The overwhelming majority of pharmaceuticals fail when tested on humans, particularly in oncology, where 95% of drugs tested in phase-I trials never make it to market. Kim Thomas speaks with Italo Poggesi, scientific director of global clinical pharmacology/quantitative sciences at Janssen, about how that figure might be brought down.
Researchers have been working for years on improving the design of pre-clinical experimental models. Italo Poggesi, scientific director of global clinical pharmacology/quantitative sciences at Janssen, suggests that one reason for failure at the clinical stage could be due to the fact that “we often looked at individual preclinical experiments”.
For example, he says, until relatively recently, the systemic drug exposure (the plasma concentrations achieved and maintained after dosing) of candidate drugs in pre-clinical models was rarely measured: “The main focus was to observe a significant tumour reduction in the group of treated animals compared with a control group of untreated animals.”
Could the failure to work in humans have been predicted earlier? When Poggesi looked retrospectively at pre-clinical pharmacokinetic and anti-tumour activity data, he found, for instance, that in one case, at the maximum tolerated dose defined in the phase-I trial in human subjects, “The systemic exposure was much lower than the active exposure observed in experimental models, which indicated that the probability of success of that compound was minimal.”
It is now recognised, says Poggesi, that to successfully observe anti-tumour activity in clinical trials, the systemic exposure in human subjects needs at least to reproduce that achieved in pre-clinical models where significant anticancer effect was observed.
Progress made
In the past few years, oncology researchers have reached a much better understanding of the pharmacology of targeted anti-tumour candidate drugs.
This, in turn, helps to improve understanding of why certain drugs fail and will, eventually, improve the success rate in the development of targeted therapies. “When the candidate drug provides an adequate level of target engagement, a cascade of events follows which may lead to the demonstration of the regulatory clinical end point, typically represented by longer survival of patients,” Poggesi says.
“The discovery research programmes are now generating a larger amount of information on this cascade of events, providing confidence on the pharmacology of the new compounds and on the prerequisites to elicit a clinically relevant response, such as dosing, systemic and tumour exposure to anticancer agents.”
The use of mathematical models could provide a big step forward in the ability to translate pre-clinical knowledge to clinical situations. These models, which “provide the link between systemic exposure, target engagement and the subsequent cascade of events” can, says Poggesi “provide a powerful tool to improve the success rate of anticancer candidate drugs.” By complementing the mathematical models with statistical models, researchers can account for variability and uncertainties in the parameters and in the translation from pre-clinical to clinical settings.
Historically, mathematical models have had an important impact on our understanding of cancer. Poggesi cites the pioneer work of Anna Laird in the 1960s on the dynamics of tumour growth in mice and rats, as well as the more recent papers from Richard Simon and Larry Norton testing the dose-density hypothesis, which states that the rate of cancer cell death in response to treatment is directly proportional to the tumour growth rate at the time of treatment.
The dynamics of modelling
In recent years, PK/PD modelling in particular has become an integral part of the drug development process, because it combines two elements: the pharmacokinetics (PK) element of the model describes how the drug concentration increases and decays over time, while the pharmacodynamics (PD) element describes the relationship between the drug concentrations and its anti-tumour efficacy.
What mathematical models offer now is the ability to estimate “experiment-independent parameters describing the anti-tumour potency of the anticancer candidate drugs.
These parameters can then be translated from a pre-clinical setting to a clinical one. “You can use different mathematical models – for instance, those describing the time course of tumour size, those describing the time-course of pharmacological end points or tumour markers and those describing the clinical response,” Poggesi adds. By doing this, researchers can establish if and how all these observations are consistent and correlated.
Poggesi argues that there are no alternatives to mathematical models when it comes to predicting the effect of a candidate drug in human trials based on pre-clinical experiments: “The use of models can provide a better insight into cancer, confirming and driving new hypotheses,” he says. “The multi-scale models integrating macroscopic tumour growth with what is happening at a biochemical and cellular level are a good example. In the later phase of clinical development, pharmaco-statistical models can be used to build correlations between exploratory markers of activity and surrogate clinical end points with the overall survival, which can speed up drug development and the access to new therapies.
The mathematical models can also be implemented to optimise the experimental design, thus decreasing the use of animals in the discovery phases.”
Although they offer a good deal of promise, the increased use of mathematical models hasn’t so far been reflected in an improved success rate in clinical trials. Poggesi points out that the attrition rate (the percentage of the compounds entering the clinical development that fail before achieving the market) has remained unchanged: for compounds developed in the period 1991–2000, the failure rate was 95.0%; for 2006–2015, it has been estimated at 94.9%.
He hopes, however, that this will change in the near future: the number of published modelling approaches has increased from fewer than 50 papers a year before 2005, to 450 papers published in 2016, of which approximately half are related to experimental models in pre-clinical species.
It’s important, he says, to find the right balance between different mathematical models. There is a ‘spectrum’ of models used at the pre-clinical stage, from empirical ones, which are typically descriptive, to systems pharmacology models, which are designed to include all the information available on complex biological processes, such as “tumorigenesis or potential responses of tumours to therapies”.
Empirical models, he says, are limited in scope and can rarely be used for extrapolating results from pre-clinical to clinical, or from one indication to another, while the systems pharmacology models are suited to providing “extrapolations and translations, but suffer in many instances from a lack of quantitative knowledge on the many parameters they are built on”. The performance of these models, he adds, depends on a complete knowledge of the processes under evaluation, which can never be guaranteed.
Other useful approaches, he adds, include semi-mechanistic or semi-physiological models: “These models include key features of our knowledge, and this allows us to disentangle the model parameters that are only dependent on the drug – for instance, the potency of a compound in inhibiting or activating a particular pathway – from the parameters that are dependent on the experimental system adopted, quantifying for instance how tumour growth inhibition translates to survival in xenografted mice or in human subjects with cancer.”
These semi-mechanistic models, he explains, generally include a smaller number of parameters than the system pharmacology models. This makes them suitable for translation from pre-clinical to clinical trials, as the drug-specific portion is the same, and only the system-related portion has to be updated. The more complex systems-pharmacology models, Poggesi says, are instead typically used to explore scientific scenarios, such as ‘What if I block this pathway?’ or ‘What if I engage these two targets simultaneously or sequentially?’
Integrate the knowledge
The more we integrate clinical data with experimental models, the more we will be able to predict with some accuracy what will happen in the subsequent drug development. In the long term, Poggesi believes, the quantitative integration of knowledge, including pharmacological response, tumour response and survival end points via pharmaco-statistical modelling, will be “key for reducing the attrition rate in drug development and giving priority to the programmes having the highest probability to benefit patients”.
Janssen’s own work shows that mathematical modelling at the pre-clinical stage can have a strong predictive power. “Analysis of data from pre-clinical experiments predicted that we should detect clinical activity at a certain threshold concentration,” says Poggesi. “Based on the first pharmacokinetic data in humans, it was predicted that this threshold concentration was achieved and maintained at dose levels equal or above 4mg daily. When the compound was given at 6mg daily we had the first partial responses in our phase-I trial. That was a pretty accurate prediction.”
Poggesi adds that use of predictive models will increase, and that they will eventually make the drug development process more efficient – with less attrition – and more effective in tackling cancer. “The proposed pharmacostatistical modelling approaches are increasing in number and this improves the comprehension we have of cancer and of the effect of therapeutic interventions,” he says.
“These approaches already influence decision-making in the drug research and development processes. Pre-clinical predictive modelling is making drug discovery and development more efficient and objective.”