Pancreatic ductal adenocarcinoma (PDA) remains one of the deadliest forms of cancer, in part, because it is largely refractory to current therapies. The failure of most standard therapies in PDA, as well as promising immune therapies, may be largely ascribed to highly unique and protective stromal microenvironments that present significant biophysical barriers to effective drug delivery, that are immunosuppressive, and that can limit the distribution and function of antitumor immune cells. Here, we utilized stromal reengineering to disrupt these barriers and move the stroma toward normalization using a potent antifibrotic agent, halofuginone. In an autochthonous genetically engineered mouse model of PDA, halofuginone disrupted physical barriers to effective drug distribution by decreasing fibroblast activation and reducing key extracellular matrix elements that drive stromal resistance. Concomitantly, halofuginone treatment altered the immune landscape in PDA, with greater immune infiltrate into regions of low hylauronan, which resulted in increased number and distribution of both classically activated inflammatory macrophages and cytotoxic T cells. In concert with a direct effect on carcinoma cells, this led to widespread intratumoral necrosis and reduced tumor volume. These data point to the multifunctional and critical role of the stroma in tumor protection and survival and demonstrate how compromising tumor integrity to move toward a more normal physiologic state through stroma-targeting therapy will likely be an instrumental component in treating PDA.
This work demonstrates how focused stromal re-engineering approaches to move toward normalization of the stroma disrupt physical barriers to effective drug delivery and promote antitumor immunity.
See related commentary by Huang and Brekken, p. 328
Pancreatic ductal adenocarcinoma (PDA) is one of the most lethal and therapeutically nonresponsive cancers, with nearly identical annual incidence and mortality rates (1). Despite an improved understanding of PDA, 5-year survival has scarcely improved from 5% to 9% (1). Among other factors, including early metastasis and robustly immunosuppressive tumor microenvironments, the difficulty in treating PDA may be largely attributed to protective stromal barriers that effectively limit drug delivery and distribution of antitumor immune cell populations (2–6). Indeed, PDA exhibits a striking lack of vasculature (3, 4), in contrast to virtually all other solid tumors, and extremely elevated interstitial fluid pressure (IFP; refs. 4, 7). This pressure gradient contributes to vascular collapse and effectively blocks transvascular convective (i.e., pressure-driven) transport of small molecules from perfused vessels (4, 7, 8), while dense collagen and glyocosaminoglycan networks also hinder distribution of therapeutics by diffusive (i.e., concentration-driven) transport (4, 7, 9, 10). Collectively, these barriers severely restrict intratumoral drug distribution, resulting in large drug-free regions in PDA.
An overwhelming majority of the dense extracellular matrix (ECM) in PDA is deposited by activated pancreatic stellate cells (PSC, i.e., myofibroblasts or cancer-associated fibroblasts, CAF), which can be activated via inflammation, paracrine signaling, and oxidative stress (11, 12) and also promote carcinoma cell proliferation, migration, and survival (11, 13, 14) and inception of metastatic lesions (14). Furthermore, activated PSCs can also promote immunosuppression through cytokine secretion (15, 16), and recent evidence suggests that these αSMA+ CAFs can display significant heterogeneity with some CAFs possessing stronger inflammatory functions (15). Collectively, these results make PSCs and the ECM attractive therapeutic targets in PDA. However, notably, there are conflicting reports regarding the relative benefit or harm of stroma-targeting therapies. Pharmacologic and/or genetic depletion of the stroma by targeting Hedgehog (Hh) signaling in preinvasive disease does not prevent PDA and may actually result in more aggressive disease (17), while separate studies report a benefit from targeting Hh in preclinical models (3, 18, 19) and human patients (20). Distinct from approaches targeting Hh signaling, other studies have indicated a strong therapeutic benefit from stroma targeting. Enzymatic targeting of hyaluronan in authochonous PDA (4) and angiotensin inhibition in grafted tumors (9, 10) lead to disruption of physical barriers, reduced intratumoral pressure, and restored vascular functionality to facilitate drug delivery and, in some cases, long-term stromal remodeling (4). These studies suggest that our understanding of stromal signaling and physical barriers, and how and when to alter the stroma, remains incomplete, and that novel combinatorial molecular and immune approaches are likely needed in concert with an increased understanding of how stromal reengineering alters pro- and antitumor immunity.
Here, we investigate the ability of the potent antifibrotic agent halofuginone (HF), an analogue of febrifugine, to move the tumor toward stromal normalization by targeting activated PSCs and inhibiting ECM production, thereby disrupting barriers to effective molecular and immune therapies. Febrifugine has been used as an antimalarial and antiparasitic agent, but more recently discovered to have strong antifibrotic capabilities. Although the exact mechanism of action remains somewhat unclear, HF blocks TGFβ signaling through inhibition of Smad2 and Smad3 phosphorylation and can reduce expression of TGFβ type II receptor (21), which may result from inhibition of prolyl-tRNA synthetase activity (22). As a result, HF can decrease pancreatitis-induced fibrosis, slow growth of grafted pancreatic tumors (23–25), and has elicited resolution of pathologic liver fibrosis (26). Furthermore, HF has been administered safely in human patients with advanced solid tumors (27). Yet, while the antifibrotic effects of HF have been studied in various contexts, they have not been previously tested in an autochthonous model of PDA. Here, we show that HF is able to disrupt physical barriers in established stroma to facilitate increased therapeutic delivery. Notably, HF treatment also promotes infiltration and distribution of antitumor immune populations, including classically activated inflammatory macrophages and CD8+ cytotoxic T cells. In concert with a direct effect on carcinoma cells, HF treatment culminates in widespread intratumoral necrosis and reduced tumor volume growth in a genetically engineered mouse model of PDA.
Materials and Methods
All murine primary cell lines were generated from freshly excised KPC tumors (the KPC genetically engineered mouse model is described in detail in the following section) in our laboratory and were confirmed to be Mycoplasma-free. Primary carcinoma cells were confirmed to have the correct genotype and to have undergone Cre-mediated recombination resulting in activation of the Kras oncogene and point mutant p53. To isolate primary carcinoma cells and activated PSCs (i.e., CAFs), tumors were washed with a 10% penicillin–streptomycin solution, minced into small pieces, submerged in 2 mg/mL collagenase and 100 U/mL hyaluronidase at 37°C for 30 minutes with agitation, washed and resuspended 5 times in DMEM, then plated onto FBS-coated tissue culture plates. Pancreatic stellate cells were isolated from cell mixtures by differential trypsinization of mixed populations of cells to highly enrich PSCs, followed by resuspension and incubation with in a biotinylated mouse E-cadherin antibody solution (PlusCellect Mouse E-cadherin+ Cancer/Stem Cells Kit) in combination with anti-β4 integrin and anti-CD11b, followed by resuspension and incubation with Dynabeads Biotin Binder (Thermo Fisher Scientific) to remove carcinoma cells and macrophages. Following the initial purification, differential trypsinization and selection with only E-cadherin beads was used to maintain high PSC purity as needed. Cells were used between 3 and 15 passages. Human primary PSCs were generated using an outgrowth method (28). Deidentified resectable human PDA was finely minced. Tissues were initially cultured in DMEM (Invitrogen) with 20% FCS (“CP-medium”), then second, in KSF medium (Invitrogen), and third, Dresden modified DME medium (“Dresden-medium”), consisting of CP medium and KSF medium at a ratio of 2:1. All media were supplemented with penicillin (100 U/mL) and gentamicin (2.5 mg/mL; Invitrogen). Differential trypsinization was implemented to isolate PSCs from human tumors, and no carcinoma cells were detected in the culture that were confirmed to be Mycoplasma-free. Cells were used within 10 passages. All human samples were deidentified and obtained through the UMN BioNet program in accordance with the US Department of Health and Human Services Common Rule and University of Minnesota Institutional Review Board approval.
Genetically engineered mouse model
All animal studies were approved by the Institutional Animal Care and Use Committee of the University of Minnesota. In vivo studies utilized the KrasG12D/+;p53R172H/+;Pdx-1-Cre (KPC) genetically engineered mouse model, which has proven to be highly reflective of the human cognate condition genetically, clinically, and histopathologically (29). KPC mice conditionally express endogenous, physiologic levels of the respective oncogene and tumor suppressor gene mutations in pancreatic progenitor cells, resulting in the stochastic development and spontaneous progression of preinvasive ductal lesions to invasive and metastatic carcinomas. A subset of studies were done using fluorescent reporter KPC mice [i.e., KrasG12D/+;p53R172H/+;Pdx-1-Cre;ROSATdtomato/+ (KPCT) or KrasG12D/+;p53R172H/+;Pdx-1-Cre;ZsGreen+/+ (KPCG) as we have previously described, ref. 30].
KPC study endpoint protocol
At the endpoint of the drug studies, doxorubicin was injected i.v. 15–20 minutes prior to euthanasia to visualize small-molecule distribution in tumors. This was followed by a terminal surgery, in which animals were anesthetized using isofluorane, and an intracardiac (i.c.) injection of high-molecular-weight (2,000 kDa) FITC- or TRITC-conjugated lysine-fixable dextran was administered 5 minutes prior to euthanasia to label perfused blood vessels, followed by a terminal i.c. injection of 10% formalin. The organs were harvested and fixed in 10% neutral buffered formalin, followed by dehydration in 70% ethanol before paraffin-embedding and sectioning (4 μm) for subsequent histochemical or IHC staining. Separate sections of each sample were frozen in OCT compound.
Ultrasound for enrollment and monitoring of disease progression
Weekly ultrasound was performed on KPC mice using the Vevo 2100 Imaging System for both enrollment and disease monitoring. When tumors reached a diameter of 4 to 9 mm, mice were randomly enrolled in study groups. The Vevo 2100 software was used to reconstruct 3-D tumor volumes for quantification of tumor volume growth over time.
Drug treatment plans and dosages
Unless specifically noted, the mice described in this study were enrolled in 10-day intervention studies using halofuginone, an analogue of the plant alkaloid, febrifugine. A small subset of mice were enrolled in 24-hour intervention studies to evaluate early immune dynamics following one dose of 0.75 mg/kg halofuginone hydrobromide (Sigma-Aldrich) administered intraperitoneal (i.p.), followed by euthanasia 24 hours later (n = 3). Mice enrolled in the control group received no treatment, but were continually monitored, then euthanized after 10 days (n = 6). Mice in the 10-day HF group were treated 3×/week with 0.75 mg/kg HF, receiving a total of 5 doses (n = 5).
Tumor pressures were measured as previously described (3). Briefly, IFP measurements were performed using a Millar Mikro-Tip system (pressure catheter transducer SPR-1000, 0.33 mm diameter) connected to a PCU-2000 Pressure Control Unit and the ADInstruments PowerLab data acquisition system (Millar Instruments, Inc.). The system was calibrated to 0, 25, and 100 mm Hg prior to each measurement following the manufacturer's recommendations.
All stained tissue samples were digitally scanned at high-resolution and viewed using Aperio ImageScope. Using this software, we identified PDA-positive area within hematoxylin and eosin (H&E)-stained sections. Likewise, this approach was used to identify and quantify necrotic area as a percentage of total PDA area. To quantify IHC-stained tissue sections, 10 regions of interest were randomly selected within the PDA-positive area, defined by referencing previously identified PDA-positive areas in adjacent H&E-stained sections. Macros were written in FIJI to analyze each histologic stain, using batch image processing to isolate and quantify the relevant positive signal. To identify perfused blood vessels, 2000 kDa FITC- (Thermo Fisher Scientific) or TRITC-conjugated (Life Technologies) lysine-fixable dextrans were injected i.c. 5 minutes prior to death, followed by a terminal i.c. injection of 10% formalin. Perfused blood vessels were imaged using an inverted Olympus IX81ZDC spinning disk confocal microscope. Corresponding regions were identified in adjacent tissue slices stained for CD31 and used to calculate the total number of vessels. Percent perfusion was calculated as the number of fluorescently labeled, perfused blood vessels as a percentage of the total number of CD31+ vessels in that region. To estimate immune infiltrate in H&E sections, a quantitative image-processing algorithm was created in FIJI. A built-in color deconvolution plugin was used to separate the H&E color channels and isolate the nuclei. Immune cells were distinguished by thresholding for specific nuclear characteristics (immune cells were presumed to have very small, round, and dark nuclei). Nuclear size parameters were determined by iterating the pixel range until the maximum particle size was small enough to exclude the visibly larger nuclei of carcinoma cells, and the minimum was small enough to exclude random debris. Circularity was set to 0.75 to 1.00 to exclude the elongated nuclei of CAFs. An example of the algorithm's input and output is included in Supplementary Fig. S4.
Formalin-fixed paraffin-embedded (FFPE) tissue sections were stained for αSMA (1:25; Dako, M0851), CD31 (1:50; Dianova, DIA310), iNOS (1:150 Abcam; ab15323), and HTI-601 to stain for hyaluronan (1:500). For CD31, antigen retrieval was performed using 1× citrate buffer pH 6 (Sigma) for 20 minutes, washing with 1× TBS/0.1%Tween-20 (TBST; all washes were done in TBST), blocking with 5% goat serum for 1 hour, then incubation with primary antibody overnight at 4°C. This was followed by an endogenous peroxidase block with 3% H2O2 for 15 minutes, incubation with a polymer secondary antibody for 30 minutes (Rat Probe, BioCare Rat-on-Mouse anti-CD31 Polymer Kit), followed by Rat-HRP (BioCare) for 30 minutes, and then counterstaining with freshly filtered Mayer's Hematoxylin (Fisher Scientific) for 5 minutes, followed by dehydration and clearing with increasing concentrations of ethanol and xylenes. For αSMA, antigen retrieval was performed using a 1× Dako Tris/EDTA pH 9 solution for 20 minutes. This was followed with an endogenous peroxidase block of 3% H2O2 for 10 minutes, blocking for avidin/biotin (Vector kit), blocking with TCT buffer (0.1% trypsin, 0.1% CaCl2, 20 mmol/LTris-Cl pH 7.8) for 10 minutes, then incubation with primary antibody for 1 hour at room temperature. Slides were then incubated in MACH2 Rabbit HRP, followed by incubation with DAB for 10 minutes, counterstaining with hematoxylin, and finally dehydration and clearing. For iNOS, antigen retrieval was performed using 1× citrate buffer pH 6 (Sigma) for 30 minutes, washing with 1XTBS/0.1%Tween-20 (TBST), blocking with 5% goat serum for 1 hour, then incubation with primary antibody overnight at 4°C. This was followed by an endogenous peroxidase block with 3% H2O2 for 15 minutes, incubation with anti-rabbit HRP polymer secondary antibody for 30 minutes (BioCare Polymer Kit), followed by incubation with DAB, and then counterstaining with freshly filtered Mayer's Hematoxylin. To stain for HTI-601 (Halozyme): after dewaxing and rehydration, endogenous peroxide block of 0.3% H2O2 for 2 minutes, serum block of 2% FAF-BSA-2% normal goat serum for 30 minutes at room temperature, blocking with avidin/biotin (15 minutes each), incubation with primary antibody for 30 minutes at room temperature. For detection, incubate in HRP-streptavidin (BD Biosciences) for 15 minutes at room temperature, followed by incubation with DAB for 10 minutes, counterstaining with hematoxylin, then dehydration and clearing. All washes were done in TBST.
OCT frozen tissue sections were used to stain for CD11b (1:25; Abcam, ab8878), Ly6G (1:25; Bio X Cell, BP0075-1), and anti-pan cytokeratin (1:200; Millipore Sigma, F3418). Slides were air dried for 15 minutes at room temperature, incubated in acetone for 15 minutes, then air dried for 15 minutes. Slides were rehydrated with 1× PBS for 10 minutes, blocked with 2% normal goat serum for 1 hour at room temperature, and incubated with primary antibody for 2 hours at room temperature. Following this, secondary antibody (1:1,000; Life Technologies,1749750) and directly conjugated antibodies (PanCK) were added for 1 hour at room temperature, followed by counterstaining with Bisbenzimide (1:10,000; Sigma-Aldrich) for 10 minutes at room temperature, and mounting with Prolong Gold (Life Technologies). Washes between steps were done using 1× PBS. Immunofluorescence samples were imaged on a Nikon Ti-U fluorescence microscope. A subsection of FFPE tissue sections were sent to Reveal Biosciences for IF staining, which were dual-stained for CD3/CD8, CD3/CD4, and CD4/FoxP3. For CD3/CD8 and CD3/CD4 staining, CD3 (1:100; Abcam) primary antibody was detected with donkey anti-Rabbit IgG (H+L) secondary antibody, Alexa Fluor 546 conjugate (Life Technologies); CD8 (1:200; eBioscience) primary antibody was detected with goat anti-Rat IgG (H+L secondary antibody, Alexa Fluor 647 conjugate (Life Technologies); CD4 (1:200; eBioscience) primary antibody was detected with goat anti-Rat IgG (H+L secondary antibody, Alexa Fluor 647 conjugate (Life Technologies). Heat-induced antigen retrieval was performed with Epitope Retrieval Buffer 2 for 20 minutes. Primary antibodies were incubated overnight at 4°C. Nonspecific background was blocked with 3% normal donkey serum + 3% normal goat serum in PBS-T. For CD4/FoxP3 staining, FoxP3 (1:100; Cell Signaling) primary antibody was detected with donkey anti-Rabbit IgG (H+L) secondary antibody, Alexa Fluor 546 conjugate (Life Technologies). CD4 (1:50; eBioscience) primary antibody was directly conjugated to eFluor 660. Heat-induced antigen retrieval was performed using Leica Bond Epitope Retrieval Buffer 1 (citrate solution, pH 6) or Epitope Retrieval Buffer 2 for 20 minutes. Primary antibodies were incubated overnight at 4°C. Nonspecific background was blocked with 3% normal donkey serum in PBS-T. Whole slide fluorescent images were generated using a Pannoramic SCAN (3D Histotech). Cell staining for p53 was performed using the CM5 rabbit polyclonal antibody against p53 (1:200 Leica Biosystems; NCL-L-p53-CM5p) followed by Alexa Fluor 488 conjugate secondary antibody and Bisbenzimide.
Imaging and analysis of collagen fiber networks and small-molecule distribution
Multiphoton excitation microscopy and second harmonic generation (SHG) imaging were implemented to visualize doxorubicin distribution and fibrillar collagen in FFPE tissue sections that were rehydrated with xylenes and serial dilutions of ethanol, then mounted with Prolong Gold (Life Technologies). Visualization of collagen and doxorubicin was done on a custom-built multiphoton laser scanning microscope (Prairie Technologies/Bruker) that was described previously (30, 31) using a Mai Tai Ti:Sapphire laser (SpectraPhysics) at an excitation wavelength of 880 nm and 800 nm, respectively.
The Cancer Genome Atlas human patient cohort data analysis
Transcript analysis was performed on human data sets publicly available through the The Cancer Genome Atlas (TCGA) database. Correlation analysis was performed using the GEPIA platform (32).
Two-group data were analyzed using a t test. Multigroup data were analyzed using ANOVA followed by the Tukey multiple comparison post-test. Percentage of fields of view (FOV) data was analyzed using Fisher exact test.
Biophysical barriers prevent effective drug delivery in PDA
As PDA progresses from preinvasive PanINs to invasive disease, an increasingly robust stromal environment develops, effectively preventing therapeutic distribution (4, 9, 12, 33). This stromal complexity, which critically contributes to the unique pathology of PDA, is faithfully reflected in the genetically engineered KrasLSL-G12D/+;p53LSL-R172H/+;Pdx-1-Cre (KPC) mouse model (Fig. 1A–T; refs. 4, 29, 34). Importantly, KPC tumors also utilize the TGFβ pathway (29), which contributes to the fibrotic and immunosuppressive desmoplastic reaction in both murine and human PDA (35, 36), and activated PSCs from KPC tumors express transcripts reminiscent of fibrosis, e.g., Tgfβ1, Col1a1, and Fn1 (16). In KPC mice, αSMA+ myofibroblasts emerge around early-stage PanIN lesions and persist throughout later stages of disease (Fig. 1A–C), depositing a robust ECM network (Fig. 1D–L). This includes a dense collagen matrix (Fig. 1D–I), which was confirmed to be largely fibrillar collagen from SHG imaging (Fig. G–I), and large quantities of hyaluronan (Fig. 1J–L), a pervasive glycosaminoglycan that imbibes fluid (7, 8), contributing to grossly elevated IFP in PDA (Fig. 1T), consistent with previous findings (4). When IFP increases to these levels, convective transvascular transport (i.e., intravascular to interstitial pressure-driven flux) is significantly hindered, and therapeutics must be transported without, or even against, a pressure gradient. Moreover, compounding the unique hypovascularity of PDA (Fig. 1S; refs. 3, 4), many blood vessels actually collapse under the pressure (4), leaving a notable paucity of functional vessels in PDA, and to some degree in preinvasive disease (Fig. 1M–O). Indeed, CD31 IHC reveals a stark difference in blood vessel patency phenotypes between healthy pancreata with abundant normal vessels with apparent lumens, and pancreas tumors with a predominantly compressed vessel phenotype (i.e., no visible lumen; ref. Fig. 1M–O). Previously, we and others have shown that these phenotypes correlate with either fully perfused or nonperfused vasculature, respectively (3, 4, 9). The collective impact of these stromal barriers on drug delivery can be visualized by analyzing doxorubicin distribution in normal pancreata, preinvasive, and invasive disease (Fig. 1P–R). A clear reduction in doxorubicin distribution and nuclear uptake is evident in diseased pancreata and is increasingly pronounced with disease progression (Fig. 1Q and R).
Analysis of transcript data from the TCGA human PDA patient cohort likewise demonstrates robust expression of stromal ECM transcripts (Fig. 1U), as expected (e.g. refs. 37, 38). Further analysis reveals a significant correlation between αSMA expression and expression of fibrotic ECMs, including type-I collagen, fibronectin, and hyaluronan synthases (Fig. 1V; Supplementary Fig. S1), with the strongest correlation between COL1A1 and αSMA (ACTA2) (Supplementary Fig. S1). We note that high expression of type-I collagen, present in the majority of PDA patients, results in poorer survival outcome compared with the patient population with low expression (Fig. 1W). In fact, the 2-year survival for patients with low collagen expression levels is relatively high and surprisingly stable, suggesting that COL1A1 may be a useful biomarker to predict outcome, and that reducing collagen levels in PDA may be beneficial.
To better understand signal transduction associated with fibrosis, we next evaluated relationships between TGFβ signaling elements, which are elevated in human PDA compared with normal pancreata (Fig. 1U), and stromal ECM. Analysis demonstrates significant relationships between TGFβ pathway (i.e., SMAD2, SMAD3, TGFB1, and TGFB2) and stromal ECM transcripts (COL1A1, COL1A2, and HAS1, 2, and 3), with the strongest correlations for collagen (Fig. 1X–Z; Supplementary Fig. S1). Similar relationships were observed between transcripts for type-I collagen or hyaluronan synthases and TGFβ receptor transcripts TGFBR1 or TGFBR2. Thus, these combined data in human and murine pancreas cancers suggest that targeting activated PSCs and collagen through the TGFβ pathway is a viable strategy to combat stromal fibrosis in PDA.
Halofuginone is a potent antifibrotic agent in autochthonous PDA
The potent antifibrotic efficacy of HF has been demonstrated in pancreatitis and when administered near the onset of subcutaneously grafted pancreatic tumors (24, 25). However, HF has yet to be evaluated in autochthonous PDA, particularly during intervention trials. First, to study the effects of HF in vitro, primary PSCs derived from both human and KPC autochthonous tumors (Supplementary Fig. S2A–H) were treated with 0, 50, and 100 nmol/L HF for 48 hours, demonstrating a significant, dose-dependent inhibition of per-cell αSMA expression (i.e., PSC activation) in both mouse and human cells (Fig. 2A–C). Not surprisingly, inhibition of αSMA expression corresponds to reduced expression of fibrosis-related genes, such as type-I collagen (Col1a1) and hyaluronan synthase 2 (Has2; Fig. 2D), and consistent with reports targeting PSCs with Losartan (9, 10). Note, because Has2 is expressed by activated PSCs at ∼17× and ∼85×, the level of Has1 and Has3, respectively (Supplementary Fig. S2I), we focused our analysis on Has2. Furthermore, both human and murine PSCs exhibit decreased proliferation at higher HF doses (Fig. 2E).
Compellingly, the antifibrotic effects of HF seen in vitro were maintained in vivo in physiologically relevant tumors. KPC mice were monitored and enrolled in 10-day intervention studies once tumors reached a diameter of 4 to 9 mm in largest dimension, as evaluated using high-resolution ultrasound (Fig. 2F–M). Consistent with the results observed in vitro, mice treated with HF exhibited a significant decrease in stromal αSMA+ signal, indicating a decrease in density of activated PSCs (Fig. 2F, G, N) and a profound effect on the stromal ECM. HF-treated animals showed a significant decrease in hyaluronan (Fig. 2H, I, O) and total collagen content, as determined through a Masson's trichrome stain (Fig. 2J, K, P), with SHG imaging confirming this effect in fibrillar collagen (Fig. 2L, M, Q), underscoring the utility of HF as a potent antifibrotic agent in PDA.
Antifibrotic therapy facilitates drug delivery in established PDA
Hypertension is a characteristic property of virtually all solid tumors, disrupting natural pressure gradients and dramatically reducing convective transport (7, 8, 39). As a result, diffusive transport through the viscous, charged, tortuous, and sterically challenging interstitial environments becomes the primary mechanism driving therapeutic distribution (8, 39). We note that collagen content has been identified as a main determinant of interstitial transport, with hyaluronan also limiting transport by absorbing residual fluid, resulting in increased interstitial viscosity and intratumoral pressure (7, 39). Therefore, we evaluated the impact of HF-induced depletion of collagen and hyaluronan on molecular transport in PDA. We found that stromal reengineering to normalize tumor microenvironments with antifibrotic therapy markedly disrupts transport barriers associated with PDA (Fig. 3). IFP levels are significantly reduced in HF-treated animals (Fig. 3A), substantially increasing transvascular flux of small-molecule therapy. Furthermore, relieving tumor pressure allows previously collapsed, nonfunctional vessels to open, leading to a significant increase in the percentage of functional, perfused, vessels (Fig. 3B–F) without an increase in vessel density consistent with findings following PEGPH20 therapy but distinct from increased vessel density following Hh treatment (3, 4). As a result of disrupting biophysical barriers to transport in PDA, a significant improvement in small-molecule distribution is observed (Fig. 3G–O; Supplementary Fig. S3A and B), where doxorubicin distribution (Fig. 3P) negatively correlates with fibrotic area (Fig. 3Q). Thus, these data suggest that antifibrotic therapy with HF effectively disrupts barriers to therapeutic delivery and distribution, which is increasingly needed to enhance delivery of peptide and antibody therapies (e.g., molecular and immune therapies), and thus, may represent an effective component of carcinoma–stromal–immune combination therapy strategies.
Halofuginone performs as a dual-action therapy by targeting pancreatic carcinoma cells
HF is capable of inhibiting PSC activation and collagen production in vitro, even at very low, nonterminal levels (Fig. 2). At higher concentrations, PSCs exhibit dose-dependent inhibition of proliferation (Fig. 2). Interestingly, we also observe decreased proliferation of primary pancreatic carcinoma cells in vitro following HF treatment (Fig. 4A). This indicates the possibility of a dual-action role for HF as a PDA therapy in which carcinoma cells are affected in addition to stroma-targeting effects. A number of targeted therapies have been reported to enhance carcinoma cell sensitivity to chemotherapy (40–42), including targeting CTGF, a member of the TGFβ signaling family (40). This, in combination with our proliferation data, led us to question whether HF might increase susceptibility of carcinoma cells to gemcitabine (Gem), a standard-of-care chemotherapeutic used to treat pancreatic cancer. To test this hypothesis, cells were subjected to various combination doses of HF and Gem, selected based on dose–response curves we established for each agent (Fig. 4A and B). Equal numbers of cells were allowed to adhere for 24 hours before HF-supplemented media were added for 24 hours. HF-supplemented media were then replaced by Gem-supplemented media for an additional 24 hours. Finally, supplemented media were removed and replaced with normal media for 2 hours, at which point an MTS cell proliferation assay assessed cell viability (Fig. 4C and D). Interestingly, the relative decrease in proliferation with increasing gemcitabine concentration remains fairly static (∼66%), regardless of HF concentration used to pretreat the cells, revealing that pretreatment of carcinoma cells with HF does not increase sensitivity to gemcitabine (Fig. 4D). It is, however, important to note that gemcitabine treatment following HF therapy can further reduce the carcinoma cell population, suggesting that while HF may be an effective monotherapy, greater cell death may be achieved with HF in concert with chemo and/or immune therapies.
Antifibrotic therapy promotes influx and distribution of macrophages and cytotoxic T cells in PDA
Although immunotherapies are showing great promise in a number of cancers, their impact on PDA has remained modest. This is due, in part, to low neoantigen load, strongly immunosuppressive microenvironments, and a dense stroma, which can physically limit both drug and cell infiltration (2, 5, 6, 43–45). However, studies that targeted the stroma in concert with checkpoint inhibition have observed strong immune response in PDA (16, 45), demonstrating that a robust antitumor immune response can take place in PDA under the right therapeutic conditions. Thus, stromal reengineering to normalize tumor microenvironments may not only facilitate delivery of therapeutics (to combat disease and aid antigen presentation), but also to promote distribution of immune therapies (e.g., checkpoint inhibition and cytokine blockade) and antitumor immune cell populations. Yet, our understanding of immune dynamics and surveillance following stroma targeting is currently limited. Therefore, to determine if HF results in immunomodulatory effects, we examined tissue from mice 24 hours and 10 days after initiation of HF intervention therapy. Examination of H&E tumor sections suggested that mice treated with HF for only 24 hours exhibit a striking increase in immune infiltrate, an effect that appears to subside in the 10-day group (Fig. 5A–C). To quantify histopathology observations of immune infiltrate in H&E sections, we created an image-processing algorithm to distinguish and count immune cells based upon nuclear characteristics. The algorithm selected for nuclei that were smaller than the comparably larger nuclei of carcinoma cells, but larger than random debris, with circularity set to 0.75-1.00 to filter out the elongated nuclei of fibroblasts (Supplementary Fig. S4). During development, the validity of this algorithm was confirmed in a subset of slides stained for CD45. This analysis confirmed a statistically significant increase in immune infiltrate 24 hours after HF treatment that appears to normalize, in terms of cell numbers, by 10 days (Fig. 5D), demonstrating that HF treatment can alter the immune landscape in PDA. Despite this compelling result, this metric did not tell us which immune cell subpopulations were being altered, differences in immune composition between groups, or whether the composition in the 10-day group was altered. We therefore performed a more focused examination of distinct immune cell populations that have been implicated as key regulators of the immune response in PDA.
Inspection of specific immune populations (Fig. 5E–J) through immunofluorescence (IF) and IHC staining of KPC tumors (Supplementary Fig. S5A–S5R) revealed a significant increase in CD11b+ myeloid cells (Fig. 5E) and CD8+ cytotoxic T cells (Fig. 5J and K). Because CD11b is a broad myeloid cell marker, expressed on both protumorigenic [e.g., myeloid derived suppressor cells (MDSC), “M2” alternatively activated macrophages] and antitumorigenic (e.g., “M1” classically activated macrophages) cells, we sought to further define this population. Analysis of IF staining for Ly6G, a validated marker for a primary immunosuppressive population of CD11b+ MDSCs in PDA (5), shows no significant difference in the number of Ly6G+ cells between groups (Fig. 5F). However, there is an insignificant trend suggesting a decrease in Ly6G+ cells in both HF treatment groups, indicating that the increase in CD11b+ cells cannot be attributed to increased infiltration of immunosuppressive MDSCs. In contrast, analysis of IHC staining for iNOS, a functional marker of classically activated inflammatory macrophages, shows significant increase in the presence and distribution of iNOS+ cells following HF treatment (Fig. 5G; Supplementary Fig. S5G–I). The iNOS data are consistent with the trends observed in the Cd11b+ data, suggesting that inflammatory macrophages are a chief contributor to the immune influx observed at 24 hours and remain significantly elevated at 10 days, demonstrating a shift in the immune composition during HF treatment. Furthermore, analysis of T-cell populations showed no significant difference for T helper cells (CD3+CD4+) or T regulatory cells (CD4+Foxp3+; Fig. 5H and I), but reveals a significant increase in CD8+ cytotoxic T cells (Fig. 5J). This is particularly compelling when considered in the context of overall intratumoral distribution. Over half of the FOV analyzed in untreated tumors show a complete absence of CD8+ cells (Fig. 5J), which may be related to reports that the robust ECM in PDA may impede T-cell infiltration or distribution (46, 47). In stark contrast, the 24-hour HF group was completely negative for CD8+ cells in only 23% of FOVs (Fig. 5J), indicating a robust increase in the distribution of CD8+ T cells. These data show that targeting the stroma has the capacity to increase numbers, distribution, and thus sampling, of CD8+ T cells, which are capable of killing multiple cancer cells in a region. Thus, although the mere presence of intratumoral T cells does not guarantee an antitumor response, the ability to increase CD8+ cell infiltration in PDA is an indicator of overcoming two key immunosuppressive pathways in PDA, i.e., physical exclusion by ECM barriers and chemical exclusion via cytokines (16, 45, 47). Therefore, these data suggest that stromal reengineering approaches can generate an enhanced antitumor immune response, but may require additional immune modulating therapy, including checkpoint inhibition, to enhance and maintain an antitumor immune response.
Regions of decreased hyaluronan are associated with increased immune infiltrate
To better understand the stromal changes that influence immune infiltrate into PDA, we revisited our investigation of the ECM (Fig. 6A–I). SHG imaging ruled out fibrillar collagen content as an explanation, because there is no significant difference between the 24 hours and control groups (Fig. 6G–J). However, interestingly, we observed a “relaxation” of the collagen matrix, made evident by increased fiber waviness (i.e., greater crimp under low stress conditions that likely produce the trend of modestly increasing area fraction; Fig. 6H and J)). This is consistent with decreased myofibroblast behavior and decreased hyaluronan swelling pressures that are constrained by collagen networks (4, 8), suggesting that the more relaxed collagen matrix may pose less of a barrier to immune cell infiltration and that changes in the hyaluronan matrix may be occurring. Indeed, we discovered a profound relationship between overall and regional hyaluronan levels and immune infiltrate (using nuclear phenotype algorithm analysis and visual confirmation in adjacent H&E sections; Fig. 6A–F). In control mice, analysis of hyaluronan levels and regions of immune infiltration show an inverse correlation, with greater infiltrate associated with lower hylauronan (Fig. 6K and L). Moreover, we observed, perhaps surprisingly, that rapid changes in hylauronan emerge, with more regions containing low levels of hyaluronan just 24 hours after HF treatment (Fig. 6D–F and K), suggesting that some degradation of hyaluronan has occurred. Interestingly, in the 24-hour group the relationship between hyaluronan levels and immune infiltration holds, as in Fig. 2L, and therefore greater frequency of hyaluronan-low areas in 24-hour-group tumors is associated with a greater immune cell numbers, showing that decreased hyaluronan facilitates greater immune infiltration into PDA. Indeed, comparison of immune cells in “low HA” and “high HA” conditions (bottom and top 50% of signal for each group, respectively) demonstrates that there are greater numbers of immune cells in HA-low regions compared with HA-high regions for both the control- and HF-treated mice, and significant increases in immune cells due to HF treatment (Fig. 6M), consistent with the increase in HA-low regions following HF treatment (Fig. 6E and K). Thus, we conclude that HF treatment results in a rapid impact on HA composition and, thereby, collagen matrix architecture (i.e., matrix loosening likely from decreased hyaluronan-generated swelling pressure), promoting immune infiltration into PDA.
Halofuginone treatment leads to widespread necrosis and reduced tumor volume growth
Although HF alters immune dynamics and is an effective inhibitor of fibrosis and proliferation, macroscopic changes in tumor morphology speak to a more global, clinically relevant outcome. To assess such changes, we measured tumor volume growth via high-resolution ultrasound in KPC mice enrolled in the 10-day intervention group (Fig. 7A–D). HF-treated mice exhibit significantly lower tumor volume growth compared with control mice, with one tumor decreasing in volume (Fig. 7E). Notably, and consistent with our quantitative measures of tumor volume, widespread necrosis is present in tumors from mice treated with HF for 10 days (Fig. 7F–H). Indeed, over 50% of the treated mice show necrosis in >20% of the tumor area, with one extreme responder displaying necrosis in over 50% of the evaluable tumor (Fig. 7H), with no observable toxicity or necrosis to adjacent normal pancreas (Supplementary Fig. S6A and S6B) or significant differences in body weight (Supplementary Fig. S6C). Overall, these data suggest that HF monotherapy significantly affects PDA, but may be particularly useful in concert with single-agent or combinations of molecular and/or immune therapies. Thus, we conclude that HF treatment results in reduced PSC activation inducing alterations in ECM levels and architecture and decreased carcinoma cell and CAF viability; this facilitates infiltration of antitumor immune populations, and in particular macrophages that may influence necrosis and help orchestrate T-cell responses, and culminates with large regions of necrosis and reduced tumor growth. Therefore, while a series of reductionist and cell-depletion studies are required to parse out the specific impact of each cell population, data presented here clearly demonstrate that HF has strong potential to both improve drug delivery and promote infiltration of antitumor immune populations, warranting further study as part of a combination molecular and immune therapy approach to combat metastatic PDA.
HF appears to be a valuable therapeutic agent, disrupting barriers to drug delivery and inducing infiltration of antitumor immune populations. It improves drug transport into and through tumors by enhancing convection and diffusion, which is consistent with, and provides support for, previous conclusions that inordinately high IFP in PDA results from HA-driven swelling pressures constrained by a collagen network (4, 8). Furthermore, our analysis demonstrates that the small subpopulation of patients with low collagen expression show better survival outcome, consistent with recent IHC analysis (48) and suggesting that reducing collagen in PDA may be generally beneficial. Furthermore, the marked immune response and shift in immune composition after HF administration, suggests that HF may be applied synergistically to enhance immune therapy approaches, such as immune-checkpoint inhibitors or agonists for tumoricidal macrophages. Indeed, our findings appear consistent with recent work demonstrating that FAK hyperactivity in PDA promotes stromal collagen and plays a role in immunosuppression, partially explaining the failure of immunotherapy in PDA while also demonstrating enhanced antitumor immunity with stromal disruption and checkpoint therapy (45). Lastly, we note that while HF affected the primary tumor, equally impactful measures are effects on metastases. Unfortunately, a limited number and heterogeneity of evaluable macrometastases in distinct organ sites (n = 3 HF; n = 4 control) restricted statistically powerful conclusions. However, we note that one HF-treated mouse exhibited almost complete metastasis necrosis. Likewise, a pilot case study on one mouse treated over >6.5 months showed large, but almost entirely necrotic lesions in the liver. Thus, our data suggest that further mechanistic studies and full preclinical trials with HF in combination with chemo and immune therapies are warranted.
In light of our findings, and reports implementing diverse stroma-targeting approaches (3, 4, 10, 16, 18, 19, 45), it is clear that stromal reengineering to normalize tumor microenvironments holds promise to improve PDA outcomes, particularly when combined with effective cytotoxic agents. However, not all stroma-targeting approaches have demonstrated clinical benefit to date. For instance, following groundbreaking work defining hypovascularity and the potential for Hh inhibition to alter the stroma and improve chemotherapy delivery in autochthonous PDA (3), several clinical trials were initiated, eventually resulting in conflicting findings using either IPI-926 or vismodegib (20, 49, 50). One phase Ib/II trial (IPI-926 + gemcitabine; NCT01130142) was terminated early when data suggested adverse effects, while a separate trial (IPI-926 + FOLFIRINOX; NCT01383538; ref. 20) suggested therapeutic benefit, consistent with recent reports targeting Hh in preclinical models (18, 19). In an attempt to understand these findings, recent work utilized genetic disruption or prolonged pharmacologic inhibition of Hh starting from early preinvasive disease, and showed that resultant disease appears more aggressive, has increased vascularity, and results in decreased survival (17, 51, 52). A different approach to define and combat stromal barriers in late-stage PDA demonstrated that ablating stromal hyaluronan with PEGPH20 in the context of concomitant chemotherapy normalizes tumor pressures with concurrent reexpansion of collapsed microvasculature and a dramatic increase in drug delivery, leading to reduced metastasis and improved survival (4). This is consistent with the data here, highlighting the utility of targeting a physical property of the tumor and its clinical impact. Indeed, a multitherapeutic PEGPH20 regimen is currently in phase III trial (NCT02715804; with multiple phase II trials ongoing), following phase-II data demonstrating a significant benefit in HA-high PDA patients (53). Likewise, angiotensin inhibition reduces collagen production in CAFs through TGFβ1 inhibition and has shown great promise in grafted tumors (9, 10). Although this approach has yet to be directly tested against autochthonous PDA, it suggests mechanistic overlap with HF and bolsters the case for antifibrotic agents against established PDA. Thus, collectively, while these findings make it clear that great care must be taken when developing rational strategies to perturb CAFs and/or normalize the stroma, they also demonstrate a need to increase our understanding of both physical and molecular mechanisms regulating stromal dynamics in order to develop effective stroma-targeting strategies. Indeed, some of the disparities between the studies may result from a protective role for fibroblasts in early PDA (17), in agreement with early reports suggesting that normal fibroblasts may help prevent or limit disease (54), and recent findings of CAF heterogeneity in PDA (15). It is also plausible that our current understanding of Hh signaling across multiple cell populations associated with PDA tumors is incomplete, particularly because Hh inhibition studies show an angiogenic response that has not been observed with other stroma-targeting approaches. Thus, the implications of targeting Hh signaling may extend beyond suppression of paracrine signaling between carcinoma cells and myofibroblasts. Overall, results suggest that different stages of PDA may respond quite differently to distinct stroma-targeting approaches, that certain stroma-targeting approaches may present a specific therapeutic window where the tumor is susceptible to subsequent therapy, and that it may be unwise to administer some agents long-term. Hence, we clearly need to increase our understanding of the physical and molecular stromal dynamics to develop sound stroma-targeting strategies that will likely be an instrumental component to treating PDA.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
The content of this work is solely the responsibility of the authors and does not necessarily represent the official views of the NIH or other funding agencies.
Conception and design: K.Y. Elahi-Gedwillo, P.P. Provenzano
Development of methodology: K.Y. Elahi-Gedwillo, J. Zettervall, P.P. Provenzano
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.Y. Elahi-Gedwillo, M. Carlson, J. Zettervall, P.P. Provenzano
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.Y. Elahi-Gedwillo, M. Carlson, P.P. Provenzano
Writing, review, and/or revision of the manuscript: K.Y. Elahi-Gedwillo, P.P. Provenzano
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Carlson
Study supervision: P.P. Provenzano
P.P. Provenzano and this work was supported by the NIH (R01CA181385 and U54CA210190 University of Minnesota Physical Sciences in Oncology Center to P.P. Provenzano and a SPORE Development award under P50CA101955 to P.P. Provenzano) and by a Research Scholar Grant, RSG-14-171-01-CSM from the American Cancer Society (to P.P. Provenzano). This work was also supported by the Randy Shaver Research and Community Fund, UMN College of Science and Engineering, Masonic Cancer Center, and grants from the UMN Institute for Engineering in Medicine (all to P.P. Provenzano). K. Elahi-Gedwillo received support from the Department of Education (GAANN fellowship through grant #P200A120198) and ARCS (Achievement Rewards for College Scientists Foundation, Inc.). The authors thank Courtney Podritz and Ben Cooper for their assistance with pilot experiments. We would also like to thank Dr. Ajay Dixit for isolating the human-derived primary PSCs used in our in vitro studies and members of the Provenzano laboratory for insightful comments regarding this work.
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.