Targeting the αvβ5 Integrin Modifies the TGF-β-Rich Tumor Microenvironment of Pancreatic Cancer

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is notorious for its aggressive, therapy-resistant nature that is in part driven by the desmoplastic, hypo-perfused, and immunosuppressive tumor microenvironment (TME). Here, we demonstrated that the αv integrin- and neuropilin-1 (NRP-1)-dual targeting iRGD peptide reverses some of these TME features by inhibiting transforming growth factor-β (TGF-β) activation in the tumor, a process mediated by the αvβ5 integrin. In addition to PDAC epithelial cells and fibroblasts, regulatory T cells (Tregs) in PDAC tumors also expressed the αvβ5 integrin and NRP-1. The αvβ5⁺ Tregs potently inhibited T cell proliferation, and systemic iRGD therapy not only depleted αvβ5⁺ Tregs from PDAC tumors but also reduced their αvβ5⁻ counterparts. Mechanistically, iRGD inhibited the activation of TGF-β mediated by the αvβ5-rich TME, thereby depriving Tregs of the cytokine essential for their development and maintenance. NRP-1-dependent tumor penetration was required for this effect because a traditional RGD peptide without an NRP-1-binding motif failed to inhibit TGF-β signaling or deplete Tregs in vivo. Treatment with iRGD induced a series of additional TME changes, such as improved vascular patency and perfusion, reduced stromal fibers, and increased CD8⁺ T cell entry into the core of the tumors. Combining iRGD with immune checkpoint blockade led to an enhanced anti-tumor effect. Together, these findings support targeting the αvβ5 integrin with affinity ligands such as iRGD as a potential approach to enhance immunotherapy efficacy against PDAC and other desmoplastic tumors with high TGF-β and αvβ5 expression.

INTRODUCTION

The αvβ5 integrin binds to proteins containing the RGD peptide motif (1,2). It mediates diverse processes related to cell survival, cell motility, and endocytosis (3–5). αvβ5 is highly expressed in pancreatic ductal adenocarcinoma (PDAC) and associated with poor prognosis in PDAC patients (6). Various cells in PDAC express this integrin, such as blood vessel endothelial cells (ECs), cancer epithelial cells, and carcinoma-associated fibroblasts (CAFs) (4,7). We have shown that CAFs and cancer epithelial cells collaborate to maintain an αvβ5-rich tumor microenvironment (TME) by producing transforming growth factor-β (TGF-β), which induces the expression of the integrin (7). Genetically deleting the β5 integrin from PDAC cells led to significantly delayed tumor growth and reduced metastasis in mice, indicating an important role of αvβ5 in PDAC progression (7).

The αvβ5-rich TME makes PDAC an optimal target for the iRGD tumor-penetrating peptide (amino acid sequence: CRGDKGPDC) (8). iRGD carries a tumor-specific RGD motif that binds to αv integrins, and a tissue/cell-penetrating RXXK/R motif that binds to neuropilin-1 (NRP-1). Systemically injected iRGD homes to tumors by targeting αv integrins expressed on tumor blood vessel ECs. It is then proteolytically processed to expose an active RXXK/R motif that now binds to NRP-1. The RXXK/R-NRP-1 interaction activates a penetration pathway that transports bystander molecules together with iRGD across the vascular barrier (9,10). In PDAC, CAFs and cancer epithelial cells that express αvβ5 (and NRP-1) and propel the penetration deeper despite the desmoplasia (7). This mechanism allows iRGD to deliver co-injected free drugs deep into the tumor, making tumor-specific drug delivery effective and versatile (11). A phase 1b clinical trial in stage 4 PDAC patients showed that iRGD co-injection therapy may enhance the efficacy of gemcitabine (Gem) and nab-paclitaxel (Nab-P) without increasing off-target side effects (12). Randomized double-blind phase 2 studies are in progress (e.g., NCT05042128, NCT06261359) (13).

TGF-β is secreted as an inactive large latent complex (LLC) consisting of the latency-associated protein (LAP), TGF-β, and latent TGF-β-binding protein (14). Active TGF-β is released from the LLC by cytoskeletal force when LLC binds to αv integrins via its RGD motif (14,15). TGF-β has a multitude of effects on the TME (16,17). For instance, it induces a chaotic network of dysfunctional blood vessels through angiogenesis, which can lead to poor tissue perfusion, hypoxia, and high interstitial fluid pressure (18). It also activates CAFs, which promote desmoplasia by producing cytokines, growth factors, and extracellular matrix (ECM) components (19,20). TGF-β promotes an immunosuppressive milieu because it suppresses various immune cells critical for anti-tumor immunity and facilitates the development and functions of immunosuppressive regulatory T cells (Tregs) (21,22). These features make PDAC resistant to therapies because poor perfusion and desmoplasia prevent drugs from entering the tumor, and the immunosuppressive TME disarms antitumor immune responses (23).

Tregs are major producers of TGF-β in the tumor (17). Tregs are believed to be one of the drivers of the immunosuppressive TME of PDAC, which protects PDAC cells from T cell-mediated anti-tumor immunity (24). A decreased ratio of tumor-infiltrating CD8⁺ T cells to Tregs (CD8/Treg ratio) correlates with resistance to conventional ICBs (25) and poor prognosis in PDAC patients (24). Depleting Tregs has been shown to enhance anti-tumor immunity in preclinical tumor models (26). However, most of the approaches lead to systemic Treg depletion, which can cause serious autoimmune toxicities. Systemic Treg depletion can also accelerate PDAC progression by facilitating tumor infiltration of immunosuppressive myeloid cells (27). Thus, Treg depletion should ideally be achieved in a tumor-selective manner (26).

In this study, we show that iRGD counteracts an αvβ5-dependent TGF-β activation mechanism to induce a dynamic TME modification in PDAC and reverse its immunosuppressive nature.

MATERIALS AND METHODS

Peptides

The cyclic peptides iRGD (acetyl-CRGDKGPDC-NH₂), iRGE (acetyl-CRGEKGPDC-NH₂), CRGDC (acetyl-CRGDC-NH₂), CRGDK (acetyl-CRGDK-NH₂), RPARPAR-NH₂ (acetyl-RPARPAR-NH₂) and RPARPAR-OH (acetyl-RPARPAR-OH) were synthesized in-house as previously described (8) and were also purchased from a commercial vendor (LifeTein, Hillsborough, NJ). In brief, the in-house synthesis was performed with a Liberty automatic microwave-assisted peptide synthesizer (CEM Corporation, Matthews, NC) using standard solid-phase Fmoc/t-Bu chemistry. Some of the peptides were labeled with 5(6)-carboxyfluorescein (FAM) separated with a 6-aminohexanoic acid spacer (8).

Cell culture

KPC78 PDAC cells were established from the tumors of transgenic Kras^G12D/+;LSL-Trp53^R172H/+;Pdx-1-Cre mice (KPC mice) as previously reported (7). The cells had been labeled with luciferase using a lentiviral vector (LV-Fluc-P2A-Neo; Imanis Life Sciences, Rochester, MN). hPCF1424 CAFs cells were established from fresh surgical specimens of human PDAC as previously reported (7). The M21 (RRID: CVCL_D031) and M21L (RRID: CVCL_JZ12) cell lines were a prior gift from Dr. David Cheresh. The rest of the cells were purchased through American Type Culture Collection (ATCC, Manassas, VA). hPCF1424 CAFs cells were authenticated by Labcorp (Burlington, NC), and the other cells were authenticated by ATCC. The cells were cultured in Dulbecco’s modified Eagle medium with 10% fetal bovine serum (FBS) and a penicillin-streptomycin mixture. All cell lines were routinely tested for Mycoplasma (Cat# 25235; Boca Scientific Inc, Dedham, MA). The cells were used within 10 passages after thawing.

Preparation of β5 integrin-KO KPC78 cells

Tet-on Cas9 gene was introduced into KPC78 cells by the Lentivirus vector prepared from pCW-Cas9 (Cat# 50661, RRID: Addgene_50661; Addgene, Watertown, MA), pCMV-VSV-G (Cat# 8454, RRID: Addgene_8454; Addgene), and psPAX2 (Cat# 12260, RRID: Addgene_12260; Addgene). After 3 days of treatment with 2 mg/ml of doxycycline, KPC78 (Tet-ON Cas9) cells were transfected with a plasmid carrying mouse ITGb5-specific guide RNAs (target sequence: 5’-AGGTAGGTTCCGGAGGACGT-3’) and an EGFP gene using Lipofectamine 3000 Reagent (Cat# L300015; Thermo Fisher Scientific, Waltham, MA). GFP-positive cells were isolated by BD Influx cell sorter (RRID: SCR_019593; BD Biosciences, Franklin Lakes, NJ). Limiting dilution was performed to select for clones. Deletion of the β5 integrin in the clones was confirmed by flow cytometry using a mouse anti-human/mouse αvβ5 integrin-Alexa 647 antibody (Ab) (1:50) (clone ALULA; Cat# 565836, RRID: AB_2739376; BD Biosciences) and immunoblotting using a rabbit anti-β5 integrin Ab (1:500) (Cat# PA5-50991, RRID: AB_2636439; Thermo Fisher Scientific).

Human sample collection

Human studies were approved by the Institutional Review Board (IRB) at Columbia University Irving Medical Center (CUIMC) (protocol #AAAT1231) and conducted in accordance with the Declaration of Helsinki. De-identified retrospective tissue specimens were obtained from an institutional tissue bank. Requirement for written informed consent was waived by the IRB for these samples. For prospective studies, biological specimens were collected from patients undergoing standard-of-care surgical resection at CUIMC/NewYork-Presbyterian (NYP) after obtaining written informed consent from all participants.

Tumor mouse models

All animal experiments were performed according to procedures approved by the Institutional Animal Care and Use Committee (IACUC) at Columbia University (New York, NY). Syngeneic PDAC mice were generated by orthotopic pancreatic injections of 5.0 x 10⁵ KPC78 PDAC cells into 8- to 10-week-old C57B6129SF1/J hybrid mice, which are the offspring of a cross between C57BL/6J females and 129S1/SvImJ males (RRID: IMSR_JAX:101043; Jackson, Bar Harbor, ME). The tumor mice were weighed twice daily for 2 days and then daily for 1-week post-surgery. From day 8, animals were weighed twice a week until the end of the experiment. The animals were monitored for signs of distress such as reduced body weight, cachexia, lack of mobility, reduced food and water intake or discomfort, and euthanized according to the IACUC guidelines. The allocation of groups, drug administration, and measurements were conducted independently by at least two individuals to avoid bias. No mice were excluded from the analysis. The minimal sample size required to obtain a statistically meaningful result to answer our hypotheses was estimated based on similar studies performed in the past. Tumor samples from transgenic p48-Cre, LSL-Kras^G12D, Ink4a^flox (KRAS-Ink) and KPC mice were harvested during studies published previously (7,8).

In vivo peptide homing assay in PDAC mice

C57B6129SF1/J mice bearing 22-day-old orthotopic KPC78 tumors received an intravenous (IV) bolus of PBS or FAM-iRGD (12 μmol/kg). One hour later, the mice were perfused through the heart with PBS under deep anesthesia, and the tumors were collected. The tissues were stained for CD4, Foxp3, and DAPI using Abs described elsewhere, and subjected to confocal microscopy. Homing of FAM-iRGD to tumor-infiltrating Tregs in transgenic KRAS-Ink mice was studied by re-analyzing immunofluorescence images acquired in a previously published study (8).

Mouse treatment studies

Long-term treatment studies in transgenic KPC mice were published previously (7). In brief, KPC mice with PDAC of ~4–5 mm confirmed by palpation and ultrasound imaging were randomized into 3 arms: iRGD alone (4 μmol/kg IV injection, n = 10); Gem (TSZ Chem, Waltham, MA) [100 mg/kg/intraperitoneal (IP) injection, n = 18]; and iRGD + Gem (n = 16). The treatment was given every 4 days until the mice met the criteria for sacrifice following IACUC guidelines. Tumor samples collected in the study were subjected to immunofluorescence and immunohistochemistry in this project. A PBS control arm was not included in this original survival study and samples from historical PBS-treated cohorts were stained for comparison.

The treatment studies in orthotopic syngeneic PDAC mice were performed as follows. C57B6129SF1/J hybrid mice were orthotopically implanted with 5.0 x 10⁵ KPC78 PDAC cells and randomized into paired treatment cohorts for 6 separate comparisons: IV phosphate buffered saline (PBS) alone versus IV iRGD alone (12 μmol/kg); IP PBS versus IP anti-mouse programmed cell death ligand-1 (PD-L1) monoclonal Ab (mAb) (200 μg/mouse; Ultra-LEAF™ purified anti-mouse PD-L1 Ab, clone 10F.9G2; Cat# 124339; BioLegend, San Diego, CA); IV PBS + IP PBS versus IV iRGD + IP anti-mouse PD-L1 mAb; IV iRGD + IV Gem (4 mg/kg; Cat# G6423; MilliporeSigma, Burlington, MA) versus IV iRGD + IV GEM + IP anti-mouse PD-L1 mAb; IV PBS versus IV CRGDC alone (12 μmol/kg); and IV PBS versus IV CRGDK alone (12 μmol/kg). A control IgG for the anti-mouse PD-L1 mAb was not used in the treatment studies given the negligible effects shown in historic studies (e.g., ref. 28). The treatment was given 3 times a week for 2 weeks starting 8 days after tumor cell implantation. The treatment studies were terminated according to the guidelines by the IACUC at Columbia University. In some cases, IV injection of DyLight 488-labeled tomato lectin (2 mg/kg; Cat# L32470; Invitrogen, Waltham MA) was given at the end of the treatment study prior to sacrificing the mice. At the end of the studies, tumors and major organs were harvested, weighed, and processed for flow cytometry, immunofluorescence, and immunohistochemistry.

Peptide binding to immune cells isolated from the PDAC tissue

Tumors from orthotopic PDAC mice or human PDAC tissues were minced into 2–4 mm pieces and added with an enzyme mix from a tumor dissociation kit (Cat# 130-096-730 and 130-095-929: Miltenyi Biotec, Bergisch Gladbach, Germany) in Roswell Park Memorial Institute medium 1640 (Cytiva, Marlborough, MA). The tissues were dissociated using a gentle MACS Dissociator (Miltenyi Biotec) followed by a 40-minute incubation at 37 °C. The samples were then centrifuged for 10 minutes at 3,000 x g in the presence of debris removal solution (Cat# 130-109-398; Miltenyi Biotec) and 10 minutes at 1,000 x g in cooled PBS. The supernatant was aspirated to remove the debris. Splenocytes were prepared by gently grinding the spleen and filtrating the resulting cell suspension through a 40 μm cell strainer. Red cells were lysed with an ACK lysing buffer (Cat# A10492-01; Gibco, Waltham, MA). The isolated immune cells were subjected to flow cytometry for quality check and to analyze the cell populations of interest. The isolated cells were incubated with 10 μM FAM-iRGD or FAM-CRGDC for 1 hour at 37 °C in a binding buffer containing 1.0 % BSA, 150 mM sodium, 1.0 mM magnesium, and 1.0 mM calcium. Peptide binding to T cells was analyzed by flow cytometry using Abs listed elsewhere.

In vitro co-culture of CD4⁺ T cells and KPC-derived PDAC cells

CD4⁺ T cells were magnetically isolated from the spleens of healthy C57B6129SF1/J mice using the CD4⁺ T Cell Isolation Kit for mouse cell (Cat# 130-104-454; Miltenyi Biotec). The CD4⁺ T cells were cultured for 3 days with or without KPC78 cells in the presence of low dose anti-CD3/CD28 beads (x40 dilution; Cat# 130-095-925; Miltenyi Biotec), recombinant mouse TGF-β1 (5 ng/ml; Cat# 7666-MB; R&D Systems, Minneapolis, MN), and recombinant mouse interleukin-2 (mIL-2, 100 U/ml; Cat# 11271164001; Roche, Basel, Switzerland), and subjected to flow cytometry as described elsewhere. To test FAM-iRGD binding to the T cells, CD4⁺ T cells were incubated with 10 μM FAM-iRGD for 1 hour at 37 °C in a binding buffer. Three micromolar of anti-mouse αvβ5 integrin blocking Ab or an isotype control was added 1 hour prior to FAM-iRGD. FAM-iRGD binding to T cells was analyzed by flow cytometry as described above.

In vitro induction of αvβ5 integrin⁺ Tregs from naïve CD4⁺ T cells and nTregs

Mouse naïve CD4⁺ T cells and nTregs were isolated from the spleens of healthy C57B6129SF1/J mice by magnetic separation using a mouse Naïve CD4⁺ T cell isolation kit for mouse (Cat# 130-104-453; Miltenyi Biotec). The T cells were stimulated with anti-CD3/CD28 beads (x25 dilution; Miltenyi Biotec) in the presence or absence of recombinant mouse TGF-β1 (5 ng/ml) for 3 days. iRGD (1-100 μM), iRGE (1-100 μM), CRGDC (100 μM), RPARPAR-NH₂ (100 μM), or RPARPAR-OH (100 μM) was added to the cultures in some experiments. To study the effect of TGF-β1 on the induction of αvβ5 integrin⁺ iTregs, 10 μM TGF-βR1 inhibitor (LY2157299; Cat# S2230; Selleck chemicals LLC, Houston, TX) was added during the stimulation process. In some cases, naïve CD4⁺ T cells were treated with plate-coated anti-mouse CD3 Ab (3 μg/ml; clone 17A2; Cat# 16-0032-82, RRID: AB_468851; eBioscience, San Diego, CA) in the presence of soluble anti-mouse CD28 Ab (2 μg/ml) (clone 37.51; Cat# 16–0281-82, RRID: AB_468921; eBioscience) and TGF-β1 (5 ng/ml). The cells were then subjected to flow cytometry to study the expression of αvβ5 integrin in association with the T cell markers of interest, and in some cases, apoptosis markers. To test the αvβ5 dependency of FAM-iRGD binding to iTregs, the cells were incubated with 100 μM FAM-iRGD for 1 hour at 37 °C in the presence or absence of 1 μM αvβ5 integrin-IN-1 (29) (Cat# HY-145363; MedChemExpress, Monmouth Junction, NJ), and subjected to flow cytometry.

Human naïve CD4⁺ T cells and nTregs were isolated from peripheral blood mononuclear cells by magnetic separation using the Naïve CD4⁺ T Cell Isolation Kit II for human cells (Cat# 130-094-131; Miltenyi Biotec) and human CD25 microbeads (Cat# 130-092-983: Miltenyi Biotec). The T cells were stimulated with 5 μg/ml of plate-coated anti-human CD3 Ab (clone OKT3; Cat# 317326, RRID: AB_11150592; BioLegend) in the presence of 2 μg/ml of soluble anti-human CD28 Ab (clone CD28.2; Cat# 302934, RRID: AB_11148949; BioLegend) and 5 ng/ml of recombinant human TGF-β1 (Cat# 240-B; R&D Systems) for 3 days. Recombinant human interleukin-2 (1 U/ml; Cat# 202-IL; R&D Systems) was also added to the culture. The cells were subjected to flow cytometry as described elsewhere to analyze the expression of αvβ5 integrin in T cells.

Immunosuppression assay

Naïve CD4⁺ T cells isolated from the spleens of C57B6129SF1/J mice were treated with anti-CD3/CD28 beads and TGF-β1 for 3 days as described above. mIL-2 (10 ng/ml) was added to the culture medium from day 1 to expand iTregs. After removing dead cells with a Dead Cell Removal kit (Cat# 130-090-101; Miltenyi Biotec), the cells were treated with a PE-conjugated anti-CD25 Ab (clone 7D4; Cat# 130-118-550, RRID: AB_2784088; Miltenyi Biotec), PE-conjugated anti-mouse CD198 (CCR8) Ab (clone SA214G2; Cat# 150302, RRID: AB_2566246; BioLegend), and/or Alexa Fluor^® 647-conjugated anti-β5 integrin Ab combined with an anti-PE or -AF647 magnetic separation technique (Miltenyi Biotec) to enrich for CCR8⁺ iTregs or αvβ5⁺CCR8⁺ iTregs. Responder CD4⁺CD25^- T cells and CD8⁺ T cells (Tconv) were isolated from the spleen of healthy C57B6129SF1/J mice by magnetic separation and labeled with CellTrace™ Violet (CTV; Cat# C34557; Thermo Fisher Scientific). The Tconv and iTregs were mixed at a 4 : 1 ratio and co-cultured in the presence of anti-CD3/CD28 beads (x25 dilution). Division of the Tconv was assessed by measuring the dilution of CTV by flow cytometry on day 3.

Flow cytometry

Immune cells isolated from the tumor or spleen, or cells cultured in vitro were subjected to surface marker staining with the following reagents for 20 minutes at 4 °C: Brilliant Violet 650™-conjugated anti-mouse CD3 Ab (clone 17A2; Cat# 100229, RRID: AB_11204249; BioLegend), Brilliant Violet 650™-conjugated anti-human CD3 Ab (clone OKT3; Cat# 317324, RRID: AB_2563352; BioLegend), PerCP/Cyanine5.5-conjugated anti-mouse CD4 Ab (clone GK1.5; Cat# 100434, RRID: AB_893324; BioLegend), Brilliant Violet 421™-conjugated anti-human CD4 Ab (clone RPA-T4; Cat# 300532, RRID: AB_10965645; BioLegend), PE/Cyanine7-conjugated anti-mouse CD8a Ab (clone 53-6.7; Cat# 100722, RRID: AB_312761; BioLegend), PE-conjugated anti-mouse CD25 Ab (Miltenyi Biotec), PE/Cyanine7-conjugated anti-mouse CD25 Ab (clone PC61; Cat# 102016, RRID: AB_312865; BioLegend), PE/Cyanine7-conjugated anti-human CD25 Ab (clone BC96; Cat# 302611, RRID: AB_314281; BioLegend), Brilliant Violet 421™-conjugated (clone 3E12, Cat# 145209) or 711™-conjugated anti-mouse CD304/NRP-1 Ab (clone V46-1954, Cat# 752456, RRID: AB_2917453; BioLegend), Alexa Fluor^® 647-conjugated anti-mouse integrin αvβ5 Ab, PE-conjugated anti-mouse CD198 (CCR8) Ab, Brilliant Violet 711-conjugated anti-mouse CD51 Ab (clone RMV-7; Cat# 740755, RRID: AB_2870648; BD Biosciences), or Aqua™ Fixable Viability Kit (Cat# 423101; BioLegend). Intracellular staining of immune cells was performed using the following Abs and a Foxp3 staining buffer kit (Cat# 00-5523-00; eBioscience) according to the manufacturer’s instructions: Brilliant Violet 421™-conjugated anti-mouse/Rat Foxp3 Ab (FJK-16s; 404-5773-82, RRID: AB_2925536; eBioscience), Pacific blue™-conjugated anti-mouse Foxp3 Ab (clone MF-14; Cat# 126410, RRID: AB_2105047; BioLegend), Alexa Fluor ™ 488-conjugated anti-human Foxp3 Ab (236A/E7; Cat# 53-4777-42, RRID:AB_10804652; Invitrogen), cleaved caspase-3 (Asp175) (5A1E) rabbit mAb (Cat# 9664S, RRID: AB_2070042; Cell Signaling, Danvers, MA), and Alexa Fluor™ 488-conjugated donkey anti-rabbit secondary Ab (Cat# A21206, RRID: AB_2535792; Invitrogen). Tregs were defined as CD4⁺CD25⁺ or CD4⁺Foxp3⁺ depending on the experimental design as explained in the main text. PD-L1 expression on KPC cells was assessed with an allophycocyanin-conjugated anti-mouse PD-L1 Ab (clone 10F.9G2; Cat# 124312, RRID: AB_10612741; BioLegend). Flow cytometry was performed using a Cytek Aurora (RRID: SCR_019826, Cytek, Fremont, CA) or BD LSRFortesa™ (RRID:SCR_001456, BD Biosciences), and the data were analyzed with FCS Express version 7.06.0015 (RRID: SCR_016431, De Novo Software, Pasadena, CA) or FlowJo™ v10.8 (RRID: SCR_008520, BD Biosciences).

Immunofluorescence

Mouse and human samples were fixed with 4% paraformaldehyde (PFA) overnight, washed with PBS three times, and transferred to 30% sucrose at 4°C until the tissues sank. The tissues were then embedded in optimal cutting temperature (OCT) compound to prepare 10 μm frozen sections. The sections were stained with anti-CD3 Ab (clone CD3-12; Cat# ab11089, RRID: AB_2889189; Abcam, Cambridge, UK), anti-mouse CD4 Ab (clone GK1.5; Cat# MA1-146, RRID: AB_2536856; Invitrogen), fluorescein-conjugated anti-mouse CD8 Ab (clone YTS 105.18; Cat# ab00171-2.0; Absolute Antibody, Boston MA), anti-human/mouse Foxp3 Ab (clone 1054c; Cat# MAB8214, RRID: AB_2929004; Novus Biologicals, Centennial, CO), anti-mouse NRP-1 Ab (Cat# AF566, RRID:AB_355445; R&D Systems), anti-integrin αvβ5 Ab (clone P1F6; Cat# MAB1961, RRID: AB_94464; Sigma-Aldrich), Alexa Fluor^® 647-conjugated anti-mouse integrin αvβ5 Ab, Ultra-LEAF™ purified anti-mouse PD-L1 Ab, anti-mouse CD31 Ab (clone MEC 13.3; Cat# 550274, RRID: AB_393571; BD Biosciences), anti-ER-TR7 Ab (Cat# sc-73355, RRID: AB_1122890; Santa Cruz Biotechnology, Dallas, TX), 4’,6-diamidino-2-phenylindole (DAPI; Thermo Fisher Scientific) followed by an appropriate secondary Ab with Alexa Fluor 488, 546 or 680 (Invitrogen). Images were taken with an LSM 710 confocal microscope system (RRID:SCR_018063, Zeiss, Oberkochen, Germany). A ZEN 3.0 SR black edition software (RRID:SCR_018163, Zeiss) was used for acquisition and analysis of the images. In some experiments, cells of interest were counted in five randomly chosen high-power fields (HPFs) per section to obtain an average number. CD31 and ER-TR7 positive areas were quantified using an ImageJ software version 1.54 (RRID: SCR_003070).

Immunohistochemistry

Mouse samples were fixed with 4% PFA overnight, washed with PBS three times, and transferred to 70% ethanol. The tissues were then embedded in paraffin, and processed to prepare 4 μm paraffin sections. The sections were deparaffinized and endogenous peroxidase activity was blocked with 0.3% hydrogen peroxidase. After treating them with IHC Antigen Retrieval Solution (pH 6 or 9; Cat# 00-4955 and 00-4956; Invitrogen), blocking was performed with 1% BSA. The sections were stained with phospho-SMAD2 rabbit mAb (clone; 138D4; Cat# 3108, RRID: AB_490941; Cell signaling, Danvers, MA), anti-mouse CD8 Ab (clone EPR21769; Cat# ab217344; Abcam), anti-mouse CD31 Ab (clone MEC 13.3; Cat# 550274, RRID: AB_2890649; BD Pharmingen, Franklin Lakes, NJ), or GATA6 rabbit mAb (clone D61E4; Cat# 5851, RRID: AB_10705521; Cell signaling) at 4 °C overnight. After incubation with biotin-conjugated anti-mouse or anti-rabbit IgG Ab for 1 h at room temperature, the sections were treated with Ultra-Sensitive ABC Peroxidase Staining Kit (Cat# 32050; Thermo Scientific), and colored with Vina Green Chromogen Kit (Cat# BRR807AS; Biocare Medical, Pacheco, CA) or DAB Substrate Kit (Cat# SK-4100; Vector Laboratories, Newark, CA). Histofine Simple stain Mouse Max PO (rat) (Cat# 414311F; Nichirei Biosciences, Tokyo, Japan) was used as a secondary Ab for CD31 staining. Picrosirius Red Stain Kit (Cat# 24901; Polysciences, Warrington, PA) was used to observe collagen arrangement, according to the manufacturer’s protocol. Briefly, deparaffinized sections were stained with picrosirius red for 60 minutes and washed in acidified water. The sections were counter stained with hematoxylin. The slides were imaged with a BZ-X800 fluorescence microscope (RRID: SCR_023617, Keyence, Osaka, Japan). Collagen-positive areas were quantified using an ImageJ software. CD31 staining sections were scanned to digital images via an Olympus VS200 Slide Scanner Microscope (RRID:SCR_024783) at 10x magnification with resolution of 0.55 μm/pixel [Olympus PlanXApo 10x/0.40, WD 3.1 mm (Air)]. QuPath software, version 0.2.0 or later (30), was used for analysis, and the area of interest was chosen in a blinded fashion with QuPath brush tool. Blood vessel diameters were measured with QuPath Line tool drawing from the outer edge of vessel wall to another as described previously (31). Necrotic area in tumors was quantified by analyzing hematoxylin and eosin (H&E)-stained sections using an ImageJ software. To quantify CD8⁺ T cells, periphery and core regions were marked by a clinical pathologist following a pathological criterion based on stromal content. Five random non-necrotic fields were chosen for each region and CD8⁺ cells were counted in each field under a microscope. The counts/field were combined for each region to obtain a CD8 count that represented the CD8 burden in the periphery or the core.

Immunoblot analysis

KPC78 cells or M21 cell lines were incubated in the presence or absence of recombinant mouse TGF-β1 or LAP-TGF-β1 (Cat# ab271759; Abcam) for either 3 hours or 24 hours at 37 °C. iRGD (100 μM), iRGE (100 μM), CRGDK (100 μM) or a TGF-βR1 inhibitor (LY2157299, 10 μM) was added to the cultures in some experiments. The cells were lysed on ice for 5 minutes in buffer containing 50 mM Tris-HCl, 2 mM EDTA, 150 mM NaCl, 1% Triton X-100, 1.0% Sodium deoxycholate, 0.1% sodium dodecyl sulfate, and a protease and phosphatase inhibitor cocktail (Cat# 78442; Thermo Scientific Fisher). The lysate was fractionated by SDS–PAGE and transferred electrophoretically onto a polyvinylidene difluoride membrane. The membrane was blocked with 2% non-fat milk and incubated with phospho-SMAD2 rabbit mAb, SMAD2 mouse mAb (clone L16D3; Cat# 3103, RRID: AB_490816; Cell signaling), Anti-Glyceraldehyde-3-Phosphate Dehydrogenase Ab (clone 6C5; Cat# MAB374, RRID: AB_2107445; Sigma-Aldrich) overnight at 4 °C. Blots were incubated with a peroxidase-conjugated anti-mouse IgG (Cat# 115-035-003, RRID: AB_10015289; Jackson ImmunoResearch, West Grove, PA) or anti-rabbit IgG (Cat# 111-035-003, RRID: AB_2313567; Jackson ImmunoResearch), and developed with the enhanced chemiluminescent substrate (Cat# 34096; Thermo Scientific).

Analysis of TGF-β1 activation using M21L cells

KPC78 or hPCF1424 cells were incubated with LAP-TGF-β1 for 1 hour at 37 °C. iRGD (100 μM) was added to the cultures in some experiments. For the immunoblot analysis, CM prepared from the cultures was added to M21L cells and the cells were cultured for 3 hours at 37 °C. The cells were then subjected to immunoblot as described elsewhere to analyze the phosphorylation of Smad2. For the reporter gene assay, CM was added to M21L/SBE-luc cells, which had been prepared by infecting M21L cells with the SBE Luciferase Reporter Lentivirus (Cat# 79806; BPS Bioscience, San Diego, CA). After 6 hours of culture at 37°C, the cells were lysed, mixed with Luciferase Assay Reagent (Cat# E1500; Promega), and analyzed for luminescence intensity using Varioskan LUX (RRID: SCR_026792; Thermo Fisher Scientific).

Quantitative RT-PCR

KPC78 cells were incubated in the presence or absence of LAP-TGF-β1 with or without iRGD (1-100 μM) for 3 hours at 37 °C. Total RNA extraction was carried out using RNeasy Kit (Cat# 74104; Qiagen, Hilden, Germany) according to the manufacturer’s instructions. Total RNA (4 μg) was reverse-transcribed into cDNA using Superscript III RNase H Reverse Transcriptase (Cat# 18080-044; Invitrogen) and 500 ng Oligo-d(T) primers. Each quantitative real-time RT-PCR reaction was performed using QuantStudio 5 (RRID:SCR_020240, Thermo Fisher Scientific). For a 10 μl PCR reaction, 2 μl of cDNA template mixed with the appropriate primers to a final concentration of 200 nM was combined with 5 μl SYBR Green Master Mix (Cat# A27542; Applied Biosystems). The reaction took place at 95 °C for 4 minutes, followed by 40 cycles at 95 °C for 15 seconds, and 60 °C for 1 minute. The primers were as follows: Sox4 FW, 5’-CCCACCCTTCCAACGAGCTTC-3’ and Sox4 RW, 5’-CCCTCCTTTCTTGACCATGAGGC-3’; Pdgfb FW, 5’-AATGCTGAGCGACCACTCCATC-3’ and Pdgfb RW, 5’-TCGGGTCATGTTCAAGTCCAGC-3’; Zeb1 FW, 5’-ATTCAGCTACTGTGAGCCCTGC-3’ and Zeb1 RW, 5’-CATTCTGGTCCTCCACAGTGGA-3’; Snai1 FW, 5’-TGTCTGCACGACCTGTGGAAAG-3’ and Snai1 RW, 5’-CTTCACATCCGAGTGGGTTTGG-3’; Gapdh FW, 5’-TGGTGAAGGTCGGTGTGAAC-3’ and Gapdh RW, 5’-AATGAAGGGGTCGTTGATGG-3’.

RNA-seq

Total RNA was extracted from the samples using miRNeasy kits (Cat# 74104; Qiagen). RNA integrity was assessed with an Agilent Bioanalyzer using an RNA Pico kit (Cat# 5067-1513; Agilent Technologies, Santa Clara). For STRDPOLYA, a poly-A pull-down was used to enrich mRNA from total RNA samples, and cDNA libraries were constructed using Illumina TruSeq chemistry. Libraries were fragmented, ligated to adapters, and the final PCR reaction was modified by replacing the Illumina TruSeq PCR reaction mix with KAPA HiFi HotStart Ready Mix. Libraries were sequenced using an Element Aviti (Element Biosciences, San Diego, CA) at the Columbia Genome Center. Multiplexing was applied to each lane to achieve the target number of 75 bp paired-end reads per sample.

Statistical analysis

The unpaired, two-tailed Student’s t-test, Mann-Whitney U test, and Welch’s test were used to compare two groups. One-way analysis of variance (ANOVA) was used to compare three or more groups with a normal distribution. All statistics were performed using GraphPad Prism Ver. 8.4.3 (RRID:SCR_002798).

RESULTS

A subset of PDAC-infiltrating mouse Tregs expresses the αvβ5 integrin

We previously showed that the iRGD peptide targets various cell types in PDAC that express αv integrins and NRP-1, such as blood vessel ECs, cancer epithelial cells, and CAFs (7,8). There, we often noticed that fluorescein (FAM)-labeled iRGD also entered small mononuclear cells that morphologically resembled lymphocytes (Supplementary Fig. S1; also refer to Fig. 6 in ref. 8). We hypothesized that the cells were Tregs because NRP-1, one of the iRGD receptors, is abundantly expressed on Treg surfaces as an important regulatory molecule, which might have allowed iRGD to target this cell population (32,33). Supporting this hypothesis, FAM-iRGD homed to Foxp3⁺ cells in orthotopic PDAC prepared in B6129SF1/J hybrid mice using KPC78 PDAC cells derived from transgenic Kras^G12D/+;LSL-Trp53^R172H/+;Pdx-1-Cre (KPC) mice with the same genetic background (Fig. 1A).

Fig. 1: iRGD targets tumor-infiltrating Tregs in mouse PDAC.

A, Representative confocal images of CD4⁺Foxp3⁺ Tregs (magenta, CD4; red, Foxp3) in the tumor of KPC78 orthotopic PDAC mice that received IV FAM-iRGD (green). Blue, DAPI. Arrows indicate Tregs positive for iRGD. Scale bars, 50 μm. B, Representative confocal images of CD4⁺Foxp3⁺ Tregs (Green, CD4; red, Foxp3) expressing the αvβ5 integrin or NRP-1 (magenta) in KPC78 orthotopic PDAC in mice. Blue, DAPI. Arrows, αvβ5 integrin⁺ or NRP-1⁺ Tregs. Scale bars, 20 μm. C, A representative flow cytometry analysis showing the proportion of CD4⁺CD25⁻ T cells (non-Tregs) and CD4⁺CD25⁺ T cells (Tregs) that express αvβ5 integrin in the spleen or KPC78 orthotopic PDAC in mice. The bar diagram summarizes the findings from 5 biological replicates per group. D, The proportion of αvβ5 integrin-positive cells among CD4⁺Foxp3⁺ Tregs in mouse KPC78 orthotopic PDAC analyzed by flow cytometry. Four biological replicates per group were analyzed. E, In vitro binding of FAM-iRGD to CD4⁺CD25⁻ T cells (non-Treg, blue line) and CD4⁺CD25⁺ T cells (Tregs, red line) isolated from the tumors and spleens of KPC78 orthotopic PDAC mice. The bar diagram summarizes the mean fluorescence intensity (MFI) from 6 biological replicates each. Gray shades, unstained control. Statistical analysis; one-way ANOVA (C) Welch’s test (D), or Mann-Whitney U test (E). Error bars, mean ± standard error.

The result suggested that the Tregs also expressed αv integrins because they are required as a primary receptor for iRGD to transform into an NRP-1-binding form (8). Immunofluorescence identified the expression of αvβ5 integrin in addition to NRP-1 on CD4⁺Foxp3⁺ Tregs in KPC78 tumors (Fig. 1B). The integrin and NRP-1 were also expressed in the remaining tumor tissue consistent with their known abundance in cancer epithelial cells and stromal cells (7,34,35). Flow cytometry showed αvβ5 integrin expression on about 20% of CD4⁺CD25⁺ T cells, a population known to be largely composed of Tregs (32) (Fig. 1C and Supplementary Fig. S2A). In contrast, the integrin was expressed much less on CD4⁺CD25⁻ T cells isolated from the tumor, and rarely on CD4⁺CD25⁺ T cells from the spleen. PDAC-infiltrating CD8⁺ T cells lacked αvβ5 expression. Another set of flow cytometry revealed that more than 80% of splenic and tumor-infiltrating CD4⁺CD25⁺ T cells expressed Foxp3 confirming that the T cells were mostly Tregs (Supplementary Fig. S2B), and that the Tregs in the tumor had a significantly higher level of αvβ5 expression than those in the spleen (Fig. 1D). NRP-1 was expressed on both PDAC and splenic Tregs as expected, but not on CD8⁺ T cells (32) (Supplementary Fig. S2C). FAM-iRGD bound more effectively to Tregs than to non-Tregs isolated from KPC78 tumors consistent with the αvβ5 expression profile (Fig. 1E). The binding to splenic Tregs and non-Tregs was negligible. FAM-CRGDC, a traditional RGD peptide without an NRP-1-binding motif (8), showed a similar binding profile as FAM-iRGD, but with a clearer biphasic pattern suggesting its binding to the αvβ5⁺ cells (Supplementary Fig. S3).

αvβ5 integrin is expressed on mouse Tregs after T-cell receptor stimulation

Tumor-infiltrating Tregs are believed to be a mixture of induced Tregs (iTregs) that differentiate in the periphery from naïve CD4⁺ T cells and natural Tregs (nTregs) that develop in the thymus (36,37). To study the mode of expression of αvβ5 on Tregs, we first prepared iTregs from mouse naïve CD4⁺ T cells by TGF-β1 and T cell receptor (TCR) stimulation, which are essential for iTreg induction (38,39). The starting naïve CD4⁺ T cells did not express αvβ5 (Supplementary Fig. S4A). TCR stimulation alone did not induce iTregs from them or induce αvβ5 expression on the non-Tregs (Supplementary Fig. S4B). TCR stimulation along with TGF-β1 induced CD4⁺Foxp3⁺ iTregs and nearly 30% of them expressed αvβ5 (Supplementary Fig. S4C). FAM-iRGD bound to the iTregs in an αvβ5-dependent manner because the binding was inhibited by an αvβ5-specific small molecule inhibitor (29) (Supplementary Fig. S4D). The αvβ5 expression on the iTregs correlated with that of CD25, a T cell activation marker (40) (Supplementary Fig. S4E), and CCR8, a chemokine receptor that is expressed on immunosuppressive tumor-infiltrating Tregs (41,42) (Supplementary Figs. S5A and B). Most of the αvβ5-positive CD4⁺CD25⁺Foxp3⁺ iTregs expressed CCR8, suggesting that they were a subpopulation of CCR8⁺ iTregs (Supplementary Fig. S5C). Enriching the αvβ5⁺ population among CCR8⁺ iTregs increased their ability to inhibit the proliferation of CD4⁺ and CD8⁺ T cells, suggesting that αvβ5⁺ iTregs are one of the functionally active fractions of CCR8⁺ iTregs (Supplementary Figs. S5D and E).

In nTregs, the expression profiles of αvβ5 and CD25 did not necessarily match. CD4⁺CD25⁺Foxp3⁺ nTregs enriched from mouse spleen lacked αvβ5 despite the expression of CD25 (Supplementary Fig. S6A). Only upon receiving TCR stimulation did the splenic nTregs express the integrin (Supplementary Fig. S6B). CD4⁺CD25⁻Foxp3⁺ Tregs that act as peripheral reservoirs of activated CD4⁺CD25⁺Foxp3⁺ nTregs (39) lacked the integrin. The results suggest that Tregs express αvβ5 upon activation in response to antigens. Supporting this finding, αvβ5⁺ Tregs emerged from splenic CD4⁺ T cells in the presence of co-cultured KPC78 cells (Supplementary Fig. S7A). Binding of FAM-iRGD to the Tregs was inhibited by an anti-αvβ5 integrin Ab (Supplementary Fig. S7B).

αvβ5⁺ Tregs are present in human PDAC

αvβ5⁺ Tregs were also found in human PDAC tissue especially in neighboring areas of cancer ducts (Fig. 2A). Tregs remote from cancer ducts tended to be αvβ5-negative, although this observation was not quantitatively assessed (Supplementary Fig. S8A). αvβ5⁺ Tregs were minimally present in the spleen from matching PDAC patients. Flow cytometry showed compatible results in a set of patient-derived samples from which we could successfully isolate tumor and splenic Tregs (Supplementary Fig. S8B). In vitro assays using naïve human CD4⁺ T cells revealed that approximately 50% of CD4⁺CD25⁺Foxp3⁺ Tregs induced from the CD4⁺ T cells with TGF-β1 and TCR stimulation expressed αvβ5 (Fig. 2B). Human CD4⁺CD25⁺Foxp3⁺ nTregs isolated from the peripheral circulation were negative for αvβ5 (Supplementary Fig. S9). However, 15% of them expressed αvβ5 after TCR stimulation. In addition to Tregs, human PDAC tissues were noted to widely express αvβ5 (Fig. 2C and Supplementary Fig. S10). While the signal intensity varied between the samples, αvβ5 expression was consistently present in cancer ducts and surrounding stromal cells in line with our previous finding that the interaction between CAFs and cancer epithelial cells mediated by TGF-β contributed to the maintenance of αvβ5 expression in the TME (7).

Fig. 2: αvβ5 integrin⁺ Tregs are present in human PDAC.

A, Representative confocal images of Tregs in human PDAC and spleen specimens collected from an identical patient. The dotted lines outline cancer ducts. Red, CD3; white, Foxp3; green, αvβ5 integrin; blue, DAPI. Arrowheads, αvβ5 integrin⁺ Tregs. The boxed area is magnified. Scale bars, 20 μm (large panels); 10 μm (small panels). The bar diagram summarizes the average number of αvβ5⁺ Tregs in the PDAC and spleen from 5 patients. The cells were counted in 5 randomly chosen HPFs per section. B, Flow cytometry analysis of naïve human CD4⁺ T cells before (left) and after (right) a 3-day-treatment with anti-CD3/CD28 Abs and TGF-β1. Note the induction of CD25⁺Foxp3⁺ iTregs (red) that express αvβ5 integrin. CD25⁻Foxp3⁻ non-Tregs (blue) remain negative for the integrin. Three biological replicates per group were analyzed. C, Representative microscopic images of αvβ5 integrin (blue) expressed in one of the five human PDAC specimens analyzed. Red box, with cancer ducts; green box, without cancer ducts. Black arrowhead, cancer duct; white arrowhead, stroma cell. Scale bars, 500 μm (left); 50 μm (middle); 20 μm (right). Statistical analysis, Student’s t test. Error bars, mean ± standard error.

Systemic iRGD therapy reduces Tregs in the tumor but not in the spleen in PDAC mice

The expression of αvβ5 (and NRP-1) on PDAC-infiltrating Tregs suggested that iRGD could direct co-injected therapies toward the Tregs. To this end, we analyzed mouse PDAC tumors collected from transgenic KPC mice that were treated with iRGD with or without Gem for up to 40 weeks in a prior study (7). The tissue samples were randomly chosen from mice that underwent varying lengths of treatment (Supplementary Fig. S11A). Foxp3⁺ cells were greatly decreased in PDAC tissue exposed to iRGD + Gem therapy (Fig. 3A). Surprisingly, tumors from mice that received iRGD monotherapy for an average of 8 weeks also had reduced Foxp3⁺ cells. This finding suggested an intrinsic function of iRGD to deplete these cells. Analysis of CD8⁺ T cells in randomly picked PDAC samples from the same study revealed a higher number of the cells in the iRGD + Gem and iRGD alone arms than in the Gem alone arm (Supplementary Figs. S11B and C).

Fig. 3: iRGD reduces Tregs in PDAC.

A, Representative confocal images of Foxp3⁺ cells (magenta, arrowheads) in the PDAC of transgenic KPC mice that underwent long-term treatment with Gem, iRGD, or iRGD + Gem (7). PBS-treated samples were from a historical cohort. Blue, DAPI; scale bars, 25 μm. The bar diagram summarizes the number of Foxp3⁺ cells per mm² in 3 biological replicates per group. B-E, C57B6129SF1/J hybrid mice bearing KPC78 orthotopic PDAC were treated with IV PBS or iRGD (12 μmol/kg) 3 times a week for 2 weeks. Tumors and spleens were collected on days 3, 7 and 14, and subjected to flow cytometry and immunofluorescence. A schematic of the study is shown in B. The proportion of CD4⁺CD25⁺ Tregs among CD4+ T cells in the tumor (C, left panel) or the spleen (E, left panel), and the CD8/Treg ratio in the tumor (C, right panel) or the spleen (E, right panel) were analyzed by flow cytometry. n = 3 (day 3, PBS), 4 (day 3, iRGD; day 7) or 5 (day 14) of biological replicates per group were analyzed. In D, representative confocal images of CD8⁺ (green) and Foxp3⁺ (red) cells in day 14-tumors and the CD8/Foxp3 ratio quantified by counting the cells in 5 random HPFs per section for 3 individual mice per group are shown. Blue, DAPI. Scale bars, 50 μm. Statistical analysis, one-way ANOVA (A), Mann-Whitney U test (C and E), or Welch’s t test (D). Error bars, mean ± standard error. The schematic was created in BioRender. MIYAMURA, N. (2026) https://BioRender.com/pbnlqhs.

To study the intrinsic effect of iRGD on tumor-infiltrating Tregs, we treated syngeneic PDAC mice bearing orthotopic KPC78 tumors with iRGD monotherapy and analyzed tumor-infiltrating Tregs in a time-dependent fashion by flow cytometry (Fig. 3B). Treatment for 7 days significantly decreased Tregs in the tumor (Fig. 3C). CD8⁺ T cells increased longitudinally, but the change did not reach significance within this limited time frame (Supplementary Fig. S12A). These changes led to a significant increase in the CD8/Treg ratio. Immunofluorescence revealed similar changes (Fig. 3D and Supplementary Fig. S12B). In contrast, iRGD minimally affected Tregs or the CD8/Treg ratio in the spleen, correlating with the lack of αvβ5 expression on splenic Tregs (Fig. 3E). In line with previous studies (7,8,11,34,43), iRGD monotherapy did not affect the tumor growth (Supplementary Fig. S12C). Both the integrin-binding RGD motif and the NRP-1-binding RXXK motif were necessary for iRGD to deplete Tregs in vivo because systemic treatment with neither CRGDC, which lacks an RXXK motif, nor CRGDK, the NRP-1-binding fragment of iRGD with negligible integrin affinity (8), affected tumor-infiltrating Tregs (Supplementary Fig. S13).

iRGD depletes both αvβ5⁺ and αvβ5⁻ iTregs in vitro

αvβ5-dependent killing was a possible cause of the Treg depletion because αvβ5⁺ Tregs in the tumors showed a trend toward reduction following iRGD treatment (Supplementary Fig. S14A). αvβ5⁺NRP-1⁻ Tregs exhibited a downward trend in iRGD-treated tumors, whereas αvβ5⁻NRP-1⁺ Tregs were not affected (Supplementary Fig. S14B). However, interestingly, the minor non-Treg population that expressed αvβ5 was not affected (Supplementary Fig. S14C). In vitro studies also led to mixed results that could not be explained by αvβ5-dependent killing. iRGD reduced the proportion of overall iTregs, and increased apoptosis of the iTregs (Supplementary Figs. S15A and B). However, the proapoptotic effect was somewhat limited, and was only observed with a higher dose of iRGD. Among the iTregs, the αvβ5⁺ population was significantly reduced and exhibited increased apoptosis (Supplementary Fig. S15C). However, αvβ5⁻ iTregs were also affected in a similar manner (Supplementary Fig. S15D). Puzzlingly, the RGD motif was still necessary for the iRGD effect because CRGDC significantly reduced cultured iTregs (Supplementary Fig. S15E). In contrast, NRP-1 binding was not essential for these in vitro effects because RPARPAR-OH, an NRP-1-binding peptide that lacks affinity to integrins (10), had no effect on the iTregs (Supplementary Fig. S15F).

iRGD inhibits the activation of TGF-β in an RGD-dependent manner

TGF-β is essential for the development and maintenance of Tregs in general (44). In fact, the TGF-β receptor type 1 (TGF-βR1) kinase inhibitor LY2157299 (7) depleted cultured iTregs without selectively affecting the αvβ5⁺ population (Supplementary Fig. S15G). We speculated that iRGD depleted Tregs through a similar mechanism. RGD-binding integrins mechanically activate TGF-β, and RGD peptides have been shown to inhibit this process (14,15). We therefore hypothesized that iRGD inhibited integrin-dependent activation of TGF-β in an RGD-dependent manner and that the Treg depletion in vitro and in vivo was one of the outcomes of TGF-β deprivation. While αvβ5⁺ Tregs were suspected to be the activators of TGF-β in the in vitro studies, they are a minor population at tissue level. As αvβ5-expressing cancer epithelial cells and CAFs are more abundantly available in the PDAC tissue as potential activators of TGF-β, we used KPC78 PDAC cells to test our hypothesis.

Culturing the KPC78 cells for 3 hours in the presence of exogenous latent TGF-β1 (LAP-TGF-β1) led to a detectable increase in phosphorylated Smad2 (p-Smad2), a primary mediator of the canonical TGF-β signaling pathway (Supplementary Fig. S16A). The results indicate that the cells were able to activate TGF-β and mediate downstream signaling. Quantitative PCR (qPCR) showed corresponding increase in mRNA expression of Sox4, one of the TGF-β target genes (Supplementary Fig. S16B). iRGD inhibited the Smad2 phosphorylation induced by exogenous LAP-TGF-β1 (Fig. 4A). An iRGD variant with a disrupted RGD motif (iRGE: CRGEKGPDC) (8) was less potent suggesting RGD-dependency. CRGDK had no effect, suggesting that NRP-1 binding was not required for iRGD to inhibit the signaling (Supplementary Fig. S16C). qPCR revealed that iRGD, in a dose-dependent manner, suppressed the expression of downstream target genes induced by LAP-TGF-β1, such as Pdgfb, Zeb1, and Snai1 (Fig. 4B). iRGD did not reduce p-Smad2 when an active form of TGF-β1 was exogenously added to the culture supporting our hypothesis that iRGD inhibits the activation rather than the function of TGF-β (Fig. 4C). When the KPC78 cells were cultured for 24 hours, p-Smad2 became detectable even in the absence of exogenous LAP-TGF-β1, indicating the bioactivity of endogenous TGF-β (Supplementary Fig. S17A). iRGD reduced p-Smad2 under this condition, and even when active TGF-β1 was added, reflecting the effect of iRGD on accumulating endogenous TGF-β. The effect of LY2157299 was more pronounced consistent with its capacity to directly inhibit TGF-βR1 (Supplementary Fig. S17B).

Fig. 4: iRGD inhibits αvβ5 integrin-dependent activation of TGF-β.

A-C, KPC78 PDAC cells were treated with LAP-TGF-β (A and B) or active TGF-β (C) for 3 hours in the presence or absence of iRGD or iRGE. In A and C, the samples were subjected to immunoblot to analyze (p-)Smad2. Representative immunoblot images and quantified data are shown. In B, qPCR was performed to analyze the expression of TGF-β target genes, Pdgfb, Zeb1, and Snai1. The values were normalized to Gapdh. n = 5 (A), 3-6 (B), or 3 (C) biological replicates per group. D-F, M21L cells were cultured for 3 hours in the presence of CM from KPC78 cells (D and E) or hPCF1424 human PDAC CAFs (F) that were treated with LAP-TGF-β for 3 hours in the presence or absence of iRGD. Immunoblot was performed in D and F. The M21L cells in E carried a TGF-β/Smad responsive luciferase gene and luminescence was used as a read out for TGF-β-dependent Smad activity. n = 5 (D and E) or 3 (F) of biological replicates per group were analyzed. G and H, p-Smad2 analyzed by immunoblot in WT or β5 integrin-KO (β5 KO) KPC78 cells treated with LAP-TGF-β (G) or active TGF-β (H) for 3 hours. n = 4 (G) or 3 (H) of biological replicates per group were analyzed. I, Sox4 expression analyzed by qPCR in WT and β5-KO KPC78 cells treated with LAP-TGF-β. The values were normalized to Gapdh. Three biological replicates were analyzed. Statistical analysis, one-way ANOVA (A and B); Student’s t test (C-I). Error bars, mean ± standard error.

To further study the effect of iRGD on TGF-β activation, we developed an assay using M21L human melanoma cells, which detects the conversion of LAP-TGF-β1 into active TGF-β (Supplementary Fig. S18A). M21L is a variant of αv integrin-rich M21 cells that spontaneously lost surface αv expression (8,45). Thus, M21L cells are no longer able to activate LAP-TGF-β1 but remain responsive to activated TGF-β1 evidenced by the differential Smad2 phosphorylation in immunoblot analyses (Supplementary Fig. S18B). The findings were confirmed using M21L cells that express a TGF-β/Smad responsive luciferase gene [M21L/SMAD Binding Element (SBE)-luc cells; Supplementary Figs. S18C and D]. Immunoblot revealed that conditioned media (CM) prepared from LAP-TGF-β1-treated KPC78 cells induced p-Smad2 in M21L cells indicating the presence of activated TGF-β1 in the CM (Fig. 4D). Adding iRGD to the KPC78 culture during the LAP-TGF-β1 treatment reduced the ability of the CM to induce p-Smad2, suggesting that iRGD inhibited the conversion of LAP-TGF-β1 to active TGF-β1. Reporter assays using the M21L/SBE-luc cells confirmed the findings (Fig. 4E). Similar results were obtained using hPCF1424 human PDAC CAFs that express αvβ3 and αvβ5 integrins (7) (Fig. 4F and Supplementary Fig. S18E).

The αvβ5 integrin is a major activator of TGF-β in PDAC cells

RNA-seq suggested that αvβ5 was the dominant αv integrin expressed on KPC78 cells (Supplementary Fig. S19A). We therefore prepared β5 integrin-knock out (KO) KPC78 cells to study the role of αvβ5 in TGF-β activation (Supplementary Figs. S19B and C). β5-KO cells showed a dramatically reduced response to LAP-TGF-β1 compared to wild type (WT) cells (Fig. 4G). Both WT and β5-KO cells responded equally to active TGF-β1, indicating that the KO did not affect TGF-β signaling (Fig. 4H). qPCR showed that Sox4 expression remained undetectable in β5-KO cells after LAP-TGF-β1 treatment (Fig. 4I). These results indicated that αvβ5 was a major activator of TGF-β in PDAC.

iRGD modifies the PDAC tumor microenvironment

To investigate the anti-TGF-β properties of iRGD in vivo, we treated syngeneic mice bearing orthotopic KPC78 tumors with IV iRGD alone 3 times a week for 2 weeks, the dosing regimen that reduced PDAC-infiltrating Tregs (Fig. 5A). Consistent with the earlier study, the treatment did not induce a meaningful anti-tumor effect. However, we noted a dramatic inhibition of Smad2 phosphorylation throughout the tumor (Fig. 5B). CRGDC, which does not penetrate tumors due to the lack of NRP-1 affinity (8), did not inhibit Smad2 phosphorylation (Supplementary Fig. S20). Consistent with the reduced TGF-β signaling, iRGD-treated tumors had fewer CD31⁺ blood vessels suggesting reduced angiogenesis (16,46) (Fig. 5C). The remaining blood vessels were more patent (inset, Fig. 5C). Measurements based on immunohistochemistry confirmed a wider lumen (Fig. 5D). Underpinning the morphological change, blood vessels in iRGD-treated tumors were better perfused (Fig. 5E). Stromal fibers detected by an ER-TR7 Ab were greatly reduced in the iRGD-treated tumors (Fig. 5F). Quantification of collagen fibers showed a similar trend (Fig. 5G).

Fig. 5: iRGD inhibits Smad2 phosphorylation and modifies the tumor microenvironment in PDAC mice.

C57B6129SF1/J hybrid mice bearing KPC78 orthotopic PDAC were intravenously treated with PBS or iRGD (12 μmol/kg) 3 times a week for 2 weeks. The tumors were harvested on day 14 and subjected to subsequent analyses. A, A schematic of the treatment regimen and the tumor weight at the end of the study are shown. Each dot in the bar diagram represents the tumor weight in an individual mouse. n = 4 per arm. B, p-Smad2 (blue) in the tumors analyzed by immunohistochemistry. Scale bars, 100 μm (left); 50 μm (middle). n = 4 (PBS) or 3 (iRGD) of biological replicates were analyzed. C, CD31⁺ angiogenic blood vessels (red) in the tumors analyzed by immunofluorescence. Blue, DAPI. Scale bars, 50 μm. Four biological replicates per arm were analyzed. D, CD31⁺ angiogenic blood vessels (brown) in the tumors analyzed by immunohistochemistry. Scale bars, 50 μm. The minor axis was measured using the images. n = 8 (PBS) or 5 (iRGD) biological replicates were analyzed. E, The orthotopic KPC78 mice treated with PBS or iRGD received an IV injection of tomato lectin (green) at the end of the study to identify functionally perfused vessels. Red, CD31; blue, DAPI. The proportion of tomato-lectin+ vessels (arrowhead) was quantified based on manual counts. Ten biological replicates per arm were analyzed. F, ER-TR7⁺ stromal fibers (red) in the tumors analyzed by immunofluorescence. Blue, DAPI. Scale bars, 50 μm. Four biological replicates per arm were analyzed. G, Collagen (red) in the tumors stained with picrosirus red. Scale bars, 100 μm. n = 4 (PBS) or 3 (iRGD) of biological replicates were analyzed. In B-G, representative micrographs and quantified data are shown for each data set. Image-based quantification was performed in 5 randomly selected HPFs and the values were averaged per section. Statistical analysis, Mann-Whitney U test (A), Student’s t test (B-G). Error bars, mean ± standard error. The schematic was created in BioRender. MIYAMURA, N. (2026) https://BioRender.com/pbnlqhs.

iRGD promotes T cell entry into the tumor core and improves the anti-tumor efficacy of an ICB in PDAC mice

Considering the TME changes iRGD induced, we hypothesized that adding an ICB would provide synergistic efficacy because TGF-β inhibitors have been shown to potentiate ICBs (47). Similarly, iRGD conjugated to or co-injected with an ICB has been shown to be efficacious against several different tumor types (48,49). Given the high expression of PD-L1 in our syngeneic PDAC model (Supplementary Fig. S21A), we elected to use an anti-PD-L1 mAb. Consistent with the basal-like nature of the orthotopic KPC78 tumors evidenced by low GATA6 expression (50) (Supplementary Fig. S21B), treatment with an anti-PD-L1 mAb alone did not affect the tumor size (Supplementary Fig. S21C). However, combining the anti-PD-L1 mAb with iRGD led to significantly smaller tumors (Fig. 6A). Here, the anti–PD-L1 mAb was administered intraperitoneally to isolate the TME-modulating effects of iRGD, rather than its role in facilitating antibody delivery through the circulation (11). The anti-PD-L1 mAb also enhanced the anti-tumor effect of iRGD + Gem combination therapy (Supplementary Fig. S21D).

The iRGD + anti-PD-L1 mAb therapy induced wider necrosis in the tumors consistent with its enhanced anti-tumor effect (Supplementary Fig. S22A). A moderate negative correlation was noted between the tumor size and necrotic area suggesting that smaller tumors had a trend to have increased necrosis (Supplementary Fig. S22B). While statistical significance was not reached with the limited number of samples, tumors treated with anti-PD-L1 mAb showed a propensity to harbor more CD8⁺ T cells supporting its ability to recruit T cells to tumors (51) (Figs. 6B and C). Of note was that the proportion of CD8⁺ T cells in the core of the tumors was significantly increased in both the iRGD alone and combination arms indicating that iRGD therapy grants T cells access to the tumor interior (Fig. 6D). The absolute number of CD8⁺ T cells in the tumor core was increased in the combination arm correlating with the wider necrosis in these tumors (Fig. 6E).

Fig. 6: iRGD promotes T cell entry into the tumor core and improves the anti-tumor efficacy of an ICB in PDAC mice.

KPC78 orthotopic PDAC mice were treated with PBS or iRGD (12 μmol/kg) + anti-PD-L1 mAb (200 μg/mouse) 3 times a week for 2 weeks. The tumors were harvested on day 14 and subjected to subsequent analyses. A, A schematic of the treatment regimen and the tumor weight at the end of the study are shown. Each dot in the bar diagram represents the tumor weight in an individual mouse. Representative tumor images are shown to the right. Scale bars, 5 mm. Four or 5 biological replicates per arm were analyzed. B-E, Tumors from A along with those from Fig. 5A and Supplementary Fig. S21C that received iRGD or anti-PD-L1 mAb treatment alone as controls were stained for CD8⁺ T cells. Panel B shows representative micrographs of the periphery and core of the tumors. The dotted lines indicate the tumor boundary. Arrowheads, CD8⁺ T cells. Scale bars, 20 μm. CD8⁺ T cells in randomly selected HPFs were counted to quantify the number of CD8⁺ T cells in each tumor (C), and the proportion (D) and absolute number (E) of CD8⁺ T cells in the core of the tumors. Three biological replicates per arm were analyzed. Statistical analysis, Mann-Whitney U test (A); one-way ANOVA (C-E). Error bars, mean ± standard error. The schematic was created in BioRender. MIYAMURA, N. (2026) https://BioRender.com/pbnlqhs.

DISCUSSION

We show that tumor-infiltrating Tregs are an important αvβ5⁺ cell population in the PDAC tissue, and that the αvβ5-rich TME of PDAC serves as a matrix for TGF-β activation. The iRGD peptide alone and in combination with chemotherapy reduces the number of Tregs and inhibits αvβ5-mediated TGF-β activation in the tumor, bringing about improved vascular perfusion, reduced stroma, and enhanced T cell entry into the tumor. Thus, iRGD, in addition to its role in enhancing drug uptake to tumors, may be useful as a TME modifier and potentiator of tumor immunotherapy.

Our study adds Tregs to the list of αvβ5⁺ cells in the PDAC TME in mice. We initially showed the expression of αvβ5 on CD4⁺CD25⁺ T cells. While CD25 is a T cell activation marker that is not specific to Tregs, CD4⁺CD25⁺ T cells are known to largely consist of Tregs (32,40). Indeed, most of the CD4⁺CD25⁺ T cells both in the tumor and spleen expressed Foxp3, and the Foxp3⁺ subset in the tumor expressed αvβ5. Of note, neither the CD4⁺CD25⁺ T cells nor the Foxp3⁺ subset in the spleen showed meaningful αvβ5 expression. This tissue-dependency of αvβ5 expression correlated with the ability of iRGD to target Tregs, suggesting that αvβ5 can be an affinity target to direct therapies to tumor-infiltrating Tregs. However, the αvβ5 expression found on minor non-Treg populations requires attention when delivering cytotoxic drugs as they may serve as a source for off-target toxicity. The expression profile of αvβ5 on Tregs in other organs and disease contexts remains to be elucidated.

The correlation of αvβ5 and CD25 expression in mouse iTregs suggests that αvβ5 is an activation marker. Indeed, enriching αvβ5⁺ cells among CCR8⁺ iTregs significantly enhanced their ability to inhibit T cell proliferation. While αvβ5 was not expressed on CD25⁺ splenic nTregs, TCR stimulation induced it, indicating that activation of the Tregs was a key trigger. TCR stimulation is critical for antigen priming in Tregs (52) and may explain why αvβ5⁺ Tregs were primarily found in the tumor and induced in vitro in the presence of PDAC cells. These findings suggest that αvβ5⁺ Tregs are functional and important targets for PDAC therapy. Our in vitro studies confirmed that the αvβ5⁺ Tregs can be targeted using affinity ligands because iRGD binding was blocked by an αvβ5-specific small molecule inhibitor and an anti-αvβ5 Ab. However, the expression of other RGD-binding integrins, such as αvβ8 (53), has yet to be profiled, particularly in light of the incomplete inhibition observed with the αvβ5 blockades in this study.

The mode of αvβ5 expression on human Tregs follows the same pattern as in mouse Tregs. αvβ5⁺ Tregs were found in the tumor but not in the spleen collected from human PDAC patients. αvβ5⁺ Tregs were induced in vitro from CD4⁺ T cells and nTregs isolated from human PBMCs. While we did not test whether iRGD targets the human αvβ5⁺ Tregs, it is likely to be achieved because iRGD binds to the human αvβ5 protein (8). The αvβ5 profiling also revealed widespread expression of this integrin in human PDAC tissue particularly in cancer ducts and surrounding stromal cells. This profile supports our previous data that PDAC cells and CAFs stimulate each other to express αvβ5 by producing TGF-β (7). The communication mediated by TGF-β may also support the induction and maintenance of αvβ5⁺ Tregs given that they were often found around the cancer ducts pretending as guardians against anti-tumor immunity. Their presence may also help maintain local TGF-β concentration because Tregs are major TGF-β producers (17).

We show in two PDAC models, transgenic KPC and syngeneic PDAC mice, that systemic iRGD treatment reduces Tregs in the tumor but not in the spleen. The preferential Treg depletion in the tumor is encouraging as it is generally accepted that Treg depletion should be achieved in a tumor-specific manner to avoid autoimmune toxicities (26,27). However, to date, there are limited ways to achieve it. One of them is to target CCR8⁺ Tregs (42). These Tregs are highly immunosuppressive and exist in various mouse and human cancers. In some cancers, up to 80% of tumor-infiltrating Tregs were CCR8-positive. Being a functional subpopulation of CCR8⁺ Tregs, αvβ5⁺ Tregs may serve as a novel target to improve the tumor specificity and efficiency of Treg depletion.

We initially hypothesized that selective targeting/killing of αvβ5⁺ Tregs was the primary mechanism underlying the Treg depletion in iRGD-treated PDAC. Every circumstantial evidence seemed to support this hypothesis. For example, FAM-iRGD targeted tumor-infiltrating Tregs in vitro and in vivo; these Tregs expressed both iRGD receptors (αvβ5 and NRP-1); iRGD preferentially depleted Tregs in the tumor correlating with the expression profile of αvβ5 on tumor-infiltrating Tregs; and both the integrin-binding RGD motif and NRP-1-binding RXXK motif were required to recapitulate the iRGD-mediated Treg depletion in vivo. However, in vitro studies showed that iRGD depleted not only αvβ5⁺ iTregs but also αvβ5⁻ iTregs, clearly contradicting this hypothesis. It was particularly confusing that the iTreg depletion depended on the RGD motif even though the effect was seemingly integrin-independent. We resolved this discrepancy by showing that iRGD counteracted TGF-β activation in an RGD-dependent manner. We propose a model in which iRGD inhibits αvβ5-dependent TGF-β activation, thereby depriving both αvβ5⁺ and αvβ5⁻ Tregs of bioactive TGF-β required for their development and survival. This model also explains why iRGD primarily reduced Tregs within PDAC tumors while minimally affecting non-Tregs even when the cells expressed αvβ5. In the in vitro setting, we suspect that αvβ5⁺ iTregs were responsible for the activation of TGF-β produced by the iTreg populations. The experiments were conducted over a 3-day period, which is sufficient for Tregs to produce endogenous TGF-β, and for the effects of iRGD to be detectable even in the presence of the exogenous TGF-β, which was used to induce iTreg differentiation (38,39). Although direct evidence that αvβ5⁺ iTregs mediated TGF-β activation is lacking, the possibility is supported by previous reports showing that Tregs can produce and self-activate TGF-β via αv integrins to maintain their functions (53).

Whether direct killing of αvβ5⁺ Tregs still contributed to the iRGD-induced Treg depletion is an important question. The αvβ5 integrin mediates survival signals, and blocking them with Abs and RGD peptides can cause apoptotic events (3,5). iRGD did promote apoptosis of αvβ5⁺ iTregs in vitro, but the effect was quite limited even at a relatively high concentration. No increased apoptosis was noted at lower concentrations that are more physiologically relevant in light of standard preclinical and clinical dosing regimens of iRGD (7,11,12). This finding agrees with earlier studies that iRGD, at similar doses, did not induce cell death (e.g., ref. 11). Collectively, while the possibility that iRGD directly killed αvβ5⁺ Tregs cannot be explicitly ruled out, this mechanism was unlikely to be central to the Treg depletion observed in our studies.

In the PDAC TME, various cells beyond Tregs, such as PDAC epithelial cells and CAFs, express the αvβ5 integrin, establishing a broad matrix for TGF-β activation. In KPC78 PDAC cells, αvβ5 was the central TGF-β activator. At mRNA level, the expression of the β5 integrin was very high compared to the negligible expression of β subunits such as β3, β6, and β8 that form other RGD-binding αv heterodimers. This finding again supports our previous data that PDAC cells express high levels of αvβ5 triggered by TGF-β (7). Deleting the β5 subunit inhibited the cells from responding to exogenous LAP-TGF-β evidenced by reduced phosphorylation of Smad2 and expression of a downstream TGF-β target gene. CAFs, which make up most of the tumor volume alongside PDAC epithelial cells, also activated TGF-β in our in vitro assay. These findings indicate that PDAC has assembled a vicious cycle that keeps facilitating TGF-β activation and αvβ5 expression in its TME (Fig. 7A). This mechanism may explain the excessively desmoplastic and immunosuppressive nature of PDAC (23) and the poor prognosis of PDAC patients associated with high αvβ5 expression in the tumor tissue (6).

Fig. 7: Maintenance of TGF-β activation and αvβ5 integrin expression in PDAC.

A, Positive feedback loop between TGF-β activation and αvβ5 integrin expression in PDAC. αvβ5 integrin mechanically facilitates the release of active TGF-β from LAP-TGF-β. The activated TGF-β in turn induces the expression of αvβ5 integrin on cells in the PDAC tissue. B, Various cells such as cancer epithelial cells, CAFs, and Tregs in the PDAC tissue produce TGF-β and express αvβ5 integrin to maintain high TGF-β activity and αvβ5 integrin expression in PDAC. Targeting the αvβ5 integrin with the iRGD tumor-penetrating peptide inhibits the integrin-mediated TGF-β, leading to a dynamic TME modification characterized by reduced ECM components, improved vascular perfusion, Treg depletion, and CD8+ T cell infiltration. The Schematics were created in BioRender. MIYAMURA, N. (2026) https://BioRender.com/pbnlqhs.

We show that iRGD disrupts the αvβ5-dependent TGF-β activation cycle and normalizes some of the TME features in PDAC (Fig. 7B). iRGD, but not its RGD-disrupted variant, inhibited TGF-β activation by KPC78 cells suggesting αvβ5-dependency. We confirmed the effect in several ways by measuring p-Smad2 in KPC78 cells and detecting the conversion of LAP-TGF-β to active TGF-β using M21L responder cells. The in vivo effect was striking. iRGD reduced p-Smad2 throughout the tumor in PDAC mice causing dynamic TME modifications represented by improved vascular patency and perfusion, reduced stromal fibers, and deeper infiltration of CD8⁺ T cells into the tumor core. Similar TME changes have been reported using TGF-β blockades (16,54). The wide TGF-β suppression was likely a major cause of the Treg depletion seen in our earlier studies, which was achieved with an identical dosing regimen. Concurrently, the tumor selectivity of the Treg depletion is explainable by the abundance of αvβ5 in the tumor, which enhances the tumor specificity of iRGD targeting. Apparently, NRP-1-dependent tumor penetration was critical for iRGD to induce these in vivo events because CRGDC failed to deplete Tregs in vivo, despite its ability to reduce iTregs in vitro. CRGDC did not inhibit Smad2 phosphorylation within the tumors either, which was supportive. NRP-1 is frequently co-expressed on αvβ5-positive cell populations in PDAC and found in up to 80% of human cases (8,34,35). This abundance provides a robust platform for iRGD to achieve tumor-penetrating TGF-β antagonism.

The enhanced CD8/Treg ratio induced by iRGD in syngeneic KPC78 tumors mainly reflected the depletion of Tregs rather than an increase in CD8⁺ T cells because iRGD did not dramatically increase CD8⁺ T cells within the short treatment period. Longer treatment with iRGD may further increase the CD8 burden as the results in transgenic KPC mice suggest. Our data show that the main effect of iRGD on CD8⁺ T cells was that it facilitated their local spreading into the tumor interior. We postulate that the improved perfusion and reduced stroma were the key mechanisms of this effect. This is in line with prior studies, which showed that restoring tumor perfusion by inducing vascular normalization leads to increased T cell infiltration (18,54,55). Accumulating evidence suggests that vascular normalization has profound effects on improving anti-tumor therapy. In addition to establishing a physical conduit for immune cells and drugs to enter the tumor, it reduces hypoxia to overcome hypoxia-induced immunosuppression by stimulating chemokine release, suppressing Tregs, and promoting CD8⁺ T cells to release cytotoxic factors (55,56). iRGD may prove to be an effective tumor-specific inducer of the beneficial changes.

Despite the ability to improve the CD8/Treg ratio in the tumor, iRGD monotherapy did not show significant anti-tumor effects consistent with previous studies (e.g., refs. 7,11,34,43). We suspect that it was due to insufficient activation and/or exhaustion of the expanded CD8⁺ T cells caused by the high PD-L1 expression in the PDAC model we used. PD-L1 can inhibit naïve-to-effector differentiation of CD8⁺ T cells and induce exhaustion after differentiation (57,58). Thus, blocking the PD-1/PD-L1 axis can be a necessary step to gain enhanced and prolonged effector function of CD8⁺ T cells against cancers, or infections. Indeed, we found that combining a PD-L1 blockade rendered iRGD therapy effective against PDAC tumors, which were refractory to both iRGD and anti-PD-L1 monotherapies. This result agrees with a previous study, which showed that delivery of cytotoxic CD8⁺ T cells into PD-L1^high gastric tumors using iRGD yielded only a moderate anti-tumor effect, and genetic deletion of PD-1 in the CD8⁺ T cells significantly augmented the therapeutic efficacy (48). A recent study performed in hepatocellular carcinoma mice corroborated this observation. It showed that iRGD combined with an anti-PD-L1 mAb led to enhanced activation of CD8⁺ T cells and improved anti-tumor effects compared to either treatment alone, thereby further supporting our premise (49).

The area of necrosis in the tumors from the iRGD + anti-PD-L1 arm showed a moderately negative correlation with the tumor size, implicating a role of cell death in the improved anti-tumor effects. Considering the increased CD8⁺ T cells in the tumor core, it is reasonable to suspect that they contributed to these changes. Some tumors in the single treatment arms also showed necrosis correlating to the upward trend in CD8 burden. However, the necrotic response was quite variable in all the arms. Between tumors, and even within a single tumor, there is a vast degree of heterogeneity at genetic, epigenetic, and microenvironmental levels (59,60). PDAC can be classified into basal and classical subtypes (or a varying mixture of the two) with the former often being richer in TGF-β and poorer in therapeutic response (61). Antagonizing TGF-β signaling in basal-like PDAC can reverse the heterogeneity and sensitize them to various therapies by shifting the tumor toward a classical subtype (62). We speculate that iRGD induced a similar effect in the KPC78 model, which showed a GATA6-low basal-like signature (50). Nonetheless, the response was still heterogenous as some tumors showed only marginal increase in vascular patency, anti-PD-L1 mAb efficacy, and necrotic burden. PDAC tumors that grow in transgenic KPC mice can be even more heterogenous showing coexisting features of basal-like and classical subtypes (63). This strong heterogeneity might have contributed to the variable therapeutic response to the iRGD + Gem combination therapy in the original study (7). Further studies are desired to optimize iRGD dosing for maximal effect and understand the differential effects of iRGD under basal-like and classical backgrounds. Finally, it is also worth noting that intravenously co-administering the anti–PD-L1 mAb with iRGD may improve outcomes in future studies by facilitating iRGD-mediated transvascular transport and accumulation of the antibody within extravascular tumor tissue (11).

The present study extends the utility of iRGD from a drug delivery molecule that achieves acute tumor penetration within minutes of injection (8,11) to a TME modifier that may provide prolonged therapeutic benefits through TGF-β antagonism. The novel properties of iRGD may improve immunotherapy for various cancers that are negatively impacted by TGF-β activity. While the exact iRGD-induced TME changes responsible for the enhanced anti-PD-L1 efficacy remain to be elucidated, our data indicate that combining iRGD with ICIs is beneficial in PDAC and provide a rationale for the ongoing iLSTA clinical trial that studies the safety and preliminary efficacy of iRGD along with Gem, Nab-P, and durvalumab (ACTRN12623000223639) (64). Importantly, some of the TME changes we observed may account for the improved therapeutic response noted in the PDAC patients treated with standard-of-care-chemotherapy in combination with iRGD because remodeling the TME using TGF-β blockades can potentiate anti-PDAC chemotherapy (12,13,62). In this context, expression profiling of αvβ5, NRP-1, and TGF-β could be considered as an approach to stratify patients for iRGD-based chemo- and immunotherapies in future clinical trials.

Statement of significance.

The iRGD tumor penetrating peptide inhibits αvβ5 integrin-dependent TGF-β activation in pancreatic cancer to normalize the desmoplastic and immunosuppressive tumor microenvironment.

Data availability

Raw sequencing data (FASTQ files) and processed count matrices generated from RNA-seq analyses have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE330526. Other data generated in this study are available upon request from the corresponding author.