Patient enrollment

From 3 March 2016 to 14 January 2019, 39 patients were randomized to Arms A and B (Figs. 1 and 2, Fig Supplementary Figure 1). From 15 February 2019 to 9 September 2020, Arm C enrolled 12 patients consecutively (Figs. 1 and 2, Supplementary Figure 1). During this time, randomization was held to ensure that Arm C met its accrual goal while urelumab was still available prior to its discontinuation by BMS (Supplementary Figure 1). In October 2020 the remaining supply of urelumab expired, and any Arm C patients who remained on study were transitioned to Cy-GVAX and nivolumab (Arm B regimen) with the same schedule and dosing (without urelumab). This impacted the treatment of three Arm C patients: the 1st received urelumab with their initial four study therapy cycles, the 2nd with their initial 2 cycles, and the 3rd with only their 1st (neoadjuvant) cycle. Once enrollment in Arm C was complete, randomized enrollment to Arms A and B restarted with an additional three patients enrolled, between 25 February 2021 and 10 September 2021, before the decision to close these respective Arms due to the plans to add new treatment Arms to the platform trial. The date of data cutoff for the final analysis for Arms A, B, and C was 25 May 2022.

Fig. 1: J1568 study treatment schema. Eligible patients with clinically resectable PDA received the first priming study treatment Cy-GVAX-based therapy (alone [Arm A], + PD-1 [Arm B], + PD-1 and CD137 [Arm C]) 2 weeks before the surgical resection, and the 2nd priming treatment 6–10 weeks following definitive surgical resection. Patients began SOC adjuvant therapy ~4 weeks following the 2nd study treatment. SOC adjuvant chemotherapy was administered as per the standard of care at the time at the discretion of the primary treatment oncologist. The 3rd (and up to 6th) priming study treatment was administered every 28 days beginning four weeks after the completion of SOC adjuvant therapy. Study treatment was given as follows: Day 1–Cyclophosphamide (Cy) 200 mg/m2 IV (Arms A, B, C), nivolumab (PD-1) initially, 3 mg/kg, and later 480 mg IV following approval of every 4 week flat dose (Arms B, C), urelumab (CD137) 8 mg IV (Arm C Only); Day 2–GVAX intradermal (Arms A, B, C) was injected equally into six intradermal areas in both lower limbs and the non-dominant upper limb. This study began randomized enrollment to Arms A and B in March 2016. In October 2018, the study protocol was amended to add Arm C (due to limited supply of urelumab, Arm C had to enroll consecutively) as well as an optional “extended-treatment” phase. In this “extended-treatment” phase, all patients with no evidence of recurrence following the initial six priming doses of study treatment were given the option to receive additional Cy-GVAX every 12 weeks (up to 2 additional treatments), and, for Arm B and Arm C participants only, nivolumab (without ureulmab) every 4 weeks (up to six additional treatments). Full size image

Fig. 2: CONSORT diagram of patient enrollment and on-study participation. aCause of death was unknown, occurred during standard of care adjuvant course, and was outside time range of reporting SAE; bGrade 3 immune-colitis; cprotocol amendment with extended-treatment phase approved in October 2018. Full size image

Upon final analysis, 46 participants (n = 17 [A], n = 18 [B], n = 11 [C]) were enrolled and received the first dose of study treatment and were thus included in the safety cohort (Fig. 2). Forty (n = 16 [Arm A], n = 14 [Arm B], n = 10 [Arm C]) underwent subsequent, definitive (R0/R1) resection with surgical pathology confirming PDA and were thus evaluable for efficacy endpoints based on our pre-specified criteria (Fig. 2, Supplementary Note). Though the target number of evaluable patients in Arm A and B was not met, this did not inflate the type I error of the comparisons. Of the 40 evaluable patients, 37 went on to receive their 2nd treatment dose at the postoperative timepoint and standard of care (SOC) adjuvant therapy (patient patients were removed from trial for prolonged surgical recovery [>10 weeks] due to complications not related to the study drugs and another patient was removed due to metastatic progression found during recovery from surgery) (Fig. 2). During SOC adjuvant phase, one patient withdrew from treatment due to moving out of state, nine patients developed progressive disease, 1 patient died from an unknown cause, and 1 patient came off for grade 3 colitis. Twenty-five patients (n = 9 [Arm A], n = 9 [Arm B], n = 7 [Arm C]) received study treatments following the completion of SOC adjuvant therapy (Fig. 2). At the time of data cutoff, 23 patients (n = 7 [A], n = 9 [B], n = 7 [C]) completed all six priming treatments, and 9 (n = 2 [A] n = 2 [B] n = 5 [C]) entered and completed the extended-treatment phase (Fig. 2).

Demographics and clinicopathologic characteristics

Demographics and tumor characteristics were similarly balanced across Arms A, B, and C (Table 1). Study participants evaluable for efficacy endpoints had a median age of 68 years old, with a majority having R0 resections (92.5%), pT-stage 2 disease (70%), moderate (65%) or high (30%) tumor grade, and regional nodal involvement (70%) (Table 1). All patients enrolled had low tumor mutation burdens (median 0.94 mut/Mb [range 0–3.97]). While all arms had similar median adjuvant SOC therapy durations, patients in Arms A and B more often received gemcitabine (Gem) + capecitabine (Cap) as adjuvant chemotherapy (62.5% and 64.3%, respectively) while most of Arm C patients received adjuvant (m)FOLFIRINOX (70%) (Table 1). One patient in Arm A (6.3%) and 4 patients in Arm B received (28.6%) received additional adjuvant chemoradiation (cRT), compared to 0 patients in Arm C (Table 1).

Table 1 Baseline demographic and clinicopathologic disease characteristics (efficacy cohort [N = 40]) Full size table

Primary immunologic endpoints

The primary endpoint for Arms A and B were met by demonstrating tumor specimens from resected PDA patients treated with nivolumab and Cy-GVAX had significantly increased IL17A expression/TH17 infiltration in TLAs compared to tumor specimens from PDA patients treated with Cy-GVAX alone and was reported as part of correlative studies with Arms A and B, previously (19). For the biologic endpoint of CD8+ CD137+ tumor-infiltrating T cells, multiplex immunohistochemistry (mIHC) was performed on surgical specimens from 8 patients in Arm C. Tumors without an identifiable regions of interest (ROI) that contained epithelial neoplastic cells in the vicinity of TLAs were excluded from the analysis following the same standard established previously9. The results were analyzed and compared with those previously obtained from Arms A and B (n = 7 [A], n = 8 [B], n = 8 [C]). As CD8+ CD137+ T cells were very rare on pre-treatment biopsy specimens9, reporting the fold change between pre-treatment baselines and post-neoadjuvant immunotherapy samples would not be meaningful; therefore, this study only reported the density of CD8+ CD137+ T cells in the post-neoadjuvant immunotherapy PDA resected tumors.

Surgically resected tumor specimens from Arm C showed a significantly increased density of CD8+ CD137+ T cells within TLAs compared to those from Arm A (p = 0.007) and Arm B (p = 0.003), respectively (Fig. 3). Mean density of CD8+ CD137+ T cells within TLAs for Arms A, B, and C was 3.72%, 0.183%, and 27.9%, respectively. With a 152.5-fold difference in mean density of CD8+ CD137+ T cells within TLAs between Arms C and B, the primary endpoint was met. Using the overall median value (0.41% across arms) to stratify patients, >0.41% of CD8+ CD137+ T-cell density within TLAs correlated with improved disease-free survival (DFS) (HR 0.30, 95% confidence interval [CI] 0.11,0.86, p = 0.026) (Supplementary Table 1, Supplementary Figure 2) but did not reach statistical significance with OS (HR 0.61 [95% CI 0.22,1.70], p = 0.349) (Supplementary Table 1, Supplementary Figure 2).

Fig. 3: Combination GVAX, Nivolumab, and Urelumab increase infiltrating CD3+ CD8+ CD137+ and CD3+ CD8+ CD137+ GZMB+ T Cells. a Shown was one representative ROI that contains TLA and epithelial neoplastic cells in the vicinity; quantification was done within TLA and the tumor vicinity area outside TLA, respectively; mIHC marker pseudocolors: green = CD137, yellow = CD8, pink = CD3, blue = nuclei; b Comparison of the density of CD3+ CD8+ CD137+ T cells within the TLA among treatment arms as indicated. GVAX (Arm A) vs GVAX+PD-1+CD137 (Arm C): p = 0.007; GVAX+PD-1 (Arm B) vs GVAX+PD-1+CD137 (Arm C): p = 0.003. Arm A: n = 7; Arm B: n = 8; Arm C: n = 8. c Comparison of the density of CD3+ CD8+ GZMB+ T cells within TLA among treatment arms as indicated. Arm A: n = 10; Arm B: n = 10; Arm C: n = 8. d Comparison of the density of CD3+ CD8+ CD137+ GZMB+ T cells within TLA among treatment arms as indicated, GVAX (Arm A) vs GVAX+PD-1+CD137 (Arm C): p = 0.004, GVAX+PD-1 (Arm B) vs GVAX+PD-1+CD137 (Arm C): p = 0.002. Arm A: n = 7; Arm B: n = 8; Arm C: n = 8. e Representative co-registered images of multiplex IHC showing CD3+ CD8+ CD137+ T cells within a tumor ROI; mIHC marker pseudocolors: green = CD137, pink = CD3, red = CD45, yellow = CD8, blue = nuclei. f Representative co-registered images of multiplex IHC showing CD3+ CD8+ GZMB+ T cells within a tumor ROI; mIHC marker pseudocolors: green = Granzyme B, pink = CD3, red = CD45, yellow = CD8, blue = nuclei. Two-sided Mann–Whitney were performed. Significance codes are displayed as follows: *<0.05; **<0.01; ***<0.001, ns = non-significance. All data shown as the mean ± SEM (standard error of the mean). Multiplex IHC analysis was repeated twice with consistent results. Full size image

A Spearman correlation coefficient of 0.54 (Supplementary Figure 3) suggested a high correlation between CD8+ CD137+ T cells and CD8+Granzyme B (GZMB)+ T cells, a cytotoxic effector T cell subtype (p = 0.008). While CD8+ GZMB+ T cell density did not differ significantly across treatment arms (Fig. 3, Supplementary Table 1, Supplementary Figure 4) and did not correlate with survival (Supplementary Figure 4), we observed significantly increased CD8+ CD137+ GZMB+ T-cell density in TLAs in Arm C samples compared to specimens obtained from Arm A (p = 0.004) and B patients (p = 0.002) (Fig. 3). Using the overall median value (0.1% across arms) to stratify patients, >0.1% of CD8+ CD137+ GZMB+ T-cell density within TLAs, while not reaching statistical significance due to small sample size, was associated with a favorable HR for both DFS and OS compared to tumors with a CD8+ CD137+ GZMB+ T-cell density of 0.1% or less (DFS HR = 0.41[95% CI 0.14,1.17], p = 0.095; OS HR = 0.41 [95% CI 0.13,1.29], p = 0.127) (Supplementary Table 1, Supplementary Figure 5).

Efficacy

At median follow-up times of 23.1 [Arm A], 26.1 [Arm B], and 31.6 [Arm C] months (mo), median DFS (95% CI) was 13.90 mo (5.59, NR), 14.98 mo (7.95, 44.09) and 33.51mo (16.76, NR) for Arms A, B, C, respectively (Table 2, Fig. 4). Detecting true statistical significance was limited due to the small number of patients within each treatment arm. In context of this, compared to Cy-GVAX alone (Arm A), adding nivolumab to Cy-GVAX (Arm B) did not improve DFS (HR 1.09 [95% CI 0.50, 2.40], p = 0.829) (Table 2, Fig. 4). Patients treated with the combination of urelumab, nivolumab, and Cy-GVAX (Arm C) demonstrated numerically-improved DFS when compared against those treated with Cy-GVAX alone (HR 0.55 [95% CI 0.21,1.49], p = 0.242) or Cy-GVAX with nivolumab (HR 0.51 [95% CI 0.19,1.35], p = 0.173) (Table 2, Fig. 4), but did not reach statistically significance. This favorable HR trend, though again not statistically significant, persisted after controlling for age, nodal spread, and adjuvant chemo regimen (HR = 0.64 [95% CI 0.19–2.19], p = 0.478 compared with Arm A; HR = 0.48 [95% CI 0.15–1.60], p = 0.232 compared with Arm B) (Supplementary Table 2).

Table 2 Disease-free survival and overall survival comparisons between treatment arms (efficacy cohort [N = 40]) Full size table

Fig. 4: Disease-free (DFS) and overall survival (OS) by treatment arm. a DFS Kaplan–Meier curve stratified by treatment arm (efficacy cohort [n = 40]); b OS Kaplan–Meier curve stratified by Treatment Arm (Efficacy Cohort [n = 40]). Both DFS and OS were measured starting at time of first study therapy treatment. For DFS, individuals were censored at the date of last restaging scan with documented disease status if they had no evidence of disease. For patients who died within 3 months of the last scan showing no recurrence, death was counted as an event. Otherwise, patients were censored at the time of last scan showing no recurrence. Full size image

Median OS (95% CI) was 23.59 mo (13.27, NR), 27.01 mo (20.76, NR), and 35.55 mo (17.74, NR) for Arms A, B, C, respectively (Table 2, Fig. 4). Compared to Cy-GVAX alone, adding PD-1 to Cy-GVAX did not improve OS (HR = 1.11 [95% CI 0.47, 2.63], p = 0.813) (Table 2, Fig. 4). Patients treated with the combination of CD137+ PD-1 + Cy-GVAX showed a numerically-improved OS when compared against those treated with Cy-GVAX alone (HR 0.59 [95% CI 0.18, 1.91], p = 0.377) and in combination with PD-1 (HR = 0.53 [95% CI 0.17, 1.67], p = 0.279) (Table 2, Fig. 4), but did not reach statistical significance. Similar to DFS, this favorable HR persisted after controlling for age, nodal spread, and adjuvant chemo regimen (HR = 0.75 [95% CI 0.18, 3.10], p = 0.692 compared to Arm A; HR = 0.41 (95% CI 0.10, 1.62), p = 0.202 compared to Arm B) (Supplementary Table 3), but did not reach statistical significance.

On multivariate analysis (MVA), presence of nodal spread at time of surgery correlated with worse OS (HR = 2.92 [95% CI 1.02, 8.32], p = 0.045) and trended towards worse DFS (HR = 2.21 [95% CI 0.88, 5.53], p = 0.091) (Supplementary Table 2, Supplementary Table 3). Type of SOC adjuvant chemo regimen was not significantly correlated with DFS or OS in our study sample nor was tumor-stage (Supplementary Table 2, Supplementary Table 3, Supplementary Table 2, Supplementary Figure 6). Following one neoadjuvant dose of Cy-GVAX-based study therapy, 3 patients in Arm A (18.8%), 1 in Arm B (7.1%), 3 in Arm C (30%) displayed moderate pathologic responses (CAP grade 2)19 upon surgical resection (Supplementary Figure 7). Due to confounding factors affecting a significant number of these patients including incomplete SOC adjuvant treatment courses, stage pT4 disease, and limited follow-up time, a correlation between pathologic response and survival could not be meaningfully assessed (Supplementary Figure 7).

To further address the potential confounder of SOC adjuvant chemo selection, Arm C patients were also compared to a matched-historical control cohort of resected PDA patients from the Johns Hopkins Pancreatic Cancer Registry during the time when Arm C was enrolling. When matched 3:1 on adjuvant chemo regimen, age, and nodal disease status with propensity score matching (Supplementary Table 4, Supplementary Figure 8), Arm C patients reproduced a numerically favorable, although not statistically significant, HR for DFS compared to matched-historical controls: Arm C mDFS = 33.02 mo; Historical Control mDFS = 20.83 mo; stratified HR 0.72 [95% CI 0.29, 1.80], p = 0.480 (*DFS was measured starting the day of surgery for both groups, analysis stratified by adjuvant chemo type [FOLFIRINOX vs non-FOLFIRINOX]) (Supplementary Table 5, Supplementary Figure 8). We did not anticipate that this HR of DFS would reach statistical significance due to the small sample size in Arm C. Additionally, the DFS HR may have been underestimated due to the follow-up/surveillance imaging schedule being more stringent for patients on the trial compared to SOC DFS assessment in the historical cohort which carries a potential lead-time bias.

Safety

All patients had mild (grade 1–2) vaccine injection site reactions such as local soreness, induration, erythema, and/or pruritus. One patient in Arm B had their treatment complicated by grade 3 immune-related diarrhea and colitis occurring while on treatment with SOC adjuvant FOLFIRINOX (Table 3). There were no other serious adverse events related to the study regimens (Table 3). In Arm C, 1 patient had a grade 3 rash that resulted in a one-time treatment delay and there was 1 instance of a grade 2 AST/ALT elevation that resolved without intervention (Table 3). There were no unusual patterns of postoperative complications, and there were no delays in surgery due to study regimen-related adverse events.

Table 3 Summary of study treatment-related adverse eventsa (safety cohort [N = 46]) Full size table

Explorative immune analysis

Additional immune cell subtypes within TLAs following neoadjuvant immunotherapy in Arm C were analyzed (Supplementary Figure 9) and compared with previously reported results from Arms A and B9. The general CD8+ T cells increased in Arm B, but did not further increase in Arm C. Although PD-1+CD8+ T cells decreased in Arm B compared to Arm A as previously reported9, PD-1+CD8+ T cells modestly increased in Arm C compared to Arm B likely as a result of T cell activation by CD137 agonist. Interestingly, Foxp3+CD4+T regulatory cells (Tregs) were significantly increased in Arm C compared to Arms A and B, consistent with the role of CD137 in Tregs as previously suggested9. Whether this induction of Tregs would suppress antitumor immune response remains to be investigated. Analysis of myeloid cell subtypes showed that the CD137 agonist decreased both M1 and M2-like tumor-associated macrophages, but did not change tumor-associated neutrophils significantly.

We also examined TIGIT+CD8+ T cells in TLAs in post-neoadjuvant immunotherapy tumors in Arm C (Fig. 5a) and found that higher density of CD137+ T cells in TLAs is associated in a trend with lower density of TIGIT + CD8+ T cells, supporting our previously developed hypothesis that CD137 agonist treatment may overcome T cell exhaustion9. Although the general CD8+ T cells in the TLAs did not increase in Arm C compared to Arm B, the percentage of CD137+ CD8+ T cells, but not GZMB+ CD8+ T cells, among CD8+ T cells significantly increased in Arm C. This seems to suggest that a subset of CD8+ T cells, likely a subset of GZMB+ cytotoxic T cells (considering their strong correlation with CD137+ CD8+ T cells [Supplementary Figure 3]) are converted to activated effector T cells following CD137 agonist treatment (Fig. 5b, c). As previously reported9, CD8+ CD137+ T cells were essentially restricted in TLAs with minimal-to-no CD8+ CD137+ T cells seen in the vicinity outside the TLAs in Arms A and B. In contrast, this activated T-cell subtype made up 2–4% of cells in the tumor vicinity outside the TLAs within the same ROIs in PDAs from Arm C (Figs. 3e, 5d), suggesting that activated T cells may have migrated from TLAs to the vicinity of neoplastic cells.