Role of TGF-β in pancreatic ductal adenocarcinoma progression and PD-L1 expression

Abstract

Purpose

The transforming growth factor-beta (TGF-β) pathway plays a paradoxical, context-dependent role in pancreatic ductal adenocarcinoma (PDAC): a tumor-suppressive role in non-metastatic PDAC and a tumor-promotive role in metastatic PDAC. We hypothesize that non-SMAD-TGF-β signaling induces PDAC progression.

Methods

We investigated the expression of non-SMAD-TGF-β signaling proteins (pMAPK14, PD-L1, pAkt and c-Myc) in patient-derived tissues, cell lines and an immunocompetent mouse model. Experimental models were complemented by comparing the signaling proteins in PDAC specimens from patients with various survival intervals. We manipulated models with TGF-β, gemcitabine (DNA synthesis inhibitor), galunisertib (TGF-β receptor inhibitor) and MK-2206 (Akt inhibitor) to investigate their effects on NF-κB, β-catenin, c-Myc and PD-L1 expression. PD-L1 expression was also investigated in cancer cells and tumor associated macrophages (TAMs) in a mouse model.

Results

We found that tumors from patients with aggressive PDAC had higher levels of the non-SMAD-TGF-β signaling proteins pMAPK14, PD-L1, pAkt and c-Myc. In PDAC cells with high baseline β-catenin expression, TGF-β increased β-catenin expression while gemcitabine increased PD-L1 expression. Gemcitabine plus galunisertib decreased c-Myc and NF-κB expression, but induced PD-L1 expression in some cancer models. In mice, gemcitabine plus galunisertib treatment decreased metastases (p = 0.018), whereas galunisertib increased PD-L1 expression (p < 0.0001). In the mice, liver metastases contained more TAMs compared to the primary pancreatic tumors (p = 0.001), and TGF-β increased TAM PD-L1 expression (p < 0.05).

Conclusions

In PDAC, the non-SMAD-TGF-β signaling pathway leads to more aggressive phenotypes, TAM-induced immunosuppression and PD-L1 expression. The divergent effects of TGF-β ligand versus receptor inhibition in tumor cells versus TAMs may explain the TGF-β paradox. Further evaluation of each mechanism is expected to lead to the development of targeted therapies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13402-021-00594-0.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a malignancy of which treatment continues to present significant challenges for patients and clinicians alike. It presents as metastatic disease in over 50% of patients [1]. The American Cancer Society forecasts that PDAC will likely become the second leading cause of cancer-related deaths in the next few years. With few advances in the clinic, the death toll from PDAC is expected to rise further [1, 2]. While the vast majority of patients has a dismal survival (characterized as ‘progressive disease’ herein), there are patients with longer-term survival (defined as ‘indolent disease’), a small minority of whom appear to be cured when tumors induce an immunogenic response leading to long-term tumor control [3, 4].

While the genomic landscape of PDAC is diverse, the transforming growth factor-beta (TGF-β) pathway is a critically important component [2, 5–8]. In PDAC, similar to other desmoplastic cancers, TGF-β is ubiquitously expressed and plays a context-dependent, yet paradoxical role, i.e., it is tumor suppressive in early, non-metastatic PDAC and tumor promotive in more advanced, metastatic PDAC [9]. The exact mechanism underlying this paradox is unknown, but general consensus is that TGF-β-mediated epithelial to mesenchymal transition (EMT) represents a critical phenotypic switch that impacts tumor progression, metastasis and prognosis. This switch may be related to the balance between SMAD and non-SMAD tumor signals. The tumor suppressive role of TGF-β is thought to be controlled through the SMAD family of proteins [10–12]. Although the specific mechanism of TGF-β-mediated tumor promotion is unclear, it may act via non-SMAD signaling proteins such as β-catenin [7, 13, 14], a molecule with diverse functions during normal and cancer development [15]. Non-SMAD signaling through the TGF-β receptor may be pathogenic via aberrant regulation of p38 MAPK (MAPK14), ERK and others [16–18]. Important effector molecules regulated by β-catenin and other non-SMAD intracellular signals include Programmed Death-Ligand 1 (PD-L1) and cellular Myc (c-Myc) [19], both of which are associated with cancer progression and a poor prognosis. It has been reported that ligand-independent TGF-β receptor signaling may occur primarily via non-SMAD-mediated pathways [18], leading to p38 MAPK and Akt phosphorylation [16, 20–22]. Recent work by Batlle et al. has shown that inhibition of the TGF-β receptor with galunisertib does not completely inhibit non-SMAD intracellular signaling due to p38 MAPK modulation [23].

It has been reported that PDAC tumor associated macrophages (TAMs) can induce inflammation and immunosuppression through multiple mechanisms [24–27]. In PDAC, TGF-β polarizes macrophages to the M2 phenotype leading to pro-tumor activity and the release of more TGF-β via a positive feedback loop [28, 29]. TGF-β has various effects on immune cells in the tumor microenvironment (TME) [30–32]. In other cancers, β-catenin has been found to play a critical role in polarizing macrophages to TAMs, resulting in EMT and tumor progression [25, 33, 34]. Mechanistically, it remains to be established whether in PDAC TAMs lead to β-catenin expression through TGF-β or some other cytokine.

Here, we investigated the relationship between TGF-β, the β-catenin complex and downstream effector molecules such as PD-L1 and c-Myc in PDAC tumors and TAMs in order to understand their relationship, since it has been important in other contexts, such as inducing cancer stem cell characteristics and expressing PD-L1 in melanoma [35–38]. Our hypothesis is that non-SMAD TGF-β signaling induces metastatic PDAC through activation of downstream effector molecules (i.e., β-catenin). To test this hypothesis, we investigated the non-SMAD TGF-β signaling pathway in patient-derived tissues and various in vitro and in vivo PDAC models in order to understand the effects of combined galunisertib (a TGF-β inhibitor) and gemcitabine (clinically approved chemotherapeutic agent) treatment on disease progression. We report a novel axis in which TGF-β receptor inhibition leads to increased PD-L1 and c-Myc expression via β-catenin (Fig. 1).

Fig. 1.

This novel axis describes TGF- β receptor inhibition leading to c-Myc and PD-L1 expression

Materials and methods

Reagents, antibodies and treatments

Gemcitabine (Gem, Sigma, Cat. #G6423) is a chemotherapeutic agent used to treat patients with PDAC. Galunisertib (Gal, LY2157299; Selleckchem, Cat. #S2230) is a small molecule inhibitor directed against the TGF-β receptor type I. MK-2206 (Selleckchem, Cat. #S1078) is a small molecule inhibitor preventing phosphorylation and activation of Akt; it inhibits the phosphorylation of Thr308 and Ser 473 of Akt [39]. Human recombinant TGF-β1 (Biolegend, Cat. #580702) was used for cell treatment as listed. Cells were treated with gemcitabine (100 nM), galunisertib (10 μM), MK-2206 (20 μM), human TGF‐β1 (10 ng/ml), or combinations thereof in complete cell culture media. All the primary and secondary antibodies used for Western blot analyses were obtained from Cell Signaling Technology.

Cell lines and mice

Human pancreatic cancer cell lines derived from PDAC tumors, Panc-1 (ATCC CRL 1469) and Capan-2 (ATCC HTB-80), human pancreatic cancer cell lines derived from metastatic PDAC tumors, Capan-1 (ATCC HTB 79) and AsPC-1 (ATCC CRL-1682), and a human macrophage cell line (ATCC, CRL 9855) were purchased from the ATCC. All cell lines were authenticated by an outside company twice yearly. Panc-1 and AsPC-1 cells are wild type for SMAD4, while Capan-1 is mutant for SMAD4. Capan-2 cells are wild type for SMAD4, but express this protein at low a level [40].

The macrophages were polarized to TAMs using 20 ng/ml IL-4 for 24 h and mixed with 50% PDAC conditioned media. In vitro studies were carried out using the respective cell lines. The cell lines were cultured and maintained in complete media supplemented with 10% fetal bovine serum (FBS) and penicillin-streptomycin at 37 °C with 5% CO₂ as recommended by the supplier (ATCC, Manassas, VA, USA). Equivolume PBS, gemcitabine (100 nM), TGFβ1 (10 ng/ml) or galunisertib (10 μM) were utilized as indicated.

For the in vivo mouse experiments, we orthotopically implanted murine PDAC KPC cells into immunocompetent C57BL/6 J mice (B6 mice, The Jackson Laboratory, Bar Harbor, ME, USA; 5 mice per treatment group). These cells were derived from tumors spontaneously forming in the pancreas of KPC (KrasLSL.G12D/+; p53R172H/+; PdxCretg/+) mice [41] and subsequently stably transfected with luciferase. For the in vivo experiments, the cells were treated in vitro for 24 h and subsequently orthotopically implanted (10⁶ cells/50 μl; 1:1, v/v with Matrigel with the assigned treatment embedded). Next, the mice were treated as per assigned group three times per week, i.e., gemcitabine 100 mg/kg intraperitoneally (i.p.), and TGF-β1 0.05 mg/kg and galunisertib 25 mg/kg by oral gavage. Luciferase activity was measured at 15 days post-implantation and tumors were harvested at 28 days post implantation (IVIS Lumina XRMS Series III, Perkin-Elmer). Since the tumors were implanted in the tail of the pancreas (anatomic left upper abdominal quadrant of the animal), we considered any isolated luciferase activity on the right side or lower part of the abdomen to be metastatic disease. We evaluated for overall relative luminescent flux (luminescence, photons/s) as well as focal activity in the anatomic right upper quadrant (liver metastases) or lower abdomen (pelvic metastases). Metastases were confirmed at necropsy.

Separately, we compared the effects of gemcitabine treatment to galunisertib treatment on the percentage of TAMs, tumoral PD-L1 expression and PD-L1 expression on TAMs isolated from pancreatic and metastatic (liver) tumors using the same model (KPC cells orthotopically implanted into B6 mice as above) with KPC cells being treated in vitro for 24 h prior to orthotopic implantation. The treatment drug was also embedded in Matrigel at the above in vitro concentration. Methods for fluorescent immunohistochemistry and flow cytometry are detailed below. Depending on the sizes of the tumors, between 20 and 80 representative images were taken of each tumor for analysis.

SDS-PAGE and Western blotting

For the in vitro studies, we used Western blotting to study protein expression. Briefly, the media were removed, and adherent cells were collected using trypsin/EDTA. Next, the cells were washed once with PBS and lysed using Pierce RIPA buffer (Thermo Scientific, Cat. #89901) containing 1x Pierce Protease and Phosphatase inhibitors (Thermo Scientific, Cat. #88668)). Subsequently, they were stored at −20 °C or immediately processed for protein isolation. To this end, the cell lysates were pelleted at 4 °C/1500 RPM for 15 min, after which the supernatants were transferred to fresh Eppendorf tubes and the protein concentrations were quantitated using Quick Start Bradford Dyed Reagent (Biorad, Cat. #500–0205). Seventy-five micrograms of each protein sample were analyzed by SDS-PAGE (10% gel) followed by Western blotting and autoradiography. Precision Plus Dual Color (Biorad, Cat. #161–0374) was used as protein standard marker. Clarity Western ECL Substrate (Biorad, Cat. #170–5060) was used to develop the probe signal, and Classic Blue BX film (MidSci, Ref. 6,045,983) was used for autoradiography.

Patient selection, specimens and fluorescent immunohistochemistry

Clinical data and specimens were obtained from the electronic medical record (Methodist Hospital System, Memphis, TN, USA) and the cancer center biorepository after institutional review board approval. Consecutive patients were identified based on survival (“progressive” patients survived less than 6 months while “indolent” patients survived longer than 6 months), tumor characteristics (stage of disease and high-risk features) and demographic data (age, gender, ethnicity). Formalin-fixed, paraffin-embedded (FFPE) tissue blocks were obtained from the institutional biorepository (University of Tennessee Health Science Center, UTHSC). All primary antibodies used for fluorescent immunohistochemistry (IHC) assays were obtained from Cell Signaling Technology, except c-Myc (ThermoFisher Cat. #13–2500), CD274 / B7-H1 / PD-L1 (LSBio Cat. #LS-C338364–100), EMA (MUC1, Origene Cat. #TA800938) and S100A4 antibodies (Abcam Cat. #ab41532), and CD68 monoclonal antibody (KP1, Invitrogen Cat. #14–0688-82). Secondary antibodies (all from ThermoFisher) used in these assays included goat anti-mouse IgG1 conjugated to AlexaFluor488 (A-21121), goat anti-mouse IgG2a conjugated to AlexaFluor647 (A-21241), goat anti-mouse IgG2b conjugated to AlexaFluor488 (A-21141), and goat anti-rabbit IgG conjugated to AlexaFluor555 (A-21429). The mounting medium used was Prolong Diamond mounting solution (P36961, Invitrogen), and DAPI (D21490, Invitrogen) was used to stain the nuclei. Slides were imaged on a Zeiss 710 LSM at 200x magnification and fluorescent intensities quantified using ImageJ software. Two representative images per patient sample were acquired and used for analysis.

Flow cytometry

Panc-1 and Capan-1 cells were treated with gemcitabine (100 nM), galunisertib (TGF-β receptor 1 kinase inhibitor, 10 μM) and MK-2206 (pan-Akt phosphorylation inhibitor) for 24 h. Single cell suspensions were obtained using gentle dissociation with trypsin/EDTA. Cells were stained according to the manufacturer’s recommendations (BD Biosciences) for Annexin V (apoptosis) and 7-AAD (cell death/necrosis). Cells were analyzed using a FACSAria system (BD Biosciences).

Tumors were harvested, and single cell suspensions were obtained for TAM identification and PD-L1 expression analysis. Briefly, pancreatic and metastatic liver tumors were excised and processed using a tumor dissociation kit (Miltenyi Biotec cat#130–096-730) to obtain single cell suspensions. Next, the cells were stained with the following antibodies: anti-Hu/Ms., CD11b-RedFluor 710 (TONBO, Cat # 80–0112-U025), anti-mouse Ly-6C-APC (Biolegend cat #128015), anti-mouse CD45-VF450 (TONBO Cat #75–0451-U025), anti-mouse CD274(PD-L1)-BV711 (Biolegend Cat #124319), anti-mouse F4/80-PE (TONBO Cat #50–4801-U025), anti-mouse Ly-6G-PerCP-CY5.5 (TONBO Cat #65–1276-U025), anti-mouse CD279(PD1)-BV421 (Biolegend Cat #109121), anti-mouse CD11c-BV650 (Biolegend Cat #117339), anti-Hu/Ms. Arginase1-PE-Cy7 (Invitrogen Cat #25–3697-80) and anti-mouse NOS2-APC-eFulor780 (Invitrogen Cat #47–5920-80). Ghost dye was used as live/dead stain. The cells were acquired using a ZE5™ Cell Analyzer (Propel Labs) and data were analyzed using FlowJo. TAMs were gated as CD45+, CD11b+, F4/80+, Ly6C lo and Ly6G lo and further analyzed for PD-L1 expression.

Gene expression analysis

Quantitative real time PCR (qRT-PCR) was used to investigate Snail and c-Myc expression in Panc-1 cells after treatment with recombinant TGF-β or in TAMs (ATCC, CRL 9855) prepared as described above. Isolation of total RNA was performed using TRIzol (Invitrogen) according to the manufacturer’s protocol. This RNA was treated with DNA-free DNA Removal Kit (Ambion) and converted to cDNA using a Maxima First Strand cDNA Synthesis Kit (Fermentas) according to the manufacturer’s protocol. The cDNA obtained was subjected to quantitative PCR (qPCR) using a StepOnePlus system (Applied Biosystems). qPCR was conducted in a final volume of 20 μl using a Maxima SYBR Green/ROX qPCR Master Mix (Fermentas) according to the manufacturer’s protocol. Amplification conditions were: 95 °C (5 s), 40 cycles of 95 °C (30 s), 55 °C (60 s) and 72 °C (60 s). Primer sequences for hSnail (5’-ACTGCAACAAGGAATACCTCAGCC-3′) and hc-Myc (5’-TCT CCACACATCAGCACAACTACG-3′) were determined relative to H18S ribosomal RNA (5’-CCGCGGTTCTATTTTGTTGGT-3′). All experiments were conducted using both three biological and three technical triplicates.

Statistical analysis

Statistical analyses were performed using STATA 14 (STATA Corp, College Station, TX, USA). Results are reported as mean ± standard deviation unless otherwise reported. Means were compared using a two-sided Student’s t test or Analysis of Variation (ANOVA) as appropriate with α = 0.05 for statistical significance. Frequencies were evaluated for statistical significance using a χ² test.

Results

Clinical PDAC samples and clinicopathological data

PDAC tumors with adequate quality from 39 patients from the UTHSC tumor biorepository were identified and matched for similar age, sex, ethnicity, body mass index and stage of disease (Table 1). Of all patients included, 46% was African American and 54% was Caucasian. Most tumors were well differentiated, and all tumors came from patients who underwent pancreatic resection with curative intent. 10 of 39 patients died from PDAC in less than 6 months, representing the progressive disease group. Of these 10 patients, one was identified with metastatic PDAC at the time of resection. 29 patients were found to have a longer survival, termed indolent disease, with a mean survival of 31 ± 23 months. Neoadjuvant therapy was not used in the treatment of these patients.

Table 1.

Clinicopathologic characteristics of the patients

Clinical feature		Indolent Disease	Progressive Disease	P value
		n = 29	n = 10
Age, mean ± SD		64.6 ± 10.4	68.8 ± 7.5	0.27
Sex, male n (%)		15 (52%)	4 (40%)	0.52
Ethnicity	African American n (%)	13 (45%)	5 (50%)	0.78
	Caucasian n (%)	16 (55%)	5 (50%)
BMI, mean ± SD		27.5 ± 5.2	30.5 ± 7.0	0.16
AJCC Stage	I, n (%)	6 (21%)	1 (10%)	0.34
	II, n (%)	20 (69%)	7 (70%)
	III, n (%)	3 (10%)	1 (10%)
	IV, n (%)	0 (0%)	1 (10%)
Survival, mean ± SD days		923 ± 704	84 ± 49	0.0006

SD: standard deviation of the mean

Phospho-p38MAPK and PD-L1 expression levels are higher in tumors from patients with progressive disease

In order to understand the initial steps in non-SMAD TGF-β signaling, we investigated p38 MAPK expression in primary PDACs. Tumor samples from 39 PDAC patients (Table 1) were identified from the UTHSC biorepository. Because the TGF-β receptor phosphorylates p38 MAPK, an early and critical component in non-SMAD TGF-β signaling, we investigated phosphor-p38 MAPK expression in the patient-derived tumor samples (Fig. 2a). We further sought to understand differences in PD-L1 expression in tumors from our patient cohort. We found that PDAC tumors from patients with progressive disease had an over 2-fold higher PD-L1 protein expression level compared to those from patients with indolent disease (p = 0.006, Fig. 2b). Tumors from patients with progressive disease had a 3-fold higher phospho-p38 MAPK protein expression level compared to those from patients with indolent disease (p = 0.01, Fig. 2c). Importantly, in these patient samples the PD-L1 and phospho-p38 MAPK protein expression levels were well correlated on a tumor by tumor basis (r = 0.48, p = 0.0019, Fig. 2d). This PD-L1 - phospho-p38 MAPK correlation was higher in the progressive disease group compared to the indolent group (p < 0.01).

Fig. 2.

Tumors from patients with progressive PDAC exhibit greater phospho-p38MAPK and PD-L1 levels. (a) Cell surface pancreatic cancer marker MUC1 (red) expression shown with PD-L1 (green) and p38 MAPK (blue) expression along with nuclear staining (DAPI, white) in PDAC tumors. (b) PD-L1 and (c) phospho-p38 MAPK proteins quantified in indolent (n = 29) and progressive (n = 10) PDAC samples. (d) Correlation of PD-L1 and phosphor-p38 MAPK in tissue samples from PDAC patients (Pearson r = 0.48, p = 0.0019). Scale Bars in (a) represent 50 μm. Representative images are shown

Phosphorylated Akt (pAkt) increases active β-catenin due to increased TGF-β expression in PDAC

Based on the above findings in patient samples, we next set out to investigate protein expression in 5 human PDAC cell lines, i.e., non-metastatic PDAC cell lines (Panc-1 and Capan-2) and metastatic PDAC cell lines (Capan-1 and AsPC-1) to interrogate whether gemcitabine, galunisertib or MK-2206 (an Akt inhibitor) affect the expression of key proteins in tumor progression. In addition, a patient-derived cell line from a 57 year old male with PDAC was tested. We first investigated pAkt, as it is a known target of active p38 MAPK, and both are known to exhibit critical interactions with the β-catenin complex protein GSK-3β [42–45]. Two major non-SMAD TGF-β signaling pathways, i.e., the NF-κB and β-catenin pathways, are known to be activated via pAkt [9, 10, 20, 46]. We found that MK-2206 potently inhibited TGF-β induced pAkt expression in both Panc-1 and Capan-1 cells (Fig. 3a and b), with similar expression levels in cells with a high pAkt expression at baseline (Supplemental Fig. 1). Next, we sought to better understand which treatment was predominantly responsible for PDAC cancer cell control (cytotoxicity). To this end, we investigated apoptosis and cell death using flow cytometry (Fig. 3c). After treating Panc-1 and Capan-1 cells individually with gemcitabine, galunisertib or MK-2206, we found that MK-2206 treatment led to the lowest viabilities in both Panc-1 and Capan-1 cells compared to controls (13% and 68%, respectively). Panc-1 cells were more sensitive to treatment than Capan-1 cells (Fig. 3c). Since selective and differential regulation of NF-κB by β-catenin and Akt have been described in multiple settings [47–50], we next investigated whether NF-κB is regulated by pAkt to explain the differences in aggressive PDAC cell phenotypes. We found that the expression of NF-κB was dependent on Akt activation in Panc-1 cells, but was independent on Akt activation in Capan-1 cells (Fig. 3a and b), consistent with differential mechanisms of action of non-SMAD TGF-β signaling.

Fig. 3.

Modulation of non-SMAD TGF-β signaling impacts multiple proteins resulting in cancer cell proliferation. Panc-1 cells (a) and Capan-1 cells (b) were treated with gemcitabine (Gem, 100 nM, DNA synthesis inhibitor), galunisertib (Gal, 10 μM, TGF-β type 1 receptor inhibitor) and MK-2206 (Akt inhibitor, 20 μM) for 24 h alone or in combination. Signaling pathways were examined by Western blotting for the indicated proteins. (c) Apoptosis and cell death assessment in both Panc-1 and Capan-1 cells by Annexin V and 7-amino-actinomycin D (7-AAD) staining using flow cytometry

Since Akt is critical in many cancer pathways and especially related to β-catenin and NF-κB expression [21, 51], we next compared pAkt protein expression in tumor samples derived from patients with progressive disease and those from patients with indolent disease. We found that tumors from progressive disease patients exhibited significantly higher pAkt expression levels based on immunofluorescence assays (p = 0.035, Fig. 4a). Since patients with progressive disease showed a higher pAkt expression (Fig. 4a) and chemotherapy increased pAkt expression in Capan-1 cells (Fig. 3b), we next investigated whether TGF-β increases pAkt expression in PDAC cells. We found that TGF-β increased pAkt expression in both cell lines tested (Capan-1 and Panc-1; Fig. 4b). Whereas TGF-β treatment increased NF-κB expression in Panc-1 and Capan-1 cells, MK-2206 only decreased NF-κB expression in Capan-1 and Capan-2 cells. This suggests that NF-κB regulation by TGF-β is brought about by two distinct mechanisms in PDAC cells, i.e., an Akt-mediated mechanism in metastatic/progressive PDAC cells and a non-Akt-mediated mechanism in non-metastatic/indolent PDAC cells.

Fig. 4.

Differences in Akt, NF-kB and β-catenin signaling distinguishes primary from metastatic PDAC. (a) pAKT expression levels were found to be lower in indolent (n = 29) compared to progressive (n = 10) PDAC tumors by immunofluorescence (p = 0.035). (b) PDAC (Panc-1) versus metastatic PDAC (Capan-1) cells were treated with vehicle (PBS), recombinant human TGF-β1 (10 ng/ml) and/or AKT inhibitor MK2206 (20 μM) for 24 h, after which phospho-Akt (pAkt) and NF-κB expression levels were measured. (c) Panc-1 and Capan-1 cells were treated with vehicle (PBS), recombinant human TGF-β, or TGF-β co-cultured with TAMs for 24 h. Active and total β-catenin were measured by Western blotting. β-tubulin served as loading control. Experiments were repeated 3 times. Representative blots and images are shown

In addition, β-catenin was investigated in Panc-1 and Capan-1 cells. In the metastatic cell line Capan-1, TGF-β induced total and active β-catenin with a slight contribution from TAM polarized macrophages (Fig. 4c). Panc-1 cells express low levels of β-catenin (Fig. 4c), likely reflecting its less invasive phenotype [52], since β-catenin expression has been reported to be associated with cancer invasiveness/aggressiveness [35]. Capan-1 cells express higher levels of β-catenin at baseline (Fig. 4c). We found that TGF-β increased total and active β-catenin expression levels in Capan-1 cells, but not in Panc-1 cells. We validated our results by co-treating these two cell lines with TAMs (Fig. 4c) in addition to TGF-β, which did not dramatically change the results, again suggesting two distinct responses to TGF-β in these cell lines. Finally, we found that gemcitabine increased active β-catenin expression in Capan-2 and AsPC-1 cells (Supplemental Fig. 1).

PD-L1 expression varies with inhibition of the TGF-β receptor and Akt in PDAC cells

Since the above data showed that progressive disease PDAC patients exhibited higher PD-L1 expression levels and since PD-L1 may be regulated by Akt, GSK-3β or the β-catenin complex, we next investigated whether PD-L1 expression can be manipulated using a standard and clinically used chemotherapeutic agent (gemcitabine), a TGF-β receptor inhibitor (galunisertib) or a pAkt inhibitor (MK-2206). We were interested in understanding the effect of Akt inhibition on PDAC cells based on previously reported data that pAkt may interfere with SMAD-mediated tumor suppression [53]. We found that inhibition of the TGF-β receptor by galunisertib still allowed for phosphorylation of p38 MAPK, a critical component of the SMAD-independent TGF-β signaling pathway, in Panc-1 and Capan-1 cells (Fig. 3), concordant with work of others showing similar results in non-cancer settings [18, 54]. We found that galunisertib alone decreased PD-L1 expression in Panc-1 cells and slightly increased PD-L1 expression in Capan-1 and Capan-2 cells (Fig. 3 and Supplemental Fig. 1). MK-2206 inhibited PD-L1 expression in Panc-1, Capan-1 and Capan-2 cells (Fig. 3a and b, Supplemental Fig. 1). In Capan-1 cells, a trend was noted across all treatment groups, i.e., increased active β-catenin expression was associated with increased PD-L1 expression in all treatment groups containing galunisertib, suggesting an important mechanism that may be at work in at least some PDAC patients (Fig. 3).

TGF-β induces c-Myc expression in PDAC cells

The proto-oncogene c-Myc has been reported to be regulated both by β-catenin-dependent and -independent mechanisms in various cancers and to serve as a poor prognostic indicator [55–58]. We found that a high c-Myc expression in PDAC patient tumor samples was associated with progressive, rather than indolent, disease (p = 0.007, Fig. 5a). High expression levels of active β-catenin and c-Myc were found to be highly correlated in the PDAC patient tumor samples as well (p = 0.0002, Fig. 5b). Since active β-catenin is not highly expressed in Panc-1 cells (Fig. 4c), we hypothesized that TGF-β treatment may increase c-Myc expression, likely through a β-catenin-independent signaling pathway, and may eventually lead to tumor progression. Interestingly, we found that c-Myc expression was increased in Panc-1 cells after treatment with TGF-β alone and even more so after co-treatment with TAMs (p < 0.001, Fig. 5c). The c-Myc protein was found to be uniformly expressed in Panc-1 cells treated with TAMs, TGF-β or both (Fig. 5d), but not in control cells, in agreement with our prior data. In Capan-1 cells, however, TGF-β alone induced c-Myc expression (Fig. 5c), potentially via the NF-κB pathway and regulated by pAkt, as previously demonstrated (Fig. 3), but TAMs did not induce the same effect.

Fig. 5.

TGF-β induces c-Myc expression. (a) c-Myc measured in indolent (n = 29) and progressive (n = 10) PDAC tumors by immunofluorescence (p = 0.007). (b) β-catenin and c-Myc protein expression levels are highly correlated (Pearson correlation r = 0.57, p = 0.0002). (c) qRT-PCR quantification of c-Myc expression in Panc-1 cells treated the same as in Fig. 4c. Data are shown as relative units of delta-delta Ct normalized to H18S ribosomal RNA. N = 3 experimental replicates shown, mean ± SEM. (d, e) Panc-1 and Capan-1 cells were treated the same as in Fig. 4c with the addition of co-culture with TAM (M0) alone. c-Myc expression was measured by Western blotting with β-tubulin as loading control

TGF-β induces PDAC progression through EMT

One mechanism for TGF-β-induced PDAC tumor progression is through EMT. Despite evidence that TGF-β acts as a tumor suppressor in patients with non-metastatic PDAC [7, 13], we found that Panc-1 cells migrated more aggressively when they were both treated with TGF-β and co-cultured with TAMs (Fig. 6a). This observation was consistent with increased Snail mRNA and protein expression in Panc-1 cells treated in the same manner (Fig. 6b and c). A synergistic increase was observed in Snail expression after co-culture with TAMs and co-treatment with TGF-β. In Capan-1 cells (metastatic PDAC cells) the expression of TWIST1, another EMT marker, was increased when they were treated with TGF-β alone or treated with TGF-β and co-cultured with TAMs (Fig. 5c). The baseline Snail protein expression is very low in Capan-1 cells and was not increased upon TGF-β treatment (data now shown).

Fig. 6.

TGF-β induces PDAC progression through EMT. (a) Panc-1 cells were treated with vehicle (PBS), TAM co-culture, recombinant TGF-β, or combined TGF-β + TAM for 48 h. Next, the cells were subjected to a cell migration assay and followed over 48 h. (b) qRT-PCR quantification of Snail expression in Panc-1 cells treated the same as in Fig. 6a (at endpoint of 48 h). Data are shown as relative units of delta-delta Ct normalized to H18S ribosomal RNA. N = 3 experimental replicates shown, mean ± SEM. (c) Western blot analysis of EMT markers Snail and Twist1 with β-tubulin as loading control. Representative blots are shown

Gemcitabine plus galunisertib treatment decreases tumor burden and increases PD-L1 expression in a syngeneic mouse model

In order to understand the relationship more fully between non-SMAD TGF-β signaling and PDAC progression, we used a syngeneic mouse PDAC model to resolve differences in data obtained from Panc-1 cells, Capan-1 cells and patient-derived tumor samples. To this end, we implanted luciferase transfected KPC cells (murine PDAC cells) into C57BL/6 mice (B6 immunocompetent mice) in order to assess critical biologic variations. The respective treatment groups were: PBS control, gemcitabine alone, gemcitabine + TGF-β, or gemcitabine + galunisertib (Fig. 7). At 7 days post-implantation, no difference in tumor burden or metastasis based on bioluminescent imaging was observed (p > 0.3). At 14 days post-implantation, the gemcitabine + galunisertib group showed significantly fewer metastatic deposits as measured by focal luminescence distinct from the pancreas (p = 0.046) and significantly decreased overall bioluminescence as a marker for overall tumor burden (p = 0.018, Fig. 7b).

Fig. 7.

Gemcitabine (Gem) and galunisertib (Gal) reduce pancreatic cancer burden while Gal increases PD-L1 expression in murine PDAC tumors. C57BL/6 J mice (n = 5) were orthotopically injected with luciferase-transfected KPC cells in Matrigel with the assigned treatment and subsequently treated three times weekly with gemcitabine (100 mg/kg i.p.), TGF-β1 (0.05 mg/kg i.p.) or galunisertib (25 mg/kg oral gavage). (a) IVIS images taken 14 days post orthoptic implantation of cells. Metastatic disease was measured by visual inspection of the liver at the end of the experiment (overall p = 0.046 by ANOVA). (b) Total luminescence flux was quantified by treatment group (p = 0.018 PBS versus Gem + Gal). Mean ± standard deviation are shown (n = 5 per group). (c, d) PD-L1 (red) and DAPI (blue) were quantified by immunofluorescence in KPC tumors orthotopically implanted as in (a-b), but treated with galunisertib, gemcitabine, both, or PBS control (n = 5 per group). Representative images are shown (scale bar is 100 μm). Depending on the size of the tumors, between 20 and 80 representative images were taken of each tumor for analysis in Fig. d. (e) Next, the effect of TGF-β treatment versus galunisertib treatment was investigated (n = 5 mice/group with PBS control, as previously described) on tumor associated macrophages (TAMs) by isolating TAMs from single cell suspensions of KPC pancreatic tumors and metastatic liver tumors using flow cytometry after treating the mice three times weekly for 2 weeks. (f) As percentage of all CD45+ cells, there were more TAMs in liver metastatic tumors compared to pancreatic tumors (p = 0.001) after KPC orthotopic implantation. (g) TAMs more highly expressed PD-L1 in the TGF-β treated group compared to either the control group or the galunisertib group in metastatic tumors (53% increase in PD-L1 mean fluorescent intensity compared to the PBS group, p < 0.05) and primary tumors (69% increase in PD-L1 mean fluorescent intensity compared to the PBS group, p = 0.036)

To further understand putative therapeutic responses, we investigated tumoral PD-L1 expression in luciferase transfected KPC cells orthotopically implanted into B6 mice (Fig. 7). We found that galunisertib treatment induced a higher PD-L1 expression in the tumors based on immunofluorescence analysis (p < 0.0001, Fig. 7d). Gemcitabine treatment did not significantly alter the expression level of PD-L1 in these tumors (p > 0.5). The combined treatment group (gemcitabine plus galunisertib) exhibited higher tumoral PD-L1 expression levels compared to controls (p < 0.0001), but similar to the galunisertib only group (Fig. 7d, p > 0.5).

We next compared the effects of TGF-β treatment versus galunisertib treatment on the percentage of TAMs and the expression of PD-L1 on TAMs isolated from pancreatic and liver (metastatic) tumors in the same model (KPC cell line orthotopically implanted into B6 mice) using flow cytometry (Fig. 7e). We found that there were more TAMs (as a percentage of all CD45+ cells) in liver metastatic tumors compared to pancreatic tumors for all treatment groups (p = 0.001, Fig. 7f). We also found that the TAMs more highly expressed PD-L1 in the TGF-β treated group compared to either the control group or the galunisertib treated group in both the metastatic (53% increase in PD-L1 mean fluorescent intensity, p < 0.05, Fig. 7g) and primary (69% increase in PD-L1 mean fluorescent intensity, p = 0.036) tumors.

Discussion

TGF-β remains an important disease defining molecule for PDAC. Due to its ubiquitous expression, multiple molecular targets, co-signals and cellular effects, it remains a challenge to harness its therapeutic potential. We and others have previously shown that the tumor-suppressive and tumor-promotive effects of TGF-β may vary based on the primary versus metastatic location of the cancer cells [5, 9, 59]. Here, we add to our understanding of oncogenic TGF-β signaling through the non-SMAD pathway and identify treatment modalities that lead to PD-L1 expression. We specifically show through in vitro, in vivo and ex vivo studies that even when blocked, the TGF-β receptor can still activate p38MAPK, leading to increased c-Myc, β-catenin and PD-L1 expression. This pathway mechanistically involves Akt and NF-κB. All of these molecular intermediaries are clinically important molecules leading to worse overall prognoses. Whereas this complex cascade leads to PDAC progression through EMT and cellular proliferation, secondary effects such as immune cell regulation in the TME are as yet not fully understood.

Our work explicitly demonstrates that PD-L1 expression is increased on PDAC cells when treated with gemcitabine and galunisertib, suggesting a therapeutic response, a potential therapeutic target for combined or sequential therapy, and a critical mechanism of therapeutic resistance. Previous work by Reyes et al. has shown that neoadjuvant (pre-operative) chemotherapy may not only have cytotoxic effects on cancer cells, but also depletes tumorigenic immune cells [60]. Their study involved tumors from only non-metastatic PDAC patients and revealed a shift both towards a proinflammatory immune TME (anti-tumor) and a decrease in myeloid-derived suppressor cells. Our current work revealed a distinct effect between PDAC tumor cells and TAMs with a mixed effect between metastatic and non-metastatic TMEs, suggesting an immunologic response to both cytotoxic chemotherapy and anti-TGF-β treatment. While the work of Reyes, et al. suggests that this event is a response to chemotherapy, our work suggests that location (pancreas or liver) is an important factor in determining the tumor immune response to therapy. This therapeutic context-dependent response is critical to the development of effective therapies.

We also show that TGF-β receptor activation increases PD-L1 expression in TAMs. This contrasting result, i.e., TGF-β receptor inhibition increases PD-L1 expression on PDAC cells, whereas TGF-β receptor activation increases PD-L1 expression on TAMs, leads to a clinical challenge in which therapeutics (i.e., TGF-β receptor inhibitors such as galunisertib) may have mixed biological responses. PD-L1 expression in TAMs may be a mechanism underlying therapeutic resistance, that may potentially lead to a rational combinational therapy, but for this further mechanistic understanding of intracellular TAM signaling is needed.

Our in vitro studies revealed an increased PD-L1 expression after gemcitabine treatment in both non-metastatic (Panc-1, Capan-2) and metastatic (Capan-1, AsPC-1) pancreatic cancer cells, indicating the potential for a combination treatment for PDAC by using anti-PD-1 or anti-PD-L1 immune checkpoint therapies along with chemotherapy treatment in appropriate patients at the appropriate time when their tumor may be responsive. Alternatively, sequential therapy may be more clinically relevant and effective. Recent studies [61, 62] have shown initial safety and efficacy of galunisertib in patients with unresectable PDAC. Although gemcitabine treatment of Panc-1 cells led to upregulated PD-L1 expression, it did not have a major impact on β-catenin expression in these cells. This result is not surprising as β-catenin is expressed at a very low level in Panc-1 cells at baseline [63], confirmed by our work. These studies did not compare galunisertib to the as yet most effective therapy, FOLFIRINOX [64], for patients with PDAC but they did combine galunisertib with gemcitabine, a well-reasoned approach [61, 65]. An underappreciated limitation of the clinical approach so far is that since the TGF-β receptor can still activate the non-SMAD pathway upon binding to galunisertib, there remains a possibility that this will only inhibit SMAD signaling and induce a predictable mechanism of resistance through non-SMAD TGF-β signaling via e.g. c-Myc, NF-kB and/or β-catenin.

Challenges seen in the aforementioned phase I trials of anti-TGF-β receptor-based therapy are addressed by our in vivo data, which support the use of gemcitabine plus galunisertib in the setting of a decreased primary and metastatic tumor burden. Our data also suggest that resistance may be due to pAkt activity, PD-L1 expression, or both. Both resistant patterns avail themselves to effective therapies with pAkt inhibition or PD-L1 blockade, but require an in-depth molecular understanding of tumor dynamics.

Tumor heterogeneity is another defining characteristic of PDAC [2, 66, 67], and our assessment of the mechanisms underlying EMT demonstrates this as well. Through differences in phosphorylation of Akt, activation of β-catenin, and expression of NF-κB, two distinct mechanisms of EMT are described. Through Snail expression, Panc-1 cells undergo TGF-β-mediated EMT. Capan-1 cells, on the other hand, invoke TWIST1 as a marker of EMT when treated with TGF-β and exposed to TAMs (a known inducer of more aggressive PDAC). We show that these differences are mediated by differences in pAkt activity, a potential therapeutic target [68, 69].

Inhibition of Akt phosphorylation by MK-2206 in in vitro studies resulted in induction of a high percentage of cell death in Panc-1 and Capan-1 cells. Treatment with Akt inhibitors has had mixed results in early phase clinical studies involving PDAC, but these studies did not discriminate patients with tumors that may be Akt dependent versus Akt independent [69, 70]. Given the data on the tumor-suppressive nature of TGF-β in select settings [7], a rational approach might be to direct MK-2206-based therapy only to patients with an upregulated non-SMAD TGF-β signaling pathway, as described herein.

There are some limitations to this study. First, there are multiple types of molecularly defined PDACs with both prognostic and therapeutic implications [2, 71]. Our cell line data represent a small window into the mechanisms that may vary between the different PDAC subtypes. Intrinsic differences between the Panc-1 and Capan-1 cell lines may contribute to the differences we have shown here, but that heterogeneity is present in patients with PDAC as well. Our most important findings were also seen in additional cell lines (Supplemental Fig. 1). The use of a KPC mouse model may help to define a more generalizable data set, but it is not all inclusive. Further studies with patient-derived PDAC organoids are planned along with patient-derived xenograft animal studies. A specific molecular subtype of PDAC, i.e., those with SMAD4 mutations, does not permit normal nuclear internalization of the SMAD2/3/4 complex for normal SMAD pathway activation. These patients, in a sense, have tumors that may only undergo non-SMAD signaling. Supporting this notion is the recurrent observation that SMAD4 mutations are markers for a worse overall survival. [72, 73] Others have reported that patients with SMAD4 mutations have a worse overall prognosis compared to SMAD4 wild type tumors [9, 74], suggesting that competition between SMAD and non-SMAD pathway activation dictates whether TGF-β is a tumor promotor or suppressor, which bears clinical importance. Another limitation of our study is that p38 MAPK may be activated by other receptors/intracellular kinases independent of the TGF-β receptor. While this may limit the therapeutic efficacy of TGF-β receptor modulation, it does not explain our finding that PD-L1 expression can be manipulated by TGF-β receptor inhibition alone, suggesting that multiple molecular mechanisms and potential adaptations are occurring.

A second limitation of this study relates to the nature of metastatic versus primary PDAC. While many believe that PDAC is a systemic disease from its onset, molecular differences in the transition from primary to metastatic disease are not well understood. Another limitation relates to the diverse nature of non-SMAD TGF-β signaling. There are multiple important pathways such as the SMAD, β-catenin, NF-κB, JNK-STAT and Rho pathways, that are modulated by TGF-β receptor activation [75]. The SMAD pathway has been extensively studied [76–79] and here, a critically important non-SMAD pathway is explored. Given the diverse nature of Akt activity and the numerous effects of β-catenin on cellular activity, there may be other gene targets and intermediate proteins that should be explored. However, given the ongoing clinical investigations with small molecule inhibitors directed against Akt, β-catenin and NF-κB, we felt that the investigations undertaken have the highest probability of leading to interesting and insightful results to improve PDAC patient survival.

Conclusions

Non-SMAD TGF-β pathway activation leads to increased EMT and cancer progression through multiple mechanisms in PDAC. Despite TGF-β receptor blockade, intracellular signaling proteins such as Akt, β-catenin, c-Myc and NF-κB are activated either as a response to TGF-β receptor or p38 MAPK phosphorylation. We theorize that this seemingly aberrant phosphorylation of p38 MAPK leads to PD-L1 and c-Myc expression, both leading to PDAC progression. Our current results suggest a new mechanism to target PDAC through the non-SMAD signaling pathway. Specifically, a profound understanding of the immune context of PDAC cell therapeutic responses and TAM responses to TGF-β may provide a rationale for investigating TAM inhibitors in the clinical setting. Our results also suggest a way for personalized therapeutic PDAC approaches. Further studies may lead to (1) personalized therapy based on the degree of SMAD versus non-SMAD TGF-β pathway activation, (2) an improved treatment of patients with uncontrolled non-SMAD TGF-β signal activation, and (3) anti-PD-L1-based therapy in combination with chemotherapy in selected patients based on their response to initial anti-TGF-β-based therapy.

Authors’ contributions

Conception and design: SMH, DS, & ESG.

Development of methodology: SMH, DS, & ESG.

Acquisition of data: SMH, GMC, RGK, MAA, PVD, JLD, DS, ESG.

Analysis and interpretation of data: SMH, GMC, MK, DNH, DS, LM, & ESG.

Writing, reviewing and/or revising: all authors.

Study supervision: ESG.

Funding

This work was supported by the UTHSC Center for Cancer Research, the UTHSC Cancer Biorepository (ESG) and a Society for Surgery of the Alimentary Tract Career Development Award (ESG). LM was also supported by NIH NCI CA253329.

Declarations

Study approval

All research has been approved by the Institutional Review Board (IRB) for human research and the Institutional Animal Use and Care Committee (IACUC) at UTHSC.

Disclosure of potential conflicts of interest

There are no real or potential conflicts of interest.