When the COVID-19 pandemic hit, there was massive investment into the discovery of new treatments. Given the urgent need, repurposing of approved or clinically pretested drugs appeared especially attractive because that strategy promised fast initiation of antiviral clinical studies. On page 541 of this issue, the study by Tummino et al. (1) raises concerns that many drug candidates that showed antiviral activity in hypothesis-free cellular screens and were then repurposed to treat severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infections may be scientific dead ends. Their study is a warning that even amid the pressure of a pandemic, scientific diligence is still essential.
Early in the pandemic, an international team of scientists undertook a comprehensive study to identify human proteins that interact with proteins from SARS-CoV-2 (2). The idea was that if one or more of these human proteins were required for viral production, then there may be drugs that target them тАЬon the shelfтАЭ that could be repurposed to treat COVID-19 patients. The team identified several candidate drug targets, among which were the sigma receptors. They went on to show that drugs targeting these receptors potently inhibited viral production in cell culture, providing preliminary тАЬvalidationтАЭ of sigma receptors as COVID-19 drug targets.
However, as members of the same team began to investigate their potential for clinical studies, they became concerned. In testing 50 different sigma receptor drugs to find the most suitable one, they found no correlation between the potency with which the drugs inhibited the receptor and their antiviral activity. What was going on? Tummino et al. show that the antiviral activity in cell-culture assays had nothing to do with modulation of the sigma receptors but rather correlated with certain chemical properties of the compounds. The subset of drugs that had antiviral activity were all cationic and amphiphilic, features known to induce phospholipidosis (3), which is an aberrant accumulation of phospholipids in the lysosome. Potent sigma receptor drugs that did not induce phospholipidosis showed no antiviral activity. Drug-induced phospholipidosis is a side effect of cationic amphiphilic drugs (CADs), and these drugs are known to inhibit the production of many other viruses in cell culture.
Tummino et al. then turned their attention to 310 drugs found to inhibit SARS-CoV-2 production in cellular screens in the literature. To these, they applied a computational filter of physicochemical properties and found that 60% of the repurposed drugs were predicted to be CADs. Computational filters are known to miss many compounds that induce phospholipidosis, so they concluded that it is likely that many more repurposed SARS-CoV-2 drugs identified in cellular screens act through this nonspecific mechanism.
In theory, induction of phospholipidosis might still be a viable therapeutic strategy to treat COVID-19 patients, even if repurposed drugs were not acting through their intended biological target. However, this is unlikely. The most potent four CADs in the cell screen carried out by Tummino et al. had no antiviral activity in mice at doses exceeding the concentrations that were effective in cell culture. This was true whether the drug was given only once or chronically for 12 days before viral infection. It is possible that at these doses, phospholipidosis was not induced in the mice. If it were possible to induce phospholipidosis in patients by further increasing the doses, this would likely inflict serious harm to patients, in part because the same general compound properties that lead to phospholipidosis are often associated with cardiotoxicity. Perhaps most compelling, 33 of the predicted phospholipidosis-inducing COVID-19 drugs, including the infamous hydroxychloroquine, have been tested in 316 clinical trials, and none have shown efficacy.
The main lesson from this study is less about phospholipidosis itself, for it is but one of many confounders in drug screens. Rather, the lesson is that all screening hits should be treated with extreme skepticism, whether they derive from a biochemical or cell-based assay and whether they are new compounds or approved drugs. Indeed, traditional small-molecule drug discovery is structured to avoid such misleading starting points (4, 5). Counter assays are regularly used to rule out nonspecific mechanisms (6).
Hypothesis-free drug-repurposing efforts require scientific diligence in ruling out artifacts before true antiviral drug candidates can move into clinical studies.
PHOTO: AKOS STILLER/BLOOMBERG/GETTY IMAGES
Unfortunately, although counter screens can readily identify compounds that are overtly toxic, they are less able to flag compounds that act subtly or through yet unknown confounding mechanisms. This challenge can often be overcome by carefully exploring structure-activity relationships (SARs). This systematic correlation of structural changes of the compounds with biological activity against the target in vitro and in cells takes time and requires medicinal chemists to synthesize multiple compounds. If the SAR correlates well, there is more confidence as to the validity of both the target and compound. If there is no correlation, flags should be raised. Exploring the SAR is exactly what Tummino et al. did (with the benefit of compounds that had already been synthesized), and their diligence was rewarded by falsifying a compelling target hypothesis without investing money and the hopes and health of patients into a clinical study.
The findings in this study are a reminder of potential pitfalls when relying on assays that monitor a decrease in a signal in cell-based drug screens (7). In a complex and interconnected system like a cell, interfering with any essential function may indirectly affect a signal and lead to false-positive hits. Viral production is a particularly sensitive readout because viruses exploit many cellular processes as part of their life cycles.
It is important to note that mechanism-informed drug repurposing can work. This strategy revealed remdesivir, which may be somewhat effective at reducing death from COVID-19 (8, 9). Conversely, repurposing drugs based on hypothesis-free screens has not yet yielded any effective treatments for COVID-19, nor for any disease (10). Perhaps it is not the concept itself that is at fault, but rather the notion that approved drugs represent a privileged class of molecules that do not require as much scientific due diligence. Whatever the reason, given that billions of dollars have already been spent on hypothesis-free drug repurposing for COVID-19 antivirals with no impact on patients, it would seem wise to increase investments toward the rational development of new direct-acting antiviral drugs. Although this approach may take more time, it is less risky and can even be used to develop drugs for newly emerging viruses.
The study of Tummino et al. highlights how the seduction of the concept of hypothesis-free drug repurposing can corrupt the scientific method. Drug discoverers know that no compound, whether an approved drug or not, is monospecific and that all have side effects or toxicities, especially at higher concentrations commonly used in cell-based screens. All hits in cell-based screens should be considered artifacts until conclusively proven otherwise (7). Unfortunately, when the hit is a known drug and the cellular effect is the one wished for, and especially amid the urgency of a pandemic, warning cries are easily dismissed, even if they come from experienced drug discoverers who have seen it all before.
