Remarkable sex-specific differences at single-cell resolution in neonatal hyperoxic lung injury

Abstract

Exposure to supraphysiological concentrations of oxygen (hyperoxia) predisposes to bronchopulmonary dysplasia (BPD), which is characterized by abnormal alveolarization and pulmonary vascular development, in preterm neonates. Neonatal hyperoxia exposure is used to recapitulate the phenotype of human BPD in murine models. Male sex is considered an independent predictor for the development of BPD, but the main mechanisms underlying sexually dimorphic outcomes are unknown. Our objective was to investigate sex-specific and cell-type specific transcriptional changes that drive injury in the neonatal lung exposed to hyperoxia at single-cell resolution and delineate the changes in cell-cell communication networks in the developing lung. We used single-cell RNA sequencing (scRNAseq) to generate transcriptional profiles of >35,000 cells isolated from the lungs of neonatal male and female C57BL/6 mice exposed to 95% $\mathrm{\text{F}}{\mathrm{\text{I}}}_{{\mathrm{\text{O}}}_2}$ between PND1-5 (saccular stage of lung development) or normoxia and euthanized at PND7 (alveolar stage of lung development). ScRNAseq identified 22 cell clusters with distinct populations of endothelial, epithelial, mesenchymal, and immune cells. Our data identified that the distal lung vascular endothelium (composed of aerocytes and general capillary endothelial cells) is exquisitely sensitive to hyperoxia exposure with the emergence of an intermediate capillary endothelial population with both general capillaries (gCap) and aerocytes or alveolar capillaries (aCap) markers. We also identified a myeloid-derived suppressor cell population from the lung neutrophils. Sex-specific differences were evident in all lung cell subpopulations but were striking among the lung immune cells. Finally, we identified that the specific intercellular communication networks and the ligand-receptor pairs that are impacted by neonatal hyperoxia exposure.

INTRODUCTION

The male disadvantage for neonatal mortality and major morbidities in preterm neonates is well known (1–4). Sex-chromosome-based, hormonal, imprinting, or epigenetic mechanisms may modulate male susceptibility or female resilience (5). Susceptibility to diseases and the subsequent repair and recovery may differ between the sexes due to the sex-based differences at baseline or pathways activated or inhibited in response to the injurious stimuli.

Poor lung health in premature neonates is a major factor underlying an inferior quality of life, economic costs related to health care, and risk of developing adult-onset chronic lung diseases (6, 7). Respiratory morbidity including the development of bronchopulmonary dysplasia (BPD) is common in preterm neonates with long-term impact. Even in the postsurfactant era, extremely premature male neonates (born between 24 and 26 wk of gestation) displayed a significantly increased risk of respiratory complications (8). The incidence of respiratory distress syndrome (RDS), BPD, and moderate to severe BPD was significantly increased in males after adjusting for multiple confounding factors, particularly in the 750–999 g birth weight group (9). Similarly, the need for respiratory support, respiratory medications, and home oxygen use was higher in males (10, 11). Male sex was an independent risk for tracheostomy among preterm neonates and in infants who received a tracheostomy, an independent predictor of mortality (12). Hospital readmissions and outpatient visits due to respiratory issues in the 1–4 yr and 5–9 yr age groups were also increased in male premature neonates (13). Male sex is thus an independent predictor for the development of BPD and the subsequent morbidities related to lung disease in premature infants, but the underlying mechanisms behind these sexually dimorphic outcomes are unknown.

Animal models for human BPD use several approaches to replicate the major findings of alveolar simplification and abnormal vascular remodeling in the murine lung including postnatal hyperoxia exposure in the term mouse (14, 15). Exposure to hyperoxia and the resulting oxidative stress to the developing lung contributes to the pathophysiology of this disease. The murine lung is in the saccular stage of lung development from birth to postnatal day (PND4-5), which is equivalent to 26–36 wk in human preterm neonates (16). Most preterm neonates are exposed to varying degrees of hyperoxia in the neonatal intensive care unit during this period, with the sicker neonates receiving the highest concentrations of oxygen support (17).

Our laboratory and others have reported sex-specific differences in alveolar and vascular development in neonatal mice using this model (18–20). Furthermore, we have highlighted differences involving epigenetic mechanisms, micro-RNA-mediated effects, and the role of sex-chromosome and gonadal hormones in this model, all of which show striking sex-specific differences (21–24). However, the sex-specific differences in individual lung cell subpopulations at single-cell resolution at the saccular stage of lung development in the setting of neonatal hyperoxia exposure are not known. In this investigation, we tested the hypothesis that there are marked sex-specific differences in different lung cell subpopulations in the developing murine lung during the saccular stage of lung development upon exposure to hyperoxia.

MATERIALS AND METHODS

Mice

All animal experiments were performed under an approved protocol by the IACUC at the Children’s Hospital of Philadelphia. Timed pregnant C57BL/6N WT mice were obtained from Charles River Laboratories (Wilmington, NC). The sex in neonatal mouse pups was determined by both the anogenital distance and pigmentation in the anogenital region method. In neonatal male mice, a pigmented spot on the scrotum is visible to the naked eye from postnatal day 1 (PND1), whereas female pups lack visible pigmentation in the anogenital region. This was also verified by the expression of Y-chromosome transcript abundance in male samples.

Mouse Model of BPD

An arrest of alveolarization was induced in mouse pups by exposure to hyperoxia (95% O₂), as described previously (24). Mouse pups from multiple litters were pooled before being randomly and equally redistributed to two groups, one group exposed to room air (21% O₂) and the other group exposed to hyperoxia (95% O₂), within 12 h of birth for 5 days. The dams were rotated between air- and hyperoxia-exposed litters every 24 h.

Lung Isolation for Single-Cell Sequencing

Mice were euthanized on postnatal day 7 (PND7) with intraperitoneal pentobarbital. The right ventricle was perfused with ice-cold PBS and the lungs were harvested quickly and homogenized and dissociated using the Miltenyi MACS lung dissociation kit (Miltenyi Biotec, Cat. No. 130-095-927), using the GentleMacs program m-Lung-1 on the AUTO MaCs. The lungs were incubated for 20 min at 37°C under continuous rotation and then run on the gentleMACS m-Lung-2 program on the gentleMACS (Miltenyi Biotec, Cat. No. 130-095-937). This was followed by brief centrifugation and the passage of cells through a 70-µm filter, washed, and centrifuged again, and the pellet was treated with ACK lysis buffer, centrifuged, and resuspended in sorting buffer and passed through a 40-µm filter. Live cells were sorted and subjected to single-cell RNA sequencing as described in the section scRNA-Seq Library Preparation and Sequencing.

scRNA-Seq Library Preparation and Sequencing

Single-Cell Gene Expression Library was prepared according to Chromium Single-Cell Gene Expression 3v3.1 kit (×10 Genomics). In brief, single cells, reverse transcription (RT) reagents, gel beads containing barcoded oligonucleotides, and oil were loaded on a chromium controller (×10 Genomics) to generate single-cell GEMS (Gel Beads-In-Emulsions) where full-length cDNA was synthesized and barcoded for every single cell. Subsequently, the GEMS are broken and cDNA from every single cell is pooled. Following cleanup using Dynabeads MyOne Silane Beads, cDNA is amplified by PCR. The amplified product is fragmented to optimal size before end-repair, A-tailing, and adaptor ligation. Final library was generated by amplification. 10X Genomics Cell Ranger [version] “cellranger count” was used to perform alignment, filtering, barcode counting, and unique molecular identifier (UMI) counting. The raw data has been uploaded to NCBI GEO; GSE211356. All Supplemental material is available at https://doi.org/10.6084/m9.figshare.20520525.

Sequencing, Data Processing, Quality Control, Integration, and Cluster Annotation

For each sample, a Seurat object was created using Cellranger output matrices using Read10X method. SCTransformV2 and Get clusters (RunPCA, RunUMAP, RunTSNE, FindNeighbors, FindClusters) was run. Then we filtered using SoupX version 1.6.1 (25) with a contamination factor of 5%. We used PercentageFeatureSet to get the percent of mitochondrial genes in each cell. Resulting cells were filtered using ≥ 500 molecular identifiers (nUMI ≥ 500), removing upper number of UMI outliers (nUMI ≤ outlier_threshold_nUMI), >250 detectable genes (nGene ≥ 250), removing upper number of gene outliers (nGene ≤ outlier_threshold_nGene), log10GenesPerUMI > 0.80, < 5% of transcripts coming from mitochondrial genes (mitoRatio < 0.05). Genes were filtered, only keeping those genes expressed in more than 10 cells. Genes ‘Gm42418’, ‘S100a8’, ‘S100a9’ (26, 27) were removed from Seurat object. We ran SCTransformV2 and Get clusters, resolution 0.6 was chosen. Doublets were detected and removed using DoubletFinder version 2.0.3 (28). paramSweep_v3, summarizeSweep, find.pK, doubletFinder_v3. Get clusters ran again. We aggregated the samples using SelectIntegrationFeatures with number of features to return set to 3,000, PrepSCTIntegration, FindIntegrationAnchors with sample 1 as reference (Female Hyperoxia), and IntegrateData using SCT as normalization method. Ran Get clusters again and chose resolution 0.6. Ran PrepSCTFindMarkers with default parameters and used the FindMarkers function for each cluster with parameters set to Only return positive markers using Wilcoxon test and SCT assay. Differential expression between conditions was found using Seurat’s FindMarkers function. Genes detected in 25% of cells with an FDR (adjusted P value) value of <0.05 and a log2fold change of >0.3 were reported as significant in each comparison. Significant genes were then processed using enrichR (v. 3.0) (29) along fgsea (v. 1.22.0) (30) to find enriched biological pathways present. RNA velocity estimation was done using cellranger output files, then sorted with samtools (1.14), then used velocyto (0.17) (31), then scVelo (0.2.4) (32). Cell-cell communication was analyzed using Cellchat (v. 1.4.0) (33) on room air and hyperoxia samples separately with the software’s default parameters. See Table 1 for the software list.

Table 1.

Name, version, and source of software used

Software Name	Version	Source
dplyr	1.0.9	(34) Wickham H, François R, Henry L, Müller K; RStudio. dplyr: A Grammar of Data Manipulation (Online). https://CRAN.R-project.org/package=dplyr.
tidyverse	1.3.1	(35) Wickham H, Averick M, Bryan J, Chang W, McGowan L, François R, Grolemund G, Hayes A, Henry L, Hester J, Kuhn M, Pedersen T, Miller E, Bache S, Müller K, Ooms J, Robinson D, Seidel D, Spinu V, Takahashi K, Vaughan D, Wilke C, Woo K, Yutani H. Welcome to the tidyverse. J Open Source Softw 4: 1686, 2019. doi: 10.21105/joss.01686.
Seurat	4.1.1	(36) Hao Y, Hao S, Andersen-Nissen E, Mauck WM 3rd, Zheng S, Butler A, Lee MJ, Wilk AJ, Darby C, Zager M, Hoffman P, Stoeckius M, Papalexi E, Mimitou EP, Jain J, Srivastava A, Stuart T, Fleming LM, Yeung B, Rogers AJ, McElrath JM, Blish CA, Gottardo R, Smibert P, Satija R. Integrated analysis of multimodal single-cell data. Cell 184: 3573–3587.e29, 2021. doi: 10.1016/j.cell.2021.04.048.
patchwork	1.1.1	(37) Pedersen T. patchwork: The Composer of Plots (Online). https://CRAN.R-project.org/package=patchwork.
openxlsx	4.2.5	(38) Schauberger P, Walker A. openxlsx: Read, Write and Edit xlsx Files (Online). https://CRAN.R-project.org/package=openxlsx.
ggplot2	3.3.6	(39) Wickham H. ggplot2: Elegant Graphics for Data Analysis. New York: Springer-Verlag, 2016. doi: 10.1007/978-0-387-98141-3.
readxl	1.4.0	(40) Wickham H, Bryan J. readxl: Read Excel Files (Online). https://CRAN.R-project.org/package=readxl.
gridExtra	2.3	(41) Auguie B. gridExtra: Miscellaneous Functions for “Grid” Graphics (Online). https://CRAN.R-project.org/package=gridExtra
scales	1.2.0	(42) Wickham H, Seidel D. scales: Scale Functions for Visualization (Online). https://CRAN.R-project.org/package=scales.
data.table	1.14.2	(43) Dowle M, Srinivasan A. data.table: Extension of ‘data.frame’ (Online). https://CRAN.R-project.org/package=data.table.
magrittr	2.0.3	(44) Bache S, Wickham H. magrittr: A Forward-Pipe Operator for R (Online). https://CRAN.R-project.org/package=magrittr.
plyr	1.8.7	(45) Wickham H. The split-apply-combine strategy for data analysis. J Stat Softw 40: 1–29. https://www.jstatsoft.org/v40/i01/.
cowplot	1.1.1	(46) Wilke C (2020). _cowplot: Streamlined Plot Theme and Plot Annotations for ‘ggplot2’_. R package version 1.1.1. https://CRAN.R-project.org/package=cowplot.
ggrepel	0.9.1	(47) Slowikowski K (2021). _ggrepel: Automatically Position Non-Overlapping Text Labels with ‘ggplot2’_. R package version 0.9.1. https://CRAN.R-project.org/package=ggrepel.
DoubletFinder	2.0.3	(28) McGinnis CS, Murrow LM, Gartner ZJ. DoubletFinder: doublet detection in single-cell RNA sequencing data using artificial nearest neighbors. Cell Syst 8: 329–337.e4, 2019. doi: 10.1016/j.cels.2019.03.003.
stringr	1.4.0	(48) Wickham H (2022). stringr: Simple, Consistent Wrappers for Common String Operations. http://stringr.tidyverse.org, https://github.com/tidyverse/stringr.
CellChat	1.4.0	(49) Jin S (2022). _CellChat: Inference and analysis of cell-cell communication from single-cell transcriptomics data_. R package version 1.4.0.
ggalluvial	0.12.3	(50) Brunson JC, Read QD. ggalluvial: Alluvial Plots in ‘ggplot2’ (Online). http://corybrunson.github.io/ggalluvial/.
NMF	0.24.0	(51) Gaujoux R, Seoighe C. A flexible R package for nonnegative matrix factorization. BMC Bioinformatics 11: 367, 2010. https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-11-367.ComplexHeatmap: (52) Gu Z. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics 32: 2847–2849, 2016. doi: 10.1093/bioinformatics/btw313.
ComplexHeatmap	2.13.1	(52) Gu, Z. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics 32: 2847–2849, 2016. doi:10.1093/bioinformatics/btw313.
future	1.26.1	(53) Bengtsson H. A unifying framework for parallel and distributed processing in r using futures. R J 13: 273–291, 2021. doi: 10.32614/RJ-2021-048.
fgsea	1.22.0	(30) Korotkevich G, Sukhov V, Budin N, Shpak B, Artyomov MN, Sergushichev A. Fast gene set enrichment analysis (Preprint). bioRxiv 2021. doi:10.1101/060012.
enrichR	3	(54) Jawaid W. enrichR: Provides an R Interface to ‘Enrichr’ (Online). https://CRAN.R-project.org/package=enrichR.
msigdbr	7.5.1	(55) Dolgalev I. _msigdbr: MSigDB Gene Sets for Multiple Organisms in a Tidy Data Format (Online). https://CRAN.R-project.org/package=msigdbr. Cell communication and regulon analysis
python	3.7.12	(56) Van Rossum, G., & Drake, F. L. (2009). Python 3 Reference Manual. Scotts Valley, CA: CreateSpace.
pandas	1.3.5	(57) Reback J, McKinney W, van den Bossche J, Augspurger T, Cloud P, Hawkins S, Roeschke M, Klein A, Petersen T, Hoefler P, Tratner J, She C, Ayd W, Naveh S, Garcia M, Darbyshire JHM, Schendel J, Hayden A, Shadrach R, Saxton D, Gorelli ME, Li F, Zeitlin M, Jancauskas V, McMaster A, Battiston P, Seabold S. pandas-dev/pandas: Pandas 1.3.5 (v1.3.5) (Online). Zenodo, 2021. https://doi.org/10.5281/zenodo.5774815.
numpy	1.21.6	(58) Harris CR, Millman KJ, van der Walt SJ, Gommers R, Virtanen P, Cournapeau D, Wieser E, Taylor J, Berg S, Smith NJ, Kern R, Picus M, Hoyer S, van Kerkwijk MH, Brett M, Haldane A, Del Río JF, Wiebe M, Peterson P, Gérard-Marchant P, Sheppard K, Reddy T, Weckesser W, Abbasi H, Gohlke C, Oliphant TE. Array programming with NumPy. Nature 585: 357–362 (2020). doi: 10.1038/s41586-020-2649-2.
matplotlib	3.5.2	(59) Hunter JD. Matplotlib: a 2D graphics environment. Comput Sci Eng 9: 90–95, 2007. doi: 10.1109/MCSE.2007.55.
scvelo	0.2.4	(32) Bergen V, Lange M, Peidli S, Wolf FA, Theis FJ. Generalizing RNA velocity to transient cell states through dynamical modeling. Nat Biotechnol 38: 1408–1414, 2020. doi: 10.1038/s41587-020-0591-3.
Samtools	1.14	(60) Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, Marth G, Abecasis G, Durbin R; 1000 Genome Project Data Processing Subgroup. The sequence alignment/map format and SAMtools. Bioinformatics 25: 2078–2079, 2009. doi: 10.1093/bioinformatics/btp352
anndata	0.8.0	(61) Cannoodt R. anndata: ‘anndata’ for R (Online). https://anndata.dynverse.org and https://github.com/dynverse/anndata.
velocyto	0.17.17	(62) La Manno G, Soldatov R, Zeisel A, Braun E, Hochgerner H, Petukhov V, Lidschreiber K, Kastriti ME, Lönnerberg P, Furlan A, Fan J, Borm LE, Liu Z, van Bruggen D, Guo J, He X, Barker R, Sundström E, Castelo-Branco G, Cramer P, Adameyko I, Linnarsson S, Kharchenko PV. RNA velocity of single cells. Nature 560: 494–498, 2018. doi: 10.1038/s41586-018-0414-6; PMID: 30089906; PMCID: PMC6130801.

In Situ Hybridization

In situ hybridization was performed using the RNAscope Multiplex Fluorescent Reagent Kit v2 (Advanced Cell Diagnostics, Hayward, CA). Samples were treated with Pecam (316721), Inhba (455871-C4), Ednrb (473801-C2), Peg3 (492581-C3), Aplnr (436171), and Ly6A (427571-C2) probes (Advanced Cell Diagnostics, Hayward, CA) for 2 h at 4°C. For each round, samples were run alongside a positive control slide treated with housekeeping genes Polr2A, Ppib, Ubc, and Hprt (321811) and a negative control slide that was treated with dapB (a bacteria gene) (321831). Nuclei were counterstained with DAPI and autofluorescence was reduced using TrueBlack Plus Lipofuscin Autofluoresne Quencer 40X in DMSO (Biotum: 23014) for 30 min then washed in PBS. Slides were mounted in EverBrite TrueBlack Hardset Mounting Medium (Biotium: 23017). The following day, slides were imaged using Leica Bmi8 Thunder Imager microscope.

RESULTS

Single-Cell RNA Sequencing of Room Air- and Hyperoxia-Exposed Lungs

To generate a comprehensive study of the sex-specific transcriptional changes after hyperoxia exposure in the early lungs, 1-day-old male and female mice (day of birth) were exposed to room air (21% O₂) or hyperoxia (95% O₂) until day 5. Mice were allowed to recover in room air until day 7 when lungs were collected and processed for single-cell RNA sequencing (Fig. 1A). A total of 35,934 cell transcriptomes were profiled with cell clustering showing capture overlap between room air and hyperoxia cells in both male and female lungs. Taking advantage of dimensionality reduction techniques, such as UMAP projections (63), we were able to visualize each cell transcriptome in 2-D where cells with similar expression profiles cluster together (Fig. 1B). Epcam+ cells were identified as epithelial, Cldn5+ as endothelial, Ptprc+ as immune and Col1a2+ as mesenchymal cells (64) (Fig. 1, C and D). Consistent cell recovery was seen between each single-cell sequencing run but unfortunately, cell populations such as club cells and nonciliated epithelial cells were lost during the mechanical and enzymatic dissociation. In addition, 22 clusters were identified by comparing their expression profile to the published literature (65–70) (Fig. 1, C and E). The number of cells sequenced in each group and the highly expressed genes in each cluster are included in Supplemental Table S1.

Figure 1.

Capturing the effects of hyperoxia in the lung at single-cell resolution. A: experimental design and timeline. Mouse pups (C57BL6) from multiple litters were pooled before being randomly and equally redistributed to two groups, one group exposed to room air (21% O₂) and the other group exposed to hyperoxia (95% O₂), within 12 h of birth for 5 days, and euthanized on PND7. Equal number of pooled viable lung cells from three mice were used for each group. B: UMAP of sequenced cells labeled by experimental condition (room air in teal or hyperoxia in red) and split by sex (male or female). C: UMAP overview of cell clusters identified based on gene expression. Dotted groups encompass large cell groups of distinct populations separated by endothelial, epithelial, immune, and mesenchymal. D: UMAP plot of expression levels of cell markers for epithelial (Epcam), endothelial (Cldn5), immune (Ptprc), and mesenchymal (Col1a2) cell populations. E: dot plot showing expression levels (dot color) and percent of cells expressing each gene (dot size) of previously validated marker genes for each cluster identified. [Image created with BioRender.com and published with permission.]

Endothelial Expression Changes in Response to Hyperoxia

Sub clustering of 9,394 endothelial cells identified seven distinct populations (Fig. 2A). Alveolar-specific general capillaries (gCap) and aerocytes or alveolar capillaries (aCap) make up over 75% of the cells captured. gCap cells express the transporter for lipoprotein lipase Gpihbp1, while actively proliferating gCap cells express the cell cycle genes such as Hist1h1b (Fig. 1E) (65). aCap cells were identified by their expression of carbonic anhydrase 4 or Car4 (Fig. 1E) (65). All large vessels were marked by expression of von Willebrand factor (vWF, Fig. 1E) (71) with specific expression of Gja4 in arterial, Nr2f2 in venous, and Prox1 (36) in lymphatic cells. Interestingly, a second population of aCap cells, which we called reactive aCap, were characterized by the upregulation of Inhba (64) and greatly expand after hyperoxia exposure and represent more than 50% of the total lung endothelial cells (Fig. 2, B and C).

Figure 2.

Gene expression changes in endothelial cells in response to hyperoxia. A: UMAP of sequenced lung endothelial cells identified seven distinct clusters. B: UMAP of lung endothelial cells showing the seven distinct clusters in room air and hyperoxia. C: changes in the relative contribution of different lung endothelial cell subpopulations in room air and hyperoxia. D: trajectory analysis of lung endothelial cells showing reactive aCaPs emerging from gCaPs and leading to aCaPs. E: gene expression similarities and differences between aCaP, reactive aCaPs, and gCaPs in the distal lung microvasculature. F: number of cells sequenced and the number of up- and downregulated genes in aCaP/reactive aCaP, gCaP, and arterial endothelial cells upon exposure of the neonatal lung to hyperoxia. G: volcano plots showing the differentially expressed genes in arterial, gCaP, and aCaP/reactive aCaP endothelial cells. All the results above are obtained from equal number of pooled viable lung cells from three mice. H: in situ hybridization showing the increased expression of Inhba in lung endothelial cells in response to hyperoxia and violin plots showing increased expression of Inhba in aCaP and reactive aCaP endothelial cells. Representative image from n = 3/group. aCap, aerocytes or alveolar capillaries; gCap, general capillaries.

To understand the expression dynamics between the reactive aCaps and the rest of the endothelial cell population, we employed RNA Velocity (62) and partition-based graph abstraction. RNA Velocity results were consistent with previously published data showing that venous cells give rise to lymphatic endothelium (72) as well as proliferative gCap giving rise to mature gCap cells (65). Interestingly, we observed the gCap cluster specifying into reactive aCap cells that then further connected into aCap cells, suggesting that reactive aCap is an intermediate less differentiated cell state that occurs in response to injury (Fig. 2D). Gene expression in aCaPs, gCaPs, and reactive aCaPs is shown in a heat map in Fig. 2E. Reactive aCaPs have an expression profile that is similar in some respects to aCaPs (high Car4 and high Apln) and some to gCaPs (high CD93).

The differential expression shows the largest number of genes changing in aCap/reactive aCaP cells (Fig. 2F-G). We observed that 337 genes upregulated and 290 downregulated in hyperoxia. Inhba expression was increased in reactive aCaPs upon exposure to hyperoxia and this was validated using single-molecule fluorescent in situ hybridization (smFISH), with higher expression of Inhba among Pecam positive cells in the hyperoxia-exposed lung compared to room air controls (Fig. 2H). Our finding was reported previously by Hurskainen et al. (64) in a newborn hyperoxia model after 14 days of hyperoxia exposure (85% $\mathrm{\text{F}}{\mathrm{\text{I}}}_{{\mathrm{\text{O}}}_2}$). Another upregulated gene target that was identified and validated in the reactive aCaP population was paternally expressed gene 3 (Peg 3). Peg 3 activates autophagy in endothelial cells and attenuates angiogenesis through thromspondin-1 (73–76). We show increased expression of Peg3 in aCaP cells (labeled by endothelial receptor b; Ednrb) in the hyperoxia-exposed lung (Supplemental Fig. S1A).

The second endothelial population with the most gene expression changes were the gCap cells, where hyperoxia-induced the downregulation of 30 genes and upregulation of 40 genes (Fig. 2F). As the stem cell-like population, gCap cells are involved during lung repair (65) and their total number decrease lungs exposed to hyperoxia (Fig. 2C). Interestingly, we observed the upregulation of Ly6a or Sca1 (stem cell antigen-1) in gCap cells (Aplnr+) after hyperoxia (Supplemental Fig. S1B). Sca1 is used to mark endothelial progenitor cells (77) and is expressed in the large vessel and capillary endothelium in the lung (78). Sca 1+ endothelial gCaP cells may contribute to endothelial repair after injury (79, 80).

In arterial cells, the endothelial homeostasis regulator Socs3 (Suppressor of cytokine signaling 3) (81) is downregulated in hyperoxia whereas pro-angiogenic factor Ecm1 (extracellular matrix protein 1) (82) is upregulated (Fig. 2G).

Biological pathways that were differentially regulated in aCaPs/reactive aCaPs and gCaP cells in the distal capillary endothelium were identified. The upregulated pathways that were common to male and female aCaPs/reactive aCaPs included oxidative phosphorylation, ECM-receptor interaction, cytokine-mediated receptor signaling, and mTORC1 signaling (Supplemental Fig. S1C). The switch from glycolysis to oxidative phosphorylation in endothelial cells under pathological conditions is well described (83–85). Downregulated pathways included regulation of ERK1 and ERK2 and TGF-β receptor signaling pathway. ERK signaling plays a crucial role in angiogenesis and in maintaining endothelial quiescence (86–88). The gCaPs were positively enriched for pathways related to p53 signaling, mesenchymal transition, whereas regulation of cellular proliferation was downregulated (Supplemental Fig. S1D). Differentially expressed genes in all endothelial cell subpopulations are listed in Supplemental Table S2.

Sex-Specific Response to Hyperoxia in Endothelial Cells

Next, we separated our cells by sex to explore the gene expression changes of the lung endothelium in response to hyperoxia injury (Fig. 3A). Differentially expressed genes up- and downregulated in aCaps/reactive aCaPs, gCaPs, and arterial endothelial cells in male and female lungs are shown in Fig. 3B. Interestingly, the number of sex-specific genes was greater than the shared genes in these endothelial cell subpopulations. Volcano plots and biological pathways in male and female aCaPs/reactive aCaps and gCaPs are shown in Fig. 3, C and D, respectively. Klf 4 (Kruppel-like factor 4) was downregulated in male aCaPs/reactive aCaps, Klf 4 (89) modulates angiogenesis through the regulation of the Notch pathway. Klf4 limits the activation of Notch 1. Ddah1 (dimethylarginine dimethylaminohydrolase 1) was upregulated in the female acaP/reactive aCaPs. Loss of Ddah 1 impairs whereas overexpression (90, 91) enhances angiogenesis. Sex-specific biological pathways included cellular senescence and artery morphogenesis (downregulated) and nitric oxide biosynthesis (upregulated) in female, and mitochondrial electron transport, response to oxidative stress and cholesterol homeostasis (upregulated) in males (Fig. 3D).

Figure 3.

Sex-specific response to hyperoxia in endothelial cells. A: UMAP of sequenced lung endothelial cells identified seven distinct clusters split by sex and experimental condition (room air and hyperoxia). B: number of differentially expressed genes (upregulated on top and downregulated at the bottom) in response to hyperoxia that are either shared between male and female (purple), unique in male (blue), or unique in female (pink) aCaP, arterial, and gCaP lung endothelial cells. C: volcano plots showing up- and downregulated in the female and male aCaP/reactive aCaP endothelial cells upon exposure to hyperoxia. D: sex-specific enriched biological pathways (blue: male, pink: female) in aCaP/reactive aCaP lung endothelial cells. E: volcano plots showing up- and downregulated in the female and male gCaP endothelial cells upon exposure to hyperoxia. F: sex-specific enriched biological pathways (blue: male, pink: female) in gCaP lung endothelial cells. All of the results above are obtained from the equal number of pooled viable lung cells from three mice. aCap, aerocytes or alveolar capillaries; gCap, general capillaries.

Among gCaP lung capillary endothelial cells, CxCl12 was upregulated in males, whereas neuropilin-1 (Nrp1) and bone morphogenetic receptor 2 (Bmpr2) were upregulated in females (Fig. 3E). CxCl12 plays an important role in angiogenesis through interaction with its receptors CXCR4 or CXCR7 (92–94). Autocrine signaling in lung endothelial cells leads to the development of pulmonary hypertension (95, 96). Nrp1 mediates VEGF signaling in endothelial cells and plays a key role in angiogenesis. Nrp1 also attenuates senescence in endothelial cells (97). The role of Bmpr2 mutations in pulmonary hypertension is well established (98) and loss of Bmpr2 leads to endothelial dysfunction including disordered proliferation (99). Among sexually dimorphic pathways in gCaPs, TNF-signaling was downregulated, whereas cell migration was upregulated in females. Transmembrane tyrosine receptor kinase was upregulated in males. Apoptosis was modulated in opposite directions in males and females (Fig. 3F). Differentially expressed genes in male and female endothelial cell subpopulations are listed in Supplemental Table S2.

Lung Epithelial Gene Expression Changes in Response to Hyperoxia

Lung epithelial cells composed of AT1, AT2, Lyz1+ AT2, and ciliated epithelial cells (Fig. 4A) and their relative contribution to the lung epithelial cells in room air and hyperoxia is shown in Fig. 4B and C. The number of cells analyzed and the number of upregulated and downregulated genes in AT1, AT2, and LyZ1+ AT2 cells is shown in Fig. 4D. p21 (Cdkn1a) was upregulated in hyperoxia-exposed AT1 and AT2 cells. Peroxiredxin (Prdx) 6 was upregulated in AT1 cells, which is induced in the alveolar epithelium under oxidative stress and has an antioxidant role (100). Igfbp2 was downregulated in the AT1 cells, which is a marker for postnatal AT1 cells and increases following pneumonectomy during lung regeneration. Igfbp2 positive AT1 cells are terminally differentiated (101). Biological pathways related to oxidative phosphorylation, apoptosis, and neutrophil-mediated immunity were upregulated, whereas TNF-signaling via NF-kappa B, hypoxia, and the MAP kinase signaling pathway were downregulated (Fig. 4E).

Figure 4.

Gene expression changes in epithelial cells in response to hyperoxia. A: UMAP of sequenced lung epithelial cells identified four distinct clusters. B: UMAP of lung endothelial cells showing the four distinct clusters in room air and hyperoxia. C: changes in the relative contribution of different lung epithelial cell subpopulations in room air and hyperoxia. D: number of cells sequenced and the number of up- and downregulated genes in AT1, AT2, and AT2 Lyz 1+ lung epithelial cells upon exposure of the neonatal lung to hyperoxia. E: volcano plots showing the differentially expressed genes in AT1 cells in response to hyperoxia and enriched biological pathways. F: volcano plots showing the differentially expressed genes in AT2 and AT2 Lyz1+ cells in response to hyperoxia and enriched biological pathways. All of the results above are obtained from equal number of pooled viable lung cells from three mice.

Thioredoxin-interacting protein (Txnip) was induced in the AT2 and AT2 Lyz1+ cells upon exposure to hyperoxia. Overexpression of Txnip decreased HIF activity and VEGF expression in the murine lung, independent of its ability to bind thioredoxin (102). Apoe was downregulated in the hyperoxia-exposed AT2 and AT2 Lyz1+ cells, which is protective in lung disease models through its anti-inflammatory and antioxidant effects (103). Oxidative phosphorylation and adipogenesis were upregulated, whereas extracellular matrix organization and epithelial-to-mesenchymal transition were among the downregulated pathways in AT2 and AT2 Lyz1+ cells (Fig. 4F). Differentially expressed genes in all epithelial cells are listed in Supplemental Table S3.

Sex-Specific Response to Hyperoxia in Epithelial Cells

The lung epithelial cells in room air and hyperoxia in the male and female PND7 murine lung are shown in Fig. 5A. The number of up- and downregulated differentially expressed genes show remarkable sex-specific modulation of the cellular transcriptome with most genes showing a sex-specific expression with little overlap (Fig. 5B). ATPase Na⁺/K+ transporting subunit β 1 (Atp1b1) was upregulated in the female AT1 cells (Fig. 5C), which is crucial for alveolar fluid clearance (104), and expression is decreased in lung fibrosis (105). Egr1 (early growth response gene 1) and Fosb are downregulated in the male AT1 cells upon exposure to hyperoxia, both of which are immediate early genes and inhibit cell proliferation (106, 107). Surfactant protein D (Sftpd) was upregulated in the female AT2 cells (Fig. 5C), which is a hydrophilic collection molecule secreted by type 2 alveolar epithelial cells and plays an anti-inflammatory role by regulating lung macrophage activity (108). A similar female bias in Sftpd expression was reported in murine hearts (109). Secretory leukocyte proteinase inhibitor (Slpi) was increased in male AT2 cells, which is an antiprotease and acts on neutrophil elastase (110, 111). Cytochrome c oxidase subunit 6c (Cox 6c) was upregulated in male AT2 Lyz1+ cells (Fig. 5C), which plays a crucial role in oxidative phosphorylation and may play a role in the metabolic changes in these cells in response to hyperoxia (112). Sex-specific biological pathways enriched from genes that were exclusively differentially regulated in males and females in AT1 and AT2 cells are highlighted in Fig. 5D. Angiogenesis was downregulated in male AT1 cells, whereas apoptosis, inflammatory response, and interferon-γ response were downregulated in female AT1 cells. The aerobic electron transport chain was upregulated in male AT1 cells, whereas ECM-receptor interaction was upregulated in female AT1 cells. The aerobic electron transport chain was upregulated, whereas xenobiotic metabolism was downregulated in male AT2 cells, whereas coagulation, myogenesis, and cytokine-mediated signaling pathway were downregulated in female AT2 cells. TNF-α signaling via NF-κ B and p53 pathway were downregulated in female AT2 Lyz1+ cells (Fig. 5D). Differentially expressed genes in male and female lung epithelial cell subpopulations are listed in Supplemental Table S3.

Figure 5.

Sex-specific response to hyperoxia in epithelial cells. A: UMAP of sequenced lung epithelial cells identified four distinct clusters split by sex and experimental condition (room air and hyperoxia). B: number of differentially expressed genes (upregulated on top and downregulated at the bottom) in response to hyperoxia that are either shared between male and female (purple), unique in male (blue), or unique in female (pink) AT1, AT2, and AT2 Lyz1+ lung epithelial cells. C: volcano plots showing up- and downregulated in the lung AT1, AT2, and AT2 Lyz1+ lung epithelial cells in female and male lungs upon exposure to hyperoxia. D: sex-specific enriched biological pathways (blue: male, pink: female) in AT1 and AT2 lung epithelial cells. All of the results above are obtained from equal number of pooled viable lung cells from three mice.

Neonatal Hyperoxia Exposure Leads to Significant Changes in Cell State in Lung Immune Cells

Sub clustering of 13,103 immune cells identified 12 distinct populations (Fig. 6A). UMAP plots of the immune cell subpopulations in room air and hyperoxia are shown in Fig. 6B and the percentage contribution to the total immune cells under room air and hyperoxia is shown in Fig. 6C.

Figure 6.

Gene expression changes in immune cells in response to hyperoxia. A: UMAP of sequenced lung immune cells identified twelve distinct clusters. B: UMAP of lung endothelial cells showing the twelve distinct clusters in room air and hyperoxia. C: changes in the relative contribution of different lung immune cell sub-populations in room air and hyperoxia. D: gene expression similarities and differences between neutrophils 1, 2, and 3. E: trajectory analysis of lung neutrophils showing neutrophils 1 and 3 arise from neutrophil 2. Neutrophils 1 and 3 show gene signatures seen in PMN-MDSCs and activated PMN-MDSCs, respectively. F: number of cells sequenced and the number of up- and downregulated genes in lung immune cells upon exposure of the neonatal lung to hyperoxia. G: volcano plots showing the differentially expressed genes in interstitial macrophages and neutrophil 3 in the neonatal lung upon exposure to hyperoxia. All of the results above are obtained from equal number of pooled viable lung cells from three mice. PMN-MDSCs, PMN-myeloid-derived suppressor cells.

We identified a population of neutrophils enriched for markers associated with PMN-myeloid-derived suppressor cells (PMN-MDSCs) (neutrophils 1 and 3) (66). PMN-MDSCs possess immunosuppressive properties and suppress the function of T-, B-, and NK cells and are thought to be pathologically activated. They are considered to be protumorigenic due to their immune-suppressive functions and contribute to pathologic tissue remodeling (113). Neutrophils 3 in our study expand under hyperoxic conditions. This cell cluster is similar to the neutrophil cluster reported by Hurskainen et al. (64) after the neonatal murine lung was exposed to hyperoxia with higher expression of Basp1 and CCl4. Differences in gene expression between these three neutrophil populations are shown in Fig. 6D. Upon exposure to hyperoxia, neutrophils 3 enriched for Ccl4, Ccl3, CxCl4, Basp1, Ninj1, Hilpda, Gadd45b, Atf3, and Mif associated with activated PMN-myeloid-derived suppressor cells (66). Neutrophils 1 showed enrichment for genes associated with immature neutrophils and PMN-MDSCs (Ngp, Ltf, and Cybb) (66). Trajectory analysis using RNA velocity showed that neutrophils 1 and 3 arise from neutrophils 2 (Fig. 6E). Supplemental table showing enriched genes in these three neutrophil populations are included in Supplemental Table S4.

The number of cells sequenced and up- and downregulated genes in each immune cell population is shown in Fig. 6F. The interstitial macrophage and the type 3 neutrophils showed the greatest number of differentially expressed genes among the immune cells following exposure to hyperoxia. Interstitial macrophages showed high expression of Spp1 (Fig. 6G) or osteopontin (114). Interstitial macrophages are sources of Spp1, which functions as a profibrotic mediator in lung diseases (115, 116). Type 3 neutrophils enrich for the chemokines CCl4 and CxCl3. These are characteristics of activated PMN-MDSCs (66). Leukocyte-specific transcript-1 (Lst1) was one of the downregulated genes in type 3 neutrophils (Fig. 6G). Interestingly, this finding was also reported among lung neutrophils in an IL-β-mediated pneumonitis model (117).

The common biological pathways up- and downregulated in male and female lungs in interstitial macrophages (Supplemental Fig. S2A) and type 3 neutrophils (Supplemental Fig. S2B). Biological pathways enriched in interstitial macrophages included interferon-γ response, complement, oxidative phosphorylation, necroptosis, and cellular response to interleukin-1. Neutrophils 3 showed upregulation of TNF-α signaling via NF-κB, inflammatory response, IL-2/STAT 5 signaling, apoptosis, and neutrophil degranulation, whereas adipogenesis was downregulated. Differentially expressed genes in all immune cells are listed in Supplemental Table S5.

Sex-Specific Changes in the Lung Immune Cells in Response to Neonatal Hyperoxia

The lung immune cells show remarkable sex-specific differences in their transcriptional state upon exposure to hyperoxia. The UMAP plots of the lung immune cells in male and female lungs in room air and hyperoxia are shown in Fig. 7A. The number of unique and shared differentially expressed genes in the immune cell subpopulations are shown in Fig. 7B. Basophils, B cells, neutrophils 3, and classical monocytes are predominated by the female response, whereas interstitial macrophages show a predominant response in males. Volcano plots showing the differentially expressed genes (DEGs) in male and female alveolar macrophages, interstitial macrophages, and neutrophils 3 are shown in Fig. 7C. DEGs in female basophils and B cells are shown in Fig. 7D. The top two upregulated genes in basophils in the female lung (Fig. 7D) upon exposure to hyperoxia were Mcpt8 (mast cell protease 8) (118) and Prss34 (protease serine 34) (119) both of which are specific to basophils. Upregulated pathways in basophils included interferon-γ response and Myc targets, whereas TNF-α, IL-6, PI3-Akt signaling, and cellular response to TGF-β were among the downregulated pathways (Fig. 7E).

Figure 7.

Sex-specific response to hyperoxia in immune cells. A: UMAP of sequenced lung immune cells identified twelve distinct clusters split by sex and experimental condition (room air and hyperoxia). B: number of differentially expressed genes (upregulated on top and downregulated at the bottom) in response to hyperoxia that are either shared between male and female (purple), unique in male (blue), or unique in female (pink) lung immune cells. C: volcano plots showing up- and downregulated genes in female and male alveolar macrophages, neutrophil 3, and interstitial macrophages upon exposure to hyperoxia. D: volcano plots showing up- and downregulated genes in female basophils and B cells. Sex-specific enriched biological pathways (blue: male, pink: female) in basophils (E), interstitial macrophages (F), and neutrophils 3 (G). All of the results above are obtained from equal number of pooled viable lung cells from three mice.

CCl4 or macrophage inflammatory protein-1 β was upregulated in the male IMs, which plays a role in leukocyte recruitment following injury (Fig. 7C) (120). Biological pathways upregulated in male lung interstitial macrophages upon exposure to hyperoxia included type-1 interferon signaling, inflammatory response, antigen processing and presentation, IL-2/STAT 5 signaling, and TNF-α signaling via NFκB. On the contrary inflammatory response was downregulated in female interstitial macrophages (Fig. 7F). Gadd45g was upregulated in male neutrophils 3, which plays a role DNA repair (Fig. 7C) (121, 122). Female-specific pathways downregulated in neutrophils 3 included oxidative phosphorylation, DNA repair, Myc targets, whereas neutrophil chemotaxis and mTORC1 signaling were upregulated. Interferon-γ response, IL-6 signaling, cellular response to type 1 interferon, and TNF were upregulated in male neutrophils 3 (Fig. 7G). S100a4 was upregulated in male classical monocytes, which leads to increased production of inflammatory mediators from monocytes (123), whereas CxCl2 was downregulated in females in classical monocytes, which recruits neutrophils and leads to neutrophil accumulation (124) (Supplemental Fig. S2C). Classical monocytes in male lungs showed upregulation of neutrophil-mediated immunity, PI3-Akt signaling, and complement, whereas in female lungs MAPK signaling pathway, inflammatory response, TNF-α signaling via NFκB, and regulation of apoptosis were downregulated (Supplemental Fig. S2D). Differentially expressed genes in male and female lung immune cell subpopulations are listed in Supplemental Table S5.

Cell-Cell Communication

To infer, visualize, and analyze intercellular communications from the scRNA-seq data, we used the Cell Chat tool (33). We interrogated whether the cell-cell interaction between the cell types was significantly changed and if the major sources and targets were changed between room air and hyperoxic conditions. We analyzed the cell-cell communication in the endothelial, epithelial, and immune cell subpopulations and also studied the interactions between epithelial-endothelial and endothelial-immune subpopulations.

Novel ligand-receptor pairs and the role of the distal lung endothelium in neonatal hyperoxia.

The cell-cell communication between the different endothelial cell subpopulations is enhanced, but the interaction strength is decreased compared to normoxia (Supplemental Fig. S3A). Under room air conditions, among all endothelial cell subpopulations, the aCaPs have significant outgoing and incoming signaling to and from gCaPs and proliferating gCaPs (as demonstrated by the edge width of the ligand-receptor pairs) and serve as the dominant communication hub (Fig. 8A). Strikingly, upon exposure to hyperoxia, the differential number of incoming and outgoing signaling networks are increased from the reactive aCaP population whereas the signaling networks among other endothelial cell subpopulations are decreased compared to normoxia (Fig. 8, A and B). There was a strong component of autocrine signaling in the reactive aCaPs (Fig. 8B). Within the endothelial cell subpopulation, we identified the global communication patterns that connect cell groups with signaling pathways in the context of outgoing or incoming signaling under room air and hyperoxia conditions (Fig. 8, C and D). We uncovered three outgoing and incoming signaling patterns under hyperoxia. There are notable differences and similarities between the communication patterns in room air and hyperoxia. The communication patterns in room air are shown in Fig. 8C, where the outgoing signaling shows four patterns with the aCaPs, gCaPs, arterial/venous, and lymphatic endothelial showing distinct outgoing signaling pathways. Under hyperoxia, a large portion of the outgoing signaling from the distal lung endothelium (with gCaPs, aCaPs, and reactive aCaPs) is characterized by pattern #1, which represents multiple pathways, including but not limited to Kit, Apelin, Semaphorin, and collagen pathways. The lymphatic and arterial/venous endothelial cells have distinct outgoing signaling pathways. The incoming signaling patterns in hyperoxia grouped the gCaP/proliferating gCaP (pattern #3), arterial/venous/lymphatic (pattern #1), and the aCaPs (pattern #2) in three groups (Fig. 8D). Collagen, Kit, Semaphorin (Sema) 7, and protein tyrosine phosphatase receptor type M (Ptprm) were the autocrine-acting pathways in the gCaP/proliferating gCaP population. Since the reactive aCaPs population was significantly increased in the hyperoxia-exposed PND7 lung, we focused on the incoming and outgoing signaling pathways in this cell subpopulation. Tweak (Tnf superfamily member 12) and Ptprm pathways were identified in the incoming and outgoing signaling pathways in this population of cells, respectively (Fig. 8, E and F). Tweak belongs to the TNF ligand family and is a ligand for the Fn14/TweakR receptor. It plays a role in angiogenesis by modulating endothelial cell proliferation and migration and has been shown to promote oxidative stress in endothelial cells (125–128). Ptprm (protein tyrosine phosphatase receptor type M) is highly expressed in the lung and expressed in the lung endothelium. It binds directly to VE-cadherin, modulates its phosphorylation, and maintains barrier integrity (129, 130). The Ptprm pathway was not among the inferred networks under room air conditions in lung endothelial cells. Network centrality analysis of the inferred Ptprm signaling network identified that reactive aCaPs are the most prominent sources and dominant mediators for this ligand and has a predominant autocrine and paracrine effect on gCaPs and proliferating gCaPs, with no discerned effects on other endothelial cells. The TGF-β network between room air and hyperoxia, displayed some striking differences. The aCaPs are the predominant targets under room air and hyperoxia (Fig. 8, G and H). TGF-β 1 was the main ligand with expression from all endothelial cells under room air and hyperoxia. However, TGF- β 2 was induced in reactive aCaPs and lymphatic endothelial cells upon exposure to hyperoxia (Fig. 8I). These two isoforms have different tissue-specific expression (131, 132) and may have distinct biological functions in disease pathophysiology (133).

Figure 8.

Intercellular communication changes in response to hyperoxia in lung endothelial cells. A: circle plot showing the number of interactions between different endothelial cell subpopulations in room air and hyperoxia. Edge colors are consistent with the sources as sender, and edge weights are proportional to the interaction strength. Thicker edge line indicates a stronger signal. Circle sizes are proportional to the number of cells in each cell group. B: circle plot showing the number of differential interactions between different endothelial cell sub-populations in room air (blue) and hyperoxia (red). The inferred outgoing patterns of secreting cells and incoming communication patterns of target cells among the lung endothelium, in room air (C) and hyperoxia (D). The thickness of the flow indicates the contribution of the cell group or signaling pathway to each latent pattern. Heatmap showing the relative importance of each cell group based on the network centrality measures of Tweak signaling network in hyperoxia (E), Ptprm signaling network in hyperoxia (F), TGF-beta signaling in room air (G), and hyperoxia (H). Violin plots showing expression of Tgfb1 and Tgfb2 (I) in lung endothelial cells in room air and hyperoxia. All of the results above are obtained from equal number of pooled viable lung cells from three mice.

Neonatal hyperoxia alters cell-to-cell communication among lung epithelial cells.

The number of cell-cell communication networks is decreased but their strength increased in hyperoxia compared to normoxia among lung epithelial cells (Supplemental Fig. S3B). Under both room air and hyperoxic conditions, AT1 cells enrich in outgoing signaling pathways whereas AT2 cells enrich in incoming signaling pathways (Supplemental Fig. S3C). However, the number of cell-cell communication from the AT1 cells was decreased in hyperoxia compared with room air conditions (Fig. 9, A and B). The communication patterns between alveolar epithelial cells in room air and hyperoxic conditions are highlighted in Fig. 9, C and D. Under hyperoxia, AT1 cells are the senders for most of the outgoing signaling networks (pattern #1).

Figure 9.

Intercellular communication changes in response to hyperoxia in lung epithelial cells. A: circle plot showing the number of interactions between different lung epithelial cell subpopulations in room air and hyperoxia. Edge colors are consistent with the sources as sender, and edge weights are proportional to the interaction strength. Thicker edge line indicates a stronger signal. Circle sizes are proportional to the number of cells in each cell group. B: circle plot showing the number of differential interactions between different endothelial cell subpopulations in room air (blue) and hyperoxia (red). The inferred outgoing patterns of secreting cells and incoming communication patterns of target cells among the lung epithelial cells, in room air (C) and hyperoxia (D). The thickness of the flow indicates the contribution of the cell group or signaling pathway to each latent pattern. Heatmap showing the relative importance of each cell group based on the network centrality measures of Bmp signaling network in room air (E) and relative contribution of each ligand-receptor pair in room air to the overall communication network of Bmp signaling pathway (F). Heatmap showing the relative importance of each cell group based on the computed four network centrality measures of Bmp signaling network in hyperoxia (G) and relative contribution of each ligand-receptor pair in hyperoxia to the overall communication network of Bmp signaling pathway (H). I: violin plots showing expression of Bmpr1a and Bmpr2 in lung epithelial cells in room air and hyperoxia. All of the results above are obtained from equal number of pooled viable lung cells from three mice.

The bone morphogenetic protein (BMP) signaling network, under room air conditions, originates from the AT1 cells (sender) with autocrine and paracrine effects on the AT1 and AT2 cells (receiver). Under hyperoxia, the AT1 cells are the main receivers with AT1 and AT2 cells becoming the ligand sources (Fig. 9, E and F). Bmp4 was the main ligand, with Bmpr2 and Bmpr1a being the inferred receptors (Fig. 9, G and H). Bmpr1a and Bmpr2 expression was increased in AT1 cells after exposure to hyperoxia (Fig. 9I). The role of BMP signaling in the alveolar niche has been highlighted in a pneumonectomy model with increased BMP signaling, enhancing the differentiation of AT2 cells to AT1 cells (134). In primary cultures of mouse AT-2 cells, however, recombinant BMP4 inhibited trans differentiation of AT2 to AT1 cells (135).

Cell-cell communication patterns among lung immune cells are altered by neonatal hyperoxia.

The number of cell-cell communication networks among the immune cells is increased in hyperoxia compared with normoxia, whereas the interaction strength decreased (Supplemental Fig. S4A). Classical monocytes and interstitial macrophages enrich in pathways related to outgoing signaling both under room air and hyperoxia whereas dendritic cells enrich in incoming signaling networks. Basophils emerge as a cell subpopulation with an increase in outgoing interaction strength upon exposure to hyperoxia (Supplemental Fig. S4B). Interstitial macrophage and dendritic cells also show strong autocrine signaling in the hyperoxia-exposed lung (Fig. 10A). The communication patterns between immune cells in the lung in room air and hyperoxia are shown in Fig. 10, B and C. Interstitial macrophages have a distinct set of outgoing and incoming signaling patterns both in room air and hyperoxia. Interestingly, enriched among the outgoing signaling patterns in basophils in hyperoxia is GM colony-stimulating factor (Csf), with macrophages and dendritic cells as receivers (Fig. 10D). Csf1 expression is increased in basophils, whereas Csf1r expression is increased in lung dendritic cells upon exposure to hyperoxia (Fig. 10D). Basophils are imparted a unique transcriptional signature in the lung microenvironment partly by GMCSF and polarize the alveolar macrophages towards an anti-inflammatory state (136). The fibronectin signaling pathway was activated in hyperoxia with classical monocytes and interstitial macrophages as sources and CD44 as the receptor (Fig. 10E). Increased expression of fibronectin in lung macrophages has been shown in lung diseases and upon exposure to hyperoxia (137, 138).

Figure 10.

Intercellular communication changes in response to hyperoxia in lung immune cells. A: circle plot showing the number of differential number of interactions between different lung immune cells in room air (blue) and hyperoxia (red). The inferred outgoing patterns of secreting cells and incoming communication patterns of target cells among the lung immune cells, in room air (B) and hyperoxia (C). The thickness of the flow indicates the contribution of the cell group or signaling pathway to each latent pattern. Heatmap showing the relative importance of each cell group based on the network centrality measures of Csf1 signaling network in hyperoxia, and relative contribution of each ligand-receptor pair in room air to the overall communication network of Csf1 signaling pathway (D), and violin plots showing expression of Csf1 and Csf1r in lung immune cells in room air and hyperoxia (E). Heatmap showing the relative importance of each cell group based on the computed four network centrality measures of Fn1 signaling network in hyperoxia, and relative contribution of each ligand-receptor pair in room air to the overall communication network of Fn1 signaling pathway (F), and violin plots showing expression of Fn1 in lung immune cells in room air and hyperoxia (G). All the results above are obtained from equal number of pooled viable lung cells from three mice.

Neonatal hyperoxia decreases and alters ligand-receptor interactions between endothelial and epithelial cells.

The number of inferred interactions and the interaction strength decreases upon exposure to hyperoxia (Supplemental Fig. S4C). aCaPs, reactive aCaPs, and AT1 cells showed the highest outgoing interaction strength, whereas the AT2 cells showed the highest incoming interaction under normoxia and hyperoxia (Supplemental Fig. S4D). Reactive aCaP cells display the most differential number and strength of both outgoing and incoming signaling networks from AT1, AT2, and AT2Lyz1+ cells (Fig. 11, A and B). The communication patterns between epithelial and endothelial cells in room air and hyperoxia are shown in Fig. 11, C and D. Under room air, most of the outgoing signaling was dominated by pattern #2 (from AT1 epithelial cells) and pattern #3 was from ciliated epithelial cells. Pattern #2, in room air from AT1 epithelial cells, included VEGF with aCaP cells as receivers. Under hyperoxia, four predominant outgoing patterns were observed with pattern #1 (reactive aCaPs/aCaPs) and pattern #2 (AT1 cells) being the major secreting cells. The activin signaling network [ligand-receptor pair: Inhba-(Acvr1b + Acvr2a)] signals from the reactive aCaPs to AT2 and AT2 Lyz1+ cells (Fig. 11E). Endothelial activin-A signaling accelerated the development of pulmonary hypertension (139). Activin signaling also exacerbated pathology in murine models of idiopathic pulmonary fibrosis (140–142), and overexpression led to inflammation and alveolar cell death (143). Tweak (Tnf superfamily member 12) is another inferred endothelial-to-epithelial communication with the ligand from the gCaPs, aCaP, and arterial endothelial cells and signaling to the AT1 and AT2 cells (Fig. 11F). TNFRSF12A^hi cells clustered within terminal bronchioles and exhibited enriched clonogenic distal lung organoid growth activity (144). Blockade of the Tweak receptor decreased lung fibrosis and mortality in a sepsis-induced acute lung injury model (126). The density of epithelial-to-endothelial interactions in the Notch signaling pathway is decreased in hyperoxia (Fig. 11, G and H). One of the Notch ligands identified in the epithelial cells was Dlk1 (δ-like noncanonical Notch ligand 1). Dlk1 expression was high in AT2 cells in room air with a significant decrease in hyperoxia (Fig. 11I). Dlk1 regulated lung epithelial cell proliferation and differentiation during fetal lung development (145). Dlk1 signaling was also necessary for timed inhibition of Notch signaling to allow AT2-AT1 differentiation (146). Interestingly, an inhibitory role in angiogenesis has been ascribed to Dlk1 (147).

Figure 11.

Neonatal hyperoxia decreases and alters ligand-receptor interactions between endothelial and epithelial cells. Circle plot showing the number (A) and strength (B) of differential interactions between epithelial and endothelial cell subpopulations in room air (blue) and hyperoxia (red). The inferred outgoing patterns of secreting cells and incoming communication patterns of target cells between the epithelial and endothelial cells, in room air (C) and hyperoxia (D). The thickness of the flow indicates the contribution of the cell group or signaling pathway to each latent pattern. E: chord diagram showing senders and targets for the activin signaling network under hyperoxia. The activin signaling network signals mainly from the reactive aCaPs to AT2 and AT2 Lyz1+ cells. F: chord diagram showing senders and targets for the Tweak signaling network under hyperoxia. This was another inferred endothelial-to-epithelial communication with the ligand from the gCaPs, aCaP, and arterial endothelial cells and signaling to the AT1 and AT2 cells. Chord diagram showing senders and targets for the Notch signaling network under room air (G) and hyperoxia (H). The epithelial-to-endothelial interaction in the notch signaling pathway is decreased in hyperoxia. I: violin plots showing expression of Dlk1 (notch ligand) in lung epithelial and endothelial cells in room air and hyperoxia. All of the results above are obtained from equal number of pooled viable lung cells from three mice. aCap, aerocytes or alveolar capillaries; gCap, general capillaries.

Neonatal hyperoxia increases and alters ligand-receptor interactions between endothelial and immune cells.

The number of inferred interactions is increased between the endothelial and immune cells with a decrease in the strength of interactions in hyperoxia compared to room air (Supplemental Fig. S5A). aCaPs and reactive aCaPs showed the highest outgoing interaction strength, whereas dendritic cells, interstitial macrophages, and classical monocytes showed the highest incoming interaction under normoxia and hyperoxia (Supplemental Fig. S5B). Under exposure to hyperoxia, the predominant endothelial cell population serving as the signaling hub to immune cells were the reactive aCaPs (Fig. 12A). The communication patterns between immune and endothelial cells in room air and hyperoxia are shown in Supplemental Fig. S5, C and D. Under room air, four outgoing patterns are observed with all the immune cells except NK cells, B cells, neutrophils 2, and T cells forming the predominant secreting cell group (pattern #2). Under hyperoxia, the outgoing signaling patterns condensed to five main groups with all the immune cells forming the predominant secreting cells group (pattern #2). Among the endothelial cells, the aCaP/reactive aCaP (pattern #1) and the gCaP and arterial endothelium (pattern #5) formed distinct groups of secreting cells. There was no major VEGF signaling between endothelial and immune cells in room air (Fig. 12B). Under hyperoxia, the neutrophil 3 population signals to the distal capillary endothelium with VEGF as the proposed ligand (Fig. 12, C and D). Neutrophils as a source of VEGF and mediating angiogenesis in pathological conditions have been reported previously (148–150). Another proposed ligand from basophils, interstitial macrophages, and neutrophils 3 signaling to the distal lung endothelium upon exposure to hyperoxia was TNF-α (Fig. 12, E and F). TNF- α induces expression of endothelial cell adhesion molecules such as ICAM1 (151, 152). It also modulates blood vessel remodeling in the setting of inflammation, playing a role in pathological angiogenesis (153, 154). Tnf expression has been reported in lung basophils (136) and protective in injury models as well (155).

Figure 12.

Neonatal hyperoxia increases and alters ligand-receptor interactions between endothelial and immune cells. A: circle plot showing the differential number of interactions between different lung endothelial and immune cell sub-populations in room air (blue) and hyperoxia (red). B and C: hierarchical plot showing the inferred intercellular communication network for Vegf signaling between the endothelial and immune cells in room air (B) and hyperoxia (C). D: violin plots showing expression of Vegf in lung endothelial and immune cells in room air and hyperoxia. E: hierarchical plot showing the inferred intercellular communication network for Tnf signaling between the endothelial and immune cells in hyperoxia. This plot consists of two parts: Left portions highlights signaling from immune cells to endothelial cells. The right portion shows signaling within immune cells. Solid and open circles represent source and target, respectively. Circle sizes are proportional to the number of cells in each cell group and edge width represents the communication probability. Edge colors are consistent with the signaling source. F: violin plots showing expression of Tnf in lung endothelial and immune cells in room air and hyperoxia. All of the results above were obtained from equal number of pooled viable lung cells from three mice.

DISCUSSION

We present significant sex-specific differences in all lung cell subpopulations by scRNA-seq of the PND7 murine lung at the early alveolar stage of lung development following hyperoxia exposure during the saccular stage of lung development. Significantly, we show that distal lung vascular endothelium composed of the aCaP (aerocytes) and the gCaP (general capillary) endothelial cells, is exquisitely sensitive to hyperoxia exposure with the emergence of an intermediate capillary endothelial population with both aCaP and gCaP endothelial cells and the identification of a myeloid-derived suppressor cell population from the lung neutrophils. Sex-specific differences were evident in all lung subpopulations but were striking among the lung immune cells. Finally, we identified the specific intercellular communication networks and the ligand-receptor pairs that are impacted by neonatal hyperoxia exposure.

We and others have previously shown that the female lung is more resilient to neonatal hyperoxia exposure with better preservation of alveolarization and vascular development compared with similarly exposed neonatal male lungs (18–20). Sex-specific differences by whole lung bulk RNA-seq revealed some differences at PND7, which were increased during recovery at PND21. Xia et al. (156) have reported sex-specific differences in the lung cell subpopulations using different hyperoxia exposure models using 85% $\mathrm{\text{F}}{\mathrm{\text{I}}}_{{\mathrm{\text{O}}}_2}$ for 14 days after birth.

The reactive aCaPs seem to be a transient endothelial cell population arising from the gCaP cells and leading to the aCaP cells with markers of both aCaP and gCaP cells. Whether this endothelial cell subpopulation is seen in later time points during recovery and repair and in the human disease in preterm neonates needs to be elucidated. Inhba was upregulated in the reactive aCaP capillary endothelial cells upon exposure to hyperoxia. Inhibins belong to the TGF-β family and comprise a functional heterodimer with an α- and a β-subunit. Inhibin promotes angiogenesis through TGF-β receptors (157) and increases vascular permeability through the internalization of VE-cadherin (158). Under hypoxic conditions, Inhibin expression is modulated HIF-1a and HIF-2a (159). However, an abundance of Inhba impairs endothelial cell function by reducing BMPPR2 levels in endothelial cells (139).

We identified three distinct neutrophil subpopulations in our dataset and highlight the existence of the polymorphonuclear neutrophil myeloid-derived suppressor cells (PMN-MDSC; neutrophils 3) in the neonatal lung and the emergence of activated PMN-MDSCs upon exposure to hyperoxia, highlighting the heterogeneity in lung neutrophils at baseline, after exposure to hyperoxia, and differences between the male and female lung. Neutrophils 3 were enriched for genes related to cell activation and inflammation. MDSCs play a crucial role in immunosuppression in cancer, by the production of reactive oxygen species and immunosuppressive cytokines and adversely impacting T-cell proliferation (160). Several mediators including M-CSF, which was upregulated in the hyperoxia-exposed lung at PND7 can promote the development of MDSCs. Velocity analysis showed that neutrophils 3 arise from neutrophils 2. Neutrophils 3 showed the greatest number of differentially expressed genes in response to hyperoxia and showed increased VEGF expression upon exposure to hyperoxia and signal to the lung endothelium.

Significantly, we highlight the sex-specific differences in gene expression in different lung cell subpopulations in the epithelial, endothelial, and immune cells. Differences in the secretome and transcriptome of human umbilical venous endothelial cells have been described (161) and so is the response to hyperoxia in human pulmonary microvascular endothelial cells (162). The immune cells stand out in the degree of sex-specific differences in the hyperoxia-exposed neonatal lung. The sex-based differences in both innate and adaptive immune responses in many diseases and organs are well known (163, 164). Interestingly, many immune-related genes and regulatory elements that play a role in both the innate and adaptive immune responses are located on the X-chromosome (165). One of the genes present on the X-chromosome that was upregulated in the female interstitial macrophages was thiol-specific peroxidase peroxiredoxin-4 (Prdx4). Mice lacking Prdx4 in myeloid cells have increased mortality in LPS-induced sepsis. Prdx4 restricts inflammasome-mediated signaling that thus might temper inflammation (166).

Apart from the highlighted differentially expressed genes that are distinct between the male and the female lung, several sex-specific biological pathways, which were common between different lung cell subpopulations, were discovered. Of note, apoptosis, TNF-α signaling pathway, p53 pathway, inflammatory response, interferon-γ response, and cytokine-mediated signaling was downregulated in females. Electron transport chain, inflammatory response, apoptosis, and TNF-α signaling were upregulated in males. Differences between male and female cells in changes in mitochondrial respiration upon exposure to various stressors and the sex-dependent impacts have been reported in many previous studies (167–170). Male-specific upregulation of TNF-α was described in response to other injuries and other organs (171–174) in newborns and adults. The differences between the male and female-specific biological pathways correspond with the phenotype of greater lung injury with decreased alveolarization and vascular development in the hyperoxia-exposed male lung compared to the female (19).

We highlighted cell-cell communications between the major lung cellular subpopulations and have highlighted novel autocrine and paracrine-acting ligands and ligand-receptor pairs. This resource will be made available to the scientific community at https://www.lingappanlab.com/resources. The aCaPs and reactive aCaPs function as the main senders among the lung endothelial cells, whereas the AT1 cells serve that role among epithelial cells. Interstitial macrophages and classical monocytes serve as the signaling hubs among the lung immune cells. The number of intercellular communications as well as the strength of communication is generally decreased upon exposure to hyperoxia except among endothelial cells and immune cells, where the number of interactions is increased. We highlight several known and well-established signaling networks such as Notch, TGF-β, Bmpr2, and Tnf-α signaling networks but at the same time were also able to reveal novel ligand and ligand-receptor interactions such as the Tweak and Ptprm signaling networks.

We recognize the limitations of the present study with the use of pooled lung cells from three biological replicates, which precludes us from conducting an interaction analysis between biological sex and hyperoxia exposure. The lung dissociation methods may have biased our yield with increased isolation of certain lung cell subpopulations while decreasing the yield of others. However, the strengths of this study include the detailed study of sex-specific differences in the major lung cell subpopulations at single-cell resolution and the change in cell-cell communication patterns in the neonatal lung upon exposure to hyperoxia.

Future studies would investigate the sex-specific transcriptional state in the lung at prenatal and at P1 (at baseline) before the onset of injury, which may predispose the male and female lungs to different patterns of lung injury postnatally. We had previously identified that the sex-specific differences in gene expression increase at late alveolar stage of lung development (PND21) following hyperoxia exposure during the saccular stage of lung development. Cell states during recovery and repair after early injury at single-cell resolution need to be elucidated. Sex-specific differences may be mediated through sex hormones or through genes on the X-chromosome having differential expression in the male and female lungs. Exploring the basis behind sex-specific differences will be crucial to explain the female sex resilience in human BPD and will suggest new therapeutic modalities and guide the right therapy to the right patient.

DATA AVAILABILITY

The raw data have been uploaded to the National Center for Biotechnology Information Gene Expression Omnibus database (NCBI GEO), accession number GSE211356 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE211356).

GRANTS

This work was supported in part by National Institutes of Health Grants R01-HL144775, R01-HL146395, and R21-HD100862 (to K.L.).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

K.L. conceived and designed research; X.D., C.L., and K.L. performed experiments; A.C., M.C., C.L., E.S., and K.L. analyzed data; A.C., M.C., E.S., and K.L. interpreted results of experiments; A.C., M.C., C.L., and K.L. prepared figures; A.C., M.C., and K.L. drafted manuscript; A.C., M.C., E.S., and K.L. edited and revised manuscript; A.C., X.D., C.L., E.S., and K.L. approved final version of manuscript.