Facts and Hopes in Immunotherapy of Pancreatic Cancer

Pancreatic ductal adenocarcinoma (PDAC), lying at the other end of the spectrum from melanoma with regards to response to immunotherapy, has stolen from it the dubious distinction of being called “the cancer that gives cancer a bad name.” While the rise of agents targeting the immune system has led to a renaissance in cancer care overall, immunotherapy remains largely ineffective in PDAC. Due in part to our improved understanding of basic cancer biology, mechanisms of drug resistance, and advances in drug, vaccine, and cell therapy discovery, better therapeutic options are on the horizon. Further, an increasing number of rationally designed clinical trials featuring novel treatment combinations are becoming available for patients with PDAC, paving the way for a brighter future.

While the incidence of PDAC continues to rise, the 5-year overall survival (OS) rate has remained nearly stagnant at 11%; as such, PDAC is expected to become the second most common cause of cancer-related mortality by 2030 (1, 2). Multiagent chemotherapy regimens have led to only modest improvements in patient outcomes (3, 4). Various immunotherapy trials in PDAC have failed to show meaningful results despite strong preclinical backing. Nevertheless, many of these studies offer important insights into the biology of the disease, potential biomarkers, patient selection strategies, and how to best combine existing and novel immunotherapy agents with other treatment modalities to trigger an effective, and hopefully, curative response for more patients with this disease.

In this review, we discuss barriers to effective immunotherapy in PDAC, ranging from cancer cell–intrinsic factors to immunosuppressive features of the tumor microenvironment (TME). We also examine lessons learned from recent immunotherapy clinical trials. Finally, we discuss our hopes for a future where novel immunotherapies will become effective treatments for patients with pancreatic cancer.

The pancreatic TME is a complex ecosystem including cancer cells, immune cells, cancer-associated fibroblasts (CAF), and other stromal elements that create a milieu hidden from the immune system, permissive of tumor growth and metastasis (5). Intratumoral effector T cells in PDAC are relatively rare, particularly in the juxtatumoral stroma located most closely to cancer cells (6), making it one of the “coldest of the cold” tumors. This contrasts with many other “warm” and “hot” solid tumors where infiltration of effector T cells is prominent and immune checkpoint inhibitors (CPI) have made significant inroads by both reinvigorating exhausted effector T cells and recruiting new cells. Aside from the TME, cancer cell–intrinsic factors may also hinder the efficacy of immunotherapy. Namely, oncogenic RAS signaling, ubiquitous in PDAC, drives an inflammatory program that establishes immune privilege in the TME (7), with TME cells in turn secreting specific cytokines to reprogram the metabolic landscape of PDAC cells, upregulating glycolytic genes and glucose uptake (8). This contributes to metabolic derangement in the TME, creating a hostile environment for T cells, which similar to cancer cells are reliant on glycolytic pathways for survival (9). The hypoxic and acidic milieu of PDAC with increased lactic acid production may further limit proper function and proliferation of tumor-infiltrating lymphocytes (10). Tumor mutational burden (TMB), a modest predictor of response to CPI, correlates with neoantigen burden and is often low in PDAC (11, 12). PDAC generally expresses an average of <100 neoepitopes per tumor; and only a small portion of these can modulate immunogenicity (13). To compound this problem, antigen presentation may be impaired by mechanisms such as MHC-I downregulation on cancer cells and a paucity of dendritic cells (DC; refs. 14, 15). In addition, tumor infiltration by immunosuppressive myeloid cells and a dense desmoplastic stroma with its related chemokine axis deflects the immune response and supports tumor progression (6, 16–18). A summary of key mechanisms underlying resistance to immunotherapy in PDAC is illustrated in Fig. 1.

Inhibitors of cytotoxic T lymphocyte—associated protein 4 (CTLA-4) and programmed death 1 (PD-1) T-cell checkpoints are the backbone of contemporary immunotherapy. While both pathways suppress T-cell responses, temporal and site-specific differences impart nonredundant functions. Notably, CTLA-4's B7 ligands are primarily restricted to professional antigen-presenting cells (APC) in secondary and tertiary lymphoid organs, while programmed death-ligand 1/2 (PD-L1/2) are inducible and broadly expressed by leukocytes, non-hematopoietic, and cancer cells (19). Further, owing to PD-1 expression on chronically stimulated T cells, PD-L1 antagonists enhance the activity of activated antigen-experienced effector T cells (19). In contrast, CTLA-4 expression is limited to naïve and regulatory T cells (Treg) such that CTLA-4 antagonists selectively enhance naïve T-cell activation (20). These nuances may explain why responses to conventional immunotherapy are limited to small subsets of patients in immunogenic ‘hot’ tumors (e.g., melanoma) but not in ‘cold’ tumors (e.g., PDAC). Notably, recent discoveries on the critical role of fragment crystallizable-Fc gamma receptor (Fc-FcγR) co-engagement for CTLA-4 antibodies have led to the emergence of next-generation Fc-enhanced anti–CTLA-4 antibodies that harness multiple mechanisms of action to overcome barriers associated with ‘cold’ tumors for more effective antitumor immunity (21, 22).

Clinical experience with first generation CTLA-4 and PD-1 inhibitors to date has been underwhelming in PDAC. The anti–CTLA-4 antibody ipilimumab was ineffective as monotherapy with only 1 patient of 27 with advanced PDAC experiencing an objective response (23). In addition, ipilimumab showed no additive benefit when combined with gemcitabine in previously treated patients with PDAC, with an objective response rate (ORR) of 14% and disease stabilization in 33% of patients (24). In a phase II trial of durvalumab, a monoclonal antibody (mAb) against PD-L1, with or without the anti–CTLA-4 antibody tremelimumab, ORR was only 3.1% for patients receiving combination therapy and 0% for those receiving monotherapy (25). Table 1 summarizes the results of a select group of completed immunotherapy trials in PDAC (23, 25–33).

Pancreatic cancer stroma is dominated by a robust desmoplastic and hypo-vascular matrix that creates a mechanical and biochemical boundary around cancer cells that prevents appropriate vascularization, diminishes exposure to chemotherapy, and limits immune cell infiltration into the tumor (34–36). CAFs contribute to this stroma and comprise a diverse population of cells that have been implicated in tumor progression, metastatic spread, and in immunotherapy failure (16, 37, 38). Many of the immunomodulatory effects of CAFs are thought to involve cellular cross-talk through the secretion of cytokines (39–41) which lead to T-cell exclusion from the tumor parenchyma and resistance to CPI (42–44). Conflicting evidence exists that certain stromal elements restrain, rather than promote tumor growth (45, 46), suggesting the existence of functionally and temporally divergent populations of immune cells and CAFs in the TME. Thus far, CAF-directed therapies have been primarily targeted to the highly expressed serine protease fibroblast activation protein-α (FAP), but without clear efficacy signals. For instance, in a phase II trial, the small molecule inhibitor of FAP, talabostat, was combined with gemcitabine (47). Of 21 evaluable patients with advanced PDAC, only 3 achieved objective responses and median progression-free survival (mPFS) was 3.5 months (47). Recent studies have noted the heterogeneity and plasticity of CAFs in PDAC, including identification of myofibroblast (myCAF), antigen-presenting (apCAF), and inflammatory fibroblast (iCAF) subtypes (48, 49). This phenotypic heterogeneity may underlie observations that unselected targeting of CAFs could paradoxically accelerate tumor progression (45, 46).

TGFβ is a cytokine released by pancreatic tumor cells, fibroblasts and other stromal cells contributing to the immunosuppressive architecture of the TME (50). Targeting of TGFβ with the small molecule inhibitor galunisertib combined with gemcitabine as first-line therapy in PDAC only slightly increased median overall survival (mOS) versus gemcitabine monotherapy and did not reach statistical significance [8.9 vs. 7.1 months; HR, 0.79; 95% confidence interval (CI), 0.59–1.09; ref. 51]. Galunisertib was subsequently tested in combination with durvalumab in 32 patients with advanced PDAC and showed limited efficacy, with only one partial response (PR; ref. 26). New strategies, for example using a bifunctional fusion of a mAb against TGFβ with other CPIs are under investigation (52).

All-trans retinoic acid (ATRA) is a metabolite of vitamin A that has pleiotropic effects on immune and nonimmune cell types. Under tumorigenic conditions, the tissue-resident fibroblasts, pancreatic stellate cells, lose their intracellular stores of retinoic acid and convert to critical mediators of desmoplasia (53). Restoring retinoic acid with ATRA limits desmoplasia and reduces cancer growth in a preclinical model (54). In addition, ATRA increases the ratio of CD8-to-Tregs in preclinical models (55). In a phase Ib trial, ATRA combined with gemcitabine and nab-paclitaxel showed a mOS of 11.7 months in the first-line setting for locally advanced and metastatic disease (56).

Stromal cell–derived factor-1 (CXCL12), a chemokine produced by CAFs, is often expressed at higher levels in the PDAC TME and creates a network of dense stroma restricting immune cell migration and recognition of cancer cell antigens (16). Preclinically, disrupting the binding of CXCL12 with its receptor, C-X-C chemokine receptor type 4 (CXCR4) improved the effect of CPI in PDAC models (6, 16). A phase II trial with the small molecule CXCR4 antagonist, motixafortide in combination with pembrolizumab, demonstrated enhanced CD8⁺ effector T-cell infiltration in the tumor, decreased numbers of intratumoral myeloid-derived suppressor cells (MDSC), and decreased circulating Tregs (57). Despite these effects, the combination of motixafortide/pembrolizumab/NAPOLI-1 regimen (liposomal irinotecan, 5-fluorouracil, and leucovorin) led to a modest ORR of 21.1%, disease control rate (DCR) of 63.2% and mOS of 6.6 months in patients with de novo metastatic PDAC who progressed after gemcitabine-based chemotherapy (33). Notably, patients without liver metastasis derived more benefit from this immunotherapy regimen, with an ORR of 37.5% and DCR of 87.5% (33).

PDAC tumors are infiltrated by tumor-associated macrophages (TAM) of the immunosuppressive phenotype (M2) defined by expression of CD206, colony-stimulating factor 1 receptor (CSF-1R), IL10 and decreased expression of MHC class II (58). The CSF-1 pathway is critical for the differentiation and survival of M2 macrophages. Blockade of the CSF-1 pathway has been shown to shift TAMs towards the M1 phenotype and induce distinct TME remodeling (59–61). Cabiralizumab, a humanized IgG4 mAb, binds to CSF-1R and blocks its signaling. In a phase I trial combining cabiralizumab and nivolumab, the ORR was 10% in a cohort of 31 heavily pretreated patients with PDAC (62). However, in a subsequent phase II trial cabiralizumab in combination with nivolumab, with or without standard of care (SOC) chemotherapy did not improve progression-free survival (PFS) compared with SOC chemotherapy alone in the second-line setting (NCT03336216; ref. 63).

Preclinical characterization of IL10 has revealed its direct regulation of MHC class II antigens, expansion of T cells, B cells, and mast cells, as well as promotion of T cell–mediated tumor rejection (64, 65). Pegylated IL10 (pegilodecakin) was tested in an early-phase study demonstrated clinical activity when combined with folinic acid, fluorouracil, and oxaliplatin (FOLFOX) in patients with gemcitabine-refractory metastatic PDAC, with an ORR of 15.8%, including two (10.5%) complete responses (CR), and a 1-year survival rate of 43% (66). Despite these promising early results, the addition of pegilodecakin to FOLFOX did not improve efficacy in the phase III SEQUOIA trial (31). Notably, exploratory analyses revealed that pegilodecakin caused immunostimulatory changes such as an increase in total IL18, IFNγ, and granzyme B, as well as a reduction of TGFβ (31). In addition, larger on-treatment increases in IL18 correlated with improved clinical outcomes in pegilodecakin-treated patients. Further investigation into whether baseline levels of IL18 may be predictive of benefit is warranted.

DCs are essential to generate antitumor adaptive immunity through their priming of de novo naïve antigen-specific T cells, but often fail to accumulate in the tumor milieu where they acquire tumor antigen (67, 68). Further, DCs that do accumulate in the tumor may be immature and less effective at antigen presentation and T-cell stimulation (69). DC dysfunction is mediated in part by dysregulated cytokines such as IL10 and TGFβ secreted directly by the tumor or tumor-recruited immune cells (69, 70). Thus, therapeutic approaches to engage DCs may improve adaptive immune responses against the tumor. CD40 ligand, a costimulatory molecule expressed on T cells, binds to its cognate receptor (CD40) on DCs and other APCs leading to lymphocyte activation (71). As demonstrated in preclinical models, CD40 signaling upregulates the expression of B7 costimulatory molecules and promotes antigen presentation by APCs to activate antigen-specific T cells (72–74). Recently, the phase I PRINCE trial with the CD40 agonist mAb sotigalimab (APX005M)/gemcitabine/nab-paclitaxel/±nivolumab, yielded a median OS of 20.1 months in the first-line for patients with locally advanced and metastatic PDAC (29). In a phase II study, gemcitabine/nab-paclitaxel/nivolumab had superior outcomes with a 1-year OS rate of 57.3%, meeting the primary endpoint. Gemcitabine/nab-paclitaxel/sotigalimab had intermediate efficacy with 1-year OS rate of 51% and unexpectedly, chemotherapy/sotigalimab/nivolumab had lower efficacy with a 1-year OS of only 41% (75). Further, a non-fucosylated IgG1 CD40 agonist mAb (SEA-CD40) with gemcitabine/nab-paclitaxel/pembrolizumab in the first-line setting demonstrated an ORR of 44% with a mOS of 15 months (76). A potential pitfall for targeting CD40 in PDAC may be the prevalence of liver metastases, in which a scarcity of DCs may directly contribute to immunotherapy resistance (77).

Another approach to augment T cell–mediated immune surveillance is to employ vaccines targeting tumor-associated antigens that are overexpressed in cancer cells relative to normal tissues or tumor-specific neoantigens derived from mutational events. GVAX is a whole-cell vaccine derived from allogeneic or autologous PDAC cell lines engineered to release GM-CSF at the vaccination site (78). In a phase I trial, previously treated patients with advanced PDAC were randomized to either ipilimumab or ipilimumab plus GVAX. Three patients in the combination arm had prolonged stable disease with a decline in carbohydrate antigen 19–9 (CA19–9; ref. 79). Median OS was 3.6 months in the ipilimumab arm and 5.7 months in the GVAX/ipilimumab arm, with encouraging induction of mesothelin-specific T cells and enhancement of the T-cell repertoire with GVAX (79). However, a phase II trial of GVAX in combination with CRS-207, a live-attenuated listeria vaccine expressing mesothelin, showed no survival advantage over physician's choice of single-agent chemotherapy in the second-line for metastatic PDAC (80). These data underscore the challenges in eliciting cancer-specific T-cell responses in PDAC and suggest that a combination of vaccines with immune modulators beyond first generation anti–CTLA-4 and/or anti–PD-1 antibodies will be necessary to achieve clinical benefit.

Oncolytic viruses can similarly promote tumor-directed innate and adaptive immune responses. Pelareorep, an oncolytic reovirus, was tested in a phase Ib trial with pembrolizumab and chemotherapy (gemcitabine, 5-fluorouracil, or irinotecan) in previously treated patients with PDAC. Some clinical benefit was observed, with 3 of 11 patients experiencing a PR or stable disease (81). Biomarker analysis revealed changes in the peripheral T-cell repertoire, with early and durable clonal expansion correlating with longer OS (81). Additional investigation is warranted to determine if these dynamic changes in peripheral T-cell clonality may serve as a predictive biomarker of response to oncolytic viral therapy. VCN-01, another oncolytic virus, was designed to replicate in cancer cells with a dysfunctional RB1 pathway and express hyaluronidase (82). A phase I trial of gemcitabine/nab-paclitaxel in combination with intratumoral injection of VCN-01 showed that the agent was well tolerated, patients had increasing levels of serum hyaluronidase and injected lesions achieved stabilization or regression (82).

The use of appropriate models is critical to determine the dominant drivers of cancer immunity in pancreatic cancer and better inform clinical trials. Studies employing the genetically engineered mouse KPC model (LSL-Kras^G12D/+;LSL-Trp53^R172H/+;Pdx-1-Cre) have enabled the elucidation of several mechanisms of stromal-driven immunosuppression in PDAC (83). Unfortunately, immunotherapy approaches including CXCR4 and hedgehog inhibitors, among others, which demonstrated efficacy in the KPC model have failed in the clinic (33, 84–86). Potential explanations for these discrepancies may include dissimilarities in the immune cell composition of human pancreatic tumors and KPC, such as a higher abundance of effector memory CD8⁺ and Tregs in some human tumors (6, 87), whereas murine tumors have a more myeloid-predominant microenvironment (87). Hence, employing KPC as the only model for preclinical testing of immunotherapy may be inadequate to predict clinical responses. Other factors including the size and location of the tumor in the model (e.g., subcutaneous, orthotopic, or liver) as well as the magnitude of the response in the model may be important, that is, a modest increase in PFS may be less predictive in the clinic than a curative response in a majority of treated mice.

Patient-derived xenograft (PDX) avatar models have been developed to expand translational research and personalize the treatment of pancreatic cancer (88). A fundamental shortcoming of PDX models for immunotherapy studies is that the tumors are propagated in immunocompromised mice (88, 89). PDAC avatar tumors do not retain donor immune cells, including T cells and macrophages. To enable translational immunotherapy research, avatar tumors can be grown and propagated in mice with a “humanized” immune system (90, 91), such as NOD/SCID gamma mice engrafted with human CD34⁺ hematopoietic stem cells (91). While these humanized models have been used to evaluate responses to CPI in lung cancer and melanoma with varying success (91, 92), immune cells of humanized avatar PDAC models are not derived from the same patient as the tumor and this may hamper their ability to predict responses in the clinic.

Ex vivo tissue cultures maintain tumor histology, proliferation, and viability for up to 7 days (6, 93). In these cultures, the integrity of the PDAC TME is evidenced by viable stromal myofibroblasts, CD8⁺ T cells, and CD68⁺ macrophages (6, 93), and may serve as a close surrogate of the patient's tumor. A recent study using an ex vivo culture evaluated the immune effects of the CXCR4 inhibitor plerixafor and an anti–PD1 antibody, showing enhanced effector function of CD8⁺ T cells as well as enhanced migration through the stromal microenvironment (6), recapitulating a previous study in KPC models (16). Unfortunately, the clinical impact of this therapeutic strategy was more limited (33), again underscoring the challenges in translating preclinical findings to the clinic. Implementing methods to maintain viability beyond 7 days may be needed because many relevant immune responses take longer than a week to develop. In addition, this model does not account for immune cell migration from the periphery to the tumor.

Combining more than one of these methods may potentially help avoid late-stage clinical trial failures. Further, taking scientific insights derived from early-phase trials and applying them backwards to preclinical models as a means of reverse translation may help better characterize and develop more relevant preclinical models for immunotherapy of PDAC.

Multiple promising therapies impinging on a range of mechanisms are under preclinical and clinical development. These include novel approaches to promote T-cell responses, modify myeloid and stromal compartments, and recruit new immune cells to the PDAC TME. In addition to the CD40 agonistic antibodies discussed earlier, agonism of CD11b has emerged as an approach to reprogram myeloid cells (94). GB1275 is a first-in-class oral allosteric agonist of CD11b shown to blunt both monocyte and granulocyte infiltration in the TME, while simultaneously repolarizing tumor-resident TAMs to an M1 phenotype overcoming resistance to CPI in murine models of PDAC (18). These preclinical data have led to a clinical trial of GB1275 in combination with anti–PD1 or chemotherapy in PDAC and other tumors (NCT04060342; ref. 95). Beyond targeting the myeloid compartment directly, other approaches add to the immune repertoire via allogeneic natural killer (NK) and invariant natural killer T (iNKT) cells to bypass antigen presentation and reprogram the TME. These cellular therapies may overcome the immunosuppressive TME in PDAC with their innate ability to recognize tumor cells by bypassing downregulated antigen-MHC complex presentation on PDAC cells and modulating TAMs (96–98). A phase II multimodal study of an off-the-shelf allogeneic PD-L1–targeted high-affinity NK cell infusion (t-haNK) with an IL15 agonist mAb, radiation, and chemotherapy (NCT04390399), and a phase I trial of iNKT cells in combination with CPI in solid tumors including PDAC are underway (NCT05108623).

Aside from targeting other elements of the immune system, there is renewed interest in recruiting and activating effector T cells to cold tumors such as PDAC with a new generation of CPIs in combination with other modalities. Botensilimab (AGEN1181), an Fc-enhanced CTLA-4 inhibitor designed to strengthen APC/T cell co-engagement thereby promoting optimal T-cell priming, activation and memory formation has yielded curative responses combined with gemcitabine and nab-paclitaxel in the preclinical KPC model (99). In a phase I trial, objective responses with botensilimab have been reported in pancreaticobiliary, microsatellite stable (MSS) colorectal cancer and several other immunotherapy-resistant tumors (100) prompting development of a phase II combination trial with gemcitabine and nab-paclitaxel in PDAC. T-cell immunoglobulin and ITIM domain (TIGIT) has also been implicated in immune escape in PDAC (101), and blockade of the CD155/TIGIT axis in combination with other CPIs, such as CD40, may prove beneficial (101). Targeting stroma along with CPI is also a common theme, including a focus on the TGFβ and adenosine pathways (102, 103). Given prior failures with nonspecific targeting of CAFs, a novel approach entering phase I clinical trials is using antibody–drug conjugates to target FAP-expressing fibroblasts with the FAP protein as a dock for the local delivery of a cytotoxic payload to kill bystander tumor cells (92). Finally, microbial host factors have been shown to affect tumor responses in the KPC preclinical model following fecal microbiota transplantation from PDAC long-term human survivors, a concept currently being tested in clinical trials (NCT04975217; ref. 104). Table 2 provides a select group of ongoing clinical trials testing novel immunotherapy approaches that may hold promise in PDAC.

The few biomarkers in clinical practice that may predict response to immunotherapy include PD-L1 expression, TMB, and microsatellite status. Of these, in PDAC, PD-L1 has no established role, and TMB and microsatellite status are of limited utility as only a small proportion of PDAC tumors are microsatellite instability-high (MSI-H), DNA mismatch repair deficiency, or high TMB (105). Response rates to CPI in these patients are modest (106, 107) and inferior to other gastrointestinal MSI-H tumors (106, 108). In PDAC, homologous repair deficiency (HRD) may predict response to platinum-based therapy (109) as well as to PARP inhibitors (110), and HRD is also an emerging biomarker of response to combined checkpoint inhibition including CTLA-4 inhibitors in PDAC (111). While HRD represents a larger segment of the PDAC population than MSI-H, it still represents only a minority of patients with PDAC even if the definition is extended beyond germline mutations. While not yet clinically relevant, efforts are underway to classify PDAC based on genomics and transcriptomics and link these molecular subtypes, including an “immunogenic” subtype to clinical activity (112, 113).

In addition to information gleaned from tumor and blood samples, patient intrinsic factors such as the neutrophil to lymphocyte ratio as represented in the Gustave Roussy Immune Score (GRIm) score (114), along with sites of metastatic disease are also emerging as important biomarkers of response to immunotherapy across solid tumors. Liver metastases are known to be particularly recalcitrant to immunotherapy (115) and because they predominate in PDAC, they likely contribute to primary resistance to CPI as they do in MSS colorectal cancer (116). Aside from selecting patients with PDAC without liver metastases for immunotherapy, another potential approach would be to locally treat liver metastases in conjunction with effective systemic immunotherapy. A study of ipilimumab, nivolumab, and stereotactic body radiation therapy (SBRT) in PDAC may support this approach with a DCR of 29% (5 of 17) and 1 patient experiencing a CR (117). In addition, in the trial of the CXCR4 inhibitor motixafortide with anti–PD-1 and chemotherapy discussed above, patients with PDAC without liver metastases had greater benefit compared with those with liver metastases (ORR 37.5% vs. 21.1%, respectively; ref. 33). Whether sites of metastatic disease are simply prognostic or predictive of benefit for immunotherapy in PDAC can be answered in future trials with more effective immunotherapy combinations. These trials should be thoughtfully coupled with translational research and biomarker development to maximize our ability to identify patients who will benefit from treatment.

“Cure” is a laden word in the context of metastatic cancer, with no universally accepted definition. As oncologists, we are loathe to use the word with patients out of fear that a late relapse will occur, and we will be guilty of having instilled false hope. Ultimately, what matters to most patients is a quality life that is not physically or temporally constrained by the presence of cancer. This goal is reflected in the statistical concept of the “tail” of the survival curve, which represents prolonged survival. James Allison, in his acceptance speech for the Nobel Prize, which he received for his role in the discovery of inhibition of negative immune regulation (CTLA-4), famously described the goal of immunotherapy combinations as “raising the tail” of the curve (118). The question then becomes, what practical surrogate endpoints will most accurately translate into greater long-term survival for patients with PDAC? While mOS has been the gold standard historically, using it as the primary endpoint in early-phase trials would unacceptably delay progress.

For late-phase immuno-oncology clinical trials, novel designs using milestone-based endpoints have been proposed to account for the uncertainty of survival kinetics with immunotherapy (119). These designs are informed by ipilimumab/nivolumab versus nivolumab (120) in melanoma where an improvement in ORR (with durability) gave an early signal of efficacy, leading to improved landmark PFS at 12 to 18 months, and finally an improvement in OS that became more pronounced over time. When the clinical effect of a therapy may be delayed and the cure rate is relatively small, traditional endpoints such as mPFS and mOS may be initially misleading. This is apparent in patients with MSI-H PDAC treated with pembrolizumab, where there was only an 18% ORR and median PFS of 2.1 months, but the durability of responses was substantial (median duration of response 13.4 months, range 8.1 to 16.0+; ref. 106).

The challenge in phase one immuno-oncology trials in a resistant disease like PDAC is even greater: how to appropriately select aggressive endpoints with near term readouts that will de-risk and set an appropriately high bar for combinations worth pursuing in subsequent trials. Drawing on experience in melanoma, we propose that these trials should aim for a substantial improvement in ORR with durability (e.g., 90% ORR in first-line PDAC), and if this is not achieved, then landmark PFS at 12 months should be at least doubled compared with SOC (greater than 30% to 40%; ref. 3, 4). If the bar is met, the phase II trial should incorporate an active control arm. In addition to imaging-based endpoints, we hope that circulating tumor DNA and other blood-based biomarkers will add dynamic range to early read outs, thereby improving trials in the metastatic setting, and allowing for more practical trials of immunotherapy in the neoadjuvant and minimal residual disease setting (e.g., NCT04853017).

Pancreatic cancer exhibits multiple immunotherapy resistance mechanisms, and therapies targeting only one mechanism at a time have failed. While new immunotherapy agents with monotherapy activity in this disease would be welcome, we believe it is possible to maximize the benefit of existing agents through rational combinations. Indeed, a framework has been suggested for successful next-generation immunotherapy-based treatment of PDAC featuring combinations that boost immune priming and activation, block immune checkpoints, and optimize the TME, in concert with standard cytotoxic therapy (121). MSS colorectal cancer serves as an example: it is a disease where PD-1 inhibition alone has not shown activity (122) and regorafenib has an ORR of only 1% (123). Despite this, the REGONIVO trial combining regorafenib and nivolumab demonstrated encouraging results (124), igniting interest in PD-1 combinations that led to ongoing registration trials where the individual agents have no single agent activity (NCT04776148, NCT05064059).

Prior to the advent of modern immunotherapy, there were notable successes curing complex metastatic solid tumors and hematologic malignancies with novel combinations intended to overcome resistance to single agents (125–127). Testicular cancer and lymphoma exemplify the rapid progress made with this approach (126, 127). More recently, investigators have focused on “one step at a time” development but given substantial obstacles of time and cost and a sense of desperation for patients with cancer without viable options, members of the oncology research community are calling for a return to bold combination designs (128, 129). While there remain several unanswered questions with respect to optimal immunotherapy combinations for PDAC (e.g., optimal sequencing with chemotherapy to move into earlier lines) experience curing testicular cancer and lymphoma tells us it is not necessary that all these questions be answered to design successful combination trials if basic principles are followed. Combinations should have a strong biological rationale and supporting preclinical evidence, preference should be given to agents with individual activity, safety must be acceptable, the patient population should allow for adequate assessment of efficacy based on the expected time to response with immunotherapy, and endpoints should be selected appropriately as discussed above. To encourage such trials, there must be a path to approval for a highly effective combination of multiple novel agents. Progress necessitates that efficacy in an area of unmet need must take precedence over requirements for dose optimization and assessment of contribution of components. To succeed, there must be cooperation between regulatory authorities, academic investigators, biotech engines of drug discovery, and patient advocacy groups, allowing clinical research the flexibility necessary to achieve rapid and meaningful treatment advances for patients.

B. Bockorny reports grants from NanoView Biosciences, nonfinancial support from Erytech Pharma, and personal fees from Blueprint Medicines outside the submitted work. J.E. Grossman reports personal fees and other support from Agenus outside the submitted work. M. Hidalgo reports personal fees from BMS, InxMEd, Champions Oncology, MinKy, VelaVigo, Oncomatrix, and Kahr and other support from Nelum outside the submitted work; in addition, M. Hidalgo is a member of BMS Board of Directors.