Abstract
Early in life, the lung mesenchyme is extremely dynamic. We profiled the transcriptomes and locations of mesenchymal cells (MC) in the perinatal murine lung and observed cellular progressions of fibroblasts, myofibroblasts, airway smooth muscle (ASM), and mural cells, poised to modulate the extracellular matrix (ECM) with both subtype- and temporal-specific patterns. Within one day after birth, embryonic fibroblast precursors branched into two subtypes, alveolar and adventitial fibroblasts. ASM and myofibroblasts derived from a common precursor population that specifically expressed Crh, a hormone central to glucocorticoid release. Vascular smooth muscle cells and pericytes differentiated gradually. Fibroblast, myofibroblast, and pericyte proliferation peaked at 7 days after birth. Paracrine signaling among cell types decreased after birth and was hierarchical, with pericytes at the interface between endothelial cells and other MC. Postnatal hyperoxia delayed the developmental progression in cell type abundance and gene expression, mirroring the arrested development that characterizes the neonatal lung disease, bronchopulmonary dysplasia. Hyperoxia decreased pericyte and myofibroblast abundances and proliferation, dysregulated signaling, altered ECM modulation and, in males specifically, gave rise to a novel population of contractile fibroblasts. This study identifies the mesenchymal populations orchestrating lung development at the critical transition to air-breathing life and their distinct reactions to hyperoxia.
Introduction
In utero, the lung is fluid filled and the pulmonary circulation develops in a relatively low oxygen tension environment with low blood flow and high pressure [1]. A few moments after birth, the perinatal lung undergoes a remarkable transition that enables gas exchange with establishment of a gas-liquid interface, a ten-fold increase in pulmonary blood and a marked decrease in pulmonary arterial pressure [2,3]. After birth, the distal lung undergoes structural remodeling including the formation of alveoli by secondary septation and a transition from a double to single capillary layer to increase gas exchange efficiency [4]. Alveolarization, the final stage of lung development, characterized by rapid pulmonary parenchymal and vascular growth, continues through at least the first decade of life in humans.
Lung development requires coordinated interactions between multiple cell types in the epithelial, endothelial, immune and mesenchymal populations [5,6]. Temporally and spatially regulated changes in gene expression of specific cell types mediate branching morphogenesis and the stereotypical proximal to distal outgrowth of the airways [7], vasculature and lymphatics that occurs with maturation [8]. Lung mesenchymal cells play a central role in secondary septation, thinning of the interstitium, and in transmitting mechanical forces that promote alveolarization [9]. Though the lung mesenchyme includes multiple, distinct cell types, knowledge surrounding the degree of cellular heterogeneity, dynamic changes in gene expression, and cell-cell interaction during the alveolar stage of lung development remains limited.
Insight into the lung mesenchyme during early postnatal life has significant implications for lung injury and regeneration, especially in neonates. Since the respiratory tract continues to mature in late gestation and through the first decades of life [10], premature infants are uniquely susceptible to life-threatening lung disorders, including bronchopulmonary dysplasia (BPD), a lung disease characterized by compromised alveolarization [11]. Given the importance of the mesenchyme in both physiologic and pathophysiologic lung development, we sought to interrogate the lung mesenchyme relative to cellular diversity, dynamic changes in gene expression and cell-cell communication in the perinatal lung, during alveolarization, and in the context of hyperoxia-induced lung injury, a preclinical model of BPD [12].
In this report, we combined single cell transcriptomics (scRNA-Seq) with fluorescent multiplexed in situ hybridization (FISH) to characterize changes in composition, localization, and function of mesenchymal cells in the murine lung from just before birth through the first three weeks of postnatal life. Moreover, we applied the same strategies to cells derived from neonatal murine lung after 7 days of hyperoxia (80% oxygen). Mesenchymal cells fell into three broad categories: fibroblasts, airway smooth muscle (ASM)/myofibroblasts (MyoF), and mural cells, with 14 distinct subtypes overall. The most dynamic changes in cell phenotype occurred at birth as bipotent precursors for fibroblast and ASM/MyoF matured into distinct cell subtypes between E18.5 and P1. P7 represented the most proliferative time point across the lung mesenchyme, when a novel proliferating population of MyoF emerged. From P7 to P21, the transcriptomic profiles of MyoF and proliferative cells disappeared completely, while alveolar and adventitial fibroblasts and ASM underwent deep transcriptional changes towards cell quiescence. Hyperoxia decreased pericyte abundance and delayed maturation-related transcriptomic changes in the mesenchyme, mirroring the arrested development of the lung that characterizes BPD. Further, in hyperoxic males, but not females, a previously undescribed population of contractile fibroblasts emerged. Taken together, our data demonstrate that in the neonatal lung, the mesenchyme demonstrates remarkable cellular diversity, dynamic changes in gene expression, plays a critically important role in the communication with endothelial and immune cells, and demonstrates a marked sex-related difference to hyperoxia-induced neonatal lung injury.
Results
The perinatal lung mesenchyme is composed of three coarse cell types
To construct a single cell map of the perinatal lung mesenchyme, we generated single cell RNA-Seq (scRNA-Seq) data from eight mice at different stages of perinatal development, with two mice (one female and one male) from E18.5 (early saccular stage), P1 (late saccular), P7 (early alveolar) and P21 (late alveolar). Lung tissue was resected, perfused, and dissociated as described previously [13]. Fluorescence activated cell sorting (FACS) was used to enrich mesenchymal cells by negative selection (i.e. by depleting cells positive for CD326, CD31, and CD45 that constitute the epithelium, endothelium, and immune compartments, respectively) (Figure 1A). Smart-Seq2 library generation and sequencing were then constructed [13].
(A) Experimental design including time points sampled, tissue dissociation, and cell sorting to isolate mesenchymal cells. (B) Embedding of lung mesenchymal cells, colored by relative expression of four distinguishing marker genes. (C) As in B, but colored by coarse cell type. (D) Relative abundance for each coarse cell type during normal development, with ASM/MyoFs peaking at P7 and mural cells gradually increasing over time. (E) As in B, but colored by time point of each cell. ASM: airway smooth muscle. MyoF: myofibroblast.
Mesenchymal cells are among the most diverse and least well characterized cells of the body [14]. We initially built a coarse representation of the perinatal lung mesenchyme by computing a t-distributed stochastic embedding (t-SNE) [15] and coloring each cell based on the expression of relatively established marker genes (Figure 1B). We thereby distinguished three coarse mesenchymal cell types: fibroblasts marked by either Col13a1 or Col14a1 [16,17], airway smooth muscle cells or myofibroblasts (ASM/MyoF) marked by Tgfbi [18], and mural cells marked by Pdgfrb [19] (Figure 1C). No cluster demonstrated high expression of both Tgfbi and Pdgfrb, previously described as vascular smooth muscle cells (VSM) markers [20], arguing for the importance of clarifying VSM annotations. Cell type abundance formed an interesting temporal pattern: fibroblasts were the majority cell type at all time points except P7, ASM/MyoF peaked at P7, and the relative abundance of mural cells gradually increased with maturation (Figure 1D). Temporal patterns were evident in the embedding with each cell colored by time point (Figure 1E).
Distinct mesenchymal subtypes differentially modulate the extracellular matrix composition
With this coarse cell ontology in mind, we clustered the cells [21] to identify each subtype. We identified a total of 14 distinct cell populations (Figure 2A). Based on our initial ontology, clusters 1-6 resembled fibroblasts, clusters 7-11 ASM/MyoF cells, and clusters 12-14 mural cells. The phenotypic richness of the pulmonary mesenchyme posed a challenge for cell type identification. We therefore harmonized our data with mesenchymal cells from Tabula Muris Senis (TMS) [22] via a batch-balanced k-nearest neighbors graph [23] and computed a separate embedding [24] colored by data source (Figure 2B, top) and cluster (bottom). Among our clusters, the ones that included only or mostly P21 cells (4, 6, 9, 12, 13, 14) merged best with TMS, which includes data from mice 1 month old or older. Clusters 13 and 14 overlapped with TMS pericytes, cluster 4 with TMS alveolar fibroblasts, cluster 6 with TMS adventitial fibroblasts, cluster 9 with TMS myofibroblasts, while cluster 12 embedded between smooth muscle-like cells and pericytes. The other clusters (1-3, 5, 7-8, 10-11) included cells predominantly from earlier developmental time points and did not overlap with adult clusters, supporting the notion of distinct mesenchymal cell phenotypes in the perinatal developing lung.
(A) Embedding as in Figure 1B, but colored by fine cell type and annotated with cluster numbers 1-14. This color code is used in subsequent figures. (B) Harmonized embedding of perinatal mesenchymal cells with mesenchymal cells from TMS colored by source (top panel, grey TMS cells drawn on top) and cell type (bottom, merging concordant cell types between our data and TMS). (C-D) Dotplots of all mesenchymal cell types in this study for (C) marker genes and (D) components of the extracellular matrix. TMS: Tabula Muris Senis.
For each cluster, we identified marker genes by differential expression (Figure 2C). Unambiguous annotation of almost every cell subtype could be achieved by combining: (i) marker gene expression; (ii) TMS harmonization; (iii) our initial coarse ontology; and (iv) the most common time point of the cells populating the cluster. For cluster 12, the coexpression of both smooth muscle (Tagln, Acta2) and mural (Prgfrb) genes identified this population as vascular smooth muscle (VSM) cells, a cell type that was not annotated in TMS. The annotation of cluster 9 was the most challenging: while it segregated far from other ASM/MyoF cells in our embedding (Figure 2A) and consisted of cells derived from P21 mice (Figure 1E), it colocalized with cells annotated in TMS as myofibroblasts (MyoF) in the harmonized embedding (Figure 2B). Although TMS did not identify an airway smooth muscle (ASM) cell type, ASM but not MyoF are abundant in healthy, adult mice [25]. At postnatal time points, the expression of the marker genes Hhip and Pdgfra was mutually exclusive, and Hhip is an ASM marker [26]: we therefore labelled the Hhip+ Pdgfra- clusters 8 and 9 as early ASM and ASM, and the Hhip-Pdgfra+ cluster 10 as MyoF. At E18.5, a single MyoF/ASM cluster was found (cluster 7), which coexpressed both Hhip and Pdgfra as well as Tgfbi: it was therefore annotated as MyoF/ASM precursors. Clusters 2, 11, and 14 expressed Mki67 and were therefore labeled as proliferating fibroblasts, MyoF, and pericytes, respectively (Figure 2C-D, bottom labels). These cell subtypes corresponded only partially to categories identified in recent publications [27] (Figure 2 - Supplementary 1) and preprints [28] (Figure 2 - Supplementary 2). In particular, in those studies, similar to TSM, there was no distinction made between ASM and VSM.
A canonical function of mesenchymal cells is extracellular matrix (ECM) modulation. We therefore examined the expression of several ECM-related genes (Figure 2D). Fibroblasts shared ECM components Mfap4, Fn1, Vcan, and Ogn. In alveolar fibroblasts (AlvF), subtype-specific expression included Spon1 which encodes a secreted adhesion protein. Adventitial fibroblasts (AdvF) specifically expressed Podn, encoding a molecule that constrains SMC proliferation and migration [29], Mfap5, which contributes to tissue elasticity [30], and Dcn, which confers resistance to tissue compression. ASM expressed Aspn, which inhibits canonical TGFβ and SMAD signaling [31], and late ASM expressed Lum, which is responsible for collagen binding. Interestingly, early ASM did not express Lum, suggesting a change in fibril diameter regulation around conducting airways at the end of alveolarization [32,33]. Conversely, Aspn, a modulator of ECM mineralization that competes with Dcn for collagen binding [34], was highly expressed in early ASM from P1 and P7 but downregulated at P21, suggesting higher activity in the temporal window surrounding late alveolarization. Pericytes expressed Mcam, Mfge8, Postn, and Eng, which modulates vascular tone, endothelial cell (EC) migration and angiogenesis [35]. VSM had lower expression of many ECM genes but the highest expression of elastin (Eln), which modulates tissue stiffness during alveolarization, with all cell types except pericytes expressing moderate levels of Eln as well [36,37].
Fibroblast and ASM/MyoF bipotent precursors split at birth
The lung mesenchyme changed dramatically between E18.5 and P1, a time window of just 48 hours. Both fibroblast precursors (Fib precursor) and ASM/MyoF precursors, the two most abundant mesenchymal types at E18.5, completely disappeared by P1 (Figure 3A). Fib precursors (cluster 1) are replaced by two distinct types of postnatal fibroblasts (clusters 3 and 5) (Figure 3B) which expressed Col13a1 and Col14a1 and resembled adult AlvF and AdvF fibroblasts, respectively. Both genes are coexpressed in the E18.5 precursors, indicating a common origin and a cell fate commitment during late gestation up to birth (Figure 3C). In terms of ECM secretion, fibroblast precursors appear to be more similar to AlvF (Figure 2D). ASM/MyoF precursors (cluster 7) split into ASM (cluster 8, Hhip+) and MyoF (cluster 10, Pdgfra+) (Figure 3D): only ASM/MyoF precursors coexpressed Hhip and Pdgfra (Figure 3E). Shared expression of Tgfbi and Tagln underscored a common lineage (Figure 2D). Unlike early AlvF and AdvF, which are already present in small numbers at E18.5, almost no ASM or MyoF cells were observed before birth.
(A) Changes in cell type composition in the lung mesenchyme between E18.5 and P1. Colors and cluster numbering as in Figure 2A. (B) Zoom-in of the embedding on fibroblast precursors (cluster 1), early alveolar (3), and early adventitial fibroblasts (5) at E18.5 (left) and P1 (right). (C) Gaussian kernel density estimate of expression for genes shared between fibroblast precursors and either one of the progenies. (D) As in B, but focused on ASM/MyoF precursors (7), early airway smooth muscle (8), and myofibroblasts (10). (E) As in C but for ASM/MyoF precursors versus their progenies. (F) Negative log-P values of gene set enrichment tests for each precursor against a balanced sample of their two postnatal subtypes [38], limited to the top 4 up-regulated (red, top) and 4 down-regulated (blue, bottom) pathways each. (G) Geometric averaged expression of biologically relevant genes in the bipotent precursors versus their progenies. (H) Zoom-in of the embedding on ASM/MyoF precursors showing the expression of Tgfbi and Crh. (I) In-situ imaging of Crh+ Tgfbi+ cells (arrows) in E18.5 lungs.
We then used pathway enrichment analysis to compare prenatal precursors and postnatal subtypes [38]. Both fibroblast and ASM/MyoF bipotent precursors expressed higher levels of genes related to vascular development and muscle cell proliferation. In contrast, fibroblasts precursors specialized in mesenchyme development compared to postnatal fibroblasts, while ASM/MyoF precursors specialized in response to hormones compared to postnatal ASM and MyoF (Figure 3F, upper, red bars). Pathways down-regulated in precursors compared to distinct postnatal counterparts also differed between the two cell lineages (Figure 3F, lower, blue bars). MyoF/ASM precursors were the only cell type expressing corticotropin releasing hormone (Crh, Figure 3G), an early factor within the hypothalamic–pituitary–adrenal (HPA) axis that stimulates glucocorticoid release. No genes encoding Crh receptors were expressed by any cell type in the perinatal lung, including in immune [13], endothelial [20] and epithelial cells (Figure 3 - Supplementary 1). Further, no cells in the adult lung expressed Crh or its receptors [22], raising the possibility that MyoF/ASM precursor-secreted Crh might be signaling through the HPA axis. Interestingly, Crh plays an essential role in lung maturation without significantly impacting surfactant production by epithelial cells [41,42]. Embeddings of ASM/MyoF precursors at E18.5 colored by expression of Tgfbi and Crh indicate higher heterogeneity for the latter gene (Figure 3H). Crh-expressing MyoF/ASM precursors were localized by in-situ imaging via RNA-Scope within the lung parenchyma (Figure 3I). Fibroblast precursors expressed high levels of Hsd11b1 (Figure 3G), the main enzyme responsible for converting cortisone into its bioactive form cortisol, indicating a cell-type-specific amplification of glucocorticoid action [43]. Several genes related to response to hypoxia were lost in both lineages, including Hif3a, a negative regulator of Hif1a, suggesting tight regulation of this pathway at the critical transition between fetal environment and air-breathing life. Lox, a key modulator of elastin fiber assembly, was also down-regulated after birth except in AdvF, indicating a cell function specialization in that direction (Figure 3G).
Mesenchymal cell proliferation increases transiently during postnatal development
Through the first seven days of life, the relative abundance of fibroblasts decreased, while ASM and particularly MyoF increased, with MyoF also branching into a distinct proliferating population representing the only novel cell type at P7 (cluster 11, see Figure 4A, left). The subsequent temporal transition from P7 to P21 indicated a widespread maturation of early postnatal fibroblast and ASM cell types into the adult types (Figure 4A, right) that harmonized with TMS (Figure 2B). While MyoF were the most abundant cell type at P7, no cells with this transcriptomic profile were observed at P21, while Hhip+ cells remained. This is consistent with MyoF playing a specific function during alveolarization [44].
(A) Cell type composition changes between P1, P7, and P21. Colors as in Figure 2A. (B) Negative log-P values of gene set enrichment analysis for pathways up-regulated (red, top) and down-regulated (blue, bottom) in adult versus early postnatal Alv F, Adv F, and airway SM. (C) Fraction of proliferative fibroblasts, myofibroblasts, and pericytes over total number of cells of that type over time. MyoF (proliferating or not) were not detected at P21. (D) In-situ imaging of proliferating myofibroblasts at P7 (left), of distal Tgfbi+ Hhip+ (dashed arrow) and Tgfbi+ Pdgfra+ cells (solid arrow), indicating two distinct ASM/MyoF subpopulations in agreement with the scRNA-Seq (center), and of Tgfbi+ Hhip+ Pdgfra- cells around conducting airways, consistent with ASM function (right). (E) Zoom-in of the embedding on early postnatal ASM and MyoF cells, colored by gene expression.
To understand the gene expression changes underlying the P7 to P21 transition, we performed pathway analysis of differentially expressed genes in ASM, AlvF and AdvF (Figure 4C). In ASM cells, this analysis indicated a loss of cell proliferation and reduced expression in pathways related to alveolar septation (trabecula formation, molecules associated with elastic fibers). In both AlvF and AdvF, up-regulated pathways included immune functions, suggesting an increased role in immunity for these cells in adults compared to neonates. Down-regulated pathways indicated a stabilization of the ECM.
In mice, during the first week of life the lung has transitioned from the saccular to the alveolar stage by P7, corresponding to an explosive growth in lung surface area which entails substantial cell proliferation. We therefore computed the fraction of proliferating cells within fibroblasts, MyoF, and pericytes at all three postnatal time points and observed a peak at P7 in all populations (Figure 4C), consistent with our previous report in endothelial cells [20]. At P21 there were no MyoF, proliferating pericytes, or proliferating fibroblasts.
We then used fluorescence in-situ RNA hybridization via RNA-Scope to define the location of ASM, MyoF, and proliferating MyoF within the P7 lung (Figure 4D). First, we detected Mki67+ Pdgfra+ Tgfbi+ proliferating MyoF in the distal lung (left, blue is nuclear DAPI stain). We then aimed to distinguish ASM from MyoF by probing Tgfbi, Pdgfra, and Hhip and detected both cell types in the distal lung parenchyma (middle): Tgfbi+ Hhip+ Pdgfra-cells (dashed arrow) and Tgfbi+ Pdgfra+ Hhip- cells (solid arrow). We also found Tgfbi+ Hhip+ cells lining conducting airways (right). This suggests that all three cell populations are present in the murine lung at P7: (i) Hhip- MyoF populating the distal lung, (ii) Hhip+ ASM cells around proximal airways, and (iii) Hhip+ cells that share a transcriptomic profile with ASM but, like MyoF, are found in the distal parenchyma. To better delineate the gene expression boundary between MyoF and ASM, we plotted the expression of several differentially expressed genes between these two populations on top of the embedding (Figure 4E) and found that the mutually exclusive expression of Hhip and Pdgfra is accompanied by additional gradients in gene expression, with one subset of ASM expressing Actc1 (top left corner) and one MyoF subset expressing Cd34 (bottom right corner), which has been implicated in MyoF activation [45,46].
Mural cells change gradually during perinatal development
The transcriptomes of mural cells changed more slowly and gradually between E18.5 and P21 compared to other mesenchymal cells. Relative cell type abundance of mural cells as a whole (Figure 1D) and of both pericytes and VSM (Figure 5A) increased over time. Like MyoF, pericytes were accompanied by a twin, proliferating population at P7 (Figure 4C). Unlike other mesenchymal cells, pericytes and VSM did not split into clear subclusters based on developmental stage. Nonetheless, the embedding of pericytes (Figure 5B) and VSM (Figure 5C) colored by time point indicates a transcriptomic shift over time (arrows are guides to the eye). To better understand this shift, we visualized the expression of ECM-related genes stratified by time points in pericytes (Figures 5D) and VSM (Figure 5E). In pericytes, Pdgfrb was stable over time while Postn showed widespread expression with a steady increase over time (green to yellow dots). Genes involved in cell contractility (Acta2, Myl9) were up-regulated at both E18.5 and P7. A number of genes were expressed only or mostly at P21 (Enpp2, Gap43, Npnt, Prelp, Cxcl14). In VSM cells, expression of Sparc was constant throughout, while elastin (Eln) and contractility genes peaked at P7. Sparcl1, Mgp and Timp3 instead dipped at P7. Like in pericytes, several genes show similar expression at E18.5 and P7 (Fbn1, Mfap4, Fn1). Other ECM modulators (Aspn, Mfge8 and Postn) peaked earlier in development. Overall, this evidence supports the notion of mural cells undergoing a process of slow maturation to gradually modulate the ECM.
(A) Mural cell type composition changes between E18.5 and P21. Colors as in Figure 2A. (B-C) Zoom-in of the embedding on pericytes (B) and vascular SM (C), colored by time point. Arrows are guides to the eye, indicating temporal shifts. (D-E) Dot plots divided by time point for pericytes (D) and vascular smooth muscle (E) VSM: vascular smooth muscle.
Cell-cell communications decrease after birth and offer insight into tissue organization
Mesenchymal cells often provide support to neighboring cell types. To quantify the magnitude of this process, we counted the number of ligand-receptor pairs in the CellPhoneDB database [47] expressed in at least 10% of a mesenchymal cell type and a partner cell type among immune and endothelial cells (EC) and found a hierarchical structure, especially at P7 (Figure 6A and Figure 6 - Supplementary 1). The largest number of interactions were found between AdvF and macrovascular EC, especially lymphatics. Pericytes play a natural mediating role between EC and parenchyma: we quantified the number of interactions with a more stringent cutoff (20% of cells expressing each gene) and confirmed more interactions with macro-than microvascular cell subtypes at all time points (Figure 6B). The number of interactions with most cell types decreased from E18.5 to P7, and this conclusion was further supported when we included other pairs of cell types beyond pericytes (Figure 6C), indicating less intense cell-cell communications at P7. To further understand at the single gene level what kind of communications pericytes are entertaining, we computed a dot plot of several highly expressed genes involved in paracrine communications (Figure 6D). Expression of partner genes across EC types was heterogeneous, with the highest expression of Pdgfb in Car4+ microvascular EC, while interactions with both MyoF and ASM included the angiotensin pathway (Agtr1a-Agt) which regulates and responds to blood pressure, the Wnt (Wnt5a-Fzd1) pathways which is a central developmental signaling hub, and Axl-Pros1, which might be important during lung injury [48].
(A) Number of potential ligand-receptor interactions between mesenchymal and other cell types at P7 with a threshold of 10% expressing cells. (B) Number of interactions between pericytes and endothelial cells at each time point, with a threshold of 20% expressing cells. (C) Histogram of percentage changes in number of interactions across many pairs of lung cell types from E18.5 to P7. (D) Dot plot of specific genes involved in pericyte interactions with the endothelium (red arcs connecting ligand-receptor pairs) and the rest of the mesenchyme (blue arcs). (E) Dot plot of Gap junction genes across many cell types in the neonatal lung. Immune cells do not express high levels of Gap junction genes and are excluded. Mesenchymal cell types are numbered as in Figure 2A. Other cell type short names: EC = endothelial cell, NP emb EC = nonproliferative embryonic EC, AT1/2 = epithelial alveolar type 1/2. Data sources: endothelial cells [20], epithelial cells [39], and immune cells [13].
Paracrine signaling via secreted ligands is not the only way cells communicate with one another. We investigated the developmental expression patterns of connexins, which enable direct exchange of biochemicals across the cytoplasmic membranes of two adjacent cells (Figure 6E). Four connexins were expressed in the neonatal pulmonary mesenchyme. Gja5 (aka CX40) marked arterial endothelial cells (ECs) while Gja4 (CX37) was expressed by both arterial ECs and pericytes, which are in physical contact, as well as a few additional cell types. Gjc1 (CX45) was expressed by multiple cell types including pericytes and MyoF. Gja1 (CX43) was not detected in ASM/MyoF precursors at E18.5 but was expressed in ASM at P1 and later time points, as well as in ECs of lymphatic vessels, and in epithelial alveolar cells types 1 and 2. Gja1 expression is consistent with homotypic junctions (i) within the contractile ASM rings surrounding airways; (ii) in the alveolar epithelium, mirroring previous studies in the eye lens epithelium [49]; and (iii) in lymphatic valves preventing lymph backflow [50]. Taken together, these data show that specific connexins are expressed by distinct groups of cell subtypes within the lung and at specific time points, indicating that the direct exchange of cytoplasmic biomolecules between adjacent cells through these channels might contribute to the orchestration of tissue development.
Hyperoxia delays development, dysregulates communications, and gives rise to a new male-specific population of fibroblasts
Having characterized the distinct mesenchymal populations during perinatal development, we used a murine model of bronchopulmonary dysplasia (BPD) in which newborn mice were exposed to 80% oxygen (hyperoxia) from birth through P7, whereupon mice were sacrificed and tissue obtained for scRNA-Seq and in-situ analysis. We quantified relative abundances of mesenchymal populations and observed a marked expansion of fibroblasts including proliferating ones at the expense of MyoF and pericytes (Figure 7A). The distribution of cell type abundances resembled that observed in healthy P1 lungs (Figure 7A, right), suggesting a global developmental arrest. To test this hypothesis, we computed 1,000 bootstraps over cells of cell type abundance distributions at P1, P7, and P7 hyperoxia and asked whether the hyperoxia composition was more similar to a healthy P1 lung (right, blue) or to a healthy P7 lung (left, red). In 999 out of 1,000 randomizations, hyperoxic mice were closer to P1 than healthy P7 (Figure 7B), which is consistent with previous histological analyses [51,52]
(A) Cell type composition in between normoxia and hyperoxia at P7 and in normoxia at P1. Colors as in Figure 2A. (B) Kernel density estimate of 1,000 bootstraps over cells (balanced at each time point) for the difference in distance on cell composition between hyperoxic mice and P7, minus hyperoxic mice and P1. A positive number indicated that the hyperoxic mesenchyme composition is closer to a healthy P1 mouse than a P7 one. Only 1 out of 1000 simulations showed hyperoxia closer to P7 than P1. (C) Embedding as in Figure 2A but coloring only P7 cells from healthy (red) and hyperoxia-exposed mice (purple), indicating a distinct separation between the two in many cell types. Several subtypes are highlighted by dashed lines. Cells from other time points are colored in grey. (D) Distribution of expression of Btg2 and Spry2 in alveolar fibroblasts at P1 (yellow solid), P7 healthy (red solid) and P7 hyperoxia-exposed mice (purple dashed). (E) Expression changes of genes involved in cytoskeleton, ECM adhesion, contractility, and ECM modulation from normoxia to hyperoxia across cell subtypes. (F) Heatmap of cell-cell interactions as in Figure 6A but for P7 mice exposed to hyperoxia. (G) Embedding as in Figure 2A but with the male hyperoxic fibroblast (MHF) population added in red. Other cells are colored in grey. (H) Dot plot of genes used to identify (top) and mark (bottom) MHF. (I) In-situ imaging of P7 lungs showing detection of Tubb3+ Wnt2+ MHF only in male, hyperoxia-exposed mice (M-H, top left) but not in females (F-H, top right) or healthy mice (bottom, M-N male left and F-N female right).
We next asked how gene expression within distinct kinds of mesenchymal cells is affected by hyperoxia. First, we found that cells from hyperoxia-exposed mice segregated in the embedding (Figure 7C). Embedding shifts represent changes in hundreds of genes at the same time. To exemplify these changes at the level of single genes, we plotted the distributions of expression of Btg2 and Spry2 in AlvF (Figure 7D, all cell types in Figure 7 - Supplementary 1). Btg2, an inhibitor of proliferation, is highly expressed at P1 (yellow line) and down-regulated in healthy P7 mice (red), but much less affected upon hyperoxia exposure (purple dashed line, Figure 7D, top), indicating residual inhibition leading to reduced cell proliferation, in agreement with the observed loss of several proliferative cell subtypes such as pericytes and MyoF (Figure 7A). Spry2, a modulator of fibroblast growth factor (FGF) pathway-related mesenchymal cell differentiation [53], is even higher in hyperoxia P7 mice than in P1 and P7 healthy mice (Figure 7D, bottom), indicating that hyperoxia can affect gene expression beyond a simple developmental delay.
Second, hyperoxia caused changes in the expression of genes involved in tissue stiffness and ECM secretion (Figure 7E). We measured an increase in Myh9, Tpm3, Itga5 across multiple cell types, and of Itga6 specifically in the remaining pericytes, suggesting the mesenchyme becomes more reactive to mechanical forces. Genes that contribute to or modulate the ECM were also mostly upregulated, often across multiple mesenchymal subtypes such as Mgp, Mmp2, Eng, and Mfge8. Expression of some genes was increased in a cell specific manner.
For example, Cd34 increased mostly in the remaining MyoF and to a lesser degree in both AlvF and AdvF, but did not change in other cells.
A direct consequence of the observed severe depletion in MyoF and pericytes and the changes in gene expression is the risk of an imbalanced paracrine signaling network across the tissue. Indeed, when we plotted the heatmap of interaction numbers in hyperoxic mice (Figure 7F) and compared it with healthy P7 mice (Figure 6A), we observed a more active but less organized network, suggesting widespread dysregulation of paracrine signaling within the mesenchyme and with other cell types.
Although most cells from hyperoxic mice fit into previously characterized cell types (Figure 2A), we identified a population of 55 cells specific to our male mouse that formed a novel cluster (Figure 7G, red circles, also visible in Figure 7C). Most of these cells clustered together next to AlvF. No such cells were found in the female hyperoxic mouse or in the normoxic mice. Therefore, we called these cells male hyperoxic fibroblasts (MHF). Like AlvF, MHF expressed Col13a1 and Wnt2 but not Col14a1, Tgfbi, or Pdgfrb (Figure 7H). They also expressed a number of marker genes, including contractility genes Acta1 and Actc1 and microtubule-related genes Tubb3 and Tuba1c, surface markers Cd9 and Aqp3, the enzyme Aldoart2 and the lectin Lgals3, which plays a role in fibrosis [54]. In-situ imaging of lungs from healthy and hyperoxia-exposed mice at P7 confirmed the presence of cells double positive for Tubb3 and Wnt2, which are markers of MHF, only in male hyperoxia-exposed mice (Figure 7I, top left). No double positive cells were observed in female (top right) or healthy mice of the same age (bottom). Overall, these data indicate that a new type of contractile cells related to AlvF emerge in the lung parenchyma of males only after hyperoxia exposure.
Discussion
In mammalian life, organismal survival depends upon sufficient lung development to permit establishment of an air-liquid interface, a ten-fold increase in pulmonary blood flow, and marked decrease in pulmonary arterial blood pressure as gas exchange moves suddenly from the placenta to the lungs [3]. Lung development continues after birth with a rapid increase in the number of alveoli [10]. While insight into the contributions of the cellular constituents of the lung across development is increasing [13,20], definitive information surrounding the composition, interactions and functions of lung mesenchymal cells remains elusive. In particular, there is still great ambiguity in the definition and naming of cell subtypes across the lung mesenchyme, especially during development [27,28,39]. The transcriptomic and imaging data presented in this study provide sufficient molecular texture to increase the consistency of mesenchymal cell type identification, motivating unambiguous annotations across development and in a model of neonatal lung injury.
VSM, though identified as a distinct cluster by Zepp et al. [55], have not been annotated specifically in most previous reports, including TMS [22]. Nonetheless, some TMS cells harmonize on top of our VSM (Figure 2B). Cohen et al. [39] and Liu et al. [28] report no VSM either, although in Liu et al.’ s embedding in Figure 1, Ebf1+ cells are compatible with a distinct subpopulation that may have been overlooked owing to clustering parameters [21]. Guo et al. report a cell type called Pericyte-1, which had low expression of Pdgfrb but high expression of Acta2, a marker of smooth muscle cells [27]. As shown in Figure 2 - Supplementary 1, that population aligns with our VSM, so these authors likely isolated the cluster successfully, but did not annotate it as VSM.
In contrast to VSM, there is broad acceptance of the existence of a heterogeneous population of lung fibroblasts. However, consensus surrounding the nomenclature has not been reached. Col13a1+ fibroblasts have been recently labeled matrix fibroblasts by Cohen et al. [39], MatrixFB-1 by Guo et al. [27], and lipofibroblasts by Liu et al. [28]. We named them alveolar fibroblasts or AlvF, in agreement with TMS [22] and others [16] and to highlight their anatomic location. Further, there is no direct evidence that AlvF always store large lipid reserves. Col14a1+ fibroblasts have been labeled MatrixFB-2 [27], intermediate fibroblasts [28], or aggregated into fibroblast progenitors [39], while we call them adventitial fibroblasts to emphasize their spatial niche within the lung parenchyma, consistent with TMS. The presence of a distinct pericyte fraction is also generally acknowledged, however Liu et al. opted to label them Ebf1+ fibroblasts at all developmental stages, as opposed to explicit recognition of pericytes. The embedding in Figure 1 of Cohen et al. appears to imply that pericytes are transcriptionally contiguous with fibroblast precursors, which is inconsistent with our data from E18.5 and later times. Guo et al. report two pericyte subtypes, with only Pericyte-2 mapping to our pericyte transcriptomic profile. None of those studies report proliferating pericytes, although they constitute >10% of pericytes in our P7 data (Figure 4C).
Myofibroblasts and ASM have also been annotated inconsistently. In healthy adult mice, TMS annotated a relatively abundant Tgfbi+ Hhip+ Pdgfra-Actc1- population labeled as MyoF and a rare Tgfbi-Hhip-Pdgfra-Actc1+ population labeled as ASM [22]. However, MyoF are absent or rare in healthy adults and emerge only with injury or disease [25]. Moreover, Hhip is considered an ASM marker [26] and Pdgfra appears necessary for MyoF function [56]. Independently, Liu et al. found specific expression of Hhip in lipofibroblasts at P7 (their Figure 4B) [28] and Guo et al. describe smooth muscle cells that matches our early ASM (Figure 2 - Supplementary 1) and two clusters of MyoF, one of which (MyoFB-2) does not express specific markers (their Figure 1b) [27]. Taken together, these observations indicate that previous annotations of MyoF may merit reconsideration on the basis of our present results. First, before birth, Pdgfra+ Hhip+ precursors are transcriptionally relatively homogeneous. Second, there is a large transcriptional shift in ASM/MyoF between P7 and P21, with the latter time point being essentially adult-like and much less heterogeneous, as supported by our harmonization with TMS (Figure 2B). Third, around birth a very diverse group of ASM/MyoF cells emerges (our clusters 8 and 10): while it becomes more proliferative at P7 (Figure 4C), its transcriptional profile is lost by P21. These cells are characterized by a subspecialization into Hhip+ Pdgfra- cells, which are found both in the distal lung and circling airways, and Hhip-Pdgfra+ cells, found exclusively in the distal lung (Figure 4D). A small fraction of the Hhip+ Pdgfra- cells also express Actc1, while a larger fraction of the Hhip-Pdgfra+ cells express Cd34; further transcriptional gradients are observed (Figure 4E), perhaps indicating additional layers of specialization.
The dramatic temporal shifts in the transcriptome of fibroblasts and ASM/MyoF resemble those seen in Mac I-II-III macrophages [13] and Car4- capillary endothelial cells [20]. In contrast, phenotypic shifts in mural cells were gradual but progressive, without marked changes between time points, reminiscent of alterations in multiple immune subtypes including monocytes (Mac V) [13] and in Car4+ capillaries [20,57], macrovascular cells, and lymphatic cells within the endothelium [20]. These findings indicate that select cell types perform a similar biological function throughout this developmental window, while others need to adapt transcriptionally to the distinct needs of each stage of tissue growth.
The present data identify a previously unknown role for a specific subpopulation of mesenchymal cells, as precursors of airway smooth muscle cells and myofibroblasts express corticotropin releasing hormone (Crh). Crh expression is absent in all other lung cells but high in ASM/MyoF precursors. The expression in these cells, which are positioned adjacent to the air spaces, is limited in time, present at E18.5 immediately prior to the onset of air-breathing life, and virtually lost by P1. For mammalian organisms, maturation of the surfactant system is biologically imperative [58]. Crh expression in precursors of airway smooth muscle cells and myofibroblasts supports the notion that these specific mesenchymal cells may sense and respond to select physiologic signals to release Crh and promote systemic cortisol release and maturation of alveolar type 2 cells [59] to increase surfactant production and release [41,59–61]. Moreover, we identified Hsdb11b1, a gene that encodes a protein that metabolizes cortisone into active cortisol [43,62], in fibroblasts, which may serve to amplify these effects locally. Overall, these observations suggest that specialized cells in the lung might be signaling to the brain, through corticotropin releasing hormone, to promote surfactant production and release at a time point when the molecule is most essential.
Correct tissue development requires orchestration by dozens of cell types via paracrine cell-cell interactions, which are infiuenced by ECM composition. Our finding that the number of detected interactions between mesenchymal and other cells show hierarchical structure (Figure 6A) mirrors our previous findings in the endothelium [20] and suggest that AdvF and lymphatic EC might play pivotal roles as “signaling hubs” of early alveolarization at P7. The observation that the number of interactions at P7 is consistently lower than E18.5 (Figure 6C) is remarkable considering that P7 is a time of rapid growth and cell proliferation (Figure 4C). Additional analyses to determine what specific interactions are lost will be necessary to fully understand the physiological implications of this result. Algorithmic improvements beyond current practices [47] are also needed to efficiently prioritize biologically relevant interactions from the huge tables of expressed ligand-receptor pairs found in scRNA-Seq data.
The expression patterns of connexins in our data support a coordinated action between pericytes and endothelium via Gja4 (encoding CX37). Differential expression of specific connexins enabled the distinction between VSM, which only express Gjc1 (aka CX45) and ASM, which express both Gjc1 and Gja1 (aka CX43), a gene directly involved in heart [63], bladder [64], and uterine contraction [65]. CX43 plays an important role in multiple lung cell types, including endothelial cells, alveolar type 1, alveolar type 2 cells, and fibroblasts. Global deletion of Gja1 in murine models leads to hypoplastic lungs, marked increases in the lung interstitium and neonatal death [66]. Taken together, these data indicate an untapped potential to utilize single cell transcriptomic data to investigate the direct cell-cell communication channels established by homo- and heteromultimers of connexins across tissues and development.
We employed a well-established, preclinical model of chronic lung disease of infancy to investigate whether neonatal hyperoxia possesses a specific molecular signature [12]. Approximately 75% of infants born at less than 750 grams possess chronic lung disease of infancy, initially described as bronchopulmonary dysplasia (BPD) in 1967 [11]. The transcriptomic data generated in this study was entirely consistent with this construct and indicates that throughout all mesenchymal cell types, the lung transcriptome of the P7 hyperoxic lung more closely resembled the normoxic P1 than the normoxic P7 lung (Figure 7B). Moreover, the observed increase in the expression of genes that help the cell adhere to an unstable ECM substrate provides indirect evidence of stiffness changes across the pulmonary ECM (Figure 7E), as previously observed in pulmonary fibrosis [67]. The cellular constituents of the lung demonstrated differential susceptibility to hyperoxia-induced injury. Pericytes, and not endothelial cells, were the most markedly affected by hyperoxia, with an almost complete loss of pericytes at 7 days of hyperoxia (Figure 7A). Pericytes seem to play a pivotal role in receiving signal from the pulmonary artery endothelial cells and delivering signals to the ASM/MyoF, cells that are critically important for postnatal secondary alveolar septation (Figure 6D). From a translational standpoint, the observation of differential vulnerability among the lung mesenchymal or vascular constituents has implications for emerging cell based, or exosome-driven therapeutic interventions [68].
Interestingly, in human infants, male infants generally are more frequently and severely affected than female infants [69]. The underlying reasons for the sexually dimorphic outcomes remain unknown. In these experiments, we identified a previously undescribed cluster of fibroblasts (MHF), present only in male hyperoxia exposed mice (Figures 7G-I). No novel cell types, including sex-specific ones, were observed in the immune [13] or endothelial [20] compartments in our previous studies. The spatial distribution of MHF, superimposed upon expression of cell contractility genes, support the proposition that the cells possess meaningful pathophysiologic significance. Positioned in the distal lung, these cells might negatively affect both lung compliance and gas exchange, both clinical hallmarks of BPD. The marked increase in Col13a1, Actc1 and Acta1, all of which likely contribute to lung stiffness, in the male hyperoxic fibroblasts compared to the early alveolar fibroblasts further highlights the potentially pathogenic role of these cells.
Overall, our data portray the perinatal lung mesenchyme as both extremely heterogeneous at each developmental stage and temporally dynamic, with entire transcriptomic profiles appearing or disappearing within 48 hours as well as upon hyperoxia exposure. Although much remains to be discovered, the combination of single cell resolution, careful cell type annotation, in-situ experiments, and comparison with publicly available data employed here provides a blueprint to tissue and disease profiling that could be used to systematically improve our understanding of mesenchymal cells across organs, conditions, and organisms.
Methods
Key resources table
Mouse lung cell isolation
C57BL/6 mice were obtained from Charles River Laboratories. For studies using E18.5, P1, and P7 murine lungs, pregnant dams were purchased, and pups aged prior to lung isolation. At E18.5, the dam was asphyxiated with CO2 and pups extracted. At P1, P7, and P21 pups were euthanized with euthanasia solution (Vedco Inc). Genetic sex of mice at developmental stages E18.5 and P1 was determined by performing PCR amplification of the Y chromosome gene Sry. P7 and P21 mice were sexed through identification of a pigment spot on the scrotum of male mice (Wolterink-Donselaar et al., 2009). For all timepoints, female and male mice were randomly selected for the studies. For all timepoints, except E18.5, the pulmonary circulation was perfused with ice cold heparin in 1x PBS until the circulation was cleared of blood. Lungs were minced and digested with Liberase (Sigma Aldrich) in RPMI for 15 (E18.5, P1, and P7) or 30 (P21) minutes at 37C, 200 rpm. Lungs were manually triturated and 5% fetal bovine serum (FBS) in 1x PBS was used to quench liberase solution. Red blood cells were lysed with 1x RBC lysis buffer (Invitrogen) as indicated by the manufacturer and total lung cells counted on Biorad cell counter (BioRad). Protocols for the murine studies adhered to American Physiological Society/US National Institutes of Health guidelines for humane use of animals for research and were prospectively approved by the Institutional Animal Care and Use Committee at Stanford (APLAC #19087).
Immunostaining and fluorescence-activated cell sorting (FACS) of single cells
Lungs were plated at 1 × 106 cells per well and stained with Fc block (CD16/32, 1:100, Tonbo Biosciences) for 30 min on ice. Cells were surface stained with the endothelial marker CD31 (1:100, eBioscience/ThermoFisher), epithelial marker Epcam (1:100, eBioscience/ThermoFisher), and immune marker CD45 (1:100, eBioscience/ThermoFisher) for 30 min on ice. The live/dead dye, Sytox Blue (Invitrogen), was added to cells and incubated for 3 min prior to sorting into 384-well plates (Bio-Rad Laboratories, Inc) prefilled with lysis buffer using the Sony LE-SH800 cell sorter (Sony Biotechnology Inc), a 100 μm sorting chip (Catalog number: LE-C3110) and ultra-purity mode. Single color controls were used to perform fluorescence compensation and generate sorting gates. 384-well plates containing single cells were spun down, immediately placed on dry ice and stored at −80C.
cDNA library generation using Smart-Seq2
RNA from sorted cells was reverse transcribed and amplified using the Smart-Seq2 protocol on 384-well plates as previously described [13]. Concentration of cDNA was quantified using Quant-it Picogreen (Life Technologies/Thermofisher) to ensure adequate cDNA amplification and cDNA was normalized to 0.4 ng/uL. Tagmentation and barcoding of cDNA was prepared using in-house Tn5 transposase and custom, double barcoded indices (Tabula Muris Consortium et al., 2018). Library fragment concentration and purity were quantified by Agilent bioanalyzer. Libraries were pooled and sequenced on Illumina NovaSeq 6000 with 2 × 100 base kits and at a depth of around 1 million read pairs per cell.
Data analysis and availability
Sequencing reads were mapped against the mouse genome (GRCm38) using STAR aligner (Dobin et al., 2013) and gene expression was quantified using HTSeq (Anders et al., 2015). FZ has been the main developer of HTSeq since 2016. To coordinate mapping and counting on Stanford’ s high-performance computing cluster, snakemake was used (Koster and Rahmann, 2012). Gene expression count tables were converted into loom objects (https://linnarssonlab.org/loompy/) and cells with less than 50,000 uniquely mapped reads or less than 400 genes per cell were discarded. Doublets were discarded by excluding small clusters and single cells that coexpress markers for incompatible cell tyes. Counts for the remaining NNN cells were normalized to counts per million reads. For t-distributed stochastic embedding (t-SNE) (van der Maaten and Hinton, 2008), 500 features were selected that had a high Fano factor in most mice, and the restricted count matrix was log-transformed with a pseudocount of 0.1 and projected onto the top 25 principal components using scikit-learn (Pedregosa et al., 2011). Unsupervised clustering was performed using Leiden (C++/Python implementation) (Traag et al., 2019). Singlet (https://github.com/iosonofabio/singlet) and custom Python3 scripts were used: the latter are available at https://github.com/iosonofabio/lung_neonatal_mesenchymal/. Pathway analysis on the differentially expressed genes was performed via Metascape (Zhou et al., 2019) on the top 100 most differentially expressed genes for each comparison: precursors versus sample-balanced joint progenies (Fig 3), subsequent time points within each cluster (Fig 4 and 5). The most enriched pathways against a permutation test are shown ordered by significance from top to bottom (negative log of the P-value). Batch-corrected KNN [23] was used to compare our P21 data with Tabula Muris Senis [22]. Raw fastq files, count tables, and metadata are available on NCBI’ s Gene Expression Omnibus (GEO) website: GSE172251.
In-situ validation using RNAscope and immunofluorescence (IF)
Embryonic and post-natal mice were euthanized as described above. Female and male mice were randomly selected from the litter, and at least two litters were used to source the lung tissue for all validation studies. E18.5 lungs were immediately placed in 10% neutral buffered formalin following dissection. P1, P7, and P21 murine lungs were perfused as described above, and P7 and P21 lungs inflated with 2% low melting agarose (LMT) in 1xPBS and placed in 10% neutral buffered formalin. Following 20 hr incubation at 4C, fixed lungs were washed twice in 1xPBS and placed in 70% ethanol for paraffin-embedding. In situ validation of genes identified by single cell RNA-seq was performed using the RNAscope Multiplex Fluorescent v2 Assay kit (Advanced Cell Diagnostics) and according to the manufacturer’ s protocol. Formalin-fixed paraffin-embedded (FFPE) lung sections (5 μm) were used within a day of sectioning for optimal results. Nuclei were counterstained with DAPI (Life Technology Corp.) and extracellular matrix proteins stained with hydrazide [70]. Opal dyes (Akoya Biosciences) were used for signal amplification as directed by the manufacturer. Images were captured with Zeiss LSM 780 and Zeiss LSM 880 confocal microscopes, using 405 nm, 488 nm, 560 nm and 633 nm excitation lasers. For scanning tissue, each image frame was set as 1024 × 1024 and pinhole 1AiryUnit (AU). For providing Z-stack confocal images, the Z-stack panel was used to set z-boundary and optimal intervals, and images with maximum intensity were processed by merging Z-stacks images. For all both merged signals and split channels were collected.
Statistical analyses
To identify differentially expressed genes within cell populations, Kolmogorov Smirnov tests on the distributions of gene expression were performed on all genes, and either the genes with the largest absolute value of the test statistic or the genes above a certain KS statistic (e.g. 0.3) with the largest fold change were chosen.
For context, on top of each heatmap is plotted the relative abundance of each subtype (e.g. AlvF) over the total of the cell type (e.g. mesenchymal), with a log-scaled y axis. The cell types are: endothelial, immune, and mesenchymal. Endothelial data from [20], immune data from [13].
While hyperoxia delayed down-regulation of expression in some cell types (e.g. ASM), it was associated with distinct up-regulation in MyoF and pericytes.
Acknowledgements
We thank Sai Saroja Kolluru (Stanford University) for assistance with library submission to the Chan Zuckerberg Biohub, Yuan Xue (Stanford University) for assistance with the initial single cell RNA-seq data acquisition, Astrid Gillich for technical support with the RNAscope experiments, and Maya Kumar for providing hydrazide. We also thank the Stanford Shared FACS Facility, Lisa Nichols, Meredith Weglarz, and Tim Knaak for assistance with the flow cytometry instrumentation and antibody panel design. Flow cytometry data was collected on an instrument in the Stanford Shared FACS Facility obtained using NIH S10 Shared Instrument Grant (S10RR027431-01). This work was supported by National Institutes of Health grants HL122918 (CMA), HD092316 (CMA, DNC), the Stanford Maternal Child Health Institute Tashia and John Morgridge Faculty Scholar Award (CMA), the Stanford Center of Excellence in Pulmonary Biology (DNC), Bill and Melinda Gates Foundation (SRQ), and the Chan Zuckerberg Biohub (DNC. and SRQ). MAS is supported by the NSF-GRFP.
References
