Celgene: High glucose levels increases pancreatic cancer susceptibility to chemotherapy

Pancreatic cancer is refractory to conventional chemotherapy in the majority of cases. The median survival is 7-11 months in the most fit patients with advanced disease. While novel therapeutic targets are the focus of pancreatic cancer scientific research, understanding how to enhance the efficacy of current chemotherapeutics would be s significant advance, with immediate implications for patients. We now have compelling pre-clinical, as well as clinical evidence, that low glucose levels in the tumor microenvironment confer chemotherapy resistance. Conversely, high glucose levels may confer chemo-sensitivity. If these findings are validated with ongoing research, then these data would justify clinical trials seeking to drive up glucose levels around the time that chemotheraphy is delivered. Broadly speaking, the biologic evidence from our data supports the concept that under austere conditions, pancreatic cancer cells favor pro-survival pathways and deprioritize proliferative biology. In contrast, under nutrient abundance, pancreatic cancer cells deprioritize adaptive pro-survival (“figh-or-flight”) pathways and have the luxury and nutrient supply to engage proliferative biology. Chemotherapy is only effective under these conditions, where nucleotide synthetic metabolism becomes vulnerable to cytotoxic chemotherapy. This is supported empirically by published clinical data familiar to Celgene oncologists, where diabetic patients have enhanced response rates to nab-paclitaxel (Hirsh et al. Clinical Lung Cancer, Vol 17, No. 5, 367-74).We have just submitted a paper detailing our pre-clinical and clinical data supporting the concept that low glucose levels drive chemotherapy resistance to pancreatic cancer (under review at Gastroenterology). In this study, we demonstrate the following. First, we show that under low glucose conditions, proliferative biology is deprioritized. For instance, cell cycle data show a reduction in the proportion of cells in the S-phase. Moreover, gene expression array data show that the expression of nucleotide synthesis enzymes are downregulated under low glucose. Along these lines, DNA damage is substantially diminished in pancreatic cancer cells upon treatment with cytotoxic chemotherapy when the cells are cultured under low glucose. Moreover, reactive oxygen species levels are diminished with chemotherapy treatment when the cells have been incubated for a period of time under austere conditions. These cells are primed and ready for subsequent oxidative insults. Importantly, we demonstrate in multiple cell lines, and using multiple different chemotherapeutic agents, that PDA cells have improved survival (i.e., resistance) with chemotherapy, in the context of low nutrient culture media. These data translated in to the in vivo mouse model. A high carbohydrate data which drives up peripheral glucose levels, resulted in faster growing xenograft tumors compared to mice fed a ketogenic and calorie restricted diet (with lower peripheral glucose levels). However, the mouse xenografts associated with higher glucose levels were remarkably more sensitive to chemotherapy than tumors in mice with lower peripheral glucose levels.Most important, we analyzed clinical data from patients treated at our institution with resectable pancreatic cancer. Similar to the pre-clinical model, the tumors of resected patients (n=330) with higher glucose levels (high HgbA1C) had more aggressive features, characterized by larger size and higher metastatic lymph node burden. However, patients with poorly controlled glucose levels (despite the fact that they had more aggressive cancers) had improved survival if they received adjuvant therapy. These data collectively support the concept that tight glucose control may slow tumor growth in the absence of chemotherapy, but loose or poorly controlled glucose levels may be advantageous in the peri-chemotherapy period in order to encourage proliferative pancreatic cancer biology.While our data our extensive and provide the foundation for the described model, conclusive evidence in the clinical setting is lacking. The ideal dataset would be derived from a prospective and randomized trial of patients with advanced pancreatic cancer. Such a cohort would be large, have homogeneous first line therapies in each study arm, rigorously annotated with regards to outcomes and well vetted with regards to study design, multi-institutional, and have chemotherapy response data in addition to survival data. The MPACT study uniquely fits this description. While the statisticians from Celgene did a cursory review of the data at our request and did not find any significant associations, we request data sharing capabilities because we believe this analysis requires greater scrutiny of glucose levels prior to analysis. Since HgbA1C levels are unknown and study labs are not provided as fasting glucose levels, a comprehensive analysis of multiple glucose levels is required.

Pancreatic ductal adenocarcinoma (PDA) is now the third leading cause of cancer-related death in the United States. The disease afflicts more than 50,000 individuals each year in the U.S., and causes nearly as many deaths. Recent reports suggest that modern multi-agent regimens improve survival by roughly 5 months compared to gemcitabine monotherapy in both the palliative and adjuvant settings. Despite the marginal clinical progress, current PDA therapies have substantial shortcomings. The best chemotherapy regimens rarely extend overall survival beyond two years in patients with advanced disease (the best results approach 5%). Cures are virtually non-existent in patients with advanced PDA, and are still rare after resection. Moreover, clinical progress is attributable to additive benefits of combination chemotherapy, as opposed to scientific discoveries. Front-line drugs target nucleotide integrity or metabolism, hinting that cell proliferation is hardly a metabolic vulnerability in PDA. Herein, we detail a novel strategy to enhance chemotherapeutic efficacy in PDA by converting nucleotide metabolism into a bona fide metabolic vulnerability. PDA survives and thrives in the context of a hypovascular and nutrient deprived microenvironment, stemming from its dense stromal compartment. It stands to reason that living systems (both unicellular or multicellular) will favor survival, over proliferative pathways under severe conditions. However, conventional chemotherapy targets the nucleotide synthetic machinery, and remains the mainstay of treatment for nutrient deprived PDA. Molecular and biologic data indicate that PDA cells likely deprioritize proliferative metabolism under low nutrient conditions. As indirect evidence, PDAs on average have markedly lower KI-67 indices than chemo-responsive tumors (e.g., lymphoma and testicular cancer) with much less stroma (35% and 70%, respectively). In support of this concept, we examined cell cycle kinetics under high (25 mM, HG) and low (5 mM, LG) glucose conditions and noted a substantial decrement in the S-phase fraction of PDA cells in the latter. Interestingly, a gene expression array revealed a preponderance of nucleotide metabolism genes that were down-regulated under low glucose. In fact, the few upregulated genes were predominantly involved with nucleotide turnover and degradation, consistent with the notion that nucleotide synthesis is low priority for these cells. DNA damage due to chemotherapy was assessed for biologic relevance; PDA cells had strikingly less damage caused by gemcitabine under low glucose conditions. We next compared PDA viability under nutrient abundance and deprivation, and we observed that PDA cells proliferate slower under low glucose, yet are more chemoresistant under these conditions. These findings were inferred from in vitro and in vivo experiments, as well as patient data. It was notable that patients with poor glucose control had improved survival despite the fact that their tumors had worse pathologic features (larger size and higher lymph node burden, data not shown). Thus, our data all support a model where PDA cells grow more rapidly and aggressively under high glucose, but are consequently more sensitive to chemotherapy. We currently have plans to perform additional in vivo studies to validate this model where we will induce diabetes in mice and examine chemotherapeutic efficacy compared to non-diabetic mice.We believe this line of research is particularly important, since the best available drugs all target nucleotide metabolism, and there are no alternative targeting strategies in the imminent pipeline. Indeed, we believe this line of research is realistically foundational for a phase 1 clinical trial that can be initiated with additional clinical data support, as well as validational pre-clinical data. We foresee a pilot phase 1 study in which patients glucose levels may even be adjusted prior to chemotherapy to enhance sensitivity. We further believe that a thorough review of a dataset such as the MPACT cohort is critical to validate our current findings.Our aim is straightforward- to determine if cancer-specific outcome is associated with glycemic status in the MPACT trial, in both the gemcitabine arm and the gemcitabine+nab-paclitaxel arm. We will analyzed glycemic status as a predictor of chemotherapy response, progression free survival, and overall survival in each study group. The critical preparation step prior to analysis is data cleanup to stratify patients by glycemic status. In our experience with our own institutional datasets, in the absence of HgbA1C levels, this requires a comprehensive filtering of multiple glucose levels in order to ensure that patients in the high glucose group are truly poorly controlled consistently across the study. Likewise, patients with consistently normal glucoses must be restricted to the normoglycemic group, with minimal contamination from the high glucose group. While a superficial attempt to stratify patients was performed by Celgene statisticians, we are requesting data sharing capabilities so that we can devote added attention to this step, and maximize the opportunity to see a signal if one exists. We plan on examining the available data in the dataset first, prior to determining the best strategy to categorize patients as ‘hyperglycemic’. Important data points include serial glucose levels, patient comorbidities (e.g., diagnosis of diabetes), and medications (e.g., diabetic medicines). After appropriate stratification using strict and objective criteria, we would analyze the data in each treatment arm, looking for an association between glycemic status and cancer-specific outcome. The statistical analysis will be performed by in-house Thomas Jefferson University collaborators and investigators, and presented to Celgene investigators. Any final analysis would require approval by the Celgene staff prior to publication or presentation. Once approved, the findings will be submitted for publication.

The study will include patients data from the MPACT study, and include patients in both the gemcitabine and the gemcitabine and the nab-paclitaxel arms. Patients will be excluded if their glycemic status is determined to be uncertain based on available data.

This is an exploratory analysis, based on our results in resected PDA patients which show a survival advantage to patients with high glycemic index receiving adjuvant chemotherapy. We recognize that the adjuvant setting is not the ideal context to assess chemotherapy response, since response is just one of several factors that affect survival after resection. Thus, a thorough analysis in the metastatic setting is preferred, if a cohort is available. MPACT represents a unique dataset with this information, although the glycemic stratification is limited based on non-availability of HgbA1C levels.In preparation for a survival analysis we calculated the required sample size using the following assumptions:a (two-tailed) = .05, ß = .20Proportion of exposed (High Glycemic index) = 0.4 Proportion of unexposed (Normal Glycemic index)= 0.6Estimated Relative hazard ratio = 0.75This requires 395 events in total to reach the defined statistical significance. Patients with metastatic PDA will be categorized by blood glucose values as High/Low Glycemic index, previous diagnosis of diabetes, or diabetes medicine usage. Actual criteria will be determined after reviewing glucose level distributions across the cohort. Ideally, patients will be stratified into 2 groups defined as below or above a certain average glucose level that approximates a hgbA1C of 6.0% (a commonly used threshold for diabetes). Extreme categories will also be tested (e.g., very poor glucose control vs. others). Since HbA1C, which correlates with overall long term glycemic status, was not routinely measured. We will require repeated blood glucose measurements to be in agreement in order to define a time point as being High/Low glycemic status. Cases with agreement in blood glucose levels in 3 sequential blood test will be included in the analysis.The primary analysis will be performed in the dataset as a whole. Subgroup analyses will be performed for each treatment arm separately (1) gemcitabine and (2) gemcitabine + nab-paclitaxelSurvival Analysis (Overall and Progression free)Kaplan-Meier analysis will be performed testing for prognostic factors, including age>65, Karnofsky performance-status score, Ca 19-9, and average glycemic index (generated through criteria including glucose levels, a diagnosis of diabetes, and diabetic medication useage).Parameters which will be found to have a P value <0.2 will be included in a multivariate Cox regression model.Overall response rate will first be compared between Glycemic groups using non-parametric tests (such as Mann-Whitney). Then univariate analyses will be performed to identify all significant parameters associated with best response rate. Parameters which will be found to have a P value <0.2 will be included in an ordinal regression model to predict best response.Secondary Endpoints:For continuous parameters, at each time point, equality of variance between glycemic groups will be assessed using Levene’s Test for equality of variances. Accordingly, T-Tests will be used for comparison. Further univariate tests will be performed and all significant factors will be entered into a univariate regression model to evaluate impact of glycemic status and, subsequently, all time points will be entered in to a multivariate regression model.For Nominal parameters, at each time point, Chi-Square Tests and McNamer will be used for comparison. Further univariate tests will be performed and all significant factors will be entered into a univariate regression model to evaluate impact of glycemic status and, subsequently, all time points will be entered into a multivariate regression model. For Ordinal parameters, at each time point, non-parametric tests (Mann-Whitney U test, KS tests) will be used for comparison. Further univariate tests will be performed and all significant factors will be entered into a univariate regression model to evaluate impact of glycemic status and, subsequently, all time points will be entered in to a multivariate regression model.

 

Publication Plan

Research Team: None

Other Information

Primary and Secondary Endpoints for this Research

Publication Citation