A pan-family screen of nuclear receptors in immunocytes reveals ligand-dependent inflammasome control.

C57BL6/J (JAX strain # 000664), BALB/c (JAX strain # 001026), congenic CD45.1 strain (C57BL/6J background, JAX strain # 002014), Rosa26- Cas9 knock-in mice (JAX strain #026179) 92 , Foxp3 -IRES- Thy1.1 reporter mice 93 , Nlrp3 −/− mice (JAX strain # 021302) 94 , Casp1 −/− mice (JAX strain# 016621) 95 , Gsdmd −/− mice (JAX strain # 032410) 96 , Asc −/− mice (Millennium Pharmaceuticals) 97 , Gata6 flox/flox mice (JAX strain # 008196), LysM -cre (JAX strain # 004781) and Cd115 -ERcre mice (JAX strain # 019098)) were bred in our facilities at HMS, BCH, UMCMM or WashU. Rosa26-Cas9 knock-in mice were crossed with Foxp3-IRES-Thy1.1 reporter mice to generate Cas9/Foxp3 thy1.1 mouse line. Gata6 flox/flox mice were crossed to Lysm -cre (JAX strain # 004781) and Cd115 -ERcre mice (JAX strain # 019098)) to generate cell-specific GATA6 deficiencies mice 47 . To induce recombinase activity, Cd115-ERCre × Gata6 flox/flox × Rosa26TdTomato mice and Gata6 flox/flox × Rosa26TdTomato littermate controls were received 2 mg tamoxifen (100 μL of a 20 mg/mL stock solution, diluted in corn oil) by oral gavage every other day for a total of three doses. Mice were analyzed one week after the last dose. Mice were generally used for experiments between 6–8 weeks of age. Both male and female mice were used for experiments, after confirming no difference between sexes. All mice were bred and maintained in specific pathogen free conditions at Harvard Medical School. All experiments were performed following guidelines listed in animal HMS Institutional Animal Care and Use Committee protocols IS00000196 and IS00001257 and Washington University protocol 22–0433. C57BL/6 mice were bred with either a vitamin A deficient diet (Harlan, TD.09838) or control diet containing vitamin A (20000IU/kg diet in the TD.09838), beginning at 13.5 days of gestation. The pups were weaned at 3 week of age and maintained on the same diet until analysis was performed. 6- to 8-week-old cohoused C57BL/6 mice were used. The agonist and antagonist compounds were injected i.p., as specified in Table S6 . For LPS, mice were injected i.p. with 10 μg LPS (Sigma-Aldrich, L-2880) in 100 μl PBS (at 0.5 mg/kg weight). Controls from the same lots of mice were injected with PBS. Mice were euthanized with CO2 and perfused with 30 mL PBS. Spleen, liver, lung, epididymal VAT and colon samples were collected, minced, and digested at 37°C with shaking as follows: Splenic myeloid cells were released by enzymatic digestion in 5 mL phenol red-free Dulbecco’s Modified Eagle Medium (DMEM) (Gibco) containing 2% fetal calf serum (FCS),1 mg/mL Collagenase IV (Gibco) and 10 U/mL DNaseI (Sigma) for 20 minutes. Splenic lymphocytes were released using mechanical dissection. 8 mL sterile PBS containing 2% FCS and 2 mM EDTA was injected into the peritoneal cavity. A massage was applied to the abdomen prior to abdominal incision and collection of the fluid. LPMs were purified by sorting CD11b hi F4/80 hi cells using a BD-Aria sorter. To obtain thioglycolate elicited peritoneal macrophages (TGEMs), mice were injected i.p. with 2 mL 3% sterile thioglycolate. After 2 days, TGEMs were collected using peritoneal lavage. Liver and lung were digested in 25 mL RPMI containing 0.5 mg/mL Collagenase IV (Gibco) and 0.05 mg/mL DNase I (Sigma) for 40 minutes. The liver immune cells were subsequently purified through a 33% Percoll (GE Healthcare) gradient centrifugation using 30 mL of PBS containing 2% FCS, 15.3 mL of neat Percoll, and 1.7 mL of 10 × PBS for 10 minutes at 800× g. Epidydimal visceral adipose tissue (VAT) was digested in 5 mL DMEM containing 2% FCS and 1.5 mg/mL collagenase type II (Sigma) for 20 minutes. the entire colon was incubated in RPMI (Gibco) containing 2% FCS, 1 mM DTT and 20 mM EDTA at 37°C for 15 minutes to remove epithelial cells. The colon was then minced and dissociated in RPMI containing 1% FCS, 1.5 mg/mL collagenase II (Gibco), and 0.5 mg/mL Dispase (Gibco) at 37°C for 40 minutes, with constant stirring. MC38 cells were cultured in DMEM supplemented with 3 mM L-Glutamine and 100 U/mL penicillin-streptomycin. On day 20 after bone marrow reconstitution, each BMC mouse was subcutaneously inoculated on the abdomen with one million MC38 cells in 100 μL PBS. On day 17 after MC38 cell inoculation, solid tumor tissues were isolated, then chopped, and digested in DMEM containing 2% FCS, 1 mg/mL collagenase type IV (Gibco), and 20 μg/mL DNAse I (Sigma) at 37°C for 20 minutes with shaking. For all tissues, the digested materials were lysed using ACK lysis buffer (Gibco) to remove red blood cells and then filtered through a 40 μm cell strainer to get single cell suspensions. Cells were stained with different antibody panels as detailed in experimental descriptions or “Rainbow-CRISPR” screen section. For surface-antigen staining, single-cell suspensions were incubated in phosphate-buffered saline (PBS) containing live/dead stain and surface antibodies for 15 minutes on ice, followed by washes in 2%FCS DMEM buffer. For intracellular staining, cells were fixed in eBioscience Fix/Perm buffer for 30 minutes at room temperature, followed by permeabilization in eBioscience permeabilization buffer at room temperature for 60 minutes in the presence of intracellular antibodies. For Annexin V and propidium iodide (PI) staining, cells were incubated with 5 μL of PE- or APC- conjugated Annexin V (Biolegend) in 100 μL of Annexin V binding buffer (Biolegend) at room temperature for 10 minutes. Then 2 μg/mL of PI (Sigma) was then added to each sample and incubated for another 5 minutes. After that, 500 μL of Annexin V binding buffer was added to the cells for washing. The cells were either analyzed immediately or fixed in 1% formaldehyde for 10 minutes before flow cytometry analysis with BD FACSymphony or Cytek Aurora. The gRNA lentiviral vector was modified from pLKO.3G (Addgene # 14748) by introducing the U6 promoter-filler-gRNA scaffold-cPPT-PGK promoter fragment. EGFP was replaced with either BFP or mCherry, to generate different fluorescent reporters. The gRNAs (listed in Table S1A ) were designed using CRISPick software 98 , 99 from the Broad Institute. Three gRNAs were selected for each NR. Non-targeting control gRNAs were picked from the GeCKO v2 mouse library 100 . The intergenic control gRNAs targeting murine safe harbor genome locations were picked from the genome-wide gRNA library 101 . All guide RNA oligos with BsmBI-compatible overhangs were purchased from IDT, and annealed, phosphorylated, and cloned into the gRNA lentiviral vector by BsmBI restriction digestion as described 100 . Lentivirus was packaged using HEK293T cells (TaKaRa). Briefly, cells are split and seeded into a 6-well plate to reach ~80% confluence before transfection. Three gRNA expressing lentiviral plasmids targeting the same NR were mixed equally and co-transfected with package plasmids psPAX2 and VSVG at a ratio of 2:1:1 using TransIT-293 reagent (Mirus Bio, MIR2704) following the manufacturer’s instructions. The virus-containing supernatant was collected 48 hours after transfection and filtered through a 0.45 μm syringe filter. Lentivirus was concentrated using PEG-it lentivirus concentration reagent (System Biosciences) following the manufacturer’s instructions. Virus titers were quantified by infecting HEK293T cells with serial dilutions of the virus. Infected cells were detected by recording BFP-, EGFP-, or mCherry-positive cells using flow cytometry. The viral titer was then calculated using the formula: Titer in pfu/mL = ((# of cells at start time) * (dilution factor) * (percent of infection)) / (volume of virus solution added). An infection rate of 1~10% was used. Bone marrow cells were isolated from the tibia and femur of the Cas9/Foxp3 thy1.1 mice, flushed and ACK-lysed. After depleting the Lineage + populations with biotinylated antibodies (anti-CD4, CD8, CD11b, CD11c, CD19, Gr1, NK1.1, Ter119) and Dynabeads Biotin Binder (Thermo Fisher Scientific) following the manufacturer’s instructions, LSK cells (Lineage − c-Kit + Sca-1 + ) were sorted using Aria 561 sorter. The purified LSK cells were cultured overnight in StemPro-34 SFM (Gibco) complete medium containing 1x Supplement, 2mM L-glutamine and 100 ng/mL recombinant murine cytokines (e.g., SCF, Flt3 ligand, IL-7, and TPO (all from PeproTech)). The following day, the LSK cells were transduced with different gRNA lentiviruses at MOI=20 through spin-transduction on a RetroNectin-coated plate at 650x g for 90 minutes. In the “Rainbow-CRISPR” screen, each bone marrow chimera (BMC) mouse received a mixture of three LSK populations, identifiable by different fluorescent reporters (BFP, EGFP, and mCherry). One was transduced with a control - targeting gRNA virus and served as the internal control, while the other two populations were transduced with gRNA lentiviruses targeting a given NR. In practice, each targeting actually used simultaneously 3 different gRNAs against the same gene, with same fluorescent reporter. In addition, to guard against an effect of the double-stranded breaks created in the HSCs, we used gRNAs inducing breaks in “safe harbor” loci. The three LSK populations were mixed equally and transferred into irradiated CD45.1 + recipients through retro-orbital injection 24 hours post virus transduction. After allowing for reconstitution and differentiation to proceed (4 weeks for myeloid cells, 10 weeks for T and B cells), cells were isolated from spleen and peritoneal lavage, prepped, and stained for flow analysis. Single cell suspensions were stained with four constant panels of antibodies: 1) splenic myeloid panel contains antibodies against CD45.2, CD19, CD11b, CD11c, F4/80, MHCII, Ly6C, Ly6G, CD8a and PDCA1; 2) peritoneal myeloid panel contains antibodies against CD45.2, CD19, CD11b, F4/80,Ly6C, Ly6G and Siglec-F; 3) splenic B cell panel contains antibodies against CD45.2,CD19, TCRβ, IgM, IgD, CD23, CD93, CD21/35, CD138, Fas and CD38; 4) splenic T cell panel contains antibodies against CD45.2, CD19, TCRβ, TCRγδ, CD4, CD8a, CD90.1 (thy1.1), CD44 and CD62L (see detailed gating strategies in Figs. S2B – E ). To maintain data consistency, the antibodies were kept at a constant concentration, clone, and source. The gRNA frequencies in each immune cell-type were quantified by flow analysis of BFP/GFP/mcherry expression using Cytek Aurora. Screen data were obtained from at least two independent experiments, each using a distinct batch of gRNA lentiviruses and BMC mice. The entire set of data were analyzed by FlowJo Software. Data with strong discrepancies in staining caused by reagent drift (e.g., enzymes and antibodies, etc.), or with low quality such as those with a low number of plasma B cells, or those that involved unclear gating of fluorescent protein in γδT cells, were excluded from the analysis. A total of 35 NRs across 28 cell types were included in the data set, all of which exhibited high-quality flow data. Data processing involved one or more normalization steps, to correct for the inherent variability in such experiments (variable organ sizes and cell counts in BMC mice, different levels of engraftment, fluctuations in lentiviral titers), exploiting the control-gRNA-transduced cells differentiating in the same mouse (as ratios of the fluorescence reporters carried by the lentiviruses), as well as the relative frequencies of differentiated cell edited by a particular gRNA relative to the whole population carrying this guide in the same mouse (for instance, normalizing the ratio relative to Ctrl observed in Treg cells to the same ratio in total T cells in the same mouse). For Figure 1B , the NR/Ctrl ratio in total donor-derived CD45 + cells was calculated (frequency of cells carrying targeting gRNA vs frequency of cells carrying control gRNA in the same mouse) was calculated (see Table S1D for details), with significance tested by a one-sample t test (vs null hypothesis with this ratio= 1). In other panels, which analyze the effect of NRs on differentiated cells, the NR/Ctrl ratio of each immunological cell type was normalized to the NR/Ctrl ratio of the total donor-derived CD45 + cells carrying the same gRNA, or all B, or all αβT cells in the same BMC host. Changes in cell types that constitute a dominant proportion of cells within a compartment may artefactually induce apparent changes in other cell types (in particular, changes in large peritoneal macrophages alter the proportion of peritoneal B cells). To circumvent this issue, we normalized the NR/Ctrl ratio of peritoneal cell types to the NR/Ctrl ratio of total donor-derived CD45 + splenocytes in the same BMC host (see Table S1E for details). Statistical analysis was assessed by One-way ANOVA with Dunnett correction. The efficacy of gRNA editing was assessed by amplicon sequencing around the gRNA target sites, in both the HSC population prior to transfer and in recovered B lymphocytes, by pooled amplicon sequencing of target regions using the Illumina MiSeq platform. Briefly, on day 3 following lentivirus transduction, LSK cells with NR mutations were sorted. B cells with NR mutations were sorted from BMC mice at 10 weeks after reconstitution. Genomic DNA was extracted from the sorted cells using Arcturus PicoPure DNA Extraction Kit (ThermoFisher Scientific) and amplicon libraries were prepared by employing a 2-step PCR method using NEBNext High-Fidelity 2x PCR Master Mix (New England Biolabs). Locus-specific primers flanking the modification site and containing universal 5′ tails with partial Illumina adapters were designed (primers listed in Table S1B ). The modification site should be close to the center of the amplicon fragment and ideally at least 50 bp away from the primer-binding sites. In step1 PCR, locus-specific primers are used to amplify the genomic regions containing modification sites. In step2 PCR, the amplicons from step 1 were introduced with the remaining Illumina sequencing adaptor and sample barcodes (primers listed in Table S1B ). Samples were prepared for Illumina MiSeq according to the manufacturer’s instructions with unique barcode for each cell type. Data were analyzed using CRISPResso2 with the default parameters as previously described 102 . CRISPResso2 aligns sequencing reads to a reference sequence and quantifies insertions, mutations, and deletions to determine whether a read is modified or unmodified by genome editing. Editing frequency was calculated as the percentage of reads containing a frame-shift indel out of total reads at each target site (see Table S1C ). The code available at https://github.com/CBDM-Lab/NR-screen-Homozygote-Frequency was used to calculate the frequency of cells carrying one or more homozygous knockout mutation of the target gene. RNA-seq was performed on purified peritoneal MFs sorted from “Rainbow-CRISPR” mice 4 weeks after reconstitution, each mouse yielding two NR mutant samples and one control (for economy, only 12 Ctrl samples were profiled overall). Biological duplicates were obtained for each NR ablation, sorted on different days to reduce batch effects. The standard ImmGen ULI-RNA-seq protocol ( http://www.immgen.org/ ), on 1,000 double-sorted cells – second sort directly into 5 μL TCL (Qiagen) lysis buffer containing 1% 2-mercaptoenthonol (Sigma). NR-deficient peritoneal MFs were sorted by their corresponding gRNA-conjugated fluorescence reporter, as live CD19 − CD11b hi F4/80 hi . Multiple wild-type samples were sorted and paired with NR-deficient cells on the same day. Smart-seq2 libraries were prepared by Broad Technology Labs following previously described protocol with slight modifications 103 , 104 . Briefly, total RNA was captured and purified using RNAClean XP beads (Beckman Coulter). Polyadenylated mRNA was then selected using an anchored oligo (dT) primer (50–AAGCAGTGGTATCAACGCAGAGTACT30VN-30) and converted to cDNA via reverse transcription. First-strand cDNA was subjected to limited PCR amplification followed by transposon-based fragmentation using the Nextera XT DNA Library Preparation Kit (Illumina). Samples were then PCR amplified for 12 cycles using barcoded primers such that each sample carries a specific combination of eight base Illumina P5 and P7 barcodes and pooled together prior to sequencing. Smart-seq paired-end sequencing is performed on an Illumina NextSeq500 (two full NextSeq runs per batch of 96 samples, for 10 million raw reads/sample on average) using 2 × 38 bp reads with no further trimming. Reads were aligned to the mouse genome (GENCODE GRCm38/mm10 primary assembly and gene annotations vM25; https://www.gencodegenes.org/mouse/release_M25.html ) with STAR 2.7.3a 105 The ribosomal RNA gene annotations were removed from GTF (General Transfer Format) file. The gene-level quantification was calculated by featureCounts ( http://subread.sourceforge.net/ ). Raw reads count tables were normalized by median of ratios method with DESeq2 package from Bioconductor 106 and then converted to .gct and .cls format. Samples with less than 1 million uniquely mapped reads, or having less than 8,000 genes with >10 reads, or with transcript integrity number (TIN) < 45 were removed from the data set prior to downstream analysis and excluded from normalization to mitigate the effect of poor-quality samples on normalized counts. In addition, biological replicates were analyzed for Pearson correlation to identify poor-quality samples and remove them from the data set. Pearson correlation was calculated on transcripts with an average of > 5 reads or below the 99th percentile for number of reads in the dataset to avoid outlier effects. Any replicates that did not exhibit a correlation of 0.9 or greater were removed from the data set prior to downstream analysis. Finally, the RNA integrity for all samples was measured by median transcript integrity number (TIN) across mouse housekeeping genes with RSeQC (‘tin.py’, v2.6.4) 107 . To eliminate noise from transcripts with low expression levels, genes were retained for consideration if they had an expression >20 in at least one condition. Some genes yield intrinsically high variance even in wild-type dataset. These “noisy” genes were removed from consideration if their 90 th percentile of inter-replicate coefficient of variation across all the mutants was > 0.75. A FoldChange matrix was generated for each mutant sample relative to the Ctrl samples prepared on the same sorting date. A 1-sample t-test was performed to determine whether FCs for each NR mutant are different from 1. Potential differentially expressed (pDE) genes for each NR mutant were identified by FC 2 and nominal t.test p-value < 0.05. The final set of pDE genes was the union of all the pDE genes from each NR mutant. The expression heatmap was generated with Morpheus ( https://software.broadinstitute.org/morpheus ). Information on BMC mice and corresponding hashtags was detailed in Table S4A . Peritoneal exudate cells extracted from each BMC mouse were stained in 100 μL of FACS buffer (phenol red-free DMEM, 2% FCS, 0.1% azide and 10 mM HEPES, pH 7.9) containing 10 μL FcBlock (homemade), CD11b APC (1 μL, M1/70, BioLegend), and CD115 PE/Cy7 (1 μL, AFS98, BioLegend), 1 μL (0.5 μg) of a unique hashtag antibody for 10 minutes. Wild-type and four different NR mutants, including RARγ mutant, RXRα mutant, RORα mutant, and RXRβ mutant were sorted by their corresponding gRNA-conjugated fluorescence reporter, as live CD11b + CD115 + cells into PBS–BSA 0.1% medium using the Aria561 cell sorter (70-μm nozzle). Each NR mutant has two replicates, while three wild-type samples were paired with NR mutant cells from the same BMC mouse. Samples of the same genotype were combined and hashtagged again. The multiple hash-tagged samples (46,000 cells in total) were then combined into one tube with a final volume of 30 μL PBS-BSA 0.1%. The single-cell RNA-Seq libraries were prepared using the 10X Genomics Single Cell 3′ Reagent kit (V3 chemistry) following the manufacturer’s protocols. Hashtag libraries were made separately as described 108 . In brief, at the cDNA amplification step in the Single Cell 3′ Reagent kit protocol, the yield of HTO (Hashtag Oligo) products was increased using an ‘additive’ primer to cDNA PCR. Hashtag-derived cDNAs (300 bp) were then separated using 0.6× SPRI bead selection. The supernatant contained the hashtag-derived cDNA that was purified with two rounds of 2× SPRI beads. The sequencing oligonucleotides were added by PCR which also amplified the Hashtag library. Libraries were sequenced on the NextSeq 500 platform (28/8/0/91, Read1/i7/i5/Read2). Hashtag count matrices were obtained from CITE-Seq-Count package ( https://zenodo.org/record/2590196 ). scRNA-seq data were processed using the standard CellRanger pipeline (10x Genomics). HTO counts were obtained using the CITE-seq-Count package. Processed matrix, barcodes and feature files were loaded in R (v4.3.1) using Seurat (v5.0.1). Hashing data were used to classify each cell based on the predominant dual hashtag configuration. Cells with fewer than 500 reads or with greater than 10% of reads mapped to mitochondrial genes were excluded from the analysis. Dimensionality reduction, visualization, and clustering analysis were performed in Seurat using the NormalizeData, ScaleData, FindVariableGenes, RunPCA, FindNeighbours, RunUMAP (dims = 1:15), and FindClusters (res = 0.7) functions. The method returned 12 cell clusters which were then visualized using UMAP for dimensionality reduction. Clusters 11 (T cells) and 12 (B cells) where minor contaminants that were removed using the SubsetData function. Cluster identity was determined on the basis of expression of key marker genes ( Figs.4B and S4B ). DE genes in each cluster were determined by FindAllMarkers and FindMarkers. ChIPmentation libraries were prepared following the protocol as described 109 with some modifications. In brief, approximately 100,000 LPMs per sample were sorted from peritoneal exudate cells of 8-week-old C57BL/6 mice. Briefly, cells were fixed with 1% fresh formaldehyde (Thermo Fisher Scientific) for 10 minutes at room temperature with rotation, followed by quenching using 0.125 M glycine for 5 minutes. After washing with ice-cold PBS supplemented with cOmplete protease inhibitor (Roche), cells were resuspended in lysis buffer (10 mM Tris-HCl pH 8.0, 2 mM EDTA, 0.25% SDS, 1x protease inhibitor), and sonicated on the Covaris M220 ultra-sonicator to obtain 200–300 bp DNA fragments using the following settings: peak incident power 50, duty factor 10%, cycles per burst 200, time 8 minutes. Keep 2 μL of supernatant from two replicates as the input DNA at -20 °C. The remaining supernatant was incubated overnight at 4°C on a rotator with 3 μg of anti-RXRα (D6H10) Rabbit mAb (#3085, Cell Signaling) pre-coupled to Protein A Dynabeads (Invitrogen). The beads were washed twice sequentially on a pre-cold magnet (Invitrogen) using RIPA-LS wash buffer (10mM Tris-HCl pH 8.0, 140 mM NaCl, 1 mM EDTA, 0.1% SDS, 0.1% sodium deoxycholate, 1% Triton X-100), RIPA-HS wash buffer (10mM Tris-HCl pH 8.0, 500 mM NaCl, 1 mM EDTA, 0.1% SDS, 0.1% sodium deoxycholate, 1% Triton X-100), and RIPA-LiCl wash buffer (10 mM Tris-HCl pH 8.0, 250 mM LiCl, 1mM EDTA, 0.5% NP-40, 0.5% sodium deoxycholate) and followed by a wash with 10 mM Tris-HCl pH 8.0. After washing, the beads were then resuspended in 25 μL tagmentation reaction mix containing 1 μL Tn5 transposase (Illumina), and incubated at 37°C for 5 minutes. After removing the tagmentation reaction mix, beads were washed twice with RIPA-LS wash buffer and once with TE buffer (10 mM Tris-HCl pH 8.0, 500 mM EDTA), followed by elution with 48 μL elution buffer (10 mM Tris-HCl pH 8.0, 300 mM NaCl, 5 mM EDTA, 0.4% SDS) containing 2 μL Proteinase K (NEB) at 55°C for 1 hour and 65°C for 8 hours for de-crosslinking. The input DNA was also incubated with tagmentation reaction mix, and de-crosslinked with Proteinase K. DNA was purified using MinElute Reaction Cleanup Kit (QIAGEN). Libraries were indexed and amplified using the NEBNext High-Fidelity 2x PCR Master Mix (NEB) and ATAC-seq primers 110 , followed by purification using AMPure XP beads (BeckmanCoulter) and sequencing as for ATAC-seq. After removing adapters and low-quality reads using Trim Galore ( https://github.com/FelixKrueger/TrimGalore ), we aligned ChIP-seq reads to the corresponding genome (mm10 reference genome) by Bowtie 2 111 with the following parameters: -X 1000 --fr --no-mixed --no-discordant. Nonuniquely mapped and mitochondrial DNA reads were removed using a combination of SAMtools functions 112 . PCR duplicates were removed using Picard (‘MarkDuplicates’, v2.8.0, https://broadinstitute.github.io/picard/ ). ChIP-seq peaks for individual samples were identified using MACS2 113 (‘callpeak’, v2.1.1.20160309) with the following parameters: --keep-dup all --nomodel --shift -100 --extsize 200 -p 0.05. High-confidence, reproducible ChIP-seq peaks among replicates were then identified with a global FDR < 0.01. Peaks overlapping the suspect list of problematic regions 114 were removed from downstream analyses using BEDTools (v2.27.1) 115 To visualize individual ChIP-seq tracks using the Integrative Genomics Viewer (IGV, v2.4.14) ) 116 , the alignment file (BAM file) was converted to the read-coverage file (BigWig file) using deepTools 117 (‘bamCoverage’, v3.0.2). The peak annotation was performed using HOMER 118 (‘annotatePeaks.pl’, v4.9) (binding peaks listed in Table S6 ). Motif enrichment analysis for RXRα binding peaks were performed using findMotifsGenome.pl with the default parameters from HOMER. The FACS-purified LPMs were resuspended in DMEM F12 medium (Gibco) supplemented with 10% FBS, 3 mM L-Glutamine and 100 U/mL penicillin-streptomycin and seeded in 24-well plate. After cell adhesion for 1 hour, cells were washed twice with PBS and stimulated with indicated stimuli in Opti-MEM (Gibco) specified in the figure legends. To evaluate the effect of RARγ antagonist LY2955303 on cell death programs, LPMs were left untreated or treated with 50 μM LY2955303 (Tocris) alone, or in combination with pretreatment of 200 μM Lonidamine (Tocris, LND, ASC inhibitor), 25 μM VX765 (Tocris, caspase-1 inhibitor), or 2 μg/μL of the cOmplete protease inhibitor cocktail (Roche) for 1 hour prior to LY treatment. To assess the effect of RARγ agonist BMS961 on inflammasome activation, LPMs were either left untreated or treated with 30 μM BMS961 (Tocris) for 1 hour prior to exposure to inflammasome ligands. For NLRP3 inflammasome, B6 LPMs were primed with 500 ng/mL LPS (Sigma-Aldrich, L-2880) for 3 hours followed by stimulation with 10 mM nigericin or 5 mM ATP for 30 minutes. For AIM2 inflammasome, LPS-primed B6 LPMs were transfected with 2 μg poly(dA:dT) for 6 hours using Lipofectamine 3000 (Invitrogen, L3000001). For NLRC4 inflammasome, LPS-primed B6 LPMs were infected for two hours with wildtype Shigella 2457T which was grown as previously described 119 . Briefly, an overnight culture of Shigella 2457T was back-diluted into 5 mL of TCS (trypticase soy) broth and incubated at 37°C for 2 hours with shaking. The culture was pelleted and resuspended in Opti-MEM and spun onto cells for 10 minutes at 300xg at an MOI of 30. Infected cells were incubated at 37°C in a 5% CO 2 incubator for 20 min and then washed for three times with Opti-MEM containing 50 μg/mL gentamicin, then returned to 37°C for further incubation for 100 minutes. For NLRP1b inflammasome, LPMs from BALB/c mice that express NLRP1b were treated with different concentrations of anthrax lethal toxin (LeTx) for three hours. The low dose consisted of 2 μg/mL protective antigen (PA) and 1 μg/mL lethal factor (LF), while the high dose comprised 10 μg/mL PA and 10 μg/mL LF. Cells were pretreated and maintained in vehicle or 30 μM BMS961 throughout the experiment. After cell treatment, IL-1β secretion and LDH secretion was measured in the supernatants by Lumit ™ IL-1β mouse immunoassay kit (W7010, Promega) and LDH-Glo ™ Cytotoxicity Assay (J2380, Promega) according to the manufacturer’s instructions. The cells together with the culture supernatants (in Opti-MEM) were lysed with 5 × SDS for immunoblotting analysis. After cell treatment, the cells and culture supernatants were lysed in 5x SDS sample loading buffer containing 50% glycerol, 10% SDS, 5% 2-mercaptoethanol, 0.02% bromophenol blue, 250 mM (pH 6.8) Tris-HCl, and cOmplete protease inhibitors (Roche). The proteins were separated by 8–12% SDS-PAGE and electrophoretically transferred to PVDF membranes (Millipore IPVH00010). After blocking non-specific binding with 5% skim milk, the membranes were incubated overnight with indicated primary antibodies: anti-caspase-1 (AG-20B-0044, AdipoGen), anti-caspase-3 (#9662,CST), anti-cleaved caspase-3 (#9661, CST), anti-caspase-7 (sc-56063,Santacruz),anti-caspase-8(sc-81656,Santacruz),anti-GSDMD (sc-393581, Santacruz), anti-GSDMD (ab209845, Abcam), anti-GATA6 (sc-518050, Santacruz), anti-PARP (# 9542,CST), anti-ASC(#67824, CST), anti-NLRP3 (AG-20B-0014-C100, AdipoGen) and anti-GAPDH (ab9484, Abcam). Membranes were then washed and probed with corresponding horseradish peroxidase (HRP)–conjugated secondary antibodies (from Jackson ImmunoResearch Laboratories): donkey anti-mouse IgG (H+L) (#715-035-150), goat anti-rabbit IgG (H+L) (#111-035-144), and goat anti-mouse IgG, light chain specific (#115-035-174). Immunoblot images were acquired by Image Lab software (Bio-Rad) using SuperSignal ™ West Femto Maximum Sensitivity Substrate (#34094, Thermo Fisher Scientific). For endogenous immunoprecipitation, C57BL/6 mice were i.p. injected with either vehicle or 50 nmoles LY2955303 for 10 minutes (n=5 mice per group). Total peritoneal cells were collected and resuspended in 1mL ice-cold lysis buffer (20 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% Triton X-100, 10% glycerol, 1 mM Na 3 VO 4 , 2 mM PMSF, EDTA-free protease inhibitor cocktail) and incubated on a rocker at 4 °C for 1 hour. After centrifugation at 20,000 × g, 4 °C for 10 minutes, collect 50 μL lysate as the whole cell lysate. The remaining lysates were incubated overnight at 4°C on a rotator with 3 μg IgG control antibody (#2729, Cell Signaling), or anti-RARγ antibody (11424–1-AP, Proteintech) or anti-RARγ (sc-7387, Santacruz) antibody pre-coupled to Protein A Dynabeads (Invitrogen). Subsequently, the beads were then washed three times with lysis buffer and boiled in 2 x SDS loading buffer at 100 °C for 5 minutes. For immunoprecipitation in the overexpression system, HEK293T cells were seeded into six-well plates and transfected with the indicated combination of pCDNA3.1-Flag-mouse-RARγ, pMSCV-IRES-GFP-mouse-Caspase-1 (#183361, Addgene), or pcDNA3-HA-mouse-ASC plasmids for 24 hours. Cells were then collected and lysed in an ice-cold lysis buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1mM EDTA, 1% Triton X-100, 10% glycerol, 0.02% digitonin, 2 mM PMSF, protease inhibitor cocktail). The lysates were incubated overnight with prewashed Anti-FLAG ® M2 Magnetic Beads (M8823, Sigma-Aldrich). Subsequently, the beads were then washed three times with lysis buffer and boiled in 2 × SDS loading buffer at 100 °C for 5 minutes. The immunoprecipitated and input samples were subjected to immunoblotting analysis and stained with indicated antibodies (as listed in Table S7 ). The FACS-purified LPMs were resuspended in 150 μL of ice-cold lysis buffer (10 mM Tris-HCl, pH 7.5, 10 mM NaCl, 3 mM MgCl 2 , 0.1% NP-40, 0.1% Tween-20, 0.01% Digitonin) and incubated on ice for 5 minutes with periodic vortexing. Then, add 1 mL of wash buffer (10mM Tris-HCl, pH 7.5, 10 mM NaCl, 3 mM MgCl 2 , 0.1% Tween-20) and gently invert the tube for three times. Centrifuge the mixture at 500 × g for 10 minutes at 4°C. Collect the supernatant as the cytoplasmic fraction and wash the nuclei pellet with 1 mL PBS. Boil the cytoplasmic and nuclear protein with SDS sample loading buffer and proceed with immunoblotting analysis. ASC pyroptosome was purified using low-speed centrifugation and detected by disuccinimidyl suberate (DSS) cross-linking and subsequent immunoblotting for ASC oligomers as previously described 120 . Briefly, 4 million LPMs were pretreated with either vehicle or 30 μM BMS961 for 1 hour and primed with 500 ng/mL LPS for 3 hours, followed by 5 mM ATP stimulation for 30 minutes. Cells were washed once with ice-cold PBS and scraped off in 400 μL ice-cold lysis buffer (1% NP-40, 0.1 mM PMSF, cOmplete protease inhibitor in PBS). The cell lysate was fully disrupted by passing it through a 21-G needle for ten times on ice and then centrifuged at 250 × g for 5 minutes at 4°C to remove the nucleus and cell debris. The protein concentration in each sample was quantified using the BCA protein assay. An equal amount of total protein was used to pellet the ASC pyroptosome by centrifuging at 5000 × g for 10 minutes at 4°C. The pellet was then resuspended in 300 μL of lysis buffer containing 2 mM fresh DSS and crosslinked for 30 minutes at room temperature. The reaction was quenched by adding 6 μL of 1 M Tris-HCl for 15 minutes. The cross-linked pellet was centrifuged at 5000 ×g for 5 minutes and resuspended in SDS sample loading buffer. Immunoblotting was performed to detect ASC oligomers in both soluble and pellet samples, with and without DSS crosslinking. Statistical analyses, excluding the single-cell and population RNAseq data, were conducted using GraphPad Prism software. Data were shown as mean ± SD. Statistical significance was determined using two-tailed t test (paired or unpaired as indicated) for two groups, one-way ANOVA with Dunnett correction for three or more groups, and a Chi-square test for determining gene signature significance in volcano plots. Additionally, a one-sample t-test with a hypothetical value of 1 was used for Figs. 1B , 1D (top panel) and 1H . Statistical significance was defined as P < 0.05. The level of significance in all graphs is represented as follows: * for P < 0.05, ** for P < 0.01, *** for P < 0.001, and **** for P < 0.0001.

We set up an in vivo “ Rainbow-CRISPR” screen to comprehensively investigate how the NR family affects the differentiation and homeostatic maintenance of immunologic cell-types ( Fig.1A ). We targeted all 35 NR genes expressed in at least some immunocytes ( Fig.S1A , ImmGen data). CRISPR editing was achieved by infecting hematopoietic stem cells (HSCs) from Cas9-expressing mice with gRNA expressing lentiviral vectors. The cells were then used to reconstitute all immunologic lineages of lethally irradiated recipients ( Fig.1A ). In the Rainbow strategy, each bone marrow chimera (BMC) mouse is internally controlled, receiving of mix of three HSC populations, each identifiable with different fluorescent reporters: two are transduced with lentiviruses targeting a given gene, one with a non-targeting control ( Table S1A in practice, each targeting actually pooled 3 different gRNAs against the same gene, coded with the same fluorescent reporter). This configuration enabled a direct comparison of wild-type (WT) and mutant cells differentiated in the same environment. In addition, to guard against an effect of the double-stranded breaks created in the HSCs, we used gRNAs inducing breaks in “safe harbor” loci. After allowing reconstitution and differentiation to proceed (4 weeks for myeloid cells, 10 weeks for lymphocytes), the effect of NR inactivation was assessed by flow cytometry, as the relative abundance of progeny cells identified by the three fluorescent reporters in each of 28 innate or adaptive immunocytes (gating strategies in Figs.S1B – E ). This strategy was preferred to sequencing-based identification as it required far fewer cells for robust quantification, hence was better adapted to assay small cell subsets. Editing efficacy was verified by amplicon sequencing around the gRNA target sites, in both the HSC population prior to transfer and in recovered B lymphocytes. Only a few gRNAs seemed ineffective, with 37 to 97% (95% interval) of chromosomes showing a mutation for each gRNA (median 68%; Fig.S1F , Table S1C ). Given that each targeting involved 3 independent gRNAs, we infer that 41–99% of HSCs (median 81%) carried homozygous deletion mutations of the target gene (with the exception of Nr4a1 ). To uncover broad effects at the level of HSCs, we first analyzed the relative proportions of total CD45 + donor-derived cells derived from NR- or control-targeted HSCs within the same mouse. The results ( Fig.1B , Table S1D ) denoted several impacts of NR ablation on these pan-lineage readouts, although some NRs appeared neutral (ESR1, NR2F6 or THRα; note, however, a trend toward lower frequencies of cells carrying safe-harbor gRNAs, suggesting a slight effect of double-stranded breaks in HSCs on later differentiation). We surmise that these frequent reductions reflect, at least in part, the pleiotropic roles of many NRs in HSCs 36 . We then addressed our primary goal, the role of NRs in differentiated immunocytes, factoring out fluctuations in global reconstitution efficiency by normalizing frequencies of edited progeny in each cell-type against relative frequencies in total edited CD45 + cells in the same mouse host. Overall, a number of NR involvements were detected, where the deletion of one or several NRs led to increased or decreased proportions of specific immunocytes, as portrayed in Fig.1C (detailed in Figs. 1D – H , S2A – F , Table S1E – F ). Among the most noteworthy: B cells as a whole were affected by the deletion of several NRs ( Fig.1D , top), with the strongest dependence on RXRβ, RARα and PPARγ. Narrowing in on the more differentiated states showed that mutation of these same NRs relatively increased Marginal Zone (MZ) and Germinal Center (GC) B cell differentiation ( Figs.1D , lower rows, S2A – B ), consistent with the complex positive and negative roles of retinoids in B cells 39 . In contrast, the requirement for the orphan receptor NR2C2 was more specific to B.GC cells. Within αβT lymphocytes ( Figs.S2C – D ), RORγ had the strongest impact, which was a positive control of sorts given the known requirement for RORγ in immature double-positive pre-T cells 20 . We also noted more modest effects of NR4A3 and of NR1H2 (LXRβ) on the abundance of αβT cells in the spleen ( Fig.S2D ), aligning with their well-established functions in thymic T cell differentiation 38 , 40 . We saw no apparent impact of LXRβ ablation on Treg proportions and T cell activation ( Fig.S2E ). In Treg cells ( Figs.1E , S2F ), RORγ deletion elicited an unexpected drop of FoxP3 + Treg cells, which was not merely due to the general dearth of T cells, since the results were normalized against total αβT cells. We also identified effects of NR4A3, in line with the reported roles of NR4A family members in Tregs, although somewhat unexpected given the reported redundancy among NR4A family members 26 , 27 . The NR2C2 deficiency also marginally affected Treg numbers ( Fig.S2F ). In granulocytes, previous studies have shown that RARα functions as a bidirectional regulator of neutrophil differentiation 41 . Here, RARα appeared to be a negative regulator of spleen neutrophils, but RARβ seemed to positively influence their numbers ( Fig. 1F ). Peritoneal macrophages showed the strongest effect of any deletion, with the very deep (10–30 fold) decrease in LPMs after elimination of either of the retinoic acid receptor RARγ or RXRα ( Fig.1G ). Elimination of RXRα also impacted the population of small peritoneal macrophages (SPMs) that arise and maintain distinctly from LPMs. This effect was specific, in that no other myeloid lineage was similarly affected ( Fig.1H ). In contrast to SPMs, LPMs are known to be influenced by retinoic acid and its derivatives 29 , 42 , and the present data anchor this relationship, through mechanisms that will be detailed below. This screen thus provided a broad perspective on the roles of NRs in the differentiation and homeostasis of various innate and adaptive cell-types. As a class, the retinoic acid receptors were the most often involved, in keeping with the pleiotropic actions of RA in the immune system 43 . Other NRs, those binding specific hormone families, were less frequently implicated – surprisingly including NR3C1 (GR), the conduit for major anti-inflammatory compounds, which showed no detectable effect on levels of any cell-type. For a mechanistic understanding of these effects, we focused on the macrophages (MFs) that manifested the strongest effects in the screen. We performed gene expression profiling to better understand the very similar consequences of RXRα and RARγ ablation on LPM numbers, and to identify other NRs expressed in LPMs that might influence their phenotypes but not their numbers. Therefore, NR-deficient peritoneal MFs (CD11b hi F4/80 hi ) purified from BMC mice 4 weeks after reconstitution were used for low-input RNAseq 44 ( Figs.2A – B , Table S2 ). Perhaps surprisingly, most NRs showed little effect ( Fig.2A ), with only mild imprints by NR1H3 and NR2C2 deficiencies (we are unsure of the NR2C2-related FoldChanges (FC), because the cluster of genes affected reproducibly by NR2C2 ablation had high inter-replicate variability in samples with ablation of other NRs, suggesting non-specific noise). But RXRα and RARγ mutants dominated these changes, with two major gene clusters up- or down-regulated in concert by the two mutations, and two smaller clusters more strongly affected by RXRα deficiency (also evidenced in the FC/FC plots of Fig.2C ). These dominant clusters included genes overexpressed in LPMs relative to other macrophages in previous studies 29 , 45 , 46 and from our single-cell RNAseq data (per below), and genes activated or repressed by GATA6 as deduced from previous ablation studies 29 , 47 , 48 ( Fig.2B , right). Gata6 itself was downregulated in both RXRα- and RARγ-deficient MFs ( Fig.2C ). These effects fit with the loss of LPMs in response to RXRα- or RARγ deficiency ( Figs. 1G , H ) that would proportionally increase in SPMs, but also pointed to quantitative variations between the two. RXRα and RARγ deficiencies appeared to have very similar consequences on peritoneal MFs, in line with vitamin A’s role in regulating LPM identity 29 and with RXR’s importance in serous cavity macrophages 30 . We asked whether the strong dependence on RXRα and RARγ was a general property of all macrophages. Tissue resident macrophages (TRMs) vary across organs, adapting to unique microenvironments, exemplified by alveolar macrophages in the lung and Kupffer cells in the liver 15 , 49 – 51 . LPMs belong to the “cavity macrophage” category, contributing to rapid responses to barrier breach or organ damage 52 – 54 . All TRMs express RA receptors, albeit with different patterns 55 ( Fig.S3A ), and virtually all were affected by the RA deficiency caused by a vitamin-A deficient diet ( Fig.S3B ), expanding on previous reports 29 , 56 , 57 . Prompted by these results, we extended the CRISPR screen to systematically assess in Rainbow BMC mice the impact of RA receptors on different donor-derived TRMs, including macrophages that infiltrate the transplanted MC38 tumors ( Figs.3A – B , Table S3 , gating strategies in Fig.S3C ). The deep impact of RARγ deficiency was unique to LPMs, while the loss of RXRα impacted most other TRM populations, except colon or visceral adipose tissue (VAT) macrophages, in line with previous reports that RXRα governs the differentiation of yolk sac or erythromyeloid progenitor-derived TRMs 58 . Generally, the data revealed a patchwork of dependencies that varied between TRMs ( Fig.3B ). RARβ and RXRγ were not required by any TRMs, while RARα and RXRβ appeared to have inhibitory effects on several TRMs, again with a different distribution (RXRβ most inhibitory to alveolar MFs, RARα to red pulp MFs). Thus, retinoic acid controls the differentiation and homeostasis of TRMs through distinct receptors. GATA6 + macrophages with similar transcriptional identities are found in serous cavities like peritoneal, pleural and pericardial spaces 52 . To verify whether the unique dependency on RARγ applied to the GATA6 + macrophage class more generally, we analyzed GATA6 + macrophages in the pleural cavity of BMC mice repopulated with RARγ- and RXRα- deficient HSCs ( Fig.3C ). The same deep drop was observed as in LPMs, indicating that this unique dual dependency is a general property of GATA6 + cavity macrophages. To examine more closely the varied effects of RA receptor deletions on peritoneal MFs, and determine whether inactivation of RXRα and RARγ elicited superimposable effects on different LPM clusters (as predicted if they operate as a heterodimer, a known propensity of RXR proteins 3 ), we performed single-cell RNA sequencing (scRNAseq) analyses on total CD11b + CD115 + peritoneal cells from BMC mice edited for several RA receptors, 17 days after reconstitution (each in biological duplicate, and all multiplexed by hashtagging into the same run for optimal comparability 59 ). As shown in Fig.4A , Uniform Manifold Approximation and Projection (UMAP) and Louvain clustering identified 10 cell clusters within CD11b + CD115 + cells that could be categorized into LPM (GATA6 + F4/80 hi , clusters 1–5) and SPM (CD226 + IRF4 + , clusters 7–10), and also distinguished a LYVE1 + CD206 + cluster-6 with transcriptional characteristics intermediate between LPM and SPM, which is often annotated as “Converting macrophages” 60 , 61 ( Figs.4B , S4A – B ). Transfer of these LYVE1 + CD206 + cells confirmed that they are indeed LPM precursors. Cells with either RORα or RXRβ deficiencies yielded largely the same distribution as WT macrophages ( Fig.4C ), with essentially identical cluster frequencies ( Fig.4D represents equal numbers of cells from each condition, Fig. S4C shows the same maps downsampled to match actual numbers of deficient CD11b + CD115 + cells). Deficiencies in RARγ and RXRα both caused a marked decrease in LPM clusters, as expected ( Figs.4C – D , S4C ). Sequencing of the gRNA target regions in LPMs that remain in spite of RARγ or RXRα deficiencies showed them enriched in un-mutated loci. However, these two deficiencies differed in the distributions of SPM and precursor clusters ( Fig.4D ): RARγ-deficient cells appeared similar to those of WT (with an apparent increase in cluster-6 and -8 due to the loss of LPMs); RXRα-deficient cells showed a strong reduction in cluster-6 and -8, with a corresponding increase in SPM cluster-9, which expresses higher levels of DC-associated CD209a, consistent with the possibility that loss of RXRα favored the more DC-like orientation of SPMs described previously 62 – 64 . These differences were verified by flow cytometric analysis of BMC mice with these deficiencies ( Fig.4E , gating strategies in S4D ). Thus, the RARγ deficiency allowed entry of precursors into the peritoneal cavity, and some differentiation into cluster-6 cells thereafter, while the RXRα deficiency entailed more immediate perturbations. It is interesting to note that the distribution of RARγ deficient peritoneal macrophages (very few GATA6 + LPM, many LYVE1 + ) resembles that found in human peritoneal cavity macrophages 61 , an observation which may be connected to the low levels of RARγ in human cells. To evaluate the transcriptional consequences of the two mutations, independently of the LPM loss that dominated the bulk profiles of Fig.2 , we analyzed differentially expressed genes (DEGs) by RNAseq profiling of sorted LYVE1 + CD206 + MFs (cluster-6 equivalent) from WT and mutant cells ( Fig.4F , Table S4 B – C ). The RARγ deficiency had little impact (only 15 and 19 genes up- and down- regulated at arbitrary cutoffs (|FC|>2 and p<0.01), not very different from experimental noise in such experiments. The loss of RXRα had more severe consequences (181 affected genes altogether). Interestingly, the expression of Alox15 , Ltc4s and Ptgs1 , which encode key enzymes controlling conversion of arachidonic acid to cysteinyl leukotrienes and prostaglandins, was specifically downregulated in RXRα, but not in RARγ mutants ( Figs.4F , S4G ), hinting that AA metabolites might influence LPM maturation 65 . We also analyzed patterns of genomic DNA-binding by chromatin immunoprecipitation (ChIPmentation 66 ). Peaks of significant binding were observed in whole peritoneal MFs of wild-type mice with anti-RXRα (n=876) ( Fig.S4E , Table S5A ), but we could not obtain convincing data with anti-RARγ. RXRα-binding peaks were significantly enriched (p<10 −100 ) in NR-binding motifs, notably those for retinoic acid receptors ( Fig.S4F ). A significant proportion of these peaks mapped in the vicinity (defined as closest gene, mostly within 30 kb) of genes positively controlled by RXRα in peritoneal MFs ( Fig.S4G , Table S5B ). Taken together, these results showed that, although they both controlled LPM numbers, RARγ and RXRα did so in a different manner: RXRα acted as a classical TF, binding and transactivating the expression of genes that are essential for LPM differentiation. In contrast, RARγ allowed normal maturation of the precursors, its absence having only limited transcriptional impact, its importance becoming apparent only at the mature LPM stage. To orthogonally validate our genetic data, we used a panel of synthetic agonists and antagonists to modulate the activity of RA receptors in wild-type mice in vivo , (i.p. injection over a week), assessing the effect on LPMs and lung alveolar MFs. In keeping with the genetic screen, the selective RARγ antagonist LY2955303 (hereafter “LY”) dramatically reduced LPM frequency, but had no effect on lung alveolar MFs ( Fig.5A ). Compounds targeting RXRα or other NRs had no such effect. We then used this specific inhibitor for mechanistic investigations of the control of LPMs by RARγ. We quickly realized that LY’s action on LPMs was extremely rapid, several manifestations appearing within 10 minutes, much faster than could be accounted for by transcriptional changes ( Fig.5B ). CD11b and F4/80 were down-modulated from the surface of LPMs; GATA6 levels were even more dramatically reduced, with evidence of proteolytic cleavage ( Fig.5C ); plasma membranes exhibited evidence of programmed cell death (PCD), as revealed by Annexin-V and propidium iodide (PI) staining ( Fig.5B ).These effects were also observed in recipients of transfers of reporter-tagged LPMs, indicating that they reflected true modifications of LPMs, rather than rapid recruitment of other cells into the peritoneum ( Fig.S5A ). At later time points (3 hours after LY injection), an influx of monocytes and neutrophils into the peritoneal cavity indicated the pro-inflammatory nature of these events. Several points argued that these outcomes denoted true biology, as opposed to simple toxicity or off-target action. First, they were also observed upon treatment with an independent RARγ antagonist, MM11253 ( Fig.S5B ), Second, LY treatment did not induce notable reductions in the numbers or viability of other peritoneal cell-types, including SPMs, B cells, and eosinophils ( Fig.S5C ). Third and most direct, the effects of LY treatment were hindered by prior administration of the RARγ agonists CD437 67 and BMS961 68 , indicating that the antagonist’s impact was specifically reversed by a matching agonist ligand ( Fig.5D ; the agonists had no discernible effects on their own in this assay). Interestingly, the dose range of LY effectiveness was very sharp, with a transition from virtually no effect (30 nmoles) to full effect (50 nmoles) over a 2-fold span ( Fig.S5D ). LPMs are known to disappear rapidly after triggering of innate receptors by LPS or live bacteria, by migration to the omentum or aggregation on the peritoneal wall within fibrin clots 29 , 69 , 70 . However, intraperitoneal injection of LPS did not induce the same downregulation of CD11b and GATA6 as RARγ antagonists did ( Fig.S5E ), indicating that the latter did not simply mimic the actions of innate immune agonists. In sum, these findings demonstrated the highly specific effects of antagonizing RARγ on LPMs. Because SPM seemed resistant to LY treatment, we investigated target cell specificity in more depth. Thioglycolate-elicited peritoneal macrophages, which are GATA6-negative and express lower levels of RARγ, were resistant to LY injection ( Fig. 5E ). Similarly, alveolar MFs and Kupffer cells were not affected by i.p. administration of LY that perturbed LPMs ( Fig.S5F ). To rule out a compound diffusion problem and for a direct comparison of macrophage sensitivity in the same in vivo environment, we transferred total lung CD45 + cells from Kaede reporter mice into the peritoneal cavity of naïve hosts, followed rapidly by i.p. LY injection. Macrophages of alveolar origin were still resistant to LY treatment, in a location where endogenous LPM were fully susceptible ( Fig.5F ). This tight specificity for GATA6 + MFs of the RARγ antagonist, coupled with the rapid degradation of GATA6 itself, prompted the hypothesis that GATA6 might participate in LY -induced events. We thus tested the effects of LY administration in mice with cell-specific deficiencies in GATA6 ( Gata6 flox/flox crossed to LyzM-cre or Cd115-creERT2 ). Indeed, GATA6 deficiency conferred resistance to LY-induced cell death, and a partial effect on CD11b cleavage ( Fig.S5G ). Proteolytic cleavage of CD11b is associated with neutrophil detachment during chemotaxis 71 . Together with the cleavage of GATA6 described above ( Fig.5C ), and the speed of events, these observations suggested that LY-induced modifications were initiated by proteolysis rather than by transcriptional modifications. Indeed, injection of a pan-protease inhibitor prior to LY administration resulted in a complete blockage of CD11b and GATA6 downregulation, and Annexin-V binding ( Fig. S6A ). The rapidity of the events induced by antagonism of RARγ, their sensitivity to pan-protease blockade, and the Annexin-V binding induced on LPMs suggested the possible involvement of PCD cascades. We therefore evaluated the activation of several molecular players associated with apoptosis, pyroptosis or necroptosis. Immunoblotting revealed the activation of these pathways in peritoneal cells in the first 20 minutes after LY injection in vivo ( Fig.6A ), involving molecules associated with several types of PCD. Related to pyroptosis were the cleavage of Caspase-1 and the appearance of the active p30 form of Gasdermin D (GSDMD), the pyroptosis effector required for membrane pore formation 72 . Related to apoptosis were the cleavage of caspases-3, -7 and -8, and nuclear poly (ADP-ribose) polymerase (PARP) ( Fig.6A ). Similar proteolytic activation was also induced by LY in purified LPMs in culture, denoting a direct effect of the antagonist ( Fig.6B ). While PCD types were originally defined as independent and segregated processes, it is now recognized that there is significant crosstalk and bridges between them 73 – 76 , and the multiple cleavages induced by the RARγ antagonist likely reflected such spreading. To identify the key initiators, we injected several PCD inhibitors prior to LY treatment. Inhibition of RIPK1, a key node of necroptosis, had no effect on the various aspects of LY-induced response in LPMs, while inhibitors of Caspase-1 or Caspase-3 effectively blocked all manifestations, as did a pan-caspase inhibitor ( Fig.6C ). Furthermore, the inhibition of Caspase-1 prevented LY-induced cleavage of Caspase-3, PARP and GSDMD ( Fig.6D ), suggesting that Caspase-1, the main driver of canonical pyroptosis, is positioned upstream in the cascade, and that RARγ is connected to inflammasome activation. The main Caspase-1 inflammasomes (NLRP1, NLRP3, AIM2, NLRC4) sense different pathogenic structures or stress-associated stimuli to induce pyroptosis. Most are strictly dependent on the ASC bridging adaptor, except for NLRP1 77 . To test the hypothesis that RARγ regulates inflammasome activation in a ligand-dependent manner, we treated LPMs with the BMS961 (selective RARγ agonist and RA mimic 68 that inhibited LY-induced effects in LPM above), prior to exposure to ligands that canonically activate different inflammasomes: (i) for NLRP3, LPS + ATP or LPS + nigericin; (ii) for AIM2, poly(dA:dT); (iii) for NLRC4, Shigella flexneri 78 . Indeed, as evidenced by Caspase-1 cleavage ( Fig.7A ) and IL1β secretion ( Fig.7B ), the RARγ agonist strongly inhibited Caspase-1 activation downstream of all of these inflammasomes, indicating that properly liganded RARγ was able to dominantly block the effects of all these pyroptosis-inducing agents. These results suggested that RARγ antagonists and agonists operate reciprocally, respectively activating or inhibiting inflammasome activation. Accordingly, examination of published structures of human RAR bound to these compounds 68 , 79 , 80 ( Fig.7C ) showed that they both associate with the RA-binding pockets of the receptors. We also performed loss-of-function experiments by examining the effects of the RARγ antagonist LY in Nlrp3 −/− , Asc −/− , Casp1 −/− 11 −/− , and Gsdmd −/− mice. In contrast to the strong blockade elicited by chemical inhibitors (in particular of Caspase-1 inhibitor), there were only minor differences in LY-induced manifestations in LPMs from these mice ( Fig.S6B ), likely due to the redundancy and compensation in cell death pathways 73 , 74 , 81 . To establish how RARγ interacted with inflammasome complexes, we conducted co-immunoprecipitation experiments, pulling down protein complexes with anti-RARγ antibodies. There was no detectable interaction of RARγ with Caspase-1, -3 or -7, or with NLRP3, but a clear binding to ASC, the key adaptor protein of inflammasome assembly and activation, was observed with two independent anti-RARγ Abs ( Fig.7D ). The interaction between RARγ and ASC was further confirmed by immunoprecipitation of lysates from transfected HEK-293T cells ( Fig.S7A ). Notably, LY treatment reduced the interaction between RARγ and ASC, indicating that their interaction was affected by the ligand bound to RARγ ( Fig.7D ). Consistent with its interaction with ASC, cell fractionation showed that RARγ was not solely nuclear, a sizeable fraction being cytoplasmic 17 ( Fig.S7B ). The involvement of ASC was also supported by the observation that pretreatment with the ASC inhibitor Lonidamine 82 blocked LY-induced Caspase-3 and GSDMD activation in LPMs ( Fig.6D ), as well as CD11b and GATA6 cleavage ( Fig.S7C ). Oligomerization of ASC is a key feature of inflammasome activation, eliciting downstream caspase recruitment 83 . Further establishing the connection between RARγ and ASC, the agonist BMS691 inhibited the recruitment of ASC into a polymeric insoluble form ( Fig.7E , top) and its polymerization ( Fig.7E , bottom panel) by NLRP3 inflammasome inducers. Should PCD cascades be connected to RARγ and its ligands through ASC, one would predict a different effect of the RARγ agonist on NLRP1 induced pyroptosis, which is considered as ASC independent. The involvement of ASC in NLRP1b inflammasome activation is complex. Although ASC is not required for pyroptosis induced by Anthrax lethal toxin (LeTx), the classic NLRP1b activator, it is necessary for Caspase-1 autoproteolysis, and partially required for IL-1β secretion 84 , 85 . We treated NLRP1b-expressing LPMs from BALB/c mice with BMS961 prior to LeTx treatment. The agonist indeed had no effect on LeTx-induced cell death, as evidenced by LDH release ( Fig.7F ). It did partially inhibit IL-1β release depending on the dose of initiating LeTx, and inhibited Caspase-1 autoproteolysis ( Fig.7F ). This disconnect between readouts mirrors what has been reported for LeTx-activated NLRP1b in ASC-deficient mice 84 , and this parallel strengthens the conclusion that RARγ interferes with ASC activity in a ligand-dependent manner. Thus, these results demonstrated that RA receptors strongly regulate the LPM population, but in very different ways. RXRα does it in a conventional manner, as a transcriptional regulator of genes important for LPM differentiation over time. RARγ, on the other hand, acts on a much faster time-scale, as a ligand-dependent rheostat of inflammasome assembly, either facilitating or hindering it in a ligand-dependent manner ( Fig.7G ).

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead Contact Prof. Christophe Benoist ( [email protected] ). This study did not generate new unique reagents. Bulk RNA-seq data, scRNA-seq data and ChIPmentation data generated in this project have been deposited in the Gene Expression Omnibus (GEO) database under accession number GSE254573 . Tabulated underlying data for Figs. 1 , 2 , 3 , 4 , S2 and S4 are available in Tables S1 , S2 , S3 , S4 and S5 . External datasets were retrieved from GEO (population RNA-seq for all major immunological cell types from GSE100738 , 12 different macrophage populations from Sequence Read Archive (SRA) PRJNA482293, and MC38 tumor associated macrophages from GSE185591 , ATAC-seq data of peritoneal macrophages from GSE63338 ). All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary materials .

NRs play diverse roles in the immune system, but a comprehensive understanding of how they regulate immunocyte biology was lacking. The in vivo screen strategy was applied to systematically determine the influence of 35 murine NRs on the differentiation and homeostatic maintenance of most immunologic cell-types, revealing a broad overall effect on repopulation potential, as well as some more focused cell-type-specific requirements. Of the latter, the most striking was the impact of RXRα and RARγ deficiencies on GATA6 + LPMs, which occurred through distinct regulatory mechanisms. This exploration revealed an unexpected function of RARγ which acts, in a ligand-dependent manner, to either facilitate or dampen an explosive program of proteomic alteration and cell death. The Rainbow-CRISPR screen uncovered several known and unrecognized functions of NRs in the immune system. Inactivation of many of the NRs mildly affected overall reconstitution frequencies, manifested only as trends for many, suggesting roles in the genetic regulatory network of HSCs and progenitors (such broad effects might also take place in differentiated cell-types but were not captured here, because they were factored out by the normalization). Of the specific effects, some were unexpected such as that of the orphan NR2C2 (a.k.a. TR4 86 ) with no previously known relationship to B.GC or Treg physiology, while others amplified prior knowledge. Consistent with its well-established functions in thymic T cell differentiation, we found that RORγ inactivation impacted αβT cells, but less expectedly also reduced the fraction of FoxP3 + Treg cells. It remains to be determined whether RORγ regulates thymic Treg differentiation or homeostasis in the spleen. In contrast, inactivation of NR3C1, the receptor for strongly immunoregulatory glucocorticoids, had surprisingly little effect. Overall, perturbing receptors for retinoic acid had the most frequent impact, with both positive and negative effects that denote balancing roles in some cells (e.g. RXRα and RARα in red pulp macrophages). Some RA receptors affected a whole lineage, or differentiated subsets thereof. For instance, total B cells were affected by deletion of RXRβ and RARα, while differentiated MZ and GC B cell states were regulated by RXRα and RXRβ. The study also highlighted the diverse influences of RXRα on tissue macrophage differentiation, consistent with the enrichment of enhancer-bound RXRα protein in TRMs of several origins 58 . Our study, which relied on HSC-differentiated macrophages, showed marked divergence between tissues in the requirement for RXRα, prompting the speculation that these differences are due to inductive effects within the tissues, but not to the stem cell of origin. PPARγ and LXRα control the functional specification of alveolar and splenic macrophages, respectively 87 – 89 . Since RXRα can heterodimerize with PPARγ and LXRα, it may function paired with them in various TRMs. From one viewpoint, one might be surprised by the relative paucity of specific effects unearthed in this screen, which was particularly striking in the gene expression profiles of mutant peritoneal MFs. Aside from the dominant effects of RXRα and RARγ deficiencies, most deletions left very subtle or no traces in LPM transcriptomes. One interpretation may be that differentiating cells adjust to the missing NR, through functional redundancy between NRs or through more complex adaptations of the genetic regulatory network. In addition, some effects may have been missed because inactivation was not complete or because the cells were only exposed to physiological amounts of hormones or NR ligands, and were not specifically challenged or triggered. The mechanistic exploration revealed the two-pronged transcriptional and inflammasome-mediated regulation of LPM fate, orchestrated by RXRα and RARγ, likely accounting for their previously reported dependence of LPMs on RA 29 , 42 . LPMs play critical roles in maintaining the integrity of body cavities and have been suggested to be evolutionarily ancient cell-types 52 , 54 . They are implicated in several diseases, such as postsurgical adhesions, endometriosis, and some cancers. Our genomic analyses demonstrated that RXRα directly controls LPM differentiation through classic transcriptional regulation, consistent with prior studies 30 . Intriguingly, RXRα-controlled genes included several that encode key enzymes in the arachidonic acid pathways, whose metabolites can influence LPM differentiation and phenotype 65 . Far less expected was the ligand-dependent control of inflammasome activation in LPMs, anchored by the interaction of RARγ with ASC. There are classic precedents for control of cell survival/death by NRs (GR, RORγ and RAR) through transcriptional modulation of pro- or anti-apoptotic protein expression like Bcl-2 family members or Fas/FasL 20 , 90 , 91 . More relevant are recently described instances of cell death affected by NR via non-genomic routes. During TNF-induced death, RARγ favors RIP1 dissociation from TNFR1 to form death-signaling complexes 17 . Most recently, NR4A1 has been reported to activate the NLRP3 inflammasome after binding LPS and cytoplasmic dsDNA 18 . Here, RARγ acts as a key checkpoint for LPM survival, flipping between pro- and anti-survival modes. We speculate that agonist and antagonist molecules, which fit the same ligand-binding pocket as RA, result in allosterically distinct RARγ structure. When bound to BMS961 or RA, RARγ binds ASC and supports LPM survival by hindering inflammasome activation by canonical stimuli. Conversely, when bound to its antagonist LY, RARγ dissociates from ASC, which then oligomerizes and provokes a Caspase-1 dependent activation of other caspases and Gasdermin. It seems unlikely that the agonist-dependent switch in ASC interaction alters a simple stoichiometric sequestering of ASC, and we suspect that a more complex regulation must be at play, in keeping with the highly non-linear “tipping point” mode of inflammasome activation. For instance, the RARγ-ASC complex might interfere with ASC polymerization as a dominant negative. In terms of physiological relevance, this mechanism may allow a finely tuned homeostatic control on LPM populations as a function of RA concentration in serous cavities 29 . Alternatively or in addition, natural antagonists may exist (endogenous or produced by microbes), that destabilize MF populations, possibly in situations when it is important to rapidly remove macrophages from the local cavities. RARγ’s influence is tightly restricted to GATA6 + MFs, across body locations or in the LPM maturation cascade. Indeed, GATA6 itself appears to be a player in the phenomenon, as indicated by the reduced sensitivity of GATA6-deficient MFs, in line with reports that GATA6 deficiency renders LPMs more vulnerable to apoptosis and downregulation of CD11b 47 , 48 . More generally, it will be interesting to see whether regulation of inflammasomes applies to other contexts in which NRs have been associated with cell survival. Starting from a broad unsupervised survey, this study ended up identifying a striking mode of ligand-regulated control of inflammasome activation, opening the door to explorations of wider involvement of this process in NR function, and to potential therapeutic applications. This investigation focused on the role of NRs in immunological homeostasis at steady state, in the presence of physiologically supplied ligands. NR functions specific to challenged contexts (inflammation, infection, tumors) would have been missed. Because the lentivirus-mediated gene inactivations were not complete, we cannot rule out that, in some instances, homeostatic drive might have led to the selective amplification of the un-edited fraction of cells (as clearly happens for RXRα-deficient LPMs), masking deeper effects. This situation precluded testing the affect of RARγ-modulating ligands in mature LPMs. Absence of effects should be interpreted with caution, as redundancy between NRs may mask true involvement.

Table S1: Technical details and tabular results of the Rainbow screen, related to Fig. 1 A) Sequences of control and NR-targeting gRNAs. B) PCR primers for amplicon library preparation for NGS (two-step PCR). C) CRISPR/Cas9 genome editing efficiency for each NR ablation in LSK cells before transfer and in recovered B lymphocytes. Editing frequency was calculated as the percentage of reads containing a frame-shift indel out of total reads at each target site and the estimated knockout efficiency was calculated by the frequency of cells carrying one or more homozygous knockout mutation of the target gene (related to Fig. S1F ). D) The relative proportions of total CD45+ donor-derived cells derived from NR- or control-targeted HSCs within the same mouse (related to Fig. 1B ).The non-targeting Ctrl (Ctrl_NT) and intergenic Ctrl (Ctrl_SH) derive from Control BMC mice receiving several Ctrl gRNAs. E) Whole “Rainbow-CRISPR” screen data (effects of targeting 35 NRs on 28 immune cell types, immune reconstitution for 4 weeks for myeloid cells and 10 weeks for lymphocytes). NR/Ctrl ratios of each cell type normalized to NR/Ctrl ratios of all CD45+ splenocytes, or all B cells, or all αβT cells. The “Ctrl_NT” derived from control BMC mice receiving several control gRNAs. F) Summary of effects of targeting 35 NRs on differentiation and homeostatic levels of 28 immune cell-types. Mean NR/Ctrl ratios of donor-derived immunological cell types, normalized to NR/Ctrl ratios of whole donor-derived CD45+ cells or whole spleen B or whole spleen αβT cells as indicated. p values were assessed by one-way ANOVA with Dunnett correction (related to Fig. 1C ). Table S2: Population RNAseq on mutant macrophages, related to Fig. 2 A) Bulk RNAseq was performed in biological duplicates on donor-derived peritoneal macrophages of “Rainbow-CRISPR” mice with deficiencies in LPM-expressed NRs. Foldchange was calculated as NR mutant /Ctrl. p values were from one sample t.test with a hypothesis value of 1 (related to Fig.2A ). B) Differentially expressed genes (DEGs) passing absolute log2(FC)>1 and nominal p<0.05 in any one NR deficiency (related to Fig.2B ). C) LPM-and SPM-specific signatures and GATA6-regulated genes (related to Fig. 2B ). Table S3: Rainbow screen in tissue macrophages, related to Fig. 3 A) Proportion of MFs carrying illustrative control or NR-targeting gRNAs in TRMs and MC38 TAMs within total donor-derived CD45+ cells in each tissue (% MFs of CD45+ ), measured by flow cytometry. Each pair from an individual mouse (related to Fig. 3A ). B) NR/Ctrl ratios of TRMs and MC38 TAMs, normalized to NR/Ctrl ratios of total donor-derived CD45+ cells in each tissue and p.values were assessed using paired t.test (related to Fig. 3B ). Table S4: Single-cell RNAseq details and data, related to Fig. 4 A) scRNAseq analyses on total CD11b+CD115+ peritoneal cells from BMC mice with deficiencies in several RA receptors, 17 days after reconstitution. Each NR mutant has two replicates, while three wild-type samples were paired with NR mutant cells from the same BMC mouse. All samples were multiplexed by hashtagging into the same run for optimal comparability. B) Population RNAseq profiling of sorted LYVE1+CD206+ MFs (cluster-6 equivalent) from WT and RARγ mutant cells. Only DEGs with |FC|>2, P2, P<0.01 were shown (related to Fig. 4F ). Table S5: Chromatin immunoprecipitation results, related to Fig. 5 A) High-confidence (FDR < 0.01), and reproducible ChIP-seq peaks among RXRα ChIPmentation replicates, generated using anti-RXRα (D6H10) Rabbit mAb (#3085, Cell Signaling) in whole peritoneal macrophages of wild-type mice. B) RXRα binding peaks in the vicinity of transcripts differentially expressed in RXRα- deficient vs WT peritoneal MFs (with |FC|>2 and p< 0.01, data from Fig. 2 ) (related to Fig.S4G ). Table S6: RAR agonist and antagonist compounds tested in vivo and in vitro., related to Fig. 6 Table S7: Antibodies used in immunoblotting assays, related to Fig. 7