Phospholipidosis is a shared mechanism underlying the in vitro antiviral activity of many repurposed drugs against SARS-CoV-2.

Repurposing drugs as treatments for COVID-19 has drawn much attention. A common strategy has been to screen for established drugs, typically developed for other indications, that are antiviral in cells or organisms. Intriguingly, most of the drugs that have emerged from these campaigns, though diverse in structure, share a common physical property: cationic amphiphilicity. Provoked by the similarity of these repurposed drugs to those inducing phospholipidosis, a well-known drug side effect, we investigated phospholipidosis as a mechanism for antiviral activity. We tested 23 cationic amphiphilic drugs-including those from phenotypic screens and others that we ourselves had found-for induction of phospholipidosis in cell culture. We found that most of the repurposed drugs, which included hydroxychloroquine, azithromycin, amiodarone, and four others that have already progressed to clinical trials, induced phospholipidosis in the same concentration range as their antiviral activity; indeed, there was a strong monotonic correlation between antiviral efficacy and the magnitude of the phospholipidosis. Conversely, drugs active against the same targets that did not induce phospholipidosis were not antiviral. Phospholipidosis depends on the gross physical properties of drugs, and does not reflect specific target-based activities, rather it may be considered a confound in early drug discovery. Understanding its role in infection, and detecting its effects rapidly, will allow the community to better distinguish between drugs and lead compounds that more directly impact COVID-19 from the large proportion of molecules that manifest this confounding effect, saving much time, effort and cost. ONE SENTENCE
SUMMARY: Drug-induced phospholipidosis is a single mechanism that may explain the in vitro efficacy of a wide-variety of therapeutics repurposed for COVID-19.

The outbreak of COVID-19 (Coronavirus Disease 2019) has led to multiple drug repurposing screens to find antiviral therapeutics; even a superficial survey of the literature reveals over 122 such studies over the last year(1). The motivation is clear—discovering and developing a new drug typically takes over a decade(2), while a drug previously approved for another indication can be rapidly brought to the clinic to treat an urgent threat like COVID-19. Conversely, the likelihood that a drug developed against a human disease will work against a novel virus might seem tenuous(3). Indeed, only a handful of these drugs are established antivirals, and these are generally restricted in activity to viruses of the same family. Thus, it has been encouraging, if not surprising, to see that over 1,974 unique drugs and investigational drugs have reported activity against SARS-CoV-2, the virus that causes COVID-19(1) (Fig. 1). Since this RNA virus encodes only 29 of its own proteins, the question of mechanism of action arises.

Fig. 1.

Representative examples of cationic amphiphilic drugs that are identified in SARS-CoV-2 drug repurposing screens.

Our interest in this question was motivated by our discovery that drugs binding to the sigma-1 and sigma-2 receptors were active in cells as antivirals. These receptors were identified proteomically as human proteins that interacted with the viral proteins Nsp6 and Orf9c, respectively(4). Encouragingly, early work showed that drugs and reagents like chloroquine, haloperidol, clemastine, and PB28—all potent against one or both sigma receptors—had cellular antiviral IC50 values in the 300 nM to 5 μM range. Subsequently, we investigated over 50 different molecules with a wide range of activity at these receptors. While this found molecules with relatively potent activity, structure activity relationships (SAR) found little correlation between receptor potency and antiviral efficacy (Fig. S1). Whereas drugs like amiodarone, sertraline, and tamoxifen had mid-to high-nM activities against viral replication, there were other sigma-active compounds, such as melperone and DTG, that were also potent on one or both sigma receptors but had no measurable antiviral effects. Intriguingly, we noticed the sigma drugs that were antiviral were all cationic at physiological pH and relatively hydrophobic, while those that were inactive against the virus were often smaller and more polar. We further noticed that this cationic-amphiphilic character was also shared among many of the hits emerging from other phenotypic screens (Fig. 1, S10), suggesting it was this physico-chemical property that might explain antiviral activity, instead of a specific on-target activity(5).

If the cationic-amphiphilic nature of this large set of molecules was itself responsible for antiviral activity, rather than the particular and widely different on-target activities of each individual drug, one would expect such a physical property to be associated with a shared cellular mechanism. Indeed, cationic amphiphilic drugs (CADs) can provoke phospholipidosis in cells and organs. This well-known side effect often encountered in late-stage drug development is characterized by the formation of lipidic vesicle-like structures in susceptible cells, and “foamy” or “whorled” membranes(6, 7). Though its molecular etiology remains a matter of ongoing research, drug-induced phospholipidosis (DIPL) is thought to arise by CAD disruption of lipid homeostasis. In particular, CADs are known to accumulate in intracellular compartments such as endosomes and lysosomes where they can directly or indirectly inhibit lipid processing(6). Importantly, modulation of these same lipid processing pathways is critical for viral replication(8), and inhibiting phospholipid production has previously been associated with inhibition of coronavirus replication(9). CADs have been previously shown to have in vitro activity against a range of viruses such as Severe Acute Respiratory Syndrome virus, Middle East Respiratory Syndrome virus, Ebola virus, Zika virus, Dengue virus, and filoviruses(10), though CAD-induction of phospholipidosis has only been proposed as a specific antiviral mechanism against Marburg virus(11). Finally, many of the drugs that are best-known to induce phospholipidosis, and for which it is associated with adverse events, are amiodarone(12) and chloroquine(13, 14), which we(15), and others (16, 17), had found to be potent inhibitors of SARS-CoV-2 replication in vitro. Other drugs from other SARS-CoV-2 phenotypic screens reported in the literature, such as chlorpromazine(18) and tamoxifen(17), had also been reported to induce phospholipidosis(19). Although the antiviral activity of CADs has been described, few, if any, of these CAD hits for SARS-CoV-2 have considered their phospholipidosis induction as a possible explanatory factor.

Here, we investigate the association between phospholipidosis and antiviral activity against SARS-CoV-2 in cell culture. This apparently general mechanism may be responsible for a large percentage of the drug repurposing hits reported for SARS-CoV-2, and an extraordinary amount of effort and resources lavished on drug discovery against this disease. As an effect that rarely occurs at concentrations lower than 100 nM, that does not appear to translate from in vitro to in vivo antiviral activity (as we show), and that can be a dose-limiting toxicity therapeutically(20), phospholipidosis may act as a confound to true antiviral drug discovery. We investigate the prevalence of this confound in SARS-CoV-2 repurposing studies, how phospholipidosis in particular correlates with inhibition of viral infection, and how to eliminate such hits rapidly so as to focus on drugs with genuine potential against COVID-19, and against other pandemics yet to arise.

Results

Lack of correlation between Sigma receptor activity and antiviral effect.

Our studies began with an effort to find drugs that modulated either or both of the sigma-1 or the sigma-2 receptors, and were therefore expected to be antiviral against SARS-CoV-2(4). Fortunately, many cationic drugs, for multiple primary targets and indications, have potent off-target activities against these receptors, offering a wide field for SAR. Initial results were promising, with multiple drugs found with antiviral activity and implication of sigma-1 as a relevant antiviral target from proteomic(21, 22) and CRISPR(15, 23) screens. After testing 72 sigma ligands in cellular antiviral assays and for sigma-1/sigma-2 ligand binding, it became apparent that potency for either or both sigma receptors had little relationship to antiviral activity (Fig. S1–S3, Table S1), suggesting that despite sigma receptors being involved in the viral life cycle, existing sigma receptor ligands may have additional cellular effects confounding their antiviral signal. This was a matter of some consternation, since by then much effort had been lavished on this project, and some of the drugs had antiviral IC50s as low as 130 nM.

Correlation of phospholipidosis and antiviral activity.

Among the most potent in the antiviral assays were drugs like amiodarone, tamoxifen, and sertraline; CADs known to provoke phospholipidosis at relevant concentrations, a connection also made previously for filoviruses (11). Since phospholipidosis disrupts intracellular lipid homeostasis, and many viruses propagate in double membrane vesicles that may exploit the lipid pathways that are disrupted by CADs, we investigated the hypothesis that the antiviral effect we had found, and potentially found in other drug repurposing studies for COVID-19, could be explained by the drug induction of phospholipidosis.

Accordingly, we tested 19 drugs (18 CADs and 1 non-CAD known to induce phospholipidosis) for their induction of phospholipidosis in A549 cells, a human lung-derived cell line widely used for testing viral infectivity. To measure phospholipidosis, we used the well-established NBD-PE staining assay(24), where the vesicular lipidic bodies characteristic of the effect may be quantified by high content imaging (Fig. 2A). Three classes of drugs and reagents were initially investigated: A. Sigma-binding antiviral CADs we had discovered, like amiodarone, sertraline, chlorpromazine, and clemastine (nine total); these molecules are predicted—or known—to induce phospholipidosis; B. Analogs of these CADs that no longer bound sigma receptors, but were still antiviral (four total); these molecules are predicted to induce phospholipidosis despite their lack of on-target activity; and C. Sigma-binding, -antiviral drugs, like melperone and DTG, that were much more polar than classic CADs (two total); these molecules are predicted not to induce phospholipidosis. Of the nine sigma-binding CADs that were antiviral (class A), six of which were also found in the literature for COVID-19, eight induced phospholipidosis, consistent with the hypothesis (Fig. 2A–B, S4–S5). The only non-phospholipidosis inducing antiviral from this set was elacridar, a promiscuous P-glycoprotein inhibitor; this investigational drug may therefore be active via another mechanism. Intriguingly, CAD analogs of the potent sigma ligand PB28 that had lost their sigma-binding activity from bulky electron-withdrawing substitutions (ZZY-10–051 and ZZY-10–061, Fig. 2B–F, S4–S7), did induce phospholipidosis, as did the antipsychotic olanzapine and the antihistamine diphenhydramine, which are weak sigma receptor ligands but are structurally related to potent sigma receptor tricyclics (e.g., chlorpromazine) and diarylethanolamines (e.g., clemastine; class B). Finally, melperone and DTG, which are potent cationic sigma receptor ligands but are not antiviral, did not induce phospholipidosis (Fig. 2A–B, S4–S5; class C). These results do not prove phospholipidosis as the antiviral mechanism, but are consistent with the phospholipidosis hypothesis.

Fig. 2.

Cellular phospholipidosis may confound antiviral screening results.

A. Examples of NBD-PE quantification of phospholipidosis in A549 cells including dose response curves. Blue = Hoechst nuclei staining, Green = NBD-PE phospholipid staining, Red = EthD-2 staining for dead cells. Scale bars = 20 μm. Amiodarone is the positive control for assay normalization; sertraline and clemastine are two examples of high phospholipidosis inducing drugs (phospholipidosis (DIPL) > 50% of amiodarone). Images of DMSO and a non-phospholipidosis inducing molecule (melperone) are included for reference. Thresholds for determining phospholipidosis power are shaded in dark grey (low phospholipidosis), light gray (medium phospholipidosis) and no shading (high phospholipidosis). B. Pooled DIPL amounts (mean ± SD) at the highest non-toxic concentration tested for each drug. Results were pooled from three biological and three technical replicates and were normalized amiodarone (100%) from the control wells in the same experimental batches. C. Structures of PB28 and its analog ZZY-10–051, the latter of which is inactive on the sigma receptors. D. Viral infectivity (red) and viability (black) data for PB28 (square) and ZZY-10–051 (circle) in A549-ACE2 cells. Data shown are mean ± SD from three technical replicates. E. Fractional binding of PB28 and ZZY-10–051 against sigma-1 (purple; S1R) and sigma-2 (maroon; S2R) normalized to a buffer control at 1.0 in a radioligand binding experiment. Data shown are mean ± SEM from three technical replicates. PB28 is a strong ligand of both sigma-1 and sigma-2 and has high displacement of the radioligands, whereas ZZY-10–051 is unable to displace the radioligands to a high degree at 1 μM. F. Dose response curves for PB28 (blue) and ZZY-10–051 (gold) show that these closely related analogs both induce phospholipidosis.

If phospholipidosis is responsible for CAD antiviral activity, then CADs known to induce phospholipidosis should also be antiviral, as should the rare non-CADs that are nevertheless known to induce phospholipidosis. We tested three CADs for antiviral activity, including ebastine, ellipticine, and Bix 01294, all of which are reported to induce phospholipidosis(25), none of which, to our knowledge, at the time of running this experiment, had been reported to inhibit SARS-CoV-2; we have since read the work of Dittmar et al. who also identified Bix 01294 and ebastine as drug repurposing hits against SARS-CoV-2 (26). We further tested the macrolide antibiotic azithromycin, also reported to induce phospholipidosis(27), but representing very different properties from typical CADs. We first confirmed phospholipidosis-inducing activity for these molecules, though it is important to note the difficulty of separating cytotoxicity from phospholipidosis and antiviral activity for both ellipticine and ebastine (Fig. 2B, S4). All four molecules were next shown to be antiviral, with IC50 values in the 400 nM to 3 μM range, overlapping with the activities of other CADs we and others have identified for SARS-CoV-2(26) (Fig. S5). This too was consistent with the antiviral phospholipidosis hypothesis.

For phospholipidosis to explain antiviral activity, we might expect a correlation between concentration-response curves for phospholipidosis and antiviral activity for a set of compounds. We compared phospholipidosis amounts to SARS-CoV-2 amounts for each drug individually (Fig. 3A). Typically, the correlations were high—not only did antiviral activity occur in the same concentration ranges where phospholipidosis occurred, but the statistically significant R values, ranging from 0.51 to 0.94, supported a quantitative relationship between the two effects (Fig. 3A).

Fig. 3.

Quantitative relationship between phospholipidosis and viral amounts.

A. Correlations between phospholipidosis (DIPL), normalized to amiodarone at 100%, and percent of SARS-CoV-2, normalized to DMSO at 100%, in the RT-qPCR assay in A549-ACE2 cells. Each dot represents the same concentration tested in both assays. A strong negative correlation emerges, with R ≥ 0.65 and p ≤ 0.05 for all high and medium phospholipidosis-inducing drugs except ellipticine, which is confounded by its cytotoxicity in both experiments, ebastine, and ZZY-10–61. The latter two examples are marginally significant. B. The SARS-CoV-2 viral loads and induced phospholipidosis magnitude for each compound and dose in A are plotted as sqrt(viral_amount_mean) ~ 10*inv_logit(hill*4/10*(log(DIPL_mean)-logIC50). Fitting a sigmoid Bayesian model with weakly informative priors yields parameters and 95% credible intervals of IC50: 43 [38, 48]%, hill: −5.6 [−7.0, −4.5], and Sigma 2.0 [0.14, 1.78]. Forty draws from the fit model are shown as blue lines. Salmon points overlaid with the model represent predicted phospholipidosis inducers from the literature (Fig. 5).

To assess if phospholipidosis could explain antiviral effect irrespective of the drug causing it, we fit a sigmoidal model through all the 107 phospholipidosis versus antiviral activity observations (comprised of six concentration measurements each for the 16 phospholipidosis-inducing drugs) and observed a strong negative relationship (R = 0.65, 95%CI [.52, 0.76]) between induced phospholipidosis and SARS-CoV-2 viral load across all observations for all 16 drugs. Importantly, compared to the linear fit (Fig. S8), the sigmoid shape fit the data as well, and exhibits saturation of the effects toward the extremes. Because the biological processes of phospholipidosis and antiviral effects are saturable, the sigmoid fit was selected to represent the data.

Concurrent measurement of viral infection and drug induced phospholipidosis

In the previous experiments, drug-induced phospholipidosis and drug antiviral activity were measured separately. To investigate how the two effects interact with each other, we next tested how viral infection affects phospholipidosis in the same cells. We dosed cells with either 1 or 10 μM of five characteristic CADs (amiodarone, sertraline, PB28, hydroxychloroquine (HCQ), and Bix 01294), followed by a mock or SARS-CoV-2 infection, and quantified phospholipidosis staining and the accumulation of viral spike protein in the same cells (Fig. 4A, S9). Compared to DMSO, drug treatments led to substantial increases in NBD-PE aggregates, indicating increased phospholipidosis (Fig. S9; 1 μM: F (5, 24) = 7.7, P < 0.001; 10 μM: F (5, 24) = 9.1, P < 0.001). At 1 μM drug concentrations, SARS-CoV-2 spike protein was readily stained, and one could visualize both spike protein and phospholipidosis staining in some of the same cells (yellow puncta), suggesting at this low concentration of drug—often close to the antiviral IC50 value—both phospholipidosis and viral infection were co-occuring, though even here viral staining was reduced relative to the DMSO treated controls. As drug concentration rose to 10 μM, viral spike protein staining dropped while staining for phospholipidosis increased (Fig. S9); there was nearly complete loss of spike protein signal with a concomitant increase in staining for phospholipidosis (Fig. 4A) for all treatments. To better quantify this, we repeated these experiments in seven point concentration-response for amiodarone, sertraline, and PB28. As expected, viral staining monotonically decreased as phospholipidosis increased (Fig. 4B–C).

Fig. 4.

Phospholipidosis and spike protein measurements in the same cellular context.

A. Representative images from a costaining experiment measuring phospholipidosis and SARS-CoV-2 spike protein in infected and uninfected A549-ACE2 cells. Five molecules (1 and 10 μM) and DMSO were measured; see Fig. S9 for Bix 01294. Blue = Hoechst nuclei staining, Green = NBD-PE phospholipid staining, Red = SARS-CoV-2 spike protein staining; Yellow = coexpression of spike protein and NBD-PE. Scale bar = 20 μm. B. Concentration-response curves for phospholipidosis induction measured by NBD-PE staining in infected cells for three characteristic CADs. Data are mean ± SEM from four technical replicates. C. Spike protein in infected cells decreases as phospholipidosis increases. Data are mean ± SEM from four technical replicates.

CADs are common among drug repurposing hits for SARS-CoV-2 and other viruses

With the strong correlation between CAD phospholipidosis and antiviral efficacy (Fig. 3), including drugs that have been found in multiple SARS-CoV-2 repurposing studies, such as amiodarone, sertraline, HCQ, and chlorpromazine, we wondered how prevalent phospholipidosis-inducing CADs might be in the literature. To quantify this, we investigated a subset of the 1,974 total reported repurposing hits by manually searching 12 major drug repurposing efforts for SARS-CoV-2 from the literature. These included two screens of the ReFRAME library(28, 29), a screen of the NCATS approved and bioactive libraries(16), among others (4, 15, 17, 26, 30–34). Together, 310 drugs, investigational drugs, and reagents were reported as active in antiviral assays against SARS-CoV-2. We used two physico-chemical features to identify likely CADs within these actives: drugs with calculated Log octanol:water coefficients above 3 (cLogP ≥ 3), and with pKa values ≥ 7.4 (35, 36), and then further filtered for drugs that topologically resembled known phospholipidosis inducers(19, 25), (using an ECFP-4-based Tanimoto coefficient (Tc) ≥ 0.4) (Table S2). Sixty percent of the 310 drugs surpassed the cLogP and pKa threshold; 34% were also structurally similar to a known phospholipidosis inducer (Fig. 5A; structures of representative Tc = 1 CADs from the literature in Fig. 1, and 1 > Tc ≥ 0.4 in Fig. S10).

Fig. 5.

Many drugs with activity against SARS-CoV-2 are CADs that induce phospholipidosis.

A. Percentage of total drug repurposing hits collected that pass CAD thresholds. B. Example repurposing hits from the literature that pass our CAD filters. C. Dose response curves for five predicted phospholipidosis inducers. All five induce measurable phospholipidosis (blue) with no impact on cell viability (black). D. Representative images of phospholipidosis quantification through NBD-PE staining in A549 cells. Blue = Hoechst nuclei staining, Green = NBD-PE phospholipid staining, Red = EthD-2 staining for dead cells. Scale bars = 20 μm. E. Viral infectivity (red) and cytotoxicity (black) data for five example literature CADs tested in A549-ACE2 cells. Data shown are mean ± SD from three technical replicates.

Although the two physical property filters do not capture typical phospholipidosis inducers such as azithromycin, they do capture 16 of the other 18 CADs we had tested up until now (missing only the medium phospholipidosis inducers olanzapine and ellipticine); intriguingly, nine of these, including amiodarone, sertraline, chlorpromazine, Bix 01294, clemastine, and benztropine also appeared in at least one of the 12 other repurposing studies. To probe the reliability of this association, we tested another five drugs that passed our filters, and had been reported as antiviral against SARS-CoV-2, for their ability to induce phospholipidosis (Fig. 5B). All five were active in the NBD-PE assay to varying extents (Fig. 5C). For example, duloxetine and toremifene were strong phospholipidosis inducers (> 50% of the strong phospholipidosis inducer amiodarone), while fluspirilene, methdilazine, and thiethylperazine were medium inducers (50% > DIPL > 25% of amiodarone). We were able to confirm SARS-CoV-2 antiviral activity for these drugs in our own hands (Fig. 5D). Additionally, these molecules fit into the sigmoidal model relating phospholipidosis amount to reduction in viral load described above (salmon points overlaid with sigmoidal model; Fig. 3B). Finally, we note a preliminary identification of 30 CADs, 19 of which overlap with the literature-derived SARS-CoV-2 list, active against other viruses including Middle East Respiratory virus and Severe Acute Respiratory virus(37), Ebola virus(38–40), Marburg virus(40, 41), Hepatitis C virus(42), and Dengue virus(43) (Table S3). It may be that most drugs identified in antiviral repurposing assays against SARS-CoV-2 and other viruses are CADs whose antiviral activities can be attributed to a phospholipidosis mechanism.

Animal efficacy for phospholipidosis-inducing and non-phospholipidosis-inducing repurposed drugs

Though phospholipidosis is considered a drug-induced side effect, it remains possible that the effect can be leveraged for antiviral efficacy. Accordingly, we tested four of the repurposed drugs most potent against SARS-CoV-2 in vitro, whose activity is seemingly explained by phospholipidosis: amiodarone, sertraline, PB28 (three molecules tested above) and tamoxifen (a drug often reported in the literature to induce phospholipidosis)(6, 19), for efficacy in a murine model of COVID-19(44). In the same model, we also tested two antiviral drugs that do not induce phospholipidosis: remdesivir, a drug that may be considered a success of repurposing, and elacridar, an antiviral compound that was not a phospholipidosis inducer (Fig. 2B). In pharmacokinetic studies, all molecules had relatively long half-lives, especially in the lung where tissue Cmax values often exceeded 10 μM after a 10 mg/kg dose; this Cmax was 10 to 1000 times higher than the in vitro efficacy of these drugs against SARS-CoV-2, suggesting that exposure would be high enough for plausible efficacy (Table S4–S8). Guided by the pharmacokinetic parameters of each drug, mice were dosed either once per day (amiodarone and elacridar) or twice per day (remdesivir, PB28, tamoxifen, and sertraline), for a total of three days (see Materials and Methods). Two hours following the first dose, mice were intranasally infected with 1 × 104 PFU of SARS-CoV-2, followed by viral titer quantification at the end of the three-day infection period. Notwithstanding their high lung exposure, the four phospholipidosis-inducing drugs—amiodarone, sertraline, PB28, and tamoxifen had no substantial effect on viral propagation in the mice. Conversely, the well-known antiviral remdesivir reduced viral load by two to three orders of magnitude in the mice, while the cationic drug elacridar, which had shown antiviral activity without phospholipidosis, also showed a modest antiviral effect (Fig. 6). Unfortunately, mice dosed higher than 3 mg/kg with elacridar exhibited toxicities that limited further study. Taken together, despite the high lung exposure of these drugs, their in vitro activities do not translate to in vivo action, at least at the tested conditions. These observations suggest that cationic amphiphilic drugs whose antiviral activity arises due to phospholipidosis may not be viable candidates for clinical progression. We do not rule out purposefully targeting phospholipidosis as an antiviral mechanism, though doing so may require a target-based, rather than a physical property-based approach.

Fig. 6.

In murine viral replication, four phospholipidosis inducing drugs are not efficacious in vivo.

Six different drugs were dosed to mice for three days with a two-hour preincubation before SARS-CoV-2 treatment. Lung viral titers were quantified. Only remdesivir, positive control, induced significant decreases in viral load (vehicle N = 5; remdesivir N = 4, *P < 0.05). The other five drugs, four of which cause phospholipidosis, did not significantly affect viral titers.

Discussion

The emergence of COVID-19 has motivated an intense effort to repurpose drugs as SARS-CoV-2 antivirals. From these studies, an extraordinary number of potentially promising hits have emerged(1). A key observation from this work is that many, perhaps most, of these repurposed drugs are active in antiviral assays because they are cationic amphiphiles that induce phospholipidosis (Fig. 1, 5, S10). Thus, despite their diverse chemotypes, a single mechanism of action may explain cellular antiviral activity of these drugs. Phospholipidosis disrupts lysosomal lipid catabolism and trafficking, as reflected in the formation of the pathological vesicular structures that characterize it. This disruption may underlie the antiviral effects of these repurposed drugs, perhaps through effects on the double membrane vesicles that the virus itself creates and on which it depends. Quantitatively, there is a close in vitro correlation between drug-induced phospholipidosis and antiviral activity, both drug-by-drug and over the set of drugs tested here (Fig. 3). The effect is predictive: molecules that induce phospholipidosis are antiviral over the same concentration range, irrespective of whether they are CADs or not (e.g., azithromycin), while molecules that are related by target activity to the CADs, but are more polar and do not induce phospholipidosis (e.g., melperone and DTG), are not antiviral. Correspondingly, one can modify an antiviral drug that acts on a particular target, like the canonical sigma receptor ligand PB28, so that it loses measurable target binding, but as long as it keeps its CAD character it retains antiviral activity (e.g., ZZY-10–051, Fig. 2D). Unfortunately, CAD induction of phospholipidosis, at least at the potencies observed here, does not appear to translate to in vivo antiviral activity in the tested conditions (Fig. 6). More encouragingly, this study illuminates a method to rapidly identify drugs that are acting as confounds in cellular antiviral screens, allowing one to eliminate them from further study and effort and to rapidly focus on other drugs with true potential.

Although the molecular mechanisms for the antiviral effects of phospholipidosis remain unclear, certain associations may be tentatively advanced. SARS-CoV-2, like many viruses, subverts the cell to produce double membrane vesicles in which it replicates(45–47)—these vesicles provide sheltered environments not only for reproduction but also for evasion of innate immunity(48, 49). Disruption of lipid homeostasis by the induction of phospholipidosis may lead to the formation of lipid structures that compete with or disrupt the formation of viral double membrane vesicles, reducing viral replication. Alternatively, the disruption of lysosomal(50) and endosomal(51) compartments by aggregating phospholipids and CAD-induced transient shifts in compartmental pH(52) may further affect viral entry and propagation(53). For example, it is possible that alterations to these compartments disrupt cathepsins that have been shown to prime the SARS-CoV-2 spike protein facilitating viral entry(54). For all these reasons, targeting the endosomallysosomal pathway has been suggested as a valuable strategy against SARS-CoV-2 infection(55), but developing potent and targeted inhibitors remains challenging. Of course, these mechanisms remain unproven, and currently are supported only by correlation, but they suggest a route for further research.

The cost to the community of investments in what appears to be a confound needs to be considered as a guidance for strategy in future pandemics. According to the DrugBank(56) COVID-19 dashboard, which collates data from the U.S. National Library of Medicine’s ClinicalTrials.gov and the World Health Organization’s International Clinical Trials Registry Platform, CADs reported to have antiviral activity have been promoted into an astonishing 316 Phase I to Phase III clinical trials against COVID-19 (Table S9). Admittedly, 57% of these trials (180 of the 316) solely study the phospholipidosis-inducing CADs hydroxychloroquine (Fig. 3A, top row) or chloroquine, both of which already failed to show efficacy in varying cell types(57), rodent models(58), and clinical trials(59), but that still leaves 136 trials across 33 other predicted or known phospholipidosis-inducing CADs. Taking conservative estimates of what general anti-infective trials cost, from a 2014 US Department of Health and Human Services report(60, 61), we estimate the cost of the clinical trials component alone, over the last year, for phospholipidosis-inducing CADs to be over $6 billion US dollars (Table S9).

Certain caveats merit airing. First, the correlation we observe between antiviral activity and phospholipidosis, as strong as it is, does not illuminate the mechanism by which phospholipidosis is itself antiviral. Phospholipidosis itself is only partly understood mechanistically, and there are no known genetic or chemical ways to inhibit drug-induced phospholipidosis, nor are there reliable target-selective reagents to induce it. Second, there is no consistent standard in the field as to the physical properties that will predict whether a molecule will induce phospholipidosis, and there are even non-CAD molecules that induce it (e.g., azithromycin). Thus, we have chosen conservative criteria to predict phospholipidosis-inducing CADs; while we believe that these will have relatively few false positive predictions, many phospholipidosis-inducing drugs may be missed. Third, phospholipidosis is a confound that only affects drugs repurposed for direct antiviral activity—it is irrelevant for drugs like dexamethasone (62) and for the CAD fluvoxamine (63) which have been repurposed for immunomodulatory treatment of COVID-19, nor is it relevant for antiviral CADs whose activity against the virus is well-below the range where phospholipidosis occurs. Fourth, our estimates of the clinical trial costs of CAD advancement for COVID-19 are clearly rough. If we have inadvertently included CADs advanced for immunomodulatory treatment, for instance, they will be too high. This would be balanced by the conservative numbers that we have used to estimate trial costs, and by our conservative method of predicting molecules that are phospholipidosis-inducing CADs (of the 35 drugs identified in our analysis, 12 have been directly shown to induce phospholipidosis). Finally, we do not exclude that phospholipidosis itself might be exploited for genuinely therapeutic antiviral effect, though we suspect that will have to go through a different, presumably more target-directed mechanism than that exploited by the CADs studied here.

These caveats should not obscure the central observation of this study. Many drugs repurposed for antiviral activity against SARS-CoV-2 are cationic amphiphiles, and despite their diverse structures and multiple targets, many likely have their antiviral effects via a single shared mechanism: phospholipidosis. Both because of the adverse drug reactions with which it is associated, and the limited efficacy to which it leads—rarely better than 100 nM—drugs active due to phospholipidosis are unlikely to translate to therapeutic efficacy in vivo (Fig. 6). What this study teaches is a rapid way to distinguish drug confounds arising from phospholipidosis, on which tremendous resources have been lavished over the last year, from the genuinely effective drugs—such as the antiviral remdesivir(64), and the immunomodulators dexamethasone(62), baricitinib(65) —that are so desperately needed. Doing so will be important not only for COVID-19, but also for the future pandemics that we can expect to emerge.