Increasing challenges to trial recruitment: Is it time to change the inclusion criteria for investigational compounds, not just for study participants?

We read with interest the paper by Kerr et al.1 on the increasing challenges to trial recruitment and conduct. The authors identified a steady reduction in the number of participants per site, leading to a need for more study sites. Solutions proposed in that paper and in a subsequent critical review focused on substantial changes in trial design, including expanded eligibility criteria and time-to-event trials that observe participants on blinded treatment until a key event occurs.2 In this commentary, we suggest that greater reliance on in silico tools can be an additional pathway for optimizing the benefit of clinical trials with investigational antiseizure medications (ASMs). We present the performance of an established tool for central nervous system (CNS) drug screening and of a novel, computational-based tool that considers the physicochemical characteristics that distinguish US Food and Drug Administration (FDA)-approved ASMs from other molecules. Investigational CNS drugs are particularly prone to fail in clinical trials.3-5 This has led to the development of dedicated in silico screening tools for CNS drugs, such as those described by Dichiara et al.6 and Ghose et al.,7 as well as Wager's CNS multiparameter optimization (MPO) score.8, 9 Such tools build upon selected sets of physicochemical characteristics for predicting the absorption, toxicity, and CNS activity of investigational compounds. Several investigational ASMs (e.g., apinocaltamide [ACT-70947810 and VX-7655) entered clinical trials despite low CNS MPO scores for the active moiety. The CNS MPO might not have been widely accepted at the time of their initial development, or development was supported by the prevailing notion of lasting blood–brain barrier (BBB) disruption in chronic epilepsy in humans11-21 and by the efficacy of the candidate compounds in preclinical epilepsy models. Even when in silico screening tools are applied, they might not be optimal for selection of ASM candidates, because ASMs differ in their physicochemical properties from other CNS drugs.22 For instance, the CNS MPO allows a molecular weight of up to 500 Da for desirability.22 Yet the "heaviest" ASM is tiagabine (376 Da). Larger molecules whose main targets are protected by the BBB such as padsevonil (433 Da, CNS MPO score = 5.2)23 and PRX-00023 (451 Da, CNS MPO 4.8)24 failed to show efficacy in clinical trials. The trend is consistent, regardless of the study design and the characteristics of the patient population. We present here an ASM MPO score, a high-sensitivity explainable screening model for ASM candidates developed by the authors using readily available physiochemical properties of candidate molecules. It has been developed based on multiplicity-corrected statistical analyses of ASM physiochemical properties in comparison with several small-molecule drug groups: a general cohort of marketed drugs and two subgroups of this cohort, BBB-permeable drugs and BBB-impermeable drugs. ASM candidates (either investigational compounds or repurposed drugs) that did not show adequate efficacy in clinical trials were not used for the MPO development; instead, they were set aside for external validation. The detailed methodology is available in Appendix S1 and Tables S1–S4. The robustness of the ASM MPO model resides in the absence of any a priori assumption with regard to the requirements for a successful ASM. Instead, it makes use of a set of structural features proven statistically to characterize ASMs and utilizes publicly available data for each molecule. We found that ASMs differ from the "other drugs" cohort in each of the six parameters that comprise the ASM MPO (Figure 1; p < .001 for the ASM-Other marketed drugs comparisons). ASMs averaged a score of 5.74, compared with 3.28 for noneffective ASMs (U = 128, p > .0001), demonstrating the ASM MPO's utility in prioritization and bulk-screening of candidate compounds. The distribution of the ASM MPO and CNS MPO scores among the investigated drug cohorts is presented in Figure 2. The ASM MPO applies a binary cutoff value of 4, in parallel with the CNS MPO scale, to bin compounds into "desirable" and "undesirable" classifications common in binary "yes/no" decisions on investigation of candidates. Using this cutoff, the ASM MPO correctly classified 39 of 42 (93%) ASMs as desirable, as compared with 40 of 42 (95%) with the CNS MPO score (p = 1.00; Figures 3). The ASM MPO correctly classified 14 of 24 (58%) of noneffective ASM candidates as undesirable, whereas the CNS MPO correctly classified seven of 24 (29%, p = .023). All seven noneffective ASMs binned as undesirable by the CNS MPO were also scored as undesirable by the ASM MPO (Figures 3 and 4). In addition, the ASM MPO identified 44% more repurposing candidates; 52% of the "marketed drugs" cohort were binned as "undesirable" versus 36% using CNS MPO (p < .001; Figure 3). No FDA-approved ASM scored low on both the CNS and ASM MPOs; the CNS MPO rejected cannabidiol and fenfluramine, whereas the ASM MPO binned retigabine (ezogabine), tiagabine, and topiramate as undesirable (Figure 4). Examples of desirability violations on the ASM MPO for topiramate, tiagabine, and retigabine include hydrogen bond acceptor count, molecular weight, and length, respectively (Figure 5). Notably, the ASM MPO predicted the failure of seven molecules in addition to those rejected by the CNS MPO, representing trials in ~750 patients. The ASM MPO tool is freely available for downloading at https://github.com/IamAmitE/ASM-MPO-Calculator. The tool is self-explanatory and requires only the compound's name. It is best used in tandem with other prediction tools, for example, the CNS MPO score. MPO scores can be used to support other methods at any stage of ASM candidate development, from in silico design to clinical trials. Additionally, they can help in deciding whether a molecule should be repurposed for treating patients with epilepsy. A low score on either the CNS MPO or the ASM MPO does not necessarily mean that the molecule should not be developed as an ASM; instead, it can help prioritize compounds with similar preclinical efficacy or support the decision to avoid advancing development when efficacy outcomes of initial trials are borderline. Yet molecules that score low on both the CNS MPO and the ASM MPO have lower chances of being efficacious in patients with epilepsy. Examples of suggested repurposing candidates that may be excluded by in silico screening include amiodarone,25, 26 thioridazine (scored low on both the CNS and ASM MPOs),25, 26 and quetiapine (binned as undesirable only by the ASM MPO).25 In addition, our analysis suggests that permeable BBB should not be assumed in human epilepsy, except for conditions shown to result in overt BBB disruption.22 In fact, the structural requirement for ASMs might be even stricter as compared with those of other CNS drugs.22 Many discontinuations identified through our searches occurred during phase 2 trials (Table S2). We suggest that recognizing the structural features characterizing noneffective ASMs could help select not only the molecules but also the patient population for the early stage clinical trials. For example, molecules that silence cortical activity in the pulse transcranial magnetic stimulation test in healthy volunteers would not necessarily distribute well into epileptogenic brain tissue. Building optimization scores upon already-approved CNS drugs or ASMs may potentially carry the bias of assigning larger weight to structural features that characterize established drug groups such as barbiturates and benzodiazepines. Therefore, we stress that at least for the ASM MPO, the model is primarily aimed at excluding molecules less likely to be effective rather than directing the design of effective molecules, such that innovation is less likely to be affected. This is demonstrated by the deviation of several newer ASMs from the ASM cluster on both the CNS and the ASM scores (Figure 4, upper right corner). Recently, Alan Talevi27 categorized ASMs based on their chemical novelty and mode-of-action novelty. The suggested innovation degree of perampanel, accepted by the ASM score as desirable, was radical. Fenfluramine (semiradical), and cannabidiol ("radical?") were also accepted by the ASM MPO. Low acceptability of in silico screening tools, the assumption of BBB disruption in epilepsy, or both, have led to trials with compounds that would be otherwise excluded from testing as CNS drugs. Moreover, several physicochemical boundaries for ASMs are even narrower as compared with those of other CNS drugs. Our analysis of trial efficacy outcomes demonstrates that greater use of in silico screening tools can advance initial assessment of ASM candidates during both development and repurposing. Eventually, this could reduce animal experimentation, improve the yield of clinical trials, and minimize patient exposure to compounds less likely to be efficacious. Both authors conceived and designed the study, analyzed and interpreted the data, and drafted the manuscript. Both authors revised the work critically. We thank Dr. Yonatan Eyal and Mr. Roi Brudo for fruitful discussions. S.E. is affiliated with the David R. Bloom Center for Pharmacy and Dr. Adolf and Klara Brettler Center for Research in Molecular Pharmacology and Therapeutics at the Hebrew University of Jerusalem, Israel, and holds a Dame Susan Garth Chair of Cancer Research. S.E. has served as a consultant to Biopass, Israel. A.D.E. has no conflicts of interest. The preparation of this article received no funding. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines. APPENDIX S1. TABLE S1. TABLE S2. TABLE S3. TABLE S4. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.