CD127 imprints functional heterogeneity to diversify monocyte responses in inflammatory diseases

Monocytes and macrophages are considered major mediators of inflammation in a plethora of human disease settings, including infectious diseases such as COVID-19 and chronic inflammatory diseases such as rheumatoid arthritis (RA; Donlin et al., 2019; Ermann et al., 2015; Merad and Martin, 2020). During the pathological processes, inflammatory monocytes from the peripheral blood accumulate at the sites of infection and/or inflammation and produce large quantities of proinflammatory mediators, exacerbating disease outcomes by promoting the vicious inflammation cycle (Nathan and Ding, 2010; Shi and Pamer, 2011). Despite the extensive efforts undertaken to understand the ontogeny, functionality, and heterogeneity of inflammatory monocytes (Cormican and Griffin, 2020; Dress et al., 2020; Guilliams et al., 2018), there remain many unanswered questions, especially pertaining to characterization of in situ monocyte phenotypes in human inflammatory tissues. One question of importance is whether functional characteristics and molecular networks of human inflammatory monocytes could be extrapolated based on the previous observations made with model organisms such as mice (Godec et al., 2016; Hagai et al., 2018; Schroder et al., 2012; Shay et al., 2013). In fact, recent advances in human immunology have begun to unveil unexpected features of human immune cells that could not be predicted based on the existing knowledge largely derived from mouse studies and in vitro cultures (Kuo et al., 2019; Liao et al., 2020; Zhang et al., 2019a). Another outstanding question is whether tissue-infiltrating inflammatory monocytes display distinct properties as a result of tissue adaptation or retain certain common features regardless of their physical locations. Lack of such understanding would impose significant conceptual and technical barriers to therapeutically targeting human inflammatory diseases (Ermann et al., 2015; van der Poll et al., 2017), a particularly prominent issue amid the COVID-19 pandemic (Vabret et al., 2020), where inflammatory monocytes play a major pathogenic role in lung pathology (Merad and Martin, 2020).

CD127, encoded by the IL7R gene, assembles with the IL2RG-encoded common γ-chain to form the heterodimeric IL-7 receptor. Upon ligand binding, onset of IL-7 receptor signaling leads to activation of receptor-associated JAK1 and JAK3 and subsequent phosphorylation and nuclear translocation of STAT5 to induce target gene expression. Expression and function of CD127 are best characterized in lymphoid lineage cells. CD127 is predominantly expressed in various stages of lymphoid progenitor cells and in mature lymphocytes, including T cells and innate lymphoid cells (ILCs). Functionally, IL-7 signaling is crucial for development of lymphoid lineages as well as maintenance of immune homeostasis, such as survival of memory T cells (Barata et al., 2019; Leonard et al., 2019). In contrast to lymphocytes, myeloid lineages are not well-recognized CD127-expressing cells (Leonard et al., 2019), albeit that one study documented transient expression of CD127 in mouse fetal monocytes (Leung et al., 2019). While investigating gene expression profiles of human peripheral blood monocytes, earlier studies, including ours, unexpectedly observed highly inducible expression of CD127 upon monocyte activation, typically several hundred–fold over baseline (Hu et al., 2008; Martinez et al., 2006). To our knowledge, these studies represented the initial experimental evidence for CD127 expression in human monocytes, which was confirmed by several subsequent studies (Al-Mossawi et al., 2019; Chen et al., 2013; Kim et al., 2020; Pickens et al., 2011). However, despite the realization of monocytic CD127 expression 15 yr ago, the functional significance and biological outcomes of IL-7 receptor signaling in myeloid cells such as monocytes remain obscure.

Recent technological advancements, especially those at the single-cell level, have enabled description of immune cell phenotypes at an unprecedented resolution (Papalexi and Satija, 2018). Here, we set out to characterize functional and molecular features of CD127^high inflammatory monocytes and first found that monocytic expression of CD127 was restricted to humans and was not observed in mice, excluding the possibility of using mice as the model organism and compelling us to investigate these cells in two human inflammatory diseases with high prevalence and significance: COVID-19 and RA. Interrogation of patient samples with multiomics approaches and subsequent mechanistic dissections led to revelation of remarkable functional heterogeneity of human inflammatory monocytes imprinted by the CD127–STAT5 axis. Importantly, advanced bioinformatics analyses integrating multiple data sources unraveled unified features of human inflammatory monocytes, regardless of tissues of origin and disease types, suggesting potentially common monocyte-mediated pathogenic mechanisms in a range of human inflammatory disorders that could serve as new therapeutics targets.

To directly assess expression of CD127 in situ at the sites of infection and/or inflammation, pulmonary autopsy samples from patients succumbing to SARS-CoV-2 infection were analyzed by immunohistochemistry. While minimal CD127 expression was detected in alveoli from individuals who died of noninfectious causes, robust CD127 staining signals were evident in SARS-CoV-2–infected lung tissues and appeared in the regions of CD68 positivity (Fig. 1 A), implying plausible expression of CD127 in monocytes/macrophages. These results were confirmed by colocalization of CD127 and CD68 signals on immunofluorescently stained sections of autopsied COVID-19 lung tissues (Fig. 1 B). Monocyte/macrophage (CD14^high CD68^high) expression of CD127 was further corroborated at the transcriptome level by single-cell RNA-sequencing (scRNA-seq) analyses of bronchoalveolar lavage fluid (BALF) from nine COVID-19 patients with clinical manifestations ranging from mild (n = 3) to severe (n = 6; Fig. S1 A). Strikingly, a distinct IL7R⁺ population was revealed to constitute 21% of BALF monocytes/macrophages, and the expression of IL7R in monocytes is comparable to that of lymphoid cells (Fig. 1, C–E). Moreover, in contrast to the predicted dominance of IL7R positivity by lymphoid cells (Fig. 1 E), the majority (64%) of IL7R⁺ cells in COVID-19 BALF were of the monocyte/macrophage lineage (Fig. 1 F).

To extend the above findings to chronic inflammation, peripheral blood mononuclear cells (PBMCs) from 16 anti-inflammatory treatment–naive RA patients were analyzed for CD127 expression. Compared with the minimal levels of CD127 in CD14⁺ monocytes from healthy donors, all RA patients examined displayed elevated CD127 expression on blood monocytes at protein and mRNA levels (Fig. 1, G and H). In RA blood monocytes, IL7R expression correlated with the expression of a major pathogenic factor, TNF (Fig. 1, H and I; Kalliolias and Ivashkiv, 2016). In addition to circulating monocytes, analyses of synovial immune cell scRNA-seq datasets from 18 RA patients (Zhang et al., 2019a; Fig. S1 B) showed that a distinct population representing nearly 22% of synovial monocytes also exhibited strong IL7R positivity at a level comparable to that in T cells (Fig. 1, J–L). Taken together, expression of CD127 on a subset of inflammatory monocytes/macrophages is likely a hallmark of human inflammatory conditions in multiple tissues (infected lungs, peripheral blood, and inflamed joints) and in multiple disease settings (COVID-19 and RA).

To model CD127-expressing human monocytes for facilitation of further mechanistic investigation, we explored the plausible range of inflammatory stimuli capable of inducing this population in human PBMCs and in CD14⁺ monocytes from healthy blood donors. Stimulation of human PBMCs or CD14⁺ monocytes with TLR ligands led to a drastic increase of monocytic CD127 at protein and mRNA levels, with the exception of TLR3 agonist polyinosinic:polycytidylic acid (Fig. 2, A and B). Up-regulation of IL7R mRNA and protein levels in activated monocytes was consistently observed in >20 healthy donors examined, with the cell surface expression peaking ∼6 h after activation (Fig. 2, C–F). Inducible CD127 expression was comparable to that of the constitutive CD127 in T cells (Fig. S2 A) and dependent on canonical TLR signaling modules such as NF-κB and MAPKs as evidenced by pharmaceutical inhibition and RNA interference (Fig. S2, B–E). Activation-induced CD127 up-regulation was observed in all three subsets of human blood monocytes defined by binary expression of CD14 and CD16 (Fig. 2 G and Fig. S2, F–H), suggesting that CD127 up-regulation is a common feature of human monocyte activation. In addition to pathogen-associated molecular patterns that might contribute to the appearance of CD127⁺ monocytes in infectious diseases, we also sought such factors in the context of sterile inflammation and pursued TNF as a plausible candidate because its levels correlated with CD127 expression in RA blood monocytes (Fig. 1 I). TNF treatment consistently up-regulated CD127 in monocytes from healthy donors, albeit to a lesser extent than LPS (Fig. 2, H and I). Importantly, clinically applied TNF blockade treatment significantly reduced IL7R expression in RA monocytes (Fig. 2 J), further solidifying a role for TNF in up-regulation of CD127 in vivo.

Because mice are widely used as model organisms to mimic human immunological processes, we assessed whether human monocytic expression of CD127 could be recapitulated in mice. As expected, mouse blood monocytes and tissue-resident macrophages did not express CD127 at homeostasis (Yoshida et al., 2019; Fig. S2 I). However, in contrast to the observations in humans, LPS stimulation up to 18 h failed to induce CD127 in murine blood CD11b⁺ monocytes and various macrophage populations examined (Fig. S2, J and K; and data not shown). Consistent with the lack of expression in vitro, in models of pulmonary inflammation (Sajti et al., 2020), CD127 remained undetectable in pulmonary myeloid cells, including alveolar macrophages, interstitial macrophages, and resident monocytes (Fig. S2 L). Taken together, pathogen-associated molecular patterns and sterile inflammatory stimulation elicit a subset of CD127-expressing monocytes in humans but not in mice, highlighting the unique features of human inflammatory monocytes and mandating the necessity to conduct the following experiments in a human system.

Given the wide range of CD127 expression among activated human monocytes (Fig. 2, E and F), we hypothesized that CD127 levels may correlate with certain monocyte phenotypes and subjected LPS-activated human CD14⁺ monocytes to scRNA-seq to reveal the potential diversities. Unsupervised clustering revealed four groups of cells within seemingly uniformly activated monocytes (Fig. 3, A and B). Interestingly, IL7R expression was highest in cluster 1 and exhibited a decreasing gradient from cluster 1 to cluster 4 (Fig. 3 C). In fact, IL7R is one of the top 15 signature genes of cluster 1, whereas cluster 4 was enriched with a number of well-defined proinflammatory mediators (Fig. 3 D). To further quantitatively assess the inflammatory phenotypes of these cells, we devised a numeric index “inflammatory score” based on an algorithm (Stuart et al., 2019) reflecting expression levels of eight prototypical inflammatory genes (see Materials and methods for details). Inflammatory score inversely correlated with the expression of IL7R, with the highest inflammatory score shown for cluster 4, which exhibited the lowest level of IL7R (Fig. 3 E). Such a trend could also be clearly visualized for individual inflammatory genes (Fig. 3 F). To validate the differences of inflammatory responses observed from single-cell analyses, we sorted CD127^high and CD127^low LPS-activated monocytes (Fig. 3 G). Consistent with the results from the single-cell analyses, CD127^low monocytes produced significantly higher levels of inflammatory mediators IL-6 and TNF than their CD127^high counterparts (Fig. 3, H and I). In contrast to divergent inflammatory phenotypes, certain effector functions, such as NO production and phagocytosis capacity, remained comparable between CD127^high and CD127^low monocyte populations (Fig. S3, A and B), suggesting relative functional specificity regarding CD127-delineated heterogeneity of inflammatory phenotypes. Together, these results implied that, within a highly defined system consisting of purified monocytes and a single stimulus, human monocytes displayed a diverse spectrum of inflammatory responses whose strengths were inversely correlated with CD127 expression.

Next, we wished to investigate the mechanisms by which CD127 modulated monocyte phenotypes. In human monocytes, another subunit of IL-7 receptor, the IL2RG-encoded common γ-chain, was constitutively expressed (Fig. S3 C). Therefore, we speculated that induction of CD127 would render activated human monocytes responsive to IL-7 that is otherwise inert for resting monocytes. Indeed, resting monocytes did not respond to stimulation with up to 10 ng/ml of IL-7 as measured by STAT5 tyrosine phosphorylation (Fig. 4 A). Interestingly, activated monocytes exhibited high sensitivity to IL-7 as shown by induction of STAT5 activation with a picogram range of the cytokine (Fig. 4 A), comparable to the concentrations of IL-7 detected in the in vitro monocyte culture system (Fig. S3 D). With IL-7 signaling in human T cells serving as a positive control (Fig. S3 E), the above data implied that activated human monocytes were competent to sense IL-7 via CD127-mediated IL-7 receptor signaling. Next, to assess whether active receptor signaling was causally related to the hypoinflammatory phenotypes, we knocked down CD127 expression by transfecting IL7R-targeting siRNA into CD14⁺ human monocytes (Fig. 4 B). Compared with control siRNA-transfected monocytes, knocking down IL7R resulted in significantly up-regulated IL6 and TNF mRNA in activated monocytes (Fig. 4 B). Moreover, addition of exogenous IL-7 to activated monocytes suppressed IL6 expression (Fig. S3, F and G), further supporting a negative role of monocytic IL-7 receptor signaling in inflammatory gene expression.

To comprehensively characterize features of CD127-imposed heterogeneity, we profiled the chromatin accessibility landscapes of sorted CD127^high and CD127^low human monocytes by assay for transposase-accessible chromatin coupled with high-throughput sequencing (ATAC-seq; Fig. 4 C). Analyses of ATAC-seq data obtained from three individual donors consistently revealed open chromatin regions (OCRs) that were differentially enriched in CD127^high versus CD127^low populations, thus termed “CD127^high feature OCRs” and “CD127^low feature OCRs,” respectively (Fig. 4, D–F). In further epigenetic characterizations, CD127^low feature OCRs implicated gene-proximal epigenetic features (Fig. S4 A), which were exemplified by the permissive chromatin regions around the transcription start site (TSS) of inflammatory marker genes of cluster 4 in scRNA-seq analyses of activated monocytes (Fig. S4, B–D). CD127^high feature OCRs were marked by H3K4me1 and H3K27ac, implicating intergenic enhancer-like features (Fig. S4, E and F). Intriguingly, CD127^high feature OCRs displayed permissive states in the resting state, and their accessibility was not further increased by LPS-mediated activation (Fig. S4, G and H), implying potentially stable epigenetic characters. Furthermore, unbiased motif analysis identified STAT5 as the most significantly enriched transcriptional module in CD127^high feature OCRs (Fig. 4 G), suggesting that STAT5 is the likely effector transcription factor that mediates the anti-inflammatory effects of CD127 in human monocytes. Indeed, in line with the observation of IL7R knockdown experiments, inhibiting STAT5 activation in human monocytes by a STAT5 inhibitor resulted in enhanced IL6 and TNF expression (Fig. 4 H). Taken together, CD127 up-regulation conferred activated human monocyte responsiveness to IL-7 and imposed a STAT5-coordinated transcriptional program to restrain inflammation.

To elucidate the mechanisms underlying CD127-STAT5–mediated effects, CD127^high and CD127^low populations were profiled by RNA-seq (Fig. 5 A), revealing that MAF-encoded c-Maf, an established anti-inflammatory transcription factor (Locati et al., 2020; Fig. S4, I and J), was enriched in CD127^high monocytes relative to CD127^low cells (Fig. 5 B). Moreover, analyses of human monocyte c-Maf chromatin immunoprecipitation followed by sequencing (ChIP-seq) datasets (Kang et al., 2017) unraveled preferential deposition of c-Maf in CD127^high feature OCRs with minimal occupancy in CD127^low feature OCRs (Fig. 5, C and D), implying that CD127^high monocytes were characterized by a c-Maf–imposed transcriptional signature. Furthermore, MAF was a direct STAT5 target gene in human monocytes evidenced by binding of STAT5 to a putative IFN-γ activation site (GAS) in the promoter region (Fig. 5 E). Importantly, STAT5 inhibition consistently led to down-regulation of MAF in activated monocytes (Fig. 5 F), supporting a positive role for STAT5 in tuning MAF expression. To gain better understanding of STAT5 functionality, we profiled global STAT5 occupancy in CD127^high monocytes by ChIP-seq (Fig. 5, G and H). In monocytes, STAT5 predominantly bound to intergenic regions (Fig. 5 I), reminiscent of the intergenic enhancer-like CD127^high feature OCRs (Fig. S4, A and B). Indeed, STAT5 occupancy exhibited a strong preference toward CD127^high feature OCRs relative to CD127^low feature OCRs (Fig. 5, J–M), suggesting that STAT5-associated epigenetic features were among the defining characteristics of CD127^high monocytes. Together, the above results implied that the CD127–STAT5 axis exploited the “M2” gene, MAF, to exert anti-inflammatory effects that contributed to the functional heterogeneity in human inflammatory monocytes.

Upon characterization of CD127-imposed monocyte functional heterogeneity in vitro, we next wished to validate these findings in vivo in human disease settings. Profiling synovial monocytes (Fig. S1 B) and peripheral blood monocytes (Fig. S5, A and B) from RA patients with scRNA-seq revealed that IL7R⁺ cells were largely nonoverlapping with the highly inflammatory counterparts and displayed minimal inflammatory features quantitated by inflammatory score and exemplified by expression of representative inflammatory genes (Fig. 6, A and B). Similarly, in COVID-19 BALF monocytes/macrophages, the IL7R⁺ population also did not largely overlap with the highly inflammatory cells and exhibited minimal inflammatory properties (Fig. 6, C and D). Importantly, disease severity inversely correlated with IL7R expression in COVID-19 patients (Fig. 6 E). In mild COVID-19 patients, on average, 44% of BALF monocytes/macrophages were IL7R⁺ cells; yet, this percentage was 18% in severe COVID-19 patients (Fig. 6 F and Fig. S5 C). The differential IL7R patterns in monocytes were not due to the global differences in expression levels, because lymphocytic IL7R did not significantly differ between mild and severe patients (Fig. S5 D). These results indicated that CD127 marked a population of hypoinflammatory monocytes/macrophages in vivo and suggested that the prevalence of IL7R⁺ monocytes/macrophages likely correlated with subdued inflammation and favorable disease outcomes.

Inspired by the similarity of IL7R⁺ cells from different disease conditions, we set forth to extrapolate common features of human inflammatory monocytes/macrophages. Aided by a recently developed in-house deep learning–driven bioinformatics tool specializing in incorporating single-cell sequencing datasets from multiple sources (Xiong et al., 2021,Preprint), we ran integrated analyses of three datasets: COVID-19 BALF monocytes/macrophages, RA synovial monocytes, and in vitro LPS-activated monocytes. Unsupervised clustering identified 10 cellular subsets (Fig. S5 E), with cluster 10 classified as FABP4-expressing alveolar macrophages from the COVID-19 samples (Fig. S5, F–H) and thus being excluded from the subsequent analyses (Morse et al., 2019). The remaining nine clusters (clusters 1–9) represented integrated monocyte subsets presenting in all three tissue sources (Fig. 6 G and Fig. S5 F), with cluster 2 prominently featured by high IL7R expression (Fig. 6, H and I) and sharing a high degree of similarity among three conditions (Fig. 6 J). Common signature genes of cluster 2 revealed a profile that was distinct from the currently known monocyte/macrophage subsets, with IL7R unambiguously identified as the top marker gene (Fig. 6 K). Of note, cluster 2 phenotypically differed from the M2-like cells defined by markers such as CD163 (Fig. S5 I). The above advanced informative analyses revealed unifying features of human inflammatory monocytes, regardless of tissue of origin and disease type.

During the attempts to investigate in situ phenotypes of human monocytes in COVID-19 and RA patients, single-cell transcriptome analyses revealed a subset of inflammatory monocytes positive for CD127, rendering them sensitive to picogram concentrations of IL-7, a (patho)physiological range detectable in human tissue environments, including blood circulation and lung (Ronit et al., 2021; Venet et al., 2012). Thus, in contrast to the IL-7–inert resting monocytes, upon induction of a receptor subunit, a subset of human monocytes gained IL-7 responsiveness under inflammatory conditions, in line with a previous study reporting active IL-7 signaling measured by JAK3 phosphorylation (Musso et al., 1995). Facilitated by modest up-regulation of the IL-7 cytokine in inflammatory disease settings such as COVID-19 (Ronit et al., 2021), CD127^high monocytes actively received environmental IL-7 cues to promote an epigenetically coupled, STAT5-coordiated transcriptional program. Amid the overall inflammatory tissue environments, the CD127–STAT5 axis delivers anti-inflammatory signals, possibly as an attempt to counteract inflammatory signals, to impose functional heterogeneity among human monocytes.

Despite use of various experimental or informatics approaches, we failed to detect CD127 expression in mouse homeostatic and inflammatory monocytes or macrophages, excluding the feasibility of using mice as model organisms for functional and mechanistic investigations. In fact, it has been reported that mouse fetal monocytes express CD127 that is drastically down-regulated in the postnatal stage (Leung et al., 2019). Hence, CD127 expression is silenced in adult mouse monocytes or macrophages, possibly through suppressive epigenetic mechanisms that are irreversible by inflammatory stimuli (Natoli, 2010; Winter and Amit, 2014). Taken together, these results revealed the species-specific nature of monocytic CD127 expression and function.

STAT5 is well recognized to mediate signaling by common γ-chain family cytokines in lymphoid lineages, especially in T cells and ILCs (Leonard et al., 2019). In myeloid lineage cells, STAT5 activation has been conventionally studied in the context of GM-CSF signaling and has been implicated in cell differentiation and survival (van de Laar et al., 2012; Zhan et al., 2019). However, the role of STAT5 in terminally differentiated myeloid cells such as inflammatory monocytes at the sites of infection remains unclear. Our results implied that STAT5-mediated IL-7 signaling imprints hypoinflammatory phenotypes in CD127^high monocytes and thus conveys functional diversity to mature monocyte populations. STAT5 preferentially occupies CD127^high feature OCRs, and a majority of STAT5 binding sites displayed enhancer-like characteristics. Interestingly, STAT5-bound putative enhancers were already permissive in resting monocytes before activation, suggesting plausibly heterogeneous epigenetic landscapes among the seemly homogeneous cell population (Carter and Zhao, 2021). Due to the difficulty of performing fate-mapping studies in the human system, comprehensive understanding of the exact lineage and origin of such predefined STAT5-prone monocytes awaits future investigations aided by new technical tools.

Our understanding of human immune cell heterogeneity has been revolutionized by the recent advances in scRNA-seq (Dress et al., 2020; Papalexi and Satija, 2018). Yet, due to difficulties in distinguishing the biologically meaningful differences from technical and/or experimental variations, it remains challenging to compare and contrast single-cell sequencing datasets from multiple sources, especially in human studies where samples have been obtained from different organs under different disease conditions that bear significant intrinsic variabilities. Here, aided by advanced deep learning–based computing tools, we successfully integrated multiple datasets that we and others generated and constructed the expression atlas of human inflammatory monocytes. Intriguingly, inflammatory monocytes derived from multiple disease conditions and multiple tissues showed remarkable similarity in terms of subset stratification and expression profiles, likely implying unifying mechanisms for monocyte activation in vivo. Importantly, the integrated human monocyte phenotypes could not be predicted based on the existing knowledge or extrapolated from model organism–based studies, signifying the necessity of understanding human diseases with human samples. It would be interesting and desirable to assess whether the phenotypes could be observed in additional human disease settings. Given the unique presence of this population in human inflammatory diseases, especially the correlation of expansion of CD127^high monocytes with favorable disease outcomes in COVID-19, it is highly conceivable to propose CD127^high monocytes and the related CD127/STAT5 pathway as potential therapeutic targets for inflammatory disorders.

PBMCs of anonymous healthy donors were isolated from buffy coats purchased from the Beijing Red Cross Blood Center using density gradient cell separation by Ficoll (Lymphoprep; STEMCELL Technologies) following the protocol approved by the institutional review board of the School of Medicine, Tsinghua University. The private information of anonymous blood donors was inaccessible to investigators. PBMCs of RA patients were obtained from Peking Union Medical College Hospital using the protocol that was approved by the institutional review board of Peking Union Medical College Hospital. CD14⁺ monocytes were further isolated from PBMCs using anti-CD14 magnetic beads (130-050-201; Miltenyi Biotec). CD14⁺ monocytes were cultured in RPMI 1640 medium (10040CM; Corning) supplemented with 10% (vol/vol) FBS (Gibco) and human recombinant M-CSF (300-25, 10 ng/ml; PeproTech). LPS (Escherichia coli O127:B8; Sigma-Aldrich), human recombinant IL-7 (200-07; PeproTech), human recombinant TNF (H8916; Sigma-Aldrich), or chemical inhibitors (SB203580 from Selleck; STAT5 inhibitor from Santa Cruz Biotechnology; and Bay 11-7082 from Sigma-Aldrich) were used as indicated for various experiments.

Two cases of uninfected lung tissues and three cases of COVID-19 lung tissues were from the biobank of Southwest Hospital, Third Military Medical University (Army Medical University). COVID-19 lung tissues were obtained during autopsy of the patients who died of SARS-CoV-2 infection. Pathologically normal lung tissues from pulmonary bulla patients were used as uninfected controls. Tissue collection and the following histological analyses were approved by the ethics committee of Southwest Hospital, Third Military Medical University (Army Medical University), and were carried out in accordance with regulations issued by the National Health Commission of China and the Helsinki declaration. Lung tissue sections were stained with hematoxylin for assessment of pulmonary architecture, and anti-CD68 (ab201340; Abcam) and anti-CD127 (PA5-97870; Invitrogen) antibodies were used for immunohistochemistry. Specifically, lung sections were deparaffinized and rehydrated. Antigen retrieval was performed with the Improved Citrate Antigen Retrieval Solution (Beyotime) and incubated with H₂O₂ in the dark for 15 min to block endogenous peroxidase activity. Slides were blocked with 10% goat serum in Tris-buffered saline for 30 min at room temperature and stained with primary antibodies overnight at 4°C. Slides were washed three times with 0.1% Tris-buffered saline with Tween before incubation with HRP-conjugated secondary antibodies (goat anti-rabbit IgG [H&L]–HRP conjugated, BE0101; Bioeasy [Beijing] Technology Co.; goat anti-mouse IgG [H&L]–HRP conjugated, BE0102, Bioeasy [Beijing] Technology Co.). Stained slides were washed again in PBS and stained with diaminobenzidine (Tiangen) in conjunction with a hematoxylin counterstain (Solarbio Life Science). After dehydration, sections were mounted in neutral balsam.

COVID-19 lung tissues were collected as described above and were washed and fixed overnight at 4°C in a solution of 1% paraformaldehyde in PBS. The tissues were incubated in a solution of 30% sucrose in PBS and a mixture of 30% sucrose and OCT compound 4583 (Sakura Finetek) separately at 4°C overnight. The samples were then embedded in optimal cutting temperature compound, frozen in a bath of ethanol cooled with liquid nitrogen, and stocked at −80°C. Frozen samples were cut at 10-µm thickness and collected onto slides. Slides were dried at 50°C for 30 min and fixed in 1% paraformaldehyde for 10 min and processed for staining. The tissues were permeabilized in PBS/0.3% Triton X-100/0.3 M glycine at 37°C for 30 min and blocked in PBS/5% goat serum at room temperature for 1 h. The tissues were then incubated with the indicated primary antibodies diluted (anti-CD68, 1:100; anti-CD127, 1:500) in PBS/5% goat serum at 4°C overnight and washed in PBS/0.2% Tween-20 at room temperature for 30 min three times. The tissues were incubated with Alexa Fluor dye–conjugated secondary antibodies (Alexa Fluor 488 goat anti-mouse IgG, 1:500, B40941, Life Technologies; Alexa Fluor 555 goat anti-rabbit IgG, 1:500, A27039, Invitrogen) and DAPI (1:200, C0060-1; Solarbio Life Science) in PBS/0.5% BSA at room temperature for 2 h and washed in PBS/0.2% Tween-20 at room temperature for 1 h five times before mounting with SlowFade Diamond Antifade Mountant (S36963; Life Technologies).

Nine COVID-19 patients were enrolled from the Shenzhen Third People’s Hospital. BALF collection from COVID-19 patients and healthy donors and the following studies were conducted according to the principles expressed in the Declaration of Helsinki. Ethical approval was obtained from the research ethics committee of Shenzhen Third People’s Hospital (2020-112). Diagnosis of COVID-19 was based on clinical symptoms, exposure history, chest radiography, and SARS-CoV-2 RNA positivity. Disease severity was defined as moderate, severe, or critical according to the Diagnosis and Treatment Protocol of COVID-19 issued by the National Health Commission of China. Approximately 20 ml of BALF was obtained from each patient. BALF was directly processed within 2 h, and all operations were performed in a biosafety level 3 laboratory. BALF cells were collected, counted, resuspended, and subsequently processed for scRNA-seq library construction as described in our previous study (Liao et al., 2020). According to the clinical diagnosis, nine enrolled patients included three moderate cases, one severe case, and five critical cases. For the subsequent analyses of scRNA-seq data, given that there was only one clinically defined severe case, the patients were stratified into mild (n = 3) and severe (n = 6, including both severe and critical cases) groups.

The laboratory animal facility at Tsinghua University has been accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International, and the institutional animal care and use committee of Tsinghua University approved the protocol used in this study for blood collection from mouse cheeks. C57BL/6J mice were bred and housed in isolated ventilated cages (maximum of six mice per cage) at the specific pathogen–free facility at Tsinghua University. The mice were maintained on a 12-h/12-h light/dark cycle in a 22–26°C, 40–70% humidity atmosphere with access to sterile pellet food and water ad libitum.

Total RNA was extracted from cells using TRIzol reagent according to the manufacturer’s procedure, and total RNA was reverse transcribed to cDNA with Moloney murine leukemia virus reverse transcription (2641B; Takara Bio). Real-time qPCR was performed in duplicates with SYBR Green Master Mix (A25742; Applied Biosystems) on a StepOnePlus thermal cycler (Applied Biosystems). Primer sequences are listed for the following genes: IL7R (forward) 5′-TCC​AAC​CGG​CAG​CAA​TGT​AT-3′, (reverse) 5′-TGA​CCA​ACA​GAG​CGA​CAG​AG-3′; IL2RG (forward) 5′-GTG​CAG​CCA​CTA​TCT​ATT​CTC​TG-3′, (reverse) 5′-GTG​AAG​TGT​TAG​GTT​CTC​TGG​AG-3′; IL6 (forward) 5′-ACC​CCC​AAT​AAA​TAT​AGG​ACT​GGA-3′, (reverse) 5′-TTC​TCT​TTC​GTT​CCC​GGT​GG-3′; TNF (forward) 5′-CCT​CTC​TCT​AAT​CAG​CCC​TCT​G-3′, (reverse) 5′-GAG​GAC​CTG​GGA​GTA​GAT​GAG-3′; MAF (forward) 5′-ACT​GGC​AAT​GAG​CAA​CTC​CG-3′, (reverse) 5′-CAC​TGG​CTG​ATG​ATG​CGG​TC-3′.

Upon the indicated treatment, cells were collected and washed with staining buffer (PBS with 0.5% BSA and 2 mM EDTA). Then, the surface markers were stained with the indicated fluorochrome-conjugated antibodies in 1:400 dilution for 30 min on ice in the dark. After staining, cells were washed three times with staining buffer and resuspended in PBS for analysis in a BD FACSFortessa sorter or for FACS in a BD FACSAria III sorter. Further data analysis was implemented using FlowJo software (BD Biosciences). For intracellular staining, cells were treated with GolgiStop (554724; BD Biosciences) for 4–5 h before collection. The routine staining for surface markers was performed, after which the cells were fixed with 100 µl/tube fixation buffer (420801; BioLegend) for 25 min at room temperature, and the fixed cells were permeabilized and stained in 1× permeabilization wash buffer with fluorochrome-conjugated antibodies for 30 min on ice in the dark. The fixed and intracellularly stained cells were washed twice with 1× permeabilization wash buffer and suspended in PBS for analysis. The fluorochrome-conjugated antibodies for targets of interest and fluorochrome-conjugated isotype control antibodies were as follows: anti-mouse CD3ε antibody PE (100307; BioLegend), anti-mouse CD11b PerCP-Cyanine5.5 (45-0112-82; eBioscience), anti-mouse CD127 PE-Cyanine7 (135013; BioLegend), anti-mouse MHC class II BV421 (48-5321-82; eBioscience), anti-human CD127 antibody PE-Cyanine5 (351323; BioLegend), anti-human CD14 antibody APC-Cyanine7 (301820; BioLegend), anti-human CD3 antibody FITC (317305; BioLegend), anti-human IL-6 antibody PE (501106; BioLegend), anti-human TNF-α antibody PE (502908; BioLegend), anti-human CD16 antibody BV605 (302039; BioLegend), PE-Cyanine7 rat IgG2a, κ isotype control antibody (400521; BioLegend), and PE rat IgG2b, κ isotype control antibody (400607; BioLegend).

Whole-cell lysates were prepared by direct lysis in SDS loading buffer. All samples for immunoblotting were denatured at 95°C for 10 min. For immunoblot analysis, denatured cell lysates were separated by 10% SDS-PAGE and transferred to a polyvinylidene fluoride membrane (EMD Millipore) for probing with specific primary antibodies and HRP-conjugated secondary antibodies. SuperSignal West Pico Chemiluminescent Substrate (34580; Thermo Fisher Scientific) was used for detection. Relative density of blotting bands was quantified using ImageJ (version 1.52a). Antibodies for probing proteins of interest were as follows: anti-GAPDH mouse mAb (BE0023; Bioeasy [Beijing] Technology Co.), anti–β-actin mouse mAb (AC026; Abclonal), anti-phospho-Stat5 (Tyr694) rabbit mAb (9314S; Cell Signaling Technology), anti-Stat5 antibody (9363S; Cell Signaling Technology), anti–c-Maf antibody (sc-518062; Santa Cruz Biotechnology), goat anti-rabbit IgG (H&L)–HRP conjugated (BE0101; Bioeasy [Beijing] Technology Co.), goat anti-mouse IgG (H&L)–HRP conjugated (BE0102; Bioeasy [Beijing] Technology Co.).

The concentrations of IL-7 in culture media were measured by ELISA. The culture media were collected after centrifugation for cell pelleting, and the levels of IL-7 were quantified by using the Human IL-7 Quantikine HS ELISA Kit (HS750; R&D Systems) according to the manufacturer’s instructions.

LS 174T human colon epithelial cells were labeled by CellTrace Violet and were subsequently administered with 200 µM H₂O₂ for 8 h to induce apoptosis. Human monocytes were stimulated with LPS for 6 h before coculturing with apoptotic cells for 1 h. Afterward, cell mixtures were stained with anti-human CD127 antibody (PE-Cyanine5) and anti-human CD14 antibody (APC-Cyanine7), and FACS analyses were performed to identify the CD14⁺ human monocytes with phagocytosed apoptotic cells.

First, CD127^high and CD127^low monocytes were sorted by FACS from human monocytes upon 6-h LPS stimulation. Approximately 10⁶ sorted cells were lysed in 100 µl commercialized Cell and Tissue Lysis Buffer for Nitric Oxide Assay (S3090; Beyotime). Cell lysates were subjected to measurements of NO production by using the Total Nitric Oxide Assay Kit (S0023; Beyotime) according to the manufacturer’s instructions. NO measurements were implemented by nitrate reductase–mediated reduction of nitrate and following colorimetric assay with Griess reagent.

Immediately after isolation, primary human monocytes were nucleofected with On-Target Plus SMARTpool siRNA purchased from Dharmacon Inc. specific for IL7R, MAF, or MAP3K3. Nontargeting siRNA from GenePharma was used as a control. Human Monocyte Nucleofector buffer (V4XP-3024; Lonza) and the Lonza 4D-Nucleofector platform were used according to the manufacturer’s instructions with the human monocyte nucleofection program. The nucleofected monocytes were cultured in RPMI 1640 medium (Corning) supplemented with 10% (vol/vol) FBS (Gibco) and human recombinant M-CSF (20 ng/ml; PeproTech) for 48 h before the following experiments.

For STAT5 ChIP-seq assays, CD14⁺ selected human monocytes were stimulated with LPS (10 ng/ml) for 6 h. Approximately 30 × 10⁶ stimulated cells were first stained for surface CD127 and then fixed in 1% methanol-free formaldehyde (Thermo Fisher Scientific) for 8 min at room temperature followed by quenching with 125 mM glycine for another 5 min. FACS-based cell sorting implemented the isolation of CD127^high monocytes from fixed cells. Approximately 15 × 10⁶ CD127^high monocytes were used for ChIP. The ChIP assay was performed using the SimpleChIP enzymatic ChIP kit (Cell Signaling Technology) according to the manufacturer’s instructions. The DNA–protein complexes were immunoprecipitated using 10 µl per sample of STAT5 antibody (9363S; Cell Signaling Technology). The immunoprecipitated DNA fragments were extracted with the QIAquick PCR purification kit (Qiagen) and subjected to ChIP-seq library construction with NEBNext Ultra II DNA Library Prep kit (E7103S; New England Biolabs). The constructed DNA library was sent for sequencing on an Illumina HiSeq X Ten platform for an average of 20 million unique reads per sample.

For STAT5 ChIP-qPCR assays, THP-1 cells were stimulated with LPS (100 ng/ml) for 6 h and subsequently with IL-7 (10 ng/ml) for 30 min. 10–20 × 10⁶ cells per condition were used for ChIP experiments as described above by using STAT5 antibody (9363S; Cell Signaling Technology) and IgG (2729P; Cell Signaling Technology) control. The immunoprecipitated DNA fragments were extracted with the QIAquick PCR purification kit (Qiagen) and subjected to qPCR assay for enrichment detection in the MAF TSS upstream GAS motif with the following primer pair: (forward) 5′-AAG​TGC​AGT​GCT​ATA​AAG​TTG​TTT-3′ and (reverse) 5′-ATG​TTC​AAG​ACG​CTG​GCT​TA-3′.

Human CD14⁺ monocytes were stimulated with LPS 10 ng/ml for 6 h, and CD127^high and CD127^low populations for each donor were sorted by FACS. Total RNA was extracted from CD127^high and CD127^low cells using TRIzol reagent (Thermo Fisher Scientific) according to the manufacturer’s procedure, and RNA samples were processed for library construction with the TruSeq mRNA-seq Sample Preparation Kit (Illumina) and sequencing by BGI Genomics Co. on a BGISEQ-500RS platform. Three independent sets of paired samples collected from three healthy donors were subjected to RNA-seq and the subsequent bioinformatics analyses.

After the isolation or treatment, cells were frozen in FBS + 10% DMSO for preservation in liquid nitrogen. The frozen cells were processed for scRNA-seq by BGI Genomics Co. Single-cell capturing and downstream library construction were performed using Chromium Single Cell 3′ Reagent kits (10x Genomics) according to the manufacturer’s protocol. The constructed libraries were sequenced on a BGI MGISEQ2000 platform.

Human CD14⁺ monocytes from healthy donors were stimulated with 10 ng/ml of LPS for 6 h, and CD127^high and CD127^low populations for each donor were sorted by FACS. Cells were pelleted by centrifugation for 10 min at 500 g and 4°C using a swing rotor with low acceleration and brake settings. Cell pellets were washed once with 1× PBS, and cells were pelleted again by centrifugation using the previous settings. Cell pellets were resuspended in 50 µl of lysis buffer (10 mM Tris-HCl, pH 7.4, 10 mM NaCl, 3.0 mM MgCl₂, and 0.5% NP-40), and nuclei were pelleted by centrifugation for 30 min at 500 g and 4°C using a swing rotor with low acceleration and brake settings. The supernatant was discarded, and nuclei were resuspended in 50 µl reaction buffer containing 5.0 µl Tn5 transposase and 10 µl 5× TruPrep Tagment Buffer L (TruePrep DNA Library Prep Kit V2 for Illumina; Vazyme Biotech). The reaction was incubated at 37°C for 30 min. After the tagmentation, the transposed DNA fragments were purified by 1× AMPure XP beads (Beckman Coulter). PCR was performed to amplify the libraries for nine cycles using the following PCR conditions: 72°C for 3 min; 98°C for 30 s; and thermocycling at 98°C for 15 s, 60°C for 30 s, and 72°C for 3 min; followed by 72°C for 5 min. After the PCR, libraries were purified with the 0.5× and 1.2× AMPure XP beads. DNA concentrations were measured with the StepOnePlus Real-Time PCR System (Life Technologies), and library sizes were determined using an Agilent 2100 Bioanalyzer. Libraries were sequenced on an Illumina HiSeq X Ten platform for an average of 20 million unique reads per sample. Three independent sets of paired samples collected from three healthy donors were subjected to ATAC-seq and the subsequent bioinformatics analyses.

ATAC-seq paired-end reads were collected. Adapter sequences were trimmed from the ends of reads by Cutadapt (version 1.14), and the reads that failed to pass the quality control (Q > 10) were discarded. STAT5 ChIP-seq paired-end reads were collected at 150 bp in length. The first-end reads were used for analyses and trimmed to 50 bp, and adapter sequences were further trimmed from the ends of reads by Cutadapt (version 1.14). The reads that failed to pass the quality control (Q > 10) were discarded. H3K27ac and H3K4me1 ChIP-seq datasets were downloaded from the National Center for Biotechnology Information Gene Expression Omnibus (GEO) dataset under the GEO accession no. GSE85245 (Novakovic et al., 2016), and MAF ChIP-seq datasets were obtained from GEO accession no. GSE98369 (Kang et al., 2017). The ATAC-seq datasets for resting and LPS 4 h stimulated human monocytes were downloaded from GEO accession no. GSE98369 (Kang et al., 2017) and GEO accession no. GSE120942 (Kang et al., 2019), respectively. Sequence Read Archive files were converted to fastq files using fastq-dump included in the Sequence Read Archive toolkit. Both single-end and paired-end ATAC-seq reads were aligned to the human genome (University of California, Santa Cruz hg38) using Bowtie2 (version 2.2.5; Langmead and Salzberg, 2012) with no more than one mismatch for each alignment seed with 15 bp in length to generate alignment files of uniquely mapped single-end reads or uniquely mapped paired-end fragments with maximum length of 1,000 bp. ChIP-seq reads in fastq files were aligned to the human genome (University of California, Santa Cruz hg38) using Bowtie (version 1.1.2; Langmead et al., 2009) to generate alignment files of uniquely mapped reads with maximum allowed mismatch of 2 (-m 1 -n 2) for each alignment seed. ChIP-seq reads aligned to the genome were extended to 150 bp from their 3′ end for further analyses. RNA-seq data were collected, and single-end reads were aligned to the human genome (hg38) using TopHat (version 2.1.0; Kim et al., 2013) with the parameters --min-segment-intron 50 --no-novel-indels --no-coverage-search, and only uniquely mapped reads were preserved.

The ATAC-seq alignment files for CD127^high and CD127^low monocytes from three donors were used to call peaks for significantly opened chromatin regions using MACS2 (version 2.1.1; false discovery rate <0.05). The peaks from six samples were merged as total OCRs in LPS-treated monocytes. To identify the differentially opened chromatin regions between CD127^high and CD127^low monocytes, the ATAC-seq fragments were counted in each OCR for each sample by using FeatureCounts (version 1.5.0; Liao et al., 2014). Subsequently, the fragment count was normalized to count per million mapped fragments for each sample. The normalized fragment count was used to identify differentially opened chromatin regions by edgeR (version 3.28.1; Robinson et al., 2010). Mean values of (fragment count + 1) fold change (CD127^high/CD127^low) among three donors were log₂ transformed, and differentially opened chromatin regions were identified by log₂-transformed fold changes (CD127^high/CD127^low) ≥1 or less than or equal to −1 for CD127^high monocyte feature OCRs or CD127^low monocyte feature OCRs with cutoff of P < 0.05. The STAT5 ChIP-seq alignment file was used to call peaks for significantly STAT5-enriched regions using MACS2 (version 2.1.1; false discovery rate <0.05). To visualize ATAC-seq and ChIP-seq signals around genome regions of interest, we first counted ATAC-seq fragments (-fragLength given) and ChIP-seq extended reads (-fragLength 150) every 10 bp from the center of OCRs or STAT5 peaks to ±2.5-kb regions for each OCR or STAT5 peak by using the annotatePeaks.pl program in HOMER (version 4.7.2; Heinz et al., 2010). The output counting matrices were visualized by using Cluster TreeView (version 1.2.0) and were further used to generate average signals around OCRs by calculating the average fragments/reads count per bin (10 bp) per OCR.

For coverage of mapped RNA-seq reads in transcripts, the expression level of each gene transcript was calculated as the normalized read count per kilobase of transcript per million mapped reads (fragments per kilobase per million mapped reads [FPKM]) using Cufflinks (version 2.2.1; Trapnell et al., 2012). Differential gene expression between CD127^high and CD127^low monocytes from three donors was identified using DESeq2 (version 1.27.9; Love et al., 2014). Genes with P < 0.05 and mean (FPKM + 1) fold changes (CD127^high/CD127^low) ≥1.5 or ≤0.67 among three donors were defined as highly expressed genes in CD127^high monocytes or highly expressed genes in CD127^low monocytes, respectively. The expression data (gene-specific transcripts per million total transcripts) of published RNA-seq datasets for macrophages and monocytes isolated from mouse lung tissues under LPS-induced pulmonary inflammation were downloaded from the GEO DataSets under accession no. GSE136914 (Sajti et al., 2020).

For scRNA-seq of COVID-19 patients’ BALF cells, the Cell Ranger software suite (version 3.1.0) was used to perform sample demultiplexing, barcode processing, and single-cell 5′ unique molecular identifier (UMI) counting. Specifically, splicing-aware aligner STAR was used in FASTQs alignment. Cell barcodes were then determined based on the distribution of UMI counts automatically, and the gene–barcode matrices were saved for downstream analyses. In addition, one additional healthy control was acquired from the GEO database under accession no. GSE128033. All samples were loaded as Seurat objects by using Seurat (version 3.2.1; Stuart et al., 2019), and quality control for each cell was conducted with the following criteria: gene number between 200 and 6,000, UMI count >1,000, and mitochondrial gene percentage <0.1. All samples were further integrated to remove the batch effects with the parameter settings of the first 50 dimensions of canonical correlation analysis and principal component analysis (PCA). The integrated Seurat project was first normalized, and the top 2,000 variable genes were then identified by using the Seurat analysis pipeline that has been described in our previous studies. Gene expression scaling and PCA were performed using the top 2,000 variable genes. Then, Uniform Manifold Approximation and Projection (UMAP) was performed on the top 50 principal components for visualization, and graph-based clustering was simultaneously performed on the PCA-reduced data with the 1.2 resolution setting. According to the clustering result, 32 clusters were identified, and the annotations for each cluster were implemented based on the expression of marker genes that were used in our previous study (Liao et al., 2020). Monocytes/macrophages (CD14^high CD68^high) were extracted from the total BALF cells after annotation, and reclustering (PCA and UMAP) was performed. The clusters showing expression of both monocyte/macrophage marker genes and T cell marker genes were excluded as doublets. Only the cells from COVID-19 patients were used for downstream analyses of both total BALF cells and monocytes/macrophages.

RA synovial scRNA-seq datasets (Zhang et al., 2019a) were downloaded from ImmPort with the study accession code of SDY998. The reduction and clustering results from the original study were used. The monocyte clusters were extracted for downstream analyses.

The Cell Ranger software suite (version 3.1.0) was used to perform sample demultiplexing, barcode processing, and single-cell 5′ UMI counting, and gene–barcode matrices were generated for LPS-treated human CD14⁺ monocytes and RA patient PBMCs. The gene–barcode matrices were loaded as Seurat objects, and quality control for each cell was performed with criteria for LPS-treated human CD14⁺ monocytes (gene number between 200 and 4,500, UMI count >1,000, and mitochondrial gene percentage <0.15) and RA PBMCs (gene number between 200 and 6,000, UMI count >1,000, and mitochondrial gene percentage <0.1). After quality control, the top 2,000 variable genes were identified, and gene expression scaling, PCA, and UMAP clustering were performed for each dataset. Marker genes for each cluster in each scRNA-seq dataset were identified by using the FindAllMarkers function in Seurat. According to the expression of well-studied PBMC marker genes summarized in the CellMarker website (Zhang et al., 2019b), each cluster of RA PBMCs was annotated with a certain cell type, and monocyte clusters were extracted for downstream analyses.

IL7R⁺ cells were identified based on the normalized expression (>0) of IL7R for each cell in each scRNA-seq dataset. Inflammatory score was calculated based on the normalized average expression of eight inflammatory genes: TNF, IL6, IL8, CCL2, CCL3, CCL4, CCL8, and CXCL10, which is implemented by the AddModuleScore function in Seurat with 100 controls for each inflammatory gene. The inflammatory genes were chosen based on their induction by LPS stimulation and inflammatory properties that have been widely used to profile the characteristics of macrophages or monocytes in diseases, especially RA and COVID-19 (Kuo et al., 2019; Liao et al., 2020). We also precluded the inflammatory genes whose products are subjected to extensive post-transcriptional regulation, such as IL1B. Inflammatory monocytes (infla^high) in each set of scRNA-seq data were identified as the top 20% inflammatory cells based on inflammatory score in monocytes/macrophages from RA synovia, RA peripheral blood, and COVID-19 BALF, respectively. IL7R positivity was prioritized over infla^high to enable the comparison between two distinct IL7R⁺ and infla^high populations in each condition.

Integration of COVID-19 monocytes/macrophages, RA synovial monocytes, and LPS-treated monocytes was implemented by a recently developed SCALEX method based on the asymmetric variational autoencoder framework as shown in Fig. S5 E (Xiong et al., 2021). In brief, the training of SCALEX was to maximize the likelihood of imputed scRNA-seq data to their original ones without batch effect, which could be transformed to maximize the evidence lower bound and expressed with a reconstruction term and a regularization term as shown in Fig. S5 E. The reconstruction term coaches the imputation of data to be similar to their inputs, and the regularization term is a Kullback-Leibler divergence to regularize the latent variable z to a Gaussian mixture model manifold. Leiden clustering and UMAP visualization were performed based on the features extracted by SCALEX for integrated monocytes/macrophages, and marker genes for each cluster were identified. Correlations between clusters among three datasets were calculated as Pearson correlation coefficients, and the negative values were normalized to 0 for heat map presentation.

Types of statistical tests are indicated in the figure legends. Statistical analyses were performed using GraphPad Prism software (GraphPad Software) for t test (Student’s t test), and the Wilcoxon rank-sum test was implemented using R (version 4.0.2). P < 0.05 was considered statistically significant.

Fig. S1 shows the results of clustering analyses for scRNA-seq datasets of COVID-19 patients’ BALF and RA patients’ synovial tissues. Fig. S2 shows the characterizations of elicited CD127 expression in activated human monocytes, including the comparable CD127 positivity in activated monocytes to T cells, dependency on TLR-mediated signaling pathways, commonly up-regulated CD127 expression among CD14-CD16 segmented human monocyte populations, and human-specific nature of such monocytic CD127 expression. Fig. S3 shows the comparable monocyte functions, such as NO production and phagocytosis, between CD127^high and CD127^low monocytes and the functional anti-inflammatory IL-7 receptor signaling in activated human monocytes. Fig. S4 shows CD127 expression coordinated epigenetic landscapes in CD127^high and CD127^low monocytes and MAF-regulated anti-inflammatory expression profile in human monocytes. Fig. S5 presents the results of further analyses for scRNA-seq datasets, including clustering analysis for the scRNA-seq dataset of RA patient PBMCs, profiling CD127 expression according to the disease severity in monocytes/macrophages and lymphoid lineage cells from COVID-19 patient BALF, integrative analysis for monocytes/macrophages from multiple inflammatory conditions, and designating specific monocyte/macrophage subsets via the integrative analysis.

Sequencing datasets are deposited in the GEO with assigned accession numbers as follows: RNA-seq and ATAC-seq (GEO accession no. GSE159118), STAT5 ChIP-seq (GEO accession no. GSE173490), healthy donor scRNA-seq (GEO accession no. GSE159113), RA scRNA-seq (GEO accession no. GSE159117), and COVID-19 BALF scRNA-seq (GEO accession no. GSE145926).

We thank the patients and their families for their contribution to scientific research. We also thank the clinical staff and the COVID-19 Pathology Team at Third Military Medical University and Shanghai Jiao Tong University for performing the autopsy work.

This research was supported by National Natural Science Foundation of China grants (31725010, 31821003, 31991174, and 32030037 to X. Hu), a Ministry of Science and Technology of the People’s Republic of China grant (2020YFA0509100 to X. Hu), Emergency Project from Chongqing Health Commission (2020NCPZX01 to X.-W. Bian), Tsinghua University COVID-19 Scientific Research Program (2020Z99CFZ024 to X. Hu), and funds from Tsinghua-Peking Center for Life Sciences (to X. Hu).

Author contributions: B. Zhang performed experiments and bioinformatic analyses, analyzed and interpreted data, and wrote the manuscript. Yuan Zhang performed experiments and analyzed and interpreted data. L. Xiong and Yuzhe Li assisted with integration of scRNA-seq datasets. Yunliang Zhang performed some of the monocyte experiments and assisted with scRNA-seq data analyses. J. Zhao, H. Jiang, and C. Li assisted with collection of RA patient samples. Y. Liu performed STAT5 ChIP-qPCR experiments. X. Liu, H. Liu, and Y.-F. Ping performed pathological analyses of COVID-19 patient tissue samples. Q.C. Zhang directed the integration of scRNA-seq datasets and provided advice on bioinformatics. Z. Zhang provided COVID-19 BALF scRNA-seq datasets and valuable advice. X.-W. Bian directed pathological analyses of COVID-19 patient tissue samples. Y. Zhao provided RA patient samples and advice on the related experiments. X. Hu conceptualized the project, supervised experiments, analyzed and interpreted data, and wrote the manuscript.