Experimental design and cell sorting of BAL

In the present study, we utilized samples from a previous animal experiment described in [44]. The experiment involved four groups of inbred Babraham pigs: one group (n = 5) was infected intranasally with pH1N1 influenza virus, another group (n = 5) received intranasal immunization with a recombinant adenoviral vector expressing hemagglutinin (HA) and nucleoprotein (NP) Ad-HA/NP, and the third group (n = 5) was immunized intranasally with Ad-HA/NP and a recombinant adenoviral vector encoding porcine IL-1β (Ad-HA/NP+Ad-IL-1β). Controls (n = 3) received PBS intranasally (Fig 1A). The animals were culled 21 days post immunization or infection and tissues harvested for immunological and virological analysis. In the previous study, immune responses of these groups were extensively analyzed and showed that IL-1β acts as a potent mucosal adjuvant, increasing antibody responses, proportions of plasma and activated B cells, cytokine production of T cells and NP tetramer-specific CD8 T cells in the Ad-HA/NP+Ad-IL-1β group [44].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Experimental design and UMAP clustering of BAL immune cells.

(A) Design of pig immunization/infection experiment. Total number of pigs per treatment is given. For scRNA-seq, 3 pigs were randomly selected for treatments 1, 2 and 3. (B) Graphical overview of applied sorting strategy. (C) Uniform manifold approximation and projection (UMAP) of all cells that passed quality control from pig BAL scRNA-seq analysis, colored by cell-type and labelled with cluster numbers.

https://doi.org/10.1371/journal.ppat.1011910.g001

Given the well-defined nature of these samples we chose to perform scRNA-seq analysis on leukocyte populations sorted from the BAL. We were particularly interested in the BAL as T cells in this niche are inaccessible to intravenous CD3 antibody and have a phenotype characteristic of TRM in other species [45,46]. As the cellular composition of BAL samples is strongly dominated by alveolar macrophages, we applied cell sorting to isolate macrophages, CD4 T, CD8 T and B cells from 3 animals per group and subjected equal proportions of these cells (12,000 cells each) to scRNA-seq (Figs 1B and S1A). Cell partitioning and library preparation were performed separately for each sorted and re-pooled BAL sample from 12 pigs (3 animals per group).

Pig BAL scRNA-seq identifies typical immune cell populations

After combining processed single-cell transcriptomes from all 11 samples that passed quality control checks, graph-based clustering identified 21 distinct cell clusters. Of these, three clusters were excluded for being of low quality or under suspicion of containing doublets. Three clusters were additionally split into subclusters to separate CD4 from CD8 T cells or split T cells from non-T cells, creating a final dataset with 22 cell clusters for annotation. To annotate each cluster with conventional cell-type names, marker genes defining each cluster were extracted and each cluster was screened for expression of a list of typical defining marker genes (S1 Table, S1B Fig). From this information, each cluster was manually annotated with an established cell-type name (Fig 1C).

Most clusters were found to be comprised of well-defined cell-types, such as macrophages, conventional dendritic cells (cDCs), monocytes, plasma cells, B cells and T cells, though particular subclusters presented more niche and complex phenotypes. These included 18a and 18b, which were annotated as ‘mitotic T cells’ based on their strong expression of E2F1 and E2F2, and very small clusters of CD4 and CD8 T cells (20a and 20b, with 18 and 8 cells, respectively) that differed from the major CD4 and CD8 T cell clusters (7 and 6, respectively) by expressing higher levels of genes associated with naïve and/or central memory T cell states (e.g. BACH2, SELL, LEF1) and lower levels of genes encoding cytoskeletal and adhesion molecules (e.g. LGALS1, S100A4, VIM) previously found to characterize lung-localized human T cells [17].

Cluster 14 expressed T cell-related genes in an atypical pattern: strong expression of CD28 and CD5, with low expression of CD4, CD8A and CD8B, bifurcated expression of CD3D and CD3E, and very high expression of EOMES. Pending further investigation, cluster 14 has been annotated as ‘T cell like’.

Cluster 21 localises near identified T cell clusters in both UMAP (Fig 1C) and TSNE (S1D Fig) space. However, cluster 21 does not express any of the conventional αβ T cell markers: CD28, CD5, CD3D, CD3E, CD4, CD8A, or CD8B. Expressed T cell-related genes include GATA3, IL1RL1, IL7R and KIT, suggesting that these cells may be innate lymphoid cells, but without corroborating evidence, we have currently classified these cells as ‘unknown’.

Cluster 19 showed strong expression of ACTG1, GPR37, NAPSA and SFTPA1, typically expressed by mucus producing epithelial cells. Cluster 9 and its subclusters were enriched for genes typically associated with NK cells such as KLRK1, KLRB1, KIT and NCR1. As our cell sorting strategy was not intended to capture NK cells, NK cell-like T cells or epithelial cells, clusters 9 and 19 risked presenting a biased representation of these larger cell populations. We therefore excluded clusters 9 and 19, leaving 18 clusters for further analysis.

Confirmation of scRNA-seq derived cell types by flow cytometry

We aimed to confirm cell types identified by scRNA-seq by flow cytometry of the same cryopreserved samples (Fig 2). By using an antibody panel designed for phenotyping of myeloid cells (S2 Table) and gating on FSC-A^highSSC-A^high cells (S2A Fig), we identified a prominent population of CD172a^highCD163⁺CD14^-/dimCD16⁺ macrophages (Fig 2A), which represent scRNA-seq clusters 1, 4 and 8. Within CD163^- cells, we identified a CD14⁺CD16⁺ subset, representing putative monocytes identified in scRNA-seq cluster 15. CD14^- cells were further analysed for CD172a/CADM1 co-expression which allowed the identification of CD172^dimCADM1^high subset and a CD172a^highCADM1⁺ subset, representing cDC1 and cDC2 cells, respectively [47], and identified in cluster 13 by scRNA-seq.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Confirmation of scRNA-seq cluster annotation by flow cytometry.

Annotation of clusters in scRNA-seq was confirmed by flow cytometry using antibody panels addressing key marker combinations. Colors used in UMAP clustering (left) match with colored gates and histograms in flow cytometry to indicate the same cell type (right). (A) Different subsets of myeloid cells. (B) B cell and plasma cell subsets. (C) CD4 T cell subsets. (D) CD8 T cell subsets.

https://doi.org/10.1371/journal.ppat.1011910.g002

The presence of B cells and plasma cells was also confirmed by flow cytometry (Fig 2B) using a specifically designed B cell panel (S2 Table). CD79α⁺ cells contained a prominent subset of IRF4^highBlimp-1⁺ plasma cells, which were also Pax-5^- and thus resembled clusters 2 and 3 of the scRNA-seq data. CD79α⁺IRF4^dimBlimp-1^- cells had variable levels of Pax-5 but overall likely represent B cell clusters 12 and 16 from the scRNA-seq data.

CD4 and CD8 T cells were investigated by a dedicated T cell panel (S2 Table). CD3⁺CD4⁺ T cells were separated based on Foxp3 expression (Fig 2C). Both CD25^highFoxp3⁺ putative regulatory T cells (Tregs) and CD25^dimFoxp3^- non-Treg CD4 T cells were largely CD8α^dimCCR7^-Bcl-6^-Eomes^- but expressed low levels of T-bet. With these phenotypes, they resembled scRNA-seq clusters 10 and 7, respectively. We also identified a subset of Ki-67-expressing CD4 T cells, which may represent cluster 18b annotated as “mitotic” CD4 T cells. CD8 T cells were identified by a CD3⁺CD8αβ⁺ phenotype (Fig 2D). After exclusion of γδ T cells, the majority of remaining CD8 T cells were CD161^-NKp46^-perforin^-, resembling CD8 T cell cluster 6 identified by scRNA-seq. Within this phenotype, some cells expressed Ki-67, which may represent “mitotic” CD8 T cells found in cluster 18a. Our sorting strategy for scRNA-seq excluded most NK and γδ T cells present in BAL. However, a flow cytometry panel designed for the characterization of nonconventional T cells (S2 Table) indicated the presence of a substantial proportion of γδ T cells, which were largely CD8α⁺CD16^-NKp46^- (S2C Fig). Within the CD3^- population, CD161/CD8α defined putative NK cell subsets, which exhibited substantial phenotypic heterogeneity in terms of CD16, NKp46 and perforin expression. This phenotypic analysis confirmed cell types identified by scRNA-seq and indicated further heterogeneity within NK cells and non-conventional T cells.

Comparison with PBMCs delineates similar and distinct cell-types

To validate our cell-type annotations and to further examine features of BAL tissue resident populations, we compared our combined treated and control pig BAL scRNA-seq dataset with a publicly available pig PBMC scRNA-seq dataset, published by Herrera-Uribe et al [23]. Both datasets were restricted to shared genes, batch corrected and merged using a mutual nearest neighbors approach (Fig 3). When plotted as a UMAP, several cell types such as monocytes, cDCs, plasma cells (antibody secreting cells, ASCs in the PBMC dataset), B cells and some T cells formed visually integrated and/or adjacent populations between BAL and PBMC (Fig 3A and 3B), implying similar patterns of overall gene expression. Cell-types without near neighbors between the datasets, i.e. BAL macrophages, the BAL ‘T cell like’ cluster 14 and PBMC γδ T cells, formed visually distant and distinct populations (Fig 3A and 3B). Looking more closely at T cell populations (Fig 3C) revealed two that partially overlap: BAL conventional T cells with PBMC αβ T cells (both CD4⁺ and CD8⁺) and BAL mitotic T cells with a separate population of PBMC αβ T cells.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Comparison of pig BAL and PBMC scRNA-seq datasets.

(A-C) UMAP of BAL and PBMC cells. (A) All cells from BAL and PBMC scRNA-seq datasets colored by cell type. (B) All cells from BAL and PBMC scRNA-seq datasets, colored by tissue. (C) T cells from BAL and PBMC scRNA-seq datasets, colored by cell type. (D) Mapping of BAL cells (left) to PBMC clusters (right) via scMAP, colored by cell type. Each node represents one cluster and is labelled with the cell type and cluster number. Unmapped PBMC nodes and links constituting less than 15% of the cells in a single BAL cluster were omitted for legibility. (E-F) Heatmaps of BAL and PBMC cluster gene expression. Each column represents one cluster, labelled with the tissue of origin, cell type and cluster number. Color scale is generated from mean logcounts. (E) Heatmap of B cell clusters and (F) Heatmap of T cell clusters from BAL and PBMC scRNA-seq datasets.

https://doi.org/10.1371/journal.ppat.1011910.g003

To further delineate how individual cell types compared between BAL and PBMC, the top 20 marker genes for each cluster in the PBMC dataset were used to generate an index for projecting the annotated BAL cells onto the PBMC clusters. Fig 3D provides a visual representation of this mapping, identifying the proportion of cells in each BAL cluster that showed gene expression similarity to each PBMC cluster. To examine which genes were driving or inhibiting mapping between clusters, all clusters and all genes used for mapping were plotted as a heatmap (S3A Fig). Subsets of B cells and T cells were separately plotted (Fig 3E and 3F) using gene clusters identified in S3A Fig alongside defining genes from S1 Table; additional subset heatmaps are also provided in S3B–S3G Fig. Broadly, cell-types in the BAL mapped to the same or closest equivalent cell types in the PBMC dataset. BAL macrophages and monocytes mapped to PBMC monocytes (there are no macrophages in PBMC), BAL cDCs to PBMC cDCs, BAL plasma cells to PBMC ASCs and BAL T cells to PBMC T cells. In some cases, mapping at the subset level was ambivalent. For example, the BAL macrophage clusters mapped to PBMC monocyte clusters 20 and 27 (Fig 3D), suggesting minimal differences in the patterns of top marker gene expression between these PBMC monocyte clusters. At the same time, this example also demonstrated preferential mapping such that BAL macrophages mapped to PBMC monocyte clusters 20 and 27 but not to monocyte clusters 13, 19 or 25. Such preferential mapping was apparent across most cell types analyzed, with only a subset of PBMC clusters identifying as close matches for BAL cells. This implies that the heterogeneity in gene expression that drove the clustering in the PBMC dataset also embodied some of the divergence between BAL and PBMC cell types. Of the 33 PBMC clusters, BAL cells only substantially mapped to 12, suggesting that many of the PBMC cell types/states are distinct from those found in the BAL or relate to cell-types not included in our sorting strategy.

There are also exceptions to expected mappings. In the BAL, both clusters 12 and 16 were annotated as B cells, but only cluster 12 mapped to PBMC B cells, while cluster 16 failed to map (Fig 3D). Fig 3E illustrates the gene expression of these B cell clusters, with BAL cluster 16 uniquely showing reduced expression of numerous ribosomal genes (RPL*, RPS*). Likewise, BAL T cells mapped to PBMC T cells, but there was heterogeneity in mapping of T cell subsets. BAL clusters 18 and 20 were previously subclustered into separate CD4⁺ and CD8⁺ populations. However, BAL mitotic T cells (cluster 18) mapped primarily to PBMC CD8⁺ αβ T cell cluster 14, and BAL cluster 20 mapped primarily to PBMC CD4⁺ αβ T cell cluster 0 (Fig 3D). For the largest conventional BAL T cell clusters (6 and 7) the primary mapping was to PBMC CD2^- γδ T cell cluster 21 and, to a lesser extent, PBMC CD4⁺ αβ T cell cluster 4 (Fig 3D and 3F). Furthermore, the ‘unknown’ BAL cluster 21 also mapped strongly to PBMC CD2^- γδ T cell cluster 21. Conceptually it is unsurprising that a TRM-dominant tissue like the BAL would exhibit distinct gene expression from the circulating T cells that inhabit the peripheral blood. These examples also demonstrate how gene expression representing cellular functions such as cell cycle regulators or cytokine regulation (Figs 3F and S3) can outweigh the relatively stable cell surface markers commonly used to distinguish different cell types when performing mapping between datasets using the whole transcriptome.

Frequencies of Treg cells

We next tested the scRNA-seq data for differential abundance of cells within the annotated cell types between treatment conditions (S3 Table). This analysis identified that pigs immunized with Ad-HA/NP+Ad-IL-1β have greatly reduced proportions of cells in cluster 10, annotated as Tregs, within the BAL. Among CD4⁺ T cells, Tregs make up an average of 0.02% in Ad-HA/NP+Ad-IL-1β immunized pigs versus an average of 0.09% in the other conditions (adjusted p = 7.67e^-8, Fig 4A). This apparent loss of Tregs was not mirrored by any of the other CD4 T cell clusters and was consistent between samples within each condition (S4A Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Abundance and test of suppressive function of Tregs.

(A) Tregs identified by scRNA-seq (cluster 10) as a proportion of all CD4 T cells for each treatment (data from 11 pigs: 5 males and 6 females). (B) Tregs identified by flow cytometry as percentage within CD3⁺CD4⁺ T cells for each treatment in cell preparations from BAL and tracheobronchial lymph nodes (TBLN) (** denotes p≤0.01). (C) Suppression of IFNγ and TNFα production in pH1N1 re-stimulated BAL CD4 (top panel) and CD8 T cells (bottom panel) from Ad-HA/NP+Ad-IL-1β immunized pigs, in the presence or absence of CD4⁺CD25^neg or CD4⁺CD25^high sorted BAL cells from Ad-HA/NP immunized pigs. Numbers in quadrants show percentages of cells with the respective phenotype.

https://doi.org/10.1371/journal.ppat.1011910.g004

Based on this significant reduction of Tregs in the BAL of IL-1β treated animals seen by scRNA-seq, we further analysed Treg frequency by flow cytometry in BAL and tracheobronchial lymph nodes from the same pigs, plus two additional pigs from immunized and infected groups. Interestingly, there was a significant reduction of Tregs, defined as CD3⁺CD4⁺CD25^highFOXP3⁺ in the tracheobronchial lymph nodes of the Ad-HA/NP+Ad-IL-1β immunized group compared to the other three groups (Figs 4B and S5). However, the reduction in Tregs that was observed in BAL scRNA-seq data was not replicated by the BAL flow cytometry analysis. This is likely because the Treg populations isolated by scRNA-seq and by flow are, whilst overlapping, not the same. In both scRNA-seq and flow cytometry, Tregs are defined as CD3⁺CD4⁺FOXP3⁺ cells, with the flow cytometry analysis also including high protein expression levels of CD25. However, in the scRNA-seq analysis, cells are grouped into clusters based on their top 5000 highly variable genes (HVGs) prior to being identified according to their expression of defining genes. Hence scRNA-seq Tregs (cluster 10, 799 cells), represent a population of T cells that are phenotypically distinct from the other clusters based on 5000 HVGs, and are characterised by higher expression of CD3, CD4, FOXP3 and IL2RA (encoding CD25), among other genes. Cells expressing FOXP3 and IL2RA transcripts also exist in other clusters, including CD4 T cell clusters 7 and 18b (S4B Fig). Assuming that transcript levels translate into protein expression, the flow cytometry analysis would include some of these FOXP3 and IL2RA expressing cells alongside cluster 10 as a single Treg population. Hence scRNA-seq Tregs and flow cytometry Tregs represent two overlapping but not identical populations. These data therefore imply that the BAL Treg population that is diminished during IL-1β immunization is a phenotypically distinct subset of BAL Treg and that others are not diminished.

We previously showed that IL-1β is a potent mucosal adjuvant and that intranasal immunization with Ad-HA/NP+Ad-IL-1β increased T cell cytokine production in Babraham pigs [44]. Based on our observations of altered Treg abundance in this condition, we next investigated whether Tregs modulate cytokine production in the BAL. To this end, sorted CD4⁺CD25^high Tregs and CD4⁺CD25^neg control cells from the BAL of Ad-HA/NP immunized Babraham pigs were co-cultured with BAL cells from Ad-HA/NP+Ad-IL-1β immunized animals in the presence of pH1N1. The expression of IFNγ and TNFα was then assayed by intracellular staining. Incubation of BAL cells derived from Ad-HA/NP+Ad-IL-1β immunized pigs with Tregs abolished IFNγ and TNFα production, while a minor effect was detected when CD4⁺CD25^neg control cells were introduced (Fig 4C). Together with scRNA-seq and flow cytometry quantification of Treg populations, these data suggest that IL-1β reduces the proportion of functionally active Tregs, contributing to increased cytokine production, in agreement with previous studies demonstrating the suppressive role of porcine Tregs [48]. Indeed, the IL-1β treated animals showed increased lung pathology following H3N2 infection [44]. Our data suggests that this is most likely due to the reduced Treg population, and associated enhanced cytokine production, in this group.

Differential expression analysis reveals condition- and cell type-specific phenotypes

To investigate differential gene expression between our experimental conditions we used a NEgative Binomial mixed model Using a Large-sample Approximation (NEBULA) [49] run discretely on all cells, B cells (clusters 12 and 16), CD4 T cells (clusters 7, 10, 18b and 20a), CD8 T cells (clusters 6, 18a and 20b) and macrophages (clusters 1, 4 and 8) (S6 Fig, S6 Table). To distinguish if differentially expressed (DE) genes were distinct or shared between cell types for each experimental condition, significantly DE genes with false discovery rate less than 0.01 were visualized as Venn diagrams (Fig 5A–5C). DE genes were also tested for Gene Ontology (GO) enrichment to approximate biological function (Fig 5F and 5G, S4 Table). For all experimental conditions, most DE genes were distinct to specific cell types, with only a minority being shared (Fig 5A–5C). For Ad-HA/NP immunization versus PBS (Fig 5A), macrophages, CD8 T cells and B cells showed significant DE genes, none of which were shared (Fig 5A). GO enrichment for Ad-HA/NP immunization versus PBS DE genes returned fewer than 5 DE genes per GO term and is therefore not shown. Ad-HA/NP+Ad-IL-1β immunization versus PBS identified considerably more DE genes and significant GO terms, with 5 DE genes shared between cell types (Fig 5B and 5D). DE genes in CD8 T cell populations were enriched for lymphocyte activation and apoptosis-related GO terms, while enrichment testing in macrophages returned only broad GO terms, and analyses in B cells and CD4 T cells returned fewer than 5 DE genes per GO term (Fig 5G). The enriched GO terms in CD8 T cells implied that immunization with Ad-HA/NP+Ad-IL-1β elicited a strong immunogenic response, potentially including activation-induced death given the presence of apoptotic terms. Comparing pH1N1 infection with PBS, in addition to numerous distinctly expressed DE genes, NEBULA identified 10 DE genes shared by cell-types Fig 5C and 5E). The Interferon Alpha Inducible Protein 6 (IFI6) gene was the only differentially expressed gene to be upregulated in all four cell types. IFI6 is one of many genes induced by interferons and plays a critical role in regulating apoptosis and early control of virus replication [50–54]. GO-term enrichment of DE genes in macrophages presented a trend towards anti-viral response mechanisms, whereas B cells, CD4 T cells and CD8 T cells returned fewer than 5 DE genes per GO term (Fig 5F).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5.

Differential gene expression with immune challenge (A-C) Venn diagrams presenting the number of overlapping and distinct differentially expressed genes for macrophages (clusters 1, 4 and 8), CD4⁺ T cells (clusters 7, 10, 18b and 20a), CD8⁺ T cells (clusters 6, 18a and 20b) and B cells (clusters 12 and 16) under each experimental condition. Numerals indicate the number of differentially expressed genes, and the number of arrows corresponds to the number of cell types in which the genes are differentially expressed. Red up arrows indicate gene upregulation, blue down arrows indicate gene downregulation. All genes shown have an adjusted p value (Benjamini-Hochberg) of less than 0.01 when comparing gene expression in each condition versus PBS. (D-E) Tables of all differentially expressed genes shared between cell types for each condition. A full table including both shared and unshared genes is available in S4 Table. (F-G) Bubbleplots presenting GO term enrichment for macrophages, B cells, CD4 T cells and CD8 T cells under each experimental condition, based on all genes with an adjusted p value of less than 0.01. GO terms annotated with fewer than five differentially expressed genes have been omitted. (A) Ad-HA/NP treated pigs versus PBS, (B, D, G) Ad-HA/NP+IL-1β treated pigs versus PBS and (C, E, F) pH1N1 infected pigs versus PBS.

https://doi.org/10.1371/journal.ppat.1011910.g005

Overall, these comparisons suggest that both Ad-HA/NP+Ad-IL-1β immunization and pH1N1 infection invoke a greater immune response than Ad-HA/NP, resulting in greater differential gene expression. These trends and widespread upregulation of IFI6 across BAL cells in pH1N1-infected animals were additionally confirmed using a pseudobulk differential expression analysis.

IFI6 marks lasting impacts of pH1N1 infection on BAL leukocytes

The differential expression analyses identified IFI6 as strongly upregulated across multiple cell types in pH1N1 infected pigs. Further investigation revealed that IFI6 was upregulated in most scRNA-seq cell clusters in pH1N1 infected pigs (Fig 6A). To assess if IFI6 expression was associated with the expression of other genes in different cell types, we used the mixed model of NEBULA to perform a separate differential expression analysis with IFI6 expression itself as the dependent variable, asking which genes were associated with IFI6 expression. Genes associated with IFI6 expression in each cell type were compared in Figs 6B and S7A–S7D, identifying numerous IFI6 related genes, predominantly in macrophages, CD4 T cells and B cells. GO enrichment (Fig 6C) of these genes in CD4 T cells and B cells identified generic immune response terms. For macrophages, IFI6-associated genes were enriched for metabolism-related GO terms.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. IFI6 abundance, associated genes and functional consequences of IFI6 expression.

(A) IFI6 expression (normalized log counts per million) from scRNA-seq data across clusters in each experimental condition (data from 11 pigs: 5 males and 6 females). (B) Venn diagram of genes whose expression is associated with IFI6 expression in macrophages (clusters 1, 4 and 8), CD4 T cells (clusters 7, 10, 18b and 20a), CD8 T cells (clusters 6, 18a and 20b) and B cells (clusters 12 and 16). Numerals indicate the number of differentially expressed genes, and the number of arrows corresponds to the number of cell types in which the genes are differentially expressed. Red up arrows indicate gene upregulation, blue down arrows indicate gene downregulation. All genes shown have an adjusted p value (Benjamini-Hochberg) of less than 0.05 and a logFC that is either more than 0.3 or less than –0.3. Accompanying table displays the genes names for all significant genes shared between cell types. A full table including both shared and unshared genes is available in S4 Table. (C) Bubbleplot presenting GO term enrichment for genes differentially expressed as IFI6 is upregulated for macrophages, CD4 T cells and B cells. CD8 T cells have been omitted due to a lack of significant genes. Based on all genes with an adjusted p value of less than 0.05. GO terms annotated with fewer than five differentially expressed genes have been omitted. (D) GFP counts of BAL cells following infection with VSVΔG-GFP 7 hrs post-infection. Comparison of BAL cells from pH1N1 and control pigs. Asterisk indicates significant difference (p<0.05).

https://doi.org/10.1371/journal.ppat.1011910.g006

We also compared the antiviral state of porcine BAL cells between pH1N1-infected and control pigs through a functional interferon assay. To this end, we infected BAL cells with recombinant virus-like particles of vesicular stomatitis virus in which glycoprotein had been replaced with GFP (VSVΔG-GFP) [55, 56]. The kinetics of GFP expression from VSVΔG-GFP was used as a measure of virus replication. The mean GFP expression observed in BAL cells from pH1N1 infected pigs was significantly lower than that in BAL cells from the control group, confirming the antiviral state of BAL cells 21 days post infection with pH1N1 (Fig 6D and S7E–S7F). This analysis indicates that although it would be expected that cells in BAL always represent a population that had migrated to this site in response to an antigenic stimulus in the respiratory tract, local factors modify gene expression of BAL cells, as demonstrated by the ubiquitous expression of IFI6 following influenza infection.

Ethics statement

Animal experiments were approved by the ethical review processes of The Pirbright Institute and the Animal and Plant Health Agency (APHA) according to the UK Animals (Scientific Procedures) Act 1986 under project license P47CE0FF2.

Infection and immunization animal studies

Samples obtained from a previous experiment conducted on inbred Babraham pigs were utilized [44]. Briefly three groups of 5 pigs each, were treated as follows: 1) infected intranasally with 8x10⁶ PFU of A/swine/England/1353/2009 (pH1N1) using a mucosal atomization device (MAD); 2) immunized intranasally with 1x10⁹ particles of recombinant adenoviral vectors expressing NP from H1N1/PR8/34 virus and HA from pH1N1/Texas/05/2009 (Ad-HA/NP) using a MAD and 3) immunized intranasally with a mix of 1x10⁹ particles of Ad-HA/NP and 1x10⁹ particles recombinant adenoviral vector encoding porcine IL-1β (Ad-HA/NP+Ad-IL-1β) using a MAD. Control pigs (n = 3) received PBS intranasally. All vaccines and PBS were delivered at 2ml per nostril using a MAD, which reaches both the upper and lower respiratory tract as previously shown using scintigraphy [66]. The animals were culled three weeks later, and bronchoalveolar lavage (BAL), tracheobronchial lymph nodes and blood collected. Virus shedding and immune responses in respiratory tissues and peripheral blood were described in [44].

Sorting of BAL cell subsets for scRNA-seq

BAL cells had been isolated as described in [44]. After two washes in PBS, cells were resuspended in freezing medium (90% foetal calf serum, 10% DMSO) and subjected to gradual temperature decline in a Mr. Frosty Freezing Container (ThermoFisher) at -80°C. Subsequently, cryovials were transferred to liquid nitrogen for long-term storage. Cryopreserved BAL cells were thawed, washed and resuspended in PBS. Cells were incubated with antibodies against CD4 (clone 74-12-4, PerCP-Cy5.5-conjugated, BD Biosciences), CD8β (clone PPT23, FITC-conjugated, Bio-Rad), CD3 (clone PPT3, APC-conjugated, Southern Biotech), CD16 (clone G7, AF647-conjugated, Bio-Rad), CD172a (clone 74-22-15, PE-conjugated, Bio-Rad) and Fixable Viability Dye eFluor780 (ThermoFisher). After two washes in PBS, cells were sorted on a FACSAria UIII (BD Biosciences), equipped with 405, 488, 561 and 640nm lasers and BD FACSDiva 8 software. After gating on FSC-A^dimSSC-A^dim cells and exclusion of doublets and dead cells, CD4⁺, CD8β⁺ and CD3^-CD16^- (to enrich for B cells) cells were sorted (12,000 cells per population). In addition, after gating on FSC-A^brightSSC-A^bright cells, doublet and dead cell exclusion and gating on CD16⁺CD172a⁺ cells, 12,000 macrophages were sorted. Representative phenotyping data is shown in S1A Fig. Cells from each fraction were pooled and subjected to partitioning and barcoding on a 10x Genomics Chromium iX Controller.

scRNA-seq library preparation and sequencing

Single cells were processed with the Chromium iX controller, using the Next GEM chip G, and single cell 3’ kit v3.1 library preparation kit (10X Genomics, Pleasanton, CA). An estimated 10,000 cells per BAL sample (total of 12 samples) were targeted for recovery during partitioning. Samples were randomly grouped and run with two libraries per sequencing run on three NextSeq High Output 150 cycle reagent kits (Illumina, San Diego, CA), with 1% PhiX, targeting approximately 200 million reads per sample. Bcl files were demultiplexed using ‘cellranger mkfastq’ and aligned/counted using ‘cellranger count’ (cellranger-7.0.0) using default parameters and Sus scrofa genome (genome assembly 11.1, Ensembl release 107).

scRNA-seq QC and pre-processing

Data analysis was run using R (version 4.1.1) and RStudio (build 382). Empty droplets, barcode-swapped droplets and ambient RNA were removed from the unfiltered cellranger output using DropletUtils (1.18.1) [67,68]. Genes with no counts and cells with outlying mitochondrial/nuclear gene log-ratios (> 3 median absolute deviations) were removed. One of twelve samples was excluded as a strong outlier due to low sequencing saturation and excessive detected cell number. All remaining samples were normalized by pooled factors [69] and combined to form the analysis dataset. Batch effects were corrected between sequencing runs using fastmnn (batchelor 1.14.0) implementation of mutual nearest neighbors batch correction [70].

scRNA-seq clustering and annotation

Shared-nearest-neighbors graph-based clustering was performed on batch-corrected data using clusterCells (bluster 1.8.0) with the top 5000 highly variable genes (HVGs), specifically using Jaccard index to weight edges and the Louvain method for community detection, with a k value of 20. This generated 21 initial clusters. Cluster 5 exhibited particularly low counts, limited numbers of genes detected and lack of any positive defining features and was therefore defined as poor-quality and excluded from further analysis. Co-expression based doublet scoring by cxds (scds 1.14.0) indicated that cluster 17 was likely comprised of a large proportion of doublets. Cluster 11 also had a substantial proportion of doublets, and manual inspection revealed co-expressed macrophage and B cell marker genes, suggesting that this cluster was also likely doublets. These clusters were therefore excluded from further analysis. Clusters 18 and 20 were found to express a dichotomy of CD4 and CD8 expression and so were split into CD4⁺ and CD8⁺ subclusters. This was achieved by identifying HVGs within the cluster of interest, finding those that correlated best with CD4 and CD8B expression using a zero-inflated Kendall’s Tau correlation metric implemented in scHOT (v1.10.0), and then re-running batch correction and subclustering (Jaccard index to weight edges and the fast greedy modularity optimization algorithm for community detection) on cells in the cluster of interest. Varied expression of genes including CD3E, CD8A, CD8B, RORC and KLRB1 within cluster 9 also prompted sub-clustering. This was done in an unsupervised manner by identifying the top 2000 HVGs in cluster 9 cells and then re-running batch correction and clustering (Jaccard index to weight edges and the fast greedy modularity optimization algorithm for community detection). This split the population into three subclusters.

Clusters were manually annotated with conventional names (cell-types) by identifying the positive and negative expression of typical defining markers, as listed in S1 Table and illustrated in Figs 1C and S1B, and looking at top marker genes within each cluster as defined by the scoreMarkers function (scran 1.26.0, S5 Table).

Cluster 19 showed strong expression of ACTG1, GPR37, NAPSA and SFTPA1, typically expressed by mucus producing epithelial cells. Cluster 9 and its subclusters were enriched for NK cell typical genes such as KLRK1, KLRB1, KIT and NCR1. As our cell sorting strategy was not intended to capture NK cells, NK-cell-like T cells or epithelial cells, clusters 9 and 19 risked presenting a biased representation of these larger cell populations and were therefore removed from further analysis.

scRNA-seq analysis, differential abundance analyses

To compare the abundance of cells within each cluster across conditions, we performed differential abundance analyses using a negative binomial generalized linear model with empirical Bayes quasi-likelihood F-tests [71,72], as implemented in edgeR (v 3.40.0). We included sex as a covariate and fitted the model with all conditions to test for differential abundance in any condition. The number of cells per cluster was normalized by sample cell number within the comparison, and trend estimation was turned off when estimating dispersion and quasi-likelihood dispersion. To account for the sort and pool strategy used to balance cell types among the sequenced cells, clusters were divided according to their membership of the 4 sorted cell types (macrophages, CD4 T cells, CD8 T cells, B cells) before running differential abundance analyses within each. Those clusters that did not obviously fall into one of these cell types were excluded in order to keep the universes for differential abundance analysis equivalent to or smaller than the sorted cell types, thereby avoiding sorting biases in intra-cell-type differential abundance analyses.

scRNA-seq analysis, differential expression analyses

Differential expression analyses for macrophages (clusters 1, 4 and 8), CD4 T cells (clusters 7, 10, 18b and 20a), CD8 T cells (clusters 6, 18a and 20b) and B cells (clusters 12 and 16) (Figs 5 and 6) were performed using a NEgative Binomial mixed model Using a Large-sample Approximation (NEBULA) [49], using PBS treated BAL as a baseline, blocking on sample and cluster of origin (to account for the effect of differential cell numbers of clusters between treatment groups). Final p values were adjusted using the Benjamini-Hochberg procedure [73] as implemented in p.adjust (stats v4.1.1). All results from this analysis are provided in S6 Table. To mitigate against potential issues with limited numbers of subjects [49], we also ran pseudobulk differential expression analyses within each cluster. Briefly, for each comparison of immune challenge versus PBS, counts from all cells in each cluster in each sample were pooled, limiting to clusters with at least 10 cells in at least 2 samples from each condition. For each cluster, we then fitted a negative binomial generalised linear model, blocking on sex, and tested for differential expression between conditions using empirical Bayes quasi-likelihood F-tests, as implemented in edgeR (v 3.40.0).

Gene Ontology Enrichment

Lists of differential expressed genes were passed to topGO (2.44.0) using the Genome wide annotation for pig (DOI: 10.18129/B9.bioc.org.Ss.eg.db, release 3.13.0), biological process subontology and a minimum node size of 5. Enrichment tests for classicFisheralong with all results are available in S4 Table.

scRNA-seq analysis, comparison with PBMCs

For comparison with pig PBMC transcriptomic characteristics, scRNA-seq data generated by Herrera-Uribe et al. 2021 [23] was downloaded under license CC BY-SA 4.0 from: https://data.nal.usda.gov/dataset/data-reference-transcriptomics-porcine-peripheral-immune-cells-created-through-bulk-and-single-cell-rna-sequencing/resource/7c096891-2807-4e4e-8f5f-0c02c6f434c4

This PBMC dataset was then filtered to include only genes shared between the BAL and PBMC datasets (13050 shared genes); clustering and cell type annotation in the PBMC dataset was left unchanged. The top 20 differentially expressed genes for each cluster within each dataset (55 total clusters) were identified using findMarkers (scran 1.20.1) and combined (402 final genes after removing duplicates) to provide a data subset for mutual nearest neighbors correction [70]. mnnCorrect (batchelor 1.8.1) was used to correct for batch effects in each dataset and then to merge the datasets prior to plotting of joint UMAPs.

To directly compare gene expression between cell types and clusters, the top 20 defining genes for each cluster in the PBMC dataset (36 clusters) were identified using findMarkers (scran 1.20.1), duplicates removed (268 final genes) and used as selected features to index the clusters in the batch corrected PBMC dataset via scmap (scmap 1.14.1). The clusters in the batch corrected BAL dataset were then projected onto the indexed PBMC dataset via scmap, using the default threshold of 0.7 (i.e. 70%, equal to 188 genes), and the results plotted as a Sankey plot using networkD3 (networkD3 0.4).

To identify the sets of genes driving the mapping between BAL and PBMC clusters, all genes and all clusters inputted into scmap were plotted as a heatmap (S3A Fig), with clusters of genes manually annotated (guided by Gene Ontology terms and the literature). Selected subsets of cell clusters and gene clusters were then plotted as separate heatmaps for Figs 3E and 3F and S3B–S3G.

Flow cytometry of BAL cells

BAL cells were thawed, washed in PBS, plated into 96-well round bottom plates for flow cytometry staining and labelled with antibodies grouped into four staining panels (S2 Table). Some antibodies were labelled with isotype or species-specific secondary antibodies conjugated to fluorochromes. Biotinylated antibodies were labelled with streptavidin conjugates. Free binding sites of secondary antibodies were blocked with mouse IgG (1μg per sample, Jackson) prior to incubation with further, directly conjugated antibodies. Live/dead discrimination was achieved by labelling of samples with VDeFluor780 (ThermoFisher). Following labelling of surface markers, cells were fixed and permeabilized with eBioscience Foxp3/ Transcription Factor Staining Buffer Set (ThermoFisher) according to manufacturer’s instructions, followed by labelling with antibodies against intracellular targets. Incubation steps for extracellular markers lasted for 20 minutes, incubation with antibodies for intracellular targets was performed for 30 minutes. Cells were kept at 4°C at all times. Cells from BAL and TBLN were analysed on a Cytek Aurora spectral flow cytometer, equipped with 5 lasers (UV 355nm, Violet 405nm, Blue 488nm, Yellow-Green 561nm, Red 640nm) and 64 fluorescence detection channels UV: 16, Violet: 16, Blue: 14, Yellow-Green: 10, Red: 8. Per sample, at least 2 x 10⁵ cells were acquired. Data was acquired using Cytek’s SpectroFlo software and later analysed with FlowJo version 10 (BD Biosciences).

Analysis of Treg function

BAL cells from Ad-HA/NP immunized animals were stained with CD4 PerCP-Cy5.5 (as in S2 Table) and CD25 (K231.3B2, AF488-conjugated, Bio-Rad). CD4⁺CD25^high and CD4⁺CD25^neg cells were sorted on a FACSAria UIII (S4C Fig). To determine the effect of Tregs, 4 x 10⁴ cells of the sorted cell populations were added to 2 x 10⁶ BAL cells from animals immunized with Ad-HA/NP+Ad-IL-1β. These cocultures were stimulated with H1N1pdm09 (MOI of 1) or medium control overnight. Intracellular staining was performed to analyse IFNγ and TNFα production by CD4 and CD8 T cells. Brefeldin A (GolgiPlug, BD Biosciences) was added, and 5 h later cells were stained with mAbs against CD4-PerCP-Cy5.5 (as in S2 Table) and CD8β-FITC (clone PPT23, Bio-Rad). The cells were fixed and permeabilized with BD Cytofix/Cytoperm (BD Biosciences) as per the manufacturer’s instructions. The cells were then stained with antibodies against IFNγ (P2G10, AF647-conjugated, BD Biosciences) and TNFα (Mab11, BV421-conjugated, BioLegend). Additionally, Near-Infrared Fixable LIVE/DEAD stain (Invitrogen) was used for identification of live cells.

Virus infectivity assays to determine antiviral state of pig BAL

Pig BAL cells from pH1N1 infected or control groups were challenged with infectious recombinant VLPs of vesicular stomatitis virus (VSV-ΔG-GFP) in which the glycoprotein was replaced with GFP which was used as a marker to monitor the replication of VSV-ΔG-GFP. The procedure for infecting cells with VSV-ΔG-GFP was previously described [55,56,74] with slight modifications in which we measured the kinetics of GFP expression from VSV-ΔG-GFP in a live cell imager (IncuCyte-S3).