Authors: Satoshi Mizoguchi (1Department of Anesthesiology, Yale School of Medicine, New Haven, CT 06520, USA; 2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA; 3Department of Surgical Oncology, Nagasaki University Graduate School of Biomedical Sciences, Nagasaki, Japan), Vi Lee (1Department of Anesthesiology, Yale School of Medicine, New Haven, CT 06520, USA; 2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA), Hahram Kim (1Department of Anesthesiology, Yale School of Medicine, New Haven, CT 06520, USA; 2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA), Sophie E. Edelstein (1Department of Anesthesiology, Yale School of Medicine, New Haven, CT 06520, USA; 2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA), Nuoya Wang (1Department of Anesthesiology, Yale School of Medicine, New Haven, CT 06520, USA; 2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA), Maria Tomas Gracia (1Department of Anesthesiology, Yale School of Medicine, New Haven, CT 06520, USA; 2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA), Colten Danelski (4Department of Biomedical Engineering, Yale University, New Haven, CT 06511, USA), Connor Haynes (2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA; 5Department of Surgery, Yale School of Medicine, New Haven, CT 06520, USA), Rachel Rivero (2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA; 5Department of Surgery, Yale School of Medicine, New Haven, CT 06520, USA), David Stitelman (5Department of Surgery, Yale School of Medicine, New Haven, CT 06520, USA), Tomohiro Obata (2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA; 3Department of Surgical Oncology, Nagasaki University Graduate School of Biomedical Sciences, Nagasaki, Japan), Allison M. Greaney (6Department of Immunobiology, Yale School of Medicine, New Haven, CT 06520, USA), Tomoshi Tsuchiya (7Department of Thoracic Surgery, University of Toyama, Toyama, 9300194, Japan), Themis R. Kyriakides (2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA; 4Department of Biomedical Engineering, Yale University, New Haven, CT 06511, USA; 8Department of Pathology, Yale School of Medicine, New Haven, CT 06520, USA), Naftali Kaminski (9Department of Pulmonary, Critical Care and Sleep Medicine, Yale School of Medicine, New Haven, CT 06520, USA), Micha Sam Brickman Raredon (1Department of Anesthesiology, Yale School of Medicine, New Haven, CT 06520, USA; 2Vascular Biology and Therapeutics Program, Yale School of Medicine, New Haven, CT 06520, USA; 10Program in Translational Biomedicine (PTB), Yale School of Medicine, New Haven, CT 06520, USA)
Categories: Article
Source: bioRxiv
Authors: Satoshi Mizoguchi, Vi Lee, Hahram Kim, Sophie E. Edelstein, Nuoya Wang, Maria Tomas Gracia, Colten Danelski, Connor Haynes, Rachel Rivero, David Stitelman, Tomohiro Obata, Allison M. Greaney, Tomoshi Tsuchiya, Themis R. Kyriakides, Naftali Kaminski, Micha Sam Brickman Raredon
Recent research has emphasized the critical role of cell state transitions in tissue homeostasis. In lung biology, transitional cells are recognized as a feature of tissue-scale processes during both normal physiology and disease. The precise way that transitional cell states emerge and are regulated remains to be determined. Engineered tissues, built in a laboratory through bioengineering approaches, allow detailed study of cellular states that are not commonly found in native biology, and allow opportunities to directly induce and manipulate cellular transitions. The following study explores and characterizes epithelial cell states that emerge via cellular reprogramming in a tissue engineering context.
Cellular transition between cell states is fundamental to tissue biology (1). Cells within tissues generally reside within highly structured biochemical milieus which constrain genetic regulatory network activation and thereby canalize gene expression and resulting cellular phenotype (2, 3). In the lung, precisely structured cell-to-cell signaling niches maintain this cellular canalization during homeostasis and guide cellular transition during both homeostasis and repair (4). In disease processes, signaling niches are disrupted and cellular transition is arrested or perturbed, leading to improper tissue homeostasis and the development of pathology (5–8). Understanding the principles of cellular transition and mapping causal relationships between extracellular milieu and intracellular state, is of utmost importance to tissue biology and pulmonary regenerative medicine.
The study of native tissues taken from animals, both healthy and diseased, has formed the foundation of biology and medicine. However, when we study in vivo tissues, we end up only studying cell states, and sets thereof, which are found during in vivo processes. There are many living, viable, cell states which do not exist within native biology. These include cultured cells (9–11), cells within organoids (12–14), and cells within perfused and engineered tissues (15–17). These cell states, while often highly different from cells found in the body, can provide significant insight into the fundamental nature of cell state regulation when studied computationally (18–20). Bioengineering allows researchers to manipulate cell states in a deliberate fashion, and therefore, engineered tissue biology provides a powerful window into the processes and control parameters for cellular transition.
In the present study, we profile the cellular and molecular profile of a specific pulmonary epithelial reprogramming and transition trajectory that we have deliberately induced as part of ongoing bioengineering efforts. We show that by placing a heterogeneous primary pulmonary epithelial population within an extracellular milieu designed to mimic the epithelial microenvironment found in fetal lung, we induce dedifferentiation, reduce cellular heterogeneity and stabilize a biphasic population of reprogrammed epithelial cells which display marked fetal character. Then, by changing the extracellular milieu to one which is designed to directly activate YAP, we induce ATI-cell associated gene expression, driving the cells directly into a transitional cell state which displays many of the transcriptional and phenotypic features known as hallmarks of transitional cells found in pulmonary fibrosis (IPF) and chronic obstructive pulmonary disease (COPD). We find that complete ATI-transition does not occur in the absence of additional alveolar niche cell types and/or on the timescale in which we have performed our experiments. Our findings confirm the ability of extracellular cues to regulate cell state and show that cells within bioengineered tissues can be driven into transitional states that highly mimic those seen in human clinical disease.
We hypothesized that adult lung epithelium might be susceptible to direct developmental reprogramming if cultured in a fetal extracellular milieu. To test this hypothesis, we dissociated native adult rat lung as previously reported (4) and enriched the resulting single-cell suspension for epithelial cells (Figure 1A). These cells were cultured in a media originally designed to maintain fetal cellular states (Fetal Growth LPM-3D Media adapted from (21), see Methods for detailed composition)(Figure 1B). Within this culture media, epithelial cells proliferated generously, forming multi-layered topologies and spheroids in 2D culture (Figure 1C). By P3, the cells began to change character, and the morphology became more irregular and less polygonal. By P4 to P5, the cells still proliferated but no longer formed complex topologies. The cells were passaged up to P12. RT-qPCR data was collected for cells in passages P0 through P5. qRT-PCR found specific genes of interest Trpm5, Sox9, and Pou2f3 to increase in bulk expression while Dclk1 expression decreased (Figure 1D-G). Cytospin slides of P0, P3, and P5 cells were prepared (n=3, biological replicates) and stained for Abca3, Ager, Epcam, Krt5, Sox2. Whole-slide stitched immunofluorescence images were captured, and images were segmented and processed using simpleSeg in R (22) (see Methods) to yield unbiased single-cell protein expression values (Figure 1H, Right). Intensity thresholds were titrated for each marker to bin positive and negative cells (Figure 1H, Left). Output data were used to estimate population-scale marker positivity at each passage (Figure 1I). These data showed a significant decrease in Abca3 expression (indicating loss of ATII-cell character), concurrent with an increase in Krt5 (gain in basal-cell character) and gain in Sox9 (gain in fetal character) from P0 up to P5.
To confirm that all of the cells we were studying were indeed epithelium, and to profile their genetic expression, we turned to single-cell RNA sequencing (scRNAseq). We captured cells from our starting epithelium-enriched population before culture (n=3 biologic replicates), Passage 0 in LPM-3D (n=3 biological replicates), Passage 3 (n=3 biological replicates), and fetal lung to use as a control. We sequenced each batch of cells to a target read depth of 50,000 reads/cell and processed the resulting data through iterative cleaning, annotation, and analysis (see Methods). Supplemental Table 1 shows data acquisition and QC metrics for all samples. We were able to identify analogous cell states in each experimental condition, which nonetheless showed great transcriptional difference at each time point. Supplemental Figure 1 show individual condition embeddings, clustering, annotations, intra-condition cell state markers and global cross-condition transcriptomic markers for each of these 4 experimental arms. Our findings fundamentally revealed that culture in fetal media induces a loss of population-scale cellular heterogeneity and a loss of differentiated adult character (Figure 2A-H). Although several cell states were still recognizable as similar to canonical native cell types, none displayed fully differentiated character. The observed decrease in cellular differentiation and drop in population heterogeneity was accompanied by increased transcription of many genes which mark fetal epithelium (Figure 2I-J). Reprogramming culture caused transcriptional upregulation of Wnt7b (23), Tgfb2 (24), Efnb2 (25), and Lamc1 (26), all of which are highly expressed by developing epithelium (Figure 2I-J). These changes occur in tandem with genes regulating cellular proliferation and inhibiting apoptosis (Mki67, Birc5, Topbp1) which also show high transcription during development (27–30)(Figure 2J). These findings demonstrated to us that we had reprogrammed adult epithelial cells to a fetal-like state, inducing many, though certainly not all, genetic programs seen within epithelial cells during normal lung development.
We next sought to explore how these reprogrammed cells would respond when seeded into decellularized lung and exposed to bare, acellular matrix. Following a protocol described in (31) and summarized in Supplemental Figure 3, we seeded 75 million P3 reprogrammed cells into the airways of decellularized rat lung matrices (n=3, biologic replicates) and cultured the constructs with 40 mL/min vascular perfusion and 4 bpm ventilation in LPM-3D media for 24 hours (Figure 3A-B). After 24 hours, the lungs were removed from the bioreactor and the lobes processed for analysis, including paraffin and frozen sectioning, bulk RNA isolation, and scRNAseq. Supplemental Figure 4 shows global histology for all replicates. H&E staining showed living cells attached throughout the airways and alveoli, with higher cellular density and collective organization occurring at the adventitial and sub-pleural boundaries (Figure 3A). The cells displayed evidence of local proliferation while maintaining cuboidal morphology regardless of observed location in either alveoli or conducting airways (Figure 3C-E). Fluid flow modeling using data from culture (32, 33) showed no significant change in global tissue mechanics during the short culture time (Figure 3F). Immunohistochemistry of alveolar regions of the lung showed clear cellular expression of Sox2, Sox9, and Tp63, but did not show clear cellular expression of ATI-markers Pdpn (RTI-40), Sema3e, or Vegfa, suggesting that alveolar matrix and LPM-3D media combination was insufficient to induce full ATI-cell program expression and that the seeded cells were maintaining fetal-like character at 24 hours.
To test this hypothesis, we performed scRNAseq of cells freshly isolated from these engineered tissues (n=3 biological replicates, 23,588 cells passing QC, see Supplemental Figure 2A-C for details), and compared the transcriptional profile to that of the P0 and P3 cultured cells. This allowed us to isolate the effect of bioreactor culture on reprogrammed cellular state. Upon transition to the 3D rat lung scaffold, we observed an unexpected increase in inflammatory response, denoted by transcriptional upregulation of the chemokine ligands and neutrophil attractants Cxcl1, Cxcl2, Cxcl3, and Cxcl6, as well as the pro-inflammatory cytokine Il1a (34–36) and the inflammatory mediator Angptl4 (37) (Figure 3K-L). We also noted elevated transcription of cyclooxegenase-2 (Ptgs2), an enzyme involved in inflammatory responses through prostaglandin generation, and activating transcription factor 3 (Atf3), a transcription factor which negatively regulates Ptgs2 (38) and has recently been established as an important regulator of lung regeneration and repair (39). In addition to this inflammatory activation, we also observed significant upregulation of the epidermal growth factor family ligands epiregulin (Ereg), amphiregulin (Areg), and heparin-binding epidermal growth factor-like growth factor (Hbegf). Epiregulin is upregulated during wound healing, inflammation, and angiogenesis, and is upregulated in lung epithelium undergoing infection or compressive stress (40). Similarly, Areg is produced during inflammation and promotes tissue repair (41), and plays a role in bronchial epithelial regeneration following infection-induced injury (42). Hbegf also partakes in tissue repair and has been implicated in post-pneumonectomy alveolar regeneration (43) and post-injury tissue remodeling (44). These findings suggested that exposure to bare matrix in the bioreactor environment, in the absence of other cell types, was inducing epithelial inflammation and the transcription of tissue repair genes but was insufficient to promote transition to mature homeostatic states.
Within the alveolar regions of the engineered lungs, we observed occasional instances of cellular flattening, which lead us to hypothesize that we might be able to drive the reprogrammed cells directly to an ATI-cell state. To test this hypothesis, we compared the effect of culturing P4 cells in LPM-3D media vs. DCIR+L media adapted from Burgess et al. (45) (Figure 4A), a YAP-pathway-activating culture medium designed to induce ATII-to-ATI transition. Figure 4B outlines the composition of each media, including the Media Foundation which they share, and the media-specific components. Brightfield microscopy images were taken on Day 1, 3, and 5 of culture for both conditions grown in parallel and are shown in Figure 4C. Compared to the LPM-3D media cultures, the cells in the DCIR+L media condition did not proliferate as rapidly, and did not reach confluence until Day 5. By Day 1, the edges of the cells were not smooth as they had been in the LPM-3D condition, and many exhibited finger-like protrusions on edges that did not have contact with neighboring cells. By Day 3, the cells grown in the DCIR+L media begin to have an observable greater average cell surface area. By Day 5, most cells have a flat morphology with a large surface area, consistent with ATI-like morphology.
Cells from both conditions were stained for Pdpn, F-actin, and YAP. Immunofluorescent images show Pdpn positivity in many of the large, flat cells cultured with the DCIR+L media, compared to few positive cells in the LPM-3D media condition (Figure 4D-I). Figure 4D-F demonstrates how cells in LPM-3D have few Pdpn-positive cells, no nuclear-located YAP signal, and are small in size and densely distributed. Figure 4G-I demonstrate how the majority of cells in the ATI Differentiation Media are positive for Pdpn and have a large surface area with cells spread out across the plastic. Figure 4G shows nuclear-translocated YAP associated with the differentiation transition. The increased expression of ATI and YAP-related genes in the DCIR+L condition was consistent with RT-qPCR data performed on the cells. The data revealed a significant increase in the expression of Pdpn, Ager, Clic5, Ankrd1, Ctgf, and Cyr61 among cells grown in DCIR+L media compared to the cells expanded in LPM-3D (Figure 4J-O), indicating transcriptional expression of ATI cell programming.
Armed with a way to induce ATI-directed transition in 2D, we sought to explore what the effect would be of culturing cells in 3D in DCIR+L within the bioreactor matrix environment. P3 cells were seeded into decellularized rat lung matrix and cultured in the LPM-3D media for 24 hours as before, but then switched to DCIR+L media on Day 1 (Figure 5A-B). Media exchanges were performed daily. Mechanical modeling showed insignificant changes in barrier properties or vascular recruitment during the 5-Day culture. At the end of the experiment, the lungs were removed from the bioreactor and processed in parallel for scRNAseq and paraffin and frozen sectioning. Supplemental Figure 5 shows global histology and metabolic monitoring for all replicates.
Gross appearance and H&E staining confirmed uniform seeding and widespread viability (Figure 5A). Unique to the DCIR+L condition, we observed marked flattening of cells against the decellularized alveolar walls, in some cases approaching the <1μm cytoplasmic thinness seen in native ATI-cells (Figure 5C-E). IHC staining showed the cells to be positive for the ATI-cell markers Pdpn/RTI-40 and Sema3e. We also observed instances of both nuclear and cytoplasmic YAP, Krt8, and Sox2, while observing almost no expression of Sox9 or Tp63 (Figure 5G-J). These findings suggested that the cells were transitioning toward an ATI-like state while losing pluripotency and basal-like character in exchange.
To test this hypothesis, we analyzed scRNAseq data from the DCIR+L lungs (n=3 biologic replicates, 15,728 cells passing QC, see Supplemental Figure 2D-F for details), using both the P3 (2D) and 24-hour (3D) samples as controls. The 5-day condition showed clear differentiation of progenitors toward ATI programming. We observed increased transcription of Krt7, a cross-species marker of ATI cells, Vegfa, which is predominantly produced by ATI-cells in the lung, and Sema3e, which is an ATI-specific spatial-guidance molecule responsible for maintaining endothelial organization in the capillary-alveolar barrier. The transition toward ATI-character was associated with significantly decreased transcription of cellular-proliferation genes and significantly elevated transcription of cell-cycle arrest proteins, PI3 kinase genes, and tumor-suppressor genes, including the transcription factor Elf3. However, we found that the transitioning cells displayed many features that are not associated with ATI cells but instead are well known from the literature as biomarkers of transitionally-arrested ‘aberrant basaloid cells’ seen in IPF and COPD. After changing the media from LPM-3D to DCIR+L and incubation until day 5, we observed the upregulation of several markers associated with epithelial states now well-described in clinical lung pathology (Figure 5K-L)(5, 46). Our 5-Day population expressed integrin beta 6 (Itgb6), an integrin that is a long-established biomarker of pulmonary fibrosis and wound healing (47, 48), Wnt7a, a ligand expressed by both ATI-cells during homeostasis and aberrant basaloid cells during fibrotic remodeling (49), and ephrin type-B receptor 2 (Ephb2), a spatial guidance cue involved in postnatal alveolar development (50, 51) that is also a prominent marker of aberrant basaloid cells (52). We noted elevated transcription of vimentin (Vim), an intermediate filament upregulated in cells undergoing EMT (53) and matrix metalloproteinase 7 (Mmp7), suggesting cell-mediated matrix remodeling, and both of which mark aberrant basal cells in clinical data (5). ATI-marker gene Aqp5 transcription was increased prominently in one of the three 24-Hour Conditions, while other ATI-associated markers, such as Pdpn showed only minimal transcription in 5-Day cells. These results collectively suggested incomplete transition toward an ATI-cell state, concurrent with adoption of an aberrant or arrested transitional phenotype similar to one seen in clinical lung pathology.
To formally test this hypothesis, we compared the cell states measured in this bioengineering context to cell states measured in native rat and human lungs. Figure 6A-B provides a visual overview of the cellular populations and tissues studied within this project. The fetal and adult tissues provided control data to contextualize the cell states observed within engineered 2D and 3D conditions, which, when proliferating in 2D culture, adopt morphology similar to that seen during native fetal tissue development. In 3D culture, we observed that the cells in the 24-hour lung culture still maintained fetal morphology, while the cells in the 5-day lung culture condition began to resemble the flattened cells lining the alveolar walls of the native adult lung without ever achieving the true physiologic thinness of fully differentiated ATI cells. This comparison suggests that the engineered transition from fetal-like to ATI-like adult cells in 2D and 3D culture may in some ways mirror the fetal to adult native transition and also suggests a structural linkage between cytoskeletal arrangement and transcriptional program activation in pulmonary epithelium.
We performed a global analysis, encompassing all test and control data generated during these experiments, to characterize the major feature shifts associated with population-scale cell-state traversal. Figure 6C-D is a manually curated heatmap designed to visualize key global transcriptomic trends of interest. The switch to DCIR+L media induced a noticeable reversion of transcription of inflammatory mediators such as Cxcl1, Cxcl2, Cxcl3, Cxcl6, and Il1a in our 5-day condition, to levels nearing or below that seen in the 2D cultures. This decrease in inflammatory gene transcription coincided with a decrease in transcription of wound repair genes Ereg, Areg, and Hbegf, and a decrease in basal cell markers Tp63, Krt5, and Krt14 (Figure 6C-D). This shift in cellular state was associated with a partial-but-incomplete differentiation towards ATI-cell character, with cells elevating their transcription of ATI-marker genes Aqp5, Pdpn, and Clic5. However, other genes which mark ATI cells in vivo, such as Akap5 and Ager, show only minimal transcription that does not approach the levels measured in native ATI-cells. These trends collectively suggested that epithelial inflammation, regenerative ligand expression, and proliferation were linked to basal/basaloid/proximal cell character, while ATI-program activation is linked to reversion of all three of these genetic programs.
To directly test this hypothesis, we developed a benchmarking strategy to quantitatively score each cell within a native archetype space defined by known in vivo cellular reference states. Manually curating gene lists from the literature, we developed 10 fully orthogonal scoring axes (see Supplemental Table 2), with no gene symbol overlap, allowing us to place each cell within a 10-dimensional space reflecting degree of differentiation toward or away from each chosen phenotypic reference state (ATI, ATII-ATI, ATII, Aberrant Basaloid, Basal, BASC, Ciliated, Hillock, Secretory, & Tuft). We iteratively titrated each reference state gene list so that module scores, when applied to both native rat (4) and human (5) reference data, preferentially marked the targeted cell states while excluding as much as possible all other non-target cell states (see Supplemental Figure 6). We then scored each individual engineered cell, from this entire project, within these 10 purpose-built and functionally validated orthogonal axes. Selected outputs from this method are shown in Figure 6E-H, with comprehensive outputs available in Supplemental Figure 7. The global analysis has been summarized in the form of 6 radar plots (Figure 6I), one for each condition, showing the globally scaled population-mean value for each orthogonal module score; overlaid radar plots can be seen in Supplemental Figure 8A-B. We note BASC and transitional ATII-ATI scores are dominant in the Fetal population. The native adult (‘St. Dclk1’) condition, as expected, scores highly for well-differentiated Tuft, Secretory, and Ciliated cell character. 2D culture in LPM-3D culture media reduced this differentiated character. In the Passage 3 and 24-Hour 3D conditions, we observed an increase in Basal and Hillock-like scores and a decrease in ATII and ATI-like character. During Five Day culture in DCIR+L, conversely, we observed an increase in ATI-like character, a decrease in Basal and Hillock-like character, and the highest Aberrant Basaloid scores measured across our entire study. Pairwise correlation analysis between all 10 module scores across all cells profiled (Supplemental Figure 8C) confirmed that ATI-directed differentiation was closely associated with the activation of aberrant basaloid programming and inversely correlated with basal and hillock gene transcription. Cell cycle scoring and cross plotting against module scores (selected plots shown in Supplemental Figure 9) confirmed that ATI-directed epithelial transition inversely correlated with cellular proliferation, and that the partially transitioned basaloid cells we created in this study displayed a transcriptomic signature consistent with cell-cycle arrest.
Replicable tissue bioengineering relies on our ability to deliberately navigate cellular state spaces for desired ends. In pulmonary biology, this task is complicated by a lack of high-quality quantitative data regarding what areas of state space are fundamentally available and/or unavailable in different experimental conditions and extracellular milieus. In particular, little is known about how native cellular state spaces (those biologically accessible and commonly observed in vivo) compared to engineered cellular state spaces (those available and either observed or inducible in vitro). This study sheds new, though certainly incomplete, light on this problem. We have shown that specific culture conditions remodel population heterogeneity, constrain epithelial cell programming, and can be predictably altered to drive cellular state transitions desirable for regenerative engineering and/or disease modeling.
The data in this paper makes clear that cellular markers which are highly specific to cell states seen commonly in native tissue contexts may not retain their meaning within an engineering context. The cell states that we observe in these experiments – within 2D culture, 24-hour matrix culture, and 5-day matrix culture, are not holistically identifiable as perfect orthologues of any cell states known to us from native lung biology in neither healthy nor diseased states, in neither human, mouse, rat, or pig biology. They represent cell states that are either specific to these precise experiments or have specific character associated with the regenerative processes that we have triggered therein. We suspect that the cellular states and state trajectories we have profiled here represent a small fraction of the state space that is truly available to living pulmonary epithelial cells when placed in different extracellular milieus.
The inflammatory patterns shown within our data are particularly interesting. Although inflammation is clinically thought of as a negative event associated with tissue destruction and dysfunction, here we observe inflammatory cues being produced by epithelial populations, in monoculture, within an engineered, sterile, non-mass-transfer-limited 3-dimensional tissue microenvironment. This is an unexpected finding, and one that we consider statistically robust given its clear and consistent presence across all replicates. Furthermore, our findings demonstrate a clear linkage between epithelial inflammatory programming, regenerative ligand transcription, and adoption of basal-cell-like character, as demonstrated by the cell state shift induced by 24-Hour bioreactor culture.
Our experiments in this area have allowed us to explore epithelial transition using a broad lens. By seeding reprogrammed epithelial cells onto decellularized, perfused extracellular lung scaffolds, and then inducing direct progenitor-to-ATI cell transition (without attempting to first induce and stabilize an intermediate ATII-cell state to mirror state trajectories found during in vivo epithelial homeostasis), we have unexpectedly driven cells into a Krt17^+^/Itgb6^+^ cellular phenotype that closely resembles transitionally-arrested aberrant basaloid epithelial cells seen in human lung pathology including IPF and COPD. This is an unexpected finding and one that was not deliberately engineered. It speaks to the ability of engineered ex vivo systems to recapitulate disease-like cellular states and opens many possible questions which might be explored and possibly answered using this system regarding how transitional cellular states are regulated and controlled.
Fundamentally, this paper shows that epithelial cells in the lung are highly plastic and are shaped by their extracellular milieu. Much recent literature has hinted that complete cellular differentiation may be inextricable from the differentiation of a community of cells — that proper tissue homeostasis may rely as much on the coordinated differentiation of local cellular niches as it does on the differentiation of the individual cells themselves. Our experiments here lend additional supportive, though far from conclusive, evidence to this broad theory. In engineered tissues lacking all non-epithelial lineages, we were able to induce ATI-cell programming but have been unable, thus far, to fully recapitulate ATI-cell differentiation, even when using culture media designed specifically for this purpose. We suspect that this may be because the epithelium requires properly structured information from neighboring alveolar niche cells to fully complete their transition into a stable ATI-cell state within the 3D culture environment. This hypothesis suggests opportunities for further experimentation in future studies leveraging multicellular engineered tissues.
Our study has several limitations. We do not know if the transitional arrest we observe is due to the lack of adjoining alveolar niche cells, improper media design or mechanical environment, or is simply a consequence of the short culture time. Future experiments should be performed to investigate each of these hypotheses. The experiments in this manuscript have used primary rat cells, as is common in the bioengineering field and for which the greatest amount of control and reference data exists, but future researchers may benefit from replicating these experiments using matched cells and scaffolds from other species, in particular mice (in which most in vivo genetic experiments have been performed), ferrets (which are an emerging model displaying physiologic lung architecture and terminal airways more closely mimetic of human) and/or human beings (which would provide the greatest evidence of clinical translational relevance). We have sought at every instance to perform highly controlled experiments and therefore maximize replicability, but cost and time limitations have forced us to limit ourselves to n=3 biologic replicates per experimental condition; additional biological replicates would strengthen the statistical power of this study and are likely warranted in the future. Finally, we have opted here, following much experimentation, to analyze our data without global cross-condition batch correction and/or integration. This is because our data displays a high-degree of (deliberately engineered) cross-condition variance in cellular state and phenotype, and we were not able to identify useful computational integration tools that could properly handle this variance without obscuring biologically relevant information and/or introducing additional unwanted artifacts. As computational tools develop that are better able to handle the broad cellular state-spaces encountered during joint analyses of engineered and native samples, we anticipate that our dataset may yield additional insights relevant both to fundamental pulmonary biology and bioengineering and might help to stress-test new computational tools seeking to bridge the gap between engineered and native tissue biology.
The LPM-3D Fetal Expansion Media was adapted from the media described in Nichane et al., 2017(21), originally designed to grow Sox9^+^ fetal epithelial progenitors and adapted here for use reprogramming adult cells to an fetal state. For use with rat cells, rat-derived growth factors were substituted for the original mouse-derived factors. This media consists of advanced DMEM/F-12 (Gibco, 12634010), 5 μg/mL Heparin (Sigma-Aldrich, H3149–10KU), 10 ug/mL Insulin (Sigma-Aldrich, 91077C), 15 The LPM-3D Fetal Expansion Media was adapted from the media described in Nichane et al., 2017 (21), originally designed to grow Sox9+ fetal epithelial progenitors and adapted here for use reprogramming adult cells to an fetal state. For use with rat cells, rat-derived growth factors were substituted for the original mouse-derived factors. This media consists of advanced DMEM/F-12 (Gibco, 12634010), 5 μg/mL Heparin (Sigma-Aldrich, H3149–10KU), 10 ug/mL Insulin (Sigma-Aldrich, 91077C), 15 ug/mL Transferrin (Sigma-Aldrich, T8158), 0.1% BSA fragment V (Gemini Bio, 700–104P), 1% P/S (Gibco, 15140122), 1% amphotericin B (Cytiva, SV30078.01), 0.1% gentamicin (Gemini Bio, 400–100P-010), supplemented with 50 ng/mL Rat Egf (Peprotech, 400–25), 50 ng/mL Rat Fgf9 (Novus, NBP2–35196-5ug), 50 ng/mL Rat Fgf10 (Peprotech, 400–42), 3 μM CHIR99021 (Cayman Chemicals, 13122–10 mg), 1 μM BIRB796 (Cayman Chemicals, 10460–10 mg), 10 μM Y27632 (Cayman Chemicals, 10005583–50 mg), and 1 μM A8301 (Cayman Chemicals, 9001799–10 mg). Media components are shown in Figure 2B.
The DCIR+L ATI Differentiation Media was adapted from the formulation described by Burgess et al., 2024(54) with the addition of 10 nM retinoic acid (Sigma-Aldrich, R2625), to further promote ATI-like differentiation(17, 55). This media consists of advanced DMEM/F-12 (Gibco, 12634010), 15 mM HEPES (Corning, 25–060-CI), 10 ug/mL Insulin (Sigma-Aldrich, 91077C), 5 ug/mL transferrin (Sigma-Aldrich, T8158), 0.1% BSA fragment V (Gemini Bio, 700–104P), 1% P/S (Gibco, 15140122), 1% amphotericin B (Cytiva, SV30078.01), 0.1% gentamicin (Gemini Bio, 400–100P-010), supplemented with 50 nM dexamethasone (Sigma-Aldrich, D4902), 0.1 mM 8-bromoadenosine 3´,5´-cyclic monophosphate sodium salt (Sigma-Aldrich, B7880), 0.1 mM 3-isobutyl-1-methylxanthine (Sigma-Aldrich, I5879), and 10 μM LATS-IN-1 (Cayman Chemicals, 36623), which is a LATS-inhibitor that activates the YAP-associated Hippo pathway(56). The media composition is shown in Figure 4B (56). The media composition is shown in Figure 4B.
Adult rat lungs were harvested from 230–250g male Sprague-Dawley rats, which were approximately 6–8-week-old. Ketamine (75 mg/kg) and Xylazine (5 mg/kg) working solution was injected intraperitoneally to anesthetize animals, followed by heparin (400 U/kg) to prevent blood clotting in the lungs. The thoracic cavity was accessed trans-diaphragmatically, and the trachea, lungs, and heart were exposed by removal of the anterior rib cage. The trachea and pulmonary artery (PA) were cannulated with Y-shaped 1/16-inch barbed fittings (Cole Parmer), and blood was cleared out by PA perfusion with PBS containing heparin (100 U/mL) and sodium nitroprusside (0.01 mg/mL). The lungs and heart were collected en bloc and transferred to a petri dish for further experimentation.
Fetal rat lungs were harvested at gestational stage E17, corresponding to the canalicular stage. Pregnant female Sprague-Dawley rats were anesthetized with inhaled isoflurane at 4% vol/vol and maintained at same condition. The rats were injected with sodium heparin (400 U/kg) intraperitoneally, followed by the administration of analgesia with subcutaneous injection of lidocaine (dose), meloxicam (1 mg/kg), and buprenorphine (EthiqaXR, 3.25 mg/kg). The abdominal cavity was accessed by a midline laparotomy incision, and the gravid uterus was exposed. Fetuses were extracted by cesarian section. The fetuses were secured to a working surface by pinning and the thoracic cavity was accessed trans-diaphragmatically, then the trachea, lung and heart were exposed by removing the anterior rib cage. The fetal left ventricle was opened with scissors, and the lung was perfused through the pulmonary artery (PA) with PBS containing 100 U/mL sodium heparin and 0.01 mg/mL sodium nitroprusside. The fetal lung and heart were extracted en bloc and placed on a petri dish for further experiment.
A dissociation solution comprised of warmed Dulbecco’s modified Eagles’s medium (DMEM) containing elastase (3 U/mL), collagenase/dispase (1 mg/mL), deoxyribonuclease I (DNase I; 20 U/mL) was used to dissociate the rat lung to a single-cell suspension. Adult rat lungs harvested as previously described were perfused with dissociation buffer through PA, then inflated through the trachea with dissociation solution. Each lung lobe was cut at the hilum and collected and submerged in dissociation solution in a conical tube and incubated for 20 min in a water bath at 37°C with gentle rocking. After incubation, the tissue was gently passed through a wire mesh strainer using a weigh spatula, then the strainer was rinsed with DMEM containing 10% FBS, 1% penicillin/streptomycin (P/S), 1% amphotericin B and 0.1% gentamicin to collect remaining cells. The tissue solution was centrifuged for 5 min at 300 × g to pellet the cells which were then resuspended with ACK lysing buffer (Gibco, A1049201) at a 1 ratio of pellet volume and incubated for 120 sec at room temperature. PBS with 0.01% BSA was added to the cell suspension, and spun down for 5 min at 300 × g. The pellet was resuspended with PBS with 0.01% BSA and filtered through a 70 μm cell strainer. The cells were centrifuged again for 5 min at 300 × g, resuspended, then filtered twice through a 40 μm filter. The single cell suspension was counted, assessed for viability, then diluted to 1 × 10^7^ cells/mL for Dclk1 enrichment.
Dclk1 antibody was conjugated with Dynabeads M-280 sheep anti-rabbit IgG (Invitrogen, 11203D) according to the manufacturer’s instructions. Cells were labeled with Dclk1 conjugated Dynabeads by incubating them together for 30 min at 4°C with gentle tilting and rotation. Tagged cells were magnetically isolated using the DynaMag-5 magnetic rack and rinsed with 0.01% BSA in PBS for 4 times then spun down for 5 min at 300 × g. Supernatant was discarded, and the pellet was resuspended in LPM-3D medium supplemented with 0.5 μg/mL phenol-red free Matrigel (Corning, 356237), then counted, assessed for viability with 0.4% Trypan Blue (Gibco, 15250061), and plated in a 6-well plate.
0.75–1 × 10^5^ cells were plated on 6-well plates after cell isolation and Dclk1 enrichment and cultured with LPM-3D supplemented with diluted growth factor reduced phenol-red-free Matrigel (5 μg/ml, Corning,356237). Cells were cultured in a CO2 incubator (37°C, 5% CO2), and media exchanges were performed the day after plating and every other day thereafter until the end of culture. P0 cells were cultured for 10 days, while cells P1 were cultured for 7 days before passaging. Passaged cells were either used for further experiments or resuspended in freezing medium (90% fetal bovine serum + 10% DMSO) and stored in −80°C overnight, then transferred into liquid nitrogen to be stored as reserves.
P3 LEPs were cultured in LPM-3D supplemented with growth factor reduced phenol-red-free Matrigel (Corning, 56237) at a concentration of 5 μg/mL until day 7 and passaged. The subsequent P4 LEPs were cultured in LPM-3D for 1 day, and the medium was replaced with DCIR+L media. Cells were cultured up to 4 days in DCIR+L, and media changes were performed every other day.
Lungs for decellularization were harvested as previously described from 300–350 g male Sprague-Dawley rats, which were approximately 8–12-week-old. The lungs, trachea, and heart were extracted en bloc and placed on a petri dish for cannulation and transfer to the decellularization apparatus. The pulmonary vein (PV) was cannulated with a 1/8-inch Y-shaped barbed fittings (Cole Parmer). Decellularization of lungs was performed based on sodium deoxycholate (SDC)-based protocol as previously described (57). In summary, the lungs were perfused through the PA with antibiotic/antimycotic solutions (10% P/S, 4% Amphotericin B, 0.4% gentamicin in PBS with Ca^2+^ and Mg^2+^), then rinsed with 0.0035% Triton X-100 (Invitrogen, HFH10) in PBS with ions. After 30 min incubation with benzonase endonuclease solution (20 U/mL) in airway at room temperature, lungs were rinsed with 1M NaOH in PBS, followed by a 3-step concentration gradient SDC solution (0.01%, 0.05% and 0.1%). Next, lungs were perfused with benzonase endonuclease solution (20 U/mL) through the PA and incubated for 1h at RT, then rinsed with 0.5% Triton X-100 (Invitrogen, HFH10) in 0.5M EDTA containing PBS. Furthermore, lungs were thoroughly rinsed with 2L of PBS and perfused with antibiotics/antimycotics solution in PBS, followed by 48-hour incubation with slow PA perfusion at 8 mL/min, and finally stored at 4°C until use. Decellularized lungs were used for further bioengineering experiments within no more than 1 week.
The bioreactor used was designed in-house and custom fabricated by Daryl Smith and Preston Smith at the Yale University Glass Shop. The chamber is made up of two a glass basin with twelve lateral ports around the perimeter and a glass lid with six vertical ports of varied sizes. The bioreactor basin contains the media, tissue, and cells, and facilitates mass transfer internally while interfacing with external apparatus components. The bioreactor lid creates a closed system and has ports to facilitate mechanical conditioning and chemical monitoring. Typically, only two vertical ports of the lid are used. One is used to interface the closed bioreactor environment with a vacuum to apply constant negative pressure. The second one is used to interface the bioreactor with a syringe pump to apply cyclical changes to the vacuum, inducing a ventilation effect. Connections are made between the external and internal environments of the bioreactor using port connectors. All port connectors consist of a hollow screwcap fitted with a glass tube. The glass tube is fitted with a short piece of silicone tubing and a female luer-to-hose barb connector fitting. The design and use of port connectors allows us to dynamically manipulate the internal bioreactor environment without compromising sterility. There are three distinct types of port connectors that are used to accomplish distinct functions. The first is a Short Flat port connector which has a glass tube with a wide internal end and no fitting for simple measurements or draining. The second is a Mid Double port connector which has a glass tube with a narrow end fitted with silicone tubing for making connections within the bioreactor, including to the lung. A Mid Double port connector is used to make a Drop Line port connector which is fitted with perpendicular tubing to draw media from the bottom of the basin for perfusion or media exchange and sampling. There are six external flow circuits that are connected to the basin, including the Pulmonary Artery (PA), Pulmonary Vein (PV), and Trachea (T) which interact directly with the lung, and the Pleura, Hollow Fiber Cartridge (HFC) and Feed Line which only directly interact with the media. The complete bioreactor apparatus consists of the bioreactor lid and basin, Perfusion Circuits, Pressure Transducers, a Hollow Fiber Cartridge, a Breathing Column, a Feed Line, a PES Filter, a hybrid Pulse Dampener and Bubble Trap device, a Vacuum Pump, and a Syringe Pump. These elements work together to achieve sterility, mass transfer, physiologic control, and real-time culture monitoring.
Whole decellularized rat lungs were cultured in the above-described bioreactor platform with a continuous pressure/flow monitoring system. Briefly, acellular lung scaffolds were mounted in a sterile glass bioreactor specially designed for whole rat lung culture and were perfused with media in a CO2 incubator (37°C, 5% CO2) at a rate of 20 mL/min with a peristaltic pump. This Pre-Conditioning process before the introduction of cells ensures thorough infiltration of media into the lung construct and allows any air bubbles to pass as to not obstruct the Seeding of cells or perfusion of media to them during culture. For cell Seeding, 90–125 × 10^6^ lung epithelial progenitors (LEPs) suspended in 10 mL of media were introduced into the airways of the lung dECM by gravity and negative pleural pressure applied within the bioreactor (target of 10–15 mmHg transpulmonary pressure). After 45 min of static incubation in a CO2 incubator, perfusion of media through the PA began at 8 mL/min, then the rate of perfusion was increased in increments of +4 mL/min every 30 mins up to 20 mL/min. It was then increased in increments of +7 mL/min, every 60 min up to 40 mL/min. Ventilation was added at the PA ramp-up rate of 20 mL/min, starting at Δ1 mmHg transpulmonary pressure (Ptr – Ppl) with a syringe pump, then increased in increments of +1 mmHg simultaneously with the perfusion rate ramp up to a final ventilation variation of Δ4 mmHg.
Six lungs were cultured in two different media and culture length conditions. Three lungs were cultured for 24 hours in LPM-3D media only, and another three lungs were cultured for 4 additional days in DCIR+L media. Close and continuous monitoring of the culture conditions from beginning to end is important for evaluating the state of the tissue and making informed decisions about tuning the conditions. Real-time data is collected by pressure transducers that are placed at key points in the bioreactor circuit to allow for monitoring of the internal pressure environment of the lung during culture. Media was also sampled every morning of culture to monitor the lactate, glucose, and pH levels.
Full-volume medium exchange was performed after 24 hours of culture for switching LPM-3D media with DCIR+L media. Half-volume media exchange was performed when media sampled with lactate less than 10 mM, while full-volume medium exchange was performed in a case of lactate greater than 10 mM. Lungs were removed from the bioreactor at Day 1 or Day 5, respectively, for each condition, and lobes were dissected for tissue dissociation for single-cell RNA sequencing and snap freezing each, and the remaining lobes were perfuse-fixed by gravity through PA with 4% paraformaldehyde for at least three hours then soaked overnight and used for histological analysis.
Throughout culture, the pressure transducers are continuously measuring the pressure of the PA, PV, Trachea, and pleura, while calculating the transpulmonary pressure which is the difference between the Trachea and the pleura pressures. Twice per day, barrier measurements were taken for the PA, PV, and Trachea. These measurements are performed by temporarily stopping flow in the circuit then restoring it to evaluate the instantaneous resistance to flow, per the exact protocol described in Engler et al. (32).
The measurements of the PA, PV, Trachea, and pleura pressure were processed using established MATLAB and R scripts from Raredon et al., 2021 to model and estimate the recruited area, hydraulic conductivity, and barrier pressures of each lung (33). The measurements of the PA, PV, Trachea, and pleura pressure were processed using established MATLAB and R scripts from Raredon et al., 2021 to model and estimate the recruited area, hydraulic conductivity, and barrier pressures of each lung (33).
Pressure transducers are included in the PA, PV, and Trachea circuits nearest to the lung and are strategically placed directly upstream from the flow from the lung for accurate intra-organ measurements. A pressure transducer is also attached to a Short Flat of the basin constituting the Pleura circuit for pseudo pleural measurements. LabChart was used to record continuous pressure measurement tracings of the PA, PV, Trachea, and Pleura, and to calculate a continuous transpulmonary pressure as the difference between readings of the Pleura from the Trachea. Twice per day, barrier measurements are taken for the PA, PV, and Trachea. These measurements are performed by temporarily stopping flow in each circuit, then restoring it to evaluate the instantaneous resistance to flow.
To measure the relative gene expression for cell-type markers of interest in the isolated and 2D cultured epithelial populations and to validate immunohistochemistry, quantitative real-time polymerase chain reaction (qRT-PCR) was performed. Cell suspensions of cultured cells were homogenized in lysis Buffer RLT (Qiagen). Total RNA was isolated using the RNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. RNA was reverse transcribed using the iScript cDNA Synthesis Kit (Bio-Rad) according to the manufacturer’s instructions. cDNA was reverse transcribed in a thermocycler using the following 5 minutes at 25 °C, 20 minutes at 46°C, 1 min at 95°C, and then held at 4°C for 5 minutes. PCR reactions were run in triplicate using 2 μL of cDNA in a 26 μL final volume with iQ SYBR Green Supermix (Bio-Rad) and 0.5 μL of each forward and reverse primer (dependent on gene being tested; primer sequences listed below). qRT-PCR samples were processed using the CFX96 Real-Time PCR Detection System (Bio-Rad) using the following initial denaturation step of 4 min at 95 °C followed by 40 cycles of PCR for 15 s at 95 °C, 30 s at 60 °C, and 30 seconds at 72 °C. Average threshold cycle values (Ct) from PCR 19 reactions were normalized against β-actin expression for all samples and are reported as a fold change against β-actin using the 2-ΔΔCt processing method. Fold-change for gene of interest transcript levels between samples (ex: ‘A’ and ‘B’) were calculated as 2-ΔΔCt where ΔCt = Ct (gene of interest) - Ct (β-actin) and ΔΔCt = ΔCt(A) - ΔCt(B).
Brightfield microscopy images were captured every other day for all 2D cell cultures using Infinity Analyze software and the Carl Zeiss^™^ Axio Vert.A1 Inverted Microscope fitted with the Lumenera INFINITY2–1RM Monochrome CCD Camera. Areas of interest were captured with 4x, 10x, and 20x magnification objective lenses each time. The Auto-Exposure feature of the Infinity Analyze software was used at a setting of 50% to achieve consistent imaging of all samples.
Cells shown in situ were fixed to the plates on which they were grown. Cells processed via cytospin were collected in cell suspension and centrifuged at 1000 x g for 5 minutes onto slides using a Cytospin 4 Centrifuge (ThermoScientific), then fixed with 4% PFA.
At the end of each bioengineering culture, the lung is removed from the bioreactor and placed in a Petri dish for imaging of the gross appearance. For the purposes of the cultures performed for validating this method, the bronchiole is ligated at the hilum before one lobe is dissected for tissue dissociation and a single-cell RNA sequencing, another lobe is dissected for snap freezing and archive storage, and the remaining lobes are fixed by gravity perfusion through the PA with 4% PFA for at least three hours then soaked overnight on a rocker. The fixed lobes are dissected and sent for paraffin embedding and sectioning by Yale Pathology Tissue Services (YPTS). H&E, EVG, and trichrome histology staining are performed, and unstained tissue section slides are returned by the YPTS.
IHC staining was performed for 2D-cultured cells and fixed tissues. Cell membranes were permeabilized using diluted Triton-X in PBS. Fixed and permeabilized cells were stabilized with Blocking Buffer by 1-hour incubation at room temperature, then incubated with primary antibodies (Table X) in Blocking Buffer at 4°C overnight. The following day, the cells were rinsed with PBS, then incubated with secondary antibodies in blocking buffer for 1 hour. After the secondary antibody incubation, the cells were rinsed, then mounted with PVA-DABCO and mounting glass.
After staining, the cells were imaged using the EVOS FL Auto 2 fluorescent microscope to record single-frame, single-channel, and multi-channel single-frame and stitched frame images.
H&E, EVG, and Trichrome IHC-stained tissue slides are sent to the Yale Pathology Tissue Services (YPTS) for digital microscopy, high-resolution, whole-slide scanning. Image files were analyzed using Aperio ImageScope and QuPath softwares.
Cytospin slides were made using LEP cells at passages P1, P3, and P5. Two IHC assays were performed on three replicates of each condition. Assay 1 tested the cells for the presence of EpCAM, Ager, and Krt5. Assay 2 tested the cells for the presence of Abca3, Sox2, and Sox9. The immunofluorescent stained cells were imaged using the EVOS Fl Auto 2 fluorescent microscope to create stitched images of representative sections of the cytospin slides. The stitched images were processed using the Bioconductor package simpleSeg (22) The stitched images were processed using the Bioconductor package simpleSeg(22) to segment and characterize the cells on a single-cell level. The general workflow of image processing using this package involves loading the images of interest, reading them into R Studio using EBImage (58), converting them into a Large CytoImage (58), converting them into a Large CytoImage List (59), segmenting them according to defined parameters using the simpleSeg function, extracting the raw single-cell information, storing the information in a workable data frame, and processing and analyzing it. List (59), segmenting them according to defined parameters using the simpleSeg function, extracting the raw single-cell information, storing the information in a workable data frame, and processing and analyzing it. To ensure the most accurate segmentation was performed, segmented images were produced following iterations of variable parameters within the segmentation function. Only after satisfactory segmentation was achieved was the single-cell data extracted. The package provides each cell within each image an index ID, and extracts the source image title, x and y-coordinates, surface area in pixels, the major axis length in pixels, eccentricity, total number of segmented cells in the image, and the average pixel intensity of each channel within a cell.
Our single cell RNA seq analysis consists of six cell condition populations titled Fetal, St. Dclk1, P0–2D, P3–2D, Engineered-24Hr, and Engineered-FiveDay, with three replicates per condition. Single cell data for samples St-1, St-2, St-3, P0–1, P0–2, P0–3, P3–1, P3–2, P3–3, BDL1, BDL2, BDL3, LEP_DL1, LEP_DL2, and LEP_DL3 were prepared using 10X Single Cell 3P v3.1 protocols. We aimed for a recovery of 10,000 cells per sample in our final library reads. Sequencing was done by the Yale Center for Genome Analysis (YCGA) on the Illumina NovaSeq 6000 platform. Fetal lung samples frl1, frl2, and NRL19 were processed using the 10X Single Cell 3P RNA Seq kit, with final libraries depths sequenced to approximately 50,000 reads/cell. FastQ files were processed through CellRanger (60).
We utilized Seurat’s FindMarkers() and intersect() function to create a set of reference lists of genes specific to an identity. We generated two overarching groups of reference lists, one with genes specific to experimental conditions, and the second with genes specific to each condition’s cell annotations. We also used FindMarkers() to construct marker lists of upregulated genes in the condition of interest for every permutation of condition comparisons in the merged Seurat object. For example, to generate upregulated marker lists for P0–2D, we made five Marker Lists comparing upregulated genes against Fetal, St. Dclk1, P3–2D, Engineered-24Hr, and Engineered-FiveDay. We created a variable titled power in all marker lists that was derived by multiplying the variables avg_log2FC and ratio (ratio itself was generated by dividing pct.1 by pct.2). We filtered the top 2000 genes in order of descending power and created a character vector containing the row names (genes) of the marker list. We filtered out several extraneous genes to make our final feature vector.We then used intersect() to create gene lists tailored to our needs. For example, to create a list of genes specific to the Fetal condition, we ran multiple intersect() functions that combined the genes upregulated in Fetal over St. Dclk1, in Fetal over P0–2D, in Fetal over P3–2D, in Fetal Lung over Engineered-24Hr, and in Fetal Over Engineered-FiveDay. The result was a character vector that contained genes highly expressed in Fetal over all other conditions.
We displayed expressed genes by using an in-house wrapper function for Complex Heatmap (63). To adjust expressed gene ordering in our heatmaps, we created data frames containing our feature vector of interest, and corresponding variables from chosen marker lists. We changed the order of genes in this data frame as needed based on pct.2 or power, with the aim of ordering genes that show more distinct expression levels between two conditions, before deriving a feature vector and creating the heatmap.
We created ten sets of native features for module score calculation (ATI, ATII, ATII-ATI, BASC, Ciliated, Secretory, Tuft, Basal, Hillock, and Aberrant Basaloid). We used an existing native rat lung cell atlas to generate our native module score features. We first generated a feature list from our cell population of interest that was upregulated against the entire native rat lung cell atlas, then a second list with genes upregulated against the total epithelial native rat lung population, a third list with genes upregulated against the most spatially distant epithelial populations in the embedding, and finally a list of all genes upregulated against the most spatially proximal epithelial populations in the embedding. We then iteratively used R’s intersect() function to select a final list of highly specific features. After generating module genes, we used Seurat’s AddModuleScore() function to calculate the module scores for each condition. To generate our Aberrant Basaloid set of module genes, we used a human IPF atlas to find genes specific to the human aberrant basaloid population, then found murine orthologs to calculate the final module score. We generated FeaturePlots of both individual conditions and the global merged Seurat object with the module scores. We created a radar plot based on the mean values of each module using the ricardo-bion/ggradar package. For an accurate representation of score changes across each condition, we used the scale() function on each group of module condition scores. For visual clarity, we split the plot by condition to highlight changes in mean score.