DISCUSSION
Although the role of the Fc region of anti–PD-L1 mAbs in enhancing the antitumor response has been highlighted in several preclinical studies (
2,
3), the relevance of these observations to the activity of human PD-L1 IgGs in clinical use has not been fully addressed. Here, we demonstrate that the IgG1 subclass of avelumab does not optimally engage beneficial human FcγR pathways in the TME. We further demonstrate that increasing the ratio of activating to inhibitory FcγR pathway engagement by PD-L1 mAbs can potentiate the antitumor immunity mediated by this IgG subclass. We characterized two such practical approaches to potentiate FcγR pathways by anti–PD-L1 treatments: (i) combined treatment with FcγRIIB-blocking mAb and (ii) IgG Fc scaffold modification, in the form of afucosylated IgG1, to increase FcγRIIIA binding.
The similar antitumor efficacy seen in the Fc-silent N297A mutated compared with the human IgG1 subclass of the same PD-L1 mAb clones is in contrast to our previous observation of enhanced antitumor activity by mouse IgG2a and rat IgG2b subclasses compared with their Fc-silent variant or with mice with genetic deficiency of activating FcγRs (
2). The mouse IgG2a and human IgG1 isotypes are widely considered as homologous subclasses because of their superior ability to engage Fc effector function, such as ADCC and ADCP, relative to the other IgG subclasses in each species. However, these subclasses exhibit significant differences in their FcγR-binding properties. The mouse IgG2a and rat IgG2b isotypes exhibit significant preferential binding capacity to the mouse activating FcγRs (A/I of 69 and 40, respectively) (
2,
31), whereas the binding ratio of human IgG1 is lower (FcγRIIIA/FcγRIIB binding ratio of 5) (
10). This A/I ratio dictates the cytotoxic activity of many types of mAb and likely contributes to these observed cross-species differences due to a dominant effect of the inhibitory FcγRIIB in the TME on human IgG compared with mouse IgG2a. Moreover, the expression and distribution of the mouse versus human FcγRs on the effector monocytes and macrophages in the TME are significantly different (
12,
32). These cross-species differences in IgG-FcγR interactions and FcγR distribution and function highlight the advantage of studies using a fully humanized FcγR-IgG in vivo system to better mimic the expected activity of IgGs in humans (
33,
34).
The importance of FcγRs in enhancing the antitumor activity of additional mAb targeting immune checkpoints has been highlighted by several previous studies (
2,
13,
35–
39). Engagement of both activating and/or inhibitory FcγRs pathways was shown to contribute to the potency of these immunotherapies. Fc engagement of activating FcγRs was shown to contribute through induction of Ab-mediated elimination of intratumoral T
reg by anti-CTLA4 (
35,
37,
38,
40,
41), anti–OX-40 (
39), anti–4-1BB (
42), and anti-GITR (
38) mAbs. Conversely, engagement of the inhibitory FcγRIIB provides an inert scaffold for Ab cross-linking to enhance agonistic mAb activities, mainly those targeting tumor necrosis factor receptors, such as anti-CD40 mAbs (
36,
43,
44). Thus, many checkpoint mAbs can work through multiple mechanisms in vivo (e.g., checkpoint blockade and T
reg depletion), which can synergize their activities for optimal therapeutic efficacy. Similarly, our study implies that Fab-mediated PD-1/L1–blocking and Fc-mediated modulation of the myeloid subsets are two modes of action engaged by anti–PD-L1 mAbs.
The importance of the Fc region in several FcγR-mediated pathways that compromise antitumor activity, as in the case of anti–PD-1 mAbs, is highlighted by our findings and those of others (
2,
3). Hence, the selection of the appropriate Ab scaffold (IgG isoform or Fc variant that selectively engages or avoids specific FcγR pathways) for each mAb is extremely important to achieve optimal antitumor activity. Moreover, these identified FcγR pathways provide opportunities to Fc-engineer checkpoint mAbs to improve their activity. Examples in clinical testing include an afucosylated variant of ipilimumab, an anti–CTLA-4 mAb modified for enhanced T
reg depletion (
45), and an Fc-modified variant of selicrelumab (2141-V11), a CD40 agonist enhanced for FcγRIIB-mediated cross-linking (
13). Our study supports the rationale of clinical development and evaluation of an afucosylated variant of avelumab and other PD-L1 mAbs as second-generation checkpoint inhibitory mAbs with enhanced Fc effector function for enhanced control of tumor growth.
FcγRIIB reduces the efficacy of mAb-mediated immunotherapy (
46), and combining an FcγRIIB antagonist with anticancer mAbs was demonstrated to increase the FcγR-dependent antitumor effect of anticancer cytotoxic mAb immunotherapy (
47). Our results demonstrating FcγRIIB expression by macrophages in human Merkel cell and renal cell carcinoma lesions are consistent with previous reports on human melanoma biopsies (
37). The dominant role of macrophages in Fc-mediated activity of therapeutic mAbs and their high abundance and FcγRIIB up-regulation in mouse and human tumors implies a possible role of macrophages in the Fc-mediated activity of avelumab and highlights these macrophage-associated pathways as potential biomarkers for a treatment response (
48). This FcγRIIB expression pattern, and the enhanced antitumor activity of avelumab in combination with intratumoral injection of FcγRIIB blockers, indicates that this FcγR has an inhibitory effect on avelumab activity. Therapeutic FcγRIIB-specific antagonistic mAbs have been developed, and our study supports clinical evaluation of these available reagents for coadministration with avelumab or other anti–PD-L1 IgG1 mAbs (
49).
Survival benefit for patients with advanced urothelial carcinoma treated with avelumab is associated with the presence of two or more of the FCGR2A
H131 and FCGR3A
V158 FcγR alleles with high affinity to IgG1 binding (
50). Moreover, FcγR-expressing cells in the TME, including NK cells, monocytes, and macrophages, were associated with enhanced survival in the avelumab-treated group in this trial, further supporting the potential of beneficial FcγR engagement by avelumab in a subset of patients with these Fc effector–permissive conditions. The huFcγR mice used in our study carry the low-affinity FCGR2A
R131 and FCGR3A
F158 FcγR alleles. We show that avelumab as a monotherapy does not engage beneficial FcγR pathways in tumor models established in these mice unless the Ab is afucosylated. Together, our preclinical findings and the cited clinical data consistently imply that the binding intensity of anti–PD-L1 IgG1 with low-affinity alleles of FcγRs is insufficient to provide antitumor Fc effector function by this mAb. An increase in the activating FcγR-IgG1 binding interactions, either with FcγRs carrying higher-affinity alleles for IgG1 binding, by Fc-engineering of the IgG scaffold, or by blocking the competing binding to the inhibitory FcγRIIB, is needed to provide improved antitumor response. It is also predicted that treatment with afucosylated avelumab will benefit the patient subset with no or a low number of high-affinity FcγR alleles, compensating for the reduced treatment efficacy by the parental IgG1 mAb in these patients’ subset.
The improved activity of aFuc-IgG1 avelumab is characterized by alteration of the myeloid subsets in the TME. The most notable effects include the reduction of the PD-L1
+ subset of DCs and MDSCs and modulation of neutrophil frequencies and cell states. Tumor-associated inflammatory PD-L1
+ DCs can contribute to immune suppression and play a critical role in the antitumor response (
22,
51,
52). Hence, we suggest that afucosylated avelumab promotes a less suppressive TME by Fc-mediated reduction of these PD-L1–suppressive cell types through engagement with FcγRIIIA-expressing effector cells, such as macrophages and NK cells. This treatment leads to an increase in proinflammatory lymphocyte infiltration to the tumors, likely mediating the killing of tumor cells.
In contrast to the reduced frequencies of suppressive DCs and MDSCs, neutrophil frequencies increased after treatment with aFuc-IgG1 avelumab. This suggests that neutrophils are not targeted for elimination by the Fc-engineered PD-L1 mAb. Moreover, our data imply that they are not affected directly by their FcγRIIIB engagement and activation of intracellular FcγR signaling. Instead, we propose that neutrophils are affected by bystander effects of the proinflammatory response mediated in the TME by the aFuc-IgG1 avelumab treatment. Neutrophils have been reported to contribute to either protumor or antitumor activities. Antitumor functions of neutrophils include preventing metastasis formation (
24), mediating the inhibition of early tumor growth (
25), stimulating T cell responses (
26,
27), and antigen processing and presenting of tumor peptides to T cells (
28,
29). Our data imply that neutrophils could be indirectly affected by aFuc-IgG1 avelumab Fc-FcγR interactions with other immune cells. These interactions lead to a decrease in PD-L1
+ suppressive cells and a subsequent increase in the proinflammatory state of the TME, which modulates the neutrophil population.
A limitation of this study is the use of mouse preclinical models to predict the clinical response of fully human Abs. To minimize this caveat, we used humanized FcγR mice to better recapitulate human IgG-FcγR interactions and validated the up-regulation of FcγRIIB we observed in mice tumors in relevant clinical tumor biopsies. However, additional significant cross-species differences in immunity to the tumors exist and may compromise the clinical translation potential of our preclinical finding.
Together, we have addressed a yet unresolved question in the field regarding the relevance of previously identified FcγR pathways in mice to the activity of human PD-L1 Abs used in the clinic. We studied the contribution of human FcγRs and Fc-mediated effector function to the activity of human FDA-approved PD-L1 mAbs. Our study concludes that Fc engineering of PD-L1 mAbs to increase affinity to activating FcγRs is a valid strategy to optimize tumor immunotherapy. This is opposite to the Fc engineering strategy previously applied in the design of this class of therapeutics (atezolizumab and durvalumab) to eliminate their interaction with FcγRs. Future studies should address whether additional types of therapeutic IgG1 can optimally engage Fc effector functions in the solid TME and whether they can be further optimized by Fc engineering to potentiate their engagement with activating FcγRs.
MATERIALS AND METHODS
Study design
The objective of this study was to elucidate how human PD-L1 mAb therapy recruits FcγR
+ immune cells in the TME to enhance their antitumor activity. Our goal was to generate and characterize next-generation Fc variants of current FDA-approved anti–PD-L1 drugs with improved efficacy that can be readily advanced to clinical evaluation. We designed and performed experiments using flow cytometry, immunofluorescence staining, scRNA-seq, and in vivo tumor challenge. In this study, we used FcγR humanized mice that recapitulate the human Fc receptor activity and expression pattern and allowed us to overcome cross-species reactivity between human Abs and murine receptors. We also used human tumor sections for clinical relevance. The sample size and number of biological replicates are indicated in each of the figure legends. In several cases, experiments were pooled. Any experiment with
n < 5 in treatment groups is a representative experiment of several repeats with similar results obtained. No data were excluded from the study except in the following: (i) In tumor challenge experiments, samples were excluded in the following cases: (a) outlier mice in the tumor model experiments that had significantly higher tumor volume compared with all other mice in the same experiment (
Fig. 2C and fig. S3E), (b) micethat unexpectedly died at an early time point (
Fig. 3D), and (c) mice that did not have a tumor at the time of sacrifice (fig. S6, G and I). (ii) In flow cytometry experiments, samples were excluded from analysis when (a) the population was too small for reliable gating (fewer than 300 cells), (b) when cell quantification was not possible because of technical issues (discrepancy in bead count), and (c) in rare cases of outlier samples that differed significantly from the group’s trend (
Fig. 4B and figs. S4D and 5S, A to C). (iii) In enzyme-linked immunosorbent assay (ELISA) experiments, two concentrations were excluded because of technical problems (fig. S1A). In all in vivo experiments shown in the study, animals were randomized and assigned to experimental groups on the basis of tumor size, sex, and age, as denoted in Materials and Methods. Tumor measurements were performed by a researcher blinded to each animal’s treatment group. In all in vivo experiments shown in the study, data collection was performed either until the humane end point approved by the relevant Institutional Animal Care and Use Committee (IACUC) was reached or until mice were tumor free for up to 100 days after tumor inoculation.
Mice
All animal studies were approved by the IACUC of Weizmann Institute of Science under license numbers 05620720-2 and 03910422-3. FcγR KO (mFcγRanull), huFcγR mice (mFcγRAnull, hFcγRI+, hFcγRIIAR131+, hFcγRIIB+, hFcgRIIIAF158+, and hFcγRIIIB+), and CD16A+/+CD16B+/+ mice (mFcγRanull, hFcγRInull, hFcγRIIAnull, hFcγRIIBnull, hFcγRIIIAF158+, and hFcγRIIIB+) were provided by J. Ravetch from the Rockefeller University and maintained in-house. All in vivo experiments were performed in the Weizmann Institute of Science specific pathogen–free facility, and animals of both sexes were used between the ages of 8 and 12 weeks, randomly assigned to experimental groups.
Cell lines
MC38 and B16-F10 GVAX were provided by J. Ravetch from the Rockefeller University. B16-F10 were purchased from the American Type Culture Collection [Research Resource Identifier (RRID): CVCL_0159].
Tumor challenge and treatment
Tumor cell lines were maintained in a humidified incubator at 37°C and 5% CO2 and cultured in complete RPMI 1640 medium containing 25 mM Hepes, 1% l-glutamine, 10% fetal bovine serum, 1% penicillin-streptomycin, 1% nonessential amino acids, and 1% pyruvate. MC38 (2 × 106) or B16F10 (4 × 105) cells were implanted subcutaneously on the left flanks of the mice, and tumor volumes were measured every 2 to 3 days with an electronic caliper by an investigator blinded to treatment group. Volume was calculated using the formula (L2 * l)/2, where L is the largest diameter and l is the smallest diameter. Mice were randomized by tumor size (day 0) and were treated as described in the legend of each experiment. huFcγR mice were treated with 50 μg of anti–PD-L1 mAb IgG1, IgG1-N297A, afucosylated IgG1 (avelumab or atezolizumab), 25 μg of anti–human FcγRIIB, or control phosphate-buffered saline (PBS) on days 0, 3, and 6 unless indicated otherwise. GVAX (1 × 106) were irradiated (150 Gy) and implanted subcutaneously on the right flanks of B16-F10 tumor–bearing mice on days 0 and 6. Mice were monitored after treatment initiation and sacrificed when tumor size reached the Weizmann Institute of Science IACUC limit.
Neutrophil depletion in vivo
Anti-mouse Ly6G/Ly6C (clone RB6-8C5) (Bio X Cell, catalog no. BE0075, RRID: AB_10312146) and the corresponding rat IgG2b IC (clone LTF-2) (Bio X Cell, catalog no. BE0090, RRID: AB_1107780) were injected intravenously through the lateral tail vein, as indicated in the figure and/or legend. Depletion efficiency was assessed by flow cytometry analysis.
Human tumor tissues
Meckel cell carcinoma and renal cell carcinoma paraffin tumor sections were provided by A. Alva from the University of Michigan. Sections were 4 μm thick. Personal patient information was not reported.
Immunofluorescence staining
Immunofluorescence staining of paraffin-embedded sections was performed after deparaffinization with xylene, EtOH, and PBS. All slides were blocked with CAS-Block (Thermo Fisher Scientific, ZY-008120) for 60 min in a humidified chamber at room temperature (RT). All Abs were diluted in CAS-Block. Slides were incubated overnight at 4°C with primary Abs against CD68 (Abcam, catalog no. ab213363, RRID: AB_2801637), CD66b (Abcam, catalog no. ab197678), anti-LY75/DEC-205 (Abcam, catalog no. ab124897, RRID: AB_10976058), and anti-CD32B (Abcam, catalog no. ab77093, RRID: AB_1523300) at 1:200 dilution. The slides were incubated with biotin-conjugated anti-goat IgG (Jackson ImmunoResearch Labs, catalog no. 705-065-147, RRID: AB_2340397) at 1:100 dilution for 90 min in humidified chambers at RT. Slides were then incubated with Cy3-conjugated donkey anti-rabbit secondary Ab (Jackson ImmunoResearch Labs, catalog no. 711-165-152, RRID: AB_2307443) at 1:200 dilution or streptavidin Cy5 (Jackson ImmunoResearch Labs, catalog no. 016-170-084, RRID: AB_2337245) at 1:150 dilution for 60 min at RT. DAPI (4′,6-diamidino-2-phenylindole) staining was performed at a dilution of 1:1000 for 60 min at RT. All slides were mounted with Thermo Fisher Scientific Shandon Immu-Mount. Slides were imaged by using a Zeiss LSM 880 Airyscan confocal microscope with a 20×/0.8 M27 Plan-Apochromat objective.
Generation of anti–human PD-L1 and anti–human FcγRIIB Abs and their Fc variants
The variable heavy and light regions of avelumab or atezolizumab were synthesized on the basis of their published sequences. The variable heavy region of anti–human FcγRIIB (clone 2B6) was provided by J. Ravetch (Rockefeller University). The variable region sequences of the parental Abs were cloned and inserted into mammalian expression vectors with human IgG1 or human kappa (atezolizumab and anti–human FcγRIIB)/lambda (avelumab) Fc backbones. For the generation of the human IgG1 N297A Fc domain variant, site-directed mutagenesis using specific primers (avelumab: 5′-GAGAAGAGCAGTACGCCTCAACTTACCGCGTAG-3′, 5′-CTACGCGGTAAGTTGAGGCGTACTGCTCTTCTC-3′; atezolizumab: 5′-GAGAAGAACAATACGCCTCAACCTATCGCGTTG-3′, 3′-CAACGCGATAGGTTGAGGCGTATTGTTCTTCTC-5′) was performed by polymerase chain reaction (PCR) (Agilent Technologies) according to the manufacturer’s instructions. Mutated plasmid sequences were validated by direct sequencing (Life Science Core Facility, Weizmann Institute of Science). To produce Abs, Ab heavy and light chain expression vectors were transiently transfected into Expi293 cells (Thermo Fisher Scientific, catalog no. A14527).
For the generation of the human IgG1 afucosylated Fc domain variant, 200 μM 2-deoxy-2-fluoro-l-fucose (Biosynth Carbosynth) was added 1 day after transfection. The secreted Abs in the supernatant were purified by Protein G Sepharose 4 Fast Flow (GE Healthcare). Purified Abs were dialyzed in PBS and sterile-filtered (0.22 μm). Purity was assessed by SDS–polyacrylamide gel electrophoresis and Coomassie staining. Percentage of afucosylated forms was assessed by mass spectrometry and was estimated to be 75%.
Enzyme-linked immunosorbent assay
Binding specificity and affinity of anti–PD-L1 mAbs were determined by ELISA using recombinant mouse PD-L1 (Sino Biological, no. 50010-M08H or BPS Bioscience, no. 71117-1), human PD-L1 (Sino Biological, no. 10084-H08H or BPS Bioscience no. 71104-1), and human FcγRs (human FCGR1A, no. 10256-H08H; human FCGR2A, no. 10374-H08H; human FCGR2B, no. 10259-H08H; human FCGR3A, no. 10389-H08H; and human FCGR3B, no.11046-H08H1, Sino Biological). The 96-well ELISA Half Area High Binding plates (Greiner Bio-One, no. 675061) were coated overnight at 4°C with replicates of recombinant PD-L1 proteins (1 μg/ml) or human FcγRs (2 μg/ml). All subsequent steps were performed at RT. After washing, the plates were blocked for 1 hour with 1× PBS containing 2% bovine serum albumin (BSA; for PD-L1 proteins) or 10% BSA (for human FcγRs) and were then incubated for 1 hour (for PD-L1 proteins) or 2 hours (for human FcγRs) with serially diluted IgGs (dilutions are indicated in the figures and were prepared in the relevant blocking solution). After washing, plates were incubated for 1 hour with horseradish peroxidase–conjugated anti-human IgG (Jackson ImmunoResearch Labs, catalog no. 109-035-088, RRID: AB_2337584). For ELISA inhibition assay, after blocking nonspecific sites, plates were incubated for 1 hour with serially diluted IgGs with 1 μg/ml of mouse (BPS Bioscience, no. 71118) or human PD-1-biotin (BPS Bioscience, no. 71109-1) in 1× PBS with 2% BSA. After washing, plates were incubated for 1 hour with horseradish peroxidase–labeled streptavidin (no. 405210, BioLegend). Detection was performed using a one-component substrate solution (trimethylboron), and reactions were stopped with the addition of 0.18 M sulfuric acid. Absorbance at 450 nm was immediately recorded using a SpectraMax Plus spectrophotometer (Molecular Devices), and background absorbance from negative control samples was subtracted. The following Abs were tested: avelumab (prepared in-house), atezolizumab (prepared in-house), and anti-mouse PD-L1 (clone 10F.9G2) (Bio X Cell, catalog no. BE0101, RRID: AB_10949073).
Pharmacokinetics assay
MC38 tumor–bearing huFcγR mice were injected with anti–PD-L1 IgG1 (avelumab), IgG1-N297A, or afucosylated IgG1 (300 μg per mouse) and bled at indicated time points. Serum was stored at −80°C until all time points were collected. Ab concentration in serum was determined using a standard colorimetric ELISA assay. Briefly, assay plates were coated with recombinant human PD-L1 (1μg/ml; Sino Biological, no. 10084-H08H) and incubated overnight at 4°C. Plates were then blocked for 2 hours with 1× PBS with 10% fetal calf serum. Serial dilutions of serum were added to the plates, and plates were incubated for 2 hours. After washing, plates were incubated for 1 hour with horseradish peroxidase–conjugated anti-human IgG (Jackson ImmunoResearch Labs, catalog no. 109-035-088, RRID: AB_2337584). Absorbance at 450 nm was immediately recorded using a SpectraMax Plus spectrophotometer (Molecular Devices), and background absorbance from negative control samples was subtracted. Ab concentration in the serum was calculated for each time point based on a calibration curve of known concentrations and the dilution factor.
Tissue processing
Peripheral blood was collected into K2E EDTA tubes (Becton Dickinson), and cells were then stained at RT for 30 min. Erythrocytes were then lysed by 10-min incubation with 1 ml of BD FACS Lysing Solution (Becton Dickinson) followed by washes with fluorescence-activated cell sorting (FACS) buffer (1× PBS with 0.5% BSA and 2 mM EDTA) at 1200 rpm before analysis. Spleens and LNs were dissected through a 70-μm nylon cell strainer. For spleen, erythrocytes were lysed by 5-min incubation with red blood cell lysis buffer (Zymo Research) and washed with PBS. Tumors were mechanically dissected into small fragments and transferred to GentleMACS C tubes (Miltenyi Biotec) with deoxyribonuclease I (0.33 mg/ml; Roche) and Liberase TL (0.27 mg/ml; Roche) in Dulbecco’s modified Eagle’s medium (DMEM) (Biological Industries). Next, tumors were dissociated twice in the GentleMACS Octo Dissociator (Miltenyi Biotec), and the cell suspension was then further incubated at 37°C, 25 rpm, for 40 min. After incubation, tumors were subjected to two additional dissociation cycles in the GentleMACS. After the second dissociation cycle, tumors were dispersed through a 70-μm nylon cell strainer and washed with PBS.
Flow cytometry
Single-cell suspensions were prepared as described above. For surface staining, cells were plated in U-shaped 96-well plates (Thermo Fisher Scientific) in PBS. Cells were first stained with LIVE/DEAD Fixable blue dead cell stain (Thermo Fisher Scientific), followed by one wash with PBS, and then resuspended in 25 μl of FACS buffer with human TruStain Fc block (BioLegend) and incubated in the dark for 15 min at RT. Surface antigens were stained in FACS buffer for 30 min in the dark on ice. Then, the cells were washed twice with FACS buffer, resuspended in 150 μl of FACS buffer, and analyzed by flow cytometry. For intracellular Foxp3 and Arginase 1 staining, an additional staining step was performed using the True-Nuclear Transcription Factor Buffer Set Kit (BioLegend) according to the manufacturer’s instructions. All samples were analyzed by CytoFLEX LX (Beckman Coulter). For cell quantification, we used CountBright absolute counting beads for flow cytometry (Thermo Fisher Scientific, catalog no. C36950), and calculation of absolute number was done according to manufacturer instructions. Unless otherwise specified, cell populations were defined by the following markers: DCs: CD45+, NK1.1−, CD11b+, CD11c+, MHC-II+, F4/80−; macrophages: CD45+, NK1.1−, CD11b+, MHC-II+, F4/80+, Ly6C−, Ly6G−; monocytes: CD45+, NK1.1−, CD11b+, Ly6C+, Ly6G−, F4/80−, CD11c−; neutrophils: CD45+, NK1.1−, CD11b+, Ly6G+, Ly6Clow, F4/80−; NK cells: CD45+, NK1.1+; CD8 T cells: CD45+, CD3+, CD8+, CD4−; CD4 T cells: CD45+, CD3+, CD4+, CD8−, Foxp3−; Tregs: CD45+, CD3+, CD4+, CD8−, Foxp3+. To determine Arginase 1, PD-L1, and CD32B expression, the following Abs were used: arginase 1 (A1exF5), PD-L1 (10F.9G2), and CD32B (2B6). The following commercial Abs were used for determining cell populations by flow cytometry: anti-mouse CD45 (clone 30-F11), BioLegend, catalog no. 103151, RRID: AB_2565884; anti-mouse CD11c (clone N418), BioLegend, catalog no. 117320, RRID: AB_528736; anti-mouse MHC-II (clone M5/114.15.2), BioLegend, catalog no. 107636, RRID: AB_2734168; anti-mouse F4/80 (clone BM8), BioLegend, catalog no. 123118, RRID: AB_893477; anti-mouse CD11b (clone M1/70), BioLegend, catalog no. 101256, RRID: AB_2563648; anti-mouse Ly6C (clone HK1.4), BioLegend, catalog no. 128012, RRID: AB_1659241; anti-mouse Ly6G (clone 1A8), BioLegend, catalog no. 127645, RRID: AB_2566317; anti-mouse NK1.1 (clone PK136), BioLegend, catalog no. 108741, RRID: AB_2562561; anti-mouse CD3 (clone 17A2), BioLegend, catalog no. 100220, RRID: AB_1732057; anti-mouse CD8α (clone 53-6.7), BioLegend, catalog no. 100730, RRID: AB_493703; anti-mouse CD4 (clone RM4-5), BioLegend, catalog no. 100526, RRID: AB_312727; anti-mouse Foxp3 (clone 53-6.7), BioLegend, catalog no. 126404, RRID: AB_1089117; anti-mouse Arginase 1 (clone A1exF5), Thermo Fisher Scientific, catalog no. 12-3697-82, RRID: AB_2734839; and anti-mouse PD-L1 (clone 10F.9G2), BioLegend, catalog no. 124308, RRID: AB_2073556.
Single-cell sorting
After staining, cells were washed and resuspended in cold FACS buffer (0.5% BSA and 2 mM EDTA in PBS), stained with fluorophore-conjugated anti-mouse CD45 Ab, and filtered through a 70-μm strainer. Before sorting, cells were stained with propidium iodide to exclude dead/dying cells. Cell sorting was performed using a BD FACSAria Fusion flow cytometer (BD Biosciences), gating for CD45+ cells after exclusion of dead cells and doublets. Single cells were sorted into 384-well capture plates containing 100 μl of lysis solution, 3 μl of mineral oil, and 20 nM barcoded poly(T) reverse transcription primers for scRNA-seq as described previously (PubMed identifier: 34017133). Immediately after sorting, plates were spun down to ensure cell immersion into the lysis solution, snap-frozen on dry ice, and stored at −80°C until further processing. Cells were analyzed using BD FACSDIVA software (BD Bioscience) and FlowJo software (FlowJo LLC).
Single-cell library preparation
scRNA-seq libraries from sorted cells were prepared with a modified version of the massively parallel scRNA-seq method (MARS-seq) (PMID:31101904). In brief, polyadenylated mRNA from single cells sorted into 384-well capture plates was barcoded during reverse transcription into cDNA. Each plate was pooled, and cDNA was fragmented and amplified to generate Illumina sequencing-ready libraries. Each plate library was tested for quality and DNA concentration.
Read alignment
Sequencing libraries were pooled at equimolar concentrations and sequenced using an Illumina NextSeq 500 or NovaSeq 6000 sequencer at a sequencing depth of 10,000 to 50,000 reads per cell. Reads were condensed into original molecules by counting same unique molecular identifiers (UMIs). We used statistics on empty-well spurious UMI detection to ensure that the batches we used for analysis showed a low level of cross single-cell contamination (less than 3%). Reads were processed as previously described for MARS-seq. Reads were mapped to murine reference genome mm10 using HISAT (version 0.1.6); reads with multiple mapping positions were excluded. Reads were associated with genes if they were mapped to an exon using the UCSC genome browser for reference. Exons of different genes that shared genomic position on the same strand were considered a single gene with a concatenated gene symbol.
Quality control
Cells were filtered on the basis of count depth (400 < #UMI), number of expressed genes (200 < #genes), and mitochondrial gene content (20 > %Mt-). Cells not meeting the latter threshold were usually nonviable or apoptotic.
Clustering and annotation
Seurat package (version 4.0) (
53) standard workflow was used to cluster. Briefly, the UMI matrix was log-normalized, highly variable genes were detected, and principal components analysis (PCA) was applied, followed by K-nearest neighbors graph construction on 50 leading components. The Louvain algorithm was used for community detection using resolution parameter 1. Clusters were manually annotated on the basis of marker genes.
Dimensionality reduction
We used Uniform Manifold Approximation and Projection (UMAP) with Seurat (
53) functions RunUMAP(). We applied it to the batch-corrected data. The PCs used to calculate the embedding were the same as those used for clustering. When calculating the UMAP coordinates, we used
a = 0.5,
b = 2 for optimal spread.
Gene set enrichment analysis
We applied GSEA (
54) developed in the Broad Institute to find gene sets enriched in different treatment arms. We provided gene log
2FC generated by FindMarkers() from Seurat and sorted. We used the Fast GSEA (“fgsea”) package implemented in R (
55). Gene sets were drawn from the mouse C5 v5p2 gene ontology (GO) collection.
Differential expression testing
To find differentially expressed genes (DEGs) between treatments, we pooled together cells from all samples and used FindMarkers() from Seurat (
53) with two-sided Wilcoxon rank-sum test on uncorrected log-transformed normalized counts. We considered genes that were expressed in >0.05 of cells and with >5 cells with >2 UMIs.
Marker genes (fig. S6A) were selected to be fold change > 1.5, Benjamini-Hochberg adjusted P value < 0.05, of which the top scoring genes were presented. DEGs in fig. S6B were selected to be |fold change > 1.25|, Benjamini-Hochberg adjusted P value < 0.05, of which the top scoring genes were labeled.
Software and visualization
All statistical analyses were conducted using R software (R Foundation for Statistical Computing, Vienna, Austria). ggplot2 code package was used for most graphics.
Statistical analysis
Flow cytometry data analysis was performed using FlowJo software. All other data analyses were performed using GraphPad Prism 9 (GraphPad Software). Quantitative data are presented as means ± SEM unless otherwise indicated. When the value is the florescence intensity of the Ab deduced from an IC, delta geometric mean intensity is presented (ΔMFI). For each dataset, normality of the population and/or population residuals (Gaussian distribution) was confirmed using Shapiro-Wilk and/or D’Agostino-Pearson tests. For normal distributions, when two groups were compared, an unpaired two-tailed Student’s t test (two-tailed, unequal variance) was used to determine statistical significance; when three groups were compared, one-way analysis of variance (ANOVA) with Tukey’s post hoc test was used to determine statistical significance. When data were not normally distributed, a nonparametric test was used, Mann-Whitney test was used when two groups were compared, or Kruskal-Wallis with Dunn’s post hoc test was used for multiple comparisons. For survival curves, log-rank test with Bonferroni corrected threshold was performed. Statistical significance is represented in figures as follows: *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P < 0.0001, ns: not significant.
Acknowledgments
We thank I. Sher for artwork and S. Schwarzbaum for editorial assistance.
Funding: This work was supported by Rina Gudinski Career Development Chair (R.D.), the Israel Cancer Association (R.D.), Israel Science Foundation (R.D.), Moross Integrated Cancer (R.D.), Flight Attendant Medical Research Institute (R.D.), Dwek Institute for Cancer Therapy Research (R.D.), David E. Stone and Sheri Hirschfield Stone 75th Anniversary Fund (R.D.), Rising Tide Foundation (R.D.), Mexican Association of Friends of the Weizmann Institute (R.D.), Miel de Botton (R.D.), Garvan-Weizmann Partnership donors (R.D.), Elie Hirschfeld and Dr. Sarah Schlesinger (R.D.), an NFBI-Teva Ph.D. fellowship (N.C.S.), Maccabim Foundation Scholarship (N.C.S.), and Cancer-Immunology MICC Ph.D. program (N.C.S.).
Author contributions: R.D. supervised this study and obtained its funding. R.D. and N.C.S. conceived and designed the study. R.D. and N.C.S. wrote the manuscript. N.C.S., A.Y., T.L., R.S., and T.F. performed experiments and/or analyzed data. A.A. provided resources for this study. I.A. supervised work and obtained funding.
Competing interests: The Weizmann Institute has filed a PCT patent application related to this work, on which N.C.S. and R.D. are inventors. R.D. receives research funding from Teva Pharmaceuticals and holds consultancy roles for Teva Pharmaceuticals, Nucleai, and Immunai. The other authors declare that they have no competing interests.
Data and materials availability: Single-cell RNA sequencing data were deposited to NCBI GEO, accession GSE224011. All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. All datasets generated during and/or analyzed during the current study are available from the corresponding authors upon reasonable request.