Can QSP Save Lives? Lessons from a Trial Debacle

Can QSP Save Lives? Lessons from a Trial Debacle

The notion that volunteers could be harmed in a clinical trial is every drug developer’s worst nightmare. Earlier this year, the drug company, Bial, investigated inhibitors of the enzyme fatty acid amide hydrolase (FAAH) in clinical trials as a treatment for pain. Tragically, one person in the volunteer group died, and six patients were hospitalized. Could a better understanding of FAAH inhibitors’ biology have potentially averted this debacle?

Quantitative systems pharmacology (QSP) is a mathematical modeling approach that integrates our understanding of biology and drug pharmacology. In this blog post, I’ll discuss why applying a QSP approach to drug programs can yield important—maybe even life-saving—insights.

Why use a systems pharmacology approach in drug development?

A systems pharmacology approach to drug discovery helps tackle the problem of high attrition and low return on investment. By all metrics, the productivity of pharmaceutical R&D is declining. One of the biggest reasons for programs failing is attrition in Phase II due to failure to show efficacy or unexpected toxicity. A clear explanation for these findings is an inadequate understanding of the consequences of perturbing a complex biological system.

The challenge of understanding complex systems is not unique to drug discovery or biology. This problem has been observed in other areas, such as engineering and physics. These fields have successfully employed mathematical modeling to better understand complex systems. We can do the same in biology. For example, mathematical pharmacology models can be used to understand the pro and cons of FAAH as a drug target.

Targeting the endocannabinoid system for treating pain

CB1 and CB-2 are cannabinoid receptors. The CB-1 receptor is primarily expressed in the brain whereas the CB2 receptor is mainly expressed in the periphery. These receptors are the pharmacological target of cannabis’ active ingredient, tetrahydrocannabinol (THC). The endogenous cannabinoids—anandamide (AEA) and arachidonoyl glycerol (2AG)—are also CB1 and CB2 agonists. Increased levels of anandamide attenuate firing of pain-sensing neurons (nociceptors), and some argue there is evidence showing efficacy of cannabis preparations for pain.

Given that anandamide is degraded by FAAH, this information led a project team at Pfizer to hypothesize that inhibiting FAAH will increase the local concentrations of anandamide which in turn will decrease nociceptor firing and attenuate pain. Subsequently, they identified FAAHi PF-04457845 as a potent FAAH inhibitor. This molecule was tested in 11 rodent models of chronic pain or acute pain. Positive results were attained in six models. The interpretation of these results was that the FAAH inhibitor addressed the specific pain types represented by the models.

So far, this drug program seemed to be on the right track: interesting biology, a promising compound, and a biomarker—anandamide—that could be measured in pre-clinical species and patients. Immediately prior to the phase I evaluation of PF-04457845, the clinical pharmacology team came to me with the following questions:

Will we test the pharmacology fully and the doses that we’re intending to use in patients? Can we predict the reception occupancy and the level of occupancy required for drug efficacy?

Building a systems pharmacology model of the endocannabinoid system

We used a systems pharmacology approach with internal and external data to address these questions. First, we performed a literature review of data on the disposition and kinetics of the biosynthesis/degradation of AEA. The literature search also revealed many open questions on this biological system. For example, anandamide is a CB-1 agonist, which should attenuate pain. Anandamide is also, some evidence suggests, a TRPV-1 agonist, which is associated with causing pain, rather than inhibiting it. Thus, it was unclear whether perturbing anandamide would cause or block pain. Likewise, 2AG and anandamide both agonize CB-1, yet how they interact with one another is unclear. Finally, although qualitative evidence linked receptor occupancy and nociceptor firing, this relationship had not been quantified. Therefore, specifying what clinical trial success would look like, in terms of receptor occupancy versus time in the patients, would be difficult.

Using the information from the literature, we next constructed a model comprised of five components.

  • The first component was a model of XEAs (the ethanolamides including anandamide that could be important to the system) disposition. This model described the disposition of those XEAs in terms of their tissue binding and compartmental localization.
  • The second component related how FAAH would bind anandamide and other XEAs.
  • The third component described the biosynthesis and degradation of XEAs.
  • The fourth component detailed the clinical PK and pharmacology of the FAAH inhibitor.
  • The fifth component described anandamide’s receptor interactions.

These components were integrated into an ordinary differential equation model that described the biology of the system.

Refining the model to fit clinical observations

After building the model, we simulated the anandamide time-concentration profiles upon dosing volunteers with the candidate molecule. We predicted that the relative anandamide plasma concentration would rise quickly, peak briefly, and then decrease rapidly. However, when the volunteers were dosed, the plasma anandamide concentration increased relative to baseline, plateaued, and then slowly decayed back to the baseline. The anandamide concentration plateau was independent of drug dose.

What was the model missing that would explain the plateau? Something other than FAAH must be clearing anandamide! Updating the model to include an additional anandamide clearance mechanism resulted in an anandamide plateau similar to the observed data.

It turned out that many potential candidates― cytochrome P450, lipoxygenase, cyclooxygenase, etc—could clear anandamide in addition to FAAH. QSP models provided a quantitative framework to examine which processes could clear anandamide as well as their kinetic properties. We matched the candidates with the known properties of the different clearance processes and asked which molecule best fit the model. The most likely candidate was the enzyme n-acylethanolamine-hydrolyzing acid amidase (NAAA).

This additional clearance process was then added to the model. We performed simulations and compared our predictions with the observed patient data. The updated model accurately described the time course and dose response for anandamide and other ethanolamides.

The riddle of receptor occupancy

Our data suggested that our FAAH inhibitor could increase the plasma concentration of anandamide to ~10 nanomolar (nM). But anandamide binds human serum albumen strongly with a free fraction of ~0.1%. In vitro experiments suggested that the EC50 of anandamide was 300 nM. Assuming that the free drug in the aqueous phase drives pharmacological effects, these assumptions translate into anandamide having much less than 1% occupancy at the CB-1 receptor. How can the FAAH inhibitor have an effect with minimal CB-1 receptor occupancy?

An explanation could be that anandamide does not align with the standard free drug hypothesis. Rather than binding the reception from the aqueous phase, anandamide could intercalate within the lipid-rich membrane, diffuse to the CB-1 receptor, and express its pharmacology.

This hypothesis would explain observing pharmacology at these concentrations of anandamide if it is very hydrophobic. In this scenario, the steady state concentration of anandamide would be much higher within this lipid environment than the hydrophobic environment. Consistent with this, it was shown that anandamide has a very high Log D7.4 (equal to approximately 6). At steady state, it would be orders of magnitude more concentrated in the hydrophobic environment than in the hydrophilic environment. In the case where an agonist partitions into a hydrophobic phase and then binds to the receptor, the observed dissociation constant (KD) can be calculated from the binding of anandamide to the CB-1 receptor, multiplied by its partition coefficient. Under these assumptions, the model predicts pharmacology with picomolar free AEA in the aqueous phase. This hypothesis may explain why low free levels of anandamide could drive meaningful CB-1 receptor pharmacology. Next, we used the model to determine whether the proposed FAAH inhibitor doses blocked FAAH and what CB-1 receptor occupancy occurred.

The model predicted that the FAAH inhibitor doses resulted in greater than 96% FAAH inhibition over extended time periods. However, the model also predicted that CB-1 receptor occupancy would plateau at ~25%, independent of the FAAH inhibitor dose. Giving more FAAH inhibitor did not increase the CB-1 receptor occupancy above 25%; it only extended the time course of the effect. CB-1 receptor occupancy was limited by the clearance of anandamide by something else, which we hypothesized was NAAA. However, the question of whether 25% receptor occupancy could produce a pharmacologically relevant effect in pain remained open.

Our recommendations based on the model

To recap, the traditional drug discovery perspective supported a clear rationale for this drug program. First, using cannabinoids in pain has clinical precedent. Second, our biomarker could be measured pre-clinically and showed a clear PK/PD response. Third, we had positive data from animal models of pain.

However, the systems pharmacology perspective was quite different. Our hypothesis suggested that attaining meaningful CB-1 receptor occupancy was possible. Yet questions remained about whether 25% receptor occupancy was adequate to decrease nociceptor firing. Also, receptor occupancy cannot be increased by raising the dose of FAAH inhibitor because CB-1 receptor occupancy was likely limited by a different enzyme clearing anandamide. In addition, we didn’t know how much nociceptor firing would need to be decreased to reduce pain. Finally, the endocannabinoid system is very complex and may contain redundancies. While we examined the interplay between anandamide and the CB-1 receptor, we didn’t consider other elements in the system such 2AG. Thus, we concluded that FAAH was a relatively high risk drug target.

The feedback to the team prior to the Phase 2 trial was that the model suggested that the planned doses would almost completely inhibit the FAAH enzyme. But the alternative clearance pathway—potentially by NAAA—for anandamide would limit the effect of any FAAH inhibitor. In addition, we had no quantitative data linking CB-1 receptor occupancy to the pain outcome. The model predicted some receptor occupancy, but we didn’t know if it was adequate for drug efficacy. Thus, the success criteria for our FAAH inhibitor was an open question.

Paul Morgan and his colleagues made a similar warning in their 2012 Drug Discovery Today paper. If you can’t show that your pharmacophore gets to the target and expresses the pharmacology, you risk running an uninterpretable experiment. In the event of a negative result, you won’t know whether it was because the drug didn’t get to the target and express the pharmacology OR whether the mechanism that the hypothesis was based on was invalid. You can’t discriminate those two explanations without having an assay of pharmacology. Ultimately, our recommendation from our model was that the project risked being uninterpretable.

The FAAH inhibitor moves into Phase 2 studies

Pfizer tested PF-04457845 in a Phase 2 study in osteoarthritis patients where it failed to show efficacy. This clinical failure raised many questions. While the results from animal models were positive, they didn’t translate clinically. Also, we don’t know whether the drug expressed pharmacology. The systems pharmacology perspective was that we needed to build our knowledge and technology before any further clinical studies occur.

In a CPT:Pharmacometrics and Systems Pharmacology paper, we stated that any future progression of FAAH inhibitors should be accompanied by developing technologies that enable demonstrating pharmacology at CB1. We also needed to better understand the relationship between receptor occupancy and attenuation of nociceptor firing together with greater knowledge of endocannabinoid system functioning in humans.

Lessons from the Bial trial

Other companies subsequently looked at FAAH as a pharmacological target for pain. Like Pfizer, they were unable to demonstrate efficacy. In the Bial trial aftermath, the French authorities published a review of the clinical trial. In my view, the authorities concluded that the FAAH inhibitor was responsible for the toxicity and that it was dosed to concentrations that would saturate FAAH. Ultimately, the toxicity’s origin remains unclear. Likewise, we don’t understanding the complex biology enough to predict the consequence of perturbing it.

The way forward

A better understanding the endocannabinoid system could support developing useful drugs. Our model of the system only addressed the anandamide facet of the biology. But other elements, like 2-AG, could drive CB-1 signaling. Deeper knowledge of endocannabinoid biology will require extending the existing systems pharmacology model.

How systems pharmacology approaches adds value to drug programs

The system pharmacology approach complements traditional drug discovery. It identifies potential risks and suggests testable hypotheses to address these risks. QSP models also highlight crucial deficits in our understanding of biological systems. By leveraging the insights provided by mathematical models, we can better define the success and safety criteria of clinical drug trials.

All information presented derive from public source materials.

Learn more about QSP

My colleagues, Drs. Steve Toon and Piet van der Graaf, wrote an article in Clinical Leader explaining how QSP has the potential to improve pharma R&D productivity. I hope that you’ll read it and let me know what you think in the comments section.

Neil Benson

About the Author

Neil has 20 years’ experience of working in modeling and simulation in the pharmaceutical industry at SmithKline Beecham and Pfizer. During this time, he held a number of senior leadership positions, most recently as Head of Systems Pharmacology at Pfizer, Sandwich. He was awarded the Pfizer Upjohn award for innovation in developing dose prediction methodology. He has extensive experience of using modeling and simulation to address questions of critical importance in drug discovery including; clinical dose prediction, optimal target identification and biomarker selection and has authored ~ 30 papers and patents. From 2011-2015 he was Founder and Director at Xenologiq Ltd, a consultancy company focused on the productive application of PKPD and Systems Pharmacology in drug discovery. In 2015, Xenologiq Ltd became part of the Certara group of companies, where he is now Head of QSP Operations. Educated at the University of East Anglia (UK) and the University of Lund (Sweden), Neil has a First Class Honors degree and a PhD in enzymology.