Discovery of re-purposed drugs that slow SARS-CoV-2 replication in human cells

COVID-19 vaccines based on the Spike protein of SARS-CoV-2 have been developed that appear to be largely successful in stopping infection. However, vaccine escape variants might arise leading to a re-emergence of COVID. In anticipation of such a scenario, the identification of repurposed drugs that stop SARS-CoV-2 replication could have enormous utility in stemming the disease. Here, using a nano-luciferase tagged version of the virus (SARS-CoV-2- DOrf7a-NLuc) to quantitate viral load, we evaluated a range of human cell types for their ability to be infected and support replication of the virus, and performed a screen of 1971 FDA-approved drugs. Hepatocytes, kidney glomerulus, and proximal tubule cells were particularly effective in supporting SARS-CoV-2 replication, which is in- line with reported proteinuria and liver damage in patients with COVID-19. We identified 35 drugs that reduced viral replication in Vero and human hepatocytes when treated prior to SARS-CoV-2 infection and found amodiaquine, atovaquone, bedaquiline, ebastine, LY2835219, manidipine, panobinostat, and vitamin D3 to be effective in slowing SARS-CoV-2 replication in human cells when used to treat infected cells. In conclusion, our study has identified strong candidates for drug repurposing, which could prove powerful additions to the treatment of COVID.


INTRODUCTION
The COVID-19 pandemic caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus is having a widespread impact on global health with substantial loss of life. SARS-CoV-2 infection in patients with  can result in pulmonary distress, inflammation, and broad tissue tropism. Some infected individuals are asymptomatic whilst others mount an exaggerated immune response, or 'cytokine storm' (Chatenoud et al, 1991;Fajgenbaum & June, 2020), which correlates with multiple organ failure and poor outcome (Ruan et al, 2020). Vaccines have been developed to help protect people from COVID-19 but the emergence of mutational variants, some with increased transmissibility, calls for additional research into alternative treatment and prevention strategies.
An early step in the infection process is interaction of the Spike protein on the surface of the virus with angiotensinconverting enzyme 2 (ACE2) on the surface of the host cell (Li et al, 2003). Using this information, first-generation vaccines have been generated against the Spike protein. SARS-CoV-2 entry can be potentiated by additional host factors including neuropilin-1 (Cantuti-Castelvetri et al, 2020). Once inside, the virus uses components of the host cell to replicate and secrete viral particles (Maier et al, 2016) and disrupt RNA handling and protein translation to suppress host defenses (Banerjee et al, 2020). The different stages of the disease, from the initial infection of host cells through to virus replication and the response (normal or extreme) of the immune system, offer opportunities to identify drugs, treatments and therapies to help stop disease progression. Furthermore, large proportions of the world's population remain at risk of contracting COVID-19 as they wait to be vaccinated, which has prompted governments to encourage people to continue to self-isolate, remain at home, and maintain social distancing. The identification of safe and easily distributed medications that can target the different stages of virus infection and replication, could reduce the spread of SARS-CoV-2 and reduce the cases of COVID-19.
The Food and Drug Administration (FDA) and the European Medicines Agency (EMA) work with pharmaceutical companies to develop safe and effective drugs for the benefit of public health. Using a library of 1971 FDA-approved compounds, here we used a traceable clone of SARS-Cov-2 to identify well-characterized drugs that could potentially slow the infection and replication of the SARS-CoV-2 virus in human cells.
We had previously developed an approach to quantitate collagen synthesis by using CRISPR-Cas9 to insert the gene encoding NanoLuciferase (NLuc) into the Col1a2 gene in fibroblasts (Calverley et al, 2020). Here, we adapted the approach to tag the SARS-CoV-2 virus with NLuc, and then used the recombinant NLuc-tagged virus to monitor virus replication and to identify drugs that slow the replication process. SARS-CoV-2 was originally recovered by culturing human specimens in the African green monkey kidney cell line, Vero (Zhou et al, 2020), and the virus readily replicates in this cell type. We have performed initial screening in the Vero cell line, and in order to draw reliable conclusions from our studies we performed a replicate screen in a human hepatocyte cell line, HUH7. We evaluated a panel of human cell types for their ability to be infected by SARS-CoV-2 and to support virus replication, and optimized culture conditions for each cell type to provide a robust system for screening in human cells. We found that hepatocytes and kidney epithelial cells were proficient in supporting SARS-CoV-2 replication whereas fibroblasts and lung epithelial cells supported only minimal replication.
Through applying stringent criteria in our hit selection and confirming efficacy of these compounds in suppressing replication in cells already infected with SARS-CoV-2, we identified a shortlist of nine drugs that were effective in inhibiting SARS-CoV-2 replication. The drugs identified included anti-cancer and anti-viral drugs, but also less toxic drugs such as atovoquone, ebastine and vitamin D3. The multiple steps involved in virus infection and proliferation, means that each of these drugs may have different molecular targets, and for some, additional targets are still being elucidated. As these drugs are FDA-approved and safe, and dosimetry has been established for use in people, clinical trials could be initiated for these drugs.

Generation of a traceable SARS-CoV-2 virus
High throughput screens have been developed to identify drugs suitable for re-purposing for treatment of COVID19 (e.g., Riva et al, 2020)). These screens have been performed in non-human cell lines, such as Vero, and rely on secondary factors such as cell viability to identify candidates, or markers that indicate virus infection. To monitor the replication of SARS-CoV-2 we have generated a modified traceable virus where Orf7a is replaced with the sequence encoding NanoLuciferase (SARS-Cov-2-ΔOrf7a-NLuc, Fig. 1A). NanoLuciferase (NLuc) is an enzyme that produces light when supplied with its substrate (coelenterazine) and is readily detectable even at low quantities (Calverley et al., 2020). Orf7a has previously been removed in SARS-CoV and SARS-CoV-2 and yielded infectious and replicative virus particles (Thi Nhu Thao et al, 2020;Xie et al, 2020a;Xie et al, 2020b). We have used a reversegenetics approach to generate virus particles (Supplemental Fig. 1), and plaque forming assays demonstrated that the replication of SARS-CoV-2-ΔOrf7a-NLuc viruses were equivalent to the wild-type SARS-CoV-2 (Wuhan-Hu-1, NC_045512.2) (Fig. 1B). Viral RNAs (Supplemental Fig. 1), and electron microscopy ( Fig. 1C-D, Supplementary Fig. 2) confirmed virus production. NLuc activity was readily detected in infected cultures upon addition of the substrate coelentrazine (Fig. 1E).

NLuc activity as a marker of virus replication
To monitor replication of the SARS-CoV-2-∆Orf7a-NLuc virus, Vero cells were exposed to increasing numbers of replicative virus particles (plaque forming units, PFU) and luminescence was used as a measure of virus replication. Luminescence measurements were compared to SARS-CoV-2-∆Orf7a-NLuc virus in the absence of cells as background ( Fig. 2A), 24, 48 and 72 hours post infection (h.p.i.). A lag in virus replication of at least 48 hrs was observed but replication was readily observed at 72 hrs, when as little as 2 PFU per well (MOI 0.0004) was added. Using these optimization experiments 72 h.p.i. was chosen as the time point at which to assess virus replication in subsequent assays, the average Z' factor (Zhang et al, 1999) for our replication assay in 96 well format was 0.8 indicating suitability for library screening. Amplification of NLuc signal upon virus infection was consistent for all passages of the traceable virus (Supplementary Fig. 3) and with varying cell number (Supplementary Fig. 4).
Earlier studies have demonstrated interferon alpha 2 (IFNA2) is able to reduce SARS-CoV-2 replication (Xie et al., 2020a). We confirmed this observation using our assay ( Fig. 2C) but noted that with higher virus load (100 PFU), IFNA2 had little effect on virus replication (Fig. 2D). These results provided confidence that the traceable virus could be used to quantify the effects of drugs on virus replication.

SARS-CoV-2 replication screen validation
High throughput screens have been developed to identify drugs suitable for re-purposing for treatment of COVID19 (e.g., (Jeon et al, 2020;Riva et al., 2020;Touret et al, 2020)). These screens have been performed in non-human cell lines, such as Vero, and rely on secondary factors such as cell viability to identify candidates, or markers that indicate virus infection. A first step in validating our viral replication screen was to measure virus replication after treatment with interferon alpha 2 (IFNA2) which has previously demonstrated to reduce virus replication (Xie et al., 2020a). When Vero cells were infected with a single PFU, pre-treatment of IFNA2 was able to reduce replication of SARS-CoV-2-ΔOrf7a-NLuc in a dose dependent manner. However, with higher virus load (100 PFU), IFNA2 had little effect on virus replication ( Fig. 2C and D). These results provided confidence that the NLuc-tagged virus system could be used to quantify the effects of drugs on virus replication, especially if reduction of bioluminescence was detected with high PFU.

Drug repurposing to target SARS-CoV-2 replication.
A 1971 FDA-approved compound library was used to pretreat Vero cells before infection with either 1PFU (MOI 0.0002) or 100 PFU (MOI 0.02) SARS-CoV-2-∆Orf7a-NLuc virus as outlined in Fig. 3A. Infected Vero cells that were untreated and background (virus only wells) were used as controls. The Z' factor was assessed for each plate of the screen averaging 0.867± 0.140 SD and 0.915± 0.058 SD, for the 1PFU and 100PFU screen, respectively demonstrating a robust assay. More hits were identified in the 1PFU screen and therefore we focused our hit selection on compounds that reduced virus replication by at least 85% in the 100PFU screen ( Fig. 3B and Supplementary Table  3). We identified 69 compounds that reduced replication (FDR<0.1 compared to untreated and DMSO controls, 55 showed no viral replication). These hits were also identified in the duplicate screen performed with 1PFU per well (Fig. 3C). Five of these compounds were false positive hits due to direct inhibition of NLuc activity (Fig. 3C, Supplemental Table 4).

Drug repurposing screen for SARS-CoV-2 replication in human cells.
Having established an effective screening strategy in Vero cells, we sought to establish a similar approach in a human cell line capable of supporting SARS-CoV-2 replication. SARS-CoV-2 has been reported to infect multiple cell lines, and replication of the viral genome has been identified (Chu et al, 2020). We tested the replicative capacity of SARS-CoV-2-∆Orf7a in the lung epithelial cell lines Calu3, A549, and 16HBEo and in lung fibroblasts. Using real-time PCR and western blotting for the SARS-CoV-2 nucleocapsid protein, we observed virus replication in these cell lines, and we confirmed that SARS-CoV-2 WT and SARS-CoV-2-ΔOrf7a-NLuc viruses infected these lung cell lines (Supplemental Fig. 5). However the Z' factor suggested that they were not suitable for high throughput screening (Fig. 4A). We assessed additional cell lines reported to be infected by SARS-CoV-2 including Caco-2 (Chu et al., 2020;Touret et al., 2020) and HUH7 (Zhou et al., 2020), as well as additional kidney cell lines. These were chosen as we were able to recover virus particles from 293T cells and kidney is reported to express high levels of ACE2 (Wysocki et al, 2020). Replication in monocytic cells and fibroblastic cell lines was also monitored due to the systemic and fibrotic nature of COVID-19. We identified virus replication in HUH7 cells, microvascular cells of the kidney glomerulus, and proximal tubule cells of the kidney and THP1 monocytic cells (Fig. 4A).
As each cell line has a specific culture medium for optimal growth, we set out to test if different media would support cell growth and virus replication in lung epithelial cells, as this is the primary site of Sars-CoV2 infection. We used 15 different medium conditions to grow lung epithelial cells prior to infection with SARS-CoV-2-ΔOrf7a-NLuc; for comparison, HUH7 cells were also included. Changes in growth conditions improved assay performance in lung epithelial and fibroblasts but signal-to-noise ratios remained low. In contrast, for HUH7 cells, these optimized conditions significantly improved the 96 well format assay as indicated by Z' factor >0.5 and signal-to-noise ratios >20 ( Fig. 4B and C). Using the 1971 FDA-compound library we identified 223 compounds that suppressed SARS-CoV-2 replication by greater than 85% whilst maintaining cell viability. We refined these hits by overlapping with positive hits from the screen in Vero cells (Fig. 4E, Supplemental Table 5). We identified 35 inhibitory compounds and 2 compounds that increased NLuc activity, that were common to the two screens (Fig. 4F). The intended clinical use or targets of the 35 inhibitory compounds included anti-virals, antibiotics, modifiers of dopamine and estrogen receptor activity, calcium ion channel inhibitors and HMG-CoA reductase inhibitors (Fig. 5A). The hits also identified vitamin D3, which is available over the counter. Multiple vitamin D related compounds were present in the screen and these too suppressed SARS-CoV-2-∆Orf7a-NLuc although these did not all meet our criteria for both Vero and human cell lines (Fig. 5B).
These initial hits were further tested to determine if treatment could also prevent replication in cells already infected with SARS-CoV-2-∆Orf7a-NLuc. We infected Vero cells with SARS-CoV-2 for 24 hrs (which is sufficient time for the virus to begin replication) then added each of the 35 inhibitor compounds. The cells were incubated for 48 hrs (i.e. a total of 72 h.p.i) prior to bioluminescence being measured. The majority of the compounds had no impact on virus replication; but 9 of the 35 compounds reduced replication relative to DMSO controls (Table 1, Supplemental Fig. 6). The effective doses for these compounds were then determined in HUH7 cells to provide a future reference for selecting dose after comparison to pharmacokinetic data established in human trials (Fig. 5, Table 1). Whilst these are pre-clinical in vitro data, they demonstrate the efficacy of the compounds in reducing virus replication post infection and warrant further investigation to determine if these could ease the burden of the virus in patients.

DISCUSSION
In this study we have shown that the SARS-CoV-2 virus infects and replicates in a range of human cells especially hepatocytes, kidney glomerulus, and proximal tubule cells of the kidney, and, that 9 drugs that have previously been shown to be safe in humans and approved by the FDA for clinical use are effective in inhibiting SARS-CoV-2 infection and replication.
The identification that liver and kidney cells are infection and replication competent for SARS-CoV-2 aligns with the observed liver and kidney abnormalities in patients with COVID-19. Liver comorbidities have been reported in 2-11% of patients with COVID-19, and 14-53% of cases reported abnormal levels of liver enzymes , with liver injury being more prevalent in severe cases (reviewed by ). Another study reported that patients with chronic liver disease, especially African Americans, were at increased risk of COVID-19 . It has been suggested that liver damage in patients with COVID-19 might be caused by viral infection of liver cells, which is supported by the presence of SARS-CoV-2 RNA in stool (Yeo et al, 2020). On a similar note, acute kidney injury is a common complication of COVID-19 and has been associated with increased morbidity and mortality (reviewed by (Braun et al, 2020)). Our data showing SARS-CoV-2 infection and replication in kidney cells helps to explain kidney pathology associated with COVID-19.
RECOVERY (Group) has set out to identify therapeutics that could be re-purposed for treatment of COVID patients. For example, dexamethasone, which is a broad spectrum immunosuppressor, has been applied clinically to inhibit the destructive effects of the cytokine storm and shown to reduce mortality and decrease the length of hospitalization (Group et al, 2021), did not demonstrate any effect on viral replication in our study. It may be possible in the future to combine the therapeutics identified in our study with dexamethasone so that both the cytokine storm and virus replication are targeted.
Of interest, other therapeutics that have been tested in the RECOVERY trial include ritonavir and lopinavir (antiretroviral protease inhibitors), which failed to show clinical benefit in the treatment of COVID-19 (Group). These drugs were present in the DiscoveryProbe library used in our study but showed no significant effect on SARS-CoV-2 replication, in our assays.
The high costs and lengthy lead-in times associated with new drug development, make repurposing of existing drugs for the treatment of common and rare diseases an increasingly attractive idea (for review see (Pushpakom et al, 2019)). The approaches used include hypothesis driven, preclinical trials including computational (e.g., computational molecular docking) and experimental (e.g., biochemical or cell-based assays of drug interactions) methods, and evaluation of efficacy in phase II clinical trials. As noted by Pushpakom et al., of these three steps, the identification of the right drug for an indication of interest (in our case SARS-CoV-2 and COVID-19) is critical.
The nine drugs that we identified here have been approved for use in the treatment of a variety of diseases. Panobinostat, for example, is a HDAC inhibitor that blocks DNA replication, and has been used to inhibit cell growth in the management of cancer. Panobinostat had the strongest effect on limiting SARS-CoV-2 replication whilst maintaining cell viability, and completely blocked replication of SARS-CoV-2 at all doses tested ( Figure 5); however, if cells were infected prior to treatment a more modest effect on replication were observed. This difference may be related to the recent observations that panobinostat can suppress ACE2 expression (He & Garmire, 2020), which would explain its beneficial effect prior to entry of the virus into cells and not thereafter. Abemaciclib (LY2835219) is another cell cycle inhibitor, suppressor of DNA replication and anti-cancer drug that emerged from the screen however these are likely to be detrimental to recovery from SARS-CoV-2 infection.
Atovaquone is of particular interest because it has been identified in other studies of SARS-CoV-2 in the context of COVID-19. It is a hydroxynaphthoquinone approved by NICE for the treatment of mild to moderate pneumocystis pneumonia and as a prophylaxis against pneumocystis pneumonia. It also has been used in combination with Proguanil (Malarone) as an antimalarial. The mechanism of action of atovaquone has been widely studied; it is a competitive inhibitor of ubiquinol, and against P. falciparum, it acts by inhibiting the electron transport chain at the level of the cytochrome bc1 complex (Fry & Pudney, 1992). A more recent study showed that atovaquone inhibits Zika and Dengue virus infection by blocking envelop protein-mediated membrane fusion (Yamamoto et al, 2020). Furthermore, in silico molecular docking strategies suggested potential binding of atovaquone to the SARS-CoV-2 spike protein (Farag et al, 2020;Ramirez-Salinas et al, 2020). These two mechanisms of action help to explain how atovaquone slows both infection and replication of SARS-CoV-2 in cells.
Our observation that bedaquiline inhibits SARS-CoV-2 replication supports evidence from an in silico repurposing study in which bedaquiline was proposed to be a promising inhibitor of the main viral protease of SARS-CoV-2 (Ferraz et al, 2020). The main protease cleaves pp1a and pp1ab polypeptides, which encode nonstructural proteins to form the replication-transcription complex (Snijder et al, 2006), and help explain how inhibitors to the main protease are effective in inhibiting replication of SARS-CoV-2 in Vero 76 cells (Ma et al, 2020). Manidipine (a calcium ion channel blocker approved for the use in treating hypertension) is proposed to reduce the activity of the main protease of SARS-CoV-2 (Ghahremanpour et al, 2020) and limit the availability of calcium ions, which are required for insertion of the coronavirus fusion peptide into the host lipid bilayers during viral fusion (Lai et al, 2017).
Our screen also identified ebastine (a second generation H1 receptor antagonist that has been approved for the treatment of allergic rhinitis and chronic idiopathic urticaria) and vitamin D3, which is a health supplement available over the counter. Ebastine is a second generation H1-antihistamine with proven efficacy for treating seasonal allergic rhinitis. However, histamine antagonists exhibit effects in addition to blockage of the histamine receptor. For example, ebastine blocks the release of anti-IgE-induced prostaglandin D2 (Campbell et al, 1996). The vitamin D receptor (VDR, a member of the nuclear hormone receptor superfamily) is proposed to be essential for liver lipid metabolism because its deficiency in mice protects against hepatosteatosis (Bozic et al, 2016). Once bound to VDR, vitamin D plays a major role in hepatic pathophysiology, regulation of innate and adaptive immune responses, and might contribute to anti-proliferative, anti-inflammatory and anti-fibrotic activates (reviewed by (Triantos et al, 2021)). Thus far, whether vitamin D supplementation reduces the risk of SARS-CoV-2 infection or COVID-19 severity is unclear (The Lancet Diabetes, 2021). Whilst vitamin D3 met the stringent cut-offs of 85% virus reduction, vitamin D2 and other Vitamin D related therapeutics also reduced virus replication but did not meet all criteria across both Vero and HUH7 cell lines.
In conclusion, our study has identified compounds that are safe in humans and show effectiveness in reducing SARS-CoV-2 infection and replication in human cells, especially hepatocytes. Their potency in stopping SARS-CoV-2 replicating in human cells in the face of the COVID pandemic, warrants further study.

MATERIALS AND METHODS Cell culture
Cell lines maintained in growth medium are shown in Supplementary Table 1.

Generation of functional SARS-CoV-2 virus
DNA encoding the genome of SARS-CoV-2 and SARS-CoV-2-ΔOrf7a-NLuc were purchased from Vectorbuilder Inc. (Chicago, US). Transfection of virus encoding the DNA failed to generate replicative virus when electroporated into 293T cells. We therefore produced RNA molecules which encoded the virus in vitro. Briefly, virus encoding DNA (1 µg) was transcribed using T7 mMessenger mMachine (Thermo) with a GTP:Cap ratio of 2:1 used in a 20 µL reaction. In addition, RNA encoding the SARS-CoV-2 nucleocapsid was also generated by PCR using primers P1 and P2 (Supplementary Table 2). It has been reported that this aids the recovery of replicative virus (Insert ref 3). Viral RNA genomes (10 µl) and 2.5 µL nucleocapsid RNA were electroporated into 293T cells within the BSL3 laboratory. Viral RNAs were electroporated (5,000,000 cells, 1100 V, 20 ms and 2 pulses) and grown in T75 cm 2 flasks and 24 well plates. Cells grown in 24 well plates for 24-120 hrs were used to monitor changes in NLuc activity.

Virus production, maintenance and assessment of titer
Culture medium was collected from 293T cells 6 days after electroporation. This virus (P0) was used to infect cells of interest. As virus replication was slow in 293T cells, virus stocks were maintained by passage in Vero cells grown in DMEM supplemented with 2% FBS. Medium (1 mL) was used to infect Vero cells in order to generate P1 virus. Replication was assessed by measuring NLuc activity over 10 days. For subsequent passage of the virus, Vero cells in T75 cm 2 flasks were infected with 2 mL of medium containing virus, after 3 days the medium was collected and passed through 0.45 µm filters using Luer-loc syringes.
To titre the virus, 200,000 Vero cells were seeded in 6-well plates overnight in growth medium. After removing growth medium Vero cells were infected with 200 µL of serially diluted virus containing medium at 37 °C. After 1 hour, wells were overlaid with 2 mL of 0.3% low melt agarose in 2xDMEM containing 1% FBS and grown for 3 days. After 3 days infected cells were fixed with 10% PFA overnight and then stained with crystal violet. Plaques were identified by imaging plates on BioDoc-It gel documentation system (UVP, Upland, US).
To detect viral RNA in medium, 0.25 mL of virus containing medium was collected 3 days post infection, 0.75 mL TriPure LS reagent (Sigma-Aldrich St. Louis, US) was added, and RNA isolated according to the manufacturer's recommendations, in the final step RNA was dissolved in 15 µL DNAse/RNAse free water. For detection of SARS-COV-2 nucleocapsid transcripts in lung epithelial cells, 200,000 cells were infected at the indicated MOI for 3 days, monolayers were lysed directly in 1 mL Trizol (Invitrogen, Paisley UK). For assessing expression of ACE2, TMPRSS2 and NLP1 expression in lung epithelial cells RNA was isolated from the cells using Trizol and RNA isolated according to the manufacturer's recommendations. For cDNA generation and real-time PCR we used conditions previously described (Pickard et al, 2019) using primers detailed in Supplementary Table 2.

Electron microscopy
After 3 days infection with P4 virus, Vero cells were scrapped and pelleted. Cell pellets were fixed using 2.5% glutaraldehyde and 4% paraformaldehyde in 0.1 M cacodylate buffer for 24 h, washed in ddH2O three times, 30 mins for each wash. Cell pellets were incubated in freshly made 2% (vol/vol) osmium tetroxide and 1.5% (wt/vol) potassium ferrocyanide in cacodylate buffer (100 mM, pH 7.2) for 1 hr at room temperature. Samples were washed in ddH2O five times each for 3 minutes. Specimens were transferred to freshly made 1% (wt/vol) tannic acid in 100 mM cacodylate buffer (pH 7.2) for 40 mins at RT and washed in ddH2O five times for 3 mins each at RT. The specimen was incubated with 1% (vol/vol) osmium tetroxide in ddH2O for 30 minutes at room temperature and washed in ddH2O three times for 5 min each at room temperature. The specimen was then incubated with 1% (wt/vol) uranyl acetate (aqueous) at 4 °C for 16 hrs (overnight) and then washed in ddH2O three times for 5 mins each time at room temperature.
Specimens were dehydrated in graded ethanol: 30, 50, 70, 90% (vol/vol) ethanol in ddH2O for 10 mins at each step. Then samples were washed four times for 10 mins each time in 100% ethanol at room temperature. Samples were transferred to propylene oxide for 10 min at room temperature.
The specimen was finally infiltrated in a graded series of Agar100Hard in propylene oxide at room temperature: first for 1 hour in 30% (vol/vol) Agar100Hard, 1 hr in 50% (vol/vol) Agar100Hard then overnight in 75% (vol/vol) Agar100Hard, and then 100% (vol/vol) Agar100Hard for 5 hrs. After transferring samples to freshly made 100% Agar100 Hard in labelled molds and allowed to cure at 60 °C for 72 hrs. Sections were imaged on an FEI Tecnai12 BioTwin.

Drug screens
The DiscoveryProbe FDA-approved library of 1971 compounds (L1021, APExBIO Boston, US) was prepared as follows. After thawing the library for 4 hours at room temperature the library was arrayed into 96 well plates at 1 mM in DMSO and stored at -20 °C. Stocks (1 mM) were thawed at room temperature for 2 hrs before compounds were added to cells.
For all drug screens and validation, 5000 cells were seeded in 50 µL of growth medium for 24 hrs in white walled microwell plates. DMEM containing 2% FBS was used for Vero cells and alphaMEM containing 10% FBS was used for HUH7 cells. The following day 0.5 µL of each compound (final concentration 10 µM) was added per well and incubated for 24 hours. Eighty-eight compounds were tested per plate and each plate contained the following controls: two untreated wells, two DMSO treated wells, and one well treated with 10 µM puromycin to kill cells. SARS-CoV-2-ΔOrf7a-NLuc virus was added to all wells. In addition, two wells did not contain any cells but were infected with the SARS-CoV-2-ΔOrf7a-NLuc virus as a measure of background NLuc activity. A final well which contained cells but were uninfected were also included. Twenty-four hours after drug treatment, cells were infected with 100 PFU per well SARS-CoV-2-ΔOrf7a-NLuc virus in 50 µL of the indicated growth medium for 72 hrs. To assess virus replication and viability, 2.5 µL of Prestoblue was added to each well and plates incubated for 10 mins at 37 °C. Coelenterazine was then added to a final concentration of 3 µM. Plates were sealed prior to reading luciferase activity and viability as described above. Validation of hits were performed using the same procedures described for the drug screen.

Study design and statistical analysis
To identify hit compounds, raw luciferase luminescence reads ( ) were normalised relative to the virus-infected druguntreated controls ( ) and plate minimum read ( ) on each plate by the formula: This meant that a normalized luciferase value of 1 implied no difference from untreated virus replication, and a value of 0 represented total inhibition of viral replication.
The PrestoBlue reads ( ) were normalised relative to the virus-infected DMSO-treated controls ( ) and plate minimum read ( ) on each plate by the formula: This meant that a normalized PrestoBlue value of 1 implied no difference in cell viability from DMSO-treated virus infected cells, and a value of 0 representing maximal reduction in cell viability.
Z-scores were calculated relative to the mean log2(fold change) for each plate.
Where indicated one-tailed Student's T-test were performed to evaluate significance in changes to NLuc activity.

Contributors
AP co-conceived the project, performed the experiments, interpretated the data, drafted the manuscript. BCC performed experiments and interpreted the data. JC performed experiments. RG performed experiments. YL performed experiments. KEK co-conceived the project, raised the funding, drafted the manuscript. All authors approved the final manuscript.

Declaration of interests
The authors have no conflicts of interest.

Acknowledgements
The research was funded by Wellcome (  Inhibitor heme polymerase activity

Figure 2: Timings of SARS-CoV-2-ΔOrf7a-NLuc virus replication.
A) Schematic showing how SARS-CoV-2 ΔOrf7a-NLuc was used to assess virus replication. During virus replication as virus particles are released from the cell, or as a result of cell death, NLuc activity is detected in the conditioned medium together with virus particles. In subsequent assays the NLuc activity is recorded in the absence of cells, and NLuc signals above background signal are used to indicate the level of virus replication.
B) The replication of SARS-CoV-2-ΔOrf7a-Nluc was monitored in a 96 well assay. Using 5000 Vero cells per well, increasing numbers of virus particles were added and monitored 24, 48 and 72 hours post infection (h.p.i). Substantial and significant increases in NLuc activity were observed when only two plaque forming units (PFU) were added. Higher NLuc activity was associated with higher viral input. Even with higher viral inputs the substantial increases in NLuc activity occurred only after 72 h.p.i.

Figure 3: Screen of 1971 FDA-approved compounds to identify therapeutics that inhibit SARS-CoV-2-ΔOrf7a-NLuc virus replication.
A) Schematic of the screening procedure used to assess whether FDA-approved compounds alter SARS-CoV-2 ΔOrf7a-Nluc virus replication.

Figure 4: Replication of SARS-CoV-2-ΔOrf7a-Nluc in human cell lines.
A) SARS-CoV-2 replication in human cell lines. Five thousand cells per well were seeded and infected as in Figure 3A. All cells were grown in the recommended growth medium, detailed in Table 1. The NLuc activity after 3 days was used to assess the degree of virus replication. NLuc activity above the baseline activity is shown as a heat map. Each box is the average of 2 biological repeats of each cell line at each virus dose, with the same findings observed when 2000, or 10,000 cells were seeded. B) Lung epithelial, fibroblasts, HUH7 and 293T cell lines were grown in 15 different growth medium compositions. NLuc activity over background three days after infection is shown. Each row represents a different growth medium as indicated in C). Each box represents 4 replicate measures of replication. Similar findings were observed in n=6 biological repeats. Z' factors were calculated for each virus dose using 'virus only' wells as negative controls, the average Z' factor for each growth medium is shown. The signal-to-noise ratio of infected cells relative to 'virus only' wells are also shown.
C) NLuc activity above background three days after infection in HUH7 cells. Each box represents 4 replicate measures of replication. Similar findings were observed in n=6 biological repeats.
D) The effects of 1971 FDA-approved compounds on the replication of SARS-CoV-2-∆Orf7a-NLuc (100 PFU/well, MOI 0.02) in HUH7 cells grown in alpha-MEM supplemented with 10% FBS. Five thousand cells were treated with 10 µM of each compound for 24 hours prior to infection with 100 PFU of virus. Virus replication progressed for 72 hours as in Figure 3A.
E) Comparison of compounds identified by screening for HUH7 and Vero cells Z-scores for luciferase activity are shown F) Comparison and overlap of inhibitors identified in the compound screening for HUH7 and Vero cells. C) Dose responses to the 9 compounds in HUH7 cells. Cells were treated and infected as described in Figure 4D. NLuc activity relative to DMSO controls are shown in blue, and viability relative to DMSO are shown in black. N=2 independent repeats are shown.

Supplementary Figure 1: Recovery of replication competent SARS-CoV-2 from synthetic DNA constructs.
A) Schematic for the recovery of SARS-CoV-2 virus particles from DNA encoding the wild type and NLuc modified SARS-CoV-2 genome. RNA was transcribed from the synthetic DNA constructs and electroporated into 293T cells, the cells were monitored daily for NLuc activity and evidence of the cytopathic effects of virus replication. Virus containing medium was collected 6 days post electroporation.

Supplementary Figure 3: Replication assays for different passages of SARS-CoV-2-ΔOrf7a-NLuc virus.
A) The plots show 5000 Vero cells infected with different volumes of virus harvested from repeat passages of the virus (P) in Vero cells. With increasing viral load, there is an increase in background as more NLuc activity is added to each well (Blue). The background signal arises when NLuc, expressed as the virus replicates, is released from infected cells that have lysed upon liberation of the virus particles. N=2 independent experiments. B) After subtracting the background noise observed in wells without cells (blue in A), the additional NLuc activity generated through viral replication showed comparable signals for each virus passage. N=2 independent experiments.
C) The signal-to-noise ratio for each batch was used to identify the optimal conditions for virus replication for subsequent screening experiments.

Supplementary Figure 4: SARS-CoV-2-ΔOrf7a-Nluc replication optimisation
NLuc activity at different times after infection with increasing numbers of SARS-CoV-2 virus particles. Enhanced NLuc signals were observed 72 h.p.i. when either A) 2000, B) 5000 or C) 10,000 cells per well were used. Maximal signals were generated with 5000 cells per well, as seeding 10,000 cells tended to reduce NLuc activity. Five thousand cells were used for all subsequent assays.

Supplemental Figure 6: Effective compounds that suppress SARS-CoV-2 replication with treatment commencing after infection.
A) Schematic for assessing whether any compounds that suppressed virus replication prior to infection could also affect replication post-infection. Figure 3 were assessed to identify dose-dependent effects on SARS-CoV-2 replication with treatment progressing after infection (as outlined in A). Nine compounds were found to suppress replication in a dose dependent manner.

A)
Schematic showing how SARS-CoV-2 ΔOrf7a-NLuc was used to assess virus replication. During virus replication as virus particles are released from the cell, or as a result of cell death, NLuc activity is detected in the conditioned medium together with virus particles. In subsequent assays the NLuc activity is recorded in the absence of cells, and NLuc signals above background signal are used to indicate the level of virus replication.

B)
The replication of SARS-CoV-2-ΔOrf7a-Nluc was monitored in a 96 well assay. Using 5000 Vero cells per well, increasing numbers of virus particles were added and monitored 24, 48 and 72 hours post infection (h.p.i). Substantial and significant increases in NLuc activity were observed when only two plaque forming units (PFU) were added. Higher NLuc activity was associated with higher viral input. Even with higher viral inputs the substantial increases in NLuc activity occurred only after 72 h.p.i.

C)
Five thousand Vero cells were pre-treated with increasing doses of IFNα2 for 24 hours prior to infection with 1PFU SARS-CoV-2-ΔOrf7a-Nluc. NLuc activity and viability were assessed 72 h.p.i. demonstrating effective inhibition of virus replication. The blue line indicates the luminescence counts per second for DMSO treated controls. N=3 replicate samples D) As in C, 5000 Vero cells were treated with increasing doses of IFNα2 for 24 hours prior to infection with 100PFU SARS-CoV-2-ΔOrf7a-Nluc. With this higher titre, virus replication was not inhibited. The blue line indicates the luminescence counts per second for DMSO treated controls. N=3 replicate samples.   A) SARS-CoV-2 replication in human cell lines. Five thousand cells per well were seeded and infected as in Figure 3A. All cells were grown in the recommended growth medium, detailed in Table 1. The NLuc activity after 3 days was used to assess the degree of virus replication. NLuc activity above the baseline activity is shown as a heat map. Each box is the average of 2 biological repeats of each cell line at each virus dose, with the same findings observed when 2000, or 10,000 cells were seeded.

B)
Lung epithelial, fibroblasts, HUH7 and 293T cell lines were grown in 15 different growth medium compositions. NLuc activity over background three days after infection is shown. Each row represents a different growth medium as indicated in C). Each box represents 4 replicate measures of replication. Similar findings were observed in n=6 biological repeats. Z' factors were calculated for each virus dose using 'virus only' wells as negative controls, the average Z' factor for each growth medium is shown. The signal-to-noise ratio of infected cells relative to 'virus only' wells are also shown.   C) Dose responses to the 9 compounds in HUH7 cells. Cells were treated and infected as described in Figure  4D. NLuc activity relative to DMSO controls are shown in blue, and viability relative to DMSO are shown in black. N=2 independent repeats are shown.

A)
Schematic for the recovery of SARS-CoV-2 virus particles from DNA encoding the wild type and NLuc modified SARS-CoV-2 genome. RNA was transcribed from the synthetic DNA constructs and electroporated into 293T cells, the cells were monitored daily for NLuc activity and evidence of the cytopathic effects of virus replication. Virus containing medium was collected 6 days post electroporation.

B)
NLuc activity in 293T cells electroporated with wild type and NLuc modified SARS-CoV-2 RNA transcripts.

C)
Real-time PCR primer validation using SARS-CoV-2 encoding DNA. NLuc coding sequences were inserted in place of the Orf7a gene. Primers designed to amplify sequences within the DNA constructs confirmed primer specificity.  A) The plots show 5000 Vero cells infected with different volumes of virus harvested from repeat passages of the virus (P) in Vero cells. With increasing viral load, there is an increase in background as more NLuc activity is added to each well (Blue). The background signal arises when NLuc, expressed as the virus replicates, is released from infected cells that have lysed upon liberation of the virus particles. N=2 independent experiments.

B)
After subtracting the background noise observed in wells without cells (blue in A), the additional NLuc activity generated through viral replication showed comparable signals for each virus passage. N=2 independent experiments.

C)
The signal-to-noise ratio for each batch was used to identify the optimal conditions for virus replication for subsequent screening experiments.

Supplementary Figure 4: SARS-CoV-2-ΔOrf7a-Nluc replication optimisation
NLuc activity at different times after infection with increasing numbers of SARS-CoV-2 virus particles. Enhanced NLuc signals were observed 72 h.p.i. when either A) 2000, B) 5000 or C) 10,000 cells per well were used. Maximal signals were generated with 5000 cells per well, as seeding 10,000 cells tended to reduce NLuc activity. Five thousand cells were used for all subsequent assays.

A)
Schematic for assessing whether any compounds that suppressed virus replication prior to infection could also affect replication post-infection.

B)
Fifty compounds identified to suppress SARS-CoV-2 replication in Figure 3 were assessed to identify dose-dependent effects on SARS-CoV-2 replication with treatment progressing after infection (as outlined in A). Nine compounds were found to suppress replication in a dose dependent manner.