Translating breakthrough science into breakthrough medicines is a daunting task. It’s also worth every nail-biting minute waiting for the next result that brings you closer to helping a patient in need. We started miRagen based on some amazing laboratory results and the promise that targeting a powerful new regulatory pathway controlled by small RNA molecules (microRNAs or miRs) offered the hope of intervening in complex diseases that were difficult to address with more traditional drug mechanisms.
Having worked away at this goal for a number of years, I thought it would be instructive to share a few of the valuable lessons we learned in developing a focused process for converting interesting phenomenology into drug leads. While the drug modality we are exploring is new, many of the key considerations are not, and often translate back to general innovation in the drug development paradigm.
Carefully reproduce any external discovery that will consume your time / capital
It has become abundantly clear that the successful reproduction of literature in life sciences is a major issue. Since breakthrough observations reported in the scientific literature form the basis for entrepreneurial pursuits in this area, it is absolutely vital that a rigorous analysis of the foundational discovery is rapidly conducted. It should also be conducted for any new external observations that would lead you down the drug development path. My last post was directed to a specific example with several interesting twists and turns, so I won’t dwell too much on the topic.
When we identify a new observation in microRNA biology that could serve as the basis for a development program, the very first step is to independently reproduce the findings. The work should be done in a blinded fashion, whether in your own shop, with an academic collaborator, or a CRO. In addition to reproducing the beneficial phenomenology, focus intensely on demonstrating drug target engagement and correlate this to the effects on disease. Employ multiple active agents and appropriate negative controls and reference compounds if available.
Our experience has shown that the likelihood of accurate reproduction is very low, in alignment with that reported by others. The low reproducibility is exacerbated for nucleic acid targeting (antisense, siRNA, microRNA, lncRNA, etc.) because of general access to oligonucleotide based tool compounds. This often prompts early studies by academic scientists with suboptimal means to truly address the problem. It is also absolutely vital to incorporate a set of robust negative controls and carefully correlate effects on the target with phenotypic outcomes. Finally, powering studies sufficiently and applying the correct statistical analysis methods is of utmost importance. We have derived a rigorous standard process in microRNA target validation that gives us enough confidence in the outcomes to make the call to kill a program and also often identify the cause of the spurious results originally reported.
It is vital that you commit to act quickly to kill projects that do not fulfill an agreed upon go / no go decision. However, when a well-designed reproduction experiment is un-blinded and recapitulates the originally reported observations it provides a solid foundation for the next step of progressively de-risking the drug / drug target pair.
Focus on measurable markers of drug / target interaction
As soon as there is confidence that modulating the drug target can lead to effects on disease drivers it is vital to embark on a robust effort to identify surrogate markers of drug action. We initiate the process by identifying “pharmacodynamic biomarkers of target engagement”. Now that’s a heck of a mouthful, but it’s simply looking for a readout that can be experimentally measured to demonstrate that the drug candidate was able to enter the desired cells and bind to the intended target(s). In more traditional drug modality approaches, this practice has become fairly standard. But the challenges in identifying the right pharmacodynamic markers when targeting new modalities (and particularly in microRNA targeting) are magnified.
This step is particularly important in nucleic acid drug development because we can use these markers to optimize lead molecules and evaluate delivery technologies where these are employed. For microRNA targeting, directly measuring binding of the drug to the microRNA is essentially impossible. This is because the direct measurement of microRNA binding by the inhibitor (antimiR) is subject to an artifact that occurs when cells are lysed prior to performing the binding measurements.
Much of the oligonucleotide in any tissue or cell after systemic administration is trapped in a non-productive compartment (endosomes, connective tissue, etc.) and when the biological sample is is solubilized for analysis, the antimiR is freed from those compartments, binds to the microRNA and grossly overestimates the amount that was actually available to bind in the intact cell and tissue. So we need to go a step deeper to ensure that we are monitoring the true fraction of oligonucleotide drug candidate that is able to engage its target in the intact cell.
After using bioinformatics to predict the genes whose expression could be controlled by the microRNA, we then empirically determine via expression profiling those genes that actually are controlled by the microRNA. By adding more of the microRNA or inhibiting the native microRNA, we identify a set of modulated genes that provide confidence that the drug candidate is binding to its intended target. The resultant set of gene markers becomes a barcode for drug uptake and target engagement as drug candidates are assessed and optimized in a host of preclinical studies and in early clinical studies.
While readouts of target engagement are a vital piece of the puzzle and may serve as mechanistic biomarkers, it is important to identify robust mechanistic biomarkers that tie drug effects to beneficial disease treatment. Because microRNAs control complex network biology, a robust marker of target engagement may be difficult to succinctly link to the disease. It is then vital to make a linkage through secondary or tertiary effects to a mechanistic biomarker of drug action for use in both preclinical and clinical studies.
Reliable biomarkers provide an invaluable tool for assessing whether the fundamentals behind your drug hypothesis are valid in humans and most importantly provide a means to mechanistic validation in early clinical studies. Obtaining that early mechanistic proof in humans is undoubtedly the most value creating event in the development of a new drug modality.
Use animal models only as absolutely required
Many of our core programs are based on genetic manipulation studies (particularly knockouts) in disease stress models that suggest microRNA manipulation has a potential therapeutic benefit. We typically follow these findings by treating normal mice with a microRNA modulator (antimiR or promiR) and then demonstrating that the potential benefit can be recapitulated. By phenocopying the genetic result with a pharmacological treatment, we feel confident that this is a good target to pursue further. Should we find human genetic evidence to support the hypothesis, it is a real nirvana.
After foundational studies are recapitulated, it is best to pursue mechanistic studies in human cells in culture to reinforce the drug action hypothesis and then conduct the minimal instructive set of studies in animals to progressively de-risk the drug candidate.
I’m not suggesting that animal testing isn’t useful, it’s the center point of pharmacokinetic and pharmacodynamic characterization, and safety assessment. However, the capacity of animal models of disease to foretell the human response to substances has been broadly questioned with some studies intimating a very low (or non-existent) predictive value, either positive or negative.
Using a highly contrived animal model of disease to make gatekeeping development decisions based on drug candidate efficacy is a bad practice. It suggests that the model is able to predict human clinical response with an extremely high negative predictive capacity (not true). As long as the drug candidate’s molecular mechanism is supported in disease animal models as well as human cells and the drug candidate has an adequate safety profile, it should be advanced into human clinical evaluation.
The contrived nature of many animal models of human disease means that employing multiple models does not enhance their predictive value. Such an approach not only increases the likelihood of difficult to interpret negative data, but also requires scarce capital to be expended for no commensurate reduction in risk. In the end, it is about the balance of confidence in the molecular mechanism at play to translate to man and a biomarker strategy to rapidly assess potential drug activity in early trials.
***
There are many additional important lessons learned from our endeavors in finding good microRNA targets and oligonucleotide based drug leads. The unique biophysical properties of short antimiR’s are neither small molecule nor biologic like and require a new way of thinking about drug properties and methods for analysis. Careful considerations need to be made about known potential off-target effects (like immune-stimulation) and of course productive delivery to the desired cell type.
Finally, the complexity of the systems under the control of microRNA’s surpasses the widespread view of gene expression control by binding just messenger RNA. Interactions between microRNA, pseudogenes, long non-coding RNA, messenger RNA, circular RNA and RNA binding proteins produce an amazingly complex, context dependent control of pathway biology. By taking a systematic approach to addressing these complexities we are beginning to define the properties of a multifaceted new drug modality like microRNAs.
Moving a potentially exciting new drug modality from promise to proof of concept requires innovative ways of rapidly addressing whether the drug candidates can work in man. Building streamlined processes to assess the potential of new targets and rapidly building the appropriate package to support their human clinical evaluation is a key to adoption of the modality and its logical use as a source of meaningful new drugs.