PD-L1ATTAC mice reveal the potential of depleting PD-L1 expressing cells in cancer therapy

Antibodies targeting the PD-1 receptor and its ligand PD-L1 have shown impressive responses in some tumors of bad prognosis. We hypothesized that, since immunosuppressive cells might present several immune checkpoints on their surface, the selective elimination of PD-L1 expressing cells could be efficacious in enabling the activation of antitumoral immune responses. To address this question, we developed an inducible suicidal knock-in mouse allele of Pd-l1 (PD-L1ATTAC) which allows for the tracking and specific elimination of PD-L1-expressing cells in adult tissues. Consistent with our hypothesis, elimination of PD-L1 expressing cells from the mouse peritoneum increased the septic response to lipopolysaccharide (LPS), due to an exacerbated inflammatory response to the endotoxin. In addition, mice depleted of PD-L1+ cells were resistant to colon cancer peritoneal allografts, which was associated with a loss of immunosuppressive B cells and macrophages, concomitant with an increase in activated cytotoxic CD8 T cells. Collectively, these results illustrate the usefulness of PD-L1ATTAC mice for research in immunotherapy and provide genetic support to the concept of targeting PD-L1 expressing cells in cancer.


Activation of T-cells involves binding of the T cell receptor (TCR) on T cells to peptide-bound major histocompatibility complexes (MHC) on antigen presenting cells (APCs). This activating signal is modulated by
membrane-bound co-stimulatory receptors that potentiate the response or by coinhibitory receptors which limit self-damage [1]. The discovery of immune checkpoint mediated by PD-1 and CTLA-4 receptors, and that targeting these pathways potentiates the antitumoral capabilities of our immune system led to the Nobel Prize in Medicine in 2018 [2]. Among all cancer immunotherapy strategies, antibodies targeting the PD-1/PD-L1 interaction have been the most intensively evaluated in clinical trials and are currently approved for a wide range of malignancies including melanoma, non-small cell lung cancer, Hodgkin´s lymphoma, head and neck squamous cell carcinoma or solid tumors presenting microsatellite instability (MSI) [3,4]. Despite the undisputable success of this therapies, unfortunately only 20-40% of the patients respond to the therapy and even fewer show long-term responses [5,6]. In addition, resistance to therapy, intrinsic or acquired, is also a frequent clinical finding in patients treated with immune checkpoint inhibitors (ICIs) [7]. Consistently, current efforts are directed to identify strategies that increase the percentage of patients that respond to cancer immunotherapy, and the efficacy of these treatments. AGING We hypothesized that, since PD-L1 expressing APCs might display additional immune checkpoint mediators on their membranes, their elimination could also exert antitumoral potential. In fact, recent studies have reported that chimeric antigenic receptor T (CAR-T) cells targeting PD-L1 reduce the growth of solid tumor xenografts in mice [8,9]. To provide genetic proof-ofprinciple support for the validity of this approach, we generated mice carrying an inducible suicidal reporter allele of PD-L1. Besides its usefulness to identify and isolate PD-L1 expressing cells (PD-L1 + , hereafter) from adult mouse tissues, our work with these mice reveals that the selective elimination of PD-L1 + cells potentiates immune responses against different stimuli such as bacterial endotoxins or immunogenic cancer cells.

Generation of an inducible suicidal mouse model of PD-L1
To generate an inducible suicidal reporter allele of PD-L1, we used the previously developed ATTAC (apoptosis through targeted activation of caspase 8) strategy [10]. In brief, we knocked in EGFP at the start codon of the mouse PD-L1 gene (Cd274), followed by FLAG-tagged catalytic domains of human caspase 8 fused to dimerizing serial FKBP domains which expression is driven by an IRES ( Figure 1A). This PD-L1 ATTAC allele allows for the identification of PD-L1 + cells on the basis of EGFP expression, as well as their selective killing through apoptosis upon treatment with the FK102 analogue AP20187. After identifying successfully recombined mouse embryonic stem cell (mESC) clones by Southern Blotting ( Figure 1B) and before proceeding into generating mice, we first wanted to further confirm the correct integration of the allele by evaluating EGFP expression. To do so, we used the synergistic activation mediator (SAM) strategy, which enables CRISPR-dependent transcriptional activation of a selected gene [11]. Upon lentiviral transduction of PD-L1 ATTAC heterozygous mESC (PD-L1 AT/+ ) with an sgRNA targeting the Cd274 promoter, EGFP expression was detected throughout the infected population ( Figure 1C). PD-L1 AT/+ mESC were then used to generate mice using standard procedures and crosses between PD-L1 ATTAC heterozygous mice yielded PD-L1 +/+ , PD-L1 AT/+ and PD-L1 AT/AT animals at Mendelian rations. Mutant mice showed no apparent phenotype when compared to wild type (wt) littermates ( Figure 1D). However, and consistent with the fact that PD-L1 AT/AT animals are knockouts for Pdl1, allografts of B16-F10 melanoma cells presented more immune infiltrates and grew less when implanted in PD-L1 AT/AT mice (Supplementary Figure 1).

In vitro validation of the PD-L1 ATTAC model
To evaluate the usefulness of the system in vitro we first generated mouse embryonic fibroblasts (MEF). Western Blotting (WB) revealed that a treatment with interferon gamma (IFNγ), a known inducer of PD-L1 expression [12,13], triggered expression of EGFP in PD-L1 AT/+ and PD-L1 AT/AT but not in wt MEF ( Figure 2A). Conversely, PD-L1 expression was detectable in wt and PD-L1 AT/+ MEF but not in homozygous mutants, which is expected as the construct was inserted at the start codon and is thus a knockout allele ( Figure 2A). Equivalent findings were made by immunofluorescence (IF) ( Figure 2B) or flow cytometry ( Figure 2C). Moreover, analysis of flow cytometry data revealed a full correlation between EGFP and PD-L1 expression in IFNγ-treated MEF ( Figure 2D). In what regards to the cell killing induced by AP20187 and, to our surprise, we could only detect a significant impact in cell death if PD-L1 AT/+ or PD-L1 AT/AT MEF were previously treated with IFNγ to trigger PD-L1 expression and also grown in low serum media (0.1% FBS) ( Figure 2E). In contrast, AP20187 failed to significantly induce cell death if MEF were grown in media containing 10% FBS ( Figure 2E). We hypothesized that this could be due to the ATTAC system being particularly efficient in killing nongrowing cells as these cells might have a lower threshold for triggering apoptosis. In support of this view, we want to note that despite the usefulness of this strategy it has only been previously used to target nondividing cells such as adipocytes, pancreatic islet beta cells or senescent cells [10,14,15].
Next, and given that immune cells are the main source of PD-L1 in vivo, we isolated splenocytes from adult mice, stimulated B cells with LPS and treated the cultures with IFNγ to induce PD-L1 expression. These experiments revealed a clear population of EGFP positive B cells (identified on the basis of B220 expression), which was selectively killed upon treatment with AP20187 ( Figure 2F). Moreover, this cell killing could be prevented with a pan-caspase inhibitor, confirming that cell death was due to apoptosis ( Figure 2F). Collectively, these data indicate that the PD-L1 ATTAC allele is efficient for the tracing of PD-L1 + cells, as well as for enabling their clearance in several cell types such as LPS-stimulated B cells or MEF.

In vivo validation of the PD-L1 ATTAC model
To evaluate the usefulness of the PD-L1 ATTAC allele as a reporter of PD-L1 expression in vivo, we first stained tissues of adult mice with an anti-EGFP antibody. As expected, EGFP expression was highest in organs from the immune system such as the spleen, lymph nodes, bone marrow or the thymus. In addition, scattered expression could also be detected in other organs such as the intestine, lungs or liver, while no expression was seen in the kidneys, pancreas, or hearts ( Figure 3A and Supplementary Figure 2A). Similar observations were also made by WB ( Figure 3B). Importantly, dual staining in lungs from wt and PD-L1 AT/+ animals identified that cells with cytoplasmic EGFP expression also presented PD-L1 on their membranes ( Figure 3C), further supporting the reporter nature of the introduced mutation. Furthermore, while EGFP expressing cells were readily seen in homozygous PD-L1 AT/AT lungs, these cells lacked PD-L1 expression, consistent with the null nature of the allele ( Figure 3C). Immunofluorescence experiments further identified cells expressing both PD-L1 and EGFP in lungs from PD-L1 AT/+ mice ( Figure 3D).
In what regards to the inducible-suicidal properties, we first evaluated the effects of an intraperitoneal (i.p.) administration of AP20187 in PD-L1 ATTAC mice. The treatment was particularly efficacious in killing EGFP+ cells in the peritoneum, although we also saw a significant effect in the lungs ( Figure 3E). In contrast, this approach had no significant impact in reducing the percentage of EGFP expressing cells in the blood, thymus, spleen or lymph nodes ( Figure 3E). FACS analyses confirmed a very efficient depletion of EGFP expressing cells from the peritoneum of PD-L1 AT/+ and PD-L1 AT/AT mice after treatment with AP20187 ( Figure 3F). We also evaluated if an intravenous (i.v.) administration of the drug could have more widespread effects. However, while i.v. delivery of AP20187 led to a significant depletion of EGFP+ cells in the blood and bone marrow, we failed to see significant effects on other tissues (Supplementary Figure 2B). On the basis of these results, we decided to focus in the adult peritoneum as a model where to study the impact of selectively targeting PD-L1 + cells.

Depletion of PD-L1 expressing cells sensitizes mice to LPS
Intraperitoneal injection of the bacterial lipopolysaccharide (LPS) is a widely used experimental model of a lethal septic shock associated to a cytokine containing IFNγ (10 ng/ml) and treated or not with AP20187 (100 nM). Cells were cultured in normal or low-serum media for 24 hours. The day after, cells were exposed or not to AP20187 for 72 hours. (F) FACS analyses of B220 and EGFP expression of splenocytes from PD-L1 +/+ and PD-L1 AT/+ mice cultured in IFNγ (10 ng/ml), LPS (10 ng/ml) and M-CSF (10 ng/ml) for 24 hours before exposition to AP20187 (100 nM) and caspase inhibitor I (20 μM) for 24 hours. Percentage of B220+ EGFP+ cells is shown. AGING storm [16]. Strikingly, while i.p. injections of AP20187 for three days did not affect LPS-mortality in wild type mice, it led to a significant reduction of the survival of PD-L1 AT/+ animals ( Figure 4A, 4B). This effect was even more pronounced in PD-L1 AT/AT mice with all animals dying by 18 hrs after LPS injection (with no wt animals being dead at this timepoint) ( Figure 4C). Consistent with survival data, treatment of PD-L1 AT/+ and PD-L1 AT/AT mice with AP20187 triggered a higher accumulation of the inflammatory cytokine IL-6 in the plasma of LPSinjected mice, confirming the increased severity of the septic shock (Supplementary Figure 3A-3C). Moreover, immunohistochemistry (IHC) analyses revealed a clear accumulation of immune infiltrates in the lungs or livers from AP20187-treated PD-L1 AT/+ and PD-L1 AT/AT mice exposed to LPS ( Figure 4D and Supplementary Figure  3D). Together, these data illustrate that the selective elimination of PD-L1 + cells increases the severity of immune responses in the mouse peritoneum.
To determine which changes in cell types were responsible for the observed effects, we conducted single cell RNA sequencing analyses of intraperitoneal cells from untreated or AP20187-treated PD-L1 AT/+ mice. These analyses revealed a drug-induced depletion of B cells and macrophages, concomitant to an accumulation of neutrophils and cytotoxic CD8/NK cells ( Figure 4E and Supplementary Figure 3E). Moreover, Gene Set Enrichment Analyses (GSEA) indicate that CD8 cells from AP20187-treated PD-L1 AT/+ mice were hyperactivated, as revealed by the significant activation of several pathways such as those related to TNFα, IFNγ or IL-6 signaling, as well as a general activation of the inflammatory response ( Figure 4F and Supplementary Figure 3F, 3G). Hence, depletion of PD-L1 + cells alters the intraperitoneal immune repertoire which includes the accumulation of activated cytotoxic CD8 cells.

Depletion of PD-L1 + cells increases survival to peritoneal tumor allografts
Finally, and given that cytotoxic T cells are thought to be the main effectors in the context of anti-PD-L1 immunotherapies [17], we tested the impact of depleting PD-L1 + cells in of cancer. To this end, we used a model of intraperitoneal metastasis by the highly immunogenic colon adenocarcinoma cell line MC-38, which can be used for allografts in immunocompetent mice and is sensitive to anti-PD-L1 therapies [18]. Furthermore, dissemination of cancer cells into the peritoneum is frequent in digestive and gynecological cancers and is associated with poor prognosis [19], highlighting the relevance of the chosen model.
To conduct these experiments, we first generated a MC-38 clone harboring constitutive expression of firefly luciferase which enables monitoring tumor progression by intravital imaging (MC-38 luc ). Intraperitoneal injection of MC-38 luc cells led to a lethal disease associated to the dissemination and growth of cancer cells in the peritoneal cavity ( Figure 5A). Treatment with AP20187 did not affect the progression of the disease in wt mice (Supplementary Figure 4A). Strikingly, while all PD-L1 AT/+ mice were dead within the first month after being intraperitoneally injected with MC-38 luc cells, a prior treatment with AP20187 prior to the injection of MC-38 cells led to survival of half of the animals ( Figure 5B). Consistently, intravital imaging showed a clear reduction of MC-38 cells in the peritoneum of AP20187-treated PD-L1 AT/+ mice ( Figure 5C). Furthermore, and similar to our previous observations in the LPS-sepsis model, intraperitoneal tumors from AP20187-treated presented a significant accumulation of infiltrating CD8 lymphocytes, highlighting the increased immune response to the tumor ( Figure 5D). Besides this prevention model, AP20187 treatment also increased the survival of PD-L1 AT/+ mice if the drug was administered 4 days after the injection of MC-38 cells (treatment model) (Supplementary Figure  4B-4D). Together with the data from the LPS-induced septic shock, these experiments indicate that the selective targeting of PD-L1 + cells intensifies immune responses in the peritoneum, which in the context of cancer is protective and favours the clearance of tumor cells.
In summary, we here present PD-L1 ATTAC as a reporter and inducible suicidal allele of mouse PD-L1. The reporter EGFP enables the identification and isolation of PD-L1 + cells from adult tissues, which we believe is useful as antibodies detecting mouse PD-L1 often give signal in PD-L1 deficient samples. As for the inducible-suicidal strategy, our experiments indicate that this effect is significantly influenced by the administration route for AP20187 and growth rates of the cells, which we wonder to what extent could also have influenced previous studies using the ATTAC strategy. Despite these limitations, i.p. administration of AP20187 yields a very efficient depletion of PD-L1 + from the peritoneum, enabling functional studies to investigate the impact of depleting PD-L1 + cells in vivo. Of note, no significant toxicities were observed in mice that were i.p. treated for up to 3 months with 3 weekly doses of AP20187. Our results confirm that the selective elimination boosts immune responses in the peritoneum, which prolongs survival against a model of peritoneal cancer metastasis. Hence, our work supports the usefulness of targeting PD-L1 expressing cells in cancer therapy, and provides the immunotherapy research community with a useful genetic tool for investigations on the PD-1/PD-L1 checkpoint in mice.

Mouse models
The PD-L1 ATTAC targeting vector was generated by recombineering (Genebridges, Germany). Recombinant ES cells were screened by Southern Blot through standard procedures, and subsequently used for the generation of chimaeras by aggregation. Animals were genotyped by PCR using the following primers (UP: 5′-TTGCTTCAGTTACAGCTGGCTCG-3′; Down_WT: 5′-CGTAGCAAGTGACAGCAGGCTG-3′; Down_ MUT: 5′-GCCGTTTACGTCGCCGTCCAG-3′). Mice were kept under standard conditions at specificpathogen free facility of the Spanish National Cancer Centre in a mixed C57BL/6-129/Sv background. 9-12week-old mice were used for all experiments. All mouse work was performed in accordance with the Guidelines for Humane Endpoints for Animals Used in Biomedical Research, and under the supervision of the Ethics Committee for Animal Research of the "Instituto de Salud Carlos III".

LPS-induced septicemia
Mice were injected i.p. with AP20817 (2.5 mg/kg) for 3 consecutive days before an i.p. injection of LPS (10 mg/kg, Sigma-Aldrich, #L2630) resuspended in PBS. Blood, plasma and tissues were isolated for further IHC and ELISA analyses. Mice were monitored for overall health and for a week after LPS injection.

CONFLICTS OF INTEREST
The authors declare no conflicts of interest related to this study.

ETHICAL STATEMENT
Mice were maintained in a mixed C57BL/6-Sv background under standard housing conditions with free access to chow diet and water, as recommended by the Federation of European Laboratory Animal Science Association. All mice work was performed in accordance with the Guidelines for Humane Endpoints for Animals Used in Biomedical Research, and under the supervision of the Ethics Committee for Animal Research of the "Instituto de Salud Carlos III", following the procedures detailed in the approved ethics protocol (PROEX 264/19).

Cell culture
293T cells (ATCC) were grown in DMEM supplemented with 10% FBS and 1% Penicillin/ Streptomycin and transfected in media containing no antibiotics. MC-38 cells (kind gift from Eduard Batllé) were cultured in DMEM supplemented with 10% FBS and 1% Penicillin/Streptomycin. For the generation of an MC-38 clone expressing luciferase, MC-38 cells were transduced with "pLenti CMV B5-Luc Blast" lentiviruses (Addgene, #21474). MEFs isolated from 13.5 d.p.c. embryos were cultured in DMEM (Sigma, D5796) supplemented with 10%-15% FBS (Sigma) and 1% Penicillin/Streptomycin (Gibco, #11548876) in lowoxygen conditions. Cells were passaged every 3 or 4 days in a 1:4 ratio. IFNγ (Peprotech, #315-05) was added to culture media at 10, 50 or 100 ng/ml concentration. For the experiments assessing the effect of AP20187, MEFs were plated in DMEM containing 10% FBS for 2 days in the presence or absence of IFNγ (10 ng/ml). To evaluate the effect of the drug in non-growing cells, MEF were washed twice with PBS and grown in lowserum media (0.1% FBS) for 3 days in the presence or absence of IFNγ. In all cases, AP20187 was added to culture media at a final concentration of 100 nM for the last 3 days. Resting splenic B cells were isolated and cultured in IFNγ (10 ng/ml), LPS (10 ng/ml) (Sigma, L2630) and M-CSF (10 ng/ml) for 24 hours before exposition to AP20187 (100 nM) with or without the caspase inhibitor I (20 μM) (Sigma-Aldrich, #627610) for 24 hours.

Lentiviral infection
Lentiviruses were produced by transfection of 293T cells with lentiviral transfer vectors and the packaging plasmids pMDL, pRev and VsVg at a 1:0.65:0.25:0.35 ratio. Transfection was performed using Lipofectamine 2000 (ThermoFisher, #11668019) and Opti-MEM Reduced Serum Medium (Gibco, #31985070) as recommended by the manufacturer. Viral supernatants were collected 48 h following transfection, filtered through a 0.45 μm filter and either added to target cells or frozen at −80°C.

Flow cytometry
Cells were trypsinized, pelleted and resuspended in PBS with DAPI. For experiments using AP20187, culture media and the PBS from the washing steps were also collected to include dead cells in the analyses. Cells were blocked with PBS +1% BSA (Roche #10735086001) for 30 minutes and stained with an anti-mPD-L1 antibody antibody (Biolegend, 124313). mPD-L1 and EGFP expression were then evaluated by flow cytometry with a FACSCanto II (BD Biosciences), and data analysed with FlowJo (BD Biosciences).

Immunoblotting
For WB analyses, cells were washed once with PBS, and lysed in RIPA buffer (Tris-HCl 50 mM, pH 7.4, NP-40 1%, Na-deoxycholate 0.25%, NaCl: 150 mM, EDTA 1 mM) containing protease and phosphatase inhibitors (Sigma-Aldrich). Samples were resolved by SDS-PAGE and analyzed by standard WB techniques. Primary antibodies against GFP, PD-L1, GAPDH and β-ACTIN were used (see the list at the end of the Methods section with the references to the antibodies). Protein blot analyses were performed on the LICOR platform (BD Biosciences).

Tissue immunofluorescence
Immunofluorescence on tissue sections was carried out as previously described [2]. Briefly, after antigen retrieval, the sections are rinsed, and permeabilised in PBS containing 0.25% Triton X100 and 0.2% gelatine. After that, they were blocked in 5% BSA in permeabilisation buffer. Sections were then incubated with the primary antibodies in 1% BSA overnight at room temperature, and with the secondary antibodies for 1 h at room temperature. Finally, after rinsing, the sections are incubated in 10 mM CuSO4/50 mM NH4Cl solution, dried, and mounted with ProLong Gold antifade (ThemoFisher, P10144) mounting media. Images were captured with a Leica SP5 WLL confocal microscope. A 40× magnification lens was used and images were taken at non-saturating conditions.

Anti-mouse PD-L1 mAb
A new anti-mouse PD-L-1 mAb (clone GOYA536A) was produced by immunizing Wistar rats with HEK293expressed extracellular domain (ec) of PD-L1 fused to Fc fragment. Wistar rats (Charles River Laboratories, France) were injected intraperitoneally (three times at 14-day intervals) with 100 μg of ecPDL1-Fc and Complete Freund's adjuvant (Difco). A 150 μg last booster of the recombinant ecPD-L1-Fc protein was injected intraperitoneally and splenocytes were isolated and fused 3 days later. Hybridoma supernatants were screened by ELISA using HEK293T cells transfected with the PCMV6-mPDL1-MYC-DDK plasmid (Origene, #MR203953). The rat mAb that was raised against mouse PD-L1 (clone GOYA536A) was cloned by the limiting dilution technique. All animal experiments were performed under the experimental protocol approved by the Institutional Committee for Care and Use of Animals from Consejería de Medio Ambiente y Ordenación del Territorio of the Comunidad de Madrid (Madrid, Spain) with reference number PROEX62.3/20. All efforts were made to minimize animal suffering.

Droplet based single-cell mRNA sequencing
Peritoneal cavity cells were obtained by peritoneal lavage following standard procedures. Cells were collected in cold PBS containing 5% FBS and 1 mM EDTA to preserve cell viability. Cell suspension was washed with PBS containing BSA and filtered through a 40 μm cell strainer. Viable cells were enriched by magnetic separation using the Dead Cell Removal Kit (Miltenyi Biotec #130-090-101). Isolated cells were next washed and finally suspended in PBS-0.04% ultrapure BSA (Invitrogen #AM2616) at 1000 cells/μl, and shown to have a viability higher than 95% by Trypan Blue exclusion (Gibco #15250061). Approximately 10 4 cells from each sample were loaded onto a 10X Chromium Single Cell Controller chip B (10× Genomics) following manufacturer's instructions (Chromium Single Cell 3ʹ GEM, Library and Gel Bead Kit v3, PN-1000075). Generation of gel beads in emulsion (GEMs), barcoding, GEM-RT clean up, cDNA amplification and library construction were all performed according to manufacturer's recommendations. Libraries were sequenced in an asymmetrical pair-end format, with 28 bases for read 1 and 56 for read 2 in a NextSeq550 instrument (Illumina) with v2.5 reagent kits.

scRNAseq data analysis
Bcl files were converted to fastq format with cellranger mkfastq (10× Genomics), a wrapper for bcl2fastq (Illumina), and subsequently analysed with/by the bollito pipeline [3]. Reads were aligned to the Gencode mouse reference (GRCm38) and quantified using the STARsolo aligner [4]. The Seurat toolkit [5] was used to perform the cell-based quality control, normalisation, integration and clustering steps. We filtered out cells with less than 750 genes detected and high mitochondrial (>10%) and ribosomal (>40%) gene content. Cells with more than 4000 detected genes were considered doublets. Low-abundance genes, those expressed in less than 2 cells, were also removed from the dataset. A total of 14347 cells were recovered. Samples were normalised using the sctransform approach [6] and integrated. The first 20 components were selected to cluster the samples and a Uniform Manifold Approximation and Projection (UMAP) was applied for their visualisation. We used the SingleR [7] annotation scores to guide the annotation of each cluster. Differential gene expression was performed using Seurat's Wilcox test. Genes expressed in less than 25% of the cells were filtered out. The estimated significance level (P value) was corrected to account for multiple hypotheses testing using Benjamini and Hochberg False DiscoveryRate (FDR) adjustment. Genes with FDR less than or equal to 0.05 were selected as differentially expressed. Significantly upregulated or downregulated genes were introduced into PANTHER