Once, Twice, Three Times a Finding: Reproducibility of Dendritic Cell Vaccine Trials Targeting Cytomegalovirus in Glioblastoma

Glioblastoma is among the most aggressive and therapeutically challenging tumors in adults and children. Despite gross total resection, high-dose radiation, and dose-limited chemotherapy, median survival in unselected populations has been essentially frozen at 15 months in adults for decades (1). Even the aggressive implementation of novel therapies such as antiangiogenic agents (2), viral vectors (3), and tumor-treating fields (TTF; 4, 5) have provided small benefits when compared with the advances in immunotherapy being seen in other cancers. Moreover, the relatively small number of patients with this diagnosis and its dire consequences relegate the clinical testing of new agents to small, nonrandomized studies that generally thwart the reliable and reproducible establishment of efficacy.

Still, new treatments are desperately needed. As a result, our group has developed vaccines that target the cytomegalovirus (CMV) protein pp65 in glioblastoma. Several groups have demonstrated that CMV proteins are expressed in more than 90% of glioblastoma (6–10). Because these proteins have been detected specifically in glioma cells and not surrounding normal brain, they provide a unique opportunity to subvert these antigens as tumor-specific immunotherapy targets (8). Efforts to understand CMV tumorigenesis in the brain has since been undertaken (11–14), and these proteins have also been isolated in other types of glioma and pediatric brain tumors (15–17).

Here, we summarize our observations from sequentially conducted clinical trials utilizing CMV-specific dendritic cell (DC) vaccines for newly diagnosed glioblastoma. In these trials, the CMV antigen pp65 was targeted with autologous DC vaccination following standard-of-care resection and chemoradiation. Autologous DC vaccine generation and administration followed the same treatment schedule with adjuvant temozolomide (TMZ) in all three trials. Our initial study was the randomized phase II ATTAC study (18); it led to an expanded cohort trial with doubled sample size in ATTAC-GM (NCT00639639; ref. 19). In these trials, approximately 30% long-term survivor rate was evident. Within the ATTAC study, randomization to one of two vaccine site preconditioning regimens revealed that migration of DC vaccines to draining lymph nodes could be augmented, and this effect was associated with survival. Observations from these two studies led to a third, double-blinded, randomized trial (NCT02366728), which has corroborated the effect of preconditioning on enhanced migration of DC vaccines. Here, we report the evolution and reproducibility of these CMV DC vaccine trials. Furthermore, we assessed the results of exceptional long-term survival rates in our vaccine trials with clinical outcomes of published trials employing standard of care.

Prior immunotherapy clinical trials for rare, highly aggressive solid cancers such as glioblastoma have illuminated several barriers to translating effective therapies in clinical research. These barriers include exceedingly small sample sizes from the rarity of the disease itself along with the already narrow therapeutic windows that aim to be maximized in scalable clinical trial designs. The current limitations and difficulties in performing high-powered clinical trials for glioblastoma are discussed herein.

Patient selection, demographics, and vaccination frequency were similar across clinical protocols for all three clinical trials using CMV DC vaccines (NCT00639639; NCT02366728). The clinical protocols and written informed consent were approved by the FDA and Institutional Review Board (IRB) at Duke University, Durham, NC (FDA-IND-BB-12839, Duke IRBPro00003877, NCT00639639; FDA-IND-BB-16301, Duke IRBPro00054740, NCT02366728). Research in human subjects was conducted in accordance with the ethical guidelines according to the Belmont Report and the U.S. Common Rule. Adults with newly diagnosed WHO Grade IV glioblastoma, who had a gross total resection and residual radiographic contrast enhancement on postresection MRI not exceeding 1 cm in diameter in two perpendicular axial planes, and a Karnofsky performance scale score of ≥ 80, were eligible for these clinical studies. Histopathology of all specimens was initially read as glioblastoma, but this diagnosis was reconfirmed by a second board-certified neuropathologist. All patients received the same prevaccination therapy in accordance with the standard of care for these patients. All patients underwent gross total resection. Thereafter, all patients completed a 6-week course of conformal external beam radiotherapy to a dose of 60 Gray (Gy) with concurrent TMZ at a targeted daily dose of 75 mg/m²/day. Upon completion of standard therapy, all patients underwent an MRI for evidence of progressive disease. Those with evidence of progressive disease or who remained dependent on oral dexamethasone ≥2 mg daily at the time of first vaccination did not continue on protocol.

In each trial, a vaccine dose of 2 × 10⁷ mature pp65 RNA-pulsed DCs in 0.4 mL of saline was given intradermally in the groin. In all three trials, patients received the first three CMV DC vaccines biweekly. In NCT00639639 and NCT02366728 at vaccine 4, patients were randomized to tetanus-diphtheria (Td) or unpulsed autologous DCs as preconditioning and received Indium-111 (¹¹¹In)-labeled DCs for migration studies. The first vaccination occurred on day 21 ± 2 of standard-dose TMZ (SD-TMZ) cycle 1 (200 mg/m²/day × 5 days). Vaccine 4 and additional monthly vaccines until tumor progression occurred on day 21 ± 2 of successive TMZ cycles. In the second clinical trial within NCT00639639, patients received the same vaccination schedule but received granulocyte-macrophage colony-stimulating factor (GM-CSF)-containing DC vaccines as an adjuvant in lieu of Td or unpulsed DC preconditioning and received vaccinations on day 23 ± 1 of dose-intensified TMZ cycles (100 mg/m²/day × 21 days) for evaluation of greater lymphopenic effect. Vaccine 4 and additional monthly vaccines were again administered until tumor progression occurred. Each of these trials maintained a minimum of six adjuvant TMZ cycles, and if no evidence of tumor progression or intolerance to TMZ-related side effects was precluding continuation of the protocol, patients continued with vaccination and maintenance TMZ at the discretion of their treating neurooncologist.

DC vaccines were generated according to previously published methods (18, 19). DC migration studies were done at the fourth vaccination. In all trials, DCs were labeled with 10 μCi/1 × 10⁷ DC of ¹¹¹In (GE Healthcare) and divided equally across the two intradermal sites. Gamma camera images (GE Infinia Hawkeye and GE Discovery NM/CT 670) were taken immediately after injection and at 24 and 48 hours after injection to compare ¹¹¹In-labeled DC migration from the inguinal injection sites to the inguinal lymph nodes. The percent of nodal uptake of ¹¹¹In signal compared with injection site on planar image region of intensity was calculated according to the same methods as published previously (18).

Statistics were reviewed by biostatisticians and tested as described in our previous Methods (18, 19). The median and interval overall survival (OS) in standard-of-care published clinical trials (1, 4) reflect the reported survival rates from the time of randomization. Reported survival from the CMV DC vaccine trials represents follow-up through the cut-off date September 27, 2019 and, where indicated, displays patient survival rates from the time of diagnosis (resection), from randomization, or from the initiation of adjuvant therapy (if no randomization). Survival rates for the CMV DC trials include interval and median OS with updated confidence intervals for comparisons with prior published trials.

The data that support the findings of this study are available from the corresponding author upon request.

We previously published in Nature (18, 20) a small but randomized trial (NCT00639639) where patients with newly diagnosed glioblastoma were randomized to one of two vaccine site preconditioning regimens that was given prior to vaccination with CMV pp65–specific autologous DCs. Patients in one arm received intradermal vaccines in the thigh consisting of DCs pulsed with CMV pp65 RNA after preconditioning the vaccine site with Td toxoid. Both arms received adjuvant STD-TMZ at 200 mg/m² × 5 days and DC vaccines following standard-of-care temozolomide/radiotherapy and continued on the vaccine protocol as well as maintenance TMZ cycles if there was no evidence of intolerability or tumor progression. The patients in the Td arm showed enhanced migration of the DCs to the draining inguinal lymph nodes bilaterally and significantly improved survival, compared with the unpulsed DC control cohort (18). Although the CCL3-dependent mechanism was identified, similar results were confirmed in mice, and no genetic differences in either arm suggested a bias, the clinical trial was subject to a type I statistical error (false positive) given its small size.

Under NCT00639639, a separate cohort was treated under the same protocol eligibility criteria and identical treatment schedule, but patients in this separate study received GM-CSF–containing autologous CMV DC vaccines and received dose-intensified TMZ (DI-TMZ) adjuvant cycles at 100 mg/m² × 21 days, based on our prior experience with GM-CSF–containing DC vaccines (21–23) and our prior observations that profound lymphopenia following DI-TMZ permits de novo expansion of vaccine-induced antigen-specific immune responses through reactive homeostatic proliferation (24, 25). In this study, targeting CMV with GM-CSF–containing DCs and DI-TMZ resulted in antigen-specific expansion of pp65 responses in patients and again resulted in long-term progression-free survival (PFS) and OS for patients with newly diagnosed glioblastoma (19). Given the profound survival response in this second trial, we updated our long-term follow-up of these two trial cohorts and compared this follow-up data with clinical outcomes of trials utilizing standard of care.

To compare main prognostic factors in our patients with those from published trials, we summarized the distribution of age, sex, MGMT promoter methylation, and IDH mutations in the seminal Stupp and colleagues (2005) randomized trial that led to standard-of-care radiotherapy with concurrent TMZ and additionally with the Stupp and colleagues (2017) randomized trial investigating the survival benefit of TTF with adjuvant TMZ (Table 1). The median age of patients in the Stupp trials was 56–57 and 62 in our historical control cohort. Individual CMV DC vaccine patient information in Table 2 shows that the median age for ATTAC-GM patients was 55 (median age of 54 in the long-term survivors) and for patients with ATTAC-Td was much higher at 65 (59 in the long-term survivors). Table 1 illustrates the predominance of glioblastoma in males across these trials. Each cohort in Table 1 includes a mixed methylator phenotype for MGMT. Patient demographics of the long-term survivors in Table 2 further demonstrate a mixed sample of promoter methylated and unmethylated phenotypes. With the now known prognostic implications of IDH mutations in primary glioblastoma (26, 27), we retrospectively analyzed our samples and compared them with the reported data in Stupp and colleagues (2017) and found that almost all of our CMV DC vaccine long-term survivors were negative for these mutations [with the exception of two not available (NA) samples for testing]. These NA patients are shown in the figure with an * for unavailable IDH mutation and ** for unavailable MGMT methylation testing.

Adjuvant TMZ in MGMT promoter–methylated patients has demonstrated a superior clinical response compared with unmethylated patients (28, 29). Given this potential caveat in our trial patients receiving CMV DC vaccines, we retrospectively assessed the total number of TMZ cycles given for patients in our vaccine trials and arranged patient data by survival times (Table 2). Long-term survivors in both ATTAC-GM and ATTAC-Td had both methylated and unmethylated tumor samples. Because the long-term survivors showed no evidence of tumor progression, patients were able to continue adjuvant TMZ, which was given monthly with CMV DC vaccines beyond the fourth vaccine (per protocol). The number of TMZ cycles given as part of our vaccination protocols does exceed the median cycles given in the Stupp trials (Stupp and colleagues, 2005; median = 3, range 0–7; ref. 1) but does have a smaller range of total cycles [Stupp and colleagues, 2017; median = 6, range 0–51 (TTF + TMZ); median = 5, range 0–33 (TMZ alone); ref. 4]. Monthly TMZ was given as part of a DC vaccination schedule in all cohorts listed in Table 2, and patients were continued on maintenance TMZ cycles if no progression or intolerability had occurred. Because patients in the unpulsed DC cohort progressed much earlier than Td + CMV DC, the number of adjuvant TMZ cycles is also reduced.

Within two separate clinical trials, DC vaccines targeting CMV have now been associated with extended OS rates and a substantial subset of long-term survivors (Fig. 1). OS rates for cohorts in Fig 1 are calculated from the time of pathologic diagnosis (resection) given that not all cohorts shown were part of a randomized trial. Long-term follow-up of the ATTAC-GM and ATTAC-Td cohorts revealed that one-half of the Td cohort and 55% of the GM cohort were alive without tumor recurrence at 40 months from diagnosis. In fact, such prolonged survival was evident in these cohorts at 5 years from diagnosis, with 33% of the Td cohort and 36% of the GM cohort alive without tumor recurrence (Fig. 1). As displayed in Table 1, these survival rates dramatically exceed those expected with current standard of care (1), with standard of care and adjuvant TMZ (4), and with historical controls treated with standard of care and various adjuvant chemotherapies and targeted therapies (19). Interval OS and median OS rates in Table 1 reflect OS from the time of randomization or initiation of adjuvant therapy (after completion of chemoradiation) in all five studies, as these events share nearly identical timing [with the exception of Stupp and colleagues (2005) that randomized patients one week prior to initiation of chemoradiation]. The observed long-term survival rates in our vaccine studies demonstrate a median OS of 37.7 (18.2–109.1) and 38.3 (17.5–∞) in the ATTAC-GM and ATTAC-Td studies, respectively. These rates surpass the reported median OS of 14.6 to 20.9 months in cohorts receiving standard of care and adjuvant therapies.

Because of the small sample size of ATTAC and inherent high false-positive rate, observations of enhanced DC vaccine migration and associated clinical responses prompted the design of a third, larger, validation study (NCT02366728) that was conducted under an identical vaccine generation and administration protocol to ATTAC with STD-TMZ. This double-blinded randomized trial has recently completed accrual and has met the primary endpoint of migration analysis. Increases in the migration of the vaccinating DCs to the draining lymph nodes as a result of Td preconditioning has also now been recapitulated (Fig. 2).

We now report reproducibility of these findings in three sequential clinical trials and long-term follow-up on the patients treated demonstrating sustainability of the survival results. As a result, we are able to report follow-up on several patients even 10 years from initial diagnosis. Even in a population of patients matched for known prognostic factors (age, performance status, IDH mutation, MGMT promoter methylation), standard of care has not resulted in these exceptional outcomes.

Reproducibility in science is a necessity that profoundly influences our ability to uncover scientific truths. It is also a crucial checkpoint within the scientific method to confirm or refute our assumptions of natural biology. Within the basic sciences, post hoc analyses have suggested that only 10%–25% of the findings are truly reproducible (30–32). Reproducibility in clinical research has not been reviewed with such intent until very recently. However, there are major implications in clinical research for reproducibility because the ability to replicate our findings should be critical for the advancement of novel therapies to wider populations.

Reproducing clinical trial data is often negated by increasing the number of exposed patients, which reveals the greater heterogeneity within a disease cohort and challenged by the consistency of highly technical interventions. In addition, reproducibility in clinical data is also harder to assess because failure to meet prespecified endpoints for success may occur for a number of reasons beyond the reproducibility of the observed effect. So unfortunately, especially in oncology, as many as 65% of medical studies appear inconsistent when retested, and only 6% can be deemed reproducible (33). Moreover, rare cancers like glioblastoma are often also aggressive which produces a narrow therapeutic window, therapeutic nihilism, and reluctance to enroll patients in placebo-controlled trials. As a result, among approximately 200 registered glioblastoma trials identified between 2005 and 2015, nearly half were single-arm studies with sample sizes less than 50 total patients, almost 90% were open-label, and only 40% had a reportable randomization scheme (34). These constraints have made demonstrating efficacy for novel agents difficult in glioblastoma and other rare and aggressive cancers like adenocarcinoma of the pancreas.

While these CMV DC trials have yielded long-term survival rates, we do acknowledge the small sample size in each of these trials. Other active immunotherapy strategies for glioblastoma have been studied first with small, single-arm designs, as in the case of targeting the mutated EGFR, EGFRvIII. Results from early phase I and II studies for primary glioblastoma (21, 22, 35) indeed showed promise but did not yield a significant survival benefit in the phase III setting (36). A progression-free survival benefit was observed, however, in a randomized phase II study of 73 patients with relapsed vIII + GBM receiving bevacizumab and the vIII peptide vaccine (37). The first three studies utilizing peptide vaccination to target EGFRvIII in primary glioblastoma were conducted in single-arm fashion. A double-blinded, randomized phase III trial was conducted on an international scale and did not translate previously regarded survival benefits, although the eligibility criteria were notably different in the ACT IV trial compared with the prior studies (36). While this trajectory remains a concern in our DC studies targeting CMV antigens, some critical differences exist between these two immunotherapy platforms, including the applicability of the tumor antigen to a more heterogeneous patient population and the extent of antigen expression within the tumor itself. Even the early low-powered studies targeting vIII had not demonstrated such prolonged long-term survival as evidenced in our CMV vaccine studies reported here.

Limited patient numbers in glioblastoma trials also prohibit robust subgroup analyses, and these analyses can offer insight into the most optimal responses to novel therapies. Without testing of these agents in the first line, we are also left with questions of potential therapeutic efficacy if tested in certain subgroups or if initiated earlier on in the disease course. To overcome such limitations, we have sequentially conducted three separate clinical trials targeting CMV in glioblastoma. In all, despite their small number, our successive clinical trials demonstrate a consistency of outcomes, which we believe increases the credibility of CMV DC vaccine therapy in glioblastoma.

K.A. Batich reports grants from NCI and is listed as a coinventor on a patent regarding the use of preconditioning a tumor antigen dendritic cell vaccine site with tetanus and with ccl3, coowned by Duke University and the U.S. NIH and licensed to Duke University. D.A. Mitchell reports grants from NIH, National Brain Tumor Society, and Accelerate Brain Cancer Cure; and is listed as a coinventor on a patent application regarding targeting CMV antigens in glioblastoma owned by Duke University and licensed to Immunomic Therapeutics, Inc. J.E. Herndon reports grants from NIH. J.H. Sampson reports grants, nonfinancial support, and other from Annias Immunotherapeutics (work in PEP-CMV DC vaccine w/tetanus); grants and other from Istari Oncology (poliovirus & D2C7 work); other from Medicenna (brain tumors), Insera (brain tumors), Celldex (brain tumors); and is listed as an inventor on patents related to PEP-CMV DC vaccine with tetanus in the treatment of glioblastoma. No potential conflicts of interest were disclosed by the other authors.