Long-Term Gemcitabine Treatment Reshapes the Pancreatic Tumor Microenvironment and Sensitizes Murine Carcinoma to Combination Immunotherapy

Pancreatic ductal adenocarcinoma (PDAC) often presents at late clinical stages and is associated with poor outcomes. As the majority of patients are diagnosed with metastatic disease and are not eligible for surgery, most are managed through broad-spectrum chemotherapy (1). While aggressive chemotherapy marginally improves survival, nearly all tumors either have or develop some degree of drug resistance. Thus, median survival remains a dismal 6–12 months (2). Gemcitabine was first approved for use in metastatic PDAC in 1996 (3–5), and is used first line with albumin-conjugated (nab) paclitaxel (6). In addition, the FOLFIRINIOX regimen (folinic acid, 5-fluorouracil, irinotecan, and oxaliplatin) is also widely used in metastatic PDAC, showing superior efficacy but a higher rate of serious adverse effects (7). While both regimens offer a survival benefit to most patients, most eventually progress on treatment. With no FDA-approved third-line medications, patients with chemorefractory PDAC are generally provided only symptomatic or hospice care. As such, there is an urgent clinical need to identify novel therapeutic approaches for patients in the second- or third-line setting.

While there has been extensive research into the mechanisms that underlie the emergence of chemoresistance in PDAC, none has successfully translated to a new treatment modality. However, recent evidence suggests that in addition to their well-documented effects on nucleotide synthesis, DNA polymerization, and translation, select antineoplastic agents may alter a variety of additional cell functions, including the processing and presentation of self-peptide (8–10). Classically, PDAC is poorly immunogenic with diminished antigen presentation (11) and a highly immunosuppressive TME that further impedes the functional activation of CTLs (12). As such, single-agent immune checkpoint inhibitors (ICI) have yet to show clear efficacy in the PDAC (13–16). As invasive strategies to promote antigen presentation such as irreversible electroporation have helped overcome the innate resistance of PDAC to ICIs (17), should cytotoxic chemotherapy similarly enhance the presentation of tumor antigen, its addition to select immunotherapies may improve drug responses.

We therefore evaluated the effects of prolonged gemcitabine treatment using both established models of murine PDAC and in vitro models of gemcitabine resistance. We determined that tumor-bearing mice treated with gemcitabine for several months developed large-scale alterations in the expression of several immune surface proteins including MHC class 1, PD-L1, and PD-L2. Gemcitabine similarly altered the composition of the tumor secretome, increasing a variety of CCL/CXCL family cytokines and TGFβ-associated signals. These secreted factors conferred a gemcitabine-resistant phenotype to tumor stromal cells in vitro, further enhancing stellate cell synthesis of immunomodulating cytokines including TGFβ1.

We next challenged transgenic models of early and advanced PDAC with a combination of gemcitabine and anti-PD-1, which failed to significantly alter disease course unless mice also had genetic or pharmacologic ablation of TGFβ signaling. In the setting of TGFβ signal deficiency, gemcitabine and anti-PD-1 led to significant regression of established disease, hallmarked by increased CTL infiltration and activity exceeding that observed with PD-1/TGFβ inhibition alone. Combined, these observations suggest that select antineoplastic agents such as gemcitabine may prime tumor cells for ICIs by enhancing the presentation of self-peptide, but require the added intervention of an immunosuppressive cytokine barrier. Given the current lack of third-line treatment options, this combined approach warrants consideration in gemcitabine-refractory PDAC, potentially offering a new therapeutic option in patients for which there is currently no effective treatment.

All antibodies were purchased from established commercial vendors, and were verified by the manufacturer for the specific species and applications for which they were used. A full list of all antibodies used as well as the vendor, clone, and product numbers can be found in Supplementary Table S1.

Nongenic B6 (wild type), Ptf1a-Cre x LSL-Kras^G12D (KC), Tgfbr1^+/−, KC/Tgfbr1^+/− (KCT), Pdx1-Cre x LSL-Kras^G12D x LSL-TP53^R172H (KPC), and Pdx1-Cre x LSL-Kras^G12D x LSL-TP53^R172H+/+ (KPPC) mice were generated as described in our previous work (18), or purchased from an in house vendor at the University of Illinois at Chicago (Chicago, IL). At roughly 12 weeks of age, KC and KCT mice were administered an intraperitoneal injection of either a PBS vehicle or daily galunisertib (75 mg/kg) and a fixed dose of 200 μg anti-PD-1 twice per week. Mice were euthanized after 4 months of treatment and tissues collected for analysis. For studies involving KPC mice, animals were administered intraperitoneal injection of a PBS vehicle, gemcitabine (100 mg/kg), galunisertib, anti-PD-1, galunisertib and gemcitabine, galunisertib and anti-PD-1, or triple combination of galunisertib, gemcitabine, and anti-PD-1 as described. KPC mice were sacrificed when moribund or showing clear signs of health decline, for example, fur loss, weight loss, or lethargy, or when they reached 8 months of age in the case of mice with prolonged survival. For euthanasia, animals were anesthetized with isoflurane until unresponsive to toe tap and/or agonal breathing. Thoracotomy served as the primary method of euthanasia and exsanguination the secondary method. For all mouse studies, males and females were randomized at a 50:50 ratio.

The G-68 cell line was established from a non-Hispanic, white female with a T3N1, moderately differentiated, PDAC using methods described previously (19), and cultured in DMEM/F12 media supplemented with 10% heat-inactivated FBS, penicillin (100 U/mL), and streptomycin (100 mg/mL). A total of 5 × 10⁶ cells were suspended in 100 μL of a 1:1 DMEM/F-12 with Glutamax media and Matrigel (Corning) and inoculated subcutaneously into the right flank of NSG mice (The Jackson Laboratory). Tumor size was measured twice weekly with digital caliper. For treatment, mice were randomized into treatment groups when the tumors reached 100–200 mm³, with an average tumor volume of 145 mm³. Once randomized, animals were treated with either PBS vehicle or 40 mg/kg gemcitabine via intraperitoneal injection. Mice were euthanized when moribund, when the maximum tumor size allowed per to institutional policy (2 cm), or when tumors became ulcerated. For euthanasia, animals were scarified by CO₂ suffocation followed by cervical dislocation, and tumors subsequently harvested and processed as described above.

All experiments involving the use of mice were performed following protocols approved by the Institutional Animal Care and Use Committee at the University of Illinois at Chicago. All human tissues were obtained from patients providing fully informed, written consent. For histology, tissues were obtained in a deidentified manner from the Northwestern University Pathcore (Chicago, IL) following local Institutional Review Board (IRB) approval. Similarly, the G-68 cell line was generated from a fully consenting patient and following local IRB approval/deidentification at the University of Florida (Gainesville, FL).

Data were analyzed by either Student t test, simple linear regression analysis, or ANOVA fit to a general linear model in Minitab express, the validity of which was tested by adherence to the normality assumption and the fitted plot of the residuals. Results were arranged by the Tukey method, and considered significant at P < 0.05 unless otherwise noted. Results are presented as either boxplot showing the median value and all other values arranged into quartiles, or as the mean of individual replicates plus SD.

For additional materials and methods, please see the Supplementary Materials, which also contains the following references (12–14, 18, 20–32).

To evaluate the effects of long-term chemotherapy within the pancreatic TME, we used the well-established Pdx1-Cre x LSL-Kras^G12D x LSL-TP53^R172H (KPC) model of invasive PDAC. This model faithfully recapitulates human PDAC histotypes with several key features including poor immunogenicity and a dense, reactive tumor stroma. KPC animals were initially reported to develop precursor pancreatic intraepithelial neoplasms (PanIN) at 6 weeks, focal PDAC lesions at approximately 15 weeks, and advanced PDAC at 18 weeks (33), although this can vary by individual colony and background strain. In our colony, KPC mice are maintained in full a C57/B6 background and have a mean survival of 124 days (N = 30). Therefore, to model the long-term effects of gemcitabine, we elected for an early intervention study in which animals were allowed to develop overt disease for a minimum of 90 days (roughly 13 weeks), at which point, mice in our colony generally have a combination of normal glandular tissue, low- and high-grade PanIN lesions, as well as scant, focal areas of PDAC (Supplementary Fig. S1A–S1C). Animals were then randomized at a 50:50 male to female ratio into one of two treatment groups (N = 4/group). Mice were treated with intraperitoneal injections of either PBS vehicle or 100 mg/kg gemcitabine twice per week, and sacrificed when showing clear signs of health decline, for example, weight loss, ascites, or lethargy. Consistent with previous reports, gemcitabine provided a modest survival benefit averaging three weeks (Supplementary Fig. S1C).

After collection, tissues were sectioned and stained either with hematoxylin and eosin (H&E), or by IHC for clinically relevant immune surface proteins. Consistent with improved antigen presentation, gemcitabine led to significant increases in MHC Class I in neoplastic tissues, as well as immune checkpoints PD-L1 and PD-L2 (Fig. 1A and B). However, despite the presumptive increase in antigen presentation, gemcitabine failed to increase the number of tumor-infiltrating lymphocytes, which were largely confined to the periphery of tumor specimens in both treatment groups (Fig. 1A and B). On the basis of these observations, we next evaluated gemcitabine-induced alterations in tumor cytokines by homogenizing tumor tissue and conducting a high-throughput array of 111 immunoregulatory proteins (Fig. 1C and D; Supplementary Fig. S1D; N = 4/group).

After normalizing to reference samples, we identified consistent alterations to the tumor secretome of gemcitabine-treated animals, as well as select immune surface proteins. Specifically, gemcitabine-treated mice had significant increases in a variety of CCL and CXCL family chemokines, with CCL6 and 17 being the highest expressed (Fig. 1D). Gemcitabine-treated mice also displayed increased expression of several interleukins, including IL10 and IL28 (Supplementary Fig. S1D). Gemcitabine similarly enhanced the expression of TNFα and IFNγ, the latter consistent with the increased expression of MHC class I and PD-L1/PD-L2 (Supplementary Fig. S1D). We also observed gemcitabine-induced upregulation of a variety of immune surface proteins, including CD26, CD40, CD54, and TGFβ receptor complex member CD105 (Supplementary Fig. S1D), as well as neutrophil-associated proteins myeloperoxidase (MPO) and neutrophil gelatinase-associated lipocalin (NGAL; Supplementary Fig. S1D). Tumor lysates were also evaluated for expression of TGFβ1 by ELISA, which showed a similar increase in TGFβ1 expression in the pancreas of gemcitabine-treated mice (Supplementary Fig. S1D).

We next used the KPC-derived PDAC cell line KPC105 and conducted a similar study in vitro. Cells were incubated with 2.5 μmol/L gemcitabine and lysates collected after 48 hours. Cell extracts were then subjected to the same high-throughput cytokine screen described above, which showed a similar upregulation of CCL and CXCL family chemokines in response to a transient course of gemcitabine, as well as TGFβ1 by ELISA (Supplementary Fig. S1E and S1F).

Given the apparent alterations in the immunogenicity of murine PDAC, we next sought to evaluate the effects of long-term chemotherapy on human tumor cells. We therefore generated a primary pancreatic cancer cell line (G-68) from a non-Hispanic, white female with a T3N1 tumor harboring both KRAS^G12D and TP53^R248W mutations similar to those used in the KPC model. A total of 5 × 10⁶ cells were subsequently injected into the flank of right flank of NSG mice, and once tumors reached 100–200 mm³ in size, animals were treated with either vehicle or 40 mg/kg of gemcitabine once per week (N = 4–5/group). Tumors were initially responsive to gemcitabine for an average of four weeks, after which, tumor growth accelerated to the normal rate. At this point, mice were considered to have developed clinical gemcitabine resistance. This temporary tumor stasis corresponded to delayed mortality (Fig. 2A), though this was not reflected in tumor size or any observable change in histopathology (Fig. 2B and C). Similar to results observed in KPC mice, while chemotherapy did not alter the frequency of neoplastic lesions, gemcitabine treatment led to significant increases in the human MHC class I analog HLA-A, B, C in the neoplastic epithelium, as well as immune checkpoints PD-L1 and PD-L2 (Fig. 2D and E).

As with KPC mice, we next evaluated gemcitabine-induced alterations in tumor cytokines by homogenizing tumor tissue and conducting a high-throughput array of 105 human immunoregulatory proteins (Supplementary Fig. S1G, N = 4/group). Again paralleling results in KPC mice, despite the lack of an adaptive immune system, G-68–derived tumors treated with gemcitabine displayed highly significant increases in a variety of CCL/CXCL family chemokines, including CCL5, CCL20, CXCL8, CXCL10, and CXCL12 (Fig. 2F). Gemcitabine-treated mice also had increased intratumoral expression of inflammatory cytokines IL1 and IL32, as well as CD54, NGAL, MIF, and the TGFβ family ligand MIC-1, (Fig. 2F). Like MIC-1, tumor lysate from gemcitabine-treated mice also had significant upregulation of the immunosuppressive cytokine TGFβ1 as determined by ELISA (Fig. 2F, N = 3/group). We then repeated this experiment using G-68 tumor cells in vitro, which confirmed the upregulation of several previously identified cytokines in response to gemcitabine (Supplementary Fig. S2A and S2B). Interestingly, when evaluating the TCGA genomic databases of patients with pancreatic cancer (N = 186), we determined that mRNA expression of several of these CCL/CXCL chemokines, as well as genetically and functionally related family members, had statistically significant associations with mRNA expression of the cytotoxic T-cell marker CD8A (Fig. 2G).

To determine that the observed alterations are indeed a consequence of gemcitabine resistance in the cancer epithelium, we next sought to evaluate the effects of both transient and long-term chemotherapy on established human tumor cell lines in vitro. We therefore used Panc1 cells and determined the IC₅₀ of the five currently FDA-approved first-line chemotherapy agents. These include gemcitabine (1 μmol/L), paclitaxel (100 nmol/L), 5-fluorouracil (2.5 μmol/L), irinotecan (2.5 μmol/L), and oxaliplatin (2.5 μmol/L). After 48 hours, dead cells were excluded by live/dead gating, and live cells were analyzed by flow cytometry for surface expression of the HLA-A, B, C, PD-L1, PD-L2, CTLA-4, and CD86, as well as intracellular expression of TGFβ1.

In response to gemcitabine or paclitaxel, Panc1 cells had enhanced surface expression of PD-L1, PD-L2, CTLA-4, HLA-A, B, C, and modestly increased expression of CD86 (Fig. 3A and B). While 5-FU reduced cell viability as expected, we observed no significant change in PD-L1 or CD86 expression, and only slight increase in PD-L2, CTLA-4 and HLA-A, B, C levels (Fig. 3A and B). Similarly, irinotecan strongly enhanced expression of PD-L2, and mildly induced CTLA-4, HLA-A, B, C, and CD86 with no observable effect on PD-L2 (Fig. 3A and B; Supplementary Fig. S2A). Oxaliplatin was the least immunogenic of the drugs evaluated, causing only slight elevations in PD-L2, CTLA-4, and HLA-A, B, C (Fig. 3A and B; Supplementary Fig. S2A). After confirming these results in other cell lines (Supplementary Fig. S3A and S3B), we incubated Panc1 cells in increasing concentrations of gemcitabine for several passages until viable in a high concentration of 10 μmol/L. These gemcitabine-resistant (GR) cells now had stronger, persistent upregulation of PD-L1, PD-L2, CTLA-4, HLA-A, B, C, and CD86, as well a modest increase in intracellular TGFβ1 (Fig. 3A and B).

After verifying these observations by immunocytochemistry (Supplementary Fig. S3C), we next incubated Panc1 cells with a transient 48-hour course of 0, 1, or 5 μmol/L gemcitabine, or used Panc1-GR cells grown in 10 μmol/L gemcitabine for several passages. After 48 hours, cells were treated with a protein transport inhibitor, and lysates subjected to a high-throughput screen of 105 tumor-derived immunoregulatory proteins analogous to that described in KPC mice. This allows for the simultaneous evaluation of 105 cytokines, chemokines, and immunomodulating proteins (Fig. 3C; Supplementary Fig. S3D). While transient incubation with gemcitabine had modest effects on cytokine production, Panc1-GR cells had significant alterations in a wide variety of functionally-related proteins, particularly those involved in inflammation/chemotaxis, angiogenesis, and/or TGFβ signaling. For instance, Panc1-GR cells displayed highly significant increases in the expression of the pro-inflammatory chemokines CCL2, CCL4, CCL20, CXCL1, CXCL5, and CXCL8, closely resembling the phenotype of KPC and xenografted mice (Fig. 3C). Panc1-GR cells also had robust expression of the IL6 regulator FGF2, as well as the proinflammatory cytokine IL32 and CD56 (Fig. 3C). In addition, GR cells had upregulation of proangiogenic factors Angiogenin and the TGFβ family ligand MIC-1 (Fig. 3C). Contrasting our observations in mice, Panc1-GR cells had diminished expression of the TGFβ receptor complex member CD105 (Fig. 3C).

Given the apparent aberrations to TGFβ signaling and the established roles for TGFβ in immune evasion in PDAC (12, 18, 34), we next evaluated the concentration of TGFβ1 in the culture media of Panc1 in response to a 72-hour course of 0, 1, or 5 μmol/L gemcitabine, or in Panc1-GR cells by ELISA. While high-dose gemcitabine led to a modest increase in TGFβ1 in the culture media, Panc1-GR cells had a near 3-fold increase in secreted TGFβ1 compared with chemo-naïve Panc1 cells (Supplementary Fig. S4A). Given the established roles of TGFβ in expanding the tumor stroma (12), which itself can also contribute to the clinical failure of gemcitabine by acting as a drug scavenger (35), we next explored the capability of Panc1-GR cells to modify the activity of stromal cells.

Next, an equal number of Panc1 and Panc1-GR cells were grown in serum-free DMEM for 24 hours, after which, media were collected, filtered, and evaluated for TGFβ1 expression (Fig. 4A). This media was then supplemented with 10% FBS and transferred to one of three stromal cell lines: cancer-associated fibroblast cells CAF2 and CAF3 or the human pancreatic stellate cell line hPSC. After 48 hours, media was collected and reevaluated for TGFβ1 by ELISA. While Panc1 conditioned media increased TGFβ1 secretion by CAF2 and hPSC cells, Panc1-GR conditioned media led to greater increases in TGFβ1 in all cell lines, surpassing the sum of TGFβ1 produced by tumor and stromal cells in isolation (Fig. 4B).

Given the prosurvival role TGFβ1 in mesenchymal cells, the experiment was repeated, and cells grown in increasing concentrations of gemcitabine ranging from 0.5 to 10 μmol/L. While Panc1 conditioned media modestly enhanced gemcitabine resistance in all cell lines, the addition of Panc1-GR conditioned media further enhanced this response, particularly in CAF2 and CAF3 cells (Fig. 4C, N = 4). Given the known immunosuppressive role of stromal-derived TGFβ in PDAC (12), we next grew hPSC cells in either control media, Panc1 conditioned media, Panc1-GR conditioned media, or Panc1-GR conditioned media supplemented with 10 μmol/L of the type 1 TGFβ receptor (TGFBR1) inhibitor galunisertib, which was not toxic to any cell line at the concentration used (Supplementary Fig. S4B). After 48 hours, we repeated the described high-throughput screen of immunoregulatory proteins (Fig. 4D; Supplementary Fig. S4C, N = 2/group). Despite the difference with respect to gemcitabine resistance, both Panc1 and Panc1-GR conditioned media had similar effects on the stellate cell secretome, leading to increases in several factors observed in our previous arrays. Both Panc1 and Panc1-GR media enhanced hPSC synthesis of CXCL8, CD56, FGF2, CD54, MIC-1, CD105, and the lymphokine MIF; however, only Panc1-GR media significantly upregulated IL17 (Fig. 4D). While galunisertib had little effect on unstimulated stromal cells, the addition of galunisertib to Panc1-GR media neutralized the induction of all upregulated immunoregulatory proteins, with the exception of MIF (Fig. 4D). Similarly, galunisertib prevented the gemcitabine resistance conferred to stromal cells by Panc1-GR media, restoring drug sensitivity to near basal levels (Fig. 4E, N = 4).

Though we observed some heterogeneity across model systems, gemcitabine appears to ubiquitously alter the expression of several tumor-derived cytokines (Supplementary Fig. S4D). Of the most frequently altered signals, TGFβ remains the only such cytokine with an inhibitor currently in clinical trial for PDAC. Furthermore, the contributions of TGFβ signals to immune evasion in PDAC are well established (12, 18), and our data to this point suggest that in the setting of gemcitabine resistance, TGFβ serves as a master regulator of the tumor microenvironment. Therefore, to determine the potential clinical relevance of our findings, we first evaluated the mRNA expression of clinically actionable immune checkpoints PD-L1 (CD247) and PD-L2 (PCDCLG2), as well as a variety of genes associated with the TGFβ signaling pathway, and leukocyte-specific genes including CD3D, CD3E, CD3G, CD4, CD8A, CD8E, and CD45 (Supplementary Fig. S5A; Supplementary Table S2). Interestingly, although only PD-L1 expression was a significant predictor of poor outcomes, a high TGFβ gene signature positively associated with expression of PD-L1 and PD-L2, as did increased expression of several leukocyte markers including CD8A and CD45 (Supplementary Fig. S5B–S5G). We therefore explored the relationship between PD-L1 expression (by IHC) and lymphocyte infiltration (by H&E staining) in a group pancreatic cancer excisional biopsies collected from Northwestern University (Chicago, IL), examining both adjacent normal (N = 27) and PDAC (N = 47) specimens. While the mean lymphocyte infiltrate PDAC exceeded that of adjacent normal sections, this was highly varied with some sections having almost no tumor-infiltrating lymphocytes, whereas others had areas of overwhelming lymphocytosis (Supplementary Fig. S5H–S5J).

On the basis of these observations, we next set to determine the combined therapeutic efficacy of gemcitabine and immune checkpoint inhibition in vivo, with and without TGFβ signal inhibition. To address this, we used both the Ptf1-Cre x LSL-Kras^G12D (KC) model of neoplastic disease, as well as KC mice with heterozygous deletion of Tgfbr1^+/− (KCT; Fig. 5A). Consistent with our previous report (18), these two models had no observable difference in tumor development at 6 months of age (Fig. 5B, N = 6/group). While KCT mice had increased expression of PD-L1 in the neoplastic epithelium (Supplementary Fig. S6A), they had no significant difference in MHC Class I or PD-L2 expression compared with age-matched KC controls (Fig. 5B). Similarly, there was no significant difference between KC and KCT mice with respect to tumor-specific localization of the cytotoxic surrogate granzyme B (Fig. 5C).

Mice were then allowed to develop extensive PanIN disease for approximately 3 months, at which time, KC and KCT mice were administered either a PBS vehicle, or a twice-weekly 200 μg injection of the PD-1 neutralizing antibody RMP1-14 (anti-PD-1) in combination with 100 mg/kg gemcitabine. After four months on treatment, animals were sacrificed and tissues examined (Fig. 5D). At the conclusion of the study, vehicle treated KC and KCT mice developed an enlarged, firm, nodular pancreas, as did KC mice administered gemcitabine + anti-PD-1 (N = 3). Contrastingly, 6/7 gemcitabine + anti-PD-1 treated KCT mice had a smaller, malleable pancreas resembling that of a wild type animal (Fig. 5E, N = 7). On histologic evaluation, gemcitabine + anti-PD-1 failed to alter disease course in KC mice, which had no observable change in normalized pancreas weight, lesion frequency, or fibrosis. However, KCT mice administered gemcitabine + anti-PD-1 had significant reductions in PanIN incidence, fibrosis, and the mass of the pancreas (Fig. 5F and G).

Despite the changes in phenotype, both drug treated KC and KCT mice displayed increased expression of PD-L1, PD-L2, and MHC Class 1 on the neoplastic epithelium, (Fig. 5H and I). Expectedly, KCT mice had substantial reductions in the phosphorylation of SMAD2, affirming the disruption of TGFβ signaling (Fig. 5H and I). These observations were paralleled by changes in cell-mediated immunity, with gemcitabine + anti-PD-1 treated KCT mice displaying increased tumor infiltrating CD3⁺ and CD8⁺ lymphocytes, as well as granzyme B deposition in remaining neoplastic tissues. This was paralleled by increased activation of the granzyme B effector and apoptotic surrogate caspase-3 (Fig. 5J; Supplementary Fig. S6B). We then confirmed the apparent increase in CTL activation by flow cytometry of regional lymph nodes, which showed an increase in CD8⁺perforin⁺ and CD8⁺CD69⁺ cells in drug-treated KCT mice (Fig. 5K).

While the combination of gemcitabine and anti-PD-1 had significant efficacy in early PanIN disease with genetic ablation of TGFβ signals, the translational relevance of these findings are limited, as they do not address the individual contributions of each drug. Furthermore, patients seldom present with such early-stage disease, and genetic suppression of TGFβ signals does not fully recapitulate the effects of a TGFβ pathway inhibitor. We therefore used the KPC model of advanced PDAC, and randomized 41 mice into one of seven treatment arms. To allow adequate time for gemcitabine-induced remodeling of the TME, mice were again enrolled at 90 days (roughly 12.85 weeks) of age, at which point, they develop extensive high-grade PanIN disease and rare, focal areas PDAC. At this time, mice were then administered either a saline vehicle every other day (N = 7), 100 mg/kg gemcitabine twice per week (N = 7), 75 mg/kg of the TGFβ signaling inhibitor galunisertib every other day (N = 4), staggered doses of gemcitabine and galunisertib (N = 5), a fixed 200 μg dose of anti-PD-1 twice-weekly (N = 4), staggered doses of galunisertib and anti-PD-1, or twice-weekly gemcitabine starting at 90 days (N = 6), with the addition of galunisertib and anti-PD-1 two weeks later (N = 8). Tissues were collected either when the animals were moribund or when mice reached 8 months of age (Fig. 6A).

Consistent with our previous results (18), neither galunisertib nor anti-PD-1 monotherapy significantly altered disease course. However, the combination of galunisertib and anti-PD-1 led to a modest reduction in tumor burden with limited fibrosis in remaining neoplastic tissues. The addition of gemcitabine further enhanced the efficacy of the combined immunotherapy, leading to improved preservation of normal gland architecture, added reductions in PanIN frequency and fibrosis, as well as comparable reductions in SMAD2 phosphorylation, mitoses, and intratumoral TGFβ1 (Fig. 6B–D; Supplementary Fig. S6C–S6F). In addition, 7 of 8 mice receiving all three drugs showed no mortality or obvious signs of disease-associated morbidity at the conclusion of the study, and tissues were collected from healthy mice after 150 days on treatment. One mouse did not meet the study endpoint, and was sacrificed 103 days postenrollment. Contrastingly, significant mortality was observed in all other groups, although this was modestly delayed in galunisertib/gemcitabine-treated mice, and more significantly delayed in galunisertib/anti-PD-1–treated mice (Fig. 6B).

While areas of residual disease were scarce compared with other groups, lesions in mice administered all three drugs had uniform, overwhelming lymphocytosis exceeding mice given galunisertib and anti-PD-1 (Fig. 6C and D; Supplementary Fig. S6D). In addition to increased expression of MHC Class 1 and PD-L1, these tumor-infiltrating lymphocytes were uniformly positive for the T-cell marker CD3 (Fig. 6E and F), negative for regulatory T-cell (Treg) surrogate FoxP3, and largely positive for CD8, contrasting the sparse but FoxP3-dominant T-cell infiltrate of control mice (Fig. 6E and F). These findings were paralleled by increased tumor-specific staining for granzyme B as well as its downstream target cleaved caspase-3, affirming the increased apoptosis in remaining neoplastic areas (Fig. 6E and F).

In addition to the reduction in lesion incidence, gemcitabine/galunisertib/anti-PD-1–treated mice had substantial changes with respect to the vasculature in remaining areas of disease. Consistent with previous reports, vehicle-treated KPC mice had severe fibrosis and clear vascular dysfunction (36). Namely, we observed scant arteries/arterioles within the tumor as well as diffuse capillaries and small, compressed veins (Supplementary Fig. S6G). Contrastingly, gemcitabine/galunisertib/anti-PD-1–treated mice had an increased frequency of large vessels, including patent arteries and distended veins, as well as increased capillarity density (Supplementary Fig. S6G). These changes were confirmed by dual staining for vascular marker CD31 and αSMA, which allowed for more accurate quantification and differentiation of veins and arteries/arterioles. In addition to the increased frequency of intratumoral arteries, gemcitabine/galunisertib/anti-PD-1–treated mice had an increased mean arteriolar diameter. Similar changes were observed in the venous system, which were also more frequent and more distended in triple treated mice, with similar results in intratumoral capillary density (Supplementary Fig. S6G).

To better evaluate the restoration of cytotoxic immunity in gemcitabine/galunisertib/anti-PD-1–treated mice, we next repeated our in vivo experiment and examined the tumor-infiltrating lymphocytes by flow cytometry. The pancreata and spleens from vehicle and gemcitabine/galunisertib/anti-PD-1–treated mice (N = 3/group) were collected at their respective study endpoints and stained for CD4 and CD8. Consistent with our prior histopathology, both CD4⁺ and CD8⁺ T cells were scarce in the pancreata of vehicle-treated mice, whereas those treated with the combination of gemcitabine/galunisertib/anti-PD-1 had highly significant increases in both T-cell subsets (Fig. 7A and B). However, gemcitabine/galunisertib/anti-PD-1–treated mice only displayed modest increases in CD4⁺ and CD8⁺ populations in the spleen, suggestive of a largely tumor-specific phenotype (Fig. 7C–F).

As the cytotoxic effects of immune checkpoint inhibitors are largely dependent on the activation of CD8⁺ T cells, we next gated to CD8⁺ cells evaluated intracellular expression of three distinct activation markers: granzyme B, perforin, and IFNγ. Although CD8⁺ T cells were again scarce in the pancreas of vehicle-treated mice, those that were present remained refractory from full activation with little to no expression of perforin and IFNγ, and moderate expression of granzyme B. Contrastingly, approximately 30% of the more robust CD8⁺ infiltrate of gemcitabine/galunisertib/anti-PD-1–treated mice were positive for all three cytokines, a phenomenon that again was exclusive to intratumoral lymphocytes and not observed in the spleen (Fig. 7G and H). CD8⁺ cells were next evaluated for the simultaneous expression of multiple activation markers, which again confirmed the presence of fully activated CD8⁺ cells exclusively in the pancreas of gemcitabine/galunisertib/anti-PD-1–treated mice (Fig. 7I and J; Supplementary Fig. S7A–S7D).

Cytotoxic chemotherapy has been the backbone of the clinical management of PDAC for several decades. Despite the emergence of new multidrug regimens such as gemcitabine/nab-paclitaxel or FOLFIRINOX, 5-year survival remains a dismal 9% with nearly all patients eventually progressing on treatment (37). As discussed, there are no standard-of-care medications that can currently be offered to patients who progress on chemotherapy. Here, we sought to address this issue by exploring the biologic consequences of long-term gemcitabine treatment within the pancreatic TME to identify potential treatment strategies in the second- or third-line setting. Through these experiments, we determined that long-term gemcitabine leads to extensive reprogramming of the tumor microenvironment, particularly with respect to tumor immunogenicity.

Our results suggest that gemcitabine enhances the expression of antigen-presenting molecules and inflammatory cytokines/chemokines, as well as negative immune checkpoints PD-L1 and PD-L2. This is consistent with prior studies that hint at a potential interplay between chemotherapy and the presentation of self-peptide (38, 39). These collective observations suggest that chemotherapy may prime pancreatic cancers for ICIs, particularly as gemcitabine has been shown to enhance the cross-presentation of tumor antigens (40). However, ICIs have yet to show substantial efficacy in clinical trials (Supplementary Table S3). Particularly, the combination of gemcitabine/nab-paclitaxel and pembrolizumab provided only a modest benefit compared with gemcitabine/nab-paclitaxel in chemotherapy-naïve PDAC patients (30). This was recapitulated by our in vivo experiments, in which the combination of gemcitabine and anti-PD-1 had a marginal effect on tumor-bearing mice. Interestingly, these observations were contrasted by experiments using immune-competent xenografts. There, gemcitabine and anti-PD-1 led to complete responses and improved survival (41). However, a likely explanation of this discrepancy is that xenograft models generally lack the desmoplastic stroma inherent to genetically modified mice (42). This may be of particular relevance, as the stroma is itself a source of several cytokines including TGFβ (12, 43), and select stromal cells express negative immune checkpoints including PD-L1 (44).

As demonstrated in our earlier works, TGFβ signals are largely produced by the pancreatic cancer stroma, and impede the functional activation of tumor-infiltrating lymphocytes (12). In this study, we determined that prolonged administration of gemcitabine enhances tumor cell secretion of TGFβ1, thereby conferring drug-resistant phenotypes to neighboring stromal cells and further enhancing the production of inflammatory cytokines/chemokines. On the basis of these observations, we challenged models of murine tumorigenesis with gemcitabine with anti-PD-1, first in the setting of genetic ablation of Tgfbr1, followed by pharmacologic inhibition of TGFβ signaling. In the setting of TGFβ signaling inhibition, gemcitabine and anti-PD-1 led to near uniform drug responses, CTL-mediated regression of disease, and improved survival.

While encouraging, it is important to note that progress for immunotherapy in PDAC has been difficult, both with ICIs (45) and vaccine-based immunotherapy (46–48). Similarly, combination immunotherapy has also shown less than promising results (25). For instance, the bispecific PD-L1/TGFβ antibody M7824 only produced a partial response in one microsatellite instability-high (MSI-H) patient, with no benefit observed in the other four enrolled (29). Interestingly, in the setting of MSI-H, patients generally have a high mutational burden, and following ICIs, responders undergo rapid expansion of neoantigen-specific T-cell clones (16). This led to the approval of pembrolizumab for MSI-H-PDAC in 2017. Similarly, despite early mortality, murine tumor models with homozygous TP53 loss (KPPC) display increased MHC class I expression and robust T-cell infiltration (Supplementary Fig. S8A–S8C). However, although high TMB is associated with poor clinical outcomes (Supplementary Fig. S8D and S8E), MSI-H is rare in PDAC and its predictive value for ICIs is not clear (16, 49–51).

As highlighted in this article, the immunosuppressive TME, particularly TGFβ signaling, may be an important consideration for pancreatic cancer immunotherapy. We have previously demonstrated that suppression of TGFβ signaling augments PD-1 inhibition in murine PDAC (18). In addition, galunisertib is showing clear efficacy in early clinical trial, particularly in combination with gemcitabine (52). Similarly, the addition of TGFβ signaling inhibition and ICIs is showing significant early efficacy in several cancers (53–57). In PDAC, the combination of galunisertib and durvalumab has shown early promise in a phase Ib trial (NCT02734160). While these results are no doubt encouraging, more work is needed to explore TGFβ as a potential immune checkpoint in PDAC, particularly in combination with anti-PD-1 in gemcitabine-refractory disease.

Our data also appear to substantiate the long-standing hypothesis that the relatively scant presentation of self-antigen limits the therapeutic efficacy of ICIs in PDAC. Hence, the use of cytotoxic chemotherapy followed by combined TGFβ inhibitor and anti-PD-1 had marked efficacy in vivo, cooperating to produce durable immune responses in murine PDAC. The combination of chemotherapy and ICIs has been highly effective in many solid tumor malignancies including non–small cell lung cancer (58–60) and triple-negative breast cancer (61, 62). In PDAC, however, the combination of gemcitabine/nab-paclitaxel and pembrolizumab provided only a modest benefit compared with gemcitabine/nab-paclitaxel alone (30). Similar results were observed using the combination of nivolumab with nab-paclitaxel, or nivolumab with GVAX-based immunotherapy (63). However, as none of these strategies contend with the immunosuppressive TME precipitated by long-term chemotherapy, the addition of galunisertib or alternate suppressive cytokine inhibitor may be a highly effective means of enhancing the efficacy of either chemoimmunotherapy in the first line, or ICIs and palliative gemcitabine in the third line.

Finally, although our data supports this potential approach, there are several added factors that must be considered when translating these findings to the bedside. While we showed that single-agent gemcitabine appears to prime the tumor microenvironment for immunotherapy, it remains to be seen whether the same will be true of other cytotoxic regimens such as FOLFIRONOX. In addition, both gemcitabine/nab-paclitaxel and FOLFIRINOX regimens are associated with grade III hematologic adverse effects (6, 7), and severe neutropenia has also been observed with gemcitabine monotherapy (64). The resultant paucity of leukocytes may limit the efficacy of immune checkpoint inhibition, thus dose adjustments may be required. However, despite these potential hurdles, our data suggest that gemcitabine primes the TME for immune evasion, providing an opportunity for intervention in the third-line setting. Given the lack of options at this stage, the combination of anti-PD-1, galunisertib, and maintenance gemcitabine warrants consideration in patients who begin to progress on cytotoxic chemotherapy.

S.D. Kamath is a consultant at OncLive. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D.R. Principe, H.G. Munshi, A. Rana

Development of methodology: D.R. Principe, H.G. Munshi, A. Rana

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.R. Principe, M. Narbutis, S. Kumar, A. Park, N. Viswakarma, M.J. Dorman, P.J. Grippo, M.L. Fishel, R.F. Hwang, D. Thummuri, P.W. Underwood, H.G. Munshi, J.G. Trevino, A. Rana

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.R. Principe, M. Narbutis, S. Kumar, A. Park, N. Viswakarma, M.J. Dorman, A. Rana

Writing, review, and/or revision of the manuscript: D.R. Principe, S.D. Kamath, R.F. Hwang, P.W. Underwood, H.G. Munshi, J.G. Trevino, A. Rana

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.R. Principe, N. Viswakarma, M.J. Dorman, H.G. Munshi, A. Rana

Study supervision: D.R. Principe, A. Rana

This work is dedicated to the memory of our friend Tami Bichrt Robson who recently passed away after a long and courageous fight with breast cancer. The authors would like to thank our friend and mentor Dr. Larry Tobacman for his guidance, leadership, and dedication to the Medical Scientist Training Program at the University of Illinois College of Medicine, and we wish him well in his retirement. We would also like to thank Dr. Enrico Benedetti and the Department of Surgery at the University of Illinois at Chicago. This work was supported by Veterans Affairs Merit Award I01BX002703 and Career Scientist Award IK6 BX004855 to A. Rana, by NIH F30CA236031 to D.R Principe, and by NIH R01CA217907 and Veterans Affairs Merit Award I01BX002922 to H.G. Munshi.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.