Trajectory modeling of endothelial-to-mesenchymal transition reveals galectin-3 as a mediator in pulmonary fibrosis

The endothelial-to-mesenchymal transition (EndMT) is an important source of fibrotic cells in idiopathic pulmonary fibrosis (IPF). However, how endothelial cells (ECs) are activated and how EndMT impact IPF remain largely elusive. Here, we use unsupervised pseudotemporal analysis to recognize the heterogeneity of ECs and reconstruct EndMT trajectory of bleomycin (BLM)-treated Tie2creER/+;Rosa26tdTomato/+ IPF mice. Genes like C3ar1 and Lgals3 (protein name galectin-3) are highly correlated with the transitional pseudotime, whose expression is gradually upregulated during the fate switch of ECs from quiescence to activation in fibrosis. Inhibition of galectin-3 via siRNA or protein antagonists in mice could alleviate the pathogenesis of IPF and the transition of ECs. With the stimulation of human pulmonary microvascular endothelial cells (HPMECs) by recombinant proteins and/or siRNAs for galectin-3 in vitro, β-catenin/GSK3β signaling and its upstream regulator AKT are perturbed, which indicates they mediate the EndMT progress. These results suggest that EndMT is essential to IPF process and provide potential therapeutic targets for vascular remodeling.


Introduction
Idiopathic pulmonary fibrosis (IPF) is a chronic and progressive lung disease with few clinical treatment options 1 . It is characterized by the deposition of extracellular matrix (ECM) material in fibroblast foci, which may lead to the loss of lung compliance and disruption of gas exchange 2 . Recent evidence suggests that the origin of lung fibroblasts plays a central role in the development of pulmonary fibrosis 3,4 , which provides potential therapeutic targets and novel ideas for treatment strategies. Among the identified sources of fibroblasts, the endothelial-to-mesenchymal transition (EndMT) [5][6][7] has been documented to occur in IPF, whereby lung endothelial cells (ECs) can transform into significant numbers of fibroblasts. This finding offers an explanation for the vascular regression that may be a result of the loss of ECs via EndMT, thereby contributing to the fibroblastic elements in these foci 8 . However, exactly how ECs undergo EndMT and the underlying molecular mechanisms remain incompletely understood, making it challenging to target vascular remodeling to ameliorate IPF.
The transition of a certain type of cells into another is usually characterized by the fine-tuning of different transcription factors, sequential expression of marker genes, and modification of cellular functions [9][10][11] . However, the transformation of ECs into the mesenchymal phenotype is not a simple binary event, which imposes great challenges for the pharmaceutical treatment of EndMT-related diseases. Recent studies attempted to define cell transformation during EndMT using singlecell RNA sequencing (scRNA-seq) approaches 12,13 , highlighting scRNA-seq as an important strategy for investing dynamic processes of cell transition. However, few studies aimed to develop therapeutic strategies or discussed the clinical relevance of their findings. Therefore, a deeper understanding of EndMT could be gained using scRNA-seq and bioinformatic tools, with more detailed portraits of pre-defined and possibly intermediate cell states in this process.
The current work aims to investigate EndMT as a dynamic process in IPF, identify key regulators of the transition process, and provide novel targets for the treatment of EndMT-related diseases such as IPF. We characterized heterogenous cell states during EndMT in a bleomycin (BLM)-induced IPF mouse model of IPF in single-cell resolution. Specifically, unsupervised pseudotemporal analysis [14][15][16][17] abled us to reconstruct EndMT trajectory, discovering several genes that are highly correlated with the transitional pseudotime. These candidate genes were further validated for their significance using an in vitro model of TGF-β-induced EndMT, as well as the aforementioned IPF model in vivo. The results revealed that galectin-3 is a strong facilitator and an effective target for inhibiting EndMT and even IPF. Overall, this study provides a meaningful strategy to suppress EndMT and the progression of vascular remodeling during the development of IPF, with great potential for further translational research.

Resident ECs decreased in BLM-induced pulmonary fibrosis
In order to investigate the prevalence of EndMT in pulmonary fibrosis, the pulmonary fibrosis model was established by intratracheal administration of BLM firstly. The overall design of EndMT tracing is shown in Fig. 1A. Following BLM treatment, the weight of the fibrosis model mice gradually decreased, accompanied by a considerable increase in mortality ( Supplementary Fig. 1A, B). The deposition of collagen significantly increased and the lung architecture was severely damaged (Supplementary Fig. 1C-E).
To track the trajectory of the changes of ECs during fibrosis, we performed lineage analysis of fibroblasts using Tie2 creER/ +;Rosa26 tdTomato/+ mice. Based on that, we used immunofluorescence staining to find that CD31-positive ECs scattered throughout the fibrotic area in the BLMtreated pulmonary fibrotic samples (BLM groups), while in the control groups CD31 expression could only be detected in pulmonary microvascular and arterial blood vessels. Interestingly, a subset of the ECs in the BLMinduced pulmonary fibrosis groups was positive for the fibroblast marker COL1, while there was almost no immunoreactivity in the control groups (Fig. 1B).
At the same time, we performed flow cytometry to isolate Tie2 + cells (cells of endothelial origin) from normal and fibrotic lungs. The result showed that the subpopulation of Tie2 + CD31 + was decreased almost 2.8-fold in fibrotic lungs compared to that in the control group (Fig. 1C, Supplementary Fig. 2). Due to the complex nature of lung fibrosis in situ, we also extracted cells from lung tissues and cultured them in vitro to study their changes via FACS analysis. And the result showed that in the BLM group, there were appropriately 18% Tie2 + Vimentin + (which is the fibroblast special marker) ECs (Fig. 1D), which offers strong evidence that the number of ECs decreased and transitioned into fibroblasts during the pathological progression of pulmonary fibrosis.

scRNA-seq reveals the transcriptional heterogeneity of ECs
To further explore the heterogeneity of ECs in PF, we identified subpopulations of the ECs that differ between normal and fibrotic lung tissues using scRNA-seq ( Fig. 2A). After removing cells that expressed less than 100 transcripts, which indicates poor viability and likely apoptosis, a total of 391 single-cell datasets were obtained, among which there were 174 ECs from normal lungs (control) and 217 ECs from BLM-induced fibrotic lungs (BLM). Among the 38,720 coding genes in the mouse genome, 8350 were expressed by at least 5 cells in the control groups and 9497 in the BLM groups, while 9078 genes were expressed by at least 10 cells in both groups. UMAP visualization and clustering heatmap showed that ECs in the control group are well separated from ECs in the BLM group (Fig. 2B, Supplementary Fig. 3). After using gene-ontology (GO) analysis with the database from the Laboratory of Human Retrovirology and Immunoinformatics (LHRI, https://david.ncifcrf.gov/), we found that clusters which aggregated in the BLM group highly expressed genes related to inflammation (8.6%), oxidative stress (6.4%), and differentiation (4.6%) (cluster 1). Violinmaps of the four clustering markers within all EC states also showed several genes with very large heterogeneity compared to the saline-treated control group and BLMtreated pulmonary fibrosis group. The fibroblast-specific marker S100a4 was obviously enriched in the fibrotic cluster (cluster 1), while Col4a1, which has been reported to be universally expressed in quiescent ECs 18,19 , was enriched in the saline-treated control cluster (cluster 3). Interestingly, Tgfbi (transforming growth factor, betainduced), an ECM protein secreted by ECs that has been reported to be related to epithelial transformation and tumor metastasis 20,21 , was widely expressed in cluster 1 (Fig. 2C). Additionally, we were surprised to find that C3ar1 (complement component 3a receptor 1) as well as Lgals3 were significantly expressed in the fibrosis cluster. Moreover, the RT-PCR results confirmed that these genes identified in scRNA-seq were significantly expressed in the fibrotic groups (Fig. 2D). Taken together, the results indicated that ECs present heterogeneity in response to IPF, which may explain the activation of lung ECs in vivo.

Reconstructing a transitional trajectory of ECs in IPF pathogenesis
To further depict the transcriptional dynamics of EC in PF, unsupervised pseudo-time reconstruction was applied using the R package Monocle 3 15 to place ECs on a virtual time axis along which the cells are presumed to travel as they differentiate or become activated. Using this approach, cells from each cluster were added sequentially per timepoint, and the data were visualized in terms of cellular states. The unsupervised pseudotime algorithm produced several branches and endpoints. As EndMT is expected to start from normal ECs, which then convert into a mesenchymal Fig. 1 Verification of EndMT in a mouse model of pulmonary fibrosis. A Schematic diagram of the construction of the IPF model in Tie2 creER/ +; Rosa26 tdTomato/+ mice based on a C57bl/6 genetic background. B Immunofluorescence imaging of ECs (CD31, red) and fibroblasts (COL1, green) in the BLM/control groups. The region in blue represents nucleus. Asterisks indicate the co-localization of ECs and fibroblasts. Scale bar = 13 μm. C Comparative analysis of the relative proportions of EC-derived cells (Tie2 + ) and ECs (Tie2 + CD31 + ) in the BLM/control groups presented as means ± SD, n = 2. x-axis: different mice derived from C57BL/6. Tie2-positive cells represent all cells of endothelial origin; Tie2 + CD31 + cells represent resident ECs that remain endothelium-feature supplemented with BLM or saline solution. y-axis: the percentage of subpopulations of Tie2-positive and Tie2CD31-positive cells. n = 2. D Gating strategy for calculating the fraction that underwent EndMT (Tie2 + Vim + ) cultured in vitro for 72 h after isolation from lungs by PBS perfusion. Specifically, cells gated by rectangles in black are subpopulation of interest. state over the course of fibrosis, we designated ECs from normal lungs (Fig. 3A, white points 1 and 2) as the starting points of the pseudotime. We also found that genes that are specifically activated in fibroblasts and ECs under stress conditions, such as S100a4, Fn1, and Hif1a, were highly expressed in the late stage of the pseudotime, as well as in the BLM group (Fig. 3B). Thus, it stands to reason that the pseudotime reflects the progression of EndMT.
We found that the two candidate genes C3ar1 and Lgals3 share highly similar expression trends with the Fig. 3 Pseudotemporal trajectory analysis of ECs revealing factors related to the EndMT process. A Merged pseudotime analysis of both groups revealing the transitional trajectory of ECs using Monocle3. Hollow circles: start/end nodes; black circles: branching nodes; black curve: pseudotemporal trajectory. B Expression of genes highly correlated with the pseudotemporal trajectory representing the EndMT process. C3ar1 complementary component 3a receptor 1, Lgals3 galectin-3, S100A4 S100 calcium binding protein A4, Fn1 fibronectin 1, Icam-1 intercellular adhesion molecule-1, Col4a1 collagen type IV alpha 1 chain, Tgfbi transforming growth factor beta induced, Hif1a hypoxia inducible factor 1 subunit alpha. C Expression levels of candidate regulators in whole-lung samples from the BLM and control groups according to RT-PCR. CD31: platelet and endothelial cell adhesion molecule 1, α-SMA: actin, alpha 2, smooth muscle, aorta. The data presented as the means ± SD, n = 3. *p < 0.05, **p < 0.01, ***p < 0.001.
fibroblast markers S100a4 and FN1 (Fig. 3C). According to previous literatures, both of them could stimulate fibrosis [22][23][24] , but few of them elucidate the dynamic tendency of these two mediators in the regulation of EC transformation in IPF. Thus, we used the RT-PCR method to quantify the expression of C3ar1 and Lgals3 over time in the BLM-induced IPF mice. At the same time, we measured the change trend of CD31 and α-SMA, which is a specific marker of fibroblasts, as reference. Along the time axis of pulmonary fibrosis, the expression trends of C3ar1 and Lgals3 were highly consistent with the results of pseudo-temporal analysis (Fig. 3C). Moreover, the expression of these two genes was increased by approximately 10-to 15-fold compared to the control group. These results demonstrate that genes that play active roles in lung ECs during the fibrosis process are highly consistent with the predictions of pseudo-temporal kinetics. Furthermore, C3ar1 and galectin-3 may play critical roles in the dynamic transition of ECs leading to the emergence of fibroblasts.

Inhibitors of C3ar1 and galectin-3 alleviate pulmonary fibrosis
To further confirm the positive role of C3ar1 and galectin-3 implied by the results of pseudotemporalordering analysis, we used lentivirus transfection and antagonists to interfere with their expression (Fig. 4A). The knockdown efficiency of C3ar1 and galectin-3 in CHO cells after transfection with the lentiviral plasmids (LV C3ar1 shRNA, LV galectin-3 shRNA and the control empty plasmid, LV vehicle shRNA) reached 91 and 52%, respectively ( Supplementary Fig. 4). After nasal intubation to inject lentivirus into BLM-induced pulmonary fibrosis mice, fibrosis was alleviated, including a reduction of the weight loss and increase of the survival rate, which reached approximately 40% in the LV C3ar1 shRNA group and LV galectin-3 shRNA group, and was significantly higher than the survival rate of the sham group (LV vehicle shRNA) (Fig. 4B). At the same time, we intraperitoneally injected the mice with the C3ar1 antagonist SB290157 or the galectin-3 antagonist GB1107 during the development of BLM-induced pulmonary fibrosis. As shown in Fig. 4C, the tendency of weight change and survival rate was consistent with the lentiviral interference experiments. Notably, the survival rate after treatment with GB1107 reached 76%, compared to 33.3% in SB290157 group and 20% in the sham group (vehicle control with only DMSO).
The destruction of normal tissue architecture and ECM deposition was assessed using semi-quantitative grading of HE and Masson's trichrome stained sections. As expected, GB1107 preserved the alveolar structure and reduced the deposition of ECM to a normal level. In the mice treated with SB290157, the intensity of the fibrotic areas and ECM deposition was more severe than in those treated with GB1107. At the same time, the alveolar thickness and ECM, especially collagen deposition, was clearly observed in the sham group which was treated with only the vehicle, DMSO (Fig. 4D-F). The immunofluorescence staining of lung tissues illustrated that the expression of C3a and galectin-3 was significantly reduced after antagonist treatment ( Supplementary Fig. 5). Notably, we were able to observe that ECs widely express galectin-3 protein (co-labeled) in the BLM-induced pulmonary fibrosis group (BLM + DMSO), which corroborates the results of the single-cell analysis.
Taken together, these results demonstrate that inhibiting mediators screened by pseudotemporal analysis could alleviate the development of pulmonary fibrosis. Furthermore, inhibiting galectin-3 expression was more effective in alleviating fibrosis than inhibiting C3ar1.

Galectin-3 stimulates EndMT in vitro
To further investigate if the role of galectin-3 in BLMinduced pulmonary fibrosis is related to EC activation and EndMT, we first used human lung micro-endothelial cells (HPMECs) to induce EndMT as described before [25][26][27][28] . We stimulated HPMECs with TGF-β as a positive control group, and additionally used recombinant galectin-3 protein to explore its function in EC activation. The morphology of ECs and the expression of classical markers, such as CD31 and α-SMA, could be used to assess the occurrence of EndMT (Fig. 5A). With the performance of CCK-8, there was no significant effect of galectin-3 compared to the treatment with TGF-β, which means that galectin-3 has little cytotoxicity for HPMVECs (Fig. 5D). On the other hand, by statistically assessing the morphology of HPMECs at different time points, we found that they progressively changed from a normal cobblestone appearance into dispersed, spindle-shaped cells in the presence of TGF-β or galectin-3 ( Fig. 5B, C, Supplementary Fig. 6). In addition, after measuring the expression of α-SMA, we found that α-SMA had a significant effect on the expression of galecin-3, which was upregulated 3.27-fold compared to the starvation group (Fig. 5E). Taken together, these results indicate that galectin-3 plays a vital role in the activation of ECs, and the results are highly consistent with the effects of TGF-β on the EndMT process.
Galectin-3 promotes EndMT during IPF by activating the AKT/GSK3β/β-catenin signaling pathway β-catenin is the central player of the canonical WNT signaling pathway, which participates in many cellular process 28,29 . MacKinnon et al. reported that galectin-3 reduced the phosphorylation and nuclear translocation of β-catenin 30 , which indicated that galectin-3 may also impact the activation of ECs in BLM-induced pulmonary fibrosis. GSK3β, an upstream regulator of β-catenin, has been confirmed to be inhibited through phosphorylation by activated phospho-AKT 31 . In order to investigate how galectin-3 activates ECs to regulate vascular remodeling in BLM-induced IPF, we examined the protein expression of ECs and mesenchymal biomarkers along with regulators of the AKT/GSK3β/β-catenin singling pathway after treatment with the galectin-3 antagonist GB1107 or vehicle (DMSO). As can be seen in Fig. 6, the expression of galectin-3 was downregulated after treatment with the inhibitor GB1107 compared to the sham group (BLM + DMSO) and normal mice (control group). The level of total β-catenin in the GB1107 group was decreased compared to that in the BLM group, which was accompanied by a significant reduction of phosphorylated β-catenin (pY654 β-catenin, Fig. 6B), phosphorylated AKT (pS473 AKT, Fig. 6C), and phosphorylated GSK3β (pS9GSK3β, Fig. 6D). Therefore, it stands to reason that GB1107 inhibited galectin-3 expression and attenuated the development of BLM-induced pulmonary fibrosis by suppressing the AKT/GSK3β/β-catenin singling pathway.
Based on these results, we suspected that galectin-3 also plays vital roles in the activation and trans-differentiation of ECs. Therefore, HPMECs were treated with siRNA to inhibit the expression of galectin-3 and stimulated with the recombinant TGF-β or galectin-3 protein as described above (Fig. 7A). As expected, the expression of galectin-3 was almost completely suppressed by the siRNA (Fig. 7B). Additionally, galectin-3 expression was dramatically enhanced when the cells were treated with recombinant proteins TGF-β and galectin-3 without siRNA (Fig. 7C), which also indirectly indicated that EndMT could promote galectin-3 expression. At the same time, the biomarkers α-SMA and SMAD2, which act as regulators downstream of TGF-β 32 , along with the mediators AKT and β-catenin, showed notably decreased levels after galectin-3 siRNA interference compared to the untreated group ( Fig. 7D-G). Additionally, Western blot analysis demonstrated that the level of phosphorylation of both β-catenin and AKT was also diminished following siRNA interference compared to that in the untreated groups (Fig. 7H). These results illustrated that galectin-3 activates the trans-differentiation of ECs via the AKT/GSK3β/ β-catenin signaling pathway.

Discussion
Only two drugs to date, Nintedanib and Pirfenidone, have been granted conditional recommendations for IPF, both of which attenuate fibrosis via anti-inflammatory effects. However, due to the limited understanding of the roles of blood vessels and vascular ECs in the pathogenesis of IPF, few studies attempted to inhibit the abnormal activation of ECs and consequent vascular remodeling. Analysis of the single-cell transcriptomes of ECs from normal/fibrotic lungs has enabled us to map the transitional trajectory of ECs for the first time. Several genes, including Tgfbi, Apoe, Hif1a, and Nfkbia, were found to be correlated with the pseudotemporal trajectory, although none as strongly as C3ar1 and galectin-3. Notably, these genes have also been reported as key regulators of IPF. For example, EndMT occurs in the early stage of radiation-induced pulmonary fibrosis and is dependent on Hif1a expression 33 . Their roles in EndMT should be further studied to identify possible new targets for the treatment of vascular remodeling in fibrosis, and such methods can also be used for other studies of celltype transition processes.
Inhibiting the expression of galectin-3 showed a stronger effect against the development of pulmonary fibrosis than inhibiting C3ar1. It is possible that galectin-3 is expressed ubiquitously, not only in vascular ECs, but also in epithelial cells, macrophages, and myofibroblasts 34,35 , in which case its inhibition would have exerted a global and stronger effect against fibrosis. Furthermore, as an  A Schematic showing the knockdown of galectin-3 expression with siRNA, and its impact on the transition of HPMECs. HPMECs were first cultured in complete medium for 24 h, and then transfected with siRNA targeting galectin-3 for 6 h. After treatment with the siRNA, the complete medium was replaced with starvation medium (DMEM containing 1% FBS), followed by further cultivation for 48 h. B The knockdown efficiency of galectin-3 siRNA according to RT-PCR. siNC: negative control. C-G HPMECs were stimulated with TGF-β (10 ng/ml, S + T) or galectin-3 (1 μg/ml, S + G) and interfered with galectin-3 siRNA (S + T + galectin-3 siRNA and S + G + galectin-3 siRNA, respectively) for 48 h. Analysis of the expression level analysis of the target genes galectin-3, β-catenin, and SMAD2, as well as the fibroblast-related genes α-SMA and S100A in different experimental groups using RT-PCR. Error bars represent standard deviations, n = 3. H HPMECs lysate was analyzed for CD31, α-SMA, and AKT/β-catenin signaling pathway regulators, total AKT and pS473 AKT, total-β-catenin and pY654 β-catenin, PI3K and GSK3β by Western blotting. *p < 0.05, **p < 0.01, ***p < 0.001. insights into fibroblast activation. Hence, it stands to reason that galectin-3 plays more vital roles in BLMinduced pulmonary fibrosis.
Our study proposes a new approach for studying the dynamic transition process of ECs and mapping EndMT states, provides important evidence for the transition of ECs and vascular remodeling during the fibrotic process, which has many strengths, but also has unique limitations. Firstly, we did not find completely novel markers that contribute to the EndMT process. Therefore, we hope to develop additional time-course analytic approaches to acquire more information on cell-type transition processes in the future. Secondly, it is well known that EndMT can also be induced by hypoxia or inflammatory factors in vivo, and it is challenging but important to prove that galectin-3 influences the progression of fibrotic disease mainly by inducing EndMT. Therefore, further research should be undertaken to establish conditional knockout models in order to track galecin-3 expression in ECs. Finally, based on the current research, we expect to be able to develop new combination therapy strategies based on small-molecule inhibitors of galectin-3, and propose new targets for translational medicine.
All cells were maintained at 37°C in an incubator with 5% CO 2 .

Cell-counting kit assay
For the CCK-8 assay (Beyotime Biotechnology, China), HPMECs were seeded into 96 well plates and treated with different recombinant proteins. After culture for different lengths of time (1, 2, 4, and 7 days), the cells were incubated with 10 μl/well of the Cell Counting Kit-8 (CCK-8) solution for 2 h at 37°C. The absorbance at 450 nm was measured using a conventional microplate reader.

RNA isolation and quantitative RT-PCR
Total RNA was extracted from cells using TRIzol reagent (Invitrogen, USA) according to manufacturer's instructions. Then, total RNA (1 μg) was reversetranscribed using the StarScriptII First-strand cDNA Synthesis Mix with gDNA Remover (GenStar, China). The 2× RealStar Green Power Mixture with ROXII (GenStar China) was used to label the cDNA with fluorescent probes. Real-time-polymerase chain reaction (RT-PCR) was performed on a 7500 Fast Real-time PCR system (Thermo Fisher Scientific, USA), using the two-step RT-PCR program parameters provided by the manufacturer. The primers were designed using Primer Bank (https://pga.mgh.harvard.edu/ primerbank/) and NCBI (https://www.ncbi.nlm.nih.gov/) and the sequences are listed in Table S3.
Cycle threshold (ΔΔCt) values were calculated by normalization to GAPDH, and the gene expression levels were compared using the 2 −ΔΔCt values.

Mice and fibrosis model
All wild-type (WT) and transgenic mice were based on the C57BL/6J background, aged 6 weeks at the beginning of experiments. WT mice were bred and housed at Tsinghua University Animal Facilities, with a 12 h light/dark cycle. Tie2 creER/+ ;Rosa26 tdTomato/+ mice were obtained and housed at the Shanghai Model Organisms Center, Inc. (Shanghai, China). Group allocation was randomized by Microsoft Excel (Microsoft Inc. USA) with no blinding.
To induce the pulmonary fibrosis model, the mice were anesthetized using tribromoethanol (350 mg/kg), and the lungs of the mice were treated with bleomycin (BLM) via intratracheal injection. BLM was prepared by dissolving sterile BLM sulfate powder (Ruitaibio, China) in sterile saline. BLM was administered via intratracheal injection (3.5 mg/kg in a total volume of 100 μl sterile saline) [36][37][38][39] . For the control group, the mice were treated with an equal volume of sterile saline.

HE and Masson's trichrome staining and scoring of lung sections
The lungs were perfused with phosphate buffer (PBS) from the right to the left ventricle of the heart. After perfusion, the lungs were surgically removed, the lobes separated, fixed in 4% paraformaldehyde (PFA), embedded in paraffin, and processed into 3 μm slices. The sections were stained with hematoxylin and eosin (HE) or subjected to Masson's trichrome staining. Stained sections were scanned under a laser scanning confocal microscope (Zeiss, USA).
For semiquantitative morphological index (SMI) analysis, the lung sections were scanned using a 20× objective. Each captured field was individually assessed for severity of interstitial fibrosis and allotted a score between 0 and 8 40,41 .
For FACS data analysis, cells were sorted using 488 nm and 633 nm lasers, and the fluorescence signals were captured on FITC, CY-7, FSC (forward scatter), and SSC (sider scatter) channels. In order to ensure the viability of cells, 2000 events were recorded for each sample at a rate of 0.5 μl/s. Cytometry data were processed to generate the percentage values of positive events using FlowJo (version 10.4) software (Becton, Dickinson & Company, USA).
For drug treatment, we first treated mice with BLM as described above. On the third day after model induction, we intraperitoneally injected the mice with SB290157 (0.1 mg/kg/day, n = 6) or GB1107 (0.1 mg/kg/day, n = 6) according to literatures 42,43 . These drugs were injected every other day until the weight of the mice fell below 15 g and were euthanized. The control group was treated with the vehicle (DMSO, n = 6).

Lentivirus production and infection on mice
We firstly designed the core fragment of the shRNA lentivirus for C3ar1 and galectin-3 using the shRNA library of Tsinghua University. The detailed sequences are shown in Table S2. Then, HEK 293 T cells were transfected with the lentivirus packing plasmids PSPAX2 and PMD2.G at 90% confluence on the day of transfection. To assess the knockdown efficiency of the lentivirus, we transfected Chinese hamster ovary cells (CHO) when they reached about 70% confluence, using the lentiviral particles together with polybrene (Millipore, Sigma, USA) overnight. After 3 days, FACS was conducted to screen positive cells. Once the positive cells were collected, we used RT-PCR to measure the target gene expression.
For lentivirus infection (LV C3ar1 shRNA and LV galectin-3 shRNA), nasal intubation was adopted to deliver 1 × 10 8 transducing units (TU) (20 μl per mouse, multiplicity of infection (MOI) = 10) into BLM-induced pulmonary fibrosis mice (n = 5 for each group) at the first day. For the control group, the empty lentivirus vector (LV vehicle shRNA) was injected into pulmonary fibrosis mice.

Total protein extraction and Western blot analysis
Total protein from lung tissues (n = 5 for each group) and cultured HPMECs (n = 3 for each group) was extracted using lysis buffer (M-PER Mammalian protein Extraction Reagent, ThermoFisher, USA) with protease and phosphate inhibitors (ThermoFisher, USA). Protein concentrations were determined using the BCA protein assay (Yeasen, China). An equivalent amount of protein (20 μg) was subjected to 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto a PVDF membrane (0.45 μm, Merck Millipore, Germany). The membranes were blocked for 30 min with the NcmBlot blocking buffer (NCM Biotech, China) at room temperature and/or stripped with stripping buffer (CWBIO, China). The blocked membranes were individually incubated overnight at 4°C with rabbit anti-mouse (for lung samples) or mouse anti-human (for HUVEC samples) primary antibodies against CD31 (1:1000, Abcam, UK), α-SMA The chemiluminescence signal was detected using Enlight buffer (Engreen, China). The optical densities of the bands were quantified using Image Lab software (Bio-Rad, USA) and normalized to GAPDH as internal control.

scRNA-seq of Tie2 + cells and data analysis
The healthy and BLM-treated (D7) Tie2 creER/+ ; Rosa26 tdTomato/+ mice Tie2 creER/+ ;Rosa26 tdTomato/+ mice (n = 2 for each group) were euthanized with CO 2 and the lungs were thoroughly perfused with PBS to remove blood from the lung vascular beds 44 . The lungs were then removed from the thoracic cavity and cleared of extraneous tissue. The tissue samples were cut into small pieces and digested with 5 ml of Collagenase II (2.5 mg/ ml, Invitrogen, USA) in a 50 ml tube for 30 min in a 37°C water bath 45 . The resulting tissue/cell suspension was filtered through a 70 μm strainer and centrifuged for 5 min at 300 g. After removal of the supernatant, the cells were incubated with 1× Red Blood Cell Lysis Buffer (BioLegend, CA) for 5 min in ice. Then, the cells were subjected to FACS sorting to isolate tdTomato + (Tie2 + ) cells.
scRNA-seq of the Tie2 + cells was performed on the 10-17 μm C1 HT IFC (Fluidigm, USA), immediately after the cells were collected by FACS. The cDNA library was sequenced on an HiSeq 2500 platform (Illumina, USA), followed by de-multiplexing using barcode sequences provided in the Fluidigm protocols. All downstream data analyses were conducted with R software. Principle component analysis (PCA), 2-dimentional data visualization by Uniform Manifold Approximation and Projection (UMAP) 46 and cell clustering were realized using the Seurat3 package 47 , while the pseudotemporal trajectory and pseudotime-based differential expression analysis were performed using the Monocle3 package 14,48,49 .

Statistical analysis
All samples were represented as biological replicates, and results are shown as means ± standard errors of the mean. GraphPad Prism 8 (GraphPad Software Inc. USA) was used to generate all charts and conduct all statistical analyses except the scRNA-seq results. One-way ANOVA followed by Tukey's or Dunnett's multiple comparison tests was applied to determine the significance of differences among different groups. Differences with P < 0.05 were considered statistically significant (*P < 0.05, **P < 0.01, ***P < 0.001).
Details of all reagents and products used in this study can be found in Table S1.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.