discussion and access to unpublished data; F. Chen for help with sample processing, and the Broad Genetic Platform for profiling in difficult times. This work was funded by grants from the Massachusetts Consortium for Pathogen Readiness (MassCPR), RO1 AI150686 to CB, and R24 AI072073 to the ImmGen consortium. The MGH/MassCPR COVID biorepository was supported by a gift from Ms. Enid Schwartz, by the Mark and Lisa Schwartz Meckiff, B.J., Ramirez-Suastegui, C., Fajardo, V., Chee, S.J., Kusnadi, A., Simon, H., Eschweiler, S., Grifoni, A., Pelosi, E., Weiskopf, D. et al. (2020) . Imbalance of regulatory and cytotoxic SARS-CoV-2-reactive CD4 + T cells in COVID-19. Cell. 183, 1340 -1353 Overacre Highlight of the severe SCTS Up signature (red) on a plot comparing transcriptome shifts in TITRs vs tissue-resident Tregs (see (Magnuson et al., 2018) ). Peripheral blood samples from 57 adults infected with SARS-CoV-2 (defined by the concomitant presence of a concordant symptomatology and a positive SARS-CoV-2 real-time reversetranscriptase-polymerase-chain-reaction (RT-PCR) using nasopharyngeal swabs) prospectively collected in Massachusetts General Hospital (MGH, Boston, USA), were included in this study (Table S1A and S1B). Samples from 11 adult healthy donors (HD) were also collected, dated before December 2019 or defined as no prior diagnosis of or recent symptoms consistent with COVID-19 and a negative PCR test within 3 days pre-collection. COVID-19 patients were split into three different groups depending on the disease severity or the time-course post infection. Patients were categorized as "severe" if admitted to the hospital for moderate to severe COVID-19 symptoms, either in floor or in intensive care unit, as "mild" group if they were outpatients, as "recovered" if they presented a prior positive COVID-19 PCR test, but the most current PCR test was negative. For inpatients, clinical data, including oxygenation status, medications and the outcome of the hospitalization (discharge or death), were recorded from the electronic medical record. Laboratory results were abstracted from the collection date, or if unavailable the closest to that of research blood collection, mostly within 48h. Clinical and demographic information including age, BMI and whether patients were on immunosuppressant or anti-inflammatory drugs are summarized across the groups and displayed individually, respectively in Table S1A and B. All participants provided written informed consent in accordance with protocols approved by the Partners Institutional Review Board and the MGH Human Subjects Institutional Review Board. De-identified Treg analysis was performed per approved HMS-IRB protocol 15-0504-04. 5 to 10 mL of whole blood was collected in K2 EDTA tubes and processed within a few hours. An equal volume of buffer (2 mM EDTA in PBS) and blood was mixed and carefully layered over 5ml Ficoll Hypaque solution (GE Healthcare). After centrifugation for 20 min at 900 g (with no break) at 25°C, the mononuclear cell layer was washed twice with excess buffer (three times the volume of the mononuclear cells layer), and centrifuged for 5 min at 400 g. To remove platelets, the cell suspension was then layered over 3ml FBS, centrifuged for 10 min at 300g. The pellet was resuspended in 90% FBS-10% DMSO, 5 million cells/mL, and cells stored in liquid nitrogen. Frozen PBMCs samples were processed in batches of 4 to 10 samples, with at least one HD by batch, with strict attention to processing time to avoid cell aggregation, which was otherwise pervasive with samples from severe COVID-19 patients. 5-10x10 5 PBMCs were resuspended into PBS with 2% FBS and 1mM EDTA. CD4+CD25hiCD127low (Treg) were isolated by positive and negative magnetic selection, CD4+CD25-(Tconv) by negative selection only. EasySep™ Human CD4+CD127lowCD25+ Regulatory T Cell Isolation Kit (StemCell, #18063) was used following manufacturer's instructions. 10% of the final isolated fraction was stained with anti-CD4, anti-CD8, anti-CD14, anti-CD19 and live dead for 15 min at 4°C, fixed in 1% PFA for 10 min at room temperature and analyzed by flow cytometry to determine purity and yield (antibodies references below). The remaining 90% of the samples were resuspended in lysis buffer (TCL Buffer Low-input RNAseq: RNAseq was performed in two different batches, including samples coming from 2 to 7 different experimental batches (different experiment dates in Table S1 ). After the magnetic isolation and using the data from the flow cytometry post-isolation, samples with purity > 65% and expected number of cells >800 were selected for RNAseq. RNA-seq was performed with 5µL of the previously described lysate following the standard ImmGen low-input protocol (www.immgen.org). Smart-seq2 libraries were prepared as previously described (Picelli et al., 2014) with slight modifications. Briefly, total RNA was captured and purified on RNAClean XP beads (Beckman Coulter). Polyadenylated mRNA was then selected using an anchored oligo(dT) primer (50 -AAGCAGTGGTATCAACGCAGAGTACT30VN-30) and converted to cDNA via reverse transcription. First strand cDNA was subjected to limited PCR amplification followed by Tn5 transposon-based fragmentation using the Nextera XT DNA Library Preparation Kit (Illumina). Samples were then PCR amplified for 12 cycles using barcoded primers such that each sample carries a specific combination of eight base Illumina P5 and P7 barcodes for subsequent pooling and sequencing. Paired-end sequencing was performed on an Illumina NextSeq 500 using 2 x 38bp reads with no further trimming. Reads were aligned to the human genome (GENCODE GRCh38 primary assembly and gene annotations v27) with STAR 2.5.4a (https://github.com/alexdobin/STAR/releases). The ribosomal RNA gene annotations were removed from GTF (General Transfer Format) file. The gene-level quantification was calculated by featureCounts (http://subread.sourceforge.net/). Raw read counts tables were normalized by median of ratios method with DESeq2 package from Bioconductor (https://bioconductor.org/packages/release/bioc/html/DESeq2.html) and then converted to GCT and CLS format. Quality control. Samples with less than 1 million uniquely mapped reads were automatically excluded from normalization to mitigate the effect of poor-quality samples on normalized counts. Samples having fewer than 8,000 genes with over ten reads were also removed from the data. We screened for contamination by using known cell type specific transcripts (per ImmGen ULI RNAseq and microarray data). Finally, the RNA integrity for all samples were measured by median TIN across human housekeeping genes with RSeQC software (http://rseqc.sourceforge.net/#tin-py). Samples with TIN < 45 were removed from the data set prior to downstream analysis. Batch correction. In order to analyze simultaneously the two RNAseq batches, the read counts tables from these two batches were first combined for all the Treg samples and for all the Tconv samples separately. Then, the batch effects were corrected by using Combat method from sva package. The SARS-Cov-2 genome sequence and annotation were obtained from NCBI (https://www.ncbi.nlm.nih.gov/sars-cov-2/). Reads were aligned by STAR 2.5.4a with the parameters suggested in Kim et al.(Kim et al., 2020) . The gene-level quantification was calculated by featureCounts (Subread 1.6.2). Read counts tables were normalized by median of ratios method with DESeq2 package (Love et al., 2014) Regression Linear model. Related to the epidemiology of the disease, there was a strong sex-bias in severe patients in our dataset (predominantly male). Thus, we performed a linear regression (glm() function in R) using log-transformed expression as the response variable, and severity and sex as explanatory variables. We removed 45 genes highly correlated with sex, which were no longer associated with severity once adjusted. After quality control and regression, genes with a minimum reads count of 20 in more than 20% of samples from a population (Treg or Tconv) were retained. We used an uncorrected t-test to compute differential gene expression between the different groups from the normalized read counts dataset. Genes with a FoldChange >2 or <0.5 and pvalue < 0.05 were selected for further analysis. Unless specified otherwise, data are presented as mean ± SD and tests of associations for different variables between COVID-19 patients and HD were computed (1) The data reported in this paper have been deposited in the Gene Expression Omnibus (GEO) database under accession no. GSEXXX (human RNAseq). Fig. S1 (related to Fig. 1 CXCR3 GZMK IFNG IL12RB1 TBX21 CCR4 GATA3 IL13 IL4 IL5 IRF4 CCR6 IL17A IL17F IL23R RORC STAT3 BCL6 CXCR5 ICOS Tissue-specific control of latent CMV reactivation by regulatory T cells A distinct function of regulatory T cells in tissue protection A special population of regulatory T cells potentiates muscle repair Integrating single-cell transcriptomic data across different conditions, technologies, and species Follicular regulatory T cells expressing Foxp3 and Bcl-6 suppress germinal center reactions The transcription factors Blimp-1 and IRF4 jointly control the differentiation and function of effector regulatory T cells Transcriptional Landscape of Human Tissue Lymphocytes Unveils Uniqueness of Tumor-Infiltrating T Regulatory Cells An inflammatory cytokine signature predicts COVID-19 severity and survival Molecular diversification of regulatory T cells in nonlymphoid tissues Tumour hypoxia promotes tolerance and angiogenesis via CCL28 and T(reg) cells Interindividual variation in human T regulatory cells Plasma proteomics reveals tissue-specific cell death and mediators of cell-cell interactions in severe COVID-19 patients Compendium of Immune Signatures Identifies Conserved and Species-Specific Biology in Response to Inflammation BACH2 drives quiescence and maintenance of resting Treg cells to promote homeostasis and cancer immunosuppression Impaired type I interferon activity and inflammatory responses in severe COVID-19 patients Clinical features of patients infected with 2019 novel coronavirus in Wuhan Regulatory T cells: mechanisms of differentiation and function The architecture of SARS-CoV-2 transcriptome A dynamic COVID-19 immune signature includes associations with poor prognosis Foxp3 + follicular regulatory T cells control the germinal center response Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2 Coordination of early protective immunity to viral infection by regulatory T cells Identification and validation of a tumor-infiltrating Treg transcriptional signature conserved across species and tumor types Control of the differentiation of regulatory T cells and T H 17 cells by the DNA-binding inhibitor Id3 Deep immune profiling of COVID-19 patients reveals distinct immunotypes with therapeutic implications Cells were first incubated in 100 μL of PBS with 2mM EDTA for 15 min with 5µL Fc Block (Human TruStain FcX™, Biolegend, cat #422301) and a 1:500 dilution of Zombie UV viability dye (Biolegend, cat# 423107) and stained at 4°C for 25 minutes using the following cell surface antibodies: CD14 Pacific Blue (clone M5E2 CD3 AF700 (clone OKT3, BioLegend, cat# 317340 CD4 PerCP-Cy5.5 (clone OKT4, BioLegend, cat# 317428 KLRG1 APC-Cy7 (clone 2FI/KLRG1, BioLegend, cat# 138426 PD-1) BV650 (clone EH12.2H7, BioLegend, cat# 329950, dilution 2:100) After cell surface staining, cells were fixed overnight at 4°C using 100 μL of Fix/Perm buffer (eBioscience), followed by permeabilization using 1X permeabilization buffer (eBioscience) for 40 minutes at room temperature in the presence of the following intracellular antibodies: FoxP3 APC (clone PCH101, Invitrogen Tbet BV605 (clone 4B10, BioLegend, cat# 644817 Bcl6 PE/Dazzle (clone 7D1, BioLegend, cat# 358510, dilution 4:100). Data was recorded on a FACSymphony TM flow cytometer (BD Biosciences) and analysed using FlowJo 10 software