Comparative single-cell analysis reveals IFN-γ as a driver of respiratory sequelae after acute COVID-19

Abstract

Postacute sequelae of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection (PASC) represent an urgent public health challenge and are estimated to affect more than 60 million individuals globally. Although a growing body of evidence suggests that dysregulated immune reactions may be linked with PASC symptoms, most investigations have primarily centered around blood-based studies, with few focusing on samples derived from affected tissues. Furthermore, clinical studies alone often provide correlative insights rather than causal mechanisms. Thus, it is essential to compare clinical samples with relevant animal models and conduct functional experiments to understand the etiology of PASC. In this study, we comprehensively compared bronchoalveolar lavage fluid single-cell RNA sequencing data derived from clinical PASC samples and a mouse model of PASC. This revealed a pro-fibrotic monocyte-derived macrophage response in respiratory PASC, as well as abnormal interactions between pulmonary macrophages and respiratory resident T cells, in both humans and mice. Interferon-γ (IFN-γ) emerged as a key node mediating the immune anomalies in respiratory PASC. Neutralizing IFN-γ after the resolution of acute SARS-CoV-2 infection reduced lung inflammation and tissue fibrosis in mice. Together, our study underscores the importance of performing comparative analysis to understand the cause of PASC and suggests that the IFN-γ signaling axis might represent a therapeutic target.

INTRODUCTION

Three years after the onset of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic, antiviral therapies and vaccines have substantially improved the management of acute coronavirus disease 2019 (COVID-19). However, more than 60 million individuals globally are now facing postacute sequelae of SARS-CoV-2 infection (PASC), characterized by persistent, recurring, or new symptoms manifesting after the resolution of acute infection (1). Because the lung is the primarily affected organ during acute infection, respiratory PASC (R-PASC), including disabling symptoms like dyspnea, cough, and interstitial lung disease, are particularly concerning (1). Furthermore, impaired gas-exchange functions may lead to systemic symptoms such as exertional dyspnea and chronic fatigue due to chronic hypoxia, with some patients experiencing these issues for more than 2 years, especially after severe acute COVID-19 (2–4).

Emerging evidence links prolonged or aberrant peripheral immune responses with multiorgan symptoms observed in PASC, mainly through peripheral blood mononuclear cells (PBMCs) and plasma analysis (5, 6). The immune status within affected organs during PASC remains largely uncharacterized because of limited tissue sampling. Furthermore, observational clinical research limits the ability to identify causal mechanisms because most clinical studies can only provide associations. A few animal models of COVID-19 sequelae have been developed (7, 8), although it is not clear whether these models can faithfully model the pathophysiology and molecular etiology of human R-PASC. Thus, there is an urgent need for a comprehensive comparative analysis of clinical and animal respiratory samples to better understand PASC etiology. In addition, developing therapeutic interventions necessitates functional studies in clinically relevant animal models.

To address these challenges, we performed single-cell RNA sequencing (scRNA-seq) of bronchoalveolar lavage (BAL) cells and PBMCs from COVID-19 convalescent individuals with or without R-PASC. In parallel, we performed scRNA-seq analysis of BAL samples from SARS-CoV-2–infected mice. Comparative scRNA-seq analyses of human and animal data revealed that R-PASC are associated with aberrant responses and interactions of macrophages and T cells resident in the respiratory tract. We further identified interferon-γ (IFN-γ) as a mediator driving T cell–macrophage interactions. Neutralizing IFN-γ after the resolution of acute SARS-CoV-2 infection reduced chronic pulmonary inflammation and fibrosis in mice. Our findings thus establish a foundation for identifying therapeutic interventions for R-PASC.

RESULTS

scRNA-seq reveals altered pulmonary immune landscape in R-PASC

We evaluated a cohort of 11 COVID-19 convalescent donors and 4 non–COVID-19 control donors. All convalescent donors were discharged and evaluated 60 to 90 days postinfection (dpi) and were polymerase chain reaction (PCR) negative for SARS-CoV-2 before recruitment (9) (Fig. 1A). Pulmonary functions, including FVC (forced vital capacity), FEV1 (forced expiratory volume in 1 s), and DLCO (diffusing capacity for carbon monoxide), were assessed. Convalescent donors with either FEV1 or FVC values below 80% of predicted were categorized as patients with R-PASC (7 of 11 donors); the rest were labeled as non–R-PASC (Fig. 1B and Table 1). Clinical parameters during hospitalization between the non–R-PASC and R-PASC groups were comparable; however, FEV1%, FCV%, and DLCO% decreased in the R-PASC group during sampling (Fig. 1C).

Fig. 1. Lung function status and BAL cell type components are altered in R-PASC.

(A) Experimental workflow: Lung function parameters were tested on non–COVID-19 or COVID-19 convalescent individuals (N = 15), and scRNA-seq was performed on PBMCs and BAL cells from donors (N = 11). (B) Three-dimensional distributions of lung function. (C) Summary of donor’s clinical data during the acute phase and lung function results during sampling at the indicated day after COVID-19 diagnosis. (D) Uniform Manifold Approximation and Projection (UMAP) plot showing integrated BAL cells. (E) Bar graphs showing the proportion of indicated cell types in BAL cells among each group. MoAM, monocyte-derived alveolar macrophage; TRAM, tissue-resident macrophage; ProAM, proliferating alveolar macrophage. Data are represented by means ± SEM or individual samples. Significance was tested by t test (C) or one-way ANOVA with Tukey’s adjustment (E); ns, not significant; *P < 0.05, **P < 0.01, and ***P < 0.001.

Table 1. Patient description and pulmonary function test.

Inclusion criteria included age between 60 and 80 with no evidence of preexisting interstitial lung disease or any prior chronic lung disease (COVID-19 convalescent cohort) and the absence of lung infiltrate, fever, or any signs of acute infection at the time of bronchoscopy. Individuals with mild chronic obstructive pulmonary disease (COPD) with FEVl > 80% predicted and FEVl/FVC < 0.7 were still eligible for enrollment. Exclusion criteria included (i) previous lung disease, including interstitial lung disease, pulmonary fibrosis, or any other chronic lung disease except for mild COPD; (ii) inability to provide consent to participate in the study; (iii) being under guardianship or curatorship; (iv) active cigarette smoking, vaping, or other inhalation use; and (v) immunocompromised host status. ICU, intensive care unit; HFNC, high-flow nasal cannula; FiO₂, fraction of inspired oxygen; n/a, not applicable.

Participant iD	Sex	Age	Hospitaiization after COVlD-19 diagnosis	Maximum O2 need (liters/min)	ICU stay (days)/mechanical ventilation (days)	Pulmonary embolism	Secondary bacterial pneumonia	Other complications (type)	Date of bronchos-copy after COVID-19 diagnosis	FEV1	FVC	DLCO	Grouping	scRNA-seq
CON-1	M	65	n/a	n/a	n/a	n/a	n/a	n/a	n/a	112/3.54/2.33	116/4.80/3.10	60/15.1/18.6	Non-COVID-19	No
CON-3	F	77	n/a	n/a	n/a	n/a	n/a	n/a	n/a	124/2.35/1.35	124/3.07/1.76	89/15.9/13.4	Non-COVID-19	Yes
CON-4	M	73	n/a	n/a	n/a	n/a	n/a	n/a	n/a	114/3.16/1.98	116/4.24/2.69	109/24.9/16.7	Non-COVID-19	Yes
CON-5	M	73	n/a	n/a	n/a	n/a	n/a	n/a	n/a	119/3.30/1.97	126/4.59/2.68	72/16.4/16.4	Non-COVID-19	No
CVD-01	M	64	Days 5–14 and 19–22	12 liters	7/−	Yes	No	No	Day 74	72/2.54/2.62	70/3.23/3.49	56/15.5/20.7	R-PASC	Yes
CVD-02	M	83	Days 10–18	3 liters	−/−	No	No	No	Day 74	85/2.14/1.73	76/2.60/2.44	75/16.3/15.7	R-PASC	Yes
CVD-03	M	63	Days 5–26	HFNC 60 liters	11/CPAP	Yes	No	No	Day 74	68/2.39/2.61	64/2.95/3.46	45/12.5/20.5	R-PASC	Yes
CVD-04	M	65	Days 3–14	HFNC 60 liters	8/+5	No	Yes	No	Day 88	81/2.42/2.20	75/2.88/2.90	92/21.8/17.6	R-PASC	No
CVD-05	M	74	Days 6–16	HFNC 60 liters	3/−	Yes	No	No	Day 74	78/2.08/1.90	69/2.45/2.59	46/10.2/16.2	R-PASC	No
CVD-06	F	66	Days 5–10	5 liters	−/−	No	No	No	Day 87	127/3.18/1.85	129/4.17/2.40	87/18.8/16.2	Non-R-PASC	Yes
CVD-07	M	62	Days 5–10	2 liters	−/−	No	No	No	Day 84	112/4.06/2.70	99/4.66/3.58	99/28.0/21.1	Non-R-PASC	Yes
CVD-08	F	76	Days 9–23	HFNC 40 liters	−/−	No	No	No	Day 80	83/1.64/1.40	77/1.99/1.83	44/8.2/13.9	R-PASC	Yes
CVD-09	F	63	Days 6–14	HFNC 60 liters	5/−	No	No	No	Day 78	53/1.36/1.77	52/1.69/2.26	60/11.9/14.9	R-PASC	Yes
CVD-10	M	68	Days 1–13	HFNC 50 liters	3/−	No	No	No	Day 67	119/4.09/2.50	114/5.17/3.39	81/22.2/20.3	Non-R-PASC	Yes
CVD-11	F	64	Days 2–21	50% FiO2	11/+9	No	No	No	Day 64	93/4/2	87/1/2.4	73/18.8/14.2	Non-R-PASC	Yes

We conducted scRNA-seq on fresh BAL and PBMCs. In total, 85,971 BAL cells and 101,296 PBMCs were analyzed from five of seven R-PASC, four non–R-PASC, two of four control, and 10 published control individuals (data file S1) (10, 11). Whereas SARS-CoV-2 viral mRNA expression was observed in acute COVID-19 BAL cells (12), it was undetectable in the BAL cells of convalescent individuals (fig. S1A). After integration, nine BAL and 10 PBMC cell clusters were identified (Fig. 1D and fig. S1). In comparison with noninfected controls and non–R-PASC individuals, R-PASC individuals displayed an increased monocyte-derived alveolar macrophage (MoAM) and T cell proportion in the BAL (Fig. 1E). Conversely, the tissue-resident alveolar macrophage (TRAM) diminished in patients with R-PASC (Fig. 1E).

Non–R-PASC BAL cells and PBMCs up-regulated pathways like interleukin-2 (IL-2) signaling, KRAS (KRAS proto-oncogene, GTPase) signaling, and glycolysis compared with controls (fig. S2, A and B). R-PASC BAL cells, however, highlighted pathways associated with cell proliferation and inflammation (fig. S2A). TRAM characteristic genes were highly expressed in non–R-PASC BAL cells (PPARG, FABP4, and MARCO). Respiratory CD8⁺ T cell–related genes were prevalent in R-PASC BAL cells (NKG7, CCL5, GZMK, and CXCR6) (fig. S2C), consistent with reports (9, 13). Moreover, proinflammatory monocyte-related genes were observed in the R-PASC PBMCs (fig. S2D). Gene set enrichment analysis (GSEA) illustrated that pathways prominent in non–R-PASC BAL cells revolved around AM-driven tissue homeostasis, whereas R-PASC BAL cells and PBMCs were enriched with pathways linked to tissue reactivity and inflammation (fig. S2, E and F). The abundant BAL MoAM and PBMC monocytes, coupled with decreased TRAM, correlated with compromised lung function in individuals with R-PASC (fig. S2, G and H). Although spike protein–specific BAL B cells and immunoglobulin G (IgG) were observed in convalescent individuals (9), IgG titers did not correlate with B cell counts (fig. S2I), indicating that BAL B cells may not be the major source of virus-specific IgG at the time of sampling. Together, these data suggest that R-PASC is characterized by altered immune cell composition and inflammatory responses in the respiratory tract.

Profibrotic monocyte-derived macrophages accumulate in R-PASC BAL

Analysis of scRNA-seq showed that R-PASC BAL T cells up-regulated terminal differentiation features like KLRG1 and GZMK, and B cells from R-PASC BAL showed differentially expressed genes (DEGs) compared with non–R-PASC counterparts (fig. S2, J and K). Subsequent analysis focused primarily on macrophages, which were suggested to be associated with tissue fibrosis development in animals (14). To gain a higher resolution of BAL macrophages, we subclustered macrophages into seven subclusters, including three TRAM (TRAM_1, TRAM_2, and TRAM_3), two proliferating AM (ProAM_1 and ProAM_2), and two MoAM (MoAM_1 and MoAM_2) clusters (Fig. 2A). Among the two MoAM clusters, MoAM_1 highly expressed APOE, CD14, and FCN1, indicating a transitional state from monocytes to macrophages. It was marked by high expression of alarmins (S100A8), inflammatory chemokines (CCL2), and chemokine receptors (CCR2 and CXCR4) (Fig. 2B). In addition, the MoAM_1 cluster was associated with inflammatory responses, macrophage activation, cytokine production, and pathogen phagocytosis (Fig. 2C). The MoAM_1 cluster also highly expressed SPP1, encoding osteopontin, a protein observed in macrophages across diverse pathologies, implicated as a pivotal factor in tissue damage and fibrosis (15). The MoAM_2 cluster expressed CD274 (encoding PD-L1), CD40, and ICAM1, primarily linking to cytokine-mediated signaling (Fig. 2, B and C). Three TRAM clusters were characterized by high expression of FBP1, FABP4, CD68, and MARCO (Fig. 2B). They were enriched with cholesterol synthesis regulation, wound healing, and lipid metabolism pathways (Fig. 2C). Furthermore, two ProAM populations were characterized by expressing cell cycle–related genes (MKI67 and NUSAP1; data file S2).

Fig. 2. Proinflammatory and profibrotic MoAMs accumulate in R-PASC.

(A) UMAP plot showing BAL macrophage populations. (B) Heatmap of signature gene expression for each indicated cluster. (C) Representative pathways enriched in each BAL macrophage cluster. (D) Pseudo-time analysis modeling BAL macrophage differentiating. (E) Contour plots showing the density of cell clusters for the indicated groups, with red indicating the area of cell enrichment. (F) Proportions of TRA M_1 and MoAM_1 clusters in BAL macrophages among the indicated groups. (G) Heatmap of predicted transcriptional regulators between TRA M_1 and MoAM_1 clusters based on DEGs. (H) Differential pathways enriched in MoAM_1 and TRA M_1 cells; NES, normalized enriched score. (I) Relative expression of pro–pulmonary fibrotic macrophage signature genes in BAL macrophage clusters. The dashed line indicates the median score of TRA M_1. (J) Enrichment of pulmonary fibrosis–related macrophage gene sets (17, 18) between non–R-PASC and R-PASC MoAM_1 cells. (K) Correlation of BAL MoAM_1 and PBMC monocyte percentages. (L) Module score of peripheral blood monocyte feature gene expression in MoAM_1 cells from the indicated groups. Data are represented by means ± SEM (bar plots), distribution (violin plots), or distribution with interquartile range (boxplot). Significance was tested by one-way ANOVA with Tukey’s adjustment (F), Simple linear test (K), or Wilcoxon test (L). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Dynamic gene change analysis indicated that MoAM_1 cells likely differentiated along a trajectory toward TRAM_1 (Fig. 2D). When compared with control or non–R-PASC donors, BAL cells from R-PASC individuals showed an increase in MoAM_1 and decreased TRAM_1 cells (Fig. 2, E and F, and fig. S3A). Furthermore, MoAM_1 abundance negatively correlated with lung function recovery (fig. S3B). Consistently, flow cytometry revealed that CD206⁺ AMs from R-PASC BAL up-regulated CD14, indicating retained monocytic features (fig. S3, C and D). Furthermore, R-PASC AMs down-regulated macrophage receptors with collagenous structure (MARCO) protein expression, consistent with scRNA-seq analysis (fig. S3, C to E). Relative to TRAM_1, DEGs of MoAM_1 cells were modulated by hypoxia-inducible factor 1 subunit alpha, von Hippel–Lindau tumor suppressor, and proinflammatory cytokine signaling [mediated by transcription factors such as signal transducer and activator of transcription 3 and 1 (STAT1 and STAT3), interferon regulatory factor 1 (IRF1), and nuclear factor kappa B subunit 1 (NF-κB1)] (Fig. 2G), aligning with IFN-γ, IL-6, and tumor necrosis factor–α (TNF-α) signaling enrichment within MoAM_1 cells (Fig. 2H).

MoAMs were reported to adopt a profibrotic phenotype during COVID-19 acute respiratory distress syndrome (ARDS) (16). However, it remains uncertain whether accumulated MoAMs in patients with R-PASC retain these characteristics after ARDS resolution. We thus evaluated the module score of a pro–pulmonary fibrosis macrophage core gene set (17–19) in BAL macrophages (data file S3). MoAM_1 consistently scored highest (Fig. 2I and fig. S4A). In addition, the R-PASC individual–derived MoAM_1 subset displayed enrichment of pro–pulmonary fibrosis macrophage features (Fig. 2J and fig. S4B).

The heightened expression of the monocyte chemoattractant CCL2 in MoAM_1 cells indicates a potential feedback loop where CCL2 produced by MoAM_1 cells enhances circulating monocyte recruitment into the respiratory tract, resulting in their subsequent differentiation into additional MoAMs (fig. S4, C to E). The positive correlation between BAL MoAMs and circulating monocytes further underscores this hypothesis (Fig. 2K). However, no correlations were observed among circulating T cells, NK cells, and their respiratory equivalents (fig. S4F). R-PASC MoAM_1 cells exhibited strong transcriptional similarity to circulating monocytes, implying MoAM recruitment from the circulation (Fig. 2L and fig. S4G). Collectively, our scRNA-seq analysis reveals that R-PASC is associated with pulmonary macrophage dysregulation, with increased proinflammatory and profibrotic MoAM presence.

Resident T cell–derived IFN-γ promotes MoAM recruitment and differentiation in R-PASC

BAL macrophages from R-PASC individuals exhibited an enrichment of inflammatory cytokine responses, including IFN-γ response, IL-6 signaling, and TNF-α signaling. In contrast, lipid metabolic pathways like adipogenesis and cholesterol hemostasis were evident in macrophages from non–R-PASC individuals. The up-regulated lipid metabolic genes including PPARG, SCD, FABP4, and FABP5 are features of TRAMs (fig. S5A and data file S2). Focusing on the MoAM_1 cells, there were more DEG counts between the R-PASC and non–COVID-19 than counts between non–R-PASC and non–COVID-19 (Fig. 3A). Direct comparison of MoAM_1 cells from R-PASC and non–R-PASC individuals showed the up-regulation of the TRAM-associated genes FABP4, PPARG, and MARCO in non–R-PASC donors. In contrast, MoAM_1 cells from patients with R-PASC exhibited increased expression of inflammatory chemokines (CCL2 and CXCL10), inflammatory regulators (FCN1, S100A12, and S100A8), SPP1, and APOE (a MoAM marker) (Fig. 3B). These data suggest that MoAM_1 cells from the R-PASC group differentiated less toward TRAMs, maintaining a more proinflammatory status.

Fig. 3. MoAMs from patients with R-PASC show increased IFN-γ responsiveness.

(A) DEG counts in MoAM_1 cells from the indicated comparisons. (B) DEGs between non–R-PASC and R-PASC MoAM_1 cells; genes that average log₂ fold change (FC) > 0.25 and adjusted P value < 0.05 are highlighted. (C) Transcriptional regulator prediction in MoAM_1 cells from the indicated groups. Red, interferon-stimulated factors. (D) Enriched pathways in non–R-PASC and R-PASC MoAM_1 cells. Red, pathways with the lowest NES. (E) Scatter plots showing the relative module score of IFN-γ–responsive gene expression and M1 (left) or M2 (right) macrophage feature gene expression in MoAM_1 cells from the indicated groups; Pearson’s correlation score between the two features is displayed. (F) Macrophage cluster proportion in lung tissue from the indicated groups (GSE224955). (G) Relative module score of IFN-γ responsiveness signature in MoAMs from the indicated groups. (H) Relative module score of pro–pulmonary fibrotic macrophage signature in MoAMs from the indicated groups. The dashed line indicates the median score of non–COVID-19. Data are represented by means ± SEM (bar plots) or distribution (violin plots). Significance was tested by Wilcoxon test (G and H) or one-way ANOVA with Tukey’s adjustment (F); *P < 0.05 and **P < 0.01.

After reconstructing the gene regulatory networks, we observed that the MoAM_1 cells in the R-PASC group showed increased STAT1, IRF1, IRF7, and NFκB2 activity (Fig. 3C), consistent with enriched proinflammatory gene sets (Fig. 3D). Moreover, a prominent trait of R-PASC MoAM_1 cells was the enriched IFN-γ response (Fig. 3D). Elevated IFNGR2, a critical determinant for IFN-γ responsiveness (20), further indicated an up-regulated IFN-γ response in R-PASC patient–derived MoAM_1 cells (fig. S5B). In addition, IFN-γ–responsive MoAM_1 cells demonstrated an elevated propensity for pro–pulmonary fibrosis in patients with R-PASC (fig. S5C). Macrophage polarization gives rise to distinct proinflammatory (M1) or profibrotic (M2) gene expression profiles (21). In general, M2 macrophages are critical in lung fibrosis compared with M1 macrophages (22). We observed MoAM_1 cells up-regulating both M1 differentiation features while gaining profibrotic and IFN-γ responsiveness features. Conversely, the M2 features of MoAM_1 cells were low (Fig. 3E and fig. S5D). Furthermore, R-PASC MoAM_1 cells had the highest M1 score and a comparable M2 score compared with non–R-PASC or non–COVID-19 (fig. S5E). Together, these data indicate that R-PASC MoAM_1 cells are relatively M1-polarized. Similar IFN-γ response enrichment was discerned in other lung and circulatory cell types (fig. S5, F to I), suggesting a widespread IFN-γ response across many respiratory cell types in individuals with R-PASC.

To explore the altered immune status in lung tissues of individuals with R-PASC, we analyzed a lung scRNA-seq dataset from patients with extensive lung fibrosis (PASC-PF) requiring lung transplantation (fig. S6A) (23). We observed markedly increased Mono1 and MoAMs and diminished TRAM2 and TRAM3 in PASC-PF lungs (Fig. 3F and fig. S6B). This shift mirrors the macrophage composition alterations observed in BAL from our cohort. Moreover, MoAMs from PASC-PF lungs displayed increased IFN-γ responsiveness, enhanced profibrotic characteristics, and biased M1 differentiation (Fig. 3, G and H, and fig. S6).

Compared with controls and non–R-PASC individuals, individuals with R-PASC had more IFNG-expressing BAL cells, predominantly T cells (Fig. 4, A and B, and fig. S7A). Of importance, BAL IFNG-expressing T cell proportions were negatively associated with lung function recovery (Fig. 4C and fig. S7B). We also observed increased CD4⁺ conventional and CD8⁺ T cells in R-PASC BAL (fig. S7, C and D), with increased IFNG expression across both subsets (fig. S7E). Furthermore, IFNG-expressing T cell subsets exhibited tissue-resident characteristics (Fig. 4, D to F, and fig. S7F). Consequently, compared with non–R-PASC donors, individuals with R-PASC appeared to have elevated BAL IFN-γ concentrations and IFN-γ–producing CD4⁺ and CD8⁺ T cells upon antigen stimulation (fig. S7, G to I). Further analysis revealed that BAL CD69⁺CD103⁺ T cells can produce IFN-γ in response to SARS-CoV-2 antigen stimulation (fig. S7, J and K), although the limited sample size prevents drawing a firm conclusion. In contrast, type 2 and type 3 cytokines associated with M2 macrophage polarization were almost undetectable among all donors (fig. S7L). In addition, T cells with increased tissue-resident characteristics were the major cellular source of IFNG in the lung tissue during PASC-PF (Fig. 4, G and H, and fig. S6D). Thus, evidence for IFN-γ–mediated pulmonary T cell–macrophage communication was observed in both R-PASC cohorts.

Fig. 4. Respiratory tract T cells are the major source of IFN-γ during R-PASC.

(A) Violin plot showing IFNG mRNA expression in whole BAL cells from the indicated groups. (B) Pie chart showing percentages of cell types within total IFNG-expressing BAL cells. (C) Correlation of IFNG high-expressed cell proportion with lung functional parameters. (D) UMAP of BAL T cells. (E and F) Tissue-resident T cell core genes module score (E) and IFNG mRNA expression (F) of BAL T cell subclusters. (G and H) Tissue-resident T cell core genes module score (G) and IFNG mRNA expression (H) in lung T cells from the indicated groups. (I) Experimental design of MoAM-like cells in vitro differentiation. (J) CC L2 concentration in the culture medium from IFN-γ–treated or vehicle (Veh)–treated human MoAM-like cells. Data are represented by distribution (violin plots) or individual samples. Significance was tested by paired t test (J) or R package corrplot (C); ***P < 0.001.

R-PASC MoAMs overexpressed CCL2 (fig. S4E), suggesting augmented monocyte recruitment. To determine a role for IFN-γ, we differentiated MoAM-like cells from human peripheral monocytes after treatment with transforming growth factor–β (TGF-β), granulocyte-macrophage colony-stimulating factor (GM-CSF), and rosiglitazone in vitro (Fig. 4I) (24). Compared with monocytes, MoAM-like cells displayed increased expression of AM markers (such as CD169 and CD68) and MoAM genes (SPP1, APOE, and FN1) (fig. S8). After exposure to recombinant human IFN-γ (10 ng/ml), circulating monocytes showed no changes in CCL2 expression; however, MoAM-like cells exhibited elevated CCL2 secretion compared with untreated cells (Fig. 4J and fig. S8). These data suggest that an IFN-γ–abundant microenvironment amplifies CCL2 production by MoAMs, boosting or sustaining monocyte recruitment to the respiratory tract in R-PASC.

Aged C57BL/6J mice manifest pulmonary sequelae after acute SARS-CoV-2 infection

To gain insight into the underlying mechanisms of R-PASC, we leveraged a mouse-adapted strain of SARS-CoV-2, MA10 (25, 26). In humans, advanced age is a known risk factor for developing chronic respiratory sequelae after COVID-19 (27). To understand age-dependent susceptibility to SARS-CoV-2 infection, we infected both young (3-month-old) and aged (21-month-old) female mice with 5 × 10⁴ plaque-forming units (PFU) of SARS-CoV-2 MA10 (Fig. 5A). Infection of young mice resulted in an approximately 15% weight loss followed by rapid recovery, whereas aged animals experienced more substantial and prolonged weight loss (Fig. 5B). Furthermore, young mice did not succumb to the infection, whereas 40% of aged mice did (Fig. 5C), consistent with the phenotypes observed in SARS-CoV-2–infected 1-year-old BALB/c mice (7, 25).

Fig. 5. SARS-CoV-2 infection in aged C57BL/6 mice induces persistent lung sequelae after viral clearance.

(A) Schematic overview of SARS-CoV-2 MA10 infection experimental design. mo, months. (B and C) Weight loss (B) and survival (C) of young and aged mice. Data are pooled from two independent experiments. (D) Viral titers in the BAL were determined. Symbols represent individual mice. The dashed line indicates the detection limit. (E) Representative images of histopathology. H&E, hematoxylin and eosin staining. Masson’s trichrome (TRI-MA) staining highlights fibrotic collagen deposition. Scale bars, 2.5 mm (for the whole lung lobe) and 100 μm (for the zoomed area). (F) Quantification of inflamed area percentage in the lungs at indicated time points. (G to M) Total cell counts (G), Ly6C^hi monocyte counts (H), Siglec-F^lo AM counts (I), CD8⁺ T cell counts (J), CD4⁺ T cell counts (K), CD8⁺ CD69⁺ CD103⁺ T_RM cell counts (L), and CD4⁺ CD69⁺ CD103⁺ T_RM cell counts (M) in BAL of MA10-infected young and aged mice at the indicated time points (n = 5). Data are represented by means ± SEM [error bar: (B), (D), and (F); shading: (G) to (M)]. Significance was tested by two-way ANOVA with Tukey’s adjustment for multiple comparisons (D, F, and G to M); *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

To elucidate the age-associated disease progression, we collected tissues at various time points in both the acute and postacute infection phase and assessed lung pathology and immune responses in the respiratory tract (Fig. 5A). At 3 dpi, lung viral loads in aged mice increased about fivefold compared with young mice, although they were largely undetectable by 10 dpi (Fig. 5D). Of interest, at 21 dpi, the viral N gene was still detectable in lung tissues, although the N protein was not stained through immunofluorescence (fig. S9, A and B). The lung pathology in young mice peaked at 10 dpi and essentially recovered at 35 dpi (Fig. 5, E and F, and fig. S9). In contrast, aged mice displayed more severe lung pathology with considerable lung inflammation at 21 dpi, when the infectious virus had been completely cleared. Lung pathology in aged mice remained noticeable at 35 dpi (Fig. 5, E and F). Similarly, heightened collagen deposition was also evident in the aged mice at 21 dpi (Fig. 5E). These data suggest that aged C57BL/6J mice manifest pulmonary inflammatory and fibrotic sequelae after acute SARS-CoV-2 infection.

We next compared immune cell recruitment in the BAL of young and aged mice. MA10 infection triggered a rapid cell infiltration into the respiratory tract; however, aged mice exhibited a higher cell count at 7 dpi (Fig. 5G). In addition, elevated neutrophils and inflammatory monocytes were seen in infected aged mice (Fig. 5H and fig. S10). Previous studies showed that influenza infection can lead to Siglec-F^hi TRAM reduction, subsequently instigating the emergence of Siglec-F^lo MoAMs (28). Our observations similarly demonstrated Siglec-F^hi TRAM reduction in both groups after SARS-CoV-2 infection, with aged mice that showed a more pronounced deficit in TRAMs and a greater prevalence of MoAMs in comparison with young mice (Fig. 5I and fig. S10). Moreover, we observed more CD8⁺ and CD4⁺ T cells in the BAL of aged mice (Fig. 5, J and K, and fig. S10). Although overall spike protein–specific CD8⁺ T cell counts were higher because of increased total lung CD8⁺ T cell numbers in aged mice, the proportion was largely reduced compared with that in young mice, indicating a more prominent nonspike protein–specific CD8⁺ T cell response in aged mice (fig. S10D). Consistent with findings from patients with R-PASC, increased CD69⁺CD103⁺ tissue–resident T cells were observed in BAL of aged mice (Fig. 5, L and M, and fig. S10). Together, these data indicate that a post–SARS-CoV-2 pulmonary sequelae model developed in aged C57BL/6 mice is characterized by respiratory cell profiles similar to those in individuals with R-PASC.

scRNA-seq analysis of respiratory immune cells in mouse R-PASC

We next sought to understand the cellular and molecular profiles associated with this R-PASC mouse model. scRNA-seq analysis identified eight cell populations in BAL cells from both young and aged C57BL/6J mice at 0, 10, 21, and 35 dpi (Fig. 6A and fig. S11A). Kinetic analysis revealed increased T cell and MoAM accumulation, which was coupled with a decrease in TRAMs in aged mice at 21dpi (fig. S11B).

Fig. 6. Comparative scRNA-seq analysis reveals similarities in respiratory immune profiles between aged mice infected with SARS-CoV-2 and individuals with R-PASC.

(A) The UMAP of BAL cells from young and aged C57BL/6J mice at 0, 10, 21, and 35 dpi. (B) Heatmap showing the top 50 variable genes in BAL cells. The three genes shown correlated with lung pathology and were consistent with the human scRNA-seq results. (C) Pathways enriched in BAL cells from indicated time points compared with day 0. EMT, endothelial-mesenchymal transition. Red, shared pathways for aged mice at 21 dpi and individuals with R-PASC. (D) Stacked bar plots showing the proportion of indicated BAL cell types in human and infected mice. (E) The UMAP of BAL macrophages. (F) Gene expression in the murine BAL macrophages. (G) The relative module score of human R-PASC MoAM_1 features in young or aged C57BL/6J MoAM1s at 21 dpi. (H) Pathways enriched in mouse MoAM1s and TRA Ms at 21 dpi. Shading indicates enrichment in TRA Ms (purple) or MoAM1s (teal). Red, shared pathways for aged mice at 21 dpi and individuals with R-PASC. (I) Relative score of pro–pulmonary fibrotic macrophage signatures in MoAM1s. (J) Enriched pathways in MoAM1s from young or aged C57BL/6J mice at 21 dpi. Shading indicates enrichment in the young (purple) or aged (blue) samples. Red, shared pathways for aged mice at 21 dpi and R-PASC individuals. (K) IFN-γ responsiveness gene features in MoAM1s from young or aged C57BL/6J mice at indicated time points. (L) Scatter plots showing IFN-γ responsiveness signatures and M1 (left) or M2 (right) macrophage features in MoAM1s from the indicated groups at 21 dpi. Pearson’s correlation score between the two features is displayed. (M) Scatter plot showing IFN-γ responsiveness and pro–pulmonary fibrosis features in MoAM1s from the indicated groups at 21 dpi. Pearson’s correlation score between the two features is displayed. (N) Cell counts of IFN-γ–producing BAL CD8⁺ and CD4⁺ T cells from young or aged mice at indicated time points after phorbol 12-myristate 13-acetate and ionomycin stimulation. Data represent the means ± SEM (curve plots) or distribution (violin plots). Data were analyzed by two-way ANOVA (K and N); **P < 0.01 and ****P < 0.0001.

DEG profiles between young and aged mice were pronounced at 21 dpi (Fig. 6B). Genes up-regulated in R-PASC MoAMs, including Apoe, Spp1, and Lyz2, were also up-regulated in aged mice at 21 dpi and subsided by 35 dpi, aligning with lung pathology recovery (Fig. 6B). Associated with these findings was persistent inflammation in BAL cells from aged mice at 21 dpi (Fig. 6C). The enrichment of TGF-β response and epithelial-mesenchymal transition pathways suggests inflammation-associated fibrosis development in aged mice at 21 dpi (Fig. 6C) (29). In addition, the patterns of BAL cell type composition in aged mice at 21 dpi exhibited similarity with those of individuals with R-PASC (Fig. 6D), reinforcing the usefulness of this model in investigating underlying immune mechanisms in R-PASC.

To delineate alterations in macrophage dynamics during lung sequelae progression, we stratified macrophages into a TRAM cluster, two MoAM clusters (MoAM1 and MoAM2), and a proliferating AM cluster (ProAM) (Fig. 6E). TRAMs predominantly expressed Siglecf, a marker of mature TRAMs. MoAM1s, characterized by elevated Spp1 and Apoe expression, were enriched in aged mice at 21 dpi. In contrast, MoAM2s expressed Klf2 and exhibited rapid lung infiltration during acute infection (Fig. 6F and fig. S11C). Pathway analysis further highlighted the superior endocytic potential of TRAM cells and the enhanced proinflammatory signature of MoAM1 cells (fig. S11D).

Subsequent assessment of transcriptome similarities between MoAMs in R-PASC and analogous MoAM1 from MA10-infected mice indicated substantial overlap, particularly in MoAM1 cells from aged mice at 21 dpi (Fig. 6G and fig. S11, E and F). A comparison between MoAM1s and TRAMs from aged mice showed inflammatory polarization in MoAM1s at 21 dpi (Fig. 6H and fig. S11G). Moreover, MoAM1s exhibited elevated chemotaxis, aligning with increased Ccl2 expression (fig. S11, H and I), indicating that they might be differentiated from recruited monocytes. A characteristic of human R-PASC MoAM_1 is elevated pro–pulmonary fibrosis features, a pattern similar to what we observed in aged mice at 21 dpi (Fig. 6I and fig. S11, J and K).

Consistent with human scRNA-seq analysis, GSEA identified sustained IFN-γ response across respiratory cells and Ifngr2 expression in aged mice at 21 dpi (Fig. 6J and fig. S12, A to D). However, IFN-γ signaling waned by 35 dpi (Fig. 6K), aligning with lung pathology resolution (Fig. 5E). Similar to the human R-PASC MoAM_1 cells, IFN-γ responsiveness in mouse MoAM1s was positively correlated with M1 macrophage and pro–pulmonary fibrosis features in aged mice at 21 dpi (Fig. 6, L and M), further suggesting a potential association between IFN-γ stimulation and postviral fibrotic response. Consistent with human MoAM-like cells, we observed IFN-γ–dependent CCL2 production after in vitro IFN-γ treatment of bone marrow–derived MoAM-like cells (fig. S12E). T cells exhibiting tissue-resident traits emerged as the primary IFN-γ source at 21 dpi, similar to human scRNA-seq analysis (Fig. 6N and fig. S12, F and G). Together, our comparative scRNA-seq analysis demonstrates that the immune landscape of aged mice 3 weeks after SARS-CoV-2 infection resembles the respiratory immune profiles observed in individuals with R-PASC.

Therapeutic targeting of persistent IFN-γ mitigates after SARS-CoV-2 lung sequelae

Our analysis uncovered IFN-γ signaling as a key potential factor contributing to R-PASC. To directly test this, we treated MA10-infected aged mice with anti–IFN-γ after primary viral clearance (10 dpi). Pulmonary pathology and respiratory immune responses were measured at 21 dpi when pronounced pulmonary sequelae were observed (Fig. 7A). Anti–IFN-γ treatment ameliorated lung inflammatory and fibrotic pathology compared with IgG treatment, indicating overall improved outcomes (Fig. 7, B and C, and fig. S13). Moreover, immunofluorescence staining revealed that anti–IFN-γ treatment reduced lung collagen I deposition (Fig. 7, D and E). IFN-γ blockade also reduced BAL cell counts, attributed to decreases in Siglec-F^lo MoAMs and inflammatory monocytes (Fig. 7, F to I, and fig. S14A). CD8⁺ and CD4⁺ T cell counts did not show differences, whereas spike protein–specific CD8⁺ T cells diminished after anti–IFN-γ treatment, potentially associating with decreased BAL CXCL9, an IFN-γ–dependent chemokine capable of recruiting T cells to the lung (fig. S14, B to D).

Fig. 7. Therapeutic targeting persistent IFN-γ mitigates pulmonary pathology after acute SARS-CoV-2 infection.

(A) Experimental setup for evaluating the role of IFN-γ in R-PASC mice. (B) Representative images of lungs stained for H&E and trichrome from IgG/αIFN-γ–treated MA10-infected aged mice. (C) Quantification of inflamed lung area from the indicated group; each dot represents one mouse, pooled from two independent experiments. (D) Immunofluorescent staining collagen I in lungs from the indicated group. (E) Quantification of collagen I–positive area as a percentage of the total area analyzed; each dot represents one mouse, pooled from two independent experiments. (F to I) BAL cell counts (F), Siglec-F^lo AM and Siglec-F^hi AM counts (G), Ly6C^hi monocyte counts (H), and neutrophil counts (I) in BAL of MA10-infected aged mice treated with IgG/αIFN-γ; pooled from two independent experiments. (J) UMAP of BAL cells from IgG/αIFN-γ–treated MA10-infected aged mice. (K) Stacked bar plots showing the proportion of indicated BAL cell types analyzed with scRNA-seq. (L) UMAP of BAL macrophages from IgG/αIFN-γ–treated aged mice. (M) The proportion of indicated macrophage clusters in BAL cells from the indicated group. (N) Relative module score of pro–pulmonary fibrotic macrophage signatures in MoAM1s from the indicated group. (O) DEGs of MoAM1s from IgG/αIFN-γ–treated infected mice. (P) GSEA of MoAM1s from IgG/αIFN-γ–treated infected mice. (Q) The MoAM_1 features from patients with R-PASC were assessed in MoAM1s from the indicated mouse. Data represent the means ± SEM (curve plots) or distribution (violin plots). Data were analyzed by the Mann-Whitney test (C and E to I) or Wilcoxon test (N and Q); *P < 0.05.

We performed scRNA-seq on BAL cells isolated from control or anti–IFN-γ–treated mice at 21 dpi. After IFN-γ neutralization, we noted an increased TRAM frequency (Fig. 7, J and K, and fig. S14E). To further explore the role of IFN-γ in MoAM recruitment and differentiation, macrophages were reclustered into one TRAM, two MoAMs, and one ProAM subcluster (Fig. 7L). Anti–IFN-γ administration resulted in an augmented TRAM and diminished MoAM1 presence (Fig. 7M), mirroring observations from flow cytometric analysis (Fig. 7G). Pseudo-time trajectory analysis suggested that IFN-γ blockade promoted MoAM differentiation into TRAMs by increasing the presence of transitional MoAMs (fig. S14, F and G). After anti–IFN-γ treatment, MoAM1s exhibited a decrease in pro–pulmonary fibrosis features (Fig. 7N and fig. S14, H to J), which was characterized by diminished Arg1, Spp1, Ccl8, and Ccl2 expression, and elevated expression of genes favoring tissue homeostasis and repair, including Pparg, Cebpb, and Plet1 (Fig. 7O) (30). Moreover, anti–IFN-γ treatment dampened inflammatory pathways within MoAMs and curtailed typical MoAM1 features observed in patients with R-PASC (Fig. 7, P and Q). In conclusion, direct targeting extended IFN-γ responses ameliorated proinflammatory and profibrotic macrophage development and dampened lung pathology in this R-PASC murine model.

IFN-γ is dispensable for post–SARS-CoV-2 lung sequelae in BALB/c mice

Reports showed that MA10 infection in aged BALB/c (1-year-old) mice triggered potent respiratory inflammation and pathology 1 month after infection, which can persist for 120 days (7). Similarly, we observed that MA10-infected aged BALB/c mice displayed pronounced chronic lung anomalies 1 month after infection, including immune cell infiltration and collagen deposition (fig. S15, A to D). scRNA-seq analysis of BAL cells revealed decreased TRAMs and accumulated MoAMs (fig. S15, E and F). Consistent with reports, increased B cell responses were observed in infected mice (fig. S15G) (7). However, despite persistent lung pathology, pathway analysis did not reveal IFN-γ or proinflammatory responses nor Infgr2 up-regulation in MoAMs from infected mice (fig. S15, H to K). In addition, MoAMs in infected BALB/c mice showed no congruence with R-PASC MoAM_1 cells, and MoAM profibrotic attributes were comparable between infected and uninfected animals (fig. S15, L and M). Furthermore, T cells in BAL samples from infected BALB/c mice had similar amounts of Ifng expression as uninfected mice (fig. S15N). Last, IFN-γ signaling blockade from 10 dpi did not ameliorate infection-induced chronic lung pathology in this model (fig. S15, O and P). These results suggest that chronic lung conditions developed after SARS-CoV-2 infection in BALB/c mice might not be mediated by IFN-γ or IFN-γ–responsive MoAMs. Collectively, our research revealed that MA10-infected aged C57BL/6 mice exhibited a respiratory immune atlas similar to that of patients with R-PASC, and targeting IFN-γ may offer a promising strategy to mitigate persistent lung sequelae after acute SARS-CoV-2 infection (fig. S16).

DISCUSSION

Dysregulated peripheral immune responses are associated with PASC (5, 31); however, the lung immune landscape in PASC remains unclear. By integrating scRNA-seq and clinical lung function evaluations, we noted a marked dysregulated MoAM response in R-PASC individuals. Pathway analysis underscored that respiratory resident T cell–derived IFN-γ drives MoAM precursor recruitment and polarization, promoting a profibrotic state. Murine models further established IFN-γ as a driver in chronic pathology and functional decline after acute COVID-19.

Although IFN-γ treatment promoted viral clearance in immunocompromised individuals after acute SARS-CoV-2 infection (32), evidence has linked IFN-γ with alveolar injury during acute COVID-19 (33), and elevated serum IFN-γ concentrations were found in individuals with PASC (5). Nonetheless, these studies have yet to determine whether IFN-γ acts as a “driver” or a “passenger” in PASC pathogenesis. Our scRNA-seq analysis, coupled with functional neutralization at the postacute infection stage, identified IFN-γ as a driver of R-PASC, potentially exacerbating systemic symptoms like fatigue because of chronic hypoxia. We found two potential IFN-γ–mediated mechanisms driving R-PASC: enhancing inflammatory monocyte recruitment through the promotion of CCL2 production and facilitating profibrotic MoAM polarization and differentiation.

During acute COVID-19, αβ T cells and NK cells are the main sources of IFN-γ, and persistent IFN-γ–producing NK cells have been observed in nonhuman primate R-PASC models (34, 35). However, our analysis revealed lung-resident αβ T cells as the primary IFN-γ–producing cells in human R-PASC. SARS-CoV-2–specific IFN-γ–producing T cells correlated with worse outcomes, indicating that aberrant virus-specific memory or “long-lived effector” T cells are a culprit for pulmonary sequelae after acute COVID-19, consistent with previous findings by us and others (9, 13). IFN-γ production by T cells usually requires concurrent TCR signaling. Although viral protein was undetectable, low expression of SARS-CoV-2 N gene mRNA in infected mouse lung at 21 dpi suggests that viral remnants persisted in the lung after infectious virus clearance, which may drive IFN-γ production and chronic lung sequelae. The presence of viral remnants alone is likely insufficient to drive lung sequelae, as indicated by minimal sequelae in young mice harboring residual N mRNA at 21 dpi. The host environment and long-term viral remnants may co-determine R-PASC development. Viral mRNA was not detected in BAL cells from COVID-19 convalescents by RT-PCR (9) or scRNA-seq, indicating that viral remnants may reside in the lung parenchyma, because respiratory epithelial cells, the major target of SARS-CoV-2, are present at very low frequency in the BAL. Chronic autoantigen release due to ongoing tissue injury or innate inflammatory signals may also contribute to respiratory IFN-γ production (36). In addition, SARS-CoV-2 infection alters host cell epigenetic status (37), and the chromatin accessibility profiles in aged T cells are markedly different from their young counterparts (38). Therefore, altered respiratory T cell epigenetic landscapes in patients with R-PASC may enhance TCR-independent IFNG expression. Regulatory T (T_reg) cells suppress pathogenic immune responses after respiratory viral infection (39), and their quality and quantity are dysregulated during aging (40). It is possible that dysfunctional T_reg cells in patients with R-PASC may fail to suppress IFN-γ production in pathogenic T cells.

A few animal models with persistent tissue inflammation and fibrosis after acute SARS-CoV-2 infection have been developed. Sustained pathology was described in hACE2-expressing humanized (MISTRG6-hACE2) mice (8), although assessing dysregulated immune responses in this model is challenging because of the lack of intact functional immune responses. An aged BALB/c model exhibiting persistent sequelae after acute SARS-CoV-2 infection has also been previously reported (7). Chronic pathology in this model was independent of IFN-γ. Possible reasons for the ineffectiveness of anti–IFN-γ treatment in dampening chronic tissue pathology in SARS-CoV-2–infected BALB/c mice may include moderate CD8⁺ T cell presence (7) and lack of IFN-γ responsiveness in macrophages in the model. Through rigorous comparative analyses, we identified a transient window in aged C57BL/6J mice emulating the respiratory immune profile of individuals with R-PASC. Lung conditions in patients with R-PASC can persist for more than 2 years (1). Given the short life span of rodents [one aged mouse day may equal 20 to 170 human days (41, 42)], the R-PASC window duration in aged C57BL/6 mice partially mimics clinical conditions. Nevertheless, the brief window of persistent IFN-γ production and signaling likely explains the limited R-PASC time window in mice. Thus, it is of interest to examine whether the extension of IFN-γ signaling by biologics or genetic means could sustain pulmonary sequelae after SARS-CoV-2 infection. Furthermore, some patients recover from R-PASC over time (1); whether diminished IFN-γ signaling underlies the timely recovery requires future studies with longitudinal BAL sampling.

TRAMs self-renew during homeostasis. Under conditions like infections or irradiation injuries, circulating monocytes recruited to the alveolar space gradually adopt an AM signature and differentiate into MoAMs (43–45). Compared with TRAMs, MoAMs have enhanced inflammatory attributes and may play a critical role in pathogen clearance (43, 46). Patients with severe COVID-19 exhibited aberrant monocytes and MoAM accumulation in the respiratory tract during acute hospitalization (12, 47). Nevertheless, previous work has not explored prolonged MoAM presence during the postacute stage. Our data and reports from others showed that increased monocytes and MoAMs correlate with poor clinical outcomes and lung fibrosis in patients with R-PASC (9, 13, 48). However, the molecular and cellular cues maintaining the MoAM population and shaping their phenotypes remained elusive. Our comparative analysis and functional validation suggest that MoAMs are maintained through persistent monocyte recruitment, and their proinflammatory and profibrotic phenotypes are driven by persistent IFN-γ signaling. M2-polarized macrophages are generally considered mediators of fibrosis (49), and M2-polarized MoAMs in idiopathic pulmonary fibrosis or bleomycin-injured lung tissues correlate with collagen deposition (50, 51). Reports have also found that MoAMs transiently up-regulate both M1 and M2 genes during lung fibrosis development (46). Nevertheless, our study connects IFN-γ–responsive M1-like MoAMs with tissue fibrosis and lung functional impairment, possibly due to highly polarized type 1 responses in the lung microenvironments after viral infection.

Our study has some limitations. The limited R-PASC cohort size and accessibility restrict conclusive statements regarding specific patient comorbidities or treatment influences. Evaluating published lung tissue datasets from a distinct PASC-PF cohort (23) corroborated our observations across different cohorts. A larger R-PASC cohort in a recent preprint manuscript also supports our findings (48), although without functional investigations. Because of cell type availability in the BAL, only AMs were characterized in this study, although lung interstitial macrophages may also mediate pulmonary fibrotic changes during acute COVID-19 (52). In addition, directly neutralizing IFN-γ could increase the risk of other infections, and targeting downstream of IFN-γ signaling in macrophages may offer safer therapeutics. Last, it remains to be determined whether SARS-CoV-2 infection leads to long-term pulmonary gas exchange decline in aged mice, as observed in R-PASC individuals or in influenza-infected aged mice (53).

In summary, we have identified IFN-γ as a central node mediating T cell monocyte–derived macrophage interactions in R-PASC, which drives persistent pulmonary inflammation and tissue fibrosis. Our work emphasizes performing a comparative analysis of human specimens and relevant animal-derived samples to probe the cellular and molecular etiology of PASC. Furthermore, our data indicate that the Janus kinase inhibitors, such as baricitinib, already granted emergency use for acute COVID-19 (54), may merit consideration for treating ongoing R-PASC.

MATERIALS AND METHODS

Study design

The objective of this study was to understand the immune mechanisms underlying the development of R-PASC. A cohort of non–COVID-19 controls (n = 4) and COVID-19 convalescent individuals who have recovered from acute COVID-19 more than 60 to 90 days prior (n = 11) were recruited. Furthermore, on the basis of the lung physiological function parameters during recruitment, the COVID-19 convalescent individuals were divided into R-PASC (n = 7) and non–R-PASC groups (n = 4). scRNA-seq analyses were performed on BAL cells and PBMCs from two non–COVID-19 control donors, five R-PASC donors, and four non–R-PASC donors. In addition, we established a mouse model of R-PASC by infecting 21-month-old C57BL/6 mice with SARS-CoV-2 MA10 strain. Through comparative scRNA-seq and flow cytometry analyses, we provided evidence that the animal model likely mimicked the respiratory immune features of patients with R-PASC and further identified IFN-γ as a potential key factor mediating R-PASC. Mice aged 21 months or 8 weeks were randomly assigned to each study group, and every animal experiment was repeated two or three times except for the scRNA-seq experiment. Investigators were not blinded during experimental setup or sample acquisition.

Ethics statement and biosafety

This study was approved by the Mayo Clinic Institutional Review Board (protocol ID 20-004911). Informed consent was obtained from all enrolled individuals. All animal experiments were performed in animal housing facilities at University of Virginia (UVA; Charlottesville, VA). The animal experiments were approved by UVA Institutional Animal Care and Use Committees. All work with SARS-CoV-2 infection was approved under Animal Biosafety Level 3/Biosafety Level 3 conditions and was performed with approved standard operating procedures and safety conditions by the UVA Institutional Review Board.

Statistical analysis

Individual-level data are presented in data file S4. To compare the two sample groups, the Mann-Whitney test or t test was applied for unpaired comparisons. For analysis between multiple groups, normality distribution was firstly tested with the D’Agostino-Pearson test, Anderson-Darling test, or Kolmogorov-Smirnov test, and then one-way or two-way analysis of variance (ANOVA) was performed; Wilcoxon rank sum test and model-based analysis of single-cell transcriptomics (MAST) with false discovery rate adjustment were performed during scRNA-seq data analysis. Correlations were assessed by Pearson’s correlation coefficient running under R (V4.0.3). All statistical tests were performed using GraphPad Prism (V10) or R (V4.0.3).

Supplementary Material

This PDF file includes:

Materials and Methods

Figs. S1 to S16

Tables S1 and S2

References (55–65)

Other Supplementary Material for this manuscript includes the following:

Data files S1 to S4

MDAR Reproducibility Checklist

Data and materials availability:

All data associated with this study are present in the paper or the Supplementary Materials. scRNA-seq data are available from the Gene Expression Omnibus under accession number GSE263817.