Single-Cell Multiome Impact of Prenatal Heavy Metal Exposure on Early Airway Development

Abstract

Prenatal exposure to cadmium (Cd) and arsenic (As) can severely impair fetal lung development, leading to lifelong adverse effects. As two of the most common and toxic heavy metals, Cd and As pose risks to many communities through food and water consumption. We have shown that prenatal co-exposure to Cd and As at levels relevant to human intake inhibits branching morphogenesis, yet cell-type-specific mechanisms remain elusive. Here we examined early embryonic lungs (E12) from mice exposed prenatally to either 0 (control) or 250 (treated) ppb of both Cd and As. Through single-cell multiome sequencing (scATAC-seq+scRNA-seq) and high-resolution metabolomics, we present a multifaceted landscape of Cd and As-induced molecular and cellular disruption. We identified 19 cell states exhibiting state-specific changes in gene expression related to cell proliferation and differentiation. Velocity analysis integrating RNA splicing and chromatin kinetics showed profound disruptions in cell fate, particularly affecting differentiation of Sox2⁺ proximal progenitors and Wnt2+ mesenchymal progenitors. Gene regulatory network analysis pinpointed the diminished function of Gata6 and Gli2 as central to these disruptions, which was further confirmed by their reduced protein expression in exposed E12, E14.5 and E17 lungs. Additionally, metabolomic alterations in polyamine, tyrosine and fatty acid biosynthesis correlated with changes in gene expression of catalytic enzymes. These findings demonstrate that Cd and As at levels relevant to human exposure impair early airway formation across multiple regulatory levels, including chromatin accessibility, transcription and cell metabolism, and provide insights into the factors central to cell resilience during this vulnerable stage of lung development.

Introduction

Lung development follows a complex, highly orchestrated progression through five stages: embryonic, pseudoglandular, canalicular, saccular and alveolar, each marked by distinct cellular events. The lung arises from the foregut endoderm in mice at Embryonic Day 9.5 (E9.5) and undergoes branching morphogenesis with repeated branching and outgrowth (pseudoglandular, E10.5-E16.5), laying the foundation for the preacinar airway, alveolar and vascular structures. It also marks the emergence of cellular diversity, as multipotent progenitor cells differentiate into various lineages of epithelial, mesenchymal, endothelial and other specialized types. Genetic alteration or ablation in these progenitor cells lead to profound developmental defects [1], and the pseudoglandular phase is highly vulnerable to environmental stressors before pregnancy awareness. Early-life exposure to nicotine, ozone, and various pollutants can have lasting consequences on lung development and function throughout life [2, 3]. Despite this, the precise effects of such exposures on early lung cell fates and dynamics remain poorly understood, hampering development of mitigation strategies targeting the affected mechanisms.

Previously we have reported the effects of cadmium (Cd) and arsenic (As) on inhibiting branching morphogenesis and disrupting pyrimidine biosynthesis at the pseudoglandular stage [4]. Cd and As, being the most prevalent and toxic environmental heavy metals, are ranked 1^st and 7^th on the priority list of toxicants by the U.S. Agency for Toxic Substances and Disease Registry [5]. Chronic exposures for non-smokers, often geospatially clustered, mainly come from diet and drinking water [6]. Alarming levels of Cd and As in drinking water and food have been emphasized as a global concern [7, 8], especially in certain regional and ethnic communities [9-11]. Cd and As can cross the placenta, accumulate in fetal tissue, and have been detected in umbilical cord blood and meconium, correlating with maternal exposure [12, 13]. The 2003-2012 National Health and Nutrition Examination Survey (NHANES) showed that 98.7% children had detectable As and 73.9% had Cd [14], with the highest As level found in the youngest age group studied (3-5 yo) [15]. Although linked to compromised lung health in children [16-20], only a few studies have investigated the molecular impact on lung development [21, 22] and very few have addressed any mechanistic effects of metal mixtures [23].

Early lung development is characterized by cellular complexity, diverse spatial arrangements and intricate temporal sequences of events. Single cell (sc-) sequencing technology has revolutionized the ability to study those dynamics. Recent study shows that hyperoxia exposure alters all cells, particularly affecting alveolar epithelium, stromal fibroblasts, capillary endothelium and macrophage populations [24]. The scRNA-seq analysis of mouse embryonic lungs spanning E12 to postnatal day 14 [25] reveals the rapid diversification of cell lineages in early lung development. Integration of single cell Transposase-Accessible Chromatin with high-throughput sequencing (scATAC-seq) further uncovers chromatin-level regulations of gene expression and cell fate decisions. The human lung multiomic cell atlas [26] combining scRNA-seq, scATAC-seq, spatial transcriptomics and single-cell imaging has identified 144 cell states/types samples across 5-22 pcw (post conception weeks), including new progenitor cell states and transition populations. Single cell multiome sequencing, which allows simultaneous data acquisition from the same cell, showed perturbation of a distinct FOXF1 gene regulatory network in endothelial and fibroblast progenitors in a developmental disorder of human lung morphogenesis [27].

The present work extends previous studies by employing advanced computational tools to examine the early pseudoglandular stage of lung development and its response to environmental exposure —an area that remains underexplored. Our single-cell multiome analyses showed cell-specific susceptibilities characterized by alterations in transcriptional and epigenetic landscapes with changes driving metabolic effects in specific cell populations [28]. By integrating differential gene expression analysis (DEG), gene regulatory network (GRN) mapping, cell fate trajectory modeling, and metabolic profiling, we characterized the multi-layered mechanisms by which heavy metals disrupt the normal path of lung development.

Methods

Animal treatment.

Animal protocols were approved by the Institutional Animal Care and Use Committee at Emory University and Atlanta VA medical center. Female C57BL/6J mice (Jackson Laboratory, Bar Harbor, ME) were provided ad libitum with AIN-93G diet, a purified diet minimizing uncontrollable dietary heavy metal intake [29]. The mice were exposed to either 0 (control) or 250 (treated) ppb of both Cd and As in drinking water for 10 weeks prior to gestation and until euthanization. Pregnancy was timed by the presence of a cervical plug on day 0. Gestational age was further confirmed by fetal landmarks.

Single cell multiome data processing and clustering

At least five embryonic lungs were pooled to undergo single nuclei isolation and multiome sequencing (see details in Supplemental Material). Raw sequencing data is available at open-access data repository [30]. The weighted nearest neighbor (WNN) graph was then constructed based on preprocessed scRNA-seq and scATAC-seq and the Leiden algorithm [31] was used for community detection resulting in the final 19 clusters. Pooling biological replicates is a common strategy in single-cell sequencing [25, 26, 32] to reduce cost, ensure unbiased sampling, and capture biological heterogeneity. In our study, statistical testing was conducted using individual cells (>7000 per condition) as units of observation.

Annotation

We used two approaches for cell type annotation. First, we applied the CellRef algorithm [33] referencing barcodes to specific cell types from a previously published E12.5 mouse lung dataset [34]. Second, we manually annotated the Leiden cell clusters based on established cell identity markers in a recent study [25] (Supplemental Table S1).

Single-cell computational analysis

Differential gene expression was assessed using the Wilcoxon rank-sum test to identify cluster-defining genes and treatment-specific changes (False Discovery Rate [FDR] adjusted P value < 0.05). Fisher’s exact test [35] was used to examine the statistical significance of normalized cell proportions associated with each cluster or gene marker. We can then test if the proportion of treated cells is higher in cell type A compared to other cell types.All genes were ranked by log2 fold change and analyzed via gene set enrichment analysis (GSEA) pre-ranked test [36]. MultiVelo [37] was used to predict cell fate. Single-cell trajectory of mesenchymal cells (cell cluster 4 to 14) was inferred using Moncle3, with cluster 8 being the “root” of the trajectory. Enhancer-driven Gene Regulatory Networks were inferred using SCENIC+ [38]. In silico perturbation was performed to test the impact of GATA6 on cell fate. Detailed steps are described in online Supplemental Material.

Immunohistochemistry

Paraffin embedded sections of embryonic lungs collected at E12, E14.5, E16, E17 (n=3-4 per group) were processed for immunohistochemistry for SOX2, SOX9, GATA6 and GLI2 and imaged with a Zeiss Axio Imager 2 fluorescence microscope system.

High-resolution untargeted metabolomics

E12 Mouse embryonic lungs (n=32 in total, pooling four into one sample for sufficient tissue amount) were processed for metabolomics as previously described [39-42]. Dual liquid chromatography coupled with Orbitrap Q Exactive HF mass spectrometer (Thermo Fisher) operating under negative and positive electrospray ionization (ESI) generated two sets of data: C18/-ESI and HILIC/+ESI. Features was selected by student’s t test (raw P < 0.05) and tested for enriched pathways using Rodin [43]. Metabolite identities were confirmed by co-elution relative to authentic standards, with additional annotations made conservatively in search against Human Metabolome Database (HMDB) [44].

Results

Cell populations affected by Cd and As in early pseudoglandular lungs

We employed a comprehensive approach to characterize the single-cell transcription and chromatin landscapes in early murine airway formation (Fig. 1A). A total of 4871 sequenced nuclei from control and 6046 from treated lungs were clustered into 19 distinct cell state/types (Fig. 1B). Annotation of the 19 clusters largely matched major cell types based on CellRef mapping to slightly older mouse scRNA-seq data [33, 34] and manual curation [25], with minor inconsistencies for more differentiated cell types (Supplemental Table E1). To avoid potential inaccuracies in classifying cells with low differentiation marker expression at this early stage, and to incorporate chromatin-level variations, we decided to use a data-driven approach and study the cell clusters defined by the single cell multiome data, categorizing them using broader classification: epithelial (clusters 1-3), mesenchymal (cluster 4-14, with cluster 14 being mesothelial), immune, endothelial, and other miscellaneous cells. These lineages were clearly distinguished by both UMAP embedding and gene expression markers (Fig. 1B and 1C).

Figure 1. Overview of E12 mouse embryonic lung gene expression and chromatin accessibility at single cell resolution:

(A) Scheme of experimental design showing dams were exposed to 0 (control) or 250 parts per billion (ppb) Cd and As for 10 weeks prior to mating and until harvest. Embryonic lungs were collected at E12 and processed for single-cell multiome. (B) Unified Manifold Approximation and Projection (UMAP) representing the integrated single-cell transcriptomic and epigenetic data, displaying 19 distinct clusters of cell populations. (C) Hallmark genes characterizing each of the 19 cell clusters. Dot size indicates the percentage of cells expressing the gene within each cluster, the intensity of the color indicates the average expression level. (D) Expression profiles of top differentially expressed genes across clusters. Purple indicates low and yellow indicates high expression. (E) The association of accessible chromatin regions with the hallmark genes. Dot size indicates the proportion of cells within a cluster showing activity of the corresponding gene, the intensity of the color indicates the level of activity. The activity is the sum of the peaks per cell in the gene body and promoter region or 2 kb upstream of TSS.

At E12, the predominant epithelial cells were defined by Sox9 (cluster 2 and 3) with a smaller fraction marked by Sox2 and Krt8 (cluster 1). A small percentage of cluster 2 and 3 cells have started expression of Sftpc and Sftpa1, the alveolar type II epithelial (AT2) markers, and only a few cells in cluster 2 have started expression of Ager, the alveolar type I epithelial (AT1) marker. Mesenchymal cell population, accounting for about 85% of the total cells, was characterized by expression of Col1a1 with more specific markers defining airway smooth muscle cells (ASM, cluster 4: Acta2), myofibroblast (cluster 5: Tgfbi, Acta2), mesenchymal multipotent progenitors (cluster 8: Wnt2), fibroblast (cluster 10-12: Col1a1), mesothelial cells (cluster 14: Wt1, Upk3b). The lungs showed remarkable gene expression heterogeneity across all cell clusters even at this early stage, with varying levels of many known growth factors and regulators (Fig. 1D). Variation in gene expression levels is corroborated by corresponding differences in region accessibility (Fig. 1E).

The proportions of defined epithelial clusters did not differ significantly between control and treated lungs based on unsupervised clustering (Fig. 2A). To account for differences in the total number of nuclei collected per sample—specifically, the higher number of nuclei recovered from Cd and As treated lungs—we normalized the number of cells in each cluster or lineage to the total number of nuclei per sample. This normalization allows for proportional comparisons across conditions independent of cell recovery yield. Following normalization, the proportion of Sox2⁺ epithelial cells remained unchanged between groups, whereas the proportion of Sox9⁺ epithelial cells was significantly reduced in the treated group (Fig. 2B). Statistical significance for changes in cell-type proportions was assessed using a Fishe’s exact test (Fig. 2). In contrast, mesenchymal cell composition shifted markedly (Fig. 2A). Notably, cluster 8, marked by Wnt2⁺ expression (Fig. 1D), increased by over 30% in the treated lungs, largely at the expense of cluster 9, 12 and 13. Lineages defined by Wnt2⁺ (primarily cluster 8), Acta2⁺ (primarily cluster 4) and Pdgfrb⁺ (primarily cluster 9-11) were segregated. Despite the rise in cluster 8, the overall proportion of Wnt2⁺ cells decreased following Cd+As exposure (Fig. 2B), likely due to reduced Wnt2 expression in cluster 12 (Fig. 2C). Pdgfrb⁺ cell proportions also declined, likely reflecting the reduction in cluster 9 (Fig. 2A).

Figure 2. Cell subpopulations of E12 mouse embryonic lungs affected by low dose Cd and As at E12.

CT – control; TRT – Cd and As exposed. (A) Normalized cell proportions of epithelial and mesenchymal cells (*P <0.05, **P <0.01, *** P <0.001, **** P <0.0001 comparing CT vs TRT by Fisher’s exact test). (B) Gene expression of lineage progenitor markers among cell clusters, distributed by UMAP embedding plots (*P <0.05, **P <0.01, *** P <0.001, **** P <0.0001 comparing CT vs TRT by Fisher’s exact test). (C) Gene expression profiles shown of select genes implicated in cell differentiation and development. Bonferroni FDR adjusted *P <0.05, **P <0.01, *** P <0.001, **** P <0.0001 comparing CT vs TRT by Wilcoxon’s test. (D) Pre-ranked (ranked by fold change of TRT vs CT) Gene Set Enrichment Analysis (GSEA) of genes differ between CT and TRT. Blue shows positive enrichment and red shows negative enrichment with the value showing Normalized Enrichment Scores (NES). (E) Immunofluorescent imaging of SOX2 and SOX9 shows reduced SOX9 protein level in exposed mouse lungs from E12 through E16. Scale bar: 100 μm.

Many key genes regulating distal lung epithelium in lung development [45] were broadly disrupted in all epithelial cells. Notably, Nkx2.1, Shh and Wnt7b decreased, while Foxp1 increased, with no change in Foxp2. Although partially redundant to Foxp2, Foxp1 is recognized for its unique role in some lung disease [46, 47]. Other genes, such as Mycn, Etv5, Notch1 were most significantly decreased in cluster 1 (proximal progenitors) whereas Spry2 was only reduced in cluster 3 (AT2 proximal progenitors) (Fig 2C). In mesenchymal cells, Cd+As caused a lineage specific downregulation, with cluster 8 (annotated as mesenchymal progenitors) being the most affected, showing downregulation of Col1a1, Zeb2 and Robo2.

Genes differentially expressed between treated and control groups were enriched in a wide range of cellular pathways (Fig. 2D), particularly those related to development and signaling. Genes involved in epithelial-to-mesenchymal transition (EMT) were upregulated across all clusters, most evidently in mesenchymal cluster 12 to 14 as well as epithelial cluster 2 and 3. Because branching morphogenesis requires reciprocal signaling between the epithelium and surrounding mesenchyme and even multiple rounds of conversion between cell states through processes of EMT and mesenchymal-to-epithelial transition (MET) [48], upregulation of this pathway likely indicates disrupted cell plasticity caused by the Cd and As treatment. A broad downregulation of Myc target genes and upregulation of many signaling pathways such as KRAS, hedgehog and TNF-α/NFκB showed the profound impact on cell proliferation and differentiation (see gene list in Table E2).

We confirmed protein-level changes in SOX2 and SOX9 using embryonic mouse lungs collected independently from the single-cell multiome experiment. In regions lacking airway lumens, SOX9 was broadly expressed in control lungs, consistent with active airway emerging. In treated lungs, SOX9 expression was reduced and more disorganized. Although SOX2-positive cell numbers were not significantly changed, many treated cells showed perinuclear or cytoplasmic SOX2 localization, indicating potential functional disruption of SOX2 (Fig. 2E).

Cell fate affected by Cd and As in early pseudoglandular lung

To provide a more granular landscape of cell fate, we performed multiome velocity analysis using MultiVelo [49], which uses the switch time and rate parameters of both chromatin accessibility and gene expression for improved accuracy in cell fate prediction. Results show a striking contrast in the predicted cell developmental progression between control and treated groups. The latent time (Fig 3A) along with the stream plots (Fig 3B), inferred by triplet parameters chromatin accessibility (c) and RNA (unspliced [u], spliced [s]), shows a normal, well-ordered transition from cluster 1 to clusters 2 and 3 in the control group, consistent with the expected Sox2-to-Sox9, stalk-to-tip transition in early branching mode. In contrast, the treatment group exhibited a disrupted pattern, marked by an absence of progression from cluster 1 to cluster 2, significant backflow from cluster 3 to cluster 1, and circulatory trends within clusters 1 and 2 (Fig 3B thick grey arrows). Among mesenchymal cell clusters, the control group displayed an expected transition originating from cluster 8 (Wnt2⁺ progenitors) towards other cell clusters, whereas treated progenitors failed to differentiate into other cell types.

Figure 3. Cell fate of E12 mouse embryonic lung affected by low dose Cd and As.

CT – control; TRT – Cd and As exposed. (A) Latent time in epithelial cells and mesenchymal cells estimated by parameters of RNA splicing and chromatin kinetics using MultiVelo. (B) Stream plots of epithelial and mesenchymal showing velocity predicted cell progression. Thick grey arrows show the major perturbations in the treated cells. (C) Heatmaps showing expressions of genes contributing most to velocity prediction. The cells are ordered by latent time from left to right (top bar color indicates the cell cluster as in panel B). Color of the heatmap represents smoothed spliced counts (yellow being the highest). The full list of genes is listed in Supplemental Figure E2. (D) Three-dimensional phase portraits of (c, u, s) (chromatin accessibility (c), unspliced pre-mRNA (u) and spliced mature mRNA (s)) triplets for genes Adamts18, Lama3, and Hmga2.

Spliced RNA counts of the genes contributing most to the predicted velocity showed diminished transcription of Bmp4, Lama3, Adamts18, Hmga2 in cluster 1 (Sox2⁺ progenitors) epithelial cells (Fig. 3C). While the control cluster 8 Wnt2⁺ progenitors were defined by active transcription of Sdc2, C1qtnf7, Zeb2, Hmga2, Tbx3, Gpc6, Ism1, Ror1, Cped1 and Can, the altered cell state continuum in the treated group was mostly attributed to an increased presence of cells with higher levels of Itga8, Wnt2, Sema3a, C1qtnf7, Ednra and reduced levels of Zeb2 and Hmga2.

Chromatin-level regulation affected by Cd and As in early pseudoglandular lung

Single-cell multiome offers a key advantage: the ability to infer chromatin-level regulation. Without correction of batch effects which corresponded to control vs treatment, scATAC-seq separated control and treated cells further than scRNA-seq data, indicating that the chromatin-level regulation is more sensitive to the effects of Cd and As exposure (Supplemental Fig E1B). Furthermore, velocity vectors inferred solely on scRNA-seq predicted incongruous flow in control cluster 2 (AT1 progenitor cells) and an unlikely segregation separating cluster 8 (mesenchymal progenitors) and cluster 9 and 10 (fibroblasts) (Supplemental Fig E2), which underscores the enhanced accuracy when chromatin-level information is considered in understanding cell fate.

Chromatin can either open or close while transcription is being either induced or repressed. For example, chromatin closing can occur either before or after transcription ends, corresponding to as model 1 and model 2 genes respectively, as described in Fig 3D. These events can be visualized by phase portraits of chromatin accessibility (c) and RNA splicing (u, s) [37]. We analyzed three genes that contributed most to the fate of epithelial cells, including Adamts18 (a member of the ADAMTS family of proteases involved in extracellular matrix remodeling and cell signaling), Lama3 (a component of Lamin-332 important for epithelial cell function), and Hmga2 (a non-histone chromatin protein regulating cell differentiation and growth). In control epithelial cells, chromatin closing occurs before transcription ends for the model 1 gene Hmga2, achieving its highest accessibility during the transcriptional induction. Conversely, for the model 2 genes Adamts18 and Lama3, chromatin closing occurs after transcription ends, with peak accessibility reached during the transcriptional repression. Exposure to Cd and As reversed the order of events for all three genes, indicating a disruption in the chromatin-level transcriptional machinery (Fig. 3D).

Lineage specific transcriptional regulators affected by Cd and As in early pseudoglandular lung

To further study the gene transcription mechanisms, we used SCENIC+ [50], a method that predicts enhancer-driven gene regulatory network (GRN) by linking transcription factors (TF) to the most plausible target regions and genes, identified as a set of enhancer-driven regulon (eRegulon). The most distinctive eRegulons between the Cd+As treated and control groups are shown in Figure 4A (full list in in Supplemental Figure E3).

Figure 4. Transcription factor (TF) activity and related gene regulatory network (GRN) of E12 mouse embryonic lungs affected by low-dose Cd and As.

CT – control; TRT – Cd and As exposed. (A) Heatmaps presenting the expression and activity of the most Cd and As affected TF in epithelial and mesenchymal cells. The color intensity and dot size on the heatmap indicates the level of expression and activity of each TF, respectively. †extended eRegulons (TF-region-gene links) based on extended (e.g. orthology based) motif-to-TF annotations. RSS - regulon specificity score, an indicator of predicted TF activity. The full heatmap with all TFs is shown in Supplemental Figure E3. (B) Gene regulatory network (GRN) of Gata6 and its connected regions and genes that differ between CT and TRT in epithelial cells. (C) In Silico perturbation analysis illustrating the effect on velocity predicted cell progression by a simulated perturbation turning off Gata6 in control cells. Thick grey arrows indicate major perturbations caused by either Cd+As in treated cells or the lack of Gata6 expression. (D) Trajectory inference among mesenchymal cells along pseudotime calculated by Monocle3 showing the effect on diminished differentiation of cluster 12 and 13, indicated by the thick grey arrows. (E) Gene regulatory network (GRN) of Gli2 and its connected regions and genes that differ between CT and TRT in mesenchymal cells. (F) Protein expression levels of GATA6 (red) and GLI2 (green) visualized by immunohistochemistry in control and exposed mouse embryonic lungs collected at E12, 14.5 and 17. Scale bars, 30 μm.

In epithelial cells, Gata6 exhibited the most significant reduction in estimated activity and gene expression, amidst a broader attenuation of TF activity (Fig. 4A). Cd and As reduced the number of connections among TFs, targeted DNA regions, and genes with the highest variability across all cell clusters (e.g. potential hallmark genes for different cell types and lineages). This reduction was most striking for Etv5 and Gata6. A similar, yet more pronounced, reduction in connections was found in mesenchymal cells from the exposed group compared to control (Supplemental Figure E3).

We tested the regulatory relationship between Gata6 and DEGs (control vs exposed) in the epithelial cells (Fig. 4B). In control lungs, Gata6 were linked to 15 DEGs involved in cellular structure integrity and migration (Itga3, Calm2, Podxl, Smim38, Plp2), signaling (Creb3l2, Ryr2, Akap5), and transcription and cell cycle (Tceal9, Mxi1, Bex1, Bex3, Bex4). Gata6 and other key TFs, such as Foxa1, Foxa2, Nkx2-1 and Etv5 were linked to shared genes or DNA regions, indicating a closely coregulatory network. In contrast, in exposed lungs, Gata6 was linked only to Peg3 (a paternally imprinted gene with a major role in fetal growth and cell lineage) and Sparc (a matricellular protein that functions in cell differentiation and migration), with no shared linkage with other TFs. These networks highlight the importance of Gata6 in mediating the adverse effect of Cd and As on fetal lung growth and maturation, consistent with previous findings on its essential role in distal lung differentiation [51]. Indeed, in silico perturbation of Gata6 resulted in a velocity pattern closely resembling that of the treated epithelial cells, presenting the lack of progression from Sox²⁺ proximal progenitors (cluster 1) to Sox⁹⁺ distal cells (cluster 2 and 3) (Fig. 4C), confirming its central role in Cd and As induced dysregulation.

Given the diversity of mesenchymal cell states which cannot be fully explained by known gene markers (Fig. 1B and 1C), we used lineage inference to identify five trajectories stemming from cluster 8 (mesenchymal progenitors). In contrast to the control, the trajectory leading to cell clusters 12 and 13 (Fig. 4D) is absent in exposed cells, consistent with the reduced proportion of these clusters (Fig. 2A). Gli2 was the most affected eRegulon in cluster 11 to 13, showing significant reduction in both activity and expression (Fig. 4A). In control mesenchymal cells, Gli2 was linked to Hsd11b2 (known Gli2 target regulating glucocorticoid metabolism [52]), Rbms3 (possible EMT effector), Slit3 [53] and Unc5c [54] (receptor-ligand crosstalk), and Zcchc3 [55] and Cacna2d1 (calcium channel protein). In treated mesenchymal cells, Unc5c, Slit3, Rbms3 remained linked to Gli2, while Celf2 (regulator of RNA processing) emerged as a new link, replacing Hsd11b2, Zcchc3 and Cacna2d1 (Fig. 4E).

The importance of GATA6 and GLI2 in response to Cd and As exposure was confirmed at protein level, exhibiting a consistent reduction in GATA6 and GLI2 protein levels throughout the pseudoglandular stage. This reduction was evident as early as E12 and became particularly striking at E14.5, when GATA6 protein expression peaked. Notably, in contrast to the relatively ubiquitous expression seen in controls, far fewer cells in exposed lungs expressed GATA6 and GLI2, exhibiting a spatially scattered pattern. This suggests that Cd and As exposure reduced GATA6 and GLI2 expressions in a manner that is potentially spatially or cell-type dependent.

Metabolome affected by Cd and As in early pseudoglandular lung

High-resolution metabolomics showed the perturbations at a different molecular level, presenting a clear distinction of metabolic profiles between the control and exposed lung samples (Fig. 5A) that were collected independently from the aforementioned experiments (Fig. 1A). Affected metabolic pathways included metabolism of tyrosine, ascorbic acid, β-alanine (mostly polyamine), xenobiotics, and purine (Fig. 5A). Changes in ascorbic acid and xenobiotics likely suggests a cell stress response indicated by decreased UDP-glucuronate, hypoxanthine and urate, while the decrease in polyamine products, spermine and 4-aminobutanal, aligned the downregulation of catalytic enzyme gene expression: Odc1, Srm and Sms in exposed mesenchymal progenitors (cluster 8) and some epithelial progenitors (cluster 1 and 3, Fig. 5C, D and E). Metabolites derived from dopamine were broadly affected. DOPA decarboxylase (Ddc), which synthesizes dopamine from DOPA, was broadly expressed across different cell types, showing a significant Cd+As induced reduction in mesothelial cells (cluster 14) (Fig. 5D and F).

Figure 5. Metabolomics of whole mouse embryonic lung at E12 affected by low-dose Cd and As.

CT – control; TRT – Cd and As exposed. (A) Heatmaps presenting normalized levels of HILIC+ metabolites that are different between CT and TRT. Unsupervised two-way hierarchical clustering shows clear distinction between the groups. Full details are in Supplemental Table E3. (B) Pathways enriched by significantly different HILIC+ and C18− metabolites (raw P <0.05 by Student’s t test). Only pathways with significant enrichment (P<0.05) are shown. (C) Schematic of polyamine (left), tyrosine (right) and metabolism. Enzymes are in italics, significantly altered metabolites are colored (red indicates decrease, and orange indicates increase) and non-detected metabolites or enzyme genes are in light grey. (D) Cell type-specific expression of catalytic enzymes in polyamine and tyrosine metabolism. Values indicate significant change, presenting Log2 fold change of average TRT vs CT levels in that cell cluster. Yellow indicates detection of the genes, but no significant difference determined by Wilcoxon’s test at Bonferroni adjusted P<0.05. (E) and (F) Levels of metabolites in polyamine and tyrosine pathways, respectively (*raw P <0.05, ** raw P <0.01, by Student’s t test). Full data is included in Supplemental Table E3.

Both Gata6 and the Gli family, have been implicated in upregulation of fatty acid biosynthesis in pathological context such as esophageal adenocarcinoma and liver steatosis [56, 57]. Cd and As caused downregulation of two genes in cell cluster 8 for enzymes catalyzing de novo fatty acid synthesis: citrate synthase (Cs) and fatty acid synthase (Fasn). Acot1, thioesterase that regulates levels of coenzyme A (CoA), free fatty acids and CoA esters, was downregulated in cell cluster 1 (Fig. 5D). Consistently, we found higher levels of CoA in treated lungs, and marginally increased acetyl-CoA and decreased palmitic acid levels (Fig. 5F). We also noted that those changes were connected to tyrosine metabolism alterations through decreased acetoacetate and to spermine synthesis through increased aminopropanol (Fig. 5C). Our analysis showed that multiple subnetworks of metabolism were affected by Cd and As, likely driven by the two progenitor clusters (cluster 1 and 8) that exhibited the greatest changes in metabolic gene expression.

In summary, our investigation systematically characterized the cellular impact of Cd and As exposure in early pseudoglandular lungs. We delineated distinct, cell-type-specific changes at regulatory levels of chromatin, transcription and metabolism, and showed that epithelial proximal and mesenchymal Wnt2⁺ progenitors were the most susceptible cell populations. Gata6 and Gli2 emerged as key transcriptional regulators central to the affected gene regulatory networks, presenting cell-specific reduction throughout the entire phase of airway formation.

Discussion

Developing organisms are constantly influenced by their environment. While high-dose exposure to toxicants are well-known for their teratogenic effects, most contemporary real-world exposures occur at low levels. Although the effect size of individual low-dose exposures may be small, their widespread and frequent occurrence contributes significantly to the population-level disease burdens. As shown by research supporting the Developmental Origins of Health and Disease (DOHaD) hypothesis, prenatal environmental exposure can shape long-term lung health trajectories. Thus, the goal of this study was to elucidate how environmentally relevant doses and mixtures of Cd and As disrupt early lung development at molecular and cellular levels.

The exposure levels used in our study (i.e. 250 ppb, or 1.25 μg daily intake) are comparable to real-world human exposure: Daily dietary Cd intake for average consumers were 10.6 and 11.2 μg in Swedish and French populations, and 23 and 18.9 μg for high-Cd consumers (i.e. exposure above 95 percentile), respectively [58]. In at least 61% Southwest community water systems in the U.S., As exceeded the U.S. EPA’s 10 ppb maximum contaminant level (MCL), with levels as high as 120 ppb in counties like Woodson, KS [11]. Our findings showing the cellular and molecular perturbations caused by prenatal exposure to low dose Cd+As, raises a concern that population-level exposures during early lung development may have unrecognized, potentially long-term effects on lung function and health, especially in communities with exposures in the top 10 percentiles.

During the pseudoglandular stage, lung development relies on tightly orchestrated, recursive programs that respond dynamically to environmental cues to ensure precise morphological and functional arrangements. Recent single-cell atlases of developing mouse and human lungs have delineated the temporal emergence and transcriptional dynamics of epithelial and mesenchymal lineages during branching morphogenesis [25, 26, 33, 59], while studies of hyperoxia-induced neonatal lung injury during the alveolarization stage have shown altered cell compositions, and profound cellular changes primarily driven by inflammation in a sex-dependent manner [60, 61]. However, few, if any, studies have investigated how environmental exposures disrupt these tightly regulated developmental trajectories, especially during the early stage of lung development. Our study extends this field by demonstrating that prenatal exposure to Cd and As perturbs epithelial lineage specification and destabilizes distal progenitor identity as revealed through integrative analysis of single-cell transcriptomes and epigenetic states. These findings demonstrate the primordial fetal lung’s vulnerability to environmental insults and highlight the power of single-cell multiomic tools in decoding exposure-induced reprogramming. Future work incorporating spatial transcriptomics, multi-organ models, and human organoid systems are essential to delineate conserved versus exposure-specific regulatory mechanisms.

We found that Cd and As exposure disrupted a broad network of known key factors regulating lung epithelial cell fate, including Sox9, Nkx2.1, Foxp1, Mycn, Etv5, Spry2, Shh, Wnt7b and Notch1 [45]. Two major fate alterations were observed: (1) a regression from distal AT2 progenitors (cluster 3) to proximal progenitors (cluster 1) as predicted by RNA splicing and chromatin dynamics (Fig. 3B), and (2) upregulation of EMT signatures in cluster 2 and 3 (AT1 and AT2 distal progenitors). These changes likely arise from perturbations in a multifaceted regulatory architecture composed of both coordinated and autonomous signaling mechanisms that together govern proximal-distal patterning and preserve epithelial plasticity and lineage identity. For instance, Nkx2.1 is a master regulator that preserves epithelial identity and suppresses transdifferentiation [62]. Sox9 maintains the undifferentiated state of distal progenitor and drives proliferation independently of Wnt/β-catenin signaling [63]. Wnt7b was among the most downregulated factors in epithelial clusters and Wnt2⁺ cell numbers were reduced, although Wnt/β-catenin signaling remained largely unchanged (Fig 2D). Nevertheless, Wnt7b-stimulated Pdgfrb⁺ [64] cell number was reduced in treated lung. The deficiency in the multiple regulators that promote epithelial differentiation may underlie the observed upregulation of EMT-related gene programs (Fig 2D). Taken together, our findings support a model in which Cd and As exposure destabilize epithelial lineage identity, potentially through the combined loss of distal markers and activation of EMT-associated reprogramming pathways.

Cd and As, both been studied as carcinogens, cause alterations in cell identities through complex molecular mechanisms. Both Cd and As independently affect epigenetic stability and gene expression through DNA methylation, histone modification, and signaling pathways [65-67]. As directly depletes the cellular methyl donor pool, causing DNA hypomethylation and aberrant gene expression [68]. Our data indicates that Cd and As impact chromatin dynamics of opening and closing, leading to gene expression changes that rewire cellular signaling and metabolic activities, which can further impact cell fate decisions.

Cd and As reduced polyamine biosynthesis in both epithelial Sox2⁺ and mesenchymal Wnt2⁺ progenitors. Polyamines like spermine are aliphatic polycations that bind to macromolecules, including proteins, nucleic acids and phospholipids, and regulate critical cellular process such as DNA replication, transcription and RNA translation, cell cycle, progression and apoptosis [69, 70]. Our results align with prior studies where polyamine depletion impaired embryonic growth and cell proliferation. For instance, DFMO (α-difluoromethylornithin, an ODC inhibitor) induce inhibition of polyamines increased embryo resorption, reduced DNA synthesis and cell proliferation in rodents [70], zebrafish [71] and human chondrocytes [72]. In particular, DFMO blocked upregulation of Sox9 and other chondrogenic markers, supporting a mechanistic link between polyamine synthesis and Sox9-mediated differentiation. In our study, the decrease in Sox9⁺ cells in differentiating epithelium at E12, 14.5 and 16 supports the speculation that reduced polyamine synthesis via ODC downregulation may impair proliferation of Sox9⁺ epithelial progenitors. These findings complement our previous work showing that Cd+As also impair pyrimidine biosynthesis and branching morphogenesis [4].

Our metabolomics data further implicates impaired dopamine biosynthesis by Cd+As. Alveolar epithelial cells produce dopamine from circulating DOP, as tyrosine hydroxylase (Tyr, the enzyme converting tyrosine to DOPA) is absent in early lung development (Figure 5D). In contrast to the unchanged DOPA levels, dopamine and its downstream metabolites were reduced by Cd and As, alongside decreased expression of Ddc (DOPA decarboxylase, the enzyme converting DOPA to dopamine) in mesothelial cells, suggesting a cell-type-specific bottleneck in dopamine synthesis. Dopamine regulates ion transport [73] and fluid clearance via Na⁺, K⁺-ATPase in the lung [74]. The implications of disrupted dopamine biosynthesis warrant further study, particularly in light of its known roles in lung epithelial physiology.

By constructing the gene regulatory network, we identified Gata6 and Gli2 as the most affected TFs by Cd+As at early pseudoglandular stage. Further experiments showed that this impact will likely reach maximum around E14.5 when both TF protein levels peaked over the developmental course. Gata6 is critical for the differentiation of the distal lung epithelium, especially in the formation of airway proximal tubules and the proper differentiation of AT2 cells. Its decreased expression and activity correlated with impaired AT2 cell trajectories (Fig. 3B) and echoes our earlier findings of reduced branching morphogenesis ex vivo [4]. Gli2, a key mesenchymal effector that transduces sonic hedgehog (Shh) signaling from epithelial cells, promotes cell proliferation by inducing cyclins D1, D2, and E1 [75]. Reduced Gli2 activity in cluster 11-13 may explain the diminished differentiation from mesenchymal progenitors.

Our study is limited by the lack of spatial and temporal data for single-cell lineage analyses. Branching morphogenesis during pseudoglandular stage operates as a recursive program, with each branching stage repeated multiple times. This inherent characteristic allows a single snapshot in time to potentially capture cells at various stages of morphogenesis and differentiation, making the study of maturation through single-cell data feasible even without temporal follow-up [76]. In vivo lineage tracing typically requires temporal control via agents like tamoxifen or doxycycline, which could interfere with the low-level Cd and As exposures that are central to our study. The computational strategies we employ are advantageous in providing data-driven, unbiased insights, especially in the absence of ground-truth directionality and for modeling complex expression kinetics beyond simple transcriptomics similarities. Nevertheless, our findings underscore the need for future studies that incorporate experimentally observed spatial and temporal dynamics in both animal models and human organoids to fully validate and contextualize our findings.

In conclusion, our study depicts profound, multi-layered molecular and cellular disruptions by Cd and As during the very early stage of lung development. By highlighting the impacts on chromatin accessibility, gene regulation, progenitor cell differentiation, and metabolism, we underscore the critical need for continued research and policy efforts to understand the environmental impact on developmental biology and to mitigate associated health risks.