Neutrophils are isolated using anti-Ly-6G microbeads according to the manufacturer’s instructions.

Labeled cell suspensions were partitioned into single-cell droplets with a Chromium Controller (10x Genomics) and gene expression and cell surface protein libraries were further generated using Chromium Single Cell 3’ Reagent Kits with Feature Barcoding Technology (10x Genomics) following the manufacturer’s recommendations. Samples were sequenced on a NovaSeq 6000 Sequencing System (Illumina) and the resulting FASTQ files were processed with CellRanger (version 6.0.1, 10X Genomics). Unless noted, all the analysis for the single-cell RNA-sequencing was done using the Python Scanpy framework (Scanpy version 1.9.1) (7). Cells were further filtered to include cells with log1p values of isotype control antibody reads (< 6), number of antibodies detected per cell (< 30), mitochondrial reads (< 30%), total RNA reads (< 40000), number of genes (< 6000 and > 200). To normalize antibody reads in the dataset, denoised and scaled by background (DSB) program (8) using the default parameters implemented in MUON package (version 0.1.2) was used (9). Filtered raw RNA reads were further normalized and log-transformed using normalize_total and log1p functions, respectively and data were further scaled using scale function. Before data scaling, log-normalized data were stored separately for further downstream analysis such as differential expression test. DSB-normalized protein reads were clustered using Leiden algorithm after generating a neighborhood graph with default parameters and for three different resolution settings (0.25, 0.5, 0.75). For each resolution, a heatmap was plotted for the average expression of proteins per cluster to determine at which resolution biologically relevant clusters can be visualized. To assign the respective anatomical location to each cell, we first used histograms of DSB-normalized CD45 and CD45.2 antibody reads to determine the cut-offs. Respective cut-offs were applied to each separate anatomical location (0.3 for CD45.2⁺CD45^- airway cells; > 0.28 for CD45.2^-CD45⁺ intravascular leukocytes; < 0.3 and < 0.28 for CD45.2^-CD45^- parenchymal cells). For differential expression (DE) test per cluster, tl.rank_genes_groups function with Wilcoxon rank-sum test was used.

All data were analyzed with GraphPad Prism software (GraphPad Software). P values below 0.05 were regarded as statistically significant (*p<0.05, **p<0.01, ***p<0.001). Unless otherwise indicated, figures display means ± standard deviation (SD). Experiment sample sizes (n), experiment replicate numbers and statistical tests used are included in the respective figure legends.

For in vivo labeling of immune cells in the vasculature and the airways, two different clones of pan-leukocyte anti-CD45 antibodies, clone 30-F11 and clone 104, are administered intravascularly or intratracheally, respectively. To ensure the amount of antibody used is sufficient to label all cells present in the respective anatomical compartment, we administered fluorescently labeled antibody versions via the designated route. In addition, we stained the labeled BAL or blood cells with the other anti-CD45 clone to gate on all CD45⁺ cells. Potential steric competition between antibodies was assessed beforehand by preincubation of mouse splenocytes with either fluorescent clone 104 or 30-F11 followed by staining with the respective other clone and comparison to single-stained cells ( Supplementary Figure 1 ). 30-F11 staining intensity, depicted as geometric mean fluorescence intensity (gMFI), was not affected by prior labeling of cells with clone 104 ( Supplementary Figure 1A ). When cells were prelabeled with 30-F11, staining intensity of 104 was significantly lower compared to the single-labeled cells indicating some binding interference ( Supplementary Figure 1B ). However, frequencies of double-positive cells remained comparable to the 104-only stained control indicating both clones can be used for simultaneous labeling of leukocytes ( Supplementary Figures 1C, D ). Importantly, flow cytometric analysis following in vivo antibody instillation showed that all immune cells in the airway and vasculature were labeled with clone 104 or 30-F11 respectively, indicating that the used concentrations and experimental conditions were suitable to efficiently tag leukocytes in vivo ( Figures 1A, B ). Studies were performed on infected as well as uninfected animals to ensure consistent labeling under both conditions and antibody concentrations should be tested beforehand when a different model is used. To determine whether CD45 labeling remains stable during downstream processing steps, we labeled airway and blood leukocytes with fluorescently labeled antibodies in vivo. Isolated BAL and blood cells further underwent enzymatic digestion with Liberase and dsDNase and CD45 gMFI was compared to that of undigested cells ( Figures 1C, D ). Double strand-specific DNase was used to preserve the integrity of the TotalSeq B-barcodes since regular DNase I also utilizes single-stranded DNA as a substrate. gMFIs did not change with digestion indicating that anti-CD45 antibodies remain stably bound to their epitopes during processing and can be confidently used under the tested conditions for scRNA-seq analysis.

During downstream lung processing, the airway and intravascular boundaries are disintegrated, and excess antibodies need to be neutralized to avoid cross-labeling of blood cells with unbound airway anti-CD45 or vice versa. We therefore instilled recombinant murine CD45 (rmCD45) intratracheally after antibody labeling and added additional recombinant protein to the lung digestion medium. The amount of rmCD45 needed to neutralize potentially unbound antibody was assessed by titration experiments where we incubated the respective fluorescent antibodies with increasing concentrations of rmCD45. This was followed by addition of mouse splenocytes and analysis of CD45 labeling by flow cytometry. As expected, all splenocytes were stained with either antibody in the absence of rmCD45 ( Figures 2A, B ). Clone 104 staining was completely abolished with 2.5 µg rmCD45 indicating that all antibody in solution had been effectively neutralized ( Figure 2A ). Similar observations were made with clone 30-F11, albeit 10 µg of rmCD45 were needed to achieve a similar degree of neutralization ( Figure 2B ). We next assessed whether 10 µg rmCD45 was a suitable concentration to neutralize unbound antibodies in vivo. We first analyzed neutralization efficiency in the airways by consecutive intratracheal administration of fluorescent anti-CD45 (clone 104) and rmCD45. To assess neutralization of anti-CD45 (30-F11), we administered a fluorescent version via tail vein injection, allowed the antibody to circulate and injected rmCD45 via the same route. Neutralization efficiency was assessed by the ability of BAL supernatant or plasma to label naïve mouse splenocytes and was analyzed by flow cytometry ( Figure 2C ). As expected, when only anti-CD45 was administered, all splenocytes were labeled, whereas the presence of rmCD45 in either airway or blood completely neutralized unbound antibody, evident by the lack of splenocyte labeling. We therefore chose 10 µg rmCD45 to proceed with our evaluation. It should be noted that this is a reference value for our model system. Component concentrations for other in vivo models should be tested and confirmed empirically before RNA-seq analysis.

To test the validity of our method we used an acute bacterial pneumonia model where we intranasally infected mice with a high dose of NTHi or administered PBS as a control. We additionally employed an acute LPS-induced lung inflammation model to further validate the specificity of our in vivo staining procedure. Blood and airway leukocytes were labeled with either fluorescent or TotalSeq-B anti-CD45 antibodies at 48 hours post-infection or 24 h after LPS challenge. Depending on the cell population of interest, enrichment is usually performed preceding scRNA-seq to increase sequencing resolution. This often entails positive selection for CD45⁺ leukocytes using magnetic bead enrichment. We therefore tested whether in vivo CD45 labeling would interfere with positive selection via bead-bound anti-CD45 antibodies using lung cells labeled with anti-CD45-FITC (clone 30-F11), anti-CD45.2-APC (clone 104) or a combination thereof. While the purity of the CD45⁺ fraction was comparable in all conditions (>97%) after enrichment, the flowthrough (CD45^- fraction) of cells previously labelled with anti-CD45 (clone 30-F11) contained significantly higher frequencies of CD45⁺ cells than the unstained control or cells pre-labeled with anti-CD45.2 (clone 104) ( Supplementary Figure 2A ). In accordance, numbers of CD45⁺ cells ( Supplementary Figure 2B ) were markedly lower in CD45⁺ fractions from 30-F11-stained cells as opposed to unstained or 104-stained cells. On the other hand, there was a >25-fold increase of CD45⁺ cells recovered in the flowthrough, indicating insufficient retention of 30-F11-stained cells in the column. CD45⁺ enrichment should therefore be avoided, and enrichment should be performed by negative selection or with magnetic beads of a different specificity. We therefore directly enriched for neutrophils with Ly6G microbeads and using fluorescently labeled antibodies showed they are almost exclusively found in the lung vasculature of uninfected animals ( Figures 3A, B ). As expected, inoculation with NTHi or LPS challenge resulted in neutrophil recruitment from the blood to the site of infection indicated by distinct populations of CD45.2(104)⁺CD45.2(30-F11)^- airway, and smaller populations of CD45.2(104)^-CD45.2(30-F11)⁺ intravascular and CD45.2(104)^-CD45.2(30-F11)^- parenchymal neutrophils ( Figures 3A, C ; Supplementary Figure 5A ). Importantly, almost no double-positive cells were observed indicating the absence of cross-labeling. This can be attributed to efficient neutralization of excess antibodies since all cells were labeled with intratracheal anti-CD45 antibody in the absence of rmCD45 after NTHi infection ( Supplementary Figure 3 ). No staining with intravascular antibody could be detected on Siglec-F⁺CD11c⁺ airway-resident alveolar macrophages after NTHi infection, confirming the absence of antibody cross-contamination under the indicated conditions ( Supplementary Figure 3C ). Distinct neutrophil populations could also be distinguished following infection based on distribution of CD45 barcodes after labeling with TotalSeq-B antibodies ( Figure 3B ) validating the compartmentalized in vivo labeling approach. Double positive signals exclusively stemmed from non-single cells and were filtered based on total antibody and RNA reads per cell. Specifically, cells that have detectable reads of 30 or more surface proteins used for labelling were deemed to be cell clumps or doublets and removed from the downstream analysis. In addition, cells with high RNA reads per cell (> 40,000) were also filtered out to remove low-quality or doublets from the analysis.

To analyze location-specific gene expression profiles, filtered scRNA-seq data were projected into two dimensions using Uniform Manifold Approximation and Projection (UMAP) based on their CD45 surface label. Unbiased clustering identified three distinct neutrophil subpopulations, cluster 0 (airway), cluster 2 (blood) and cluster 1 (parenchyma) ( Figure 4A ). Higher transcript levels of inflammatory and antibacterial neutrophil response genes such as Sod2, Cd274, Ncf1, Lcn2, FceR1g and the cytokines/chemokines Il1a, Ccl3, Ccl4 and Cxcl2 suggest increased activation of airway over tissue and blood subpopulations ( Figures 4B, C ). Downregulated genes in airway neutrophils including Sell, Zyx, Selplg, Coro1a, S100a8, S100a9 and Cxcr2 were mainly associated with neutrophil locomotion and adhesion ( Figure 4C ). In general, tissue and blood neutrophils were more similar in their expression profiles. In accordance, subclustering based on transcriptomic data alone and subsequent analysis of CD45 surface labeling in the resulting clusters showed that airway neutrophils are transcriptionally distinct while gene expression of blood and parenchyma neutrophils was partially overlapping ( Supplementary Figures 4A, B ). Like airway neutrophils, upregulated genes in blood and tissue neutrophils were involved in inflammatory and antibacterial responses (Cd14, Acod1, Ccrl2, Cxcl2, Il1rn, Ncf1, Ptafr, Clec4d, B2m) while downregulated genes were associated with cytoplasmic translation (Rpl7, Rpl30, Rps9, Rps27a, Rplp0) ( Figure 4C ). We further assessed PD-L1 surface expression in neutrophil populations by flow cytometry after in vivo labeling with fluorescent anti-CD45 antibodies. Surface PD-L1 was highly expressed on airway neutrophils, with intermediate and low levels found in lung parenchyma and blood neutrophils, respectively ( Figure 4D ). This finding was validated in an LPS-induced acute lung inflammation model showing high expression of PD-L1 on lung and parenchymal neutrophils while their blood counterparts only exhibited low PD-L1 levels ( Supplementary Figure 5B ). These findings correlated strongly with transcript levels after NTHi infection further validating our in vivo labeling approach for scRNA-seq.