Chemotherapy Combines Effectively with Anti–PD-L1 Treatment and Can Augment Antitumor Responses

This article is featured in In This Issue, p.2191

Immunotherapy with checkpoint inhibitors has proved to be highly effective in certain indications with durable clinical responses for a subset of individuals. However, the majority of patients still show reduced or no clinical benefit (1–5). Clinical responses to checkpoint inhibitors like anti-PD1/PD-L1 are generally observed in cancers with increased tumor mutational burden, pre-existing immunity, and higher expression of PD-L1, although this is not always the case (6–8). In an effort to broaden the number of responding individuals, overcome resistance to single-agent therapy, and extend the duration of response, a combination of chemotherapy with checkpoint inhibitors is currently being tested in multiple clinical trials. Combining checkpoint blockade with standard of care chemotherapy could also bring the benefits of immunotherapy into earlier lines of treatment in which the cancer-immune set point might have a lower threshold for clinical response (9).

Some chemotherapies can augment antitumor responses through the induction of immunogenic cell death (10–12) or through the depletion of immunosuppressive cell subsets such as myeloid-derived suppressive cells or T regulatory cells (Tregs). Alternatively, other chemotherapies are hypothesized to be detrimental for the establishment of antitumor immunity by reducing the function and number of effector T cells (13–15). Additionally, chemotherapies may impact different stages of the immune response, including differential effects on T cell priming in draining lymph nodes (dLNs) or T cell effector functions in tumor. Therefore, combining chemotherapy with checkpoint inhibitors could have various effects, depending on the balance between the beneficial and antagonistic effects of chemotherapy on components of the immune system. The summation of these factors will eventually determine whether a particular combination therapy will show enhanced combinatorial activity or reduced responses. Therefore, a better understanding of the potential interactions between specific chemotherapies and immune cell subsets and how these interplay with checkpoint inhibitors could yield valuable insights into future combinatorial modalities.

Given that checkpoint inhibitors like anti–PD-L1 are being tested in earlier lines of treatment concurrent with chemotherapy, we wanted to explore the effects of such combinatorial approaches on the antitumor activity and immune modulation mediated by anti–PD-L1 treatment in a responsive syngeneic tumor model. This approach enabled us to assess the pharmacodynamics of different chemotherapies on various immune cell components in different tissues and how these influenced anti–PD-L1 responses. We found that most chemotherapeutic agents reduced T cell numbers and activity in peripheral tissues such as blood and dLNs but did not antagonize responses in the tumor tissue, which in many cases was augmented in combination with anti–PD-L1. This disconnect between the effects in peripheral tissues and antitumor activity was partly due to responses in established tumors relying mostly on infiltrating T cells at the time of treatment initiation. Additionally, synergy with anti–PD-L1 was observed through different mechanisms that ultimately led to establishing a new threshold for antitumor activity. Overall, our results suggest that chemotherapy can effectively combine with checkpoint inhibitors like anti–PD-L1 with various effects that may be chemotherapy specific and dependent on individual tumor microenvironmental components.

All animal studies were reviewed and approved by Genentech’s Institutional Animal Care and Use Committee. Mice whose tumors exceeded acceptable size limits (2000 mm³) or became ulcerated were euthanized and removed from the study. All individuals participating in animal care and use are required to undergo training by the institution’s veterinary staff. Any procedure, including handling, dosing, and sample collection, mandates training and validation of proficiency under the direction of the veterinary staff prior to performing procedures in experimental in vivo studies. All animals were dosed and monitored according to guidelines from the Institutional Animal Care and Use Committee on study protocols approved by Genentech’s Laboratory Animal Resource Committee.

Eight- to ten-week-old female C57BL/6 mice were obtained from Charles River Laboratories (Hollister, CA). Only animals that appeared to be healthy and that were free of obvious abnormalities were used for the studies. For pharmacodynamic studies (tissue collection followed by flow cytometry analysis), IFN-γ–YFP reporter mice (IFNG.IRES.eYFP.ki.B6N) were used. IFN-γ–YFP reporter mice are of BL/6J background and have an IRES–EYFP cassette knocked in to the endogenous IFN-γ locus 3′ to the stop codon. IFN-γ promoter activity thus results in expression of a bicistronic mRNA encoding both the endogenous IFN-γ gene and EYFP. For the RAG^−/− study, female B6.129S6-Rag2^tm1FwaN12 (RAGN12-F) mice from Taconic Biosciences (Albany, NY) were used.

For animal studies, murine IgG1 anti–PD-L1 clone 6E11 and murine IgG1 anti-gp120 isotype control Abs were used. Abs were stored in 20 mM histidine acetate, 240 mM sucrose, and 0.02% polysorbate 20 (pH 5.5) and diluted in PBS prior to use.

For flow cytometry analysis, the following fluorochrome-conjugated Abs were used: anti-mouse CD45 (clone 30-F11, 1:100), anti-mouse I-A/E (clone M5/114.15.2, 1:5000), anti-mouse CD11b (clone M1/70, 1:100), anti-mouse CD11c (clone N418, 1:50), anti-mouse Thy1.2 (clone 30-H12, 1:200), anti-mouse CD19 (clone 6D5, 1:500), anti-mouse Ly-6C (clone HK1.4, 1:800), and anti-mouse ICOS (clone C398.4A, 1:100) were purchased from BioLegend. Anti-mouse CD4 (clone RM4-5, 1:100), anti-mouse CD8 (clone 53-6.7, 1:100), anti-mouse CXCR3 (clone CXCR3-173, 1:200), anti-mouse SiglecF (clone E50-2440, 1:50), anti-mouse Ly-6G (clone 1A8, 1:200), and anti-mouse PD-L1 (clone 10F.9G2, 1:200) were all from BD Biosciences (San Jose, CA). Anti-mouse Ki-67 (clone SolA15, 1:100), anti-mouse CD62L (clone Mel14, 1:80), anti-mouse CD206 (clone MR6F3 1:200), anti-mouse F4/80 (clone BM8, 1:50), and anti-mouse CD44 (clones IM7, 1:200) were from eBioscience. Anti-human granzyme B (GZMB) (cross-reacts with mouse) (clone MHGB05, 1:100) was purchased from Life Technologies. LIVE/DEAD Fixable Dead Cell Stain from Life Technologies was used to gate on live cells.

Paclitaxel was purchased from LC Laboratories (Woburn, MA), dissolved in 50% Cremophor EL and 50% ethanol at 20 mg/ml to be stored at 4°C, and then further diluted in saline immediately before administration. Oxaliplatin was purchased from Winthrop (Bridgewater, NJ) and diluted in water immediately before administration. Gemcitabine from Hospira (Lake Forest, IL), carboplatin from TEVA Pharmaceuticals (Sellersville, PA), and cisplatin from Fresenius Kabi USA (Lake Zurich, IL) were diluted in saline immediately before administration. Docetaxel from ChemShuttle (Hayward, CA) and cyclophosphamide (CTX) from Baxter Healthcare (Deerfield, IL) were dissolved in saline immediately before administration. For dosing concentration, route, and regimen, please see Supplemental Table II.

MC38 murine colon adenocarcinoma cells were obtained from American Type Culture Collection (Manassas, VA). Cells were cultured in RPMI 1640 medium plus 1% l-glutamine with 10% FBS (HyClone, Waltham, MA). Cells in log-phase growth were harvested, washed once with HBSS, counted, and resuspended in 50% HBSS and 50% Matrigel (BD Biosciences) at 1 × 10⁶ cells/ml for injection into mice.

MC38 cells were harvested in log-phase growth and resuspended in HBSS containing Growth Factor Reduced Matrigel (BD Biosciences) at a 1:1 ratio. A total of 0.1 million (100 μl) MC38 cells were then implanted s.c. in the right unilateral thoracic area. Tumors were monitored for ∼14 d until they measured between 130 and 250 mm³, at which time mice were then randomized into treatment groups based upon their tumor’s volume. The next day, treatment was initiated with either isotype control Ab (gp120 mIgG1) or anti–PD-L1 mIgG1 (clone 6E11) at an initial loading dose of 10 mg/kg i.v., followed by 5 mg/kg i.p. thereafter twice a week for 10 d (pharmacodynamic studies) or 3 wk (efficacy studies). Chemotherapies were also administered the day after group assignment at the dose, routes, and schedules depicted in Table I. Tumor volumes and body weights were measured twice per week up to the end of the study. Tumor volumes were measured in two dimensions (length and width) using Ultra-Cal IV calipers (Fred V. Fowler, Newton, MA), and volume was calculated using the following formula: tumor size (mm³) = (length × width²) × 0.5. Mice body weights were measured using an Adventura Pro AV812 scale (Ohaus, Pine Brook, NJ). Abs were diluted in histidine buffer [20 mM histidine acetate, 240 mM sucrose, and 0.02% polysorbate 20 (pH 5.5)]. For FTY720 studies, stock solution was made by dissolving FTY720 (Cayman Chemical, Ann Arbor, MI) in ethanol at 50 mg/ml and stored at −80°C. Immediately before administration, the stock solution was diluted in PBS and orally administered by gavage at 1 mg/kg every day for 21 d.

For CD8 depletion studies, once tumors reached an average size of ∼190 mm³, mice were randomized into treatment groups, and CD8 depletion was started by dosing an anti-CD8–depleting Ab (clone ATCC-2.43) or control rat IgG2b at 10 mg/kg i.p. on days 0, 3, 8, and 15. Isotype (gp120 mIgG1) or anti–PD-L1 mIgG1 (clone 6E11) treatment was initiated on day 1, and the Abs were dosed at 10 mg/kg i.v., followed by 5 mg/kg i.p. twice a week for 3 wk.

For studies with anti–CSF-1R, female C57BL/6N mice (6 wk of age; Charles River Laboratories, Sulzfeld, Germany) were inoculated s.c. into the right flank with MC38 (10⁶ cells). Tumor growth was monitored by perpendicular caliper measurement, and tumor volume was calculated by using the following formula: V = (length × width²) / 2. Group allocation and treatment started on day 7, and the average tumor size was 120 mm³. Animals received 30 mg/kg of either mIgG1 (clone MOPC-21; Bio X Cell) or anti-CSF-1R [clone 2G2 (16)] i.p.; treatment with these Abs was continued weekly for 4 wk maximum. On day 9, anti-PD-L1 (6E11; Genentech) was given, starting with a loading dose of 10 mg/kg i.v. and continued with 5 mg/kg i.p. every 3 to 4 d (total of seven administrations) in the presence of mIgG1 or anti-CSF-1R Ab. Monotherapies of mIgG1 and anti-CSF-1R groups received additional matching volumes of saline on the days of PD-L1 Ab administration. Mice were graphically censored when tumor volume reached ≥700 mm³ (Kaplan–Meier plots), n = 10 mice per group. All procedures were performed upon approval by the Regierung Oberbayern, Weilheim (approval number: 55.2-1-54-2531.2-32-10).

To generate single-cell suspensions, tumors were collected, minced into 2- to 4-mm pieces, and digested for 30 min using the mouse Tumor Dissociation Kit from Miltenyi (Miltenyi Biotec, Auburn, CA). Tumor homogenates were then filtered through a 70-μm nylon filter (Corning) and washed twice with RPMI 1640 medium. After the last wash, cells were resuspended in staining buffer (PBS + 0.5% FCS + 5 mM EDTA) and 1–2 million cells were transferred to 96-well V-bottom plates. Cells were then surface stained for 30 min at 4°C and washed twice. For intracellular staining, cells were then fixed and permeabilized using the eBioscience Foxp3 Fix/Perm buffer (Thermo Fisher Scientific, Waltham, MA).

All data were presented as means ± SD. Comparisons between treatment groups were generated using nonparametric, Mann–Whitney U tests. Prism 6.0 (GraphPad) was used to process all the statistical analyses.

To start delineating how different chemotherapies affect the antitumor activity mediated by anti–PD-L1, we made use of the MC38 tumor model that partially responds to treatment (Fig. 1A, 1B). We first characterized the pharmacodynamic changes mediated by anti–PD-L1 treatment in tumor, dLNs, and blood. Anti–PD-L1 treatment increased the number of CD8⁺ T cells in both tumor and dLNs 9 d after treatment initiation, consistent with the pharmacodynamic effects reported in patients treated with MPDL3280A (2) (Fig. 1C). We also observed a significant increase in the activation level of CD8⁺ T cells in dLNs as measured by expression of the activation marker inducible T cell costimulator (ICOS) (Fig. 1D). To investigate the induction of IFN-γ expression, an effector cytokine linked to the PD1/PD-L1 axis (2, 17, 18), we used IFN-γ–YFP reporter mice to circumvent the need to restimulate T cells ex vivo (see 2Materials and Methods). Consistent with anti–PD-L1 data from the clinic, the frequency of IFN-γ⁺ CD8⁺ T cells in the tumor increased after treatment (Fig. 1E). Additionally, we observed an increase in the frequency of activated intratumoral CD8⁺ T cells expressing ICOS, the proliferation marker (Ki-67), and the cytolytic marker GZMB (Fig. 1E). Furthermore, anti–PD-L1 did not appear to affect the overall frequency of Tregs in the tumor tissue, but there was an overall increase in the CD8/Treg ratio given the increase in CD8⁺ T cells (Fig. 1F). We did not observe any difference in the number of intratumoral neutrophils, monocytes, or macrophages following treatment (data not shown). These results demonstrate that anti–PD-L1 treatment significantly modulates CD8⁺ T cell responses in peripheral and tumor tissue by increasing the number of activated effector CD8⁺ T cells, leading to tumor growth inhibition and complete responses (CRs) in a subset of mice.

To further corroborate that CD8⁺ T cells are required for the antitumor activity mediated by anti–PD-L1 treatment, we treated mice harboring established MC38 tumors with anti–PD-L1 in the presence of a CD8-depleting Ab. Twenty-four hours after anti-CD8 treatment, CD8⁺ T cells were significantly depleted in blood and present at almost undetectable levels (Fig. 1G). Importantly, depletion of CD8⁺ T cells completely abolished the activity of anti–PD-L1, demonstrating that CD8⁺ T cells are critical in driving the response to treatment (Fig. 1H).

Given that carboplatin is used for the treatment of a wide range of cancers, we wanted to further investigate the effects of this chemotherapy in combination with anti–PD-L1. To determine whether carboplatin suppressed T cell responses, we focused on CD8⁺ T cell numbers and phenotype as these cells drive the antitumor activity mediated by anti–PD-L1 in MC38 tumors. The dose for carboplatin was selected to approximate clinical exposure (area under the curve μM × h) after running individual pharmacokinetic studies (data not shown). The particular dosing schedule was determined to approximate clinical exposure, and the mouse regimen was selected based on clinical use (once weekly in humans was equivalent to once every 4 d in mice, and once every 3 wk in humans was equivalent to once weekly in mice) (Table I). Carboplatin treatment was initiated 1 d after tumors became established (Fig. 2A). We observed that carboplatin treatment significantly decreased the number of CD8⁺ and CD4⁺ T cells in dLNs, although these trends were appreciable but NS in tumors (Fig. 2B). There was also a significant decrease in the frequency of activated ICOS⁺ CD8⁺ T cells in dLNs but no change in the phenotype of these cells in tumor tissue following carboplatin treatment (Fig. 2C). However, there was a trend toward increased frequency of Tregs, which altered the CD8/Treg ratio (Fig. 2D), whereas no change was observed in the myeloid compartment (data not shown). Together, these results suggest that the effects of carboplatin on immune cells in the tumor were significantly muted when compared with dLNs and did not significantly affect intratumoral CD8⁺ T cell numbers or phenotype.

Given that carboplatin treatment had an apparent detrimental effect on dLNs, we wanted to further investigate whether these effects would translate into diminished anti–PD-L1–mediated responses in the combination setting. Interestingly, carboplatin not only showed single-agent activity when compared with the control group (tumor growth inhibition of 71% relative to the control group and 2× time to progression of 9 d versus 3.5 d, respectively) but also showed enhanced combinatorial activity with anti–PD-L1, resulting in 30% CRs (Fig. 2E). Analysis of the immune infiltrate revealed that combination treatment did not significantly affect the phenotype of CD8⁺ T cells in tumors (Fig. 2F). Surprisingly, combination treatment still led to a significant reduction in the frequency of activated CD8⁺ T cells in dLNs despite the strong combinatorial activity observed (Fig. 2G). These results led us to speculate that immune changes in peripheral tissues do not significantly affect the antitumor responses in established tumors and response is likely dependent on direct changes to the tumor-infiltrating lymphocytes (TILs).

To further investigate the disconnect between the detrimental effects of chemotherapy in peripheral tissues (dLNs and blood) and increased antitumor responses, we treated established MC38 tumor-bearing mice with carboplatin alone or in combination with anti–PD-L1 in the presence or absence of FTY720 (fingolimod), an immune modulator that inhibits T cell egress from lymphoid organs (19). Treatment with FTY720 significantly decreased the number of circulating T cells in the blood of treated mice but had only a partial effect on the activity of either agent alone or in combination (Fig. 3A, 3B). These results suggest that, at least in mice that harbor already established MC38 tumors, the antitumor activity is majorly dependent on TILs at the time of treatment initiation.

Having observed that carboplatin had no antagonistic effect on the antitumor activity mediated by anti–PD-L1 and actually enhanced responses, we wanted to determine whether this was a carboplatin-specific effect or related to platinum-based chemotherapy. To evaluate this, we performed single-agent and combination efficacy studies with two additional platinum analogs, cisplatin and oxaliplatin. In contrast to carboplatin, neither cisplatin nor oxaliplatin decreased the number of CD8⁺ or CD4⁺ T cells, but both agents decreased the frequency of ICOS⁺ CD4⁺ or CD8⁺ T cells in dLNs, respectively (Fig. 4A). Similar to carboplatin, cisplatin and oxaliplatin did not show an antagonistic effect on the phenotype of intratumoral CD8⁺ T cells with oxaliplatin, increasing the frequency of ICOS⁺ CD8⁺ T cells in tumor when given as single agent (data not shown). Both cisplatin and oxaliplatin displayed lower antagonistic effects on T cells at peripheral sites when compared with carboplatin (Supplemental Table I).

When these agents were combined with anti–PD-L1 treatment, oxaliplatin significantly decreased the frequency of ICOS⁺ CD8⁺ T cells in dLNs, but this was not the case in tumor tissue (Fig. 4B, 4C). Contrary to carboplatin, cisplatin significantly decreased the number of CD8⁺ and CD4⁺ T cells in dLNs in the combination setting and also led to a reduced number and activation of T cells in blood (Supplemental Table II). Oxaliplatin, contrary to carboplatin and cisplatin, led to a decrease in IFN-γ⁺ CD8⁺ T cells in the tumor, with GZMB levels also trending down in the combination setting (Fig. 4C). Despite these changes, neither agent antagonized the activity of anti–PD-L1 and, on the contrary, enhanced antitumor responses (Fig. 4D, 4E). These results suggest that the effect of chemotherapy, at least for these platinum-based agents, is not class-specific, but rather each treatment has a particular effect on immune cell subsets in different tissues.

Given the wide use of taxanes, such as paclitaxel and docetaxel, for the treatment of various cancers, we also wanted to determine the effect of these chemotherapies on the antitumor activity mediated by anti–PD-L1. Similar to platinum-based chemotherapies, we observed a more pronounced and antagonistic effect in dLNs and blood when these taxanes were given as a single agent or in combination with anti–PD-L1, with minimal effects observed in TILs (Supplemental Tables I–II). Additionally, paclitaxel and docetaxel led to differential effects in peripheral tissues without a generalized class-specific effect, and, when combined with anti–PD-L1, these chemotherapies did not antagonize antitumor responses (Supplemental Fig. 1). From these results, it is evident that the effects of chemotherapy are agent specific and do not appear to generally antagonize the activity of anti–PD-L1, largely preserving the phenotype of TILs.

Because immune modulation by chemotherapy is agent specific, we wanted to investigate the activity of additional chemotherapies with well-known immune modulatory effects like CTX. CTX has been shown to enhance dendritic cell expansion and cross-presentation, stimulate NK and CD8⁺ T cells, deplete Tregs, and help modulate the myeloid compartment (20–24). When tested as single agent, CTX significantly reduced the frequency of activated ICOS⁺ CD8⁺ T cells in dLNs without decreasing their absolute numbers (Supplemental Table I). However, the frequency of Ki-67⁺ CD8⁺ T cells in the tumor tissue was significantly reduced, whereas their activation state was preserved. Interestingly, in the combination setting, CTX significantly increased the frequency and number of activated CD8⁺ T cells in the tumor as measured by IFN-γ, ICOS, and GZMB expression without significantly reducing the level of Ki-67–expressing cells (Fig. 5A, 5B). Not surprisingly, these changes translated into significantly enhanced antitumor activity, resulting in 90% CRs in the combination setting despite only modest single-agent CTX activity (Fig. 5C). This pronounced effect on CD8⁺ T cell activation in the tumor was not observed with the other chemotherapies analyzed and was quite striking given the detrimental effect on activated CD8⁺ T cells in dLNs and the reduction in Ki-67–expressing cells in the tumor when given as single agent. Additional changes in immune cell subsets are summarized in Supplemental Table II. These results suggest that CTX is indeed a strong modulator of T cell responses with observed beneficial effects in the tumor tissue when combined with anti–PD-L1 treatment, but similar to the other chemotherapies, it could antagonize responses in secondary lymphoid organs.

Gemcitabine is another potent immune modulator with known myelosuppressive effects (25–28). Similar to CTX, treatment with gemcitabine showed a trend toward decreasing the frequency of ICOS⁺ CD8⁺ T cells in dLNs with a significant decrease on the level of intratumoral Ki-67–expressing cells (Supplemental Table I). There was no change in the number of CD8⁺ and CD4⁺ T cells in dLNs following treatment, and, unlike combination with CTX, combination with gemcitabine did not show enhancement in the frequency of activated CD8⁺ T cells in the tumor but rather significantly decreased the frequency of Ki-67– and GZMB-expressing cells (Fig. 6A). This was accompanied by a decrease in the absolute number of intratumoral CD8⁺ T cells (Fig. 6B) that led to a decrease in the CD8/Treg ratio (Fig. 6C). Given these antagonistic effects on CD8⁺ T cells in both dLNs and tumor tissue, we did not expect gemcitabine to combine effectively with anti–PD-L1 treatment. However, despite modest single-agent activity, combination with gemcitabine led to enhanced antitumor responses, resulting in 40% CRs in the combination setting (Fig. 6D), demonstrating that gemcitabine significantly improves the activity of anti–PD-L1 treatment despite reducing the levels of activated intratumoral CD8⁺ T cells.

To demonstrate that the activity of gemcitabine in combination with anti–PD-L1 was driven by T cells, we evaluated antitumor responses in immunodeficient RAG^−/− mice that lack mature T cells and B cells. As shown in Supplemental Fig. 2A, anti–PD-L1 treatment had no activity in RAG^−/− mice, whereas gemcitabine treatment alone or in combination showed only partial activity (driven by gemcitabine), which was short lived as tumors quickly regrew, resulting in no CRs as was observed in wild-type mice. To further corroborate that the combination efficacy was dependent on CD8⁺ T cells despite their lower numbers following gemcitabine treatment, we depleted CD8⁺ T cells in established tumors from wild-type mice. As expected, anti–PD-L1 activity was lost, and single-agent gemcitabine and combination treatment showed only partial activity, which was quickly followed by tumor regrowth, demonstrating a lack of effective tumor control in the absence of CD8⁺ T cells (Supplemental Fig. 2B). These results corroborate that despite the effect of gemcitabine on CD8⁺ T cells, these cells are still required to drive the antitumor activity when combined with anti–PD-L1.

The strong antitumor activity of gemcitabine in combination with anti–PD-L1 was unexpected, given its antagonistic effect on intratumoral CD8⁺ T cells. This led us to speculate that gemcitabine must be affecting a major immunosuppressive component in the tumor microenvironment that compensated for the reduced numbers of activated CD8⁺ T cells while maintaining robust antitumor activity. In addition to direct antitumor effects, gemcitabine is known to be myelosuppressive and could potentially be reducing the numbers of tumor-associated myeloid cells. At the time of treatment initiation, established MC38 tumors were mostly infiltrated by macrophages defined as CD11b⁺CD11c^−/intLy-6G⁻Ly-6C⁻F4/80⁺ cells (Fig. 7A, 7B). These tumor-associated macrophages (TAMs) also expressed the highest levels of PD-L1 relative to the other cell subsets (Fig. 7C) and are likely main contributors to PD1/PD-L1–mediated suppression in the tumor tissue. Consistent with its known myelosuppressive effects, gemcitabine alone or in combination with anti–PD-L1 led to a significant decrease in the frequency of TAMs (Fig. 7D).

To further corroborate that in the MC38 tumor model macrophages have a suppressive phenotype, we subdivided the macrophage population based on MHC class II (MHC-II) and mannose receptor (CD206) expression (Fig. 7A). MHC-II expression is normally associated with a more proinflammatory, M1-like phenotype, whereas CD206 expression is associated with suppressive, M2-like macrophages (29–31). By using these two markers, we were able to subdivide the TAM population into three main subsets: an M1-like subset (CD206⁻MHC-II⁺), an intermediate subset (CD206⁺MHC-II⁺), and an M2-like subset (CD206⁺MHC-II⁻); we determined that most macrophages showed an intermediate and M2-like phenotype in established MC38 tumors (Fig. 7E). Additionally, this enriched CD206⁺ TAM population also expressed the highest levels of arginase 1 (ARG1), an immunosuppressive enzyme that processes l-arginine into l-ornithine and urea and can profoundly suppress T cell responses (31–33). Altering the macrophage population in the MC38 tumor model could therefore have beneficial consequences for antitumor activity by affecting a major immunosuppressive component in the tumor microenvironment. Indeed, gemcitabine treatment alone or in combination with anti–PD-L1 significantly decreased the number of intratumoral CD206⁺MHC-II⁻ macrophages relative to anti–PD-L1 treatment (Fig. 8A). This effect on TAMs could have important implications for antitumor activity as the relative abundance of CD206⁺ macrophages was correlated with tumor burden (Fig. 8B). The effect of gemcitabine on TAMs could therefore outweigh some of its negative impacts on CD8⁺ T cells by allowing fewer activated CTLs to effectively control tumor growth. Together, these results demonstrate that combination treatment with gemcitabine, although it reduces the number of activated CD8⁺ T cells, also reduces the number of suppressive TAMs in the tumor, thereby allowing fewer activated CD8⁺ T cells to effectively control tumor growth.

Given that the MC38 tumor model is highly enriched in macrophages with a suppressive phenotype, we hypothesized that using other approaches to target this population while preserving CD8⁺ T cell functionality should help augment the activity of anti–PD-L1 treatment. To test this hypothesis and to further corroborate that the improved activity following gemcitabine treatment is partly due to its effect on TAMs, we treated MC38 tumor-bearing mice with anti–PD-L1 alone or in combination with anti-CSF1R to deplete macrophages (16). Because gemcitabine not only targets TAMs but also has a plethora of effects on other immune cell components and tumor cells, we were not expecting to replicate the effects observed with gemcitabine combination, but we were expecting to observe combinatorial activity with anti–PD-L1 mediated by the depletion of suppressive TAMs. Indeed, anti-CSF1R treatment combined effectively with anti–PD-L1, leading to increased tumor control and delaying tumor growth relative to either treatment alone (Fig. 8C). These results highlight that in the particular immune contexture of this tumor model, targeting TAMs, which represent a major immunosuppressive component of the tumor microenvironment, augments tumor killing and that gemcitabine could indeed be conferring improved antitumor responses by significantly decreasing the population of suppressive TAMs in the tumor tissue.

Chemotherapies can have various effects on the immune system. Some agents can have immunosuppressive side effects by directly depleting immune effector cells or by inhibiting their function (11), whereas others can enhance the immunogenic properties of tumor cells, thus favoring T cell immunosurveillance over suppression (10, 12, 23, 34). Chemotherapies can also modulate different immune cell subsets, conferring enhanced antitumor activity (22, 26, 27). In this study, we show that combining anti–PD-L1 with different chemotherapy agents did not antagonize responses in established MC38 tumors and in many cases augmented antitumor activity. Despite apparent antagonistic effects on T cells in peripheral tissues like dLNs and blood, the various chemotherapies tested in this study did not significantly affect TILs.

This differential effect could be due to various factors, including differences in drug exposure and other local mediated effects (35). Similar differences between local and systemic effects of chemotherapy have been observed in a glioblastoma murine tumor model in which local chemotherapy enhanced the activity of anti-PD1, but systemic chemotherapy led to lymphodepletion in dLNs and peripheral blood (36). Another possible explanation is that chemotherapies affect T cells differently based on their differentiation state. In the MC38 tumor model, most infiltrating CD8⁺ T cells have an effector phenotype (CD62L^loCD44^hi/lo), whereas most cells in dLNs are naive (CD62L⁺CD44^lo) (data not shown). Even though the chemotherapies tested in this study are antineoplastic, it is possible that the effect of these agents is more pronounced in cells undergoing priming, activation, and expansion at peripheral sites rather than tumor-infiltrating effector cells that are less proliferative (37–39).

Despite most chemotherapies having an antagonistic effect on CD8⁺ T cells in dLNs and blood, this effect did not translate into reduced antitumor activity. CD8⁺ T cells drive anti–PD-L1 responses in the MC38 tumor model, so decreasing CD8⁺ T cell numbers and activation state should have translated into reduced antitumor activity. However, by blocking lymph node egress with FTY720 (fingolimod) (19), we showed that tumor responses to anti–PD-L1 in established tumors were not majorly dependent on incoming cells from the periphery but rather relied mostly on TILs at the time of treatment initiation. This finding suggests that when working with established syngeneic tumor models of short duration, the peripheral effect of a particular treatment might not be reflected in the overall antitumor activity. Therefore, our results reveal that even though none of the chemotherapies tested in combination with anti–PD-L1 led to a reduction in antitumor activity, the effect of these chemotherapies at peripheral sites could affect subsequent responses, and this will likely depend on how durable the antagonistic effects are and how fast the numbers and phenotype of T cells can rebound following chemotherapy treatment.

It was also evident from our studies that the effect of chemotherapy was agent specific and not class specific. This was demonstrated by differences in the pharmacodynamic regulation at both peripheral and tumor tissue for carboplatin, cisplatin, and oxaliplatin. Despite acting through DNA platination, which leads to Pt-DNA lesions that can accumulate and result in apoptosis (40), these drugs had various effects on tumor and immune cells. Cisplatin showed a more pronounced inhibition of CD8⁺ and CD4⁺ T cell numbers in blood and dLNs in combination with anti–PD-L1, whereas carboplatin and oxaliplatin decreased the frequency of activated T cells in dLNs, with only oxaliplatin significantly reducing the frequency of IFN-γ⁺ CD8⁺ T cells in the tumor. These chemotherapies, however, combined effectively with anti–PD-L1 treatment with carboplatin and oxaliplatin, showing more robust combinatorial activity.

The effects of chemotherapy are therefore drug specific and dependent on multiple factors. In this study, we focused on the effects of chemotherapy on various immune cell subsets in peripheral and tumor tissue and explored how both antagonistic and beneficial effects could balance out to establish a new threshold for antitumor response as observed with gemcitabine treatment. However, additional factors not investigated in this study are likely involved in driving the overall combinatorial activity of chemotherapy, such as the effects of drugs on tumor cell death (tolerogenic versus immunogenic) or microregional effects of chemotherapy on the tumor tissue-like changes in interstitial fluid pressure, vasculature, and desmoplastic architecture. Multiple studies have suggested that different chemotherapies can result in immunogenic cell death, leading to the release of damage-associated molecular pattern molecules like high mobility group box 1 protein (HMGB1), calreticulin, ATP, and heat shock proteins (HSPs) that can sensitize tumors to immune checkpoint blockade (10, 12, 34). Although some chemotherapies tested in this study could have resulted in immunogenic cell death, we focused our efforts on identifying the subsequent immunological changes and started delineating how these could affect overall antitumor responses.

Certain chemotherapies, such as CTX and gemcitabine, have been shown to have various effects on specific immune cell populations (20, 26, 41, 42). These two chemotherapies showed strong combinatorial activity with anti–PD-L1, whereas their single-agent effect was only modest and mostly delayed tumor growth. Both chemotherapies negatively affected T cells in dLNs and blood, but their effect in TILs was quite distinct. Combination with CTX led to a clear enhancement in CD8⁺ T cell numbers and activation state, although gemcitabine decreased CD8⁺ T cell numbers as well as the frequency of GZMB- and Ki-67–expressing cells. The MC38 tumor model is well infiltrated by CD8⁺ T cells, and increasing their number and activity was expected to further enhance the response to anti–PD-L1, which is in itself already augmenting the activity of CTLs. For gemcitabine, the strong combinatorial effect with anti–PD-L1 was more surprising, given that activity in this tumor model is dependent on CD8⁺ T cells. Upon further analysis, it was evident that even though gemcitabine reduced the number of CD8⁺ T cells, its effect on decreasing suppressive CD206⁺MHCII^lo macrophages probably balanced out the decrease in CTL numbers. In the MC38 tumor model, TAMs represent the major immune infiltrate, and these cells show a suppressive phenotype (CD206⁺ARG1⁺) (43). By reducing this inhibitory barrier, gemcitabine treatment was able to bolster antitumor responses despite reducing the number of CTLs. This likely led to the establishment of a lower response threshold, allowing fewer CTLs to effectively control tumor growth. This is likely why other chemotherapies that were largely muted in their tumor effect were still able to enhance antitumor activity. Even though few significant pharmacodynamic changes were observed for some of these chemotherapies in the tumor, there were possibly enough small changes that translated into a more favorable response threshold. It is important to point out that the strong combinatorial activity of gemcitabine is likely tumor specific and, as observed for the MC38 tumor model, dependent on the balance between tumor-infiltrating populations and how these are affected by chemotherapy treatment. In other tumor models that do not share this level of infiltrating CTLs together with a suppressive macrophage population, the effect of gemcitabine combination could be antagonistic. To further corroborate whether targeting TAMs could help modify the threshold for antitumor activity in the MC38 tumor model and enhance responses to anti–PD-L1, we treated tumor-bearing mice with anti-CSF1R, which has been shown to strongly reduce the number of tumor macrophages (16). Targeting TAMs with anti-CSF1R led to an enhancement in anti–PD-L1–mediated activity, corroborating that in this particular tumor model, depleting macrophages, which represent a major immunosuppressive population, can help augment antitumor responses.

Our results suggest that the effects of combining checkpoint inhibitors (CIT) such as anti–PD-L1 with chemotherapy will depend on the specific tumor-immune contexture at the time of treatment initiation. The antagonist effect of some chemotherapies at peripheral sites might affect the level of new Ag presentation, priming, and repopulation of tumor immune cell subsets, but these cells might recover between dosing cycles. Thus, the detrimental effect of chemotherapy at peripheral sites will likely be only transient, and recruitment of new T cells to the tumor tissue could still be achieved. In the MC38 tumor model, none of the chemotherapies tested antagonized the activity of anti–PD-L1, and in many cases they significantly improved responses as observed for carboplatin, oxaliplatin, CTX, and gemcitabine. The effects of these chemotherapy combinations were agent specific, and the pharmacodynamic changes were distinct for each of the chemotherapies. Overall, our data support the combination of CIT agents with chemotherapy in earlier lines of treatment, but further understanding the effects specific chemotherapies have on tumor and immune cell subsets as well as other microenvironmental components, such as stromal and endothelial cells, will improve the selection of such agents for combination treatment. Our results also provide some initial evidence for the potential of chemotherapy combinations that may not have been previously realized and that are not commonly used as standard of care but that should be further investigated because of their effects on specific immune cell populations. This could result in a more personalized approach to chemotherapy–CIT combinations by trying to target specific tumor-immune contextures when possible.

We thank the core dosers at Genentech as well as the cell line core group for their contribution to this project. We also thank Marcia Belvin for useful discussions of the data.

All authors are employees and stock holders of Genentech or Roche.