Airway cell-cell communication in smoking through integration of spatial and single-cell transcriptomics

Abstract

Background

Cigarette smoking has a significant impact on global health. Although cessation has positive health benefits, some molecular changes to intercellular communications may persist in the lung. In this study we created a workflow to generate hypotheses by predicting altered cell-cell communication in smoker lungs using single-cell and spatial transcriptomic data.

Methods

We integrated publicly available lung single-cell transcriptomic data (n = 104) with spatial transcriptomic data (n = 2) from never-smoker and current-smoker airway samples to create spatial transcriptomic data at virtual single-cell resolution by mapping individual cells from our lung scRNA-seq dataset to spots in the spatial transcriptomic data. Cell-cell communications altered in smoking were identified using these integrated transcriptomic data.

Results

We identified pathways altered in two current-smoker samples compared with two never-smoker samples, including the overexpressed collagen pathway. We observed increased collagen pathway activity involving the ligands COL1A1 and COL1A2 in adventitial fibroblasts and decreased activity involving COL1A2 and COL6A3 in pericytes and myofibroblasts, respectively. We also identified other pathways with structural (e.g. Fibronectin-1), immune-related (e.g. MHC-II), growth factor (e.g. Pleiotrophin) and immunophilin (e.g. Cyclophilin A) roles.

Conclusions

In this study we inferred spatially proximal cell-cell communication between interacting cell types from spatial transcriptomics at virtual single-cell resolution to identify lung intercellular signaling altered in smoking. Our findings further implicate several pathways previously identified, and provide additional molecular context to inform future functional experiments and therapeutic avenues to mitigate pathogenic effects of smoking.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12931-026-03835-8.

Introduction

Cigarette smoking has a significant impact on global health, with 480,000 smoking-related deaths in the United States each year [1]. Smoking is a major driver of morbidity and mortality and a significant risk factor for several lung diseases, including chronic obstructive pulmonary disease (COPD) and idiopathic pulmonary fibrosis (IPF). Approximately 28.8 million adults in the United States smoked cigarettes in 2022 [2], with 8.8% successfully quitting. Although cessation has positive health benefits [3, 4], some molecular changes due to smoking may persist in the lung [5–12].

Cell-cell interactions are physical (e.g. cell adhesion) or biochemical (e.g. ligand from one cell binding to receptor on another) interactions between cells, with key roles in the function of tissues [13–15]. Although not studies of smoking, Blackburn et al. sought to identify lung cell-cell interactions altered in COPD using scRNA-seq data [16] and Browaeys et al. compared healthy lungs to IPF lungs using scRNA-seq to detect disease-relevant cell-cell communications [17]. Mayr and colleagues characterized pathologic niches in IPF, reconstructing these niches within a scRNA-seq dataset using spatial information to study cell-cell communication [18]. However, studies of cell-cell communication in smoking leveraging spatial information are lacking.

Spatial transcriptomic technologies provide the ability to examine cell-cell communication with increased rigor, leveraging information regarding cell locations and proximity [19], possibly at single-cell resolution (e.g. 10x Genomics Visium HD or Xenium). However, in this study we sought to leverage publicly available data from lower-cost technologies across several lung tissue studies. We developed a workflow to integrate single-cell and spatial transcriptomic data and generate hypotheses by predicting altered cell-cell communication in smoker lungs. A better understanding of intercellular signaling altered in smoker lungs may inform therapeutic development efforts to mitigate the pathogenic effects of smoking [13, 14].

Methods

Airway spatial transcriptomic data at virtual single-cell resolution were created using the cellular (Cyto) Spatial Positioning Analysis via Constrained Expression alignment (CytoSPACE) method to efficiently map individual cells from a lung scRNA-seq dataset to spots in spatial transcriptomic data from never-smoker (NS) and current-smoker (CS) airway samples (Fig. 1A) [20]. Cell-cell communications altered by smoking were identified using CellChat with these spatial transcriptomic data [21].

Fig. 1.

A The study involved use of bioinformatic methods with publicly available lung single-cell and spatial transcriptomic data. The single-cell RNA-seq data were harmonized to create a single-cell reference dataset for use with CytoSPACE. B Each cell-type identified across the ten studies shown in the UMAP plot. CytoSPACE maps individual cells from the scRNA-seq dataset onto each spot within the spatial transcriptomic data. The cell-type proportions for a never-smoker sample (C; WSA_LngSP8759312) and a current-smoker sample (D; WSA_LngSP9258464)

Lung single-cell RNA-seq data

The lung scRNA-seq reference dataset for use with CytoSPACE was created by harmonizing data from seven prior lung tissue and three airway studies (n = 104; Supplemental Table S1) [22–32]. Prior to harmonization, quality control included removal of low quality cells and multiplets. Cell-type classifications for all remaining cells were obtained using the reference-based mapping pipeline Azimuth [33]. Harmonization included dimensionality reduction with the scVI method [34, 35] and creation of Uniform Manifold Approximation and Projection (UMAP) plots.

Spatial transcriptomics

We used publicly available Visium (10x Genomics) spatial transcriptomic data from the lung tissue study by Madissoon and colleagues, where they identified 80 lung cell types and states and gleaned details regarding lung micro-environments [32]. After excluding spots with low-quality data, spatial transcriptomic data from two NS and two CS bronchi samples were available for analysis (Supplemental Table S2). We used CytoSPACE to construct spatial datasets with high gene coverage and single-cell resolution [20], by mapping cells from the lung scRNA-seq reference dataset to spots in lung Visium transcriptomic data.

Cell-cell interactions

CellChat (v2) enabled prediction of cell-cell communication altered by smoking using data from multiple spatial transcriptomics datasets [21]. We have leveraged the capabilities of the CellChat method for use with the virtual single-cell transcriptomic data we created with CytoSPACE, where instead of RNA-seq data in each spot we have scRNA-seq data for multiple mapped cells within each spot. Likewise, the group label for each spot (typically created using a clustering method) was replaced by the cell types for each of the cells mapped from our lung scRNA-seq dataset. In this usage model, we are inferring spatially proximal cell-cell communication between interacting cell types from spatially resolved transcriptomics at virtual single-cell resolution. Additional details regarding methods are available in the Supplementary Information.

Results

Lung spatial transcriptomics at single-cell resolution

The lung scRNA-seq reference dataset for use with CytoSPACE was created by harmonizing data from ten prior lung tissue and airway studies (n = 104; Supplemental Table S1) [22–32]. The final harmonized dataset included 28 cell types and 349,109 cells (Supplemental Table S3). In the UMAP plot of the single-cell data, we observed harmonization across all studies in the clustering by cell type (Fig. 1B and Supplemental Figures S1-S11). We created lung spatial transcriptomic data at virtual single-cell resolution by mapping cells from this lung scRNA-seq dataset to spots in lung spatial transcriptomic data using CytoSPACE [20]. Scatterpie plots illustrate the cell-type proportions for each NS and CS sample (Fig. 1C and D; Supplemental Figures S13-S14), where spots lacking data are either devoid of tissue or had insufficient gene expression. In these scatter plots, the epithelium was marked by a higher abundance of multiciliated cells. The overall number of cells mapped to the available spots in the spatial samples varies from 1 to 21, with 49% of the all spots having three or fewer cells (Supplemental Figure S15). Using hypergeometric tests, we observed enriched (p < 0.05) usage of cells from four studies for NS data and enriched usage from three studies for ever-smoker (ES) data, including Madissoon et al. (Supplemental Figures S16 and S17). Cell type availability varied across the single-cell transcriptomic studies (Supplemental Figures S1-S11), enabling biased usage of scRNA-seq data during cell mapping.

Cell-cell interactions

We examined cell-cell communication altered by smoking using CellChat [21]. Spatial data for goblet cells in the NS samples were not included in the CellChat analysis, as goblet cells were not available in the ES scRNA-seq dataset, and were only present in the study by Madissoon and colleagues [32]. After combining the three NS samples and the three CS samples to create two analytical datasets, and inferring the cell-cell communication network (see Supplemental Methods), we examined the differential number of interactions and interactions strength within the network between the two subjects (NS vs. CS; Supplemental Figure S18). Fibroblasts, basal and suprabasal cells had higher overall differential interactions counts and interaction strength. In a scatter plot of differential signaling showing the differential communication probability for each cell type (Fig. 2A), adventitial fibroblasts, basal cells and myofibroblasts are among the cells with higher incoming and outgoing signaling strength in CS with respect to NS, while smooth muscle cells had relatively lower overall signaling strength.

Fig. 2.

CellChat predicts cell-cell communication altered in smoker airways and highlights relevant cell types, pathways and genes. A scatter plot of the differential signaling (NS vs. CS) for each cell type, where the x-axis and y-axis represent the differential outgoing and incoming communication probability, (B) heatmap of the relative signaling strength of the top 50 signaling pathways across all cell types, where the top colored bar plot shows the total signaling strength of a cell group by summarizing all pathways shown and the right grey bar plot shows the total signaling strength of a pathway by summarizing all cell types shown, (C) chord diagram for the overexpressed signaling pathway (COLLAGEN), where red boxes highlight the COL1A1 and SDC4 interaction in AT2 cells and adventitial fibroblasts, and (D) chord diagram for the underexpressed signaling pathway (COLLAGEN). The ordering of colors in the chord legend and chord diagrams are matched to more efficiently identify a particular cell type

In the ranked heatmap of overall (incoming and outgoing) relative signaling pathway strength across all cell types in the cell-cell communication network (Fig. 2B), we observed collagen as the top signaling pathway. Pathway identification seeks to highlight the biological roles represented by the ligands and receptors identified in the predicted cell-cell communication network. We identified over-expressed genes by smoking condition, mapped this information on the cell-cell network, and ranked pathways by overall representation in the final network (see Supplemental Methods). Chord diagrams (Fig. 2C and D; Supplemental Figures S19-S24) highlight the communication network cell types, ligands and potential receptor genes for the top seven overexpressed and underexpressed signaling pathways in CS samples vs. NS samples (Table 1). Some pathways (e.g. collagen) appear in both overexpressed and underexpressed columns of Table 1, as the cell types and genes involved are different.

Table 1.

Top seven overexpressed and underexpressed signaling pathways in cell-cell communications altered in the smoker airway

Overexpressed Pathways	Number of occurrences in network	Underexpressed Pathways	Number of occurrences in network
COLLAGEN	224	LAMININ	312
LAMININ	206	COLLAGEN	220
FN1 (Fibronectin-1)	139	PTN (Pleiotrophin)	102
MHC-II (Major histocompatibility complex class II)	59	MHC-II	18
EGF (Epidermal growth factor)	45	COMPLEMENT	15
CypA (PPIA: Cyclophilin A)	40	PTPR (protein tyrosine phosphatase receptor)	10
GDF (Growth differentiation factor)	24	IL16 (Interleukin 16)	9

In Fig. 2C and D, we observed increased collagen pathway activity involving the ligands COL1A1 and COL1A2 in adventitial fibroblasts and decreased activity involving COL1A2 and COL6A3 in pericytes and myofibroblasts, respectively. Receptors such as SDC4 and CD44 are predicted in several cell types, including alveolar type 2 cells (AT2), club cells and non-classical monocytes. We also observed pathways with structural (e.g. FN1: Fibronectin-1; Figure S20), immune-related (e.g. MHC-II; Figure S21), growth factor (e.g. PTN: Pleiotrophin; Figure S20) and immunophilin (e.g. CypA: Cyclophilin A; Figure S23) roles.

Discussion

In this study we developed a workflow for analysis of spatial transcriptomics at virtual single-cell resolution to generate hypotheses by predicting altered cell-cell communication in smoker lungs. Through analyses in airway samples, we identified cell types, ligands and potential receptors for several pathways altered in smoker lungs. We observed structural, inflammatory and growth factor pathways in our results, aligned with previous findings regarding molecular effects of smoking in lung disease [36–39] and current therapeutic development efforts [40]. Collagen was the top pathway identified, with adventitial fibroblasts, myofibroblasts and pericytes being the ligand-expressing cell types. In addition to its structural roles within the extracellular matrix, collagen can induce various cellular effects by acting as a ligand [41, 42]. Collagen has key functions in lung tissue [43], with relevance to smoking related pathogenesis, in particular lung fibrosis [44] and COPD [45–47]. With total collagen previously found constant in smokers and increased in emphysema [48], and dynamic in matrix remodeling [49], both overexpression and underexpression in the collagen pathway may indicate a shift in collagen isoform and cell-type specific activity, as part of the overall remodeling process [37, 50–52]. With respect to specific cell-cell interactions in the collagen pathway, the study by Yang et al. [53] observed enhanced COL1A1-SDC4 interactions between fibroblasts and epithelial cells in lung adenocarcinoma (LUAD). With smoking a known risk factor for development of lung cancer, our observed increase in activity of this interaction for current smoker samples (Fig. 2C; red boxes) may highlight an early suggestive signature relevant to lung cancer.

We observed overexpression of the FN1 (fibronectin-1) pathway in CS, with myofibroblasts expressing the FN1 ligand. The fibronectin-1 pathway represents another matrix remodeling finding previously observed in smoking [54] with implications in pulmonary fibrosis [55, 56], epithelial mesenchymal transition (EMT) and COPD [57, 58].

The cyclophillin (CypA) pathway was overexpressed in CS samples, with the ligand PPIA (peptidylprolyl isomerase A; gene encoding CypA) expressed in myofibroblasts, suprabasal cells and basal cells. CypA is an imunophilin with molecular chaperone functions. Although typically localized in the cytoplasm, it has intercellular communication roles when secreted in response to inflammatory stimuli [59], infection [60] or oxidative stress [61]. A proteomic study found CypA overexpressed in smokers and further enhanced in smokers with COPD [62], and another study found serum levels of CypA higher in COPD with levels further elevated during acute exacerbations [63], though the particular role of CypA in lung disease is not fully understood. The predicted receptor in the CypA pathway was BSG (basigin; CD147), a gene expressed in several cells including fibroblasts and multiciliated cells. Lung expression of CD147 was found higher in smokers with COPD and may enhance mucus secretion during exposure to cigarette smoke [64]. CD147 may also have a role in smoking initiated EMT through oxidative stress signaling [65].

Limitations of this study include the small spatial transcriptomic cohort and confounding of sex with smoking condition. Our single-cell dataset also included ever-smokers, who may be either former or current smokers. The cells were obtained from lung tissue in addition to the airway. Although this provided an extensive and diverse set of cells necessary for integration with the spatial transcriptomic data, an extensive and diverse dataset from only the airway would have been preferred and used if sufficient data had been available. The integrated spatial and single-cell transcriptomic data are from different subjects, increasing the ability to detect cell-cell communications altered in smoking by effectively expanding the study population. For both single-cell and spatial datasets, data regarding comorbidities or medication use of the study subjects was rarely reported and not included in the analyses.

Our findings further implicate several pathways previously identified, generating new hypotheses and additional molecular context to inform future functional validation experiments and inform therapeutic avenues to mitigate pathogenic effects of smoking.

Abbreviations

CS

Current-smoker

NS

Never-smoker

ES

Ever-smoker

COPD

Chronic Obstructive Pulmonary Disease

IPF

Idiopathic Pulmonary Fibrosis

scRNA-seq

Single-cell RNA-sequencing

CytoSPACE

Cellular (Cyto) Spatial Positioning Analysis via Constrained Expression alignment

UMAP

Uniform Manifold Approximation and Projection

EMT

Epithelial Mesenchymal Transition

MHC

Major Histocompatibility Complex

EGF

Epidermal Growth Factor

GDF

Growth Differentiation Factor

PTPR

Protein Tyrosine Phosphatase Receptor

FN1

Fibronectin-1

CypA

Cyclophilin A

PTN

Pleiotrophin

MHC-II

Major histocompatibility complex class II

IL16

Interleukin 16

AT2

Alveolar type 2 cells

LUAD

Lung Adenocarcinoma

Authors’ contributions

JDM: conceptualization, methodology, formal analysis, interpretation of data, manuscript preparation and approval of the final version ZWE: methodology, interpretation of data, manuscript preparation and approval of the final version JHY: methodology, interpretation of data, manuscript preparation and approval of the final version CPH: methodology, interpretation of data, manuscript preparation and approval of the final version.

Funding

This study was supported by an Alpha-1 Foundation Research Grant, a TOPMed Fellowship and NIH grants K25HL136846, R01HL166231, P01HL114501.

Data availability

Data were obtained from publicly available repositories (see Supplemental Tables S1 and S2): Single-cell RNA-seq data: Gene Expression Omnibus: GEO: GSE143868 GEO: GSE178360 GEO: GSE136831 GEO: GSE173896 GEO: GSE171541 GEO: GSE135893 GEO: GSE227136 GEO: GSE168191 SRA: SRP318548 Zenodo: https://zenodo.org/records/8393742https://5locationslung.cellgeni.sanger.ac.uk/ Spatial Transcriptomics data: EMBL-EBI BioStudies/ArrayExpress: E-MTAB-11640.

Declarations

Ethics approval and consent to participate

Not applicable.

Data used in this study were obtained from publicly available repositories.

Competing interests

No competing interests were disclosed for Drs. Morrow and El-Husseini Dr. Hersh reports research grants from Bayer, Boehringer-Ingelheim and Vertex, consulting fees from Chiesi, Sanofi, Takeda, Apogee Therapeutics, AstraZeneca, Genentech, Ono Pharma and Verona Pharma, and lecture honorarium from The Korean Academy of Tuberculosis and Respiratory Diseases, unrelated to this study. Dr. Yun reports research grants from Bayer, consulting fees from Bridge Biotherapeutics, and travel reimbursement from The Korean Academy of Tuberculosis and Respiratory Diseases in the past 3 years, unrelated to the current work.