Mapping Embryonic Mouse Lung Development Using Enhanced Spatial Transcriptomics Pengfei Zhang, Benjamin K. Law, Niles Huang, Katharine Goodwin, Celeste M. Nelson, Michelle M. Chan Abstract The mature mammalian lung contains ∼50 different cell types that are located in precise positions along and around the airway epithelial tree. Here, we apply spatial transcriptomics to early embryonic mouse lungs to investigate how these spatial patterns are progressively established during development. Specifically, we optimized a microfluidics‐enabled, deterministic barcoding‐based sequencing (DBiT‐seq) workflow to achieve sensitive spatial mapping of tissue sections, resulting in up to a twofold increase in average transcript recovery compared with previous implementations. Spatial mapping of epithelial populations revealed that Sox9 + epithelial progenitors are located throughout the epithelial tree at early stages before becoming confined to the distal ends. Moreover, we describe a previously unrecognized mesenchymal cluster that exhibits consistent spatial position, expresses high levels of extracellular matrix proteins, and appears to promote embryonic lung innervation. These findings provide new insights into the spatial organization and cellular dynamics underlying early lung development, and demonstrate the power of spatial transcriptomics to uncover hidden patterns and populations of cells. The mammalian lung contains ∼50 cell types arranged along the airway epithelium. Here, deterministic barcoding sequencing (DBiT‐seq) is optimized to nearly double transcript recovery, enabling highly sensitive spatial mapping of the embryonic mouse lung. Analysis reveals proximally located “distal” progenitors and a spatially restricted mesenchymal population linked to early lung innervation. [ADVS-9999-e77678-g004.webp] Introduction The spatial architecture of the developing lung plays a key role in its morphogenesis. Initiated at embryonic day (E) 9.5, the mouse lung buds from the ventral foregut endoderm and then recursively branches to form a tree‐shaped structure [, ]. As the epithelium branches, the surrounding mesenchyme differentiates into diverse lineages that are localized to distinct regions throughout the lung. These tissues include mesothelium enveloping the outer surface of the lung, cartilage strengthening the trachea, smooth muscle wrapping around conducting airways, and pericytes supporting capillary endothelium. Collectively, the mesenchymal populations provide inductive signals that help specify proximodistal epithelial patterning, establishing Sox2 + epithelium in the proximal airways and Sox9 + epithelial progenitors at the distal tips [, , ]. An outstanding question in the field is the precise location of multipotent epithelial progenitors and how their location shifts during the establishment of the proximodistal axis in the developing lung epithelium []. Lineage‐tracing studies suggest that multipotent Id2 + epithelial cells, located at the distal tips of the tree, contribute to both the proximal conducting airways and distal alveoli depending on the developmental stage []. Subsequent work showed that two nested waves–marked by the expression of SOX2 and SOX9 – demarcate a compartment boundary between proximal conducting airways and distal gas‐exchange epithelium []. However, the precise spatiotemporal distribution of multipotent progenitors, especially at early embryonic stages, remains largely undefined beyond insights from immunostaining or in situ hybridization of a limited set of targets [, , , ]. In parallel, a major challenge in studying the lung mesenchyme has been to classify this highly heterogeneous tissue into different subtypes. Clonal labeling experiments found that smooth‐muscle progenitors are spatially restricted to the mesenchyme positioned ahead of the branching epithelium, underscoring the importance of anatomical location in subtype specification []. Consistently, recent studies identified anatomically distinct mesenchymal subtypes that serve different roles in response to tissue injury and alveolar epithelial cell proliferation [, ]. Collectively, these findings highlight that spatial context is a key determinant of mesenchymal identity and function. Incorporating anatomical location into classification frameworks may therefore help to define mesenchymal subtypes and elucidate their roles in lung development. Spatial transcriptomics offers the opportunity to complement single‐cell analysis and provide spatial context to the different cell types within a developing tissue, even those that are difficult to isolate into single cells for single‐cell RNA sequencing (scRNA‐seq), such as the epithelium [, ]. Consistently, efforts have recently been made to obtain comprehensive spatiotemporal maps of the embryonic lung [, , , ]. Unfortunately, current whole‐transcriptomic spatial‐mapping approaches suffer from limited resolution and/or sensitivity, hindering spatial mapping at the single‐cell level with high transcript detection sensitivity. Emerging technologies, such as Visium HD [], Slide‐seqV2 [], Seq‐Scope [], Stereo‐seq [], and OpenST [], achieve single‐cell resolution but the detection sensitivity remains low due to inefficient mRNA diffusion and capture [, ]. In contrast, microfluidics‐based approaches, such as deterministic barcoding‐based sequencing (DBiT‐seq), enhance detection sensitivity by delivering spatial DNA barcodes directly into the tissue, bypassing barriers to mRNA diffusion [] while achieving close‐to‐single‐cell resolution. Nevertheless, the transcript recovery for microfluidics‐based approaches remains low compared to scRNA‐seq [, ]. Here, we improved the detection sensitivity of spatial transcriptomics by optimizing the DBiT‐seq workflow []. Specifically, we optimized the tissue fixation conditions to better preserve mRNA integrity, incorporated a randomly primed second‐strand synthesis step to overcome the inefficiency in template switching, and relocated the reverse transcription (RT) from microchannels to a bulk reservoir to improve in situ RT efficiency. In total, these changes resulted in up to a twofold increase in transcript recovery. To highlight the benefits of this optimized workflow, we generated spatial datasets for tissue sections of the mouse lung at early embryonic stages. Combining our spatial transcriptomic data with reference scRNA‐seq allowed us to spatially map known cell types and uncover previously undescribed cell populations and tissue‐tissue interactions. We confirmed that proximal and distal epithelial cells exhibit distinct patterns of gene expression, but our spatial mapping unexpectedly revealed that ‘distal’ epithelial progenitor cells are located in proximal regions of the lung at early stages of development. Moreover, we identified a subset of mesenchymal cells enriched for extracellular matrix (ECM) markers that exhibited consistent spatial location patterns, which we designate as a novel “Matrix Mesenchyme” cell type. The Matrix Mesenchyme is located near the trachea and medial boundaries of the lung lobes and interacts with neurons to promote innervation of the developing organ [, ]. Together, these findings establish our optimized workflow and datasets as valuable resources for the community, offering new insights into the spatial organization and cellular dynamics of early lung development. Results An enhanced Spatial Transcriptomics Workflow To investigate the spatial dynamics of the embryonic mouse lung, we optimized a microfluidics‐based DBiT‐seq workflow to enable unbiased, high‐resolution spatial mapping of the transcriptome (Figure and Figures ). DBiT‐seq uses microfluidic chips to deliver oligonucleotides into tissue sections for spatial barcoding at a resolution that corresponds to the width of the channels on the chip (10–50 μm). Building upon a published DBiT‐seq workflow, we performed two rounds of optimization []. In the first round of optimization (Enhanced_DBiT_v1), we improved capture efficiency by changing tissue fixation conditions and incorporating a randomly primed second‐strand synthesis step (Figure and Figures ) [, , ]. We found that adding diethyl pyrocarbonate (DEPC) in the fixation buffer increases cDNA yield, likely through inhibition of endogenous RNase activity []. We also observed that fixing with methanol/ dithiobis(succinimidyl propionate) (DSP) yields less‐fragmented cDNA compared to 4% paraformaldehyde (PFA). In addition, incorporating a random primed second‐strand synthesis step increases the cDNA recovery by mitigating the known inefficiency of template switching (Figure and Figure ). This optimized DBiT‐seq protocol resulted in a 55%–150% increase in unique molecular identifier (UMI) detection per pixel when tested on sections of whole embryos using 25‐μm‐resolution chips, after standardizing for pixel size and sequencing depth (Figure ). The Enhanced_DBiT_v1 workflow also recovers more UMIs (∼1000–1250 UMI/100 μm2) than the original DBiT‐seq protocol (∼600–1200 UMI/100 μm2) and the recently established Open‐ST (800 UMI/100 μm2) when tested on sections of whole embryos using 10‐μm‐resolution chips (Figure ) []. In the second round of optimization (Enhanced_DBiT_v2), we focused on improving the in situ reverse transcription (RT) step, which is known to be inefficient [, , ]. Performing RT within microchannels introduces an additional challenge due to nonspecific adsorption of reverse transcriptase to the polydimethylsiloxane (PDMS) surface of the chip. To address the issue, we decreased the PDMS surface area during the RT step by replacing the 10‐μm channels with a 5‐mm reservoir and modified the design of the barcoding assay (Figure and Figure ) []. This change further improved UMI and gene recovery, consistent with previous observations but achieved here at 10‐μm resolution (Figure ) []. As a result, the Enhanced_DBiT_v2 workflow recovered ∼1800 UMI/100 μm2 at 10‐μm resolution with a sequencing depth of 50 M reads/mm2. This UMI recovery is 0.5 to 2‐fold higher on average than that previously achieved using the original DBiT‐seq protocol (Figure ) and, our saturation analysis using Spacemake showed that Enhanced_DBiT has better detection sensitivity compared to Slide‐seqv2, Slide‐tags, and Visium, though it saturates faster than Slide‐tags (Figure ) [, , ]. We also note that difference in tissue type and sample preparation may contribute to the observed difference in sensitivity. Altogether, these improvements to tissue fixation and assay design enhance the detection sensitivity of microfluidic‐enabled spatial transcriptomics. Using the Optimized Spatial Transcriptomics Workflow to Profile Early Development of the Embryonic Lung We then evaluated the optimized workflow by spatially mapping the developing mouse lung. We focused on coronal sections of embryonic lungs acquired from E11.5–E13.5, when the epithelial tree is established (Figure and Figure ). Specifically, we obtained 13 total replicates: 5 at E11.5 and 4 each at E12.5 and E13.5. The E11.5 and E12.5 replicates were analyzed at 10‐μm resolution, while the E13.5 replicates were analyzed at 15‐μm resolution to ensure we covered the entire tissue section. In total, we recovered 15 037 high‐quality pixels with a median value of 1 943 UMIs and 1 251 genes per replicate, consistent with tests using whole embryos (Figure and Figure ). We observed a slightly higher recovery (2,221 UMIs and 1,480 genes per pixel, respectively) at E13.5, likely due to the larger pixel size. To annotate cell identities, we first integrated our spatial transcriptomics data with published scRNA‐seq datasets and clustered the integrated data using the Louvain algorithm (Methods) [, , , , ]. Clusters, which contain both single cells and spatial pixels, were then annotated according to canonical marker genes, identifying major cell types such as Proximal and Distal Epithelium, Smooth Muscle, Mesothelium, and Mesenchymal Progenitors (Figure and Figure ). We also identified a Heart cell cluster enriched for canonical cardiomyocyte markers, including Actc1, Tnnt2, Tnni1, and Myl7, and a Blood cell cluster marked by high expression of Hba‐x, Hbb‐y. To assess potential batch effects, we visualized the UMAP embedding split by sample and by developmental time point (Figure ). Cell‐type composition is consistently represented across both time points and samples, with no evidence of systematic bias, indicating that batch effects do not substantially confound the downstream analysis. The proportions of annotated cell types are consistent within replicates of the scRNA‐seq and spatial transcriptomic data, respectively. However, we noticed that our spatial transcriptomics dataset captured a higher proportion of epithelial cells than the scRNA‐seq dataset (Figure ). This difference might be attributed to the fact that the latter approach requires dissociation of the organ into single cells, which has been shown to be challenging for epithelial tissues, including that of the embryonic lung [, ]. Mapping of the annotated cell types onto the tissue sections showed the expected locations, with an Epcam + epithelial tube surrounded by Actg2 + smooth muscle cells and a peripheral layer of Wt1 + mesothelium (Figure ), consistent across replicates and timepoints (Figure ). We also mapped the Heart and Blood cell clusters onto the tissue sections (Figure ). The Heart cell cluster was found in only three replicates and localized near the carina of the lung, consistent with its likely origin as dissection carryover. The Blood cell cluster, meanwhile, localized adjacent to vascular endothelium, as expected for erythrocytes within the vasculature. To determine whether the optimized protocol is close to resolving single cells per pixel, we carried out conditional autoregressive‐based deconvolution (CARD) []. This analysis generated similar spatial maps of the major cell types of the embryonic mouse lung (Figures ). However, we also found a portion of pixels that presented as a mixture of different cell types. For example, 21.5% of pixels annotated as epithelium were predicted to contain <50% epithelial cells. Across all cell types, 55.1% spatial pixels have a maximum predicted cell type proportion above 0.5, and different cell types showed a distinct degree of mixing, likely due to differences in transcriptomic similarity among cell types and the complexity of their local tissue microenvironment (Figure ). The observed degree of mixed cell‐type signals is consistent with the challenge of guaranteeing single‐cell capture for spatial transcriptomics mapping []. Nevertheless, our optimized DBiT‐seq workflow enables high‐resolution (10‐μm) mapping of the major cell types of the early embryonic mouse lung with high sensitivity. To avoid any confounding issues from the possibility of cellular mixtures in spatial data, all differential gene‐expression analyses in the following sections were conducted using scRNA‐seq data, which is highlighted where it applies. Spatiotemporal Mapping of the Epithelium Reveals Unexpected Pattern of Sox9 + Progenitors We took advantage of this dataset to investigate the differentiation trajectories and spatiotemporal distributions of the major tissues in the embryonic mouse lung. We first focused on the epithelium. According to previous lineage‐tracing studies, the lung epithelium is initially patterned along its proximodistal axis, with the distal Sox9 + Id2 + epithelial cells serving as the progenitors for different mature epithelial cell types (Figure ) [, , ]. While these Sox9+Id2 + epithelial progenitors have been shown to localize to the distal regions of the epithelial tree from E11.5, it is not clear if they are spatially restricted or when that restriction may occur []. As a result, the precise spatiotemporal distribution of these progenitors, especially at early embryonic stages, remains largely undefined [, , , ]. Our spatial mapping provides a comprehensive delineation of the spatiotemporal pattern within the epithelium. Specifically, we identified two distinct epithelial clusters that express high levels of the proximal and distal epithelial markers, Sox2 and Sox9, respectively (Figure ). However, the spatial mapping of these clusters reveals that the Sox9 + ‘distal’ epithelial cells, annotated using additional markers including Etv5 and Id2, are located in both the proximal and distal regions of the epithelium at E11.5 (Figures and ). The expected distal localization of the Sox9 + cluster is observed starting at E12.5 and E13.5 (Figures and ). To validate the spatial position of the proximal and distal epithelial clusters, we profiled the expression of Sox2, Sox9, and Shh (a marker of the entire epithelium) in lung sections using fluorescence in situ hybridization. Our results confirmed that the proximal epithelium contains Sox9+ epithelial cells at E11.5, which gradually diminish from this region at later time points (Figures and ). Quantifying the levels of Sox2 and Sox9 within Shh + epithelial pixels revealed that Sox9 is expressed in a subset of Sox2+ pixels, the fraction of which decreases from E11.5 to E13.5 (Figures and ). Analysis of scRNA‐seq reference data also uncovered a small population (6‐10%) of Sox2 + Sox9 + cells from E11.5 to E13.5 (Figure ). To determine whether this transcriptional overlap is also reflected at the protein level, we performed wholemount immunofluorescence staining of E11.5 lungs, which revealed low but detectable SOX9 protein expression within the proximal epithelium (Figure ). Together, these results support the presence of a transient Sox2 + Sox9 + epithelial population during early lung development. While Sox2 + Sox9 + epithelial cells have been previously observed in the embryonic human lung, they have not been reported in the mouse, to our knowledge []. This discrepancy may reflect differences in sampling strategies, as prior studies largely concentrated on the distal epithelium and later developmental stages [, , ]. We note that the Sox2 + Sox9 + epithelium observed here in the mouse lung likely arises through a distinct mechanism from the SOX2+SOX9+ epithelium reported in the human lung, which has been attributed to distal SOX2 expression rather than the proximal Sox9 expression found here. We next subclustered the epithelial cell populations to identify possible transition states between the Sox2 + and Sox9 + cells (Figure and Figure ). This analysis revealed spatially distinct epithelial subclusters, including Proximal, Proximal Intermediate (PI), Distal Intermediate (DI), and Distal, located along the proximodistal axis of the epithelial tube (Figure and Figure ). These subclusters show gradients of Sox2 and Sox9 markers as well as other markers including Igfbp5, Ccnd2, Etv5, Id2, Bmp4, and Sftpc (Figure and Figure ). This spatial and gene expression gradient is consistent with recent findings on airway epithelium during the perinatal and postnatal stages []. There is a shift in the composition of these subclusters over time: PI and DI states are observed primarily at E11.5 and transition to proximal and distal epithelial cells at E12.5 and E13.5 (Figure ). To identify genes involved in the initial commitment towards the proximal or distal fates, we analyzed the genes that are highly expressed in both PI and DI (Figure and Figure ). We found that Smoc2 is upregulated in both PI and DI compared to the other subclusters (Figure and Figure ) and spatial maps showed that Smoc2 is specifically expressed at high levels in the epithelium immediately adjacent to the proximal and distal‐most ends of the epithelial tubes (Figure ). We confirmed the spatial pattern of Smoc2 and another intermediate‐state marker, Cldn10, using in situ hybridization. Again, we used Shh to mark the epithelium, and our results showed that both Smoc2 and Cldn10 are expressed in the intermediate regions of the epithelial tree (Figure ). Smoc2 is more specific than Cldn10, as we find a low level of Cldn10 expression in both the proximal and distal epithelium, which lacks Smoc2 (Figure , zoom‐in; Figure ). This observation is consistent with our differential gene expression analysis (Figure ). Adamtsl1, upregulated in Proximal Epithelium (Figure ), is localized to the proximal side (Figure and Figure ), while Bmp4, Adamts18, and Thbd, which are highly expressed in Distal Epithelium (Figure and Figure ), are at the distal tips (Figure ). Bmp4, Adamts18, and Etv4 have previously been shown to be critical for distal alveolar differentiation, but the exact function of Smoc2 and Adamtsl1 in epithelial differentiation requires further investigation [, , ]. Notably, Icam1and Fgfbp1, which have been used as markers for intermediate epithelium, were either not identified as a specific marker (Icam1) or barely enriched in our analysis (Fgfbp1) (Figure and Figure ). We also built a transcriptional pseudotime axis that showed the progression between Proximal and Distal epithelial subclusters (Figure ). The regression analysis and module‐enrichment analysis along the pseudotime axis revealed similar gene‐expression enrichment in each state (Figure ). Altogether, our spatial mapping highlights spatiotemporal dynamics of epithelial transitional states and suggests possible markers of the intermediate states between proximal and distal epithelium. Spatial Mapping Reveals Matrix Mesenchyme That Localizes Near Neurons and is Enriched in Genes Driving Innervation To study the highly heterogeneous embryonic lung mesenchyme, we isolated the mesenchymal cell types and subclustered the dataset. This analysis revealed eight subclusters, including both mesenchymal progenitors and mature mesenchymal states (Figure and Figure ). We first focused on the three progenitor states that express high levels of Wnt2: Mesenchymal Progenitor 1 (MP1) is in the periphery of the lung lobes and observed at all time points; Mesenchymal Progenitor 2 (MP2) is in the periphery of the lobes as well but first appears at E12.5; and Mesenchymal Progenitor 3 (MP3) is close to the vascular endothelium and appears at E12.5 and E13.5 (Figure ). The intermediate and mature cell states in our dataset include Smooth Muscle Progenitors (SMP; Wnt2 low Myocd low Ptch1 high) and Smooth Muscle (SMC; Myocd high), both surrounding the epithelium, Precartilage (Sox9 high) surrounding the trachea and upper airways, and Pericytes (Pdgfrb high) adjacent to vascular endothelial cells (Figure and Figure ). Additionally, we identified a population that we label as Matrix Mesenchyme (MatrixM; Prrx1 high Ptn high Col1a1 high) due to its high expression of ECM proteins and related factors (Figure and Figure ). Given the presence of both progenitor and mature smooth‐muscle‐cell states in our dataset, we proceeded to examine the spatial dynamics of mesenchymal progenitors as they differentiate into smooth muscle cells, a process that has been well described using clonal lineage tracing and scRNA‐seq [, , , , , , ]. We reasoned that MP1 is the progenitor source for all mesenchymal cells due to its prevalence at E11.5 compared to MP2 and MP3. We then subset the data to include the subclusters of interest and constructed a pseudotime trajectory starting from MP1, through Smooth Muscle Progenitors, and into Smooth Muscle (Figure ). To test whether our dataset could provide spatial context for the differentiation trajectories of the mesenchymal subclusters, we mapped the pseudotime analysis spatially. We observed a spatial association between smooth muscle maturation and lung topology: the most immature MP1 localizes to the periphery of the lung, while Smooth Muscle Progenitors and Smooth Muscle cells progressively align closer to the bronchi along the distal‐to‐proximal axis (Figure ). This spatial pattern aligns with previously published observations that mesenchymal progenitors adjacent to epithelial tips are recruited to the sites of branching to differentiate into smooth muscle [, ]. We next focused on the Matrix Mesenchyme population, which expresses high levels of several ECM‐related genes, including Col1a1, Tmtc2, Col1a2, and Col3a1, as well as markers Ptn, Prrx1, and Osr1 (Figure and Figures a,). Notably, previous studies identified this population of cells but grouped it together with mesenchymal progenitors [, ]. While it shares some genes with mesenchymal progenitors, such as those related to collagen, Matrix Mesenchyme expresses minimal levels of the mesenchymal progenitor marker Wnt2 (Figure , Figure ) [, , ]. Our spatial dataset revealed that Matrix Mesenchyme is consistently located around the trachea and along the medial boundaries of the lobes, distinct from Wnt2 high Mesenchymal Progenitors that localize to the lung periphery (Figure and Figure ). To verify the spatial location of these cells, we conducted fluorescence in situ hybridization analysis for Prrx1, Osr1, and Sox9, together with DAPI staining for cell segmentation. We used Sox9 to distinguish between Precartilage (Sox9 high) and Matrix Mesenchyme (Sox9 low), as Precartilage also expresses both Prrx1 and a low level of Osr1, and localizes to the trachea (Figure and Figure ). We observed a spatial pattern of Prrx1 high Osr1 high Sox9 low cells, consistent with our spatial dataset (Figure and Figure ). Similarly, we also performed fluorescence in situ hybridization analysis for Prrx1, Col1a1, and Ptn and our results revealed co‐expression of these markers in cells in both trachea and carina of the lung, corresponding to the Matrix Mesenchyme (Figure ). We therefore conclude that Matrix Mesenchyme is a distinct population of cells in the embryonic mouse lung and likely does not function as a mesenchymal progenitor. To investigate the potential function of Matrix Mesenchyme, we performed Gene Ontology (GO) analysis on the most upregulated genes in this population compared to all other mesenchymal cells (Figure ). As expected from its high expression of ECM‐related genes, Matrix Mesenchyme shows strong enrichment in gene sets associated with Extracellular Matrix Organization (Figure , p = 9.3 × 10−4). Matrix Mesenchyme is also enriched in innervation‐related terms, including Axon Guidance (p = 9.3 × 10−4), Synaptic Transmission, Glutamatergic (p = 9.3 × 10−4), Nervous System Development (p = 9.3 × 10−4), and Neuron Projection Guidance (p = 9.3 × 10−4), suggesting a potential function in the development and patterning of the lung nervous system (Figure ). Consistent with this hypothesis, neighborhood‐enrichment analysis revealed that Matrix Mesenchyme and Neurons are frequently located in close spatial proximity to each other (Figure and Figure ). To tease out the potential effect of adjacent esophageal tissue on the Matrix Mesenchyme and its proximity to neuron populations, we removed pixels in the medial part of lung lobes and adjacent pixels close to the trachea and performed neighborhood‐enrichment analysis again on the remaining spatial data (Figure ). We found that these removed pixels indeed had a high proportion of Matrix Mesenchyme but our neighborhood‐enrichment analysis on the cleaned dataset still showed significant enrichment between the remaining Matrix Mesenchyme and Neurons, suggesting a potential interaction between these two cell populations (Figure ). Supporting this idea, cell‐cell communication analysis using cellchatV2 identified significant ligand‐receptor interactions between Matrix Mesenchyme and Neurons (Figure ) [, ]. Among these predicted signaling pathways, one notable example is the PTN‐PTPRZ1 ligand‐receptor axis (Figure ). Ptn encodes for pleiotrophin (PTN), a multifunctional growth factor that has been reported to promote the migration and neurite outgrowth of embryonic neurons [, ]. In our dataset, Ptn is highly expressed in Matrix Mesenchyme (Figure ), while two known PTN receptors, Sdc3 and Ptprz1, are both specifically expressed in Neurons at these stages of lung development (Figure ). Immunofluorescence analysis further revealed that PTN‐expressing cells are positioned adjacent to neuronal cells marked by class III beta‐tubulin (TuJ1) but are largely absent from epithelial cells marked by E‐cadherin (Figure and Figure ). To confirm these PTN‐expressing cells and associated neurons do not represent contamination from the esophageal tissue, we stained for WT1 together with PTN and TuJ1. While no distinct mesothelial boundary was observed at the medial edge of the lung lobes, the PTN+ cells in this region were clearly distinguishable from esophageal tissue (highlighted cells, Figure ). Similarly, no clear mesothelial boundary was observed in the trachea, where PTN+ cells lie proximal to neuronal cells (Figure ). The lack of mesothelial boundary in the carina and trachea is maintained at E13.5 (Figure ). Our observation is consistent with previous reports that WT1+ cells are absent from both surfaces of the carina and the dorsal surface of the trachea at comparable embryonic stages [, , ]. We reason that the local absence of mesothelium in these regions may facilitate neuronal migration from the esophagus to the lung, though this possibility requires further validation. To determine the role of PTN in neuronal patterning, we cultured embryonic lung explants in the presence of a PTN function‐blocking monoclonal antibody, which led to a significant reduction in the presence of neuronal filaments innervating bronchi (Figure and Figure , p = 0.012, see Methods). Together, these data suggest that Matrix Mesenchyme forms a specialized mesenchymal niche that supports early lung innervation through multiple signaling pathways, including downstream of PTN. Our spatial mapping using the Enhanced DBiT‐seq workflow thus identifies Matrix Mesenchyme as a previously unrecognized mesenchymal subcluster with distinct gene expression and spatial locations, which may promote lung innervation during early stages of lung development. Discussion Here, we optimized DBiT‐seq and demonstrated improved transcript recovery compared to existing whole‐transcriptomic spatial transcriptomics methods that rely on mRNA diffusion. Microfluidics‐based approaches such as DBiT‐seq enhance detection sensitivity by delivering spatial DNA barcodes directly into the tissue, bypassing the need for mRNA to diffuse to capture probes []. By optimizing tissue fixation, incorporating random‐primed second‐strand synthesis, and relocating the RT step to a bulk reaction volume, we further improved detection sensitivity. Collectively, these improvements doubled UMI recovery on average and increased gene detection while maintaining 10‐μm spatial resolution. We tested whether our optimized spatial transcriptomics protocol could uncover the spatial dynamics of key cell states by using the embryonic mouse lung as a model. Our spatial analysis confirms epithelial differentiation along the proximodistal axis of the organ but surprisingly reveals that Sox9 + Etv5 + Id2 + epithelial progenitors are localized in proximal regions of the epithelial tree at E11.5. This unexpected spatial pattern contrasts with the prevailing conceptual model of lung development, which assumes an initial proximodistal patterning of the epithelial tree, with distal epithelial cells serving as the sole progenitors for different mature epithelial cell types [, , ]. Nevertheless, our findings reconcile previous lineage‐tracing studies showing that Id2 + progenitors, located both proximally and distally at E12.5 in our spatial data (Figure ), give rise to both bronchiolar and alveolar cell types when labeled before E12.5 but only alveolar cell types when labeled at E16.5 []. Additional studies using precisely time‐controlled lineage tracing are needed to validate the differentiation trajectories of the epithelium at different time points. In addition, we observed co‐expression of Sox2 and Sox9 in subsets of the epithelium at E11.5, a phenomenon that has previously been reported in the embryonic human lung but not the mouse. Our optimized pipeline was especially useful for disentangling the heterogeneous mesenchymal compartment. We uncovered a previously undescribed mesenchymal cluster, Matrix Mesenchyme, which localizes near the trachea and medial boundaries of the lobes and is characterized by high expression of ECM‐related proteins. While this cluster is present in scRNA‐seq reference data, it was previously classified as part of the mesenchymal progenitor population [, ]. Two lines of evidence support our conclusion that Matrix Mesenchyme represents a distinct cell type. First, Matrix Mesenchyme expresses low levels of Wnt2 and markers of proliferation, which are expressed at high levels in mesenchymal progenitors. Second, Matrix Mesenchyme maps to locations that are distinct from mesenchymal progenitors, which localize to the periphery of the lung. We also predicted multiple interactions between Matrix Mesenchyme and Neurons. Among these, the PTN signaling axis emerged as a notable candidate. Lung explant culture experiments blocking PTN signaling reduced neuronal filament length, consistent with a potential role for Matrix Mesenchyme in innervation of the early embryonic lung. We want to note that while Matrix Mesenchyme shows the highest expression level of Ptn, this marker is also expressed by other cell types, including mesenchymal progenitors, precartilage, pericytes, and neurons themselves (Figure ). Given that the signals that promote migration of intrinsic neurons into the lung remain unclear, we postulate that spatial transcriptomics approaches can help fill the gaps by pinpointing the critical ligand‐receptor interactions between neuronal progenitors and other cell types [, , , ]. While spatial mapping of scRNA‐seq clusters facilitates the investigation of the epithelial intermediate subcluster and heterogeneous mesenchymal populations, identifying differentially expressed genes in pixels that are spatially distinct will also enable the discovery of spatially variable genes. To that end, we annotated obvious regions of domain branching and bifurcation within E11.5 and E12.5 lung sections (Figure ). We then performed differential gene expression analysis for Epithelium, Mesenchymal Progenitors, and Mesothelium, respectively. We found no significant difference between the different branching modes for Epithelium and Mesothelium, while we observed Mesenchymal Progenitors in bifurcation sites are highly enriched in Snai2 (Figure ). Snai2 is a known regulator of ECM‐related genes but the association between this transcription factor and branching morphogenesis of the lung requires further investigation. It is important to note that the gene‐expression pattern of Matrix Mesenchyme closely resembles that of Alveolar Fibroblast 2 (AF2), a cell type previously identified in the Mouse Lung Cell Atlas, except that Mfap5, a top marker of AF2, is not enriched in Matrix Mesenchyme (Figure ) []. The similarity between Matrix Mesenchyme and AF2 is unexpected, as AF2 is primarily observed postnatally, while our data are limited to early embryonic stages [, , , ]. Moreover, the distinct locations of each cell type suggest that Matrix Mesenchyme has a different function than AF2 at this early stage. Future studies are needed to determine how Matrix Mesenchyme matures at later stages and to clarify its relationship with AF2. To link our mesenchymal subclusters with previous studies, we note that MP1 has a similar gene‐expression profile as subepithelial mesenchyme, whereas MP2 and Matrix Mesenchyme resemble submesothelial mesenchyme []. However, our subclustering in this study resulted in finer resolution with additional spatial context. Our study has limitations. First, we acknowledge that cellular purity at each spatial spot/pixel remains a limitation shared by current sequencing‐based spatial transcriptomics technologies, including the platform used in this study. Because binning or pixel‐based capture aggregates transcripts from multiple cells, spots located at cell boundaries or in densely packed tissue regions may contain mixed transcriptomic signals from more than one cell type. Technologies that achieve single‐cell resolution without pixel binning, such as Slide‐tags, are in principle less susceptible to this issue. Rigorously benchmarking cellular purity across platforms would require profiling serial sections with each technology alongside matched scRNA‐seq data for deconvolution, which is beyond the scope of the current study. Nonetheless, higher‐resolution methods (e.g., Visium HD, Stereo‐seq) are expected to achieve better cellular purity within their binned spots compared with lower‐resolution technologies, and we have taken this into consideration when interpreting our spatial data by performing all differential gene‐expression analyses using scRNA‐seq data to avoid any confounding issues. Another limitation is that the top markers of Matrix Mesenchyme are not specific, including Ptn and Prrx1, making it challenging to deconvolve how other cell types contribute to lung innervation via the PTN‐PTPRZ1 signaling pathway. The roles of other gene markers driving innervation terms remain to be investigated. As we were preparing this manuscript, a similar spatial mapping of embryonic lungs using the original DBiT‐seq protocol was published []. However, that work focuses on more mature embryonic stages and lacks the insight we have on the early embryonic stages, likely due to limited resolution (20 μm) and/or detection sensitivity. Our optimized spatial transcriptomic mapping shows a superior efficiency of transcript recovery at high resolution, which allowed us to reveal the early spatiotemporal pattern of epithelial progenitors and identify mesenchymal subpopulations with distinct distributions and developmental roles, providing a valuable resource for studying lung morphogenesis and innervation. Experimental Section/Methods Mice Embryos were isolated from euthanized timed‐pregnant mice (Mus musculus, C57BL/6J) at E11.5–13.5 and stored in chilled phosphate‐buffered saline (PBS) prior to dissection. Lungs were dissected following standard protocols in chilled PBS supplemented with 10‐U mL−1 penicillin and 10‐μg mL−1 streptomycin (Sigma‐Aldrich). All procedures involving animals were approved by Princeton University's Institutional Animal Care and Use Committee (IACUC Protocol #1855). Mice were housed in an AAALAC‐accredited facility in accordance with the NIH Guide for the Care and Use of Laboratory Animals. This study was compliant with all relevant ethical regulations regarding animal research. Embedding and Sectioning To embed samples, the dissected lungs were incubated for 1 h on a shaker at 4°C in 0.3 mL each of 20% sucrose, 30% sucrose, and 1:1 OCT:30% sucrose solution with 0.6‐μL DEPC, sequentially, before embedding in disposable base molds using dry ice. The embedded lungs were warmed up at −20°C in a cryostat and sectioned into ∼7‐μm thickness before being placed in the center of poly‐L‐lysine‐coated glass slides. The cryosections were stored at ‐80°C before testing. In total, we collected 5 sections from five E11.5 embryonic lungs, 4 sections from three E12.5 embryonic lungs (2 out of 4 sections are from a single embryonic lung), and 4 sections from four E13.5 embryonic lungs. Immunostaining Lungs were dissected and fixed in 4% paraformaldehyde (PFA) in PBS on a shaker at 4°C for 15 min. Lungs were then washed four times for 15 min with 0.5% Triton X‐100 in PBS and blocked overnight at 4°C using 5% donkey serum in PBST on a shaker. Samples were then incubated with primary antibodies against PTN (mouse, 1:100, Santa Cruz Biotechnology sc‐74443), E‐cadherin (rat, 1:400, Invitrogen 131900), TuJ1 (rabbit, 1:400, R&D Systems MAB1195‐SP, or chicken, 1:200, Novus Biologicals NB100‐1612), Sox2 (goat, 1:200, R&D Systems AF2018), Sox9 (rabbit, 1:500, EMD Millipore AB5535), WT1 (rabbit, 1:100, Abcam ab89901) for three days (E12.5 and explants) or two days (E11.5) at 4°C on a shaker, followed by Alexa Fluor‐conjugated secondary antibodies (1:200 for Thermo Fisher Scientific A21121; 1:400 for Thermo Fisher Scientific A21245, A48263, A78946, A32788, Biotium 20015; 1:500 for Thermo Fisher Scientific A78947, A32816, A32790) and Hoechst (1:1000) for another 2 days at 4°C. Lung Explant Culture and Neuronal Filament Quantification Lungs (Mus musculus, CD1) were dissected at early E11.5 with special care to preserve the structure of the trachea and the regions between medial boundaries of lung lobes, in order to maintain neuronal tissues as intact as possible. Lung explants were then cultured as previously described []. Briefly, dissected lung explants were positioned on top of a porous membrane (nucleopore polycarbonate track‐etch membrane, 8‐μm pore size, Whatman), floating on the surface of DMEM/F12 medium supplemented with 50 units mL−1 of penicillin and streptomycin, 5% fetal bovine serum (FBS, heat inactivated; Atlanta Biologicals), containing either desalted anti‐PTN antibody (6 μg mL−1; Santa Cruz Biotechnology sc‐74443) or mouse IgG1 antibody (Mouse IgG1 Isotype Control, Invitrogen, 026100) as a control. Lungs were cultured for 36 h, then fixed in 4% PFA for 15 min at 4°C, followed by immunostaining. We note that the staging of the embryonic lungs and the culture duration are critical for this inhibition experiment. We then used Imaris (v10.2.0) to quantify the neuronal filaments in the cultured lungs. Filament structures were reconstructed using the Tree AutoPath Algorithm to define neuronal segments and identify filaments. The estimated largest diameter for starting points was set to 16 μm, while the thinnest diameter was set to 0.958 μm. To minimize the effect of variability in the initial neuronal cell number introduced during dissection, analysis was restricted to neuronal filaments innervating the right and left lung lobes (see highlights in the boxed regions in Figure and Figure ), and the total filament length was measured. We then performed a two‐tailed t‐test to compare the total filament length across different conditions. Fluorescence In Situ Hybridization Adjacent lung sections were used for fluorescence in situ hybridization to validate the spatial transcriptomic mapping. Specifically, we used the RNAscope analysis Multiplex Fluorescent V2 Assay (ACD) protocol. Probes used were Mus musculus Shh (314361‐C1), Sox9 (401051‐C3), Sox2 (401041‐C2), Prrx1 (485231‐C1), Osr1 (496281‐C2), Smoc2 (318541‐C2), Cldn10 (541041‐C3), Col1a1(319371‐C2), and Ptn (486381‐C3). Sections were imaged on a spinning disk confocal (BioVision X‐Light V2) fitted to an inverted microscope. Microfabrication Our spatial tests used two microfluidic devices – Chip A for loading barcodes A1‐A50 and Chip B for loading barcodes B1‐B50 in an orthogonal direction, following a design from previous studies []. Master molds were microfabricated via standard photolithography on 4‐inch silicon wafers. Specifically, a layer of SU8‐2025 photoresist (∼25 μm) was spin‐coated and patterned using the Heidelberg DWL66+ direct write lithography system prior to developing. The mold was hard baked at 150°C and silanized for 15 min before device fabrication. Chips A and B were fabricated by adding ∼30‐g 10:1 (base to curing agent ratio) Sylgard 184 (Dow Corning 4019862) onto the mold and curing for 2 h at 65°C. The inlets and outlets of each device were then punched and cleaned using IPA for 10 min with sonication before use. Spatial Sequencing Using Enhanced_DBiT_v1 Preprocessing Sections were taken from the freezer and imaged at the desired optical resolution (4×, 10×, or 20× air objective). The sections were fixed using a mixture of 800‐μL ice‐cold methanol, 15‐μL 50‐mg mL−1 DSP in dimethyl sulfoxide, and 1.6‐μL diethyl pyrocarbonate for 15 min. The sections were then cleaned by pipetting 1 mL of PBS‐RI (998.6 μL PBS with 1.4 μL RNase inhibitor) across the tissue three times and then dried with gentle air flow. The tissue sections were then blocked for 30 min in 100‐μL BSA (10 mg mL−1) in PBS at 4°C in a humidified chamber, followed by washing three times with 100‐μL 0.5x PBS‐RI for 5 min, followed by another two quick washes using 0.5x PBS‐RI. To start spatial barcoding, the first microfluidic device was centered on the tissue. We then took a scan to record the position of the channels on the tissue. We added and loaded 1‐mL PBST into the reservoir to permeabilize the tissue section for 20 min. We then removed the remaining PBST from the reservoir and the inlets/outlets and washed the channels by adding ∼1‐mL 0.5x PBS‐RI to the reservoir and flowing for about 10 min. We applied vacuum for 5 min to remove any remaining washing buffer from the channels. Reverse Transcription We first prepared the following RT mixture: 29.2‐μL RNase‐free water, 55‐μL 5×Maxima RT buffer, 110‐μL 0.5×PBS‐RI, 1.76‐μL RNase inhibitor (Enzymatic), 3.52‐μL Superase In RNase Inhibitor (Ambion), 13.75‐μL 10‐mM dNTPs, 27.5‐μL Maxima H Minus Reverse Transcriptase, and 6.88‐μL 100‐μM template switch oligos (Table ). We then combined 4.5‐μL RT mixture and 0.5 μL of 25‐μM Barcode A1‐A50 (Table ), before introducing the barcode mixture into the inlets of the microfluidic device sequentially. We then applied the vacuum to fill all channels with solution and incubated in a humidified chamber at room temperature for 30 min followed by 42°C for 90 min for in‐tissue reverse transcription. Following reverse transcription, we removed the barcode A mixture from the inlets/outlets and cleaned the channels by flushing with 0.5‐mL NEB buffer 3.1 with 1% Enzymatics RNase Inhibitor (10 μL for each inlet) for 10 min. We then peeled off Chip A and briefly dipped the slide in a 50‐mL Falcon tube with RNase‐free water and gently dried under air. Ligation We then placed and centered Chip B on the tissue, imaging the channels to ensure proper placement. After firmly clamping Chip B on the tissue section, we prepared the ligation mixture as follows: 55.39‐μL RNase‐free water, 27.5‐μL 10× T4 ligase buffer, 11.2‐μL T4 DNA ligase (400 U μL−1), 2.24‐μL RNase inhibitor (Enzymatics), 0.71‐μL Superase In RNase Inhibitor (Ambion), 5.5‐μL 5% Triton‐X‐100, and 117.9‐μL 1× NEB buffer 3.1 with 1% RNase inhibitor (Enzymatics). We combined 4‐μL ligation buffer with 1‐μL 25‐μM barcode B1‐B50 (Table ), before introducing the barcode B buffer into the channels sequentially. We then incubated in a humidified chamber at 37°C for 30 min for in‐tissue ligation of barcode Bs. Following ligation, we cleaned the channels by flushing with wash buffer (500‐μL PBS, 5‐μL 10% Triton X‐100, 1.25‐μL Superase In RNase Inhibitor) for 10 min. We then peeled off Chip B and dipped the slide into a 50‐mL falcon tube with RNase‐free water before gently drying under air. We took a final image of the tissue section to enable mapping and alignment of the sequencing results. Lysis and Sub‐Library Generation To collect tissue‐derived cDNA, we added 10 or 15 μL of lysis buffer (0.5x PBS, 10‐mM Tris pH 8.0, 200‐mM NaCl, 50‐mM EDTA pH 8.0, 2.2% SDS, 2‐mg mL−1 Proteinase K) for 10‐μm and 15‐μm tests, respectively, to the spatially barcoded tissue sections. We then incubated the tissue section at 55°C for 2 h in a tightly sealed and humidified chamber. We collected the lysate into a 1.5‐mL DNA low‐binding centrifuge tube and then added the same amount of extra lysis solution to wash out the reservoir and retrieve as much lysate as possible. The cell lysate was immediately stored at −80°C. Cell lysate Purification To purify cDNA from cell lysates, we first prepared 40‐μL Dynabeads MyOne Streptavidin C1 beads (Thermo Fisher) per sample, following the supplier's instructions. We then added 7 volumes of DNA Binding Buffer to each volume of cDNA sample before mixing briefly by vortexing. We purified the mixture using the provided Zymo‐Spin Column in a Collection Tube. After two washes, we eluted the purified cDNA using 100‐μL RNase‐free water. We then further captured biotin‐labelled cDNAs by mixing 100 μL of resuspended Dynabeads MyOne Streptavidin C1 magnetic beads to each lysate at room temperature for 60 min. We then washed the C1 beads twice with 400 μL of B&W buffer (2.5‐mM Tris‐HCl, pH 8.0, 0.5‐M NaCl, 0.25‐mM EDTA, pH 8.0) with 5 min rotation after resuspending beads, followed by another wash once with 400‐μL 10‐mM Tris‐HCl with Tween‐20 with 5 min rotation after resuspension. Random‐Priming‐Assisted Second‐Strand Synthesis The beads with cDNA captured were first resuspended in 200 μL of ExoI mix [170‐μL water, 20‐μL ExoI buffer, and 10‐μL ExoI (NEB, M0568)] and incubated at 37°C for 50 min. Following ExoI treatment, the beads were washed three times using 400 μL of B&W buffer with 0.05% Tween (2.5‐mM Tris‐HCl, pH 8.0, 0.5‐M NaCl, 0.25‐mM EDTA, pH 8.0, 0.1% Tween‐20). The bead pellet was then resuspended in 200 μL of 0.1‐M NaOH and incubated for 5 min at room temperature. Subsequently, 400 μL of B&W buffer was added and the supernatant was then discarded, and the beads were washed three times in 400 μL of B&W buffer with 0.05% Tween. The random‐priming‐assisted second‐strand synthesis was then carried out on the beads by incubating them in 200 μL of second‐strand synthesis mix (2‐μL 1‐mM dN‐SMRT oligo (Table ), 5‐μL Klenow enzyme, 20‐μL 10‐mM dNTPs, 40‐μL 5x Maxima RT buffer, 133‐μL RNase‐free water) at 37°C for 1 h. After the reaction, the supernatant was removed, and the beads were washed three times using 400 μL of B&W buffer with 0.05% Tween‐20. PCR The beads were then washed once with 400 μL of 10‐mM Tris and 0.1% Tween‐20 solution (with resuspension) and once more with 400 μL of RNase‐free water (without resuspension). The beads were then resuspended in the 220‐μL PCR mix solution (110‐μL 2x Kapa HiFi HotStart Master mix, 8.8‐μL 10‐μM Primers (Table ), and 92.4‐μL RNase‐free water). The PCR cycles were as follows: 95°C for 3 min, followed by 5 cycles at 98°C for 20 s, 65°C for 45 s, and 72°C for 3 min. The beads were then removed, and the remaining supernatant was used to run the second PCR: 95°C for 3 min, followed by 10–15 cycles at 98°C for 20s, 65°C for 20s, and 72°C for 3 min, followed by 72°C extension for 5 min. We then purified cDNA from the PCR product using 0.8x KAPA pure beads following the supplier's instructions. Sequencing The sequencing library was then prepared with custom P5 primer (Table ) using a Nextera XT DNA Library Preparation Kit by following the supplier's instructions. The samples were sequenced using Nova‐seq SP 300nt Flowcell with 80 bp for read1, and 250 bp for read 3. Spatial Sequencing Using Enhanced_DBiT_v2 As most steps are similar to those described for Enhaced_DBiT_v1, only the major modifications are highlighted here. Reverse Transcription After the tissue sections were blocked and washed, reverse transcription was performed using a 5‐mm‐wide PDMS reservoir. Specifically, the following reaction mixture was added onto each tissue section: 11.3‐μL RNase‐free water, 12.5‐μL 5× Maxima RT buffer, 25‐μL 0.5× PBS‐RI, 0.4‐μL RNase inhibitor (Enzymatic), 0.8‐μL Superase In RNase Inhibitor (Ambion), 3.1‐μL 10‐mM dNTPs, 6.2 ‐μL Maxima H Minus Reverse Transcriptase, 1.6‐μL 100‐μM template switch oligos, and 1.6‐μL 100‐μM RT primer (Table ). Tissue sections were incubated in a humidified chamber at room temperature for 30 min, followed by 42°C for 90 min for in‐tissue reverse transcription. Following reverse transcription, the reaction mixture was removed from the reservoir, and the tissue section was washed with 150‐μL NEB buffer 3.1 with 1% Enzymatics RNase Inhibitor for 5 min. The PDMS reservoir was then peeled off, and the slide was briefly dipped in a 50‐mL falcon tube with RNase‐free water and gently dried under air. Ligation Chip A was then aligned and centered on the tissue section, and channel placement was confirmed by imaging before Chip A was firmly clamped on the tissue section. At the same time, the ligation mixture was prepared as follows: 65.45‐μL RNase‐free water, 32.5‐μL 10× T4 ligase buffer, 13.24‐μL T4 DNA ligase (400 U/μL), 2.65‐μL RNase inhibitor (Enzymatics), 0.84‐μL Superase In RNase Inhibitor (Ambion), 6.5‐μL 5% Triton‐X‐100, and 139.38‐μL 1× NEB buffer 3.1 with 1% RNase inhibitor (Enzymatics). For each channel, 4‐μL ligation buffer was combined with 1‐μL 25‐μM barcode A1‐A50_v2 (Table ), and the barcode A solutions were introduced into the channels sequentially. Slides were incubated in a humidified chamber at 37°C for 30 min for in‐tissue ligation of barcode As. Following ligation, channels were flushed with NEB buffer 3.1 with 1% Enzymatics RNase Inhibitor for 5 min. Ligation of barcode B1‐B50_v2 oligonucleotides (Table ) was then performed using Chip B following the same procedure. After ligation, channels were washed for 5 min with wash buffer containing 500‐μL PBS, 5‐μL 10% Triton X‐100, and 1.25‐μL Superase In RNase Inhibitor). All remaining steps were identical to those described for Enhanced_DBiT_v1. Computational Methods DBiT‐seq Processing Read 1 was reformatted to extract UMI, Barcode A, and Barcode B sequences. The reformatted Read 1 and Read 3 were mapped against the mouse genome (GRCh39), demultiplexed, and annotated (Gencode release M31) using the ST pipeline v1.8.1. The resulting gene‐expression matrix was then filtered by removing pixels that did not cover tissues. Saturation Analysis To compare the original DBiT‐seq and Enhanced_DBiT, raw sequencing reads were subsampled from 1M to 25M, and the same data processing workflow described above was applied using ST pipeline v1.8.1 (https://github.com/jfnavarro/st_pipeline) to obtain UMI and gene counts at each sequencing depth for all tissue pixels [, ]. To compare Enhanced_DBiT, Slide‐seqv2, Slide‐tags, and Visium, Spacemake was used to process the raw data from each technology and generate aligned sequencing reads []. We then followed a previously established computational protocol (https://github.com/rajewsky‐lab/openst/tree/benchmark) to subsample aligned sequencing reads from filtered, randomly selected spatial pixels and generate UMI and gene accounts at each subsampled read depth, with a correction accounting for reads removed during preprocessing step, including spatial barcode matching and non‐coding RNA removal []. Transcriptomic Integration Preprocessed DBiT‐seq replicates were integrated together with the previously published scRNA‐seq reference datasets [, , , ] to provide higher resolution of transcriptomic signals and facilitate cell‐state annotation of the DBiT‐seq datasets. Prior to integration, to eliminate sequencing‐depth‐based batch effects, we subsampled the reference datasets to an average of 20k reads per cell and normalized all datasets using SCTransformV2 []. We then integrated the datasets (IntegrateLayers using RPCA and indicating reference datasets), followed by clustering and UMAP calculation (RunUMAP with dims = 1:13). Once completed, the integrated object was visualized via UMAP, and cell‐state annotation was carried out using expression of canonical marker genes of expected lung‐cell states. To verify cell‐state identities, integrated annotations were compared against previous annotations per reference dataset. Cell Type Deconvolution Using Conditional Autoregressive Model‐Based Deconvolution (CARD) Cell type deconvolution was performed using CARD (https://github.com/YMa‐lab/CARD) []. Briefly, we integrated and built scRNA‐seq reference following a protocol similar to that described above [, , , ]. We then performed CARD deconvolution using the annotated scRNA‐seq reference together with the per‐sample spatial count matrices to obtain predicted cell type proportions for each spatial pixel. Spatial pixels were subsequently filtered using a threshold of 0.5 for maximum predicted cell type proportion. Gene expression correlation analysis between different cell types in scRNA reference was performed using scanpy. Sox2 + Sox9 + Epithelial Cell Quantification in scRNA‐seq Cells were counted as Sox2 + Sox9 + if the normalized and log‐scaled raw counts of both Sox2 and Sox9 were larger than 0.1 for only cells from the scRNA‐seq reference datasets. Differential Expression and GO Analysis Using the integrated dataset described above, differential expression analysis was carried out between pairwise cell‐state comparisons using the normalized and log‐scaled raw counts for only cells from the scRNA‐seq reference datasets to avoid technology‐driven batch effects. Differential expression was calculated using the Wilcoxon rank‐sum test offered in the scanpy rank_gene_groups function and the log‐fold change was calculated as the log2(Group1 + pseudocount) – log2(Group2 + pseudocount) with a pseudocount of 0.01. To calculate enriched gene sets, enriched genes (p<0.05, log fold change > 1.3) were analyzed using Enrichr. Pseudotime Analysis Pseudotime analysis was then performed using the Monocle 3 package74. Since the integrated dataset contains multiple distinct trajectories, we selected only cell states related to each trajectory prior to generation of the pseudotime axis and subsequent gene‐module analysis. The dataset subsets were selected using primary marker genes and temporal dynamics: the epithelial subclusters for the Epithelial trajectory; Mesenchymal Progenitor 1, Smooth Muscle Progenitor, and Smooth Muscle for the Smooth Muscle trajectory. The previously calculated UMAP, cell‐state assignments, and spatial localization maps were used for visualizing pseudotime trajectories. To characterize gene sets involved in each differentiation trajectory, the Monocle 3 package was used to identify gene modules that co‐vary along the pseudotime axis. First, genes that vary significantly across the full pseudotime axis were identified (graph_test, neighbor_graph = 4) using the Morans I spatial autocorrelation analysis. Then, genes were grouped into modules using Louvain community clustering based on the cells that express them (find_gene_modules). The resulting gene modules were investigated by scoring the average gene expression within different cell‐state categories and individual cells using the scanpy_score_genes function. Additional genes that vary along the pseudotime axis were identified using linear regression analysis using the scipy_linregress function. Neighborhood‐Enrichment and Localization Analysis Neighborhood‐enrichment analysis was carried out using the Squidpy [] package (gr.nhood_enrichment) to measure the neighborhood enrichment of each cell state per each individual dataset.