In vitro and in vivo identification of clinically approved drugs that modify ACE2 expression

Abstract The COVID‐19 pandemic caused by SARS‐CoV‐2 has is a global health challenge. Angiotensin‐converting enzyme 2 (ACE2) is the host receptor for SARS‐CoV‐2 entry. Recent studies have suggested that patients with hypertension and diabetes treated with ACE inhibitors (ACEIs) or angiotensin receptor blockers have a higher risk of COVID‐19 infection as these drugs could upregulate ACE2, motivating the study of ACE2 modulation by drugs in current clinical use. Here, we mined published datasets to determine the effects of hundreds of clinically approved drugs on ACE2 expression. We find that ACEIs are enriched for ACE2‐upregulating drugs, while antineoplastic agents are enriched for ACE2‐downregulating drugs. Vorinostat and isotretinoin are the top ACE2 up/downregulators, respectively, in cell lines. Dexamethasone, a corticosteroid used in treating severe acute respiratory syndrome and COVID‐19, significantly upregulates ACE2 both in vitro and in vivo. Further top ACE2 regulators in vivo or in primary cells include erlotinib and bleomycin in the lung and vancomycin, cisplatin, and probenecid in the kidney. Our study provides leads for future work studying ACE2 expression modulators.


1st Editorial Decision
Thank you again for submit ting your work to Molecular Syst ems Biology. We have now heard back from the three referees who agreed to evaluat e your st udy. Overall, the reviewers think that despit e the modest concept ual novelt y of the approach, the present ed findings seem relevant given that analysing the effect s of drugs on ACE2 is a timely topic. They raise however a series of concerns, which we would ask you to address in a major revision.
As you will see below, the reviewers make const ruct ive suggest ions on how to improve the st udy. Wit hout repeat ing all the point s raised, some of the more fundament al issues are the following: -Reviewer #1 mentions that assessing the effect of the drugs on ACE2 in non-tumour bronchial epithelial cell lines seems like a better choice compared to cancer cell lines.
-Reviewers #2 and #3 point out that the second and more relevant part of the study, reporting an unbiased analysis of all approved compounds needs to be presented more thoroughly and the related conclusions need to be better supported. Reviewer #3 recommends including additional data (e.g. from different exposure times) for the drugs proposed to be promising for further analyses.
-The methodology needs to be described in better detail in order to be better accessible to the readers.
-Importantly, along the lines of the comment by reviewer #3, we would ask you to make the code available on GitHub and provide the link in the Data Availability section.
All other issues raised by the reviewers should be convincingly addressed. Please let me know in case you would like to discuss any of the issues raised. As this is a topic of active interest, we would ask you to make sure that the lat est lit erat ure is properly referenced in the manuscript . .

REFEREE REPORTS
Reviewer #1: The st udy on ACE2 expression by a wide variet y of clinically approved drugs is an import ant cont ribut ion to bet ter underst and the consequences such drugs might have for COVID-19 pat ient s. However, there are some part s of the st udy that need to be explained in more det ails. The use of 4 cancer cell lines to test the effect of drugs on ACE2 expression seems not the best approach, as cancer cells are inhibit ed by ACE2. It would have been bet ter to use some of the non-t umour bronchial epit helial cell lines, but may be the aut hors can provide a bit more informat ion why they had chosen these cells.
Overall, the st udy has merit s to be published aft er minor correct ions. Please provide more information on the cell type and of their origin. It is unclear if these 28 cell lines result from the different studies or if the authors were performing these assays. If the latter is true, than they should provide more details of the cells.
Reviewer #2: The authors present results from a quick and simple computational analysis of public available repositories of transcriptional data to the aim of identifying modulators of the expression of ACE2: a receptor required for SARS-CoV-2 infection in humans.
To this aim, the authors queried the connectivity map (cMap) database of transcriptional responses to drug treatment in immortalised human cancer cell lines, initially focusing on anti-hypertensive and anti-diabetic drugs, i.e. ACE inhibitors (ACEi) and angiotensin II typeI receptor blockers (ARBs). Consistently with previous findings (PMID: 20838579) the authors found that ACE inhibitors increases ACE2 expression but no ARBs was associated with differential expression of ACE2.
Furthermore the authors extended this analysis to all approved compounds with data available in the cMap and performed an enrichment analysis of drug Mode-of-action and therapeutic applications among the compounds exerting an effect on ACE2 expression.
Finally, the authors confirmed some of their hits using an independent dataset including drug transcriptional responses in vivo and attempted confirmatory a gene co-expression analysis using public data from normal lung tissue.
The recent COVID-19 outbreak and the need for findings and results that might help discovering/repositioning drugs able to reduce infections rate and symptoms make the subject of this report timely and important. However the analytical approach presented in this report is not novel, there is no new data presented nor final strong statements or guidelines for experimental followups, with just percentages of drugs with confirmatory in-vivo results reported, and not even stating the names of the most promising hits.
In addition, the first part of the paper focuses too much on confirmatory results and previously reported findings related to ant hypertensive and anti-diabetic drugs, whereas the more interesting unbiased and comprehensive analysis of all approved compounds is presented quite poorly.
Particularly the following points should be addressed, in my opinion: * The methods section should be generally extended, for example is not clear whether the differential ACE2 expression analysis for a given drug was performed considering each cell line response as a replicate or applying any sort of merging strategy to dilute cell line specific responses.
In addition a short discussion on this aspect might be included, i.e. how cell line from different tissues and with different somatic mutations respond differently at the transcriptional level to the same compound.
* at the beginning of the Results the authors claim 'Individually, no major anti-HT drug was found to increase ACE2 expression'. It is not clear what 'major' refers to. Does this refer to 'widely used' or 'widely prescribed' ? * The final co-expression analysis builds on the assumption that ACEIs down-regulate the expression of their targets. At what extent this is true? there are several examples of drugs upregulating the expression of the genes coding for their targeted proteins (PMID: 20838579). This should be discussed. * A description of how the MoA enrichment analysis, whose results are presented in figure 1BD is totally missing. How the individual drugs were aggregated into classes? additionally, how the GSEA was run in this case? intuitively, by sorting drugs based on their effect on ACE2 and then using drug classes as GSEA 'signatures' ? this must be necessarily detailed in the methods. * what is the sorting criteria for the classes of drugs in the x-axis labels of figure 1BD? wouldn't make more sense to increase readability via sorting them based on effect and magnitude of ACE differential expression? Reviewer #3: The paper "Systematic cell line-based identification of drugs modifying ACE2 expression" by Sinha et al." reports results from mining public databases for available in vitro and in vivo expression data of the ACE2 gene to identify the effects of clinically approved drugs. Panobinostat and isotretinoin are found to be the top ACE2 up-and down-regulators, respectively. The authors suggest that their results provide specific leads guiding further studies aimed at carefully assessing specific drug effects on ACE2 expression.
The key motivation for studying drug effects on ACE2 at the moment is that it the key host receptor for SARS-CoV-2 entry, and that ACE2 inhibitors are used to treat hypertension and diabetes, which are known to increase risk of a serious or fatal Covid-19 disease course. In fact, a recent study (https://doi.org/10.1016/j.cca.2020.03.031) provided evidence that ACE polymorphisms known to be associated with ACE2 expression also correlate with the number of cases per million inhabitants across 25 European countries. This work may be important, since so far "solid evidence concerning whether ACE2 expression can alter the risk of COVID-19 infection is lacking", as the authors say themselves. Even more so, the authors also acknowledge "the role of ACEIs/ARBs in modulating the clinical course of viral pneumonia and COVID-19 infection is also under debate [16][17][18], both being beyond the scope of this study".
The impact of this paper therefore hinges on these aspects to be studied and clarified, an effort that is surely taken up by many at the moment. I would therefore agree with the authors' claim that there is a "need to investigate the effect on modulating ACE2 expression of a variety of prescribed drugs." Amongst the cell lines available in CMAP the authors focus on epithelial cancer cell lines, arguing that they likely best resemble airway epithelial cells in the lung, which are implicated in the Covid-19 caused pneumonia. They also mined GEO and GTEx expression data from human lung tissue to check for concordance in the much smaller subset of 10 drugs for which there was in vivo gene expression data. The tissue or cell-type specific aspects of ACE2 expression are indeed important, and the argument to focus on epithelial cells makes sense to me. Yet, I found it surprising that the ACE inhibitors (ACEI) only as a group, but not individually, resulted in a significant ACE2 upregulation (while this is their known mechanism of action to reduce hypertension). I would therefore recommend checking if their effect is stronger in cells or cell-types directly relevant for the angiotensin induced vaso-constrictive action modulating blood pressure (such as those derived from vascular smooth muscle cells). This could give an idea to what extent the in vitro effects of ACE2 expression changes captured in CMAP cultures can be used as a model for in vivo effects.
The authors noted that none of the six drugs currently being investigated in clinical trials for treating COVID0-19 show any significant alteration of ACE2 expression. While interesting, this is probably expected for the antiviral drugs Lopinavir and Ritonavir, which act as protease inhibitors. Amongst these drugs Losartan is the only one known to act as ARB (which also were found as a drug class to not alter ACE2 expression).
Generally, the paper is well written and the analyses look sound to me. I do not think that the paper provides any methodological advancement; it just applies standard analyses to a gene that is of high relevance at the moment. A lot of the significant results apply to drug groups, with some notable exceptions, such as tomoxifan and crizotinib, and the three drugs mentioned in the abstract. While the drug group findings are interesting, it is only the latter that provide "leads for further in vitro and in vivo studies", so maybe more data should be shown for those drugs (e.g. different exposure times, see comment below).

Major comments:
Since this study focuses on ACE2 I wonder if there should not be a bit more information on this gene and its role both in regulating blood pressure and as viral entry point?
Also, it would be good to have more information on CMAP. For example, many experiments there are done using the L1000 high-throughput gene expression assay, which only asses close to 1000 genes directly and imputes expression estimates for ~11k other genes (with a reduction to 80% in found relationships). Is ACE2 directly assessed by L1000? If not, how well is it imputed?
Time measurements of drug exposure were taken after 24h. Patients on HT treatment are usually roughly at steady stage of exposure, so their ACE2 expression is not in response to exposure onset, but constant exposure. Therefore if measurements taken after longer exposure than 24h are available, this could be more relevant. I suggest checking that and if available confirm that the results are robust with respect to exposure time. (Also for some of the highlighted individual drugs it would be good to know if they are metabolized by the cells in the assay or basically continue acting at the initial concentrations.) Minor comments: It would be great if the authors could publish their analysis code along with the paper, as this would allow others to expand or modify their approach and apply it to different data. Also, it's just good practise for reproducible science. 1

Reviewer# 1:
The study on ACE2 expression by a wide variety of clinically approved drugs is an important contribution to better understand the consequences such drugs might have for COVID-19 patients. However, there are some parts of the study that need to be explained in more detail. The use of 4 cancer cell lines to test the effect of drugs on ACE2 expression seems not the best approach, as cancer cells are inhibited by ACE2. It would have been better to use some of the non-tumour bronchial epithelial cell lines, but may be the authors can provide a bit more information why they had chosen these cells.
Overall, the study has merits to be published after minor corrections.
We thank reviewer 1 for the feedback and constructive suggestions. In particular, we agree that the use of 4 cancer cell lines was not ideal. The choice was partly due to limited data availability and our initial aim to test all drugs on a shared, common subset of cell-lines, as explained below and in the revised main text (Page 3-4). Nevertheless, following up on the reviewer's suggestion, we have tried to the best of our capacity to expand our analysis to additional, potentially more relevant cell types. Please refer to our specific reply to your Comment #3 below.
1. The correspondence by Fang et al hypothesized that hypertension and diabetes develop severe COVID-19 because the two diseases have been reported to express ACE2 at an increased level. In addition, these patients are often treated with drugs that further increase ACE2. Please correct the statement.
We thank the reviewer for pointing out this error. The correction has been made (Pages 2): We thank the reviewer for pointing out this error in the references. We replaced the incorrect reference with both references as suggested.

A recent publication by Fang et al. has suggested that patients with hypertension (HT) and diabetes mellitus may be at higher risk of having severe COVID-19 disease (Fang et al, 2020), as these patients have been reported to express ACE2 at an increased level and are often treated with ACE inhibitors (ACEIs) or angiotensin II type-I receptor blockers (ARBs), which have been previously suggested to increase ACE2 expression
3. Why did the authors use four cancer cell lines to determine the effect of antihypertensive drugs on ACE2. It would have been advisable to use cell lines such as: BEAS2 or NuLi or NCI First, we agree with the reviewer that the use of four cancer cell lines was not ideal. In that part of our study, we relied on the analysis of the CMAP dataset, where most of its data is in cancer cell lines and no primary bronchial epithelial cells, including the ones mentioned by the reviewer, are available. Although two non-cancerous lung cells were listed in the CMAP document from https://clue.io/connectopedia/core_cmap_cell_panel, these two cell types are actually not included in their transcriptome profiling experiments. We focused on carcinoma cell lines due to their epithelial origin, and thus they may be more similar to the airway epithelium (likely one of the primary points of viral infection) compared to some of the other cell types (eg. leukemia, sarcoma or lymphoma cells) in CMAP. We have changed the main text to make this point clearer (Pages 3-4). The specific choice of these four carcinoma cell lines is again due to the practical issue of data availability and sample size --a reasonably large proportion of clinically approved drugs are tested on all of these four cell lines, whereas choosing other sets of cell lines or including more cell lines will result in a large variety and inhomogeneity in the identify of the cell-lines that each drug has been tested on, which may act as a confounding factor in the analysis. We have now also mentioned this point in the main text (Pages 3-4) and added an Appendix Note and figure (Appendix Note 1, Figure S1) showing the trade-off between the number of cell lines to include and the number of drugs tested.
The pertaining text on Pages 3-4 reads: Among the available cell types from CMAP, we focused on carcinoma cell lines, since they are of epithelial origin and may bear more resemblance to airway epithelium, a major site of viral entry. We identified 48 clinically approved anti-HT drugs that were tested on the same four carcinoma cell lines for 24 hours in CMAP, and computed the drug-induced ACE2 expression changes averaged across the cell lines (Methods; the cell lines are A549, MCF7, PC3 and VCAP, selected because of the data available for a high number of drugs tested on all these cells, see Appendix and Table EV1A for details).
Nevertheless, we have tried our best to expand our analysis to additional, potentially more relevant sample types. First, we analyzed the data in several normal or primary cells from CMAP, although they represent a much smaller proportion of data covering fewer drugs. These include the HA1E normal kidney cell line, the PHH primary liver cell, and the neural progenitor cell (NPC) differentiated from fibroblast-derived iPSC. These cell/tissue types can also be affected by SARS-CoV-2 (Zaim et al, 2020). We found that although these cells have diverse ACE2 expression responses to drugs, there are some consistent patterns from the drug class enrichment analysis; these analyses have now been added to Results (Page 6). The new text summarizing these results read as follows:  Figure EV1A; Table EV2E). Nevertheless, we observed a consistent but insignificant trend that ACEIs tend to upregulate ACE2 expression across the three normal cell types from kidney, liver, and CNS ( Figure EV1B). Concordant with the findings from the four carcinoma cell lines above, antineoplastic agents as a group were found to be enriched for drugs down-regulating ACE2 in the normal NPC cells from CNS (GSEA adjusted P=0.01, Figure EV1C, Table  EV2F).
Second, we expanded our previous search and analysis of datasets from the GEO database to include many more datasets on non-cancerous lung and kidney cells, including primary human bronchial epithelial BEAS-2B cells and in vivo lung/kidney samples from humans or rodents. We note that although these datasets cover more relevant cell/tissue types, the number of covered drugs was limited. From this analysis, we found additional significant ACE2 modulating drugs that we highlighted and discussed in an extended paragraph (Pages 7-8) and in the new Figure 2.
The new text and the revised Figure are enclosed below, for convenience and completeness: To further extend our analysis beyond the CMAP dataset, we mined the GEO database for gene expression data of drug treatments with matched controls in lung and kidney tissue or cells (Methods). We collected a total of 74 relevant lung datasets involving 42 unique clinically approved drugs, among which 27 datasets (covering 21 drugs) were composed of non-cancerous samples including primary bronchial epithelial cells and in vivo samples from human and rodents (Table  EV3A). Similarly, for kidney, 35 datasets for 29 drugs (including 23 drugs in 28 non-cancer datasets involving in vivo samples) were identified (Table EV3B). The drug-induced ACE2 differential expression results (Methods) for the lung and kidney datasets are summarized in Figure 2A and 2C, respectively. The results for the top significant drugs identified from the more relevant non-cancer datasets are highlighted in Figure 2B and 2D.

4
For lung, the top significant drug is dexamethasone, which upregulates ACE2 in datasets of both normal and Pneumocystis-infected mice lung tissue (logFC=0.97 and 0.36, adjusted P=0.001 and 0.027, respectively, Figure 2B). Consistently, dexamethasone also increased ACE2 expression in our analysis of four carcinoma cell lines from the CMAP dataset (logFC=0.18, P=0.006, Table EV1F). Interestingly, corticosteroids, including dexamethasone, have been widely used for severe acute respiratory syndrome (SARS) but showed no survival benefit and possible harm (Russell et al, 2020), and WHO does not recommend the routine use of corticosteroids for COVID-19 patients (World Health Organization, 2020). Another top identified drug is the epidermal growth factor receptor (EGFR) inhibitor erlotinib, which is found to upregulate ACE2 in a dataset of human primary bronchial epithelial cells (logFC=1.04, adjusted P=2.95E-5, Figure 2B), a relevant cell type suggested to interact with the SARS-CoV-2 virus (Mason et al, 2020). In the CMAP analysis, we observed a non-significant trend of ACE2 upregulation at 24 hours by erlotinib (logFC=0.05, adjusted P>0.1, Table EV1F). Interestingly, erlotinib has actually been reported to inhibit the endocytosis and intracellular trafficking of multiple viruses including hepatitis C, dengue and Ebola, exerting broad-spectrum antiviral effects (Bekerman et al, 2017). The chemotherapeutic drug bleomycin is a significant ACE2 down-regulator identified in a dataset of rat lung tissue (logFC=-0.17, adjusted P=0.003, Figure 2B), in accordance with an earlier reports that it decreases ACE2 protein level in alveolar epithelial cells (Uhal et al, 2010).
Among the top significant candidates arising from the kidney cells analysis (summarized in Figure 2C), we again focused on non-cancer datasets and observed that the chemotherapy drug cisplatin up-regulated ACE2 in mice kidney samples while it down-regulated ACE2 in the renal cortex of rat (logFC=0.29 and -1.16, adjusted P=8.06E-3 and 2.36E-5, respectively, Figure 2D), suggesting a cell type and possibly species specific effect. Vancomycin, another top identified drug, is a glycopeptide antibiotic that increases ACE2 expression in mice kidney samples from two independent datasets (logFC=0.89 and 0.93, adjusted P=0.04 for both). Glycopeptide antibiotics and its derivatives have been previously shown to block MERS and SARS cell entry (Zhou et al, 2016). Probenecid, a drug for treating gout, was found to decrease ACE2 expression in a renal cortical cell line (logFC=-0.61, adjusted P=0.001, Figure 2D); this drug has been proposed to be repurposed for anti-influenza therapy (Perwitasari et al, 2013). Other top significant drugs arising from the analysis of cancer datasets of lung and kidney from GEO are shown in Figure EV3 with additional information in Appendix Note 2. 4. On page 6 the authors state that the drugs effect on ACE2 expression was verified 28 cell lines. Please provide more information on the cell type and of their origin. It is unclear if these 28 cell lines result from the different studies or if the authors were performing these assays. If the latter is true, than they should provide more details of the cells.
The data for the 28 cell types analyzed here are also from the CMAP dataset. In the analyses from the previous parts of our study where we used only 4 carcinoma cell lines, we required that all drugs be tested on all included cell lines. In the new analysis, which is now on Page 5, we extended the total number of cell types by allowing each drug to be tested on a different set of cells. We added explanations in the Methods section under the subsection "The CMAP data" (Pages 9-10). The details of these 28 cell lines are given in Table EV2A. The text added reads as follows:

Reviewer #2:
The authors present results from a quick and simple computational analysis of public available repositories of transcriptional data to the aim of identifying modulators of the expression of ACE2: a receptor required for SARS-CoV-2 infection in humans.
To this aim, the authors queried the connectivity map (cMap) database of transcriptional responses to drug treatment in immortalised human cancer cell lines, initially focusing on antihypertensive and anti-diabetic drugs, i.e. ACE inhibitors (ACEi) and angiotensin II typeI receptor blockers (ARBs). Consistently with previous findings (PMID: 20838579) the authors found that ACE inhibitors increases ACE2 expression but no ARBs was associated with differential expression of ACE2.
Furthermore the authors extended this analysis to all approved compounds with data available in the cMap and performed an enrichment analysis of drug Mode-of-action and therapeutic applications among the compounds exerting an effect on ACE2 expression.
Finally, the authors confirmed some of their hits using an independent dataset including drug transcriptional responses in vivo and attempted confirmatory a gene co-expression analysis using public data from normal lung tissue.
The recent COVID-19 outbreak and the need for findings and results that might help discovering/repositioning drugs able to reduce infections rate and symptoms make the subject of this report timely and important. However the analytical approach presented in this report is not novel, there is no new data presented nor final strong statements or guidelines for experimental followups, with just percentages of drugs with confirmatory in-vivo results reported, and not even stating the names of the most promising hits.
In addition, the first part of the paper focuses too much on confirmatory results and previously reported findings related to ant hypertensive and anti-diabetic drugs, whereas the more interesting unbiased and comprehensive analysis of all approved compounds is presented quite poorly.
We thank the reviewer for the helpful feedback and constructive criticism of our manuscript. Accordingly we have revised the parts of our manuscript on the analysis of all clinically approved drugs by clarifying the methods used (addressing the reviewer's Comment #1 and #4) and included clear descriptions of the top drugs identified. A panel was added to Figure 1 (current Figure 1D) that explicitly shows the control vs drug-treated ACE2 expression for the top significant drugs across all clinically approved drugs. We have also expanded our analysis of datasets from the GEO database, identifying additional clinically approved drugs that significantly alter ACE2 expression in relevant cell/tissue types. We have summarized these results in a new main text Figure 2 and added the corresponding description in the Results (Page 9-10, addressing the Comment #5). Please see the reply to specific comments below.
Particularly the following points should be addressed, in my opinion: 1. The methods section should be generally extended, for example is not clear whether the differential ACE2 expression analysis for a given drug was performed considering each cell line response as a replicate or applying any sort of merging strategy to dilute cell line specific responses. In addition a short discussion on this aspect might be included, i.e. how cell line from different tissues and with different somatic mutations respond differently at the transcriptional level to the same compound.
We have comprehensively expanded the Methods to include separate subsections describing the CMAP data, the differential expression analysis and the drug class enrichment analysis. In particular, we have clarified the details on how the differential expression analysis was performed for multiple cell lines. In the analysis of the four carcinoma cell lines from CMAP, we controlled for cell line identity as a covariate in the limma linear model, and therefore the results represent averaged differential expression across the four cell lines. We have made this point clearer both in the Methods (subsection titled "Identification of ACE2 modulators from the CMAP dataset") and Results (Page 4); the text on Page 4 now reads: Table EV1A for details).

We identified 48 clinically approved anti-HT drugs that were tested on the same four carcinoma cell lines for 24 hours in CMAP, and computed the drug-induced ACE2 expression changes averaged across the cell lines (Methods; the cell lines are A549, MCF7, PC3 and VCAP, selected because of the data available for a high number of drugs tested on all these cells, see Appendix and
Indeed cell lines with different cells of origin or genomic background may respond differently (PMID: 26824188, 28071740, 31990955, https://doi.org/10.1101/868752). We have extended our analysis of ACE2 expression changes in response to drugs to several normal cell types of several tissues of origin that may be affected by COVID-19 (including lung, kidney, liver, the central nervous system, and intestine, each tissue type analyzed separately). In general, we find that the concordance of ACE2 differential expression profiles across cell types are low ( Figure  EV1A), however, some similarities among the enriched drug classes could be observed ( Figure  EV1B,C). These results are now described in the Results (Page 6) and read as follows: The various analyses described above were performed by aggregating the druginduced expression changes across cell types (as explained above; Methods). We next analyzed CMAP data of additional relevant cell types separately by their tissue of origin to investigate the potential tissue-specific effects. We focused on the lung, kidney, liver, central nervous system (CNS), and intestine (Methods), which represent tissues that can be affected by SARS- CoV-2 (Zaim et al, 2020). For each of these tissues, we were only able to find one (or two, for lung) cell types where a reasonable number (>100) of clinically approved drugs were tested (at the 24 hours time point; details in Table EV2D). However, the cells identified from kidney, liver and CNS were non-cancerous or primary cells (HA1E, PHH, and NPC cells, respectively), which can be more relevant for our investigation. As expected, the drug-induced ACE2 expression changes exhibit mostly weak correlations across cells from different tissue types, with Spearman's correlation coefficients between the log fold-changes of pairs of cells ranging from -0.07 to 0.2 ( Figure EV1A; Table EV2E). Nevertheless, we observed a consistent but insignificant trend that ACEIs tend to upregulate ACE2 expression across the three normal cell types from kidney, liver, and CNS ( Figure EV1B). Concordant with the findings from the four carcinoma cell lines above, antineoplastic agents as a group were found to be enriched for drugs down-regulating ACE2 in the normal NPC cells from CNS (GSEA adjusted P=0.01, Figure EV1C, Table  EV2F).
We also added a brief summary in the Discussion section (Page 13): Analyzing the additional but limited amount of CMAP data for cells from different tissue types and for drug treatment for longer durations up to 48 hours, we find that the drug-induced ACE2 changes can be tissue-specific and time-dependent ( Figure EV1,2), and for some drugs, their effect on ACE2 can become stronger upon prolonged treatment ( Figure EV2).
2. at the beginning of the Results the authors claim 'Individually, no major anti-HT drug was found to increase ACE2 expression'. It is not clear what 'major' refers to. Does this refer to 'widely used' or 'widely prescribed' ?
Yes this is meant to be "widely prescribed", including some members of the ACEI, ARB, and calcium channel blockers. The text has been modified considering this comment (Page 4): "Individually, no widely prescribed anti-HT drug was found to increase ACE2 expression significantly in these experiments, except for methyldopa (an alpha-2 adrenergic receptor agonist) and molsidomine (a vasodilator) do significantly decrease ACE2 expression" 3. The final co-expression analysis builds on the assumption that ACEIs down-regulate the expression of their targets. At what extent this is true? there are several examples of drugs upregulating the expression of the genes coding for their targeted proteins (PMID: 20838579). This should be discussed.
One underlying assumption of this analysis is actually that the functional effects induced by a drug (which usually acts via inhibiting its target protein) tend to be similar to the effects of down-regulating the drug's target on the mRNA level. We note that this is different from the assumption that the drugs will downregulate the expression of their targets, the latter being a stronger assumption that is indeed less likely to be true in general and thank the reviewer for noting that. Nevertheless, we still think this additional analysis is valuable and chose to maintain it, but have added a clarifying sentence to this extent on (Page 12), which reads: This analysis is based on the notion that the effect of drug treatment, which usually acts on the protein level, is in general functionally similar to the effect of downregulating the drug target expression on the mRNA level.

4.
A description of how the MoA enrichment analysis, whose results are presented in figure 1BD is totally missing. How the individual drugs were aggregated into classes? additionally, how the GSEA was run in this case? intuitively, by sorting drugs based on their effect on ACE2 and then using drug classes as GSEA 'signatures' ? this must be necessarily detailed in the methods.
We thank the reviewer for pointing this out. We sincerely apologize for the missing information, which we have of course rectified now. We have expanded the description of the GSEA analysis into a separate section titled "Analysis of drug classes enrichment in ACE2 modulators" in the Methods section (Page 15). Basically, the understanding of the reviewer is correct --all the 672 clinically approved drugs were sorted by their induced ACE2 expression log fold-changes, then GSEA was applied by using the classes of drugs by MOA as "gene sets" (here actually "drug sets"). The annotation on classes of drugs by MOA was obtained from the Drug Repurposing Hub (Corsello et al, 2017). The corresponding new text now reads: Thanks. First, we updated the curation and analysis of GEO datasets to include more in vivo datasets and relevant in vitro datasets such as those in primary airway epithelial cells. In addition to the lung, we also collected data from the kidney, another tissue site that can be affected by COVID-19. Second, following the reviewer's suggestion we have added a new figure (Figure 2) summarizing the results from analyzing these datasets, where the top identified drugs that can alter ACE2 expression were also shown. We have updated the description of this analysis and included an extensive description of the top drugs in the Results (Page 7-8).
The new text is enclosed below, including the new Figure, as follows:

11
To further extend our analysis beyond the CMAP dataset, we mined the GEO database for gene expression data of drug treatments with matched controls in lung and kidney tissue or cells (Methods). We collected a total of 74 relevant lung datasets involving 42 unique clinically approved drugs, among which 27 datasets (covering 21 drugs) were composed of non-cancerous samples including primary bronchial epithelial cells and in vivo samples from human and rodents (Table  EV3A). Similarly, for kidney, 35 datasets for 29 drugs (including 23 drugs in 28 non-cancer datasets involving in vivo samples) were identified (Table EV3B). The drug-induced ACE2 differential expression results (Methods) for the lung and kidney datasets are summarized in Figure 2A and 2C, respectively. The results for the top significant drugs identified from the more relevant non-cancer datasets are highlighted in Figure 2B and 2D.
For lung, the top significant drug is dexamethasone, which upregulates ACE2 in datasets of both normal and Pneumocystis-infected mice lung tissue (logFC=0.97 and 0.36, adjusted P=0.001 and 0.027, respectively, Figure 2B). Consistently, dexamethasone also increased ACE2 expression in our analysis of four carcinoma cell lines from the CMAP dataset (logFC=0.18, P=0.006, Table EV1F). Interestingly, corticosteroids, including dexamethasone, have been widely used for severe acute respiratory syndrome (SARS) but showed no survival benefit and possible harm (Russell et al, 2020), and WHO does not recommend the routine use of corticosteroids for COVID-19 patients (World Health Organization, 2020). Another top identified drug is the epidermal growth factor receptor (EGFR) inhibitor erlotinib, which is found to upregulate ACE2 in a dataset of human primary bronchial epithelial cells (logFC=1.04, adjusted P=2.95E-5, Figure 2B), a relevant cell type suggested to interact with the SARS-CoV-2 virus (Mason et al, 2020). In the CMAP analysis, we observed a non-significant trend of ACE2 upregulation at 24 hours by erlotinib (logFC=0.05, adjusted P>0.1, Table EV1F). Interestingly, erlotinib has actually been reported to inhibit the endocytosis and intracellular trafficking of multiple viruses including hepatitis C, dengue and Ebola, exerting broad-spectrum antiviral effects (Bekerman et al, 2017). The chemotherapeutic drug bleomycin is a significant ACE2 down-regulator identified in a dataset of rat lung tissue (logFC=-0.17, adjusted P=0.003, Figure 2B), in accordance with an earlier reports that it decreases ACE2 protein level in alveolar epithelial cells (Uhal et al, 2010).
Among the top significant candidates arising from the kidney cells analysis (summarized in Figure 2C), we again focused on non-cancer datasets and observed that the chemotherapy drug cisplatin up-regulated ACE2 in mice kidney samples while it down-regulated ACE2 in the renal cortex of rat (logFC=0.29 and -1.16, adjusted P=8.06E-3 and 2.36E-5, respectively, Figure 2D), suggesting a cell type and possibly species specific effect. Vancomycin, another top identified drug, is a glycopeptide antibiotic that increases ACE2 expression in mice kidney samples from two independent datasets (logFC=0.89 and 0.93, adjusted P=0.04 for both). Glycopeptide antibiotics and its derivatives have been previously shown to block MERS and SARS cell entry (Zhou et al, 2016). Probenecid, a drug for treating gout, was found to decrease ACE2 expression in a renal cortical cell line (logFC=-0.61, adjusted P=0.001, Figure 2D); this drug has been proposed to be repurposed for anti-influenza therapy (Perwitasari et al, 2013). Other top significant drugs arising from the analysis of cancer datasets of lung and kidney from GEO are shown in Figure EV3 with additional information in Appendix Note 2. 7. what is the sorting criteria for the classes of drugs in the x-axis labels of figure 1BD? wouldn't make more sense to increase readability via sorting them based on effect and magnitude of ACE differential expression?
Thanks. The previous sorting was performed alphabetically. We agree with the reviewer that sorting based on enrichment significance may improve the interpretability and thus have modified Figure 1 accordingly. We preferred to sort by P-value rather than by effect size. (Reviewer 3, minor comment 4 also recommended sorting by P-values)

Reviewer #3:
The paper "Systematic cell line-based identification of drugs modifying ACE2 expression" by Sinha et al." reports results from mining public databases for available in vitro and in vivo expression data of the ACE2 gene to identify the effects of clinically approved drugs. Panobinostat and isotretinoin are found to be the top ACE2 up-and down-regulators, respectively. The authors suggest that their results provide specific leads guiding further studies aimed at carefully assessing specific drug effects on ACE2 expression.
The key motivation for studying drug effects on ACE2 at the moment is that it the key host receptor for SARS-CoV-2 entry, and that ACE2 inhibitors are used to treat hypertension and diabetes, which are known to increase risk of a serious or fatal Covid-19 disease course. In fact, a recent study (https://doi.org/10.1016/j.cca.2020.03.031) provided evidence that ACE polymorphisms known to be associated with ACE2 expression also correlate with the number of cases per million inhabitants across 25 European countries. This work may be important, since so far "solid evidence concerning whether ACE2 expression can alter the risk of COVID-19 infection is lacking", as the authors say themselves. Even more so, the authors also acknowledge "the role of ACEIs/ARBs in modulating the clinical course of viral pneumonia and COVID-19 infection is also under debate [16][17][18], both being beyond the scope of this study".
The impact of this paper therefore hinges on these aspects to be studied and clarified, an effort that is surely taken up by many at the moment. I would therefore agree with the authors' claim that there is a "need to investigate the effect on modulating ACE2 expression of a variety of prescribed drugs." Amongst the cell lines available in CMAP the authors focus on epithelial cancer cell lines, arguing that they likely best resemble airway epithelial cells in the lung, which are implicated in the Covid-19 caused pneumonia. They also mined GEO and GTEx expression data from human lung tissue to check for concordance in the much smaller subset of 10 drugs for which there was in vivo gene expression data. The tissue or cell-type specific aspects of ACE2 expression are indeed important, and the argument to focus on epithelial cells makes sense to me. Yet, I found it surprising that the ACE inhibitors (ACEI) only as a group, but not individually, resulted in a significant ACE2 up-regulation (while this is their known mechanism of action to reduce hypertension). I would therefore recommend checking if their effect is stronger in cells or celltypes directly relevant for the angiotensin induced vaso-constrictive action modulating blood pressure (such as those derived from vascular smooth muscle cells). This could give an idea to what extent the in vitro effects of ACE2 expression changes captured in CMAP cultures can be used as a model for in vivo effects.
Thank you. We agree with the reviewer that investigating the drug-induced effects on ACE2 in additional relevant cell types can prove informative. We note that ACEIs act via the inhibition of the angiotensin-converting enzyme (ACE) rather than angiotensin-converting enzyme 2 (ACE2), which are distinct enzymes within the same pathway but with different functions. ACEIs are not necessarily expected to inhibit (or increase) ACE2 expression, and actually the effects of ACEIs on the ACE2 gene expression are largely not well characterized. We have included more background information on the renin-angiotensin pathway, ACE2 and ACEIs in the Introduction based on the reviewer's Comment #1 below (Page 3, see Comment #1 below).
Although overall there are only limited drug-induced gene expression data available, we have tried to the best of our capacity to expand our analysis to additional relevant cell types, i.e. those that represent tissue sites that can be affected by COVID-19. First, we analyzed the data in several normal or primary cells from CMAP, although they represent a much smaller proportion of data covering fewer drugs (Page 6). These include the HA1E normal kidney cell line, the PHH primary liver cell, and the neural progenitor cell (NPC) differentiated from fibroblastderived iPSC. These cell/tissue types also represent those that can be affected by SARS-CoV-2. Second, we expanded our previous search and analysis of datasets from the GEO database to include many more datasets on non-cancerous lung and kidney samples, including primary human bronchial epithelial BEAS-2B cells and in vivo lung/kidney samples from human or rodents. Of note, kidney is also known to be a target organ of ACEIs; we found two in vivo kidney datasets for the ACEIs captopril and enalapril, although in both cases the change in ACE2 is not significant after FDR correction (Table EV3B). The corresponding descriptions in the Results have been updated (Page 7-8, see below). The paragraph added on Page 6 about the analysis of additional cell line in CMAP is attached here: The various analyses described above were performed by aggregating the druginduced expression changes across cell types (as explained above; Methods). We next analyzed CMAP data of additional relevant cell types separately by their tissue of origin to investigate the potential tissue-specific effects. We focused on the lung, kidney, liver, central nervous system (CNS), and intestine (Methods), which represent tissues that can be affected by SARS-CoV-2 (Zaim et al, 2020). For each of these tissues, we were only able to find one (or two, for lung) cell types where a reasonable number (>100) of clinically approved drugs were tested (at the 24 hours time point; details in Table EV2D). However, the cells identified from kidney, liver and CNS were non-cancerous or primary cells (HA1E, PHH, and NPC cells, respectively), which can be more relevant for our investigation. As expected, the drug-induced ACE2 expression changes exhibit mostly weak correlations across cells from different tissue types, with Spearman's correlation coefficients between the log fold-changes of pairs of cells ranging from -0.07 to 0.2 ( Figure EV1A; Table EV2E). Nevertheless, we observed a consistent but insignificant trend that ACEIs tend to upregulate ACE2 expression across the three normal cell types from kidney, liver, and CNS ( Figure EV1B). Concordant with the findings from the four carcinoma cell lines above, antineoplastic agents as a group were found to be enriched for drugs down-regulating ACE2 in the normal NPC cells from CNS (GSEA adjusted P=0.01, Figure EV1C, Table  EV2F).
The authors noted that none of the six drugs currently being investigated in clinical trials for treating COVID0-19 show any significant alteration of ACE2 expression. While interesting, this is probably expected for the antiviral drugs Lopinavir and Ritonavir, which act as protease inhibitors. Amongst these drugs Losartan is the only one known to act as ARB (which also were found as a drug class to not alter ACE2 expression).
Generally, the paper is well written and the analyses look sound to me. I do not think that the paper provides any methodological advancement; it just applies standard analyses to a gene that is of high relevance at the moment. A lot of the significant results apply to drug groups, with some notable exceptions, such as tomoxifan and crizotinib, and the three drugs mentioned in the abstract. While the drug group findings are interesting, it is only the latter that provide "leads for further in vitro and in vivo studies", so maybe more data should be shown for those drugs (e.g. different exposure times, see comment below).
We thank the reviewer for the constructive comments and overall agree with all the points made here. Indeed, many current drugs under investigation for COVID-19 target different viral proteins and are not expected to alter ACE2 directly; we examined these drugs and confirmed that they do not significantly modulate ACE2 expression. On the other hand, we agree that we should provide more detailed results on the top individual ACE2 modulators identified among all clinically approved drugs. We hence have updated the curation and analysis of GEO datasets to include more in vivo and also relevant in vitro datasets in both lung and kidney (the latter another tissue site that can be affected by COVID-19). Correspondingly we have added a new figure (Figure 2) showing the top significant ACE2-modulating drugs identified from these analyses. The description of this analysis and the top candidate drugs arising was also updated in the Results (Page 7-8). The pertaining new text and Figure are enclosed below, for convenience and completeness: To further extend our analysis beyond the CMAP dataset, we mined the GEO database for gene expression data of drug treatments with matched controls in lung and kidney tissue or cells (Methods). We collected a total of 74 relevant lung datasets involving 42 unique clinically approved drugs, among which 27 datasets (covering 21 drugs) were composed of non-cancerous samples including primary bronchial epithelial cells and in vivo samples from human and rodents (Table  EV3A). Similarly, for kidney, 35 datasets for 29 drugs (including 23 drugs in 28 non-cancer datasets involving in vivo samples) were identified (Table EV3B). The drug-induced ACE2 differential expression results (Methods) for the lung and kidney datasets are summarized in Figure 2A and 2C, respectively. The results for the top significant drugs identified from the more relevant non-cancer datasets are highlighted in Figure 2B and 2D.
For lung, the top significant drug is dexamethasone, which upregulates ACE2 in datasets of both normal and Pneumocystis-infected mice lung tissue (logFC=0.97 and 0.36, adjusted P=0.001 and 0.027, respectively, Figure 2B). Consistently, dexamethasone also increased ACE2 expression in our analysis of four carcinoma cell lines from the CMAP dataset (logFC=0.18, P=0.006, Table EV1F). Interestingly, corticosteroids, including dexamethasone, have been widely used for severe acute respiratory syndrome (SARS) but showed no survival benefit and possible harm (Russell et al, 2020), and WHO does not recommend the routine use of corticosteroids for COVID-19 patients (World Health Organization, 2020). Another top identified drug is the epidermal growth factor receptor (EGFR) inhibitor erlotinib, which is found to upregulate ACE2 in a dataset of human primary bronchial epithelial cells (logFC=1.04, adjusted P=2.95E-5, Figure 2B), a relevant cell type suggested to interact with the SARS-CoV-2 virus (Mason et al, 2020). In the CMAP analysis, we observed a non-significant trend of ACE2 upregulation at 24 hours by erlotinib (logFC=0.05, adjusted P>0.1, Table EV1F). Interestingly, erlotinib has actually been reported to inhibit the endocytosis and intracellular trafficking of multiple viruses including hepatitis C, dengue and Ebola, exerting broad-spectrum antiviral effects (Bekerman et al, 2017). The chemotherapeutic drug bleomycin is a significant ACE2 down-regulator identified in a dataset of rat lung tissue (logFC=-0.17, adjusted P=0.003, Figure 2B), in accordance with an earlier reports that it decreases ACE2 protein level in alveolar epithelial cells (Uhal et al, 2010).
Among the top significant candidates arising from the kidney cells analysis (summarized in Figure 2C), we again focused on non-cancer datasets and observed that the chemotherapy drug cisplatin up-regulated ACE2 in mice kidney samples while it down-regulated ACE2 in the renal cortex of rat (logFC=0.29 and -1.16, adjusted P=8.06E-3 and 2.36E-5, respectively, Figure 2D), suggesting a cell type and possibly species specific effect. Vancomycin, another top identified drug, is a glycopeptide antibiotic that increases ACE2 expression in mice kidney samples from two independent datasets (logFC=0.89 and 0.93, adjusted P=0.04 for both). Glycopeptide antibiotics and its derivatives have been previously shown to block MERS and SARS cell entry (Zhou et al, 2016). Probenecid, a drug for treating gout, was found to decrease ACE2 expression in a renal cortical cell line (logFC=-0.61, adjusted P=0.001, Figure 2D); this drug has been proposed to be repurposed for anti-influenza therapy (Perwitasari et al, 2013). Other top significant drugs arising from the analysis of cancer datasets of lung and kidney from GEO are shown in Figure EV3 with additional information in Appendix Note 2.