C57BL/6J mice were purchased from Jackson Laboratory (JAX stock #000664). PyMT (MMTV-PyMT) mice were obtained from Professor Pamela Ohashi, University of Health Network. PyMT mice were originally developed by Professor William Muller, McGill University (18). Mice were maintained on a C57BL/6J background by crossing C57BL/6J female mice (Jackson Laboratory, JAX strain #000664) to PyMT transgenic male mice. Mice were bred and maintained under specific pathogen-free conditions at the University of Calgary, under a 12 h:12 h light:dark cycle at 20–24°C and 30–50% humidity. Female mice expressing the PyMT transgene with confirmed mammary tumors were analyzed. Experiments were approved by the University of Calgary Health Sciences Animal Care Committee (under protocol AC23-0003, AC23-0054, and AC23-0162) and performed in accordance with the guidelines of the Canadian Council on Animal Care.

Various tissues and tumor models were used for antibody titration and validation. The MC38 colorectal cancer and MCA205 fibrosarcoma cell lines were maintained at 37°C and 5% CO₂ in complete media consisting of Roswell Park Memorial Institute (RPMI) 1640 (Gibco) containing 10% heat-inactivated FCS (Gibco), 1mM sodium pyruvate (Gibco) 2 mM L-Glutamine (Gibco), 50 mM β-mercaptoethanol (Gibco), 100 U/mL penicillin, and 100 mg/mL streptomycin (Gibco). Cells tested negative for Mycoplasma. C57BL/6J mice were shaved and 8 x 10⁵ MC38 or MCA205 cells resuspended in sterile 1X PBS (Gibco) were inoculated subcutaneously on the flank. Tumor size was measured routinely using calipers every two to three days and humane endpoint was reached when tumors were 15 mm in any direction or showed ulceration.

Mice were euthanized with CO₂ or via cervical dislocation and spleens, small intestines, lungs and tumors were collected into plates containing ice-cold 1X PBS. Tissues were kept on ice until processing. Lungs were minced using scissors in 5 mL RPMI 1640 media containing 100 µg/mL DNAse I and 1 mg/mL Collagenase IV (Worthington Biochemicals). Tissues were digested under gentle agitation at 37°C for 45 minutes. After vortexing samples, digested tissues were filtered using a 70 μm cell strainer, washed with 10 mL 1X PBS, and centrifugated at 1500 rpm for 5 minutes. Cell pellets were resuspended in 1 mL ACK buffer for red blood cell lysis. After 2–3 minutes incubation at room temperature, suspensions were topped up with 9 mL 1X PBS and cells were centrifugated at 1500 rpm for 5 minutes. After discarding the supernatant, cells were resuspended in 600 μL of 1X PBS and kept on ice until antibody staining. Tumors harvested from PyMT mice at 14–16 weeks of age were first weighed, then minced, digested, and filtered as per lungs above. Cell pellets were resuspended in 1X PBS at a final concentration of 250 mg tumor per mL and kept on ice until antibody staining. Spleens were homogenized and filtered using a 70 μm cell strainer, washed with 10 mL 1X PBS and centrifugated at 1500 rpm for 5 minutes. Following red blood cell lysis with 1 mL ACK buffer for 2–3 minutes as per lungs above, cell pellets were resuspended in 5 mL 1X PBS. Small intestines were cleared of Peyer’s patches and fat tissue was removed. Following the removal of fecal content, tissues were washed with 1X PBS and minced to 2 mm pieces. Minced tissues were washed again with 1X PBS and dissociated in 20 mL Hank’s Balanced Salt Solution (HBSS) media (Gibco) containing 2% vol/vol FBS and 5 mM EDTA under gentle agitation at 37°C for 45 minutes. After vortexing samples, dissociated tissues were filtered using a 70 μm cell strainer and washed with 1X PBS. Tissues were digested in 8 mL in RPMI containing 2% vol/vol FBS, 100 μg/mL DNAse I, 1 mg/mL Collagenase IV, and 0.2 U/mL Dispase (Worthington Biochemicals) under gentle agitation at 37°C for 45 minutes. After vortexing samples, digested tissues were filtered using a 70 μm cell strainer, washed with 1X PBS, and centrifugated at 1700 rpm for 7 minutes. Cell pellets were resuspended in 6 mL 40% Percoll (Cytiva) and 4 mL 80% Percoll was underlaid. Cells were centrifugated at 2200 rpm for 20 minutes at low acceleration and no brake. Cell fraction at the interface was collected and washed with 1X PBS, and centrifugated at 1700 rpm for 10 minutes. Cell pellet was resuspended in 200 µL 1x PBS and kept on ice until antibody staining.

To select the appropriate antibody-fluorescent conjugate pairs, we characterized each marker as either primary, secondary, or tertiary (Table 1). We paired markers to fluorescent conjugates (Table 2) with the following in consideration. First, we paired tertiary markers preferentially to bright conjugates, with consideration for new commercially available conjugates with very distinct spectral signatures. For instance, R718 in place of AF700. R718 has lower secondary peak emission than AF700, which improves unmixing quality. Furthermore, the emission maxima of fluorochromes conjugated to tertiary markers should not overlap with the natural autofluorescent signature of tissue type. We found that lymphocytes in the lung and tumor autofluoresce with peak emission in the same channels as BUV496 and BV510 (Supplementary Figure 1A). CD11b and CD90.2 have high expression profiles that are not impacted by tissue autofluorescence and were paired to these conjugates, respectively. In addition, the signal from certain conjugates can either spread into another conjugate or impact its staining index. The spillover spreading matrix (SSM) and stain index reduction (SIR) can be evaluated to ensure that the resolution of tertiary markers is preserved. For example, GATA3 staining on BV711 is negatively impacted by RB705, whereas GATA3 on PE-Cy7 shows improved resolution (Supplementary Figure 1B). When assigning conjugates for the remaining primary and secondary markers, it is important to pair co-expressed markers with conjugates that are spectrally dissimilar to each other. For instance, CD3 and CD4 are co-expressed by CD4⁺T cells and if put on overlapping conjugates, the resulting combination can significantly impact staining resolution (Supplementary Figure 1C). Finally, careful consideration should be taken regarding the use of conjugates with secondary emission peaks in channels across multiple lasers, as they significantly induce spread and reduce the resolution of affected parameters. We have incorporated RB705 in this panel, which has the same peak emission as PerCP-eF710, but significantly less background.

Viability staining. In a 96-well conical (V) bottom plate, single-cell suspensions were incubated in 50 μL 1X PBS with reconstituted Zombie NIR dye (BioLegend) at a 1:1000 dilution for 30 minutes on ice. Cells were washed with 1X PBS 2% vol/vol FBS (henceforth called flow cytometry buffer), centrifugated at 1700 rpm for 3 minutes and supernatant was discarded. Fc receptor blocking. Cells were resuspended in 50 μL flow cytometry buffer with anti-mouse CD16/CD32 (clone 93, eBioscience) at a final concentration of 2.5 μg/mL and incubated for 20 minutes on ice. Cells were then centrifugated at 1700 rpm for 3 minutes and supernatant was discarded. Surface antibody staining. Cells were resuspended in 50 μL flow cytometry buffer containing surface antibodies (Table 2) and incubated for 30 minutes on ice. Cells were washed with flow cytometry buffer, centrifugated at 1700 rpm for 3 minutes and supernatant was discarded. Cell fixation and permeabilization. Cells were resuspended in 50 μL Foxp3 Transcription Factor Fixation/Permeabilization solution (eBioscience) at a 1:3 concentrate to diluent ratio, as per manufacturer’s protocol, and incubated for 30 minutes on ice. Cells were washed with either flow cytometry buffer or 1X Permeabilization Buffer (eBioscience), centrifugated at 1700 rpm for 3 minutes and supernatant was discarded. Intracellular antibody staining. Cells were resuspended in 50 μL 1X Permeabilization Buffer containing intracellular antibodies (Table 2) and incubated for 45 minutes on ice. Cells were washed with flow cytometry buffer and supernatant was discarded. Sample acquisition and analysis. Stained samples were resuspended in 200 μL flow cytometry buffer and acquired on a 5-laser Cytek Aurora System (Cytek Biosciences) or 6-laser Sony ID7000 Spectral Cell Analyzer (Sony Biotechnology). For direct comparison between the two instruments, we divided samples into two for staining and acquisition. Single color reference controls were prepared on UltraComp eBeads™ Plus Compensation Beads (Thermo Fisher Scientific) and BD™ SpectraComp™ Unmixing and Compensation Particles (BD Biosciences), or a 1:1 live to heat killed single-cell suspension for Zombie NIR. For cell count enumeration, CountBright™ Absolute Counting Beads (Thermo Fisher Scientific) were added to samples prior to acquisition. Samples were spectrally unmixed using single color reference controls and the autofluorescent signatures extracted from unstained controls as parameters. Unmixed samples were analyzed with FlowJo software v10.10.0.

Data cleaning and import into R. Sample data were cleaned up on FlowJo and live CD45⁺ populations were exported to FCS3.0 format. The FCS3.0 files were loaded into R using FlowCore (21), FlowViz (22), and FlowVS (23) packages as a collection of flowFrame files. Formatting data using flowSet allows efficient bulk processing and a neat R file environment, especially when there is a significant number of files. Importantly, only FCS files with the same parameter configuration can be imported into the same flowSet. Transformation and Quality control. Arcsinh transformations were performed using the FlowVS (23) package, and co-factors of 2000 were used for the lung samples, 4000 and 5000 were used for the tumor samples run on Cytek and Sony, respectively. Appropriate cofactors need to be tested, and different cofactors can be applied to different fluorescence channels. Using optimal co-factors, data transformation stabilizes the variance within each fluorescence channel, homogenizing cell populations for analyzing downstream biologically relevant clusters. Quality control was then performed using the PeacoQC (24) package. Normalization of batch effects. To normalize fluorescence intensities across technical replicates, we used the CytoNorm (25) package and performed batch correction by using aggregated files from multiple samples and tissue types. Clustering and dimensionality reduction. Adapted from the approach described in Spasic et al. (26), we incorporated the Seurat (27) packages for the clustering of flow cytometric datasets. CSV files were exported from pre-processed FCS files using the Phenoflow (28) package before being imported as Seurat Objects using Seurat. After scaling of the data and principal component analysis (PCA), we used lineage markers as features to calculate neighbors and define clusters. Next, we performed dimensionality reduction using uniform manifold approximation (UMAP) to better preserve the data structure, which maintains the spatial relationship between cells and clusters. Data Integration. We used the Harmony (29) method for data integration. The R scripts and associated FCS files were deposited onto https://github.com/JacquelotLab for easy implementation and reproducibility of the analysis.

Data analyses and representations were performed either with R software or Prism (GraphPad version 10.1.0). Statistical analyses were two-sided. Unpaired comparisons between two groups were performed using Mann-Whitney tests. Paired comparisons between two groups were performed using Wilcoxon tests. Results are shown as the mean ± SD. p-values were two-sided with 95% confidence intervals and considered significant at p < 0.05.

The identification of ILCs relies on the absence of surface lineage markers and the use of lineage-defining transcription factors. As such, our panel incorporates markers to characterize a broad array of lymphocytes in addition to ILCs. In contrast to most previously published spectral flow cytometric ILC panels where lineage markers are grouped into one color for quick negative selection (6, 30, 31), our lineage cocktail consists of lineage-defining antibodies conjugated to spectrally distinct fluorochromes (Table 2). Using a 5-laser Cytek Aurora, the 25 commercially available antibodies we selected on 24 colors (Table 2) were first titrated on single-cell suspensions of murine tissues to determine staining resolution and antibody concentration for optimal staining (Supplementary Figure 2). We then tested this antibody staining panel on lung samples (Supplementary Figure 3) and were able to identify CD4⁺ and CD8⁺ T cells (CD3⁺TCRβ⁺), and B cells (CD19⁺). We assigned TCRβ and CD19 the same fluorophore due to their mutually exclusive expression profiles, reducing the complexity of the panel. By plotting against CD3, we were able to accurately identify both CD19⁺ B cells and CD3⁺TCRβ⁺ T cells. We used NK1.1 and NKp46, both expressed by group 1 ILCs, and stained for EOMES and CD49a to separate the population into NK cells and ILC1s, respectively. After ensuring the remaining lymphocytes do not express other lineage markers, CD90.2- and CD127-expressing cells were further characterized as GATA3⁺RORγt^- ILC2s, or RORγt⁺ ILC3s. We implemented this panel to identify ILCs in mammary tumor tissue (Figure 2) isolated from PyMT mice.

To identify the various ILC subsets in mammary tumor samples, we implemented a gating strategy onto the unmixed sample data (Figure 2). Briefly, after irregularities in flow rate, debris, and doublets were excluded, live CD45⁺ leukocytes were identified (Figure 2A). Using our validated manual gating strategy, we identified group 1 ILCs in tumors as CD3^-TCRβ^-CD19^-NK1.1⁺NKp46⁺, further subdivided into NK cells or ILC1s based on EOMES and CD49a expression, respectively. In both lung and tumor samples, CD11b and KLRG1 expression was assessed for the detection of immature (CD11b^-KLRG1^-) and mature M1 (CD11b^loKLRG1^-) and M2 (CD11b^loKLRG1⁺) NK cell subsets (32). Then, ILC2s were characterized as CD3^-TCRβ^-CD19^-NK1.1^-NKp46^-CD4^-CD8^-CD90.2⁺CD127⁺GATA3⁺RORγt^- while ILC3s were identified as CD3^-TCRβ^-CD19^-NK1.1^-NKp46^-CD4^-CD8^-CD90.2⁺CD127⁺RORγt⁺ (Figure 2B; Supplementary Figure 3B). In the lung, amongst the four ILC subsets, NK cells, particularly the M2 subset, and ILC2s constituted the major populations (Supplementary Figure 3C), while in mammary tumors, NK cells, largely of the M1 phenotype, and ILC1s represented most of the ILC infiltrate (Figures 2C, D).

It is now widely recognized that ILCs are heterogenous populations. Depending on signals they receive from their environment, ILCs may exhibit significant plasticity in their phenotype and surface marker expression. As such, we incorporated markers to unravel the diversity within these populations across different tissue types. We observed that immature NK cells express higher levels of Ki67 compared to the other ILC subsets, and proliferation capacity in tumors was markedly higher in all ILCs compared to those in the lung (Figure 3). CD25 expression was mostly restricted to ILC2s in both tissues, although there was a minor population of CD25⁺ ILC3s (Figure 3). Unsurprisingly, ST2 expression was exclusive to ILC2s (Figure 3). Approximately 70% of tumor and lung ILC2s expressed CD117, and we observed a higher proportion of CD117⁺ immature NK cells and CD117⁺ ILC1s in the tumor compared to their lung counterparts (Figure 3). Interestingly, KLRG1 was expressed by nearly half of lung ILC1s, whereas KLRG1⁺ ILC1s were absent in the tumor (Figure 3). 4-1BB expression was largely exclusive to group 1 ILCs, and LAG-3 was only found to be expressed on a small number of tumor ILC1s, but not NK cells infiltrating mammary PyMT tumors, as previously reported by our group (33) (Figure 3). We observed increased PD-1⁺ ILC2s in mammary tumors, similar to our previous observations in melanoma (34), although a smaller subset of ILC2s in the lung also expressed PD-1 (Figure 3). The increase of PD-1 expression on tumor-infiltrating ILC2s mirrored that seen in T cells (Supplementary Figure 4), the latter of which has been well described in the context of anti-tumor immunity and immunotherapies (35). Amongst all ILC subsets, ILC2s have the highest levels of PD-1 in mammary tumors (Figure 3).

Analysis of flow cytometric data by gating strategy is a reliable way to identify proportions of ILCs and their expression profiles, all the while preserving the ability to characterize the phenotypes of adaptive lymphocytes (Supplementary Figure 4). However, the increased number of parameters that can be accurately unmixed by spectral flow cytometry allows for the analysis and visualization of data using similar bioinformatics approaches as the ones used to analyze high dimensional single-cell datasets. To that aim, we implemented an unbiased bioinformatics pipeline for high-dimensional flow cytometric analysis. We first cleaned and exported live CD45⁺ cell data from FlowJo, removing autofluorescence parameters and the upstream channels used to gate on live CD45⁺ cells (forward and side scatter, Zombie NIR, CD45), and saved them as individual new FCS files. The data were imported into R, transformed, quality controlled, normalized, and scaled (Supplementary Figure 5). Unsupervised clustering was conducted, and cells were visualized with UMAP (Figure 4A). Clusters were merged and labeled based on their protein profiles visualized by dotplots (Figure 4B), heatmap (Supplementary Figure 6) and feature plots (Supplementary Figures 7-8). Using marker expression for immune cell labelling, this initial clustering identified myeloid cells, B cells, CD4⁺ and CD8⁺ T cells, NK cells, ILC2s, and other ILC subsets. Because of technical variation in staining among different tissue types, the same cell types showed spatial differences between tumor and lung. Therefore, we used an integration method to eliminate variation inherent to differences in autofluorescence and staining intensities between tissues (Figure 4C). Additionally, we removed the CD11b⁺ myeloid population to improve the efficiency and accuracy of lymphoid cell re-clustering, since the panel lacked sufficient myeloid-related markers. After integration, spatial variation between tissues was resolved. With the exclusion of myeloid populations, we were able to identify additional cell subtypes (Figures 4C, D; Supplementary Figures 9-10), including ILC1s, γδ T cells, and Ki67⁺ and PD1⁺ αβ CD4⁺ and CD8⁺ T cell subsets. After quantification of the proportion of each cell type, we found that the percentage of ILC2s remained similar before and after integration (Figures 4E, F), indicating that the integration process did not affect clustering accuracy but, instead, improved data resolution with the identification of additional unique subsets.

To demonstrate the reproducibility of our workflow, this panel was similarly optimized and validated on the Sony ID7000 Spectral Cell Analyzer (Supplementary Figures 11-14). We did not observe differences in lymphocyte frequencies when compared to the data acquired from the Cytek Aurora flow cytometer (Figure 5A), although we did find small discrepancies in marker expression levels, particularly CD25 expression on ILC2s (Figure 3; Supplementary Figures 13-14). Upon clustering and data integration (Figure 5B), while lower resolution of the CD4⁺ T cell subset was observed on the Sony instrument, no major differences were noted between flow cytometric data generated using the Cytek and Sony flow cytometers (Figures 5C, D). Together, these observations support the implementation and use of this bioinformatics pipeline for reproducibility and accuracy of ILCs in tissues such as tumors.