RNAi-Mediated PD-L1 Inhibition for Pancreatic Cancer Immunotherapy

The recent past has seen impressive progress in the treatment of various malignancies using immunotherapy. One of the most promising approaches involves immune checkpoint inhibitors. However, the clinical results with these agents have demonstrated variability in the response. Pancreatic cancer, in particular, has proven resistant to initial immunotherapy approaches. Here, we describe an alternative strategy that relies on combining gemcitabine and a novel programmed death-ligand 1 (PD-L1) inhibitor, termed MN-siPDL1. MN-siPDL1 incorporates small interfering RNA against PD-L1 (siPDL1) conjugated to a magnetic nanocarrier (MN). We show that noninvasive magnetic resonance imaging (MRI) could be used to monitor therapeutic response. Combination therapy consisting of gemcitabine and MN-siPDL1 in a syngeneic murine pancreatic cancer model resulted in a significant reduction in tumor growth and an increase in survival. Following optimization, a 90% reduction in tumor volume was achieved 2 weeks after the beginning of treatment. Whereas 100% of the control animals had succumbed to their tumors by week 6 after the beginning of treatment, there was no mortality in the experimental group by week 5, and 67% of the experimental animals survived for 12 weeks. This method could provide therapeutic benefit against an intractable disease for which there are no effective treatments and which is characterized by a mere 1% 5-year survival.

www.nature.com/scientificreports www.nature.com/scientificreports/ that combine: innovative checkpoint inhibitors that can be delivered efficiently to tumor cells and tumor resident macrophages, and strategies that enhance the permeation of the tumor by T lymphocytes.
Here, we explore an alternative strategy that relies on combining gemcitabine (Gem) and a novel PD-L1 inhibitor (termed MN-siPDL1). MN-siPDL1 incorporates a nanoparticle carrier that is delivered with high efficiency to tumor cells in vivo [11][12][13][14][15][16][17][18][19] , where it post-transcriptionally inhibits PD-L1 expression on tumor cells via the RNA interference mechanism. The approach is advantageous over small molecules or antibodies because the siRNA component inhibits the target antigen at the post-transcriptional level and not at the protein level. Also, the RNAi mechanism is catalytic and necessitates the delivery of only picomolar amounts of siRNA to the tumor cell for the abolition of the target antigen. By contrast, small molecules or antibodies require the achievement of at least a 1:1 molar ratio of antigen to therapeutic molecule and could be ineffective in the presence of a compensatory increase in the expression of the target antigen by the tumor cell.
A key advantage of our therapeutic approach also derives from the fact that MN-siPDL1 incorporates a superparamagnetic nanoparticle core whose accumulation in tissues over time could be monitored by noninvasive MRI. This capability is highly significant when designing and optimizing therapeutic protocols in the process of drug development.
In the current study, we administered 7 weeks of combination therapy consisting of gemcitabine and MN-siPDL1 in a syngeneic murine pancreatic cancer model. This approach resulted in significantly lower morbidity and toxicity, leading to tumor regression and a dramatic improvement in survival. In particular, following dose optimization, a 90% reduction in tumor volume was achieved 2 weeks after the beginning of treatment. Whereas 100% of the control animals had succumbed to their tumors by week 6 after the beginning of treatment, there was no mortality in the experimental group by week 5, and 67% of the experimental animals survived for 12 weeks.

Synthesis of dextran coated magnetic nanoparticles (MN).
MN was synthesized following a protocol published previously 20 . Briefly, 30 ml of Dextan-T10 (0.3 g ml −1 , Pharmacosmos A/S, Holbaek, Denmark) was mixed with 1 ml of FeCl 3 .6H 2 O (0.65 g ml −1 , Sigma, Saint Louis, MO) while flushing Argon gas for an hour. 1 ml of FeCl 2 .4H 2 O (0.4 g ml −1 , Sigma) was added into the mixture. Following, 15 ml of cold NH 4 OH (28%, Sigma) was added dropwise to the stirred mixture. The temperature was increased to 85 °C for 1 h to start the formation of a nanoparticulate dispersion and then cooled to room temperature. The magnetic nanoparticles were concentrated to 20 ml using Amicon ultra centrifugal units (MWCO 30 kDa; Millipore, Darmstadt, Germany). The resulting dextran-coated magnetic nanoparticles were cross-linked by epichlorohydrin (14 ml,8 h,Sigma) and aminated by the addition of NH 4 OH (28%, 60 ml). Aminated magnetic nanoparticles (MN) were purified by dialysis and concentrated using Amicon ultra centrifugal units. The properties of MN were as follows: concentration, 10.94 mg ml −1 as Fe; the number of amine groups per MN, 64; relaxivity (R2), 82.5 ± 1.16 mM −1 sec −1 ; size of MN, 20.3 ± 0.6 nm (NanoSight LM-10 system and Nanoparticles Tracking Analysis software (Ver. 3.2), Malvern, UK). Nanodrug Synthesis and Characterization. Nanodrug synthesis was carried out according to a previously published protocol 20 . Briefly, MN was conjugated to the heterobifunctional linker N-Succinimidyl 3-[2-pyridyldithio]-propionate (SPDP, Thermoscientific Co., Rockford, IL), which was utilized for the linkage of activated siRNA oligos. SPDP was dissolved in anhydrous DMSO and incubated with MN. The 5′-ThioMC6 of the siRNA oligo was activated to release the thiol via 3% TCEP (Thermoscientific Co., Rockford, IL) treatment in nuclease free PBS. The siRNA oligos were purified using an ammonium acetate/ethanol precipitation method. After TCEP-activation and purification, each oligo (siPDL1 and siSCR) was dissolved in water and incubated with the SPDP modified MN overnight to prepare the final product (MN-siPDL1 and MN-siSCR). Oligos were added to MN at two different ratios to obtain nanodrugs incorporating 2.1 ± 0.4 (low-dose) or 4.8 ± 0.7 (high-dose) siRNA oligos per MN, as quantified by electrophoresis 20 . The size of the final MN-siPDL1/SCR was 23.2 ± 0.9 nm. The immunohistological tissue staining was performed following the protocol for each biomarker. Briefly, excised tumor tissues were embedded in Tissue-Tek OCT compound (Sakura Finetek, Torrance, CA) and snap frozen in liquid nitrogen. The tissues were cut into 7 µm sections and fixed in 4% formaldehyde for 10 min. Detergent permeabilization was performed using 0.1% Triton X-100 in PBS, when needed. After blocking with www.nature.com/scientificreports www.nature.com/scientificreports/ 5% goat serum in 0.5% bovine serum albumin in PBS, each slide was incubated with corresponding primary antibody (dilution 1/200) at 5 °C overnight. Each slide was incubated with secondary antibody (dilution 1/200) for 60 min and mounted with Vectashield mounting medium with DAPI (Vector Laboratories, Inc. Burlingame, CA). The slides were analyzed using a Nikon E400 fluorescence microscope (Nikon, Tokyo, Japan), equipped with the necessary filter sets (MVI Inc., Avon, MA). Images were acquired using a charge coupled device camera with near-IR sensitivity (SPOT 7.4 Slider RTKE; Diagnostic Instruments, Sterling Heights, MI). The images were analyzed using SPOT 4.0 Advance version software (Diagnostic Instruments) and ImageJ (Ver. 1.51c, NIH).
Animal model. Six-week-old female mice (C57Bl/6) were implanted into the right flank with the murine pancreatic cancer cell line, Pan02 (0.25 × 10 6 cells). One week after cell injection, tumor size was monitored by caliper measurements. Tumor volume was calculated according to the equation: Volume = 0.5 x L x W 2 , where L is length, and W, width. Treatment was initiated once tumor volumes reached 50 mm 3 , as estimated using calipers. Thereafter, tumor volume was measured by MRI once mice were enrolled in the study and before and after each weekly treatment. All animal experiments were performed in compliance with institutional guidelines and approved by the Institutional Animal Care and Use Committee at Massachusetts General Hospital.
Therapy. Mice were randomly assigned to two experimental groups treated with low-dose MN-siPDL1 (10 mg kg −1 as iron; 520 nmoles/kg siRNA) in solution with gemcitabine (333.3 mg/kg)(n = 6), or high-dose MN-siPDL1 (10 mg kg −1 as iron; 937 nmoles/kg siRNA) in solution with gemcitabine (333.3 mg/kg)(n = 6) and two control groups treated with MN-siSCR + gemcitabine at the same doses (n = 6). Experiments were performed in three independent trials. The drugs were administered weekly as a mixture of nanodrug and gemcitabine (i.v.). After week 7, co-administration of gemcitabine was discontinued to avoid exceeding the maximum tolerated dose, and only nanodrug was administered until the end of the study. All mice were monitored weekly by magnetic resonance imaging to keep track of changes in tumor volume for a maximum of 12 weeks after the first treatment or until animals became moribund.
Statistical analysis. Data were expressed as mean ± s.d. or s.e.m., where indicated. Statistical comparisons were drawn using a two-tailed t-test (SigmaStat 3.0; Systat Software, Richmond, CA). A value of P < 0.05 was considered statistically significant.

MN-siPDL1 can be delivered in vivo to primary tumors and the delivery can be monitored by noninvasive MRI.
In order to successfully deliver therapeutic amounts of siRNA to tumor cells following intravenous injection, we needed to optimize the design of MN-siPDL1 in terms of hydrodynamic size, conjugation method, and number of siRNA oligos per nanoparticle. The synthetic scheme of MN-siPDL1 is illustrated in Fig. 1A. To ensure long circulation times (>4 hrs.) and efficient diffusion across the vascular endothelium and throughout the tumor interstitium, we designed MN-siPDL1 so that its final size after sequential conjugation to the SPDP linker and the oligo was 23.2 ± 0.9 nm. The number of siRNA oligonucleotides per nanoparticle was adjusted to no more than 5.5 with the goal of minimizing steric interference with bioconjugation. This design is optimized to enhance the uptake of the nanodrug by tumor tissue through the Enhanced Permeation-Retention (EPR) effect. In addition, the SPDP linker was chosen due to its reducible nature, which ensures dissociation of the oligo from the nanoparticle in cancer cells and efficient entry into the RNA-induced silencing complex (RISC). Finally, to permit detection of MN-siPDL1 by magnetic resonance imaging, the relaxivity (R2) of the final preparation was adjusted by varying the ratios of [Fe 3+ ]/[Fe 2 + ] to achieve an R2 of 82.5 ± 1.16 mM −1 sec −1 (Fig. 1B).
For the purposes of establishing an effective therapeutic protocol, we needed to confirm delivery of MN-siPDL1 to pancreatic tumor tissue and to demonstrate the capability of MRI to semi-quantitatively measure MN-siPDL1 bioavailability in tumors. We tested our hypothesis using the PAN02 syngeneic pancreatic cancer model. These animals formed tumors 2 weeks after inoculation and were monitored by MR imaging, using single-echo and multi-echo T 2 weighted protocols.
www.nature.com/scientificreports www.nature.com/scientificreports/ As shown in Fig. 2A, the localization of MN-siPDL1 in tumor tissue caused shortening of the T2 relaxation time and resulted in negative contrast as compared to the pre-contrast image. The delta R 2 -derived concentration of MN-siPDL1 showed a linear increase during the first three weeks. The accumulation rate of MN-siPDL1 was 1.5-fold faster than that of MN-siSCR during that time period (Fig. 2B). Since the concentration of the nanodrug in tumor cells reflects mostly dilution due to cell division, the faster growth rate of the control tumors treated with MN-siSCR likely led to the observed slower increase in concentration over time in this group. This difference was more pronounced at the later stages of tumor growth, further supporting this hypothesis. The rate of concentration decrease, reflective of rapid nanodrug dilution due to tumor cell division in the control group treated with MN-siSCR, was 5.1-fold greater than in the experimental group treated with MN-siPDL1, indicating a more rapid growth of the tumor in the control animals (Fig. 2C).
Interestingly, a subgroup of animals treated with the high-dose MN-siPDL1 failed to respond to treatment, as defined by rapid tumor progression and limited survival. In that group, the early increase in tissue concentration of the MN label as measured by delta-R 2 was intermediate between the animals that responded to treatment and those treated with MN-siSCR (Fig. 2B). At the later time points, the dilution of the MN label in that group was also significantly more rapid than in the responder animals (Fig. 2C). This observation suggested that indeed, the described imaging approach could represent a useful biomarker for response stratification during treatment.

Combination treatment with gemcitabine and MN-siPDL1 is effective in a model of syngeneic pancreatic cancer.
Our therapeutic studies illustrated the potential of the combination treatment with gemcitabine and MN-siPDL1 in pancreatic cancer. To determine whether treatment with gemcitabine in combination with MN-siPDL1 could inhibit tumor growth, the mice were treated with gemcitabine in solution with a low dose of MN-siPDL1 or siSCR (10 mg/kg Fe; 520nmoles/kg siRNA in both groups) or a high dose of MN-siPDL1 or siSCR (10 mg/kg Fe, 937nmoles/kg siRNA in both groups). The combination treatment was initiated when the tumor size reached >50 mm 3 as measured by anatomical MR imaging and continued for 12 weeks. In all of the therapeutic studies, the change in tumor volume was monitored by anatomical MR imaging before the administration of each weekly treatment.
The mice co-treated with MN-siPDL1 and gemcitabine demonstrated significant inhibition of tumor growth, relative to the inactive MN-siSCR controls (P < 0.05). This difference was evident at week 2 from the beginning of treatment, when tumor volume had decreased from 52.8 ± 6.7 mm 3 in week 0 to 5.3 ± 0.8 mm 3 in week 2 (p = 0.012). The difference persisted for the duration of the study (p < 0.05). Tumor volumes in the low-dose group were not different from the MN-siSCR control until week 6 (Fig. 3A,B).
The advantage of the combination treatment was clearly seen when assessing animal survival (Fig. 3C). 67% of the mice treated with gemcitabine and MN-siPDL1 (high dose) survived until week 12. 67% of the mice treated with gemcitabine and MN-siPDL1 (low dose) survived until week 8. All of the control mice treated with MN-siSCR and gemcitabine succumbed by week 6.
Interestingly, all of the mice in the group treated with gemcitabine and MN-siSCR developed large necrotic tumors, presumably due to the high rate of tumor growth. Tumor necrosis and ulceration was not seen in the experimental animals (Fig. 3D).
Importantly, in the high-dose MN-siPDL1 cohort, a subgroup of the mice failed to respond and had tumor growth rate curves and survival that were analogous to the MN-siSCR group (Fig. 4). The time constants of tumor growth stratifying the experimental animals according to response are presented in Table 1. These results indicated variability of the response, warranting further investigation.

Combination treatment prevents the inactivation of cytotoxic T cells.
In order to assess the effect of treatment on the anti-tumor immune response, we analyzed tissue biomarkers of immune cell recruitment and activation in the tumors of treated mice. After combination treatment with MN-siPDL1 and gemcitabine, there www.nature.com/scientificreports www.nature.com/scientificreports/ was an increase in the recruitment of CD8+ tumor infiltrating lymphocytes (TILs). PD-L1 expression was significantly reduced. There was evidence of an increase in cell-mediated cytotoxicity, as evidenced by higher levels of Granzyme B and a decrease in the infiltration by immunosuppressive Foxp3 + regulatory T (Treg) cells. Finally, tumor cell proliferation was inhibited (Fig. 5). Interestingly, the expression of these biomarkers in non-responsive animals treated with high-dose MN-siPDL1 and gemcitabine, was intermediate between the control animals and the regressing experimental animals, suggesting that there is a critical level of PD-L1 inhibition needed in order to observe macroscopic response (Fig. 5). These results suggested that the observed therapeutic effect was the result of successful induction of an anti-tumor immune response.

Discussion
Despite the promise of checkpoint inhibition for cancer immunotherapy, the response is generally variable, with a large number of patients failing to respond. Notable examples of FDA approved PD-L1 inhibitors include atezolizumab for metastatic non-small cell lung cancer (NSCLC) 21 and durvalumab for locally advanced or metastatic urothelial carcinoma 22 . However, despite initial encouraging results and fast-track approval of atezolizumab for bladder cancer 23,24 , the confirmatory trial failed to achieve its primary endpoint of overall survival 25 . Similarly, a phase III trial of durvalumab with tremelimumab as a first-line treatment of non-small cell lung cancer failed to meet its primary endpoint of progression-free survival 26 . In pancreatic cancer, advances in checkpoint inhibitor-based therapies have shown disappointing clinical results. In a Phase II trial of the CTLA-4 inhibitor, Ipilimumab, monotherapy was ineffective with no responders resulting from the trial 27 . Similarly, in a multicenter Phase I trial an anti-PDL-1 antibody was administered intravenously in a variety of advanced cancer patients. Out of the 14 pancreatic cancer patients recruited, there were no objective responses reported 28 .
However, recent preliminary results from a randomized Phase II study in patients with metastatic pancreatic adenocarcinoma showed that combination therapy (Gemcitabine, Nab-Paclitaxel, Durvalumab, and www.nature.com/scientificreports www.nature.com/scientificreports/ Tremelimumab) was well tolerated with 73% of patients reporting partial response. Disease control rate was 100%, median progression free survival was 7.9 months, and 6-month survival was 80% 29 . These studies highlight the potential treatment efficacy of combination therapies with chemotherapy and checkpoint inhibitors.
Here, we describe an alternative design of a PD-L1 antagonist that post-transcriptionally inhibits PD-L1 expression on tumor cells via the RNA interference mechanism. The approach is advantageous over small molecules or antibodies because the siRNA component inhibits the target antigen at the post-transcriptional level and not at the protein level. Also, the RNAi mechanism is catalytic and necessitates the delivery of only picomolar amounts of siRNA to the tumor cell for the abolition of the target antigen. By contrast, small molecules or antibodies require the achievement of at least a 1:1 molar ratio of antigen to therapeutic molecule and could be ineffective in the presence of a compensatory increase in the expression of the target antigen by the tumor cell.
An additional key advantage of our therapeutic approach derives from the fact that it presents the unique opportunity to develop a clinically-relevant, image-guided treatment protocol that provides knowledge about therapeutic outcome, expressed both as change in tumor volume and tumor growth rate. The latter capability is made possible by the fact that MN-siPDL1 incorporates a 20-nm superparamagnetic nanoparticle carrier, which ensures highly efficient delivery to tumor cells and whose disposition in tissue over time can be monitored by quantitative noninvasive MRI. As suggested by our results, tumor delta-R2 would reflect tumor growth rate, which is expected based on the fact that the loss of these nanoparticles from tumor cells is governed by cell division. In addition to the assessment of tumor growth, anatomical MRI allowed the objective measurement of tumor volume as a morphologic biomarker of response. However, the application of dynamic MR imaging protocols could readily be used to also measure physiologic variables related to tumor blood flow and microvessel permeability.
In this sense, the described methodology represents an integrated tool for drug delivery and a synchronous biomarker of therapeutic response. Such tools, if introduced into the clinic, would genuinely exemplify the essence of rational precision medicine. Given that this approach will lay the foundation for future rational designs of novel therapies, we anticipate that further combinations of targets that may work synergistically by complementary mechanisms could be interesting. For example, one could envision combination therapies that physically alter the tumor microenviroment by enzymatic degradation via recombinant human hyaluronidase (PEGPH20) 30,31 , or other alternative chemotherapy agents, and/or alternative checkpoint inhibitors that may promote a synergistic effect in activating T-cells (PD-1 and CTLA-4). However, prior to expanding to alternative www.nature.com/scientificreports www.nature.com/scientificreports/ targets and combinations, the safety of the proposed approach will be tested in large animals as we prepare for Phase I testing.
On a more concrete level, while broadly applicable to solid malignancies, the present study focuses on pancreatic cancer because of its dismal prognosis and the lack of progress against its metastatic form. Approaches, such as the one described here could advance the treatment of pancreatic cancer and potentially vastly improve treatment outcomes in patients for whom no other therapeutic options are available.