This study aimed to determine the therapeutic role of inhibiting mTOR in preclinical models of pancreatic cancer and in patients with this disease. The mTOR inhibitor, temsirolimus, induced tumour regressions in 4 of 17 (23%) freshly generated pancreatic cancer xenografts that were, in conjunction, characterised by genetic alterations, leading to an increased activation in the PI3K/Akt/mTOR pathway. As a single agent, sirolimus resulted in a 26% 6mSR in patients with previously treated pancreatic cancer, but did not result in tumour regressions in any patients. Contrary to our hypothesis, there was no indication that patients with higher activation of the pathway, as measured by the selected biomarker in this trial, did better.
The role of preclinical models in cancer drug development continues to evolve. In the era of cytotoxic agents, models were used to show tumour growth inhibition in a randomly selected group of rapidly growing xenograft models (Berger et al, 1990
). More recent efforts, however, include testing agents in larger collection of xenografts representing a disease of interest, not only to gauge potential activity but also to understand predictors of efficacy (Perez-Soler et al, 2000
; Rubio-Viqueira et al, 2006
; Houghton et al, 2007
; Jimeno et al, 2008a
). With this goal in mind, we and others have started conducing large-scale phase II-like preclinical studies in pancreatic cancer. In this work, we followed such an approach to test the activity of mTOR inhibitors in pancreatic cancer.
One important consideration in this work, however, is that the level of activity in preclinical models that predict clinical efficacy is not established. Classically, a T/C of 40% has been considered supportive of anti-tumour efficacy and used as a threshold to move drugs to the clinic. This criterion, in our opinion, is too unrestrictive and overestimates the expected clinical results. Indeed, if one applies the commonly used RECIST clinical criteria of response, a T/C of 40% would be disease progression. We have therefore applied a more restrictive criterion and consider activity if there is a tumour regression. The current situation is that there are thousands of anti-cancer agents available but yet very little work in the clinic. A more selective preclinical approach is needed to prioritise which drugs to develop in patients. Efforts to better establish levels of preclinical efficacy that predict positive clinical outcome are, indeed, needed.
Another important, and not established, question is which level of preclinical activity is required to justify conducting a clinical trial. It is remarkable to note the important parallelisms between the preclinical and clinical study with ~25% cases meeting the pre-specified primary objective in both studies. Notwithstanding that our patients were very heavily pretreated, a 25% 6mSR in the second-line pancreatic cancer is low and does not warrant further development of this drug as a single agent in this disease unless a predictive biomarker is identified. This has been indeed the finding of another recently published trial (Wolpin et al, 2009
). In retrospect, we should have established a higher threshold of activity in the preclinical study before advancing the agent to clinical development. Given the larger number of available agents in clinical development and the limited resources, we propose that only agents with preclinical activity significantly greater than the clinical activity of interest are selected for clinical development.
The second major goal of this work was to identify biomarkers of activity. The preclinical data show that tumour regression after treatment with temsirolimus was limited to xenografts with high activation of the PI3K/Akt/mTOR pathway, as shown by high activation of p70S6K. There is a whole body of literature supporting the role of PTEN
losses in activation of the PI3K/AKT/mTOR pathway in a variety of tumours including prostate, breast and glioma (Li et al, 1997
). Our finding that 1 of 17 (5%) xenografts had PTEN
deletion is in consonance with previous evidence that PTEN
losses are infrequent in pancreatic cancer (Okami et al, 1998
). The fragile histidine triad (FHIT
) gene, a tumour suppressor, is lost in most malignancies, including pancreatic cancer (Simon et al, 1998
; Huebner and Croce, 2001
). Loss of FHIT
, leads to increase AKT activity both in vitro
and in vivo
(Semba et al, 2006
). These findings are supported by the results from the GSEA, showing that sensitive xenografts were enriched in pathways with high content of genes involved in the PI3K/Akt/mTOR pathway. Thus, the striking correlation between the drug activity and pathway activation, as measured by phospho-p70S6K, a downstream mediator of the pathway, is expected on the basis of the current knowledge of this pathway.
Although the overall level of activity of mTOR inhibitors in pancreatic cancer was modest, the finding that the activity could be linked to a biomarker was critical to support the conduction of the clinical study. If the 25% of patients who are sensitive can be identified upfront, the clinical development of the drug in pancreas cancer is feasible and likely to be successful. Unfortunately, we did not observe such a relationship in the clinical trial. Several factors can be considered to explain this issue. First, it could be that the concentrations of sirolimus achieved in patients are inferior to those achieved in mice. However, we have used the maximum tolerated dose of sirolimus based on data from a previous phase I study published by our group (Jimeno et al, 2008b
). Second, PDA is characterised by an intense desmoplastic reaction that may lead to decreased intratumoural concentrations of sirolimus and thus explain lack of activity (Olive et al, 2009
). Third, it could be that the biomarker selected, that is, activation of p70S6K, is not valid. We doubt, however, this is the reason. The notion that activation of the PI3K/Akt/mTOR pathway leads to p70S6K activation is supported by multiple studies (Hennessy et al, 2005
; Riemenschneider et al, 2006
). In addition, several trials with mTOR inhibitors suggest that this is a candidate marker (Galanis et al, 2005
; Cloughesy et al, 2008
). More likely, the negative results are because of trial design and technical reasons. We have used an IHC technique applied to archival tissues and we really do not know the preservation of the phosphorylated antigen over time in these materials. This is a recurrent problem in clinical trials that relay on archival tissue for which no satisfactory solution has been proposed. In addition to this, other investigators have questioned the performance of the specific antibody we used (anti-Thr389 S6K) for reliable quantification of the activity of p70S6K on paraffin sections (Cloughesy et al, 2008
). We assessed p70S6K in primary tumours and treated metastatic disease. We do not know whether the process of generating tumour metastasis results in variations in this biomarker. In retrospect, a better strategy would have been to perform a fresh biopsy from a metastatic site and to use the same ELISA method that we have used in the preclinical study. It is clear that this approach adds morbidity, complexity and costs, but it may at the end be the only strategy to answer these questions. Finally, it could just mean that the preclinical model is not predictive at all of clinical activity and what is learnt in one does not predict what happens in the other.
In summary we have integrated a preclinical and clinical trial of mTOR inhibitors in pancreatic cancer. The agent resulted in ~25% of cases achieving a positive response. In the preclinical study, activity was linked to pathway activation. This finding, however, was not observed in the clinic. The most likely explanation for this negative result is technical in nature, but other factors cannot be ruled out. In the absence of a well-defined biomarker to select patients, the overall level of activity does not suggest that these drugs as single agents would be effective in this disease.