Inhibition of SARS-CoV-2 entry by anesthetic compounds
In order to test a membrane-disruptive mechanism for HCQ inhibition of SARS-CoV-2 viral entry, we compared HCQ to anesthetics (tetracaine and propofol) which are known to be membrane-disruptive. HEK293T cells overexpressing ACE2 were infected with a retrovirus pseudotyped with the SARS-CoV-2 spike protein (SARS2-PV). A segment of the spike protein binds to ACE2 and recapitulates viral entry47,48. A luciferase encoded in the pseudotyped virus is then used to quantitate viral entry (Fig. 1b–d).
Treatments with HCQ, tetracaine, and propofol all robustly reduced SARS2-PV entry into HEK293T cells overexpressing ACE2 (Fig. 1b). The cells were first treated with drugs (50 µM) for 1 h, then the drugs were removed. After the treatment and subsequent drug removal, SARS2-PV was applied such that the virus was never exposed to the drugs, thus avoiding potential direct effects of cholesterol on the viron. HCQ had the greatest effect on viral inhibition with almost a 90% reduction in SARS2-PV luciferase activity (Fig. 1b). We used 50 µM since that concentration was previously shown to be the minimum concentration that fully inhibited viral entry3. The concentration is ~5-fold above the concentration found in lung epithelial lining fluid after 400 mg for 1 day49 making it an appropriate concentration to see a full effect by dSTORM. Like anesthetics, the actual concentration of HCQ in the membrane is dictated by a partition coefficient and the resultant mole fraction, not external concentration.
In Vero E6 cells, a cell line that endogenously expresses ACE2 receptor and robustly facilitates SARS-CoV-2 viral entry50, HCQ, tetracaine, and propofol all significantly decreased viral entry (Fig. 1c). HCQ again showed the strongest effect decreasing viral entry by ~92%. Cell viability after HCQ treatment was assessed by Fixable Viability Dye (FVD) staining and MTT assay. The FVD staining labeled dead cells and found that HCQ treatment had no effect on Vero E6 cells, in agreement with a previous study3, but did decrease live cell number by ~23.93 ± 5% in HEK293T cells (Supplementary Fig. 1c, d). Similarly, the MTT assay showed HCQ treatment significantly decreased cell metabolism in HEK293T cells by ~36.05 ± 6% (Supplementary Fig. 1d). However, the reduction in cell viability of HCQ did not account for the full reduction in viral entry. Tetracaine and propofol had no adverse effects on cell viability in vero E6 or HEK293T cells (Supplementary Fig. 1e).
COVID-19 is often severe in obese patients and those with underlying conditions. We obtained lung samples from adult humans with chronic obstructive pulmonary disease (COPD). We found the lung tissue to have significantly higher free-cholesterol levels compared to cultured lung cell lines as measured by our fluorescent cholesterol assay (Fig. 1e). To recapitulate the physiological conditions observed in COVID-19 patients, we tested HCQ’s inhibition on viral entry in HEK293T loaded with cholesterol and overexpressing ACE2. To load cholesterol into cells, 4 µM apolipoprotein E (apoE, a cholesterol carrier protein linked to the severity of COVID-1951) was applied. ApoE binds to low-density lipoprotein (LDL) receptors in tissue and facilitates the loading of cholesterol into cells52 (Supplementary Fig. 1b). To provide a source of cholesterol to the apoE, 10% fetal bovine serum (FBS, a common source of cholesterol) totaling ~310 µg/mL was added. Importantly, apoE is not present in FBS, allowing us to carefully control cholesterol loading32,53. When apoE is in excess or in low-cholesterol conditions, it facilitates the efflux of cholesterol from the cell32,52. Cells were treated acutely (1 h) for loading or unloading cholesterol prior to viral infection.
Loading cells with cholesterol into HEK293T cells overexpressing ACE2 increased viral entry by ~56 ± 38% (Supplementary Fig. 1f), which is consistent with observations with endogenously expressed ACE2 where cholesterol loading significantly increased viral entry by ~36 ± 7% (Supplementary Fig. 1g). As expected, treatment of cholesterol loaded cells with HCQ (~85 ± 12%) and tetracaine (~43 ± 12%) reduced SARS2-PV entry in a high-cholesterol state (Fig. 1d). The efficacy of HCQ was reduced in cholesterol loaded cells compared to non-cholesterol loaded cells, but only slightly.
To confirm that apoE loads and unloads cholesterol from cultured cells, we treated HEK293T, Vero E6, and A549 cells with apoE with and without 10% FBS and measured the relative change in membrane-free cholesterol (Fig. 1f). Cells that were incubated with and without a source of cholesterol contained small but significant increases and decreases in total cholesterol respectively. The loading and unloading of cholesterol were similar in H1793 cells, although the loading of cholesterol did not reach statistical significance.
HCQ’s disruption of ordered GM1 clusters
The ability of a virus to cluster is important for its infectivity and maturation54,55,56,57. Previously, anesthetics were shown to perturb clustering in two ways. First, inhaled anesthetics tend to increase the apparent size and number of clusters, as observed using super-resolution imaging and cluster analysis, while local anesthetics tend to decrease the cluster size13,20. Second, both inhaled and local anesthetics disassociate cholesterol-sensitive proteins from GM1 clusters. The dissociation of proteins from a GM1 cluster is recorded by two-color direct stochastic optical reconstruction microscopy (dSTORM) super-resolution imaging. The GM1 and PIP2 lipid, and ACE2 protein are fixed and labeled with cholera toxin B (CTxB, a pentadentate toxin binding GM1 lipids), PIP2 antibody, and ACE2 antibody, respectively, and then ACE2 association with the lipid is measured by pair correlation analysis (Fig. 2a). The antibodies in this study were previously validated for specificity (see methods). We previously used these techniques to monitor nanoscopic movement (<100 nm) of multiple proteins between both GM1 and PIP2 clusters13,20,28,32,58 (see also Discussion).
To test HCQ’s effects on lipid membranes, the effect of HCQ was first examined on the apparent structure (size and number) of GM1 clusters by dSTORM in the membranes of HEK293T cells. This was done using density-based spatial clustering of applications with noise (DBSCAN). The use of HEK293T cells allowed us to compare the effects of HCQ to previous anesthetic studies in HEK293T cells13,20. As mentioned, 50 µM HCQ is the minimum saturating concentration that was shown to inhibit viral entry in cultured cells3.
We tested 50 µM HCQ and found it increased the number and apparent size (Supplementary Fig. 2a–c) of GM1 clusters, despite lowering the free cholesterol in the membrane (Supplementary Fig. 1h). HCQ’s perturbation to cluster size in HEK293T cells was most similar to the inhaled anesthetics chloroform and isoflurane13 (Supplementary Fig. 2a). Methyl-beta cyclodextrin (MβCD), a chemical known to deplete GM1 clusters from the cell membrane, decreased the apparent size of GM1 clusters and showed a trend of decreasing cluster numbers (Supplementary Fig. 2b, c). Interestingly, HCQ had the opposite effect on lung cells. HCQ decreased cluster size and a number of GM1 clusters, a result that is similar to what was seen for local anesthetics in nerve cells (Supplementary Fig. 2d–f)20.
HCQ’s effects on ACE2 clustering in kidney and lung cells
In cells and animals with low cholesterol, ACE2 clusters primarily with PIP2; however, in high-cholesterol and obese animals, ACE2 appears to cluster primarily with GM1 lipids28. To compare the effect of HCQ on ACE2 clustering in conditions of high and low cholesterol, we loaded and unloaded cholesterol into and from wild-type (wt.) HEK293T and Vero E6 cells (kidney) with apoE in the presence and absence of serum cholesterol. The cells were then fixed, labeled with anti-ACE2 antibody and CTxB, and imaged using two-color dSTORM. Both cell lines express low levels of endogenous ACE232,59.
We found a 50 µM HCQ treatment in HEK293T cells dramatically decreased ACE2 receptor association with GM1 clusters, despite increases in both GM1 cluster size and number (Supplementary Fig. 2a–c). Figure 2b shows representative dSTORM images comparing the disruption of HCQ on colocalization of ACE2 and GM1 clusters. At short distances (0–10 nm), pair correlation decreased 41 ± 18% (p < 0.05) (Fig. 2c, S3A), confirming that HCQ acts as a chaotrope to disrupt the ability of GM1 clusters to sequester ACE2. This result is in agreement with its anesthetic-like mechanism of action and its effect on PLD213.
As mentioned earlier, ACE2 moves to PIP2 clusters in resting/low-cholesterol conditions. PIP2 clusters reside near disordered lipids apart from GM1 clusters due to a large amount of unsaturation in PIP2’s acyl chains18,60. To determine whether ACE2 moves to PIP2 clusters after HCQ’s disruption of GM1 clusters, ACE2 and PIP2 clusters were co-labeled in HEK293T cells at resting/low-cholesterol levels, then treated with/without 50 µM HCQ.
Figure 2d shows representative dSTORM images comparing PIP2 clusters (purple labeling) before/after HCQ treatment in HEK293T cells. Surprisingly, the pair correlation between ACE2 and PIP2 clusters was decreased (Fig. 2e) at all distances (Supplementary Fig. 3b). The decrease of 77 ± 25% at short distances (0–10 nm) (p < 0.05) suggests that HCQ disrupts ACE2 association with both GM1 clusters and PIP2 clusters, presumably forcing the protein to disordered lipids in the plasma membrane. Further characterization of the PIP2 clusters showed HCQ treatment decreased both the size and number of PIP2 clusters by 20 ± 4%, and 44 ± 13% respectively (Supplementary Fig. 3c, d).
Next, we tested Vero E6 cells, which, as mentioned, endogenously express ACE2. Probing the nanoscale trafficking of ACE2 with endogenously expressed protein is important as overexpression can overwhelm the ability of GM1 lipids to sequester the receptor away from PIP2 and alter imaging results. Cholesterol-treated cells were fixed and stained for ACE2, and either GM1 clusters or PIP2 clusters. Within low-cholesterol conditions, ACE2 colocalized to both GM1 and PIP2 clusters (Supplementary Fig. 4a, b, Supplementary Table 1). Fifty micromolar HCQ treatment of ACE2 reduced the pair correlation between ACE2 and both GM1 clusters (59 ± 31%, p = 0.0789) and PIP2 clusters (32 ± 19%, p = 0.1071) (Fig. 2f, g).
In Vero E6 cells with elevated cholesterol, again, ACE2 associated with both GM1 and PIP2 clusters. However, 50 μM HCQ treatment had the greatest effect on ACE2’s association with both clusters (70 ± 17%, p < 0.01) (Fig. 2h, Supplementary Fig. 2g) compared to PIP2 clusters (52 ± 11%, p < 0.05) (Fig. 2i, Supplementary Fig. 2h). This suggests HCQ has its greatest effect on endocytic lipids when cholesterol is high (Fig. 2j).
Next, we tested A549 lung cells since HCQ appeared to affect the size of GM1 lipids differently compared to HEK293T and Vero (Supplementary Fig. 2). A549 lung cells were loaded in a manner identical to VeroE6 cells using apoE. Prior to HCQ treatment, ACE2 predominately associated with GM1 clusters in both resting cholesterol and high cholesterol. However, after loading cholesterol, ACE2 shifted out of PIP2 clusters (Supplementary Fig. 4c, d, i, j; Supplementary Table 1). Treatment with 50 μM HCQ disrupted colocalization of ACE2 receptors with GM1 clusters in both resting and high-cholesterol-treated cells (Fig. 3a, g), but not PIP2 (Fig. 3b, h). A trend toward increased PIP2 pair correlation suggested some translocation of ACE2 from GM1 to PIP2 clusters with high cholesterol may occur (p = 0.14, dashed arrow). The movement of ACE2 in A549 cells is summarized in Fig. 3e, k.
Lastly, we compared pair correlation in H1793 lung cells before and after cholesterol treatment. In resting H1793 cells (low cholesterol), most of the ACE2 was associated with PIP2 clusters (Supplementary Fig. 4e, f), but the association shifted toward GM1 clusters after loading cells with cholesterol (Supplementary Fig. 4k, l, Supplementary Table 1). After treatment with 50 μM HCQ, ACE2 was displaced from PIP2 clusters in resting cells and from GM1 clusters in high cholesterol (Fig. 3d, i), but not GM1 clusters in resting cholesterol or PIP2 clusters in high cholesterol (Fig. 3c, j). The movement of ACE2 in H1793 is summarized in Fig. 3f, l.
To estimate the distance ACE2 moves between compartments in the membrane, we labeled GM1 and PIP2 domains and performed two-color dSTORM on the lipids. We found the half maximal nearest neighbor ranges from 133 to 235 nm (Supplementary Table 1, and Supplementary Fig. 4s–v). The distribution of GM1 and PIP2 in the membrane appeared to be random, and their separation appeared uncorrelated with labeling intensity (Supplementary Fig. 4w, x).
HCQ’s disruption of PLD2
If the lipid disruption we observed with HCQ in lung and kidney cells is acting through the same membrane-mediated pathway as shown for general anesthetics, then we expect that HCQ also releases the anesthetic-sensitive protein phospholipase D2 (PLD2) from GM1 clusters. Anesthetics such as xenon, chloroform, isoflurane, propofol, and diethyl ether all displace PLD2 from GM1 clusters to activate an anesthetic pathway13,61.
To confirm HCQ’s anesthetic-like effect, clustering of PLD2 with GM1 lipids was monitored by two-color dSTORM in HEK293T cells with and without 50 µM HCQ treatment. 50 µM HCQ robustly disrupted PLD2 localization with GM1 clusters (Fig. 4a, b). Quantification of the % pair correlation at short radiuses (0–10 nm) decreased by 74 ± 13% (Fig. 4c). And this correlated with a decrease in the space between clusters (Fig. 4d). Hence HCQ’s effect on the lipid membrane is similar to general anesthetics (Supplementary Fig. 2a) in HEK293T cells. After treatment, the clusters are larger, and the ability to retain a palmitoylated protein (PLD2) is inhibited13.
Next, we tested the enzymatic activity of PLD2 in the presence of 50 µM HCQ. HCQ robustly inhibited PLD2 with a EC50 of 167 µM (standard error: 96–291 µM, Fig. 4e, f). The inhibition initially appeared similar to the direct binging that local anesthetic use to inhibit PLD20. However, when we tested HCQ’s ability to directly inhibit purified cabbage PLD, it had no effect (Fig. 4g), suggesting HCQ either has specificity for mammalian PLD2 over cabbage PLD, or HCQ inhibits by blocking its access to PIP2.
Erythromycin inhibits viral entry through perturbing GM1 clusters
Azithromycin is an antibiotic derived from erythromycin that is sometimes given in combination with HCQ. Although azithromycin has shown antiviral properties in numerous studies62,63,64,65, the results of its usage with COVID-19 patients in combination with HCQ have been mixed8. Based on the cholesterol sensitivity of SARS-CoV-2, we hypothesized that erythromycin could contribute to an antiviral effect through disruption of GM1 clusters leading us test to its effects on SARS2-PV.
We found erythromycin (100 µg/mL66) significantly inhibits SARS2-PV infection 69 ± 17% in HEK293T cells overexpressing ACE2 at normal cholesterol levels (Supplementary Fig. 5a). Consistent with the disruptive mechanism, the same treatment increased membrane fluidity 70 ± 11% (Supplementary Fig. 5b). Furthermore, using PLD2 activity as a surrogate for an effect in live cells showed a 12 ± 4% increase (Supplementary Fig. 5c), consistent with cluster disruption. When the cholesterol level was increased using apoE and serum, erythromycin was no longer effective. In fact, when the pair correlation of ACE2 and GM1 correlation was examined, the association of ACE2 with GM1 increased by 97 ± 55% (Supplementary Fig. 3e, f). This suggests erythromycin is unable to overcome the cholesterol-induced clustering of ACE2 with GM1 lipids in elevated cholesterol.
HCQ’s disruption of host defense peptides
Lastly, HCQ’s effects on host defense peptides were considered. Host defense peptides are amphipathic antimicrobial peptides that are upregulated during an immune response and perturb the membranes of microbes67,68. Cholesterol and GM1 cluster integrity show great importance to the modulation of both innate and acquired immune responses31. Amyloid-beta (Aβ) has been demonstrated to protect against microbial infection as a host defense peptide. The production of Aβ is regulated by the delivery of cholesterols to neurons by apoE32. ApoE regulates hydrolysis of amyloid precursor protein (APP) by clustering mechanism (Supplementary Fig. 1i). If HCQ disrupts GM1 lipids, then it is expected that HCQ to decrease APP pair correlation with GM1 lipids.
Aβ production was measured in HCQ-treated cells using a sandwich enzyme-linked immunosorbent assay (ELISA). HCQ was found to reduce Aβ generation by 11 ± 4% in cultured HEK293T cells (Supplementary Fig. 5d). The observed effect was statistically significant (p < 0.05). HEK293T cells were then loaded with cholesterol (apoE + serum) to better reflect the disease state of COVID-19 with severe symptoms. In the high-cholesterol state, HCQ did not inhibit Aβ production, resulting in a 24 ± 4% rise compared to production in low cholesterol. Since tetracaine and propofol also disrupt GM1 clusters, their effects on Aβ production were also tested and found to be very similar in both high- and low-cholesterol settings (Supplementary Fig. 5d, e).