LRRC15+ myofibroblasts dictate the stromal setpoint to suppress tumour immunity – Nature

Mice

Dpt^IresCreERT2 mice¹⁰ and Lrrc15^DTRGFP mice were designed, generated and bred at Genentech. Tgfbr2^fl/fl mice (012603) were obtained from the Jackson Laboratory. Age- and sex-matched mice (6–12 weeks old) were used for all studies. Mice were maintained under specific pathogen-free conditions using the guidelines of the US National Institutes of Health. The sample sizes for each study are described in the figure legends. All experiments were performed under protocols approved by the Institutional Animal Care and Use Committee at Genentech.

Generation of Lrrc15
^DTRGFP knock-in mouse

Homologous recombination and mouse embryonic stem (ES) cell technology^23,24,25 were used to generate a genetically modified mouse strain with Lrrc15 DTR–GFP knocked-in. A gene-targeting vector was constructed with a 1,704-bp 5′ arm of homology corresponding to GRCm38/mm10 chromosome 16: 30,274,520–30,276,223 and a 1,994-bp arm of 3′ homology arm corresponding to chromosome 16: 30,270,786–30,272,779.  Delete of exon 2 after ATG corresponds to chromosome 16: 30,272,780–30,274,516. DTR-EGFP-SV40-FRT-PGK-neo-FRT was inserted immediately after ATG of exon 2. The final vector was confirmed by DNA sequencing, linearized and used to target C2 (C57BL/6N) ES cells using standard methods (G418⁺ and ganciclovir^– selection)²⁶ C57BL/6N C2 ES cells²⁷ were electroporated with 20 µg of linearized targeting-vector DNA and cultured under drug selection essentially as previously described²⁸. Positive clones were identified using long-range PCR followed by sequence confirmation. Correctly targeted ES cells were subjected to karyotyping. Euploid gene-targeted ES cell clones were treated with Adeno-FLP to remove PGK neomycin, and ES cell clones were tested to identify clones with no copies of the PGK neomycin cassette, and the correct sequence of the targeted allele was verified. The presence of the Y chromosome was verified before microinjection into albino Bl/6N embryos. Germline transmission was obtained after crossing resulting chimeras with C57BL/6N females. Genomic DNA from pups was screened by long-range PCR to verify the desired gene-targeted structure before mouse colony expansion. For genotyping, the following primers were used: 5′-AGGCGAGGCGATTG-3′, 5′-CGATGAGGGCTGAAATGT-3′ and 5′-TGGTCCGTGGATACAGT-3′ amplified 408-bp wild-type and 313-bp knock-in DNA fragments. The following PCR cycle was used: 94 °C for 4 min, (94 °C for 1 min, 55 °C for 30 s, 72 °C for 1 min) for 30 cycles; 72 °C for 10 min; 4 °C ad infinitum.

Cell lines

The KPR mouse pancreatic adenocarcinoma cell line was generated by the Junttila Group at Genentech from KPR PDAC GEMMs (Kras^LSL.G12D/wt;p16/p19^fl/wt;p53^LSL.R270H/wt;Pdx1.Cre)¹¹. KPR cells were cultured in RPMI with 10% FBS (Hyclone) plus 2 mmol l^–1 l-glutamine. All cell lines were tested for Mycoplasma contamination by quantitative PCR (Lonza Mycoalert and Stratagene Mycosensor). For all injected tumours, cells were used within the first three passages.

In vivo tumour studies

For subcutaneous KPR tumours, KPR cells were trypsinized, filtered, counted and resuspended in a 1:1 mixture of Hanks’s buffered saline solution and phenol-red-free Matrigel (Corning) at a concentration of 1 × 10⁶ cells ml^–1. For all genotypes of mice used, age- and sex-matched 6–12-week-old mice were subcutaneously inoculated in the right unilateral flank with 1 × 10⁵ KPR tumour cells. Flank skin hair was shaved before implantation. Tumour volumes were measured and calculated 2–3 times per week using the following modified ellipsoid formula: ½ × (length × width²). Tumours >1,000 mm³ were considered progressed and animals were removed from the study. Similarly, animals for which tumours ulcerated greater than 5 mm were removed from the study. For subcutaneous tumour studies in Lrrc15^DTRGFP mice, when tumours reached a volume of 100–200 mm³ (about 10 days after inoculation), animals were distributed into treatment groups on the basis of the tumour volume and treatment was initiated.

For orthotopic pancreatic KPR tumours, injection of pancreatic tumour cells into the pancreas of mice was performed as previously described²⁹. KPR cells were resuspended in a 1:1 mixture of Hanks’s buffered saline solution and phenol-red-free Matrigel (Corning) at a concentration of either 2 × 10⁵ or 2 × 10⁶ cells ml^–1. Dpt^CreERT2Tgfbr2^fl/fl or Lrrc15^DTR mice were anaesthetized using inhalatory anaesthesia, placed on a heating pad and given eye drop gel. The left flank or abdominal region was shaved and sterilized using ChloraPrep (BD) before making an approximately 1-cm incision with sterile microscissors medial to the splenic silhouette. The underlying muscle layer was incised, and blunt-nose forceps were used to externalize the pancreas and spleen. A prepared 31-gauge needle containing the cell solution was inserted into the tail of the pancreas, and 50 µl of solution containing 1 × 10⁵ cells was slowly injected. The wound was closed using absorbable sutures and wound clips and the mice were allowed to recover. All animals were administered the slow-release analgesic buprenorphine SR LAB at 0.5 mg kg^–1. Mice were monitored every day after the surgical procedure for signs of infection or distress. For orthotopic tumour studies in Lrrc15^DTRGFP mice, 7 days after implantation, animals were distributed into treatment groups on the basis of the tumour volume and treatment was initiated.

Mice were collected at indicated time points after treatment for analysis or used for tumour growth studies. Sample sizes in the mouse studies were based on the number of mice routinely needed to establish statistical significance based on variability within study groups. Treatment groups were blinded when possible. All animal studies herein were approved by the Genentech Institutional Animal Care and Use Committee.

Ultrasound imaging of orthotopic pancreatic tumours

For orthotopic pancreatic tumour studies in Lrrc15^DTRGFP mice, tumour volumes were measured by ultrasound imaging. Mice were anaesthetized with 4% sevoflurane (Zoetis) in a warm induction box and positioned on their right side under a continuous flow of 2.5–3% sevoflurane during imaging. Following hair removal, ultrasound coupling gel was placed on the skin, and anatomical B-mode images were acquired on vevo2100 (Fujifilm VisualSonics-) in transverse and longitudinal planes, capturing the maximum tumour cross-sections (MS-550D probe; centre frequency of 40 MHz, axial resolution of 40 µm, lateral resolution of 90 µm and field depth of 12 mm). The pancreatic tumour volume per mouse was analysed using Vevo LAB v.5.5.1 with the following formula for an ellipsoid: volume (mm³) = π/6 × length × width × depth.

In vivo treatments

For TAM-induced Cre expression, mice were injected with 2 mg TAM (Sigma, T5648) diluted in sunflower seed oil (Sigma, 88921) for five consecutive days intraperitoneally or were fed chow containing TAM (Envigo, TD.130859). For LRRC15 CAF ablation, mice were intraperitoneally injected with 25 ng g^–1 of DT (Enzo Life Sciences, BML-G135) twice per week. For CD8-depletion studies, mice were treated with either rat IgG2b isotype control antibody or rat anti-CD8 IgG2b-depleting antibodies (BioXcell, BE0061) at a dose of 10 mg kg^–1 administered intraperitoneally three times per week. For anti-PDL1 studies, mice were treated with isotype control antibodies or anti-PDL1 (6E11) antibodies (in-house). The first dose was given at 10 mg kg^–1 followed by 5 mg kg^–1 thereafter administered intraperitoneally twice per week.

Mouse tissue digestion, cell isolation and flow cytometry

Tumours were collected, weighed and minced into small pieces. To isolate naive flank skin, hair was shaved, adipose tissue was removed and skin tissue was minced. All tissues were subsequently enzymatically digested using a cocktail of dispase (Life Technologies), collagenase P and DNaseI (Roche) for 45 min at 37 °C to obtain a single-cell suspension. Cells were counted using a Vi-CELL XR (Beckman Coulter). For cytokine staining, cell suspensions were stimulated with eBioscience Cell Stimulation Cocktail plus protein transport inhibitors (00-4975-93) resuspended in RPMI with 10% FBS plus 2 mmol l^–1 l-glutamine and 2-mercaptoethanol for 2 h at 37 °C. Cells were labelled with the following monoclonal antibodies purchased from BioLegend or BD Biosciences at 4 ºC on ice for 20–30 min, unless otherwise noted. Before cell surface staining with the following fluorescently labelled antibodies, cells were blocked with Fc block (2.4G2; 1:200, 553142). The following surface or intracellular antibodies were used: CD45 (30-F11, 103139); EPCAM (G8.8, 118218); CD31 (390, 561410); PDPN (8.1.1, 127410); CD24 (M1/69, 612832); LRRC15 (M25, in-house); CD90.2 (53-2.1, 565527); CD8 (53-6.7, 612759); PD1 (29F.1A12, 135225); TIM3 (RMT3-23, 119727); LAG3 (C9B7W, 125227); CD39 (Duha59, 143812); IFNγ (XMG1.2, 505846); granzyme B (GB11, 515408); and TNF (MP6-XT22, 506324). Live cells were identified by incubation with calcein blue (Invitrogen, C1429, 1:1,000) after surface staining. For intracellular staining, samples were fixed, permeabilized and stained using a BD Cytofix/Cytoperm Fixation/Permeabilization kit (554714) according to the manufacturer’s instructions. Data were acquired using a Fortessa, Symphony or LSRII (BD Biosciences) flow cytometer and analysed using FlowJo (Tree Star, v.10.7.1), or cells were sorted using a Fusion or Aria (BD Biosciences). instrument Data were processed using Prism GraphPad. Additional information is provided in Supplementary Table 6.

CAF–CD8 T cell co-culture

For stimulation with plate-bound anti-CD3, 96-well flat-bottomed plates were coated overnight at 4 °C with 10 µg ml^–1 of anti-CD3 (BioLegend, 100340, clone 145-2C11) and washed once with PBS. Relevant primary CAFs were sorted from digested KPR subcutaneous tumours from DTR^– or DTR⁺ mice, and 3 × 10⁴ cells were then added to the anti-CD3-coated well (100 µl). Cells were incubated for 1 h at 37 °C to facilitate attachment. During incubation, mouse CD8⁺ T cells were isolated from single-cell suspension of naive splenocytes by immunomagnetic negative selection using an EasySep Mouse CD8⁺ T cell enrichment kit from Stem Cell (19853) according to the manufacturer’s guidelines. About 6 × 10⁴ purified CD8⁺ T cells were added to the wells in the presence of soluble 0.50 µg ml^–1 anti-CD28 (BioLegend, 102115, clone 37.51) (100 µl). On the day of analysis, medium was replaced and cells were cultured with 1× Cell Stimulation Cocktail (eBioscience 500× Cell Stimulation Cocktail plus protein transport inhibitors, 00-4975) and 55 µM 2-mercaptoethanol for 4 h at 37 °C. Cells were collected, filtered and stained for surface markers. Following surface staining, cells were fixed and permeabilized with Intracellular Fixation and Permeabilization Buffer Set according to the manufacturer’s guidelines before staining for intracellular cytokines. Cells were then analysed by flow cytometry.

Immunofluorescence and image analysis of mouse tumours

Tumours were fixed overnight in 4% paraformaldehyde and embedded in optimal cutting temperature medium (Sakura Finetek) and frozen for storage at −80 °C. Sections (8–12 μm thick) were cryosectioned and stained. For staining, slides were blocked and permeabilized with normal mouse serum (1:50), mouse Fc block (1:100) and 0.3% Triton-X diluted in PBS for 30 min at room temperature. Tissue sections were incubated with primary antibodies for 1 h at room temperature or overnight at 4 °C. After washing, secondary antibodies were added for 1 h at room temperature. To counterstain, slides were rinsed and incubated with DAPI (ThermoFisher, D1306) at 300 nM in PBS for 5 min at room temperature. Details of the antibodies used can be found in Supplementary Table 6. Slides were rinsed several times in PBS, excess buffer drained and sections were mounted in Vectashield (H-1000). Images were acquired with a Nikon A1R confocal microscope equipped with a Plan apo lambda NA 0.75 ×20 lens. Lasers were set at excitation at 488 nm, 561 nm and 640 nm, and a perfect focus module. NIS Elements acquisition software was used with a digital zoom of 2 for full tissue section imaging or 7 for details, and stitching of single plane images was performed. Estimations of CD8 T cell–LRRC15 CAF interaction rates were compiled between T cells and CAFs observed among total T cells in a 500 × 500 × 10 µm³ section of tissue across 3 different tumours.

Lrrc15 in situ hybridization

Tissues for in situ hybridization were formalin-fixed and paraffin-embedded. Mouse LRRC15 in situ hybridization was performed using an ACD probe (Advanced Cell Diagnostics, 467838 with 120 min hybridization. ER2 retrieval (Leica) at 95 °C for 15 min and RNAscope 2.5 LS Protease III digestion (ACD) was performed on a Leica Bond autostainer. RNAscope 2.5 LS Reagent Kit Red (ACD) was used for detection.

Mouse scRNA-seq and cell hashing

Mouse scRNA-seq and cell hashing with unique barcoded antibodies (BioLegend) were processed using Chromium Single Cell Gene Expression 3′ v3 Library and a Gel Bead kit following the manufacturer’s instructions (10x Genomics, PN-1000075). Cells were counted and checked for viability using a Vi-CELL XR cell counter (Beckman Coulter), and then injected into microfluidic chips to form gel beads-in-emulsion in a 10x Chromium instrument. Reverse transcription was performed on the gel beads-in-emulsion, and products were purified and amplified. DNA from antibody-derived tags was separated from cDNA based on size selection using SPRIselect beads (Beckman Coulter, B23318). Expression libraries and antibody-derived tag libraries were generated and profiled using a Bioanalyzer High Sensitivity DNA kit (Agilent Technologies, 5067-4626) and quantified with a Kapa Library Quantification kit (Roche, 07960255001). All libraries were sequenced using HiSeq4000 and NovaSeq (Illumina)

Mouse scRNA-seq data processing

Initial data processing

scRNA-seq data for each library from each cell type were processed with CellRanger count (CellRanger 3.1, 10x Genomics) with a custom reference based on the mouse reference genome GRCm38 and GENCODE gene models. Counts of barcode antibodies to label individual replicates were processed using DemuxEM with default parameters to assign individual sample labels³⁰. Cells identified as doublets or HTO-negative cells were excluded from the analysis. For gene expression counts, individual samples were merged into one expression matrix and analysed using the package Seurat. Cells with fewer than 300 expressed genes or more than 5% mitochondrial counts were removed. Transcript counts were log-normalized (Seurat, NormalizeData), and the top 2,000 most variable genes were selected using variance stabilizing transformation (FindVariableFeatures), followed by data scaling (ScaleData). PCA was then performed on this gene space (RunPCA). Clustering was carried out on the basis of the shared nearest neighbour between cells (FindNeighbors, 30 PCs) and graph-based clustering (30 PCs, resolution of 0.1 for Lrrc15 depletion, 0.5 for Tgfbr2 KO experiments). We calculated markers for individual clusters using the FindMarkers function in Seurat (Wilcoxon’s rank sum test, Benjamini–Hochberg adjustment for multiple testing). To visualize gene expression levels for individual clusters, we calculated the average gene expression in each cluster and calculated a z-score value on a by-gene basis.

Filtering of cells

For the Dpt^IresCreERT2Tgfbr2^fl/fl experiments, cells in the resulting Seurat object from the initial data-processing step were further filtered based on the expression of known cell-type markers. Only fibroblast cells from clusters 0, 2, 3, 4, 5 and 8 expressing Lum and Dcn, but not Pecam1 (endothelial cells), Ptprc (immune cells) or Rgs5 (pericytes), were retained for subsequent analyses. We then performed dimensionality reduction and clustering as described above and removed remaining minor contaminant non-fibroblast cells. The final dimensionality reduction and clustering were performed using 30 PCs and a clustering resolution of 0.3.

For the Lrrc15^DTRGFP depletion experiments, fibroblasts cells (clusters 0, 1, 2 and 5) in the resulting Seurat object from the initial data-processing step were isolated by excluding clusters expressing Pecam1 (endothelial cells), Ptprc (immune cells), Rgs5 (pericytes), Krt18 (epithelial) or Myl1 (smooth muscle cells). We then performed dimensionality reduction and clustering as described above and removed remaining minor contaminant non-fibroblast cells. The final dimensionality reduction and clustering were performed using 30 PCs and a clustering resolution of 0.2.

Scoring of cells for gene expression programmes

Cells were scored for gene expression programmes using the addModuleScore function in Seurat and a gene set of interest as input. PDAC mouse GEMM programmes were derived as follows. We used genes with at least 0.6 average log fold-change upregulation in TGFβ CAFs (cluster 2) from our previous study¹ as marker genes for these conditions. To identify a normal tissue fibroblast (clusters 3 and 4) gene set, we identified the top 20 markers for clusters 3 and 4 compared these to all other cells in the dataset using the FindMarkers function in Seurat.

Population frequency analysis

To assess differences in abundance of cells from specific clusters between conditions, we used the R package speckle (https://github.com/Oshlack/speckle), which is designed for finding significant differences in cell-type proportions. In brief, speckle calculates the fraction of cells assigned to a particular cluster in each biological replicate, performs a variance stabilizing transformation on the proportions and determines whether the cell-type proportions are significant between different groups. Given that we only compared two groups in all our experiments, t-test was used by speckle to calculate P values, which were adjusted for multiple testing using Benjamini–Hochberg correction.

Pathway enrichment analysis

We used PROGENy¹⁷ to infer pathway activity from our single-cell gene expression data as previously described¹⁰ and following the single-cell tutorial provided by the authors (https://saezlab.github.io/progeny/articles/ProgenySingleCell.html). We matched progeny scores with either clusters or experimental time point/condition and summarized the data by population.

Lrrc15 gene expression in mouse tissues

Normalized fragments per kilobase of sequence per million mapped read values were retrieved from Supplementary Table 6 in ref. ¹². Data were log-transformed and expression levels of Lrrc15 and Gapdh were visualized by tissue.

TCGA data analysis

Batch-corrected normalized TCGA Pan-Cancer mRNA data were obtained from UCSC Xenabrowser (https://xenabrowser.net/) (N = 11,060). Samples containing NA expression values were removed. We additionally filtered the data to only contain samples from primary solid tumours (sample code 01; N = 9,702). Survival data were obtained from Table S1 in ref. ³¹ and linked to the Pan-Cancer dataset using the unique TCGA participant barcode. Indications with fewer than 80 patients were excluded from the analysis (final dataset: N = 9,195 patients). TGFB CAF levels were inferred by calculating the average expression of our previously published marker signature¹ within a sample after z-score transformation of each gene across samples. Association with survival across TCGA data was determined with multivariate Cox regression and TCGA indication as a covariate, as well as univariate Cox regression analysis within each indication. The hazard ratio was defined as the change in the risk of death if the TGFβ CAF score increased by one unit.

Human tumour digestion and stromal cell RNA-seq analysis

Tumour collection

Tumour samples for the Immunoprofiler Initiative (IPI) were transported from various cancer operating rooms and from outpatient clinics. All patients provided consent to the University of California, San Francisco (UCSF) IPI clinical coordinator group for tissue collection under a UCSF protocol approved by the institutional review board (UCSF IRB 20-31740). Samples were obtained after surgical excision with biopsies taken by pathology assistants to confirm the presence of tumour cells. Patients were selected without regard to previous treatment. Freshly resected samples were placed in ice-cold PBS or Leibovitz’s L-15 medium in a 50-ml conical tube and immediately transported to the laboratory for sample labelling and processing. Samples were used for either whole-tissue digestion into a single-cell suspension or a part of the tissue was sliced and preserved for imaging analysis.

Cell sorting, library preparation and sequencing

Cell sorting, library preparation, sequencing and bioinformatics data processing were performed as previously described³².

Computational analysis of sorted stromal RNA-seq

From the log-transformed matrix of normalized gene × cell expression values, we identified the 2,500 most variable genes based on their interquartile range and performed PCA in the space of these genes. We then used the 10 most strongly positively and negatively loading genes of PC1–PC6 for hierarchical clustering of samples and genes (complete linkage and Euclidean distance). The cluster dendrogram was split into k = 6 clusters based on tree height. We interpreted clusters of samples with high expression of EPCAM, KRT8 and KRT18 as epithelial-driven and TYROBP and CSF3R as myeloid-driven and excluded these samples from our subsequent analysis. Next, we performed PCA on the remaining samples. The loadings of the resulting PC space were then used to project the epithelial- and myeloid-driven samples onto PC1 and PC2.  Similarly, microdissected bulk RNA-seq samples from patients with PDAC as provided in ref. ³³ were obtained from the Gene Expression Omnibus database (identifier GSE93326) and projected onto PC1 and PC2.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.