Background

Programmed death ligand 1 (PDL1, CD274), an immune cosignaling molecule in the B7-homology family, interacts with its receptor PD1 on T cells to attenuate antitumor T cell responses, allowing tumors to escape immune surveillance.1 2 Blocking PDL1 signals with antibodies is the basis for highly effective immune checkpoint blockade (ICB) therapy that mediates anticancer efficacy, but only in a minority subset of patients.2–5 Tumor-extrinsic effects of PDL1 have been intensively studied. However, it is now clear that PDL1 also mediates important tumor-intrinsic signaling that promotes pathological outcomes including treatment resistance, reviewed in reference.6 Genetic tumor PDL1 depletion reverses pathological tumor-intrinsic signals to improve treatment efficacy of small molecules and cytotoxic agents7–10 and deletion of the PDL1 cytoplasmic tail in tumor cells can improve ICB efficacy.11 12 To overcome difficulties in translating genetic approaches to tumor PDL1 depletion, we screened libraries of agents that were enriched for FDA-approved drugs for PDL1 depletion efficacy that identified chlorambucil, a DNA alkylating nitrogen mustard drug approved to treat chronic lymphocytic leukemia and other neoplasms,13 14 as an agent that reduces cancer cell PDL1 and thus could be used clinically as a pharmacological means to reduce deleterious tumor-intrinsic PDL1 signals. We also identified 9-[2-(phosphonomethoxy)ethyl] guanine (PMEG) as a tumor PDL1 depletion agent, but did not study it in detail as it is not FDA approved.

We show that chlorambucil-mediated tumor PDL1 depletion was through distinct mechanisms in different cancer cell lines. In vivo, chlorambucil controlled tumor growth in mouse melanoma and ovarian cancer models in a tumor PDL1-dependent manner, independent of host PDL1 expression. It also paradoxically rendered otherwise αPDL1-refractory tumors to be αPDL1-responsiveness in vivo through a natural killer (NK) cell-dependent mechanism and improved antitumor immunity in a NK cell-dependent manner, which is a likely efficacy mechanism. Chlorambucil-mediated tumor PDL1 depletion phenocopied important genetic PDL1 depletion effects in vitro, including suppressing tumor mTORC1 signals and stemness, suggesting additional potential uses and mechanisms for treatment efficacy meriting additional studies. Our study suggests a novel strategy to reprogram the tumor for ICB treatment response by pharmacological tumor PDL1 signal reduction with potential additional clinical applications, requiring additional work to achieve this objective. As chlorambucil is FDA approved, orally available and generally tolerable at doses required for these effects, approaches shown can be rapidly tested clinically.

Materials and methodsGeneration of murine PDL1-red fluorescent protein construct and tumor cell expression

The pLVX-DsRed-Express2-N1 lentiviral vector was purchased from Clontech (Mountain View, California, USA, cat. 632560). pLVX-DsRed-mPDL1-red fluorescent protein (RFP) expression vector conjugating RFP to the PDL1 C-terminus was generated by PCR cloning using a murine PDL1 expression vector as a backbone (ORIGENE, cat. MG203953), using the primers PDL1-AA259F-5’-gcagatagttccctcgaggccaccatgttcttgagaaaacaagtgaga-3’; PDL1-AA290R-5’ agggttcaacagaattcgcgtctcctcgaattgtgtatcatttcg-3’. Amplified PCR product was cloned into pLVX-DsRed-Express2-N1 vector using Xho-1 and EcoR1 sites. PDL1-C terminus plasmid sequence and orientation was confirmed by sequencing. Lentivirus was produced by transfecting HEK293T cells with pLVX-DsRed-mPDL1-Cterminus-RFP plasmid with packaging vectors pMD-2G, pMDLg/pRRE and pRSV-Rev plasmids (cat. 12259, 12253, 12251, Addgene) using Turbofect Transfection Reagent (Thermo Fisher, Waltham, Massachusetts, USA) according to manufacturer protocols. The lentivirus-containing supernatants were collected and concentrated using Lenti-X (Takara). Transduction was by adding the lentivirus to medium followed by puromycin (1 µg/mL) selection for 14 days. Pooled clones were selected and expression of PDL1-C terminus-RFP was confirmed by flow cytometry for RFP expression and Western blotting.

High-throughput drug screen to identify PDL1 depleting drugs

We selected a subline suitable from PDL1 RFP reporter cells described above for drug screens by flow cytometric sorting for RFPhi cells. PDL1 RFP reporter cells were seeded at 800 cells/well into 384 well black µ-clear film plates, Griener bio-one plates, (Kremsmünster, cat. 781091) using the EL406 Washer/Dispenser (Biotekin 50 µL medium. Plates were incubated overnight at 37°C, 5% CO2. and then treated with 2.5 µM and 10 µM compounds from the Prestwick and 10 µM from the LOPAC library (1200 compounds in each library that are largely in human trials or FDA approved) using the Agilent Bravo liquid pipetting platform and incubated for 48 hours. Cells were then washed with phosphate-buffered saline (PBS), and fixed with 4% paraformaldehyde (Electron Microscopy Sciences, cat.15714S) and labeled with 2 µg/mL 4′,6-diamidino-2-phenylindole (DAPI), (Molecular Probes, cat. D1306) and 1 µg/mL Cell Mask-Blue (Invitrogen, cat. H32720). Plates were then imaged using the Operetta High Content Imaging System (Perkin Elmer) in the DAPI and Cell Mask Blue (excitation 360–400 µM, emission 410–480 µM), and RFP channel (excitation 520–550 µM, emission 560–630 µM). Images were analyzed using Perkin Elmer Harmony Software. Briefly, the number of cells in vehicle (DMSO) treated or compound-treated wells were binned into 6 RFP expression level segments based on per cell mean fluorescent intensities of RFP expression. Compounds that induced >2.6 fold PDL1 reduction versus control DMSO with >80% viability were identified as potential hits. The Cell Mask Blue stain allowed for analysis of cell shape and size and the DAPI stain allowed for analysis of nuclear shape, nuclear size and nuclear fluorescence intensity to exclude cells if these parameters were altered.

Cell lines and in vitro treatments

Human OVCAR5 ovarian cancer cells were obtained from Dr. David Curiel, Washington University. Mouse B16-F10 melanoma (herein ‘B16’) and human ES2 ovarian cancer cells were purchased from the American Type Culture Collection. ID8agg mouse ovarian cancer, ID8agg-Luc, PDL1lo ID8agg and PDL1KO B16 cells were generated as we described.7 15 All cell lines were negative for Mycoplasma in periodic testing using MycoAlert Mycoplasma Detection Kits (Lonza, cat. LT07-318), according to manufacturer directions and used in passages <5. Mouse lines were maintained in 5% fetal bovine serum (FBS)-containing RPMI-1640 medium (Roswell Park Memorial Institute). Human lines were maintained in 10% FBS-containing Gibco Dulbecco’s Modified Eagle medium, all supplemented with 1% penicillin/streptomycin, 1% L-glutamate, and 1% 4-(2-hydroxyethyl)−1-piperazine ethane sulfonic acid. Cells were treated with chlorambucil (Sigma, cat. C0253), gemcitabine (Sigma, cat. G6423), rabusertib (Selleckchem, cat. S2626), MG132 (Sigma, cat. M7449) or extracellular growth factor (Thermo Fish, cat. PHG0313) at indicated concentrations.

Cell proliferation and viability assays

Cells (500/well) were plated in 96-well flat-bottom plastic culture plates in 100 µL medium and treated with chlorambucil at indicated concentrations for 48 hours. Cell viability was determined by a cell counting kit-8 (CCK-8 (MedChemExpress; cat. HY-K0301). Absorbance was measured at 450 nm using a BioTek Synergy 2 Multi-Mode Plate Reader. Proliferation was assessed in triplicate in three separate experiments for all in vitro data shown.

In vivo tumor challenge, treatments and assessments

Wild-type C57BL/6 mice, NSG (NOD-SCID IL2Rgammanull) mice were purchased from Jackson Labs and PDL1KO mice on the C57BL/6 background were a kind gift from Dr. Leiping Chen, then at Mayo Clinic. Mice were bred in our animal facility and used at 7–20 weeks old. All mice were given ad libitum water and food and housed under specific pathogen-free conditions and matched for sex and age in each experiment. 5×105 B16 or PDL1KO B16 cells in log growth phase in 0.15 mL sterile PBS were given subcutaneously (both sexes). 4×106 ID8agg-Luc or PDL1lo ID8agg cells in log growth phase were given intraperitoneally to females only. For NSG mice, 1.5×105 B16 or 1×106 ID8agg cells were used. Intradermal tumor growth was measured with Vernier calipers and volume was calculated as (length × width2)/2. Intraperitoneal ID8agg-Luc tumor growth was determined by IVIS Lumina Imaging System (Perkin Elmer) 15 min after intraperitoneal injection of 3 mg PBS-dissolved d-luciferin K+ (Gold Biotechnology) with 1 min exposure, medium binning, and F/stop=1. Identical regions of interest were drawn over each subject’s abdomen and average radiance (photons/sec/cm2/sr) was quantified with Living Image software version 4.2. Body weight was determined weekly for PDL1lo ID8agg tumor and percentage increase body weight was used as a surrogate for tumor as we reported.16 Tumors were collected and weighed at sacrifice in confirmation. Survival was determined as spontaneous death, moribundity, tumor volume >1500 mm3 for subcutaneous tumor, or weight >130% of baseline (ascites) in ID8agg or >120% for PDL1lo ID8agg tumors. One chlorambucil cycle is 5 consecutive daily intraperitoneal injections injections using 27 gage needles (BD, cat. 305136) in 150 µL PBS, followed by 2 days of rest (doses optimized for each tumor and treatment condition in preliminary assessments as indicated) versus respective vehicle control injections. Two treatment cycles were used for B16 or PDL1KO B16, and three treatment cycles for ID8agg or PDL1lo ID8agg starting 7 days after tumor cell challenge. αPDL1 (clone 10F.9G2, BioXcell) or respective isotype control was injected intraperitoneally at 100 µg/mouse every 4 days starting on day 7 after tumor challenge for times indicated. To rechallenge cured mice, WT mice that were PDL1lo ID8agg tumor-free >100 days after initial challenge were rechallenged with 4×106 ID8agg cells. Age matched naïve WT mice were challenged as controls. Tumor volume and mice survival were measured as above.

Flow cytometry

Mice were sacrificed by cervical dislocation after induction of deep isoflurane anesthesia. Tumors and ascites were collected and cells were stained as we described.15 Cells were stained with antibodies from Invitrogen: αCD45 (30-F11); Biolegend: αCD3 (17A2), αCD11C (N418), αNK1.1(PK136), αCXCR5 (L138D7), αGr-1 (RB6-8C5), αGranzyme B (QA16A02), αTim3 (B8.2C12), αCD62L (MEL-14), αCD31 (390), αCD44 (IM7), αCD24 (M1/69), αCD133 (315–2 C11); BD Bioscience: αCD8 (53–6.7), αCD4 (GK1.5), αB220 (RA3-6B2), αIFNγ (XMG1.2), αPDL1 (MIH5), αCCR2 (475301), αCD25 (PC61), αTCF-1 (S33-966); Tonbo: αCD11b (MI/70), αFOXP3 (3G3) and Thermo Fisher: αPD1 (J43) plus matched isotype controls. We used Alexa Fluor antibody labeling kit (cat. A20186, Thermo Fisher) to add fluorescence to αFAP (ab28244, Thermo Fisher). Apoptosis assessment used Annexin V and propidium iodide (Thermo Fisher Scientific, cat. V13245) according to manufacturer instruction. Data were acquired on LSR II or Cytek Aurora hardware and analyzed by FACS Diva (BD Bioscience) or FlowJo software (Flow Jo).

To study immunogenic cell death, cells were treated with chlorambucil with indicated doses and time, harvested, washed and cell suspensions were adjusted to 2×106 cells/mL in 200 µL ice-cold PBS, 10% FCS, 1% sodium azide. We added 2 µL anti-Calreticulin antibody (ab209577) or isotype control rabbit IgG antibody (ab209478) per sample, incubated cells for 30 min in the dark at 4°C, washed cells three times by centrifugation at 400 g for 5 min and resuspend them in 300 µL in ice-cold PBS, 10% FBS, 1% sodium azide. We fixed samples with 1% PFA for immediate analysis.

Annexin V/propidium iodide staining

We harvested cells after incubation, washed in cold PBS, recentrifuged cells, discarded supernatant and resuspended in 1X Annexin-binding buffer at~1 x 106 cells/mL. The 5 µL Annexin V and 1 µL 100 µg/mL /propidium iodide working solution were added, incubated at room temperature for 15 min, and then 400 µL 1X Annexin-binding buffer was added and the mix kept on ice for immediate flow cytometry.

Quantification of reactive oxygen species

Reactive oxygen species (ROS) generation and quantification was with 2',7'-dichlorodihydrofluorescein diacetate (DCFDA)/H2DCFDA (dichlorofluorescein diacetate)-Cellular ROS Assay Kit (Abcam, ab113851) and 50 µM tert-butyl hydrogen peroxide was used as positive control per manufacturer’s instructions. In brief, PD-L1KO ID8agg cells were treated with vehicle control or chlorambucil (2.5 µM) for 48 hours, and then live cells were harvested and incubated with 20 µM H2DCFDA for 30 min at 37°C. Samples were analyzed by flow cytometry immediately with excitation and emission spectra of 485 and 535 nm, respectively.

Tumor initiating cell assessment

B16, ID8agg and ES2 cells were cultured 48 hours after reaching around 80% confluency, trypsinized using trypsin-EDTA (0.05%, GIBCO, Thermo Fisher, Waltham, Massachusetts, USA) and counted on a Vi-Cell (Beckman Coulter, Brea, California, USA). 2×106 B16 or ID8agg cells were stained with live dead viability dye (BioLegend), following by staining for tumor initiating cell (TIC) markers as previously described.17 18 Briefly, B16 TICs were CD44+CD133+CD24+ and ID8agg TIC were CD44+CD24+. In vivo TIC gating was done on tumor cells on the basis on their side and forward scatter, followed by gating for live, CD45-CD44+CD24+ for ID8agg tumors. ALDEFLUOR kits were purchased from STEMCELL Technologies and used with 2×106 ES2 cells/ml. A 5 µL of the specific aldehyde dehydrogenase (ALDH) inhibitor diethylaminobenzaldehyde was added to control cells, incubated with 5 µL 4,4-difluoro-1,3,5,7,8-pentamethyl-4-bora-3a,4a-diaza-s-indacene (BODIPY) amino acetaldehyde, a fluorescent substrate for ALDH and incubated for 45 min at 37°C, and kept in cold (4°C) PBS for remaining staining. N, N-diethylaminobenzaldehyde inhibits ALDH activity and was used as a positive control. ES2 TICs were defined as ALDHhi by flow cytometry.

Western blotting and immunoprecipitation

To prepare whole cell lysates, cells were harvested and lysed in radioimmunoprecipitation assay buffer plus 1 mM phenylmethylsulphonyl fluoride and Halt protease/phosphatase inhibitor cocktail (Thermo Scientific) at 4°C. Protein concentrations were measured by BCA Protein Assay Kit (Thermo Scientific). Samples were run on precast 4%–15% sodium dodecyl sulfate polyacrylamide gels (Bio-Rad), transferred to polyvinylidene fluoride membranes (GE Water and Process Technologies) and blocked in 5% nonfat milk for 1 hour at ambient temperature. Membranes were then incubated with 1:1000 diluted primary antibodies against indicated proteins: αP70S6K (CST, cat. 34475), αp-P70S6K(Thr389) (CST, cat. 97596), α4EBP1(CST, cat.9644), αp-4EBP1(Ser65) (CST, cat. 9451), αRaptor (CST, cat. 48648), αLC3A/B (CST, cat. 4108), αPDL1 (CST, cat. 13684), αPDL1(Abcam, cat. 213480), αβ-TRCP (CST, cat. 4394), α GSK-3β (CST, cat. 12456), αPGSK-3β (CST, cat. 5558), αCBL-B (SC, cat. 376409), αVinculin (CST, cat. 13901). Cells for (co)immunoprecipitation experiments were seeded in 15 cm2 dishes and cultured as described, using 6 million cells/immunoprecipitation reaction. For protein ubiquitination, cells were lysed in denaturing buffer (50 mM Tris-HCl, 0.5 mM EDTA and 1% SDS) followed by heating at 95°C for 10 min and then quenched with 9 volumes quenching buffer (0.5% Triton X-100, 20 mM Tris-HCl (pH 8.0), 137 mM NaCl, 10% glycerol, 2 mM EDTA). Protease inhibitor cocktail (Roche) was added fresh to all buffers. Cell lysates were incubated on a rotator for 30 min at 4°C, and then centrifuged at 20 000 × gravity for 15 min at 4°C. Supernatants were subsequently incubated with αPDL1 antibody overnight and processed using Dyna beads Protein G for immunoprecipitation (Thermo Fisher Scientific). The final eluate was processed, and Western blotting was performed with specific antibody binding detected by chemiluminescence using Western Lightening Plus regent (Perkin Elmer).

Quantitative real-time reverse transcription PCR

Total RNA was isolated from homogenized whole lung tissue with Trizol reagent (Invitrogen, cat. 15596026) according to manufacturer instructions. Subsequently, cDNA was synthesized with 1–2 µg of total RNA using the SuperScript III reverse transcriptase (Invitrogen, cat. 18080044). Quantitative PCR was conducted using the Bio-Rad Real-Time PCR System (Applied Biosystems), amplified with transcript-specific primers with SYBR Green (Thermo Scientific, cat. 4472908), according to manufacturer instructions. Mouse primers were: Tyr, 5′-CTCTGGGCTTAGCAGTAGGC-3′ and 5′- GCAAGCTGTGGTAGTCGTCT-3′; gp100, 5′-ACATTTCATCACCAGCAGGGTGCC-3′ and 5′-AACAAGTGGGTGCTGGCC-3′; Trp-2, 5′-GTCCTCCACTCTTTTACAGACG-3′ and 5′-ATTCGGTTGTGACCAATGGGT; Trp-1, 5′ CCCCTAGCCTATATCTCCCTTTT-3′ and 5′-TACCATCGTGGGGATAATGGC-3′; PDL1, 5′- GGTGGTGCGGACTACAAG-3′ and 5′- AACCCTCGGCCTGACATA-3′; and GPDH, 5′- AACGACCCCTTCATTGAC-3′ and 5′-TCCACGACATACTCAGCAC-3′ as the internal control. Human primers PDL1, 5’-GCTTTTCAATGTGACCAGCA-3’ and 5’- ATTTGGAGGATGTGCCAGAG−3’; GAPDH, 5’-CGAGATCCCTCCAAAATCAA-3’ and 5’-TTCACACCCATGACGAACAT-3’. Data were normalized to the endogenous reference gene GAPDH levels by the comparative CT (ΔΔCt) method.

Statistics

We used Prism software from GraphPad for statistical analyses. In bar graphs, data are shown as mean±SEM. We used two-way analysis of variance (ANOVA) corrected with Tukey post-tests to compare repeated means in tumor growth analyses. We used the log-rank test to analyze mouse survival from survival plots generated with the method of Kaplan-Meier. Three or more means were compared with one-way ANOVA with Tukey post-test. For comparison of two means, we used an unpaired t-test or Mann-Whitney U test. All analyses were considered p<0.05 as statistically significant.

ResultsChlorambucil reduces tumor cell PDL1 in vitro and vivo

We performed high-throughput drug screens of the LOPAC and Prestwick Libraries (online supplemental figure S1A,B) to identify novel PDL1 depleting drugs for rapid clinical translation. As the screen used mouse melanoma cells, but we also wished to define PDL1 depletion agents in human cancer cells, 15 candidate drugs from the screen were verified for PDL1 depletion in ovarian cancer cells by flow cytometry. Chlorambucil and PMEG were among top hits (figure 1A and online supplemental figure S1C). Chlorambucil reduced surface PDL1 expression by 5-fold, which was greater than PMEG-mediated PDL1 reduction. We elected to study chlorambucil in detail as it is also FDA approved and orally bioavailable, whereas PMEG is not FDA approved. To validate flow cytometry data, we used immunoblots to detect total PDL1 in various human and mouse tumor cell lines, confirming chlorambucil as a bona fide tumor PDL1 depletion drug in all cell lines tested (figure 1B).

Figure 1

Chlorambucil reduces tumor PDL1 in vitro and in vivo. (A) Flow cytometry for PDL1 percentage and mean fluorescence intensity (MFI) in VC (vehicle control) and chlorambucil treatment (10 µM, 48 hours) in OVCAR5 cells in vitro. (B) Western blots for PDL1 expression in human and mouse ovarian cancer cells treated in vitro with (+) or without (−) chlorambucil for 48 hours. Vinculin, loading control. Values are μM chlorambucil. (C) Flow cytometry analyses of WT mice challenged with ID8agg cells, treated with chlorambucil (CAMB, 1 mg/kg as described in methods) and sacrificed 4 weeks post tumor challenge. Summary of PDL1 percentage in the tumor CD45-CD31-FAP- (top) or non-immune stromal cell CD45-CD31+FAP+ (bottom) gates. PDL1 percentage in specified populations as indicated (right). (D) Summary graph of PDL1 MFI in CD45-CD31-FAP- tumor cells. (E) Summary of PDL1 percentage in CD45+ immune cells. P value determined by unpaired t-test. VC, vehicle control.

We further confirmed tumor PDL1 depletion by chlorambucil in vivo in a mouse ovarian cancer model. Chlorambucil significantly reduced the prevalence and mean fluorescence intensity of PDL1+ cells among CD45-CD31-FAP- cells which are highly enriched for tumor cells, ersus control-treated mice (figure 1C, top panel and figure 1D). By contrast, PDL1 expression and prevalence in tumor fibroblasts and endothelial cells (CD45-CD31+FAP+) and the CD45+ immune cells were unchanged versus control-treated mice (figure 1C, bottom panel, figure 1E). Together, these data are consistent with tumor cell-selective PDL1 reduction by chlorambucil in vivo. Chlorambucil also reduced PDL1 expression in B16 tumor cells in vivo, but to a lesser degree versus in ID8agg (online supplemental figure S1D), possibly due to not excluding fibroblasts and endothelial cells in these analyses.

Chlorambucil inhibits tumor cell growth in vitro in a PDL1-dependent manner

Chlorambucil inhibited control B16 and ID8agg cell growth in vitro with no, or minimal killing or apoptosis induction (figure 2A), consistent with our previous findings of tumor intrinsic PDL1 promoting cell growth.7 To determine if chlorambucil growth inhibitory inhibition was tumor cell PDL1-dependent, we tested genetic PDL1KO ES2, ID8agg and B16, all of which were less sensitive to chlorambucil-induced growth inhibition versus parental cells figure 2B, consistent with a tumor cell PDL1-dependent growth inhibition mechanism.

Figure 2

Chlorambucil slows tumor cell growth in a tumor PDL1-dependent fashion. (A) Annexin V-FITC/propidium iodide (PI) staining by flow cytometry. In given plots: Q1, necrotic cells, Q2, late apoptotic cells, Q3, live cells and Q4, early apoptotic cells. ID8agg (upper panel) and B16 cells (lower panel) were treated with 2.5 µM chlorambucil for 48 hours. Numbers in each quadrant indicate percent cells. Summary and stats for total (early+late) apoptotic cells shown on the right. (B) Tumor cells were treated with indicated concentrations of chlorambucil for 48 hours. Cell viability was detected by CCK-8 assay and relative cell viability ratios were compared between control cells and genetically PDL1-depleted cells. Data represents three independent experiments. P value determined by unpaired t-test. VC, vehicle control.

Chlorambucil is a DNA-damaging agent and DNA damage usually increases cell PDL1 expression.19 To test the possibility that chlorambucil-mediated PDL1 reduction is not through its DNA-damaging effect, we first confirmed that chlorambucil at concentrations and in cells used here induced DNA damage (online supplemental figure S2A). We then tested other DNA-damage agents including X-rays, gemcitabine and the small molecule Chk1 inhibitor, rabusertib but none of these DNA damaging agents reduced tumor PDL1, but rather all increased it as expected (online supplemental figure S2B), suggesting that chlorambucil over-rides a normal biological response to DNA damage and that its PDL1 reduction effect is independent of its FDA-approved mechanism of action.

Chlorambucil inhibits tumor growth in vivo in a PDL1-dependent manner

We challenged WT mice orthotopically with poorly immunogenic and highly aggressive ID8agg cells, as we described15 and treated with 2 mg/kg chlorambucil, which significantly reduced tumor growth and prolonged mouse survival versus PBS vehicle control (figure 3A,B). A 2 mg/kg chlorambucil also significantly improved survival in ID8agg challenged immunodeficient NSG mice versus vehicle control (figure 3C) demonstrating an immune-independent tumor growth control mechanism. By striking contrast, this same chlorambucil dose had no effect on the growth of orthotopic PDL1lo ID8agg tumors in WT mice as detected by ascites weight gain and end-experiment tumor weights (figure 3D, online supplemental figure S3A) suggesting that the chlorambucil efficacy mechanism in this model is tumor PDL1-dependent and immune-independent. To test chlorambucil in a distinct tumor model and different anatomic compartment, B16 melanoma cells were challenged subcutaneously into WT or NSG mice. Similar to peritoneal ID8agg ovarian cancer, 3 mg/kg chlorambucil significantly inhibited tumor growth and prolonged mice survival in WT mice and NSG mice, but less effectively in NSG mice (figure 3E,F). In contrast to PDL1lo ID8agg, this chlorambucil dose also inhibited PDL1KO B16 tumor growth in WT mice (figure 3G), although significantly less effectively than in control B16 (figure 3H). In NSG mice, chlorambucil efficacy against PDL1KO B16 tumor was abolished (figure 3I), again demonstrating a PDL1-dependent treatment mechanism, but in this case it was also immune-dependent in striking contrast to ID8agg. Thus, chlorambucil treatment efficacy is tumor PDL1-dependent, but at selected chlorambucil doses or in specific tumors, treatment efficacy could be dependent or independent of immune contributions.

Figure 3

Chlorambucil treats tumors in vivo in a tumor PDL1-dependent fashion. (A) ID8agg-luc tumor bioluminescence was measured in WT mice treated with chlorambucil (2 mg/kg) as described. N=5/group. P value, two-way ANOVA. (B, C) Kaplan-Maier survival curve in WT (B) or NSG mice (C) treated with chlorambucil 2 mg/kg as described. P value from log-rank test. N=5/group in WT mice and 6/group in NSG mice. (D) WT mice challenged with PDL1lo ID8agg cells and treated as described in (A). Mice body weight measured weekly and increase is relative to baseline. N=5 for PBS and six for chlorambucil, p value, two-way ANOVA. (E) Tumor growth of WT mice challenged with B16 cells and treated with 3 mg/kg chlorambucil, with the last treatment on day 18, N=5 for all groups. P value, two-way ANOVA. Kaplan-Maier survival curve in these WT mice (right panel). (F) Kaplan-Maier survival plot of NSG mice (N=5/group) challenged with B16 cells and treated with 3 mg/kg chlorambucil. (G) Tumor growth in WT mice challenged with PDL1KO B16 cells and treated with 3 mg/kg chlorambucil. N=5/group. (H) Tumor growth fold-change calculated as control treatment divided by chlorambucil treatment calculated separately for control and PDL1KO tumors on the given days shown. P value, χ2 test. (I) Kaplan-Maier survival curve of NSG mice challenged with PDL1KO B16 cells and treated with chlorambucil as in (G). N=5/group. (J) Tumor bioluminescence of PDL1KO mice challenged with ID8agg-luc cells and treated as in (A). N=6/group. P value, two-way ANOVA. (K) Kaplan-Maier survival curve of PDL1KO mice challenged with B16 cells and treated as described in (E). N=4 for both PBS and CAMB. ANOVA, analysis of variance; CAMB, chlorambucil; PBS, phosphate-buffered saline.

To test whether chlorambucil efficacy is host PDL1 dependent, we challenged ID8agg cells or B16 cells into PDL1KO mice. In both cases, chlorambucil efficacy was unaffected (figure 3J,K; online supplemental figure S3B). Thus, chlorambucil-mediated tumor growth inhibition depends on tumor rather than host PDL1, also consistent with its tumor cell-selective PDL1 depletion that is neither cancer nor anatomic compartment-specific in these models.

Chlorambucil renders αPDL1-refractory tumors αPDL1-responsive

We reported that αPDL1 is ineffective against control intraperitoneal ID8agg15 and subcutaneous PDL1KO B16 tumor7 challenged in WT mice. As sufficient chlorambucil can inhibit PDL1KO B16 tumor growth with evidence for immune contributions, we tested if a lower chlorambucil dose that was ineffective as a single agent would improve αPDL1 efficacy. Cutting the chlorambucil by 50% to 1.5 mg/kg rendered it ineffective against PDL1KO B16, but its combination with αPDL1 (also ineffective as a single agent) significantly inhibited PDL1KO B16 tumor growth and improved survival in WT mice (figure 4A). Further, this combination inhibited spontaneous lung metastases following subcutaneous challenge compared with single-agent treatment as detected by qRT-PCR for melanoma specific gene products (figure 4B). Cutting the chlorambucil by 50% to 1 mg/kg failed to inhibit control ID8agg tumor growth but greatly sensitized it to αPDL1 growth inhibition and significantly prolonged WT mice survival (figure 4C,D). Although chlorambucil alone did not treat PDL1lo ID8agg (figure 3D) adding αPDL1 significantly improved survival of WT mice challenged with PDL1lo ID8agg versus chlorambucil alone, including apparent tumor cure in 3 of 6 mice, vs 0 of 6 cures in control treated WT (right panel, figure 4E). The foregoing all together suggested that chlorambucil could elicit tumor PDL1-independent immunogenicity as a mechanism to improve αPDL1 outcomes especially in PDL1 depleted tumors in which αPDL1 alone is ineffective.

Figure 4

Chlorambucil improves αPDL1 treatment efficacy in vivo. (A) Tumor growth (left) and survival (right) of WT mice (N=5/group) challenged with PDL1KO B16 and treated with 1.5 mg/kg chlorambucil±anti-PDL1 100 µg/mouse (combination is chlorambucil plus anti-PDL1). (B) qRT-PCR for indicated melanoma specific genes in whole lung tissue harvested on day 20 after tumor challenge with PDL1KO B16 cells into WT mice and treated as in panel A. P value, unpaired t-test. (C, D) ID8agg-luc tumor bioluminescence (C) and survival (D) in WT mice (N=5–6/group) after treatment with 1 mg/kg chlorambucil as described, ±a 100 µg/mouse nti-PDL1on days 7, 11, 15 and 19. P value, two-way ANOVA. (E) Body weight gain (as surrogate measurement for tumor burden) (left) and survival (right) of WT mice (N=6/group) challenged with PDL1lo ID8agg and treated with 2 mg/kg chlorambucil±anti-PDL1 as in panel D. P value, two-way ANOVA. (F) Flow cytometry analysis of calreticulin content in vehicle control (VC) or chlorambucil (CAMB)-treated B16 or PDL1KO B16 (1.25 µM), ID8agg or PDL1KO ID8agg (1.25 µM), for 48 hours. Isotype: isotype control for anti-calreticulin antibody. Figures below panels, % positive cells. Data represent one of three independent experiments with similar results. ANOVA, analysis of variance.

Immunogenic cell death that can stimulate anticancer immunity and treatment efficacy is elicited by certain cytotoxic agents,20–22 although chlorambucil is not reported to be a major immunogenic cell death inducer. However, we found that cell surface calreticulin, a measure of immunogenic cell death, was PDL1-dependently increased by chlorambucil in ID8agg but not B16 cells, although maximal immunogenic cell death was equivalent in control or genetically PDL1 depleted cells in both cell lines (figure 4F, online supplemental figure S4A). To assess if lower chlorambucil could define an ICD difference in B16 vs PD-L1KO B16 we tested chlorambucil concentrations down to 0.625 µM but without any PDL1-dependent effect (online supplemental figure S4B). Reactive oxygen species can evoke immunogenic cell death,23 but were barely detectable in control and PDL1KO cells although they were robustly elicited by hydrogen peroxide positive control (online supplemental figure S5A). Thus, chlorambucil-induced immunogenic cell death is a candidate mechanism for improving αPDL1 efficacy in some tumor models, but not others, which requires further investigations.

Figure 5

αPDL1 improves CAMB-mediated immunity. (A) Summary statistics for flow cytometry data for CD8+CD44+CD62L+ effector memory (left) and CD8+CXCR5+TCF1+CCR2+ T cell stem cell abundance from mice ascites which challenged with ID8agg and treated with chlorambucil±anti-PDL1 or isotype as a control. Mice are sacrificed after tumor challenged 30 days, P value, one-way ANOVA, chlorambucil vs combination and anti-PDL1 vs combination using Mann-Whitney test. (B) Flow cytometric analysis of NK cell frequency and Granzyme B production in CD45+CD3-NK1.1+ NK cells in ascites after challenge with ID8agg cells and treatment with chlorambucil±anti-PDL1 treatment as described. P value, one-way ANOVA. Chlorambucil versus combination and anti-PDL1 versus combination P value by Mann-Whitney U test. (C) Mice previously cured of PDL1lo ID8agg challenge by combination treatment from figure 4E (N=3) were rechallenged with ID8agg cells 105 days after final treatment. ‘Naïve mice’ (N=5) were controls not previously challenged or treated. Kaplan-Maier plot of survival shown. (D, E) Flow cytometric analysis of PDL1KO tumor from WT mice challenged orthotopically with PDL1KO B16 cells and treated as described. (D) PD1+ (exhausted) CD45+CD3-NK1.1+ NK cells. E. PD1+Tim3+ (exhausted) CD8+ T cells. P value, one-way ANOVA, chlorambucil versus combination and anti-PDL1 vs combination by Mann-Whitney test. (F) Kaplan-Maier survival plot of WT mice challenged with ID8agg cells (N=6/group) treated with chlorambucil+anti-PDL1 combination as described in figure 4C ±Anti-NK1.1 antibody or isotype control (250 µg/mouse) on days 6, 10, 14, 18, 22, 26. (G) PDL1KO B16 cells challenged into WT mice and treated with chlorambucil as described ± α-NK1.1 to deplete NK cells. N=10 tumors/group. P value, unpaired t-test. ANOVA, analysis of variance; CAMB, chlorambucil; NK, natural killer.

αPDL1 improves immunity in combination with chlorambucil

To understand immune effects of chlorambucil contributing to αPDL1 efficacy, we evaluated the immune landscape in ascites and tumors from different treatment groups of ID8agg-challenged mice. T-distributed stochastic neighbor embedding analysis of flow cytometry data demonstrated pronounced differences in tumor immune landscapes among treatment groups (online supplemental figure S6A). Chlorambucil alone did not appreciably alter antitumor immune cells known to mediate antitumor immunity, consistent with its immune-independent mode of action in this tumor. Combination treatment with αPDL1 did not significantly increase CD8+ T cell numbers nor their production of IFN-γ+ or Granzyme B+ (online supplemental figure S6B). However, chlorambucil plus αPDL1 additively increased CD44+CD62L+ central memory CD8+ T cells and a population of CD8+ CXCR5+TCF1+ CCR2+ T cell stem cells that can contribute to ICB response24 in ID8agg (figure 5A).

Figure 6

CAMB reduces PDL1 through GSK3β/β-TrCP ubiquitination promotion. (A) RT-qPCR for CD274 mRNA in OVCAR5 and ID8agg cells treated with (+) or without (−) chlorambucil (10 µM for OVCAR5 and 2.5 µM for ID8agg) for 48 hours. (n=4). (B) Western blot for PDL1 protein in OVCAR5 cells with (+) or without (−) MG132 (2 µM) for 6 or 20 hours (upper panel). PDL1 expression in OVCAR5 cells treated with chlorambucil (2.5 µM) 48 hours