In Vivo CRISPR Screening Reveals CHD7 as a Positive Regulator of Short-lived Effector Cells

This article is featured in Top Reads, p. 1403

Immune responses to acute viral infections, such as lymphocytic choriomeningitis virus Armstrong (LCMV-Arm), are characterized by the differentiation of short-lived effector cells (SLECs) and memory precursor effector cells (MPECs) (1, 2). Both SLECs (KLRG1⁺ CD127^–) and MPECs (KLRG1^– CD127⁺) contribute to the effector response with comparable cytotoxicity and production of cytokines, such as IFN-γ and TNF-α. These populations also have distinct roles: SLECs are slightly more proliferative, whereas MPECs express higher levels of IL-2 and persist longer (2). Following viral clearance, the majority of the responding T cell populations contract, yet long-lived memory T cells capable of rapid responses to a recurrence of the original pathogen persist (1, 3). Multiple subsets of memory T cells—central, effector, and tissue-resident—have been described (4–6) with unique functions and localization in the host. Notably, although MPECs and SLECs can form all memory T cell subsets, MPECs do so more efficiently (7).

Epigenetic control of gene transcription—opening or closing chromatin, recruitment of transcription factors and associated complexes, and deposition of post-translational modifications on histones and DNA—is mediated by three main types of epigenetic regulators: writers, readers, and erasers (8, 9). MPECs, SLECs, and memory T cell subsets have distinct epigenetic profiles (10, 11) that enable fine-tuning of transcription to relevant contexts and stimuli (12, 13). Given the distinct epigenetic profiles of MPECs, SLECs, and memory subsets, multiple groups have investigated the epigenetic regulators that govern formation of these subsets. Several epigenetic regulators, such as TET2, EZH2, SUV39H1, and DNMT3A, have been implicated in MPEC, SLEC, and memory T cell generation (14–17). However, we do not have a comprehensive understanding of the regulators required for formation of these populations. Furthermore, most central memory T cells are derived from effector memory T cells (18), but the epigenetic regulators that govern this conversion are unknown. Given the importance of T cell memory to immune responses following pathogen re-exposure, it is important to define the regulators that control the differentiation of effector and memory T cell populations.

To identify the epigenetic regulators that control the 1) MPEC versus SLEC fate decision and 2) central memory versus effector memory formation, we conducted in vivo CRISPR screens in naive CD8⁺ T cells examining epigenetic writers, readers, and erasers. CRISPR-based loss-of-function screens are a powerful approach for efficiently assessing thousands of genes that affect a given phenotype (19, 20) and have been used for understanding T cell responses to pathogens. However, the majority of published T cell screens (21–24) have used preactivated T cells, which can change the expression of cytokine receptors, effector molecules, and transcription factors during T cell differentiation (25), potentially biasing screen results. In this study, we describe two CRISPR-based loss-of-function screens in naive CD8⁺ T cells to enable faithful differentiation of T cell subsets. Using these screens, we evaluated the epigenetic regulators that control the 1) MPEC versus SLEC fate decision and 2) central memory versus effector memory formation. These screens identified the ATP-dependent chromatin remodeler CHD7 (chromodomain-helicase DNA-binding protein 7) as a novel regulator of MPEC versus SLEC differentiation. Using functional, transcriptional, and epigenetic analyses to compare wild-type (WT) and Chd7 knockout (KO) T cells, we found that KO of Chd7 reduced SLECs numerically, but CHD7 was not absolutely essential for SLEC formation. In addition, we determined that CHD7 inhibits the generation of central memory T cells while promoting memory cell responses during a pathogen rechallenge. Collectively, these findings implicate CHD7 as a novel regulator of effector and memory T cell responses.

LoxP-STOP-loxP-Cas9 mice [B6J.129(B6N)-Gt(ROSA)26Sor^{tm1(CAG-cas9*,-EGFP)Fezh}/J Jax no. 026175] were backcrossed 10+ generations to C57BL/6J mice (Jax no. 000664) and then bred to Zp3-Cre mice [C57BL/6-Tg(Zp3-cre)1Gwh/J Jax no. 003651] to delete the loxP-STOP-LoxP in the female germline. We subsequently bred out the Zp3-Cre allele to create the Rosa26-Cas9–expressing (Cas9^ON) strain (26). To mark transferred cells, the Cas9^ON strain was bred to CD45.1 mice (B6.SJL-Ptprc^a Pepc^b/BoyJ Jax no. 002014). The CD45.1 Cas9^ON strain was then bred to a B6-backcrossed P14 strain to generate a P14 CD45.1 Cas9^ON strain. The B6-backcrossed P14 strain was made by crossing P14 mice (B6.Cg-Tcra^tm1MomTg(TcrLCMV)327Sdz Taconic no. 4138F) to wild-type B6 mice (C57BL/6J Jax no. 000664) for 10 generations. The CD45.1 Cas9^ON strain was also bred to OT-1 mice (C57BL/6-Tg(TcraTcrb)1100Mjb/J Jax no. 003831) to create an OT-1 CD45.1 Cas9^ON strain. The P14 CD45.1 Cas9^ON and OT-1 CD45.1 Cas9^ON strains were used for screen studies. For the primary screen, 7- to 10-week-old female wild-type B6 mice (C57BL/6J Jax no. 000664) and Cas9^ON mice were used as recipients of T cells, and 7- to 16-week-old female P14 CD45.1 Cas9^ON mice were used as bone marrow donors for the screens. For the secondary screen, 7- to 10-week-old male Cas9^ON mice were used as recipients of T cells for the screens, and 7- to 36-week-old male OT-1 CD45.1 Cas9^ON mice were used as spleen donors for the screens. CD45.1 OT-1 mice were created by breeding CD45.1 mice (B6.SJL-Ptprc^a Pepc^b/BoyJ Jax no. 002014) to OT-1 mice [C57BL/6-Tg(TcraTcrb)1100Mjb/J Jax no. 003831]. CD45.1 P14 mice were created by breeding CD45.1 mice (B6.SJL-Ptprc^a Pepc^b/BoyJ Jax no. 002014) to a B6-backcrossed P14 strain, as described above. The OT-1 CD45.1 and P14 CD45.1 strains were used for nucleofection studies. In addition, 7- to 12-week-old female wild-type B6 mice (C57BL/6J Jax no. 000664) were used as recipients of nucleofected T cells, and 7- to 20-week-old female OT-1 CD45.1 and P14 CD45.1 mice were used as donors of nucleofected T cells. The experimental sample size was chosen to ensure the possibility of statistical analysis and minimize the use of animals. Age- and sex-matched animals were used for each experiment. Experimental mice were cohoused for all experiments. All experimental mice were housed in specific pathogen-free conditions and used in accordance with animal care guidelines from the Harvard Medical School Standing Committee on Animals and the National Institutes of Health.

pXPR_053 (Addgene identifier 113591) was used for CHIME experiments (26). pRDA_526 (Addgene identifier 184859) was used for T cell transduction experiments and is a modified form of gRNA expression vector pXPR_071 (Addgene identifier 164558) with the following changes: 1) modification of the tracrRNA, 2) modification of the stuffer between BsmBI digest sites, 3) addition of a gRNA capture sequence following the tracrRNA, and 4) replacement of the fluorophore Vex with Thy1.1.

The gRNA oligonucleotides used in the 4,400-gRNA (primary) and 635-gRNA (secondary) screens were designed using the Broad CRISPR algorithm (https://portals.broadinstitute.org/gppx/crispick/public). For the 4,400-gRNA screen, we used a list of putative human epigenetic regulators curated by Dr. Bradley Bernstein (Dana-Farber Cancer Institute). We converted this list to mouse genes (1,448 genes) and designed 3 gRNAs per gene. We also added in 50 nontargeting control gRNAs and gRNAs targeting Tcf7 and Tbx21 as positive controls. Supplemental Table I contains the full list of gRNAs used in the 4,400-gRNA screen.

For the 635-gRNA screen, we analyzed the 4,400-gRNA primary screen data and identified 93 genes of interest. Briefly, for primary screen MPEC/SLEC analyses we applied a two-read count filter for MPEC and SLEC samples and Z-scored the log₂-normalized fold changes of MPEC/SLEC. We then applied a Z-score cutoff of 2 and only considered genes that had gRNAs score in at least two of the three organs (lymph node [LN], lung, and spleen). For primary screen central memory/effector memory analyses, we applied a two-read count filter for central memory and effector memory samples and Z-scored the log₂-normalized fold changes of central memory/effector memory. Due to a large number of high Z-scores, we then ranked the Z-scores and took the top and bottom 10% of ranked genes. We then consolidated the genes from our MPEC/SLEC and central memory/effector memory analyses to create our list of 93 genes. We added in 20 positive control genes to the 93 aforementioned genes and then designed 5 gRNAs per gene to these 113 genes (565 gRNAs targeting these genes). We also included 70 negative control gRNAs (35 nontargeting control gRNAs and 35 intergenic-targeting control gRNAs) such that >10% of the resultant 635 gRNA library is negative control gRNAs. Supplemental Table I contains the full list of gRNAs used in the 635-gRNA screen. For validation experiments, the following gRNAs were used: Control-1, GCGAGGTATTCGGCTCCGCG; Chd7 g1, TCTTGCGAAGCAGTTCAATG; Chd7 g2, CGTGATGGGCTCGATAAGGG; and Tet2 g1, CTCACAATAATACCCAAGGG.

pXPR_053 and pRDA_526 were predigested with BsmBI (NEB catalog no. R0580L). The 4,400-gRNA library was cloned into pXPR_053 via a two-component Golden Gate assembly. The 635-gRNA library and 96 unique molecular identifiers (UMIs) were cloned into pRDA_526 via a three-component Golden Gate assembly. Ligation products were cleaned up with an isopropanol precipitation. Ligation products were then electroporated into Stbl4 bacteria (Thermo Fisher Scientific, catalog no. 11635-018) and plated on large bioassay plates at 37°C. Plasmid libraries were then extracted using a HiSpeed Plasmid Maxi kit (Qiagen catalog no. 12662), quantified via NanoDrop, and sequenced to confirm even representation of the gRNAs (Supplemental Table II). Both libraries are available at the Broad Institute’s Genetic Perturbation Platform as custom pools: CP1272 (4,400-gRNA library in the pXPR_053 vector) and CP1646 (635-gRNA library with 96 UMIs/gRNA in the pRDA_526 vector). Supplemental Table I contains the full list of UMIs used in the 635-gRNA screen.

293x cells (HEK variant) were cultured in DMEM (Life Technologies catalog no. 11995073) supplemented with 10% FBS (Sigma) and 1% penicillin/streptomycin (VWR catalog no. 97063-708). 293x cells were used for making lentivirus, which was validated by titering. BHK-21 cells were cultured in DMEM supplemented with 10% FBS, 1% penicillin/streptomycin, and 5% tryptose phosphate broth (Sigma-Aldrich catalog no. T8159). Vero cells were cultured in Eagle’s MEM (American Type Culture Collection catalog no. 30-2003) supplemented with 10% FBS and 1% penicillin/streptomycin. B16-OVA-RFP cells (27) were cultured in DMEM containing 10% FBS and 1% penicillin/streptomycin and were validated by flow cytometry. All cell lines were confirmed to be mycoplasma negative and free of standard pathogens.

LCMV virus was produced by infecting BHK-21 cells with an LCMV-Arm or LCMV-Arm-OVA (28) virus stock at a multiplicity of infection of 0.03 and harvesting viral supernatants 48 h later. Viral titers were determined by plating diluted viral stocks or tissue samples on Vero cells with a 199 medium (Invitrogen catalog no. 31100-035)/FBS/penicillin–streptomycin/agarose (Lonza catalog no. 50004) overlay. Four days later, the Vero cells were fixed with 4% paraformaldehyde (Thermo Scientific catalog no. J19943.K2) and stained with 1% crystal violet (Sigma-Aldrich catalog no. C0775), and the plaques were quantified.

The mice were infected with 2 × 10⁵ PFU LCMV-Arm or LCMV-Arm-OVA i.p.

Lung lymphocytes were isolated by dissociation of the lung on a Miltenyi gentleMACS Dissociator followed by a 37°C incubation in collagenase I (Worthington Biochemicals catalog no. LS004194) for 30 min. Lymphocytes were enriched on a 40%/60% Percoll (GE Healthcare catalog no. 17-0891-09) gradient. Splenocytes were isolated by mechanical dissociation of the spleen on a 70-µm filter followed by enrichment with Miltenyi CD8a microbeads (Miltenyi catalog no. 130-117-044). For some of the memory time points, splenocytes were isolated by mechanical dissociation of the spleen on a 70-µm filter followed by enrichment with Miltenyi’s CD8a⁺ T cell isolation kit (catalog no. 130-104-075) followed by Miltenyi’s anti-FITC microbead kit (catalog no. 130-048-701). Lymphocytes from the inguinal lymph node were isolated by mechanical dissociation on a 70-µm filter followed by enrichment with Miltenyi CD8a microbeads. Lamina propria samples were incubated in extraction medium (DTT-containing [Thermo Scientific catalog no. P2325] and EDTA-containing [VWR catalog no. 45001-122] solution) to remove intraepithelial lymphocytes. Then, the remaining tissue was minced using forceps, followed by digestion of the remaining tissue with collagenase II (Thermo Fisher Scientific catalog no. 17101-015). Following digestion, the tissue was passed through a 40-µm filter. Lymphocytes were enriched using a 40%/70% Percoll gradient.

The samples were stained for 30 min (primary stain) or 20 min (secondary stain) on ice protected from light with Abs in FACS buffer which contains PBS^–/– (VWR catalog no. 82020-066) with 2 mM EDTA (VWR catalog no. 45001-122) and 1% FBS. Following staining, the cells were washed two times with FACS buffer. The cells were then resuspended in FACS buffer and either 1) analyzed on the following instruments: BD LSR II, BD Symphony A5, or BD Celesta or 2) sorted on a BD Aria IIu. Compensation was performed with single-color controls (Compensation Beads; Invitrogen catalog no. 01-2222-42). Gating was performed using fluorescence-minus-one controls where necessary. The samples were analyzed using FlowJo version 10.8 software.

B220 (catalog no. 103208); CD8α (catalog no. 100725); CD8β (catalog nos. 126606, 126614, 126616, 126608, 126631, and 126609); CD44 (catalog nos. 103032, 103043, 103030, and 103006); CD45.1 (catalog nos. 110741, 110714, 110730, 110731, 110708, 110706, and 110721); CD45.2 (catalog nos. 109831, 109837, 109814, and 109830); CD62L (catalog no. 104435); CD27 (catalog no. 124208); CX3CR1 (catalog no. 149020); CD127 (catalog no. 135006); Gr-1 (catalog no. 108408); IFN-γ (catalog no. 505810); IL-2 (catalog no. 503806); KLRG1 (catalog nos. 138410 and 138421); streptavidin (catalog nos. 410504 and 410505); Ter-119 (catalog no. 116208); Thy1.1 (catalog nos. 202529 and 202526); TruStain fcX (catalog no. 101320); TNF-α (catalog no. 506306); Vɑ2 (catalog no. 127814); and 7-aminoactinomycin D (7-AAD) (catalog no. 420404) were from BioLegend. Near-IR Fixable Live/Dead (catalog no. L34976), Dead Cell apoptosis kits with annexin V (catalog no. V35112), and CellTrace Violet (catalog no. NC0402709) were from Thermo Fisher Scientific. Anti-BrdU fluorophore (catalog nos. 559619 and 552598) was from BD Biosciences.

293x cells were transfected using polyethyleneimine (Polysciences catalog no. 24765-2) with the packaging plasmids pMD2.G (Addgene identifier 12259) and psPAX2 (Addgene identifier 12260), and gRNA-containing plasmids. Lentivirus-containing supernatant was collected 72 h post-transfection and spun at 854g for 5 min to remove cells. Then, the supernatant was filtered through a 45-µm low-protein binding filter (Whatman catalog no. EW-29705-54), placed in ultracentrifuge tubes (Beckman Coulter catalog no. 344058), and spun at 20,000 rpm (71,934.8 g) at 4°C in an SW28 rotor for 2 h in a Beckman Coulter ultracentrifuge. Following the ultracentrifugation, supernatant was removed and concentrated lentivirus was resuspended in SFEM medium (STEMCELL Technologies catalog no. 9600) overnight before freezing at −80°C. Lentivirus was titered on 293x cells to determine viral particles per ml.

Femurs, tibias, pelvises, and spines were isolated from CD45.1 P14 Cas9^ON donor mice and crushed with a mortar and pestle to liberate bone marrow. Bone marrow was filtered twice through a 70-µm filter and then enriched using CD117 microbeads (Miltenyi catalog no. 130-091-224). CD117-enriched bone marrow was then stained with CD117-APC (different clone than the CD117 microbeads), lineage-PE (Gr1, CD11b, CD5, CD3ε, Ter119, B220), and Sca1-BV421 Abs. The cells were sorted to purity on a BD Aria IIu. Sorted cells were then placed in a 37°C incubator overnight in polyvinyl alcohol-containing (Sigma-Aldrich catalog no. P8136-250G) medium as previously described (29). The next day, Lineage^– Sca1⁺ c-Kit⁺ (LSK) cells were spin transduced (692g, 27°C, 1.5 h) on RetroNectin-coated (Takara Bio catalog no. T100B) plates with CP1272-containing lentiviral supernatants at a multiplicity of infection of 30. This multiplicity of infection was chosen such that LSK cell transduction would be ∼30–40%. Following the spin transduction, the cells were cultured on tissue-culture treated plates in vitro as previously described (29) at a concentration of 150,000–200,000 cells/ml. After 1 week of culture, the cells were sorted for 7-AAD^– Vex⁺ cells on a BD Aria IIu. 200,000 Vex⁺ cells were injected intravenously into each of 20 CD45.2⁺ wild-type recipients that had been irradiated two times spaced 3–6 h apart with 600 rad per time (1,200 rad total) using a [¹³⁷Cs] irradiator. The mice were placed on Sulfatrim antibiotic (0.8 mg/ml sulfamethoxazole and 0.16 mg/ml trimethoprim) (Pharmaceutical Associates catalog no. NDC 00121-0854-16) in their drinking water for 1 week. The mice were then allowed to reconstitute for a minimum of 8 wk prior to use in experiments.

Following reconstitution, the chimeras were euthanized, and the bones and spleens were isolated. LSK cells were isolated from the bone marrow as above, and 100,000 LSK cells were frozen at −80°C as an input sample. Naive CD8⁺ T cells were isolated using Miltenyi’s naive CD8a⁺ T cell isolation kit (Miltenyi catalog no. 130-096-543). Naive CD8⁺ T cells were then stained with 7-AAD and Lineage (Ter119/Gr1/B220)-PE and sorted for 7-AAD^– Lineage^– Vex⁺ cells. 1 million LSK cells were frozen at −80°C as an input sample. Then, the Vex⁺ naive CD8⁺ T cells were injected intravenously (300,000 cells/mouse) into 20 CD45.2 WT recipients or 20 CD45.2 Cas9^ON recipients. One day after cell transfer, the mice were infected with 2 × 10⁵ PFU/mouse LCMV-Arm.

Eight days postinfection, the CD45.2 mice were euthanized, and the spleens, lungs, inguinal lymph nodes, and lamina propria were extracted. The spleens, lungs, lymph nodes, and lamina propria were processed as described above. The samples were sorted for Dead^– CD8β⁺ CD45.1⁺ Vex⁺ CD44⁺ KLRG1⁺ CD127^– for the SLEC population and Dead^– CD8β⁺ CD45.1⁺ Vex⁺ CD44⁺ KLRG1^– CD127⁺ for the MPEC population. The cell pellets were frozen down at −80°C. Genomic DNA was isolated using a DNeasy blood and tissue kit (Qiagen catalog no. 69504). Thirty days postinfection, the CD45.2 Cas9^ON mice were euthanized, and the spleens, lungs, and lymph nodes were extracted. The spleens, lungs, and lymph nodes were processed by mechanical dissociation as described above. The spleen and lymph node samples were sorted for Dead^– CD8β⁺ CD45.1⁺ Vex⁺ CD44⁺ CD127⁺ CD62L^– for the effector memory population from the spleen and Dead^– CD8β⁺ CD45.1⁺ Vex⁺ CD44⁺ CD127⁺ CD62L⁺ for the central memory population from the lymph node. The lung samples were sorted for Dead^– CD8β⁺ CD45.1⁺ Vex⁺ CD44⁺ CD127⁺ CD69⁺ or Dead^– CD8β⁺ CD45.1⁺ Vex⁺ CD44⁺ CD127⁺ CD103⁺. The cell pellets were frozen down at −80°C. Genomic DNA was isolated using a DNeasy blood and tissue kit (Qiagen catalog no. 69504).

293x cells were transfected using LT-1 (Mirus Bio catalog no. MIR2305) with the packaging plasmids pCag-Eco (Addgene identifier 35617), pMD2.G (Addgene identifier 12259), and psPAX2 (Addgene identifier 12260), and pRDA_526 plasmid (Addgene identifier 184859) containing the gRNA library (CP1646). Note pCag-Eco and pMD2.G were used at a 1:1 mass ratio to make pseudotyped lentivirus. Lentivirus-containing supernatant was collected 48 h after transfection and spun at 480g for 5 min at 4°C to remove cells. The supernatant was then frozen at −80°C until use.

Aliquots of lentivirus were thawed and tested prior to use in the secondary screen experiment. Lentivirus was spun on RetroNectin-coated plates for 2 h at 2,000g at 32°C. Naive OT-1 CD45.1⁺ CD8⁺ T cells were isolated using Miltenyi’s naive CD8a⁺ T cell isolation kit (Miltenyi catalog no. 130-096-543) and plated on the virus-coated plates with 1 mg/ml LentiBOOST (Sirion Biotech). The cells were cultured for 7 d in 1 ng/ml recombinant murine IL-7 (VWR catalog no. 10780-184) to enable expression of Thy1.1 and were analyzed by flow cytometry to ensure the virus led to sufficient transduction (15–30% transduced cells).

The secondary screen was performed by transducing OT-1 Cas9^ON naive CD8⁺ T cells with CP1646-containing lentiviral supernatants. Naive CD8⁺ T cells were cultured for 7 d in 1 ng/ml recombinant murine IL-7 (VWR catalog no. 10780-184). Naive CD8⁺ T cells were then stained with 7-AAD, Lineage (Ter119/Gr1/B220)-PE, and Thy1.1-BV421 and sorted for 7-AAD^– Lineage^– Thy1.1⁺ cells. A total of 635,000 cells were frozen at −80°C as an input sample. An additional 5.8 × 10⁶ cells were sorted and transferred intravenously to 15 Cas9^ON mice at 350,000 cells/mouse. Two days after cell transfer, the mice were infected with 2 × 10⁵ PFU/mouse LCMV-Arm-OVA. Eight days postinfection, mice were euthanized and spleens, lungs, and both inguinal lymph nodes were extracted and processed for CD8⁺ T cell enrichment (using CD8a microbeads). The mice were pooled into three groups of five mice for processing. The spleens, lungs, and lymph nodes were processed by mechanical dissociation as described above. The samples were sorted for Dead^– CD8β⁺ CD45.1⁺ Thy1.1⁺ CD44⁺ KLRG1⁺ CD127^– for the SLEC population and Dead^– CD8β⁺ CD45.1⁺ Thy1.1⁺ CD44⁺ KLRG1^– CD127⁺ for the MPEC population. The cell pellets were frozen down at −80°C. Genomic DNA was isolated from the 1 input sample, 6 lymph node samples, 6 spleen samples, and 6 lung samples using a DNeasy blood and tissue kit (Qiagen catalog no. 69504).

Prior to amplification of the genomic DNA samples from the secondary screen, we assessed DNA concentration using the Qubit dsDNA HS assay kit (Thermo Fisher Scientific catalog no. Q32851). We then performed a PCR (32 cycles of amplification) using Titanium Taq (Takara catalog no. 638517) on 2 µl of some of the samples to ensure the DNA was of sufficient quality for amplification. The following primers were used and are analogous to the primers that were used for amplification prior to sequencing by the Broad Institute’s Genetic Perturbation Platform: 5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTtcgatttcttggctttatatatcttgtg-3′ and 5′-CAAGCAGAAGACGGCATACGAGATGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTaccgactcggtgccactttttcaa-3′, where the bold capital letters indicate the P5/P7 flow cell attachment sequences, the capital letters indicate the Illumina sequencing primers, and the lowercase letters indicate the vector primer binding sequences. We ran amplified samples on a 1.5% agarose gel to assess formation of our indexed PCR product at 263 bp.

The Broad Institute’s Genetic Perturbation Platform amplified and indexed genomic DNA by PCR using Titanium Taq and the primers listed above. A portion of the samples was then run on a gel to estimate product amounts among the samples. The samples were then pooled based on these estimates to achieve roughly equimolar representation in the pool. The samples were then sequenced on a HiSeq2000 with a 50-cycle kit.

FASTQ files were demultiplexed using the Broad Institute Genetic Perturbation Platform’s program PoolQ (version 3.3.2) into individual samples with read counts per gRNA or gRNA-UMI pair (Supplemental Table II). PoolQ source code is located on GitHub: https://github.com/broadinstitute/poolq.

Screen data quality was assessed by determining the cumulative read distribution, the number of gRNAs recovered (>25 reads per million reads), concordance between mouse pools (Pearson correlation coefficient), and log₂-normalized fold changes of positive and negative control genes. For assessing hits, we used the hypergeometric test provided by the Broad Institute’s Genetic Perturbation Platform (https://portals.broadinstitute.org/gpp/public/analysis-tools/crispr-gene-scoring) and the MAGeCK algorithm (30). The p values generated from the hypergeometric test were normalized by performing a Benjamini–Hochberg correction. The MAGeCK algorithm directly provides Benjamini–Hochberg corrected p values in the negative and positive directions, which were plotted based on the direction of the log₂ fold change of the given gene. Gene scoring for the secondary screen can be found in Supplemental Table II. Gene scoring for the primary screen was not possible due to recovery of a single gRNA per gene for the majority of the genes. For normalization of fold changes to negative control gRNAs, we Z-scored the data using the following equation: (fold change of gRNA of interest – average fold change of negative control gRNAs)/standard deviation of negative control gRNAs. UMI recovery for control gRNAs and Chd7 gRNAs was determined by applying a read count threshold of 25 reads per million reads.

S. pyogenes Cas9 protein (IDT catalog no. 1081059) was mixed with synthetic gRNAs (Synthego) at a molar ratio of 8.1:1 (gRNA: Cas9) in nuclease-free water for 10 min to make Cas9-gRNA ribonucleoproteins (RNPs). Cas9 RNPs (25 µl) were then mixed with naive CD8⁺ T cells (isolated as above) in 100 µl of P3 buffer (Lonza catalog no. V4XP-3024) and loaded into nucleofector cuvettes (Lonza catalog no. V4XP-3024). The cells were then nucleofected using a Lonza 4D nucleofector with the DN100 program (cuvette setting). Immediately following nucleofection, 650 µl of CO₂-equilibrated complete RPMI 1640 medium (contains RPMI 1640 [Fisher Scientific catalog no. 11-875-093] with 10% FBS, 1% penicillin streptomycin, 10 mM HEPES [VWR catalog no. 97064-360], 50 µM 2-ME [VWR catalog no. 97064-588], 1 mM sodium pyruvate [Thermo Fisher Scientific catalog no. 11360070], and 1% nonessential amino acids [Thermo Fisher Scientific catalog no. 11-140-050v]) was added to the cells, which were placed in a 37°C incubator for 10 min to recover. The cells were then counted and prepared for transfer into mice intravenously. A subset of nucleofected cells was cultured at 1 million cells/ml in complete RPMI 1640 medium with 1 ng/ml IL-7 for 1 wk followed by DNA isolation for MiSeq-based KO verification.

To perform indel/frameshift analysis on the DNA of cells targeted with gRNAs, we isolated DNA from a minimum of 15,000 cells using a DNeasy blood and tissue kit (Qiagen catalog no. 69504). We then performed a PCR flanking the region of the gRNA edit. The overall PCR product was ∼220 bp, and the potential edit was within 100 bp of one end of the product. This product was then purified using a QIAquick PCR purification kit (Qiagen catalog no. 28106). Samples were then submitted to the Center for Computational and Integrative Biology DNA Core Facility at Massachusetts General Hospital (Cambridge, MA) for the CRISPR sequencing service for addition of sequencing adapters and sequencing on a MiSeq. Following sequencing, we used the CRISPR Ebert Pipeline on https://www.basepairtech.com to assess indel/frameshift formation. The following primers were used: Chd7 g1 forward, ACCCATGCTTCATGTTGCAGA; Chd7 g1 reverse, TTGAGGATGTGATAGTCTGTCCG; Chd7 g2 forward, ACCCATGCTTCATGTTGCAGA; Chd7 g2 reverse, TTGAGGATGTGATAGTCTGTCCG; Tet2 g1 forward, TGCTTTGGCCAGATTAAAGTGG; and Tet2 g1 reverse, CTTCCTACGGGATGTAGAGCTTGTT.

Congenically marked (CD45.1⁺⁺ and CD45.1⁺ CD45.2⁺) nucleofected OT-1 T cells were transferred intravenously to CD45.2 WT recipient mice at a 50:50 ratio in the following amounts: 1) 5,000 versus 5,000 or 10,000 versus 10,000 naive T cells for day 8 analyses, 2) 20,000 versus 20,000 naive T cells for day 6 analyses, or 3) 10,000 versus 10,000 or 20,000 versus 20,000 naive T cells for day 35+ analyses. To determine the exact input ratio, a portion of the mixed cells was stained with CD45.1 and CD45.2 Abs and assessed by flow cytometry. The mice were then infected with 2 × 10⁵ PFU LCMV-Arm or LCMV-Arm-OVA i.p. 7 d after naive T cell transfer. The mice were analyzed for the ratio of CD45.1⁺⁺ and CD45.1⁺ CD45.2⁺ cells 6 or 8 d postinfection (effector T cell studies) or greater than 30 d postinfection (memory T cell studies). These output ratios were normalized to the input ratios to determine competitive enrichments or depletions in vivo. To assess MPEC and SLEC, CD45.1⁺⁺ and CD45.1⁺ CD45.2⁺ cells were stained with CD127 and KLRG1, and the MPEC/SLEC ratio was calculated within the CD45.1⁺⁺ and CD45.1⁺CD45.2⁺ populations as % of CD127⁺KLRG1^–/% of CD127^–KLRG1⁺.

Congenically marked (CD45.1⁺⁺ and CD45.1⁺CD45.2⁺) nucleofected OT-1 T cells were transferred intravenously to separate CD45.2 WT recipient mice (10,000 per mouse). The mice were then infected with 2 × 10⁵ PFU LCMV-Arm-OVA i.p. 7 d after naive T cell transfer. Then, 30 d postinfection, congenically marked CD44⁺ CD127⁺ memory cells were isolated by FACS. These cells were mixed at a 50:50 ratio (10,000 versus 10,000) and transferred to C57BL6 naive wild-type recipients, which were infected with 2 × 10⁵ PFU LCMV-Arm-OVA i.p. 1 d after memory T cell transfer. Input ratios were recorded for the 50:50 memory cell mix. In addition, 8 d postinfection, spleens were harvested from the LCMV-Arm-OVA–infected mice and processed as above. Ratios of the congenically marked memory cell populations were analyzed and normalized to input ratios.

Congenically marked (CD45.1⁺⁺ and CD45.1⁺ CD45.2⁺) naive OT-1 CD8⁺ T cells were nucleofected with control or Chd7 gRNAs as above. Nucleofected T cells were cultured for 6 d in complete R10 medium with 1 ng/ml recombinant IL-7. After 6 d of culture, ∼50,000 of the nucleofected naive T cells were frozen at −80°C for indel/frameshift assessment. Nucleofected cells were then mixed at a 50:50 ratio (5,000 versus 5,000) and were transferred intravenously to CD45.2 WT recipient mice. The mice were then infected with 2 × 10⁵ PFU LCMV-Arm-OVA i.p. 1 d after naive T cell transfer. Then, 8 d postinfection, the mice were divided into three sets of five mice each. For each set, the spleens were isolated, mechanically dissociated, and combined. CD8⁺ T cells were enriched with Miltenyi CD8a microbeads (catalog no. 130-117-044). Then, congenically marked (CD45.1⁺⁺ and CD45.1⁺ CD45.2⁺) MPECs and SLECs were isolated by sorting and frozen at −80°C for indel/frameshift assessment (as above).

Naive CD8⁺ T cells were nucleofected with control or Chd7-targeting gRNAs. Nucleofected cells were plated in complete RPMI 1640 medium containing 1 ng/ml recombinant IL-7 for 6 d to allow for gene editing and protein dilution to occur. For some experiments, memory T cells (CD44⁺ CD127⁺) were sorted from mice that had cleared LCMV-Arm-OVA infection. The cells (at a concentration of 1 million cells/ml) were then labeled with 5 µM CellTrace Violet proliferation dye (Thermo Fisher Scientific catalog no. C34557) for 20 min at 37°C. A total of 50,000 naive CD8⁺ T cells were stimulated on plate-bound anti-CD3 (2 µg/ml, clone 145-2C11) (VWR catalog no. 103252-148) and anti-CD28 (2 µg/ml, Clone 37.51) (VWR catalog no. 103252-290) and were supplemented with 200 U/ml IL-2 (VWR catalog no. 103717-092) for 72 h. In addition, 50,000 memory CD8⁺ T cells were stimulated on plate-bound anti-CD3 (0.5 µg/ml) and anti-CD28 (0.5 µg/ml) and were supplemented with 200 U/ml IL-2 for 72 h. The cells were then stained, and proliferation was assessed by flow cytometry.

The mice were injected with 1 mg of BrdU i.p. 16 h before euthanasia and analysis. The cells were processed and stained using the BrdU flow kit (BD Biosciences catalog no. 552598).

Splenocytes were isolated as above from LCMV-Arm-OVA–infected mice and plated in complete RPMI 1640 medium. The splenocytes were stimulated with 0.1 µg/ml OVA_257–264 peptide (SIINFEKL) (VWR catalog no. 76198-080) and Monensin (1:1500 dilution) (VWR catalog no. 420701BL) for 4.5 h at 37°C. The cells were then stained with extracellular markers. To stain for cytokines, the cells were fixed and permeabilized using the eBioscience intracellular fixation and permeabilization buffer set (Thermo Fisher Scientific catalog no. 88-8824-00) and then stained with Abs to TNF-α, IFN-γ, and IL-2.

Sorted samples of ∼20,000 cells were spun for 10 min at 400g at 4°C. The cell pellets were resuspended in RLT buffer (Qiagen catalog no. 74104) with 1% 2-ME (VWR catalog no. 97064-588) and frozen at −80°C. RNA was extracted using a Qiagen RNeasy kit (Qiagen catalog no. 74104) and analyzed on a TapeStation2200 to ensure the RNA was of sufficient quality. Next, the Takara SMART-seq v4 kit (Takara catalog no. 634888) was used to enrich mRNA and create cDNA. An Illumina NexteraXT kit (Illumina catalog no. FC-131-1024) was used for creating libraries. Libraries were analyzed on a TapeStation2200 and by quantitative PCR to ensure of sufficient quality. The libraries were then pooled at equimolar ratios and run on a NextSeq500 sequencer at the Harvard Medical School Biopolymers Facility. For sequencing, we used a high-output 150-cycle kit and the following parameters: 36 bp read 1, 36 bp read 2, 10 bp index 1, and 10 bp index 2.

Raw sequencing data were mapped to the mm10 genome using STAR (31). Duplicate reads were identified and removed using PICARD: http://broadinstitute.github.io/picard/. A gene was considered detected if the average value of reads per kilobase of transcript per million reads mapped value for a given sample group was ≥0.5. The bioconductor package edgeR (32) was used to determine differentially expressed genes, which were defined as meeting the detection threshold, an absolute log₂ fold change of >0, and a false discovery rate (FDR) of ≤0.05. DESeq2 (33) was used as a secondary approach to confirm results and for scientific rigor (not shown). All detected transcripts were ranked by multiplying the sign of the fold change (±) by the –log₁₀ of the p value, which were used for gene set enrichment analysis (34).

ATAC sequencing (ATAC-seq) was performed as previously described (35). Sorted samples of 10,000 cells or less were spun for 10 min at 500g at 4°C. The supernatant was removed, and the cells were resuspended in a master mix containing: 12.5 µl of Illumina Tagment DNA buffer (part of Illumina kit: catalog no. 20034197), 2.5 µl of Illumina TDE1 Tagment DNA enzyme (part of Illumina kit: catalog no. 20034197), 0.02% Digitonin (BioVision catalog no. 2082-1), 0.1% Tween- 20 (Bio-Rad catalog no. 1610781), and 5 µl of molecular grade water. The samples were incubated at 37°C for 60 min. DNA tagmentation clean up buffer containing 5 M NaCl, 0.5M EDTA, 20% SDS, and 20 µg of Proteinase K (Teknova catalog no. P0701); molecular grade water was added to samples; and tagmentation clean-up proceeded for 30 min at 40°C. The samples were subjected to solid-phase reversible immobilization bead-based size selection using Agencourt AMPureXP beads (Beckman Coulter catalog no. A63880) such that only DNA of ∼200 to 400 bp in size was isolated. Briefly, high-molecular-weight DNA was removed by the addition of AMPureXP magnetic beads at 0.7× the sample volume for 15 min at room temperature, followed by magnetic separation of beads and supernatant and subsequent supernatant recollection. Lower-molecular-weight DNA was then removed by the addition of AMPureXP magnetic beads at 1.2× the sample volume for 15 min at room temperature followed by magnetic separation of beads and supernatant and the removal of supernatant. Beads containing the desired DNA fragments were washed with 80% ethanol twice for 30 s followed by 2 min of bead drying. The beads were then equilibrated in Tris-HCl (pH 8.0) for 2 min at room temperature followed by 2 min of magnetic separation for DNA elution. Size selection was repeated after DNA library amplification for the removal of lower-molecular-weight DNA following the same procedure, however, with a bead ratio of 1× the sample volume. DNA library amplification was achieved using Illumina-DNA/RNA UD Indexes, set A (catalog no. 20025019) at 10 µl per sample and 25 µl of 2× Kapa HiFi Hotstart polymerase ready mix (Roche catalog no. KK2601). The PCR was performed as follows: 72°C for 3 min initial extension; 98°C for 30 s initial denaturation; 12 cycles of 98°C for 10 s denaturation, 63°C for 30 s annealing, and 72°C for 30 s extension, followed by a final 72°C for 60 s final extension. DNA quality and concentration were assessed through 1) TS4150 Agilent Tapestation and Agilent tech D5000 high-sensitivity ScreenTape (Agilent Technologies catalog no. 5067-5593) or 2) quantitative PCR performed by the Harvard Medical School Biopolymers Facility. Samples with sufficient DNA concentrations were submitted for Illumina NextSeq next-generation sequencing at the Harvard Medical School Biopolymers Facility. Each sample was sequenced with ∼50 million paired end reads per sample. The sequencing parameters used were: 74 bp read 1, 74 bp read 2, 10 bp index 1, and 10 bp index 2.

Raw sequencing data were mapped to the mm10 genome using Bowtie2 (36). The peaks were called using MACS2 (37) and annotated using the HOMER program annotatePeaks (38). The bioconductor package edgeR (32) was used to determine differentially accessible regions, which were defined as any peak with an average RPM value of ≥1 for a given group and having an absolute log₂-normalized fold change of ≥1 and an FDR of ≤0.05.

Statistical analyses were performed using GraphPad Prism 9 software. The data were considered statistically significant with p values < 0.05. The following tests were used: a paired Student t test for comparing two groups that were not independent, a ratio paired Student t test for comparing two groups that were not independent and formed a ratio (as in competitive cotransfer assay), a one-way ANOVA for single comparisons with groups greater than two, and a two-way ANOVA for multiple comparisons within groups. Screen, RNA sequencing (RNA-seq), and ATAC-seq statistical analyses are described in the corresponding sections above.

The data, materials, and mouse strains that support the findings of this study are available from the corresponding author upon reasonable request. All screen sequencing data have been deposited on NCBI’s Gene Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra, identifier PRJNA844407). All RNA-seq and ATAC-seq data have been deposited on NCBI’s Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/, identifiers GSE274735 and GSE274736).

The epigenetic regulators responsible for the formation of MPEC, SLEC, central memory, and effector memory T cells are mostly unknown. To identify novel epigenetic regulators of effector and memory T cells, we performed an in vivo CRISPR-based loss-of-function screen of all predicted epigenetic regulators in mice infected with acute LCMV-Arm virus. To study complex differentiation states, we performed this screen in naive CD8⁺ T cells to 1) enable identification of genes important during T cell priming and 2) assess T cells that are physiologically primed in vivo. We used the CHimeric IMmune Editing (CHIME) system, which we previously developed to lentivirally deliver gRNAs to Cas9-expressing LSK bone marrow cells that are subsequently implanted into irradiated recipients (26, 39). These transduced LSK cells differentiate into mature immune cells with deletion of the genes targeted by the lentivirally delivered gRNAs in the recipient mice. For our screen, we used CD45.1⁺ Cas9-expressing LSK cells from P14 TCR transgenic mice, which recognize the LCMV glycoprotein Ag GP_33-41. We created our 4,400-gRNA screen library (Supplemental Table I) by obtaining mouse orthologs from a list of 1,448 human epigenetic regulators, targeting these orthologs with 3 gRNAs per gene, and including 50 nontargeting negative control gRNAs and positive control gRNAs targeting Tcf7 and Tbx21. We performed a primary screen of the 1,448 putative epigenetic regulators using CHIME by transducing CD45.1⁺ Cas9⁺ P14 LSK cells with the 4,400-gRNA lentiviral library. Transduced LSK cells, which are marked by the fluorophore Vex, were transferred into an irradiated WT host, and after reconstitution, naive Vex⁺ P14 CD8⁺ T cells were transferred to recipient mice that were subsequently infected with LCMV-Arm. At 8- and 30-d postinfection, transferred T cells were collected (Supplemental Fig. 1a), including splenic naive CD8⁺ T cells; MPECs and SLECs in the lung, inguinal lymph node, lamina propria, and spleen (day 8); central and effector memory cells in the spleen and inguinal LN (day 30); and memory cells in the lung (day 30). Genomic DNA was extracted from samples and sequenced to determine gRNA representation in the samples. We analyzed the quality of the screen by assessing recovery of gRNAs and correlation of the abundance of individual gRNAs in the mice. We found that only a fraction of the 4,400 gRNAs (∼500–1,000 gRNAs) was recovered in the majority of groups (Supplemental Fig. 1b), and the Pearson correlation coefficient among replicate mice was low (∼0.1–0.2) (Supplemental Fig. 1c). gRNA recovery was particularly low from the MPEC and SLEC groups from the lamina propria, so these groups were excluded from further analysis. Replicate gRNAs are vital for successfully distinguishing bona fide hits (40), yet for the majority of genes, we recovered one gRNA/gene. Thus, we instead prioritized genes of interest only if targeting gRNA(s) appeared enriched or depleted across multiple organs (spleen, LN, and lung). We calculated fold changes of gRNAs comparing MPEC (KLRG1^– CD127⁺) to SLEC (KLRG1⁺ CD127^–) and central (CD44⁺ CD127⁺ CD62L⁺) to effector (CD44⁺ CD127⁺ CD62L^–) memory and applied fold change cutoffs (detailed in methods) to identify 93 genes of interest (Supplemental Fig. 1d). To ensure rigor and focus on biologically significant epigenetic regulators, we performed a secondary screen to investigate these 93 genes (Fig. 1a, Supplemental Fig. 1e) with 7 main modifications compared with the primary screen: 1) we narrowed our focus to 113 genes (93 genes prioritized from the first screen with 20 positive control genes), 2) we targeted each gene with 5 gRNAs to ensure recovery of sufficient gRNAs for hit calling, 3) we included 96 distinct 6-bp oligonucleotide barcodes embedded within the tetraloop of the tracrRNA as UMIs (41, 42) to enable tracking of the distinct number of unique cells that contribute to a phenotypic result, 4) we directly transduced naive CD8⁺ T cells using pseudotyped lentivirus to bypass putative effects from KO of epigenetic regulators during T cell development (42), 5) we focused on a single timepoint for analyses (day 8) to assess the early establishment of the epigenetic signature, 6) we pooled sets of 5 mice together for a total of 3 pools, and 7) we used a stronger Ag-specific transgenic TCR to potentially improve signal-to-noise (OT-1 TCR transgenic T cells with OVA-expressing LCMV-Arm infection) (28). Notably, use of pseudotyped lentivirus to transduce naive CD8⁺ T cells enables evaluation of genes in naive nonactivated CD8⁺ T cells as in the primary CHIME screen but should circumvent effects of gene KOs during thymic selection and subsequent development of naive CD8⁺ T cells. In this secondary screen, we found that gRNA recovery was substantially improved, with the majority of the 635 gRNAs recovered in all groups (Fig. 1b). In addition, the correlation of pools of mice in the spleen was ∼0.6, suggesting sufficient coverage (cells per gRNA) to uncover robust biological changes in the spleen (Fig. 1c). The Pearson correlation coefficient was lower in the LN and lung, which is likely the result of overall lower cell numbers recovered, so we focused our analyses on the spleen and used the LN and lung only for confirmatory assessments. We next examined the fold changes of gRNAs between the MPEC and SLEC populations in the spleen to identify gRNAs that changed the MPEC to SLEC ratio relative to the control gRNAs. We recovered the expected biology of several positive control genes: KO of Foxo1 (43) and Mbd2 (44) led to an enrichment of SLECs, whereas KO of Stat4 (45) and Prdm1 (46, 47) led to an enrichment of MPECs (Fig. 1d, 1e; Supplemental Table II). In addition, we identified gRNAs targeting a novel putative regulator, Chd7 (chromodomain-helicase DNA-binding protein 7) enriched in MPECs relative to SLECs. Similar results were also observed for the lung samples (Fig. 1d).

To analyze our screen using a second approach, we used the MAGeCK algorithm (30) and found that the results were concordant (Supplemental Fig. 1f, Supplemental Table II). In contrast to an impact on the MPEC to SLEC ratio, KO of Chd7 did not affect the frequency of naive CD8⁺ T cells (Supplemental Fig. 1g). To test the rigor of the enrichment of Chd7-targeting gRNAs in MPECs relative to SLECs, we assessed the UMIs included in our screen (Fig. 1f); these UMIs enable tracking of the independent cells that contribute to enrichment or depletion of a gRNA. Given that 96 UMIs were included for each gRNA, UMI-level analyses will approximate performing the experiment (i.e., how a given KO cell differentiates) up to 96 times, providing strong confidence in the results. We found that relative to control gRNAs, Chd7-targeting gRNAs were represented by more UMIs in the MPEC population and fewer UMIs in the SLEC population (Fig. 1g). Thus, multiple individual cells with Chd7-targeting gRNAs displayed a bias toward MPEC differentiation relative to SLEC differentiation. Taken together, our secondary screen enabled the recovery of known regulators of MPEC and SLEC differentiation, and uncovered CHD7 as a novel regulator of the MPEC to SLEC ratio.

CHD7 is expressed in naive CD8⁺ T cells, day 7 effector CD8⁺ T cells, and memory CD8⁺ T cells, with highest expression in effector CD8⁺ T cells (https://www.immgen.org). Given this expression pattern and our screen results, we hypothesized that CHD7 has the potential to control naive CD8⁺ T cell differentiation into MPECs and SLECs and/or their maintenance, as well as potential memory CD8⁺ T cell differentiation. We first assessed the effects of Chd7 KO on CD8⁺ T cell differentiation into MPECs and SLECs by performing cotransfer assays, whereby two T cell populations are transferred at a 50:50 starting ratio to a recipient mouse, and T cell responses to OVA-expressing LCMV Armstrong (LCMV-Arm-OVA) infection are compared by measuring the ending ratio. This approach enables us to assess MPEC and SLEC ratios during LCMV-Arm-OVA infection in an internally controlled system (Fig. 2a) with the same viral load and tissue microenvironment. We used nucleofection to transiently deliver Cas9-gRNA RNPs to naive OT-1 CD8⁺ T cells to induce gene KO (48). This approach is favorable in circumstances in which the gRNA does not need to be integrated (nonscreen scenarios, for example) due to its speed and penetrance of KO compared with lentiviral approaches for naive CD8⁺ T cells. Following deletion of Chd7 using two gRNAs (Fig. 2b), we found that Chd7 KO OT-1 T cells had an increased MPEC to SLEC ratio compared with control OT-1 T cells (Fig. 2c, 2d). Furthermore, this increased ratio was comparable to the increased MPEC to SLEC ratio observed following KO of Tet2, a known regulator of MPEC to SLEC differentiation (14). To determine whether the change in the MPEC to SLEC ratio was driven by increased MPEC, decreased SLEC, or both, we examined cell counts and found that Tet2 KO and Chd7 KO did not affect the number of MPECs (Fig. 2e) but did lead to fewer SLECs (Fig. 2f). Moreover, in the Tet2 KO and Chd7 KO samples, we also observed a decrease in the number of CD127⁺ KLRG1⁺ cells (Supplemental Fig. 2a), a population that can downregulate KLRG1 to become exKLRG1 memory cells (7).

Because the LCMV-Arm-OVA model has altered MPEC and SLEC differentiation ratios relative to LCMV-Arm infection, we conducted further experiments to ensure that our validation results were not specific to the LCMV-Arm-OVA model. We used P14 T cells paired with LCMV-Arm infection and observed an increase in the MPEC to SLEC ratio using Chd7 KO P14 T cells (Fig. 2g). Similar to the LCMV-Arm-OVA studies, Chd7 KO and Tet2 KO did not change the number of Ag-specific MPECs but led to a decrease in the number of Ag-specific SLECs in the LCMV-Arm infection model (Supplemental Fig. 2b, 2c). Given that CRISPR-mediated KO is not 100% (Fig. 2b), it is possible that Chd7 KO T cells cannot form SLECs and that only the remaining non-KO T cells form SLECs. To investigate this possibility, we sorted out MPECs and SLECs generated from control or Chd7 KO naive CD8⁺ T cells and examined Chd7 deletion efficiency by evaluating formation of insertion-deletions and frameshifts in the Chd7 locus. We analyzed DNA from the MPECs and SLECs to assess KO because 1) frameshifts induced by CRISPR will lead to functional KO and 2) we are unable to recover sufficient MPECs and SLECs ex vivo for analyses of CHD7 protein expression by Western blot. We found that the MPECs and SLECs generated from Chd7 KO naive CD8⁺ T cells had equal or greater deletion efficiency compared with the input naive T cells (Fig. 2h), validating that Chd7 KO T cells are able to form both MPECs and SLECs. Collectively, these findings demonstrate that CHD7 is not absolutely required for SLEC formation but, following KO, reduces the formation and/or maintenance of SLECs.

Given that CHD7 is a chromatin regulator and Chd7 KO decreased SLECs numerically, we next assessed whether KO of Chd7 in CD8⁺ T cells affected their viability or function. To examine the impact of CHD7 on T cell proliferation, we created Chd7 KO or control naive CD8⁺ T cells via nucleofection and assessed proliferation by CellTrace Violet dilution following T cell stimulation with cross-linking Abs to CD3 and CD28 (Supplemental Fig. 2d). Chd7 KO cells proliferated to the same extent as control cells in vitro (Supplemental Fig. 2e, 2f). In addition, the viability of Chd7 KO T cells was slightly higher than control cells (Supplemental Fig. 2g). To determine whether KO of Chd7 affected CD8⁺ T cell cytotoxicity, we differentiated control or Chd7 KO naive CD8⁺ T cells into cytotoxic T cells in vitro. Chd7 KO and control CD8⁺ T cells had similar cytotoxic function (Supplemental Fig. 2h). Collectively, these data suggest that Chd7 KO CD8⁺ T cells do not have reduced function in vitro.

To assess the function of Chd7 KO CD8⁺ T cells in vivo, we performed cotransfer assays with control and Chd7 KO CD8⁺ T cells (Supplemental Fig. 3a). Chd7 KO CD8⁺ T cells were outcompeted numerically in vivo compared with control CD8⁺ T cells, similar to Tet2 KO T cells (Supplemental Fig. 3b). We also performed a BrdU pulse to assess proliferation and found that although Chd7 KO CD8⁺ T cells were able to proliferate in response to LCMV-Arm-OVA, they were less proliferative than control T cells (Supplemental Fig. 3c, 3d), indicating that this difference may be responsible for their relative reduction in vivo. In contrast, Chd7 KO CD8⁺ T cells had similar viability as control T cells (Supplemental Fig. 3e). We also assessed cytokine production following an ex vivo restimulation and found that Chd7 KO and control CD8⁺ T cells had similar frequencies of IFN-γ⁺ TNF-α^– and IFN-γ⁺ TNF-α⁺ populations, suggesting that CHD7 is not required for cytokine production (Supplemental Fig. 3f, 3g). Thus, CHD7 primarily affects the proliferative rate in vivo. Collectively, these in vitro and in vivo data show that CHD7 does not affect CD8⁺ T cell cytokine production or cytotoxic functions.

Given that Chd7 KO CD8⁺ T cells had reduced capacity to form SLECs, we next asked whether the Chd7 KO MPECs and SLECs differed transcriptionally from control cells. We generated control and Chd7 KO OT-1 T cells and either performed RNA-seq on these naive cells or transferred them into WT recipients that were infected with LCMV-Arm-OVA. Eight days later, we isolated control and Chd7 KO MPECs and SLECs and performed RNA-seq. As expected, the naive cells, SLECs, and MPECs displayed distinct transcriptional programs and clustered by cell type (Fig. 3a, 3b). Interestingly, Chd7 KO and control cells within each population (naive, SLEC, and MPEC) clustered together, suggesting that Chd7 KO did not significantly change the transcriptional profiles of naive cells, SLECs, or MPECs. To investigate this further, we performed differential expression analyses of control naive versus Chd7 KO naive, control MPEC versus Chd7 KO MPEC, and control SLEC versus Chd7 KO SLEC and found very few differentially expressed genes (Fig. 3c, Supplemental Table III), consistent with our clustering analyses. Despite the absence of differences between control and Chd7 KO MPECs and SLECs, we confirmed that both control and Chd7 KO MPECs and SLECs were enriched for gene signatures of MPECs and SLECs, suggesting that we were examining bona fide MPECs and SLECs (Supplemental Fig. 4a). Furthermore, the MPECs expressed canonical MPEC genes such as Slamf6, Il7r, Cd27, and Id3, whereas the SLECs expressed canonical SLEC genes such as Klrg1 and Cx3cr1 (Supplemental Fig. 4b). Collectively, these findings demonstrate that Chd7 KO T cells are able to differentiate into MPECs and SLECs with expected transcriptional programs.

Because CHD7 is a chromatin remodeler, we next asked whether control and Chd7 KO CD8⁺ T cells differed epigenetically. Similar to our RNA-seq analysis, we generated control and Chd7 KO OT-1 T cells and then either analyzed the naive cells by ATAC-seq or transferred them to recipient animals, which were infected with LCMV-Arm-OVA. We isolated MPECs and SLECs for ATAC-seq analysis on day 8 postinfection. Naive T cells clustered separately from MPECs and SLECs, irrespective of whether they were control or Chd7 KO samples (Fig. 3d, 3e). We also assessed differentially accessible regions for control naive versus Chd7 KO naive, control MPEC versus Chd7 KO MPEC, and control SLEC versus Chd7 KO SLEC. Remarkably, very few differentially accessible regions were observed in the Chd7 KO versus control cells (Fig. 3f, Supplemental Table IV). However, we did identify differential regions in the Il7r and Klrg1 loci when comparing MPECs versus SLECs, as expected (Supplemental Fig. 4c). Thus, Chd7 KO MPECs and SLECs are still able to remodel chromatin into the profiles characteristic of MPECs and SLECs.

Given the lack of transcriptional and epigenetic differences between control and Chd7 KO MPECs and SLECs, we hypothesized that CHD7 may play an earlier role during the differentiation of MPECs and SLECs, and by day 8 postinfection a subset of the Chd7 KO T cells is able to form bona fide SLECs. To test this hypothesis, we cotransferred control and Chd7 KO OT-1 T cells to recipients that we subsequently challenged with LCMV-Arm-OVA and evaluated at day 6 postinfection. The number of Chd7 KO SLECs was significantly lower than control SLECs at day 6 postinfection (Fig. 3g), suggesting that CHD7 functions earlier postinfection. In addition, we examined the viability of the SLECs at day 6 postinfection and found that the control and Chd7 KO SLECs had comparable annexin V staining, indicating that CHD7 is not required for SLEC viability (Fig. 3h). Taken together, these data suggest that CHD7 is not required for establishing the transcriptional and epigenetic profiles of MPECs and SLECs but that CHD7 is required for forming and/or maintaining SLECs numerically (Fig. 3i).

Given that KO of Chd7 in CD8⁺ T cells increased the MPEC to SLEC ratio and promoted formation of central memory T cells in our primary screen, we further investigated whether KO of Chd7 altered CD8⁺ T cell memory (Fig. 4a). We first assessed whether KO of Chd7 conferred a numerical advantage, relative to control T cells, at a memory timepoint, 35 d after LCMV-Arm-OVA infection. Chd7 KO CD8⁺ T cells were enriched compared with controls in the spleens of mice that had cleared LCMV-Arm-OVA infection (Fig. 4b, 4c). We also examined expression of CD62L, CD27, and CX3CR1 to distinguish central (CD44⁺ CD62L⁺ CD27^hi) and effector (CD44⁺ CX3CR1⁺) memory T cells. Chd7 KO promoted formation of CD62L⁺ and CD27⁺ cells and attenuated formation of CX3CR1⁺ cells (Fig. 4d–f, Supplemental Fig. 5a–c), suggesting that CHD7 inhibits the formation of central memory CD8⁺ T cells. Chd7 KO CD8⁺ T cells had comparable frequencies of IFN-γ–producing cells and IFN-γ/TNF-α–producing cells to control T cells (Fig. 4g, Supplemental Fig. 5d). However, consistent with an increase in central memory T cells, Chd7 KO CD8⁺ T cells had increased polyfunctional cytokine production (IFN-γ, TNF-α, and IL-2) (Fig. 4g, Supplemental Fig. 5d). In addition, we assessed the cytotoxicity of control and Chd7 KO memory CD8⁺ T cells in vitro and found that CHD7 did not alter cytotoxicity (Fig. 4h).

To examine whether KO of Chd7 affected secondary responses, we generated control or Chd7 KO CD8⁺ T cells and transferred them to recipient mice, which we infected with LCMV-Arm-OVA. We isolated control or Chd7 KO CD44⁺ CD127⁺ memory CD8⁺ T cells 35 d postinfection (Fig. 4i) and transferred these cells in a 50:50 ratio to naive mice that were subsequently infected with LCMV-Arm-OVA. Chd7 KO memory CD8⁺ T cells were outcompeted by control memory T cells by ∼3-fold (Fig. 4j). Taken together, CHD7 inhibits the generation of central memory CD8⁺ T cells but promotes memory CD8⁺ T cell expansion during a secondary response.

In response to an acute infection, CD8⁺ T cells differentiate into SLECs, MPECs, and multiple memory T cell subsets, which are distinct functionally, transcriptionally, and epigenetically. Several epigenetic regulators have been implicated in programming the distinct epigenetic profiles of these subsets, yet our understanding of the necessary regulators is incomplete. In this study, we used two in vivo CRISPR screens to identify the epigenetic regulators controlling these cell fate decisions. We found that KO of the ATP-dependent chromatin remodeler CHD7 led to an increase in the MPEC to SLEC ratio, driven by a decrease in the number of SLECs. Unexpectedly, we found that despite a decrease in SLECs numerically, the resultant Chd7 KO SLECs were epigenetically and transcriptionally similar to control SLECs. Moreover, when we examined Chd7 KO CD8⁺ T cells at memory timepoints, we found that KO of Chd7 led to an increase in the formation of central memory T cells but attenuated memory T cell expansion, following a rechallenge. Overall, these data implicate CHD7 as a novel regulator of CD8⁺ T cell differentiation during an acute viral infection.

CHD7 is expressed in SLECs and MPECs, yet we observed a specific effect of KO of Chd7 on SLEC numbers and not on MPEC numbers. The reasons for this specificity are unclear, although CHD7 has been implicated as a regulator of cell fate specification in development (49). Given that CHD7 binds to methylated histones and not specific DNA sequences (50), it is plausible that SLEC-specific or SLEC-enriched transcription factors, such as T-box transcription factor 21 (TBX21) (2), are responsible for the preferential recruitment and requirement for CHD7 in SLECs. Through interaction with SLEC-enriched transcription factors, CHD7 may be important for the initial commitment, maintenance, or proliferation of SLECs. In addition, the similar epigenetic and transcriptional programs in control and Chd7 KO SLECs suggest potential compensation of chromatin remodeling in SLECs and functional redundancy for CHD7 targets. CHD7 is part of the nine-member chromodomain helicase DNA-binding domain (Chd) family (51); thus, it is possible that another member of the Chd family or Chd5-9 subfamily can compensate for CHD7 loss. CHD2 was targeted in our secondary screen and, following KO, led to a 1.4-fold increase in the MPEC to SLEC ratio and a 0.8-fold decrease in the SLEC to input ratio. Future work could examine 1) whether CHD2 or additional members of the Chd family compensate for loss of CHD7 and 2) why CHD7 is the preferred choice for chromatin remodeling, because optimal compensation following KO of Chd7 would have prevented the decrease observed in SLEC numbers. The majority of our analyses of SLECs were at day 8 postinfection, but we also found that Chd7 KO SLECs were reduced numerically at day 6 postinfection. Given these data and our compensation hypothesis, CHD7 may be functioning early in the differentiation process, suggesting that earlier single-cell ATAC-seq analyses may illuminate epigenetic differences between control and Chd7 KO SLECs. Overall, these data demonstrate that CHD7 is a crucial positive regulator of SLEC formation and/or maintenance.

KO of Chd7 also affected memory CD8⁺ T cell responses, resulting in a decrease in the capacity of memory T cells to expand in response to a rechallenge (Fig. 4j), consistent with the depletion of Chd7 KO T cells during the effector response (Supplemental Fig. 3b). Moreover, Chd7 KO affected distinct memory T subsets differentially, leading to an increase in central memory T cells and a decrease in effector memory T cells. We did not examine the impact of CHD7 on tissue-resident memory (Trm) T cells but hypothesize that Chd7 KO would lead to a decrease in Trm responses, given the importance of Trm cell proliferation and expansion for secondary responses (52). Taken together, CHD7 may be important during periods of rapid T cell expansion in vivo.

CHARGE (Coloboma, Heart defects, Atresia of the choanae, Retardation of growth and/or development, Genitourinary abnormalities, and Ear anomalies) syndrome is a developmental disorder associated with mutations in CHD7 in 90% of patients (53). In some CHARGE patients, T cells, including CD8⁺ T cells, are reduced; however, further work is needed to understand T cells in CHARGE patients, given conflicting findings in different patient cohorts (54–56). The reduction in T cell numbers is thought to be the result of thymic aplasia and, in patients with more severe thymic aplasia, T cell lymphopenia is more severe (57). Consistently, knockdown of Chd7 in zebrafish resulted in a defect in thymic organogenesis and subsequent T cell development (58). Our KO approaches bypass T cell development and enabled us to examine the cell-intrinsic role of CHD7 in naive, activated, and memory CD8⁺ T cells. Interestingly, we did not observe a depletion of Chd7 KO naive CD8⁺ T cells nor a change in the transcriptional or epigenetic program of naive CD8⁺ T cells relative to controls following KO of Chd7. Instead, we found a specific defect in the differentiation into CD127^– KLRG1⁺ SLECs and CD127⁺ KLRG1⁺ cells. Collectively, our data demonstrate that Chd7 KO after thymic development does not lead to lymphopenia but does affect subsequent effector and memory T cell differentiation.

In summary, these findings implicate CHD7 as a positive regulator of SLEC formation and a negative regulator of central memory T cell formation, contributing to our understanding of CD8⁺ T cell differentiation. Moreover, by uncoupling the cell-intrinsic role of CHD7 in naive T cells from changes in T cell maturation and thymic development, our findings may inform future studies of the consequences of CHD7 mutations in T cells in patients with CHARGE syndrome. Lastly, our study demonstrates the power of performing in vivo CRISPR screens focused on T cell differentiation using naive T cells as a starting population.

A.H.S. has patents/pending royalties on the PD-1 pathway from Roche and Novartis. A.H.S. is on advisory boards for Elpiscience, Monopteros, Alixia, IOME, Corner Therapeutics, Bioentre, GlaxoSmithKline, Amgen, and Janssen. She also is on scientific advisory boards for the Massachusetts General Cancer Center, Program in Cellular and Molecular Medicine at Boston Children’s Hospital, the Human Oncology and Pathogenesis Program at Memorial Sloan Kettering Cancer Center, the Johns Hopkins Bloomberg Kimmel Institute for Cancer Immunotherapy, Perlmutter Cancer Center at New York University and the Gladstone Institute and is an academic editor for the Journal of Experimental Medicine. A.H.S. receives research funding from AbbVie, Taiwan Bio, and Calico Life Sciences unrelated to this project. N.H. holds equity in and advises Danger Bio/Related Sciences, is on the scientific advisory board of Repertoire Immune Medicines and CytoReason, owns equity in BioNTech, and receives research funding from Bristol Myers Squibb and Calico Life Sciences. J.G.D. consults for Microsoft Research, Abata Therapeutics, Servier, Maze Therapeutics, BioNTech, Sangamo, and Pfizer. J.G.D. consults for and has equity in Tango Therapeutics. J.G.D. serves as a paid scientific advisor to the Laboratory for Genomics Research, funded in part by GlaxoSmithKline. J.G.D. receives funding support from the Functional Genomics Consortium: AbbVie, Bristol Myers Squibb, Janssen, Vir Biotechnology, and Merck Sharp & Dohme LLC, a subsidiary of Merck & Co., Rahway, NJ. J.G.D.’s interests were reviewed and are managed by the Broad Institute in accordance with its conflict-of-interest policies. The other authors have no financial conflicts of interests.

We thank Dr. Bradley Bernstein (Dana-Farber Cancer Institute) for advice on epigenetic regulators for the CRISPR screens and the Genetic Perturbation Platform at the Broad Institute for library generation, sequencing, and deconvolution of CRISPR screens. The authors thank the Center for Computational and Integrative Biology at Massachusetts General Hospital for the use of the Center for Computational and Integrative Biology DNA Core Facility (Cambridge, MA), which performed CRISPR sequencing for indel assessment. We thank the Harvard Medical School Biopolymers Facility (Boston, MA), which performed next-generation sequencing.