Table of Contents
Establishing a quantitative screening platform to identify drug effects on embryonic and cardiovascular development
One of the earliest initiatives to promote research on drug repurposing to fight COVID-19 was supported by the Medicine For Malaria Venture (MMV). The consortium assembled and freely distributed the Covid Box for research, comprising a drug library of 160 compounds with known or predicted effects on one or multiple steps of SARS-CoV-2 infection or COVID-19 disease progression. The majority of compounds in the Covid Box are categorized as anti-infective agents, that would target biological processes of pathogens (56 agents; 35% of the library), all the other types would target biological processes in the host (Fig. 1a,b). As an extension of our screening efforts, we also purchased Molnupiravir and Sabizabulin (from MedChemExpress), which are approved agents for the treatment of COVID-19 that successfully inhibit viral replication and transmission from cell to cell, respectively19,20.
Overview of drugs included in zebrafish cardiovascular and behavioral screening setup. (a) Information on type of candidate drugs used. All primary drug indications are listed and a set of anti-infective agents is shown. (b) A representative table of drugs that are FDA-approved or in III Clinical trial with the indicated Cmax blood concentrations and risk during pregnancy.
We treated zebrafish embryos at 28 hpf with the compounds dissolved in DMSO (Fig. 2a). The baseline concentration for the screening of the drugs was 1 µM, as recommended by the MMV, which was approximately 10 × lower than used commonly in zebrafish screens, e.g.21,22,23. For those compounds also used in the clinical context for COVID-19 treatment we used compounds at Cmax (Fig. 1b). Treatment with each of 162 compounds on 10 zebrafish embryos was replicated two times or more. For Ivermectin and Sabizabulin, as 1 µM concentration elicited severe toxic effects, we lowered the dose. Treated embryos were then subjected to two parallel imaging platforms, that allowed us to observe embryonic and cardiovascular development on the one hand, and behavior on the other. We analyzed the data in a combined workflow in semi-automatic as well as automatic manner to obtain quantitative data for each drug (Fig. 2).
Schematic representation of the experimental workflow. (a) Tg(fli1:EGFP);(myl7:mRFP) embryos were used for the morphological assay (green: cytoplasmic GFP in endothelium; magenta: membrane RFP in cardiomyocytes). After drug treatment, between 96 and 104 hpf, the anesthetized larvae were transferred to 96 well plates with pre-formed agarose beds by a 3D printed mold. This is followed by image acquisition. Treated larvae positioned in a 96 well plate are imaged using a Smart Imaging Wide field Fluorescent Microscope Platform. (b) Image analysis. (1) Image analysis through a semiautomatic and blinded workflow based on a Fiji-macro. (2) To count the intersegmental vessels (ISV) the user draws a line (red line) along all ISVs. A kymograph along the Z-axis allows detection of local maxima. (3) The larval length from anterior to posterior (AP) is measured by drawing a line from the most anterior of the head to the most posterior of the tail. (4) Effusions are identified as shown in the examples. (5) For heartbeat measurements, hearts are detected with an automated threshold to generate a mask and a kymograph along the time axis detects the local maxima. For the deep learning-based analysis, image data of the ISV and the heart are manually annotated. The annotated data are used to train two different U-Net models. The ISV-Seg model is applied to segment each ISV and calculate the area. The Heart-Seg model is used to segment ventricle and atrium. The minima and maxima of the ventricular area along the time axis are used to calculate the ejection fraction. For the behavioral assay, wild type larvae were treated with the drugs and transferred to 96 well plates. The swimming of larvae with a defined light–dark exposure was recorded and the tracking data exported. (c) The results are collected and used for meta-analysis with Python. The results were presented as heat maps and in a customized online Streamlit data app.
To observe morphological and cardiovascular features, we used the homozygous double transgenic line Tg(fli1:EGFP)y1;(myl7:mRFP)ko08 in which the fli1a promotor controls expression of the enhanced green fluorescent protein encoding gene (EGFP) and the myl7 promoter drives expression of the membrane-tagged red fluorescent protein encoding gene (mRFP)24,25. At 4 days post fertilization (dpf), larvae were transferred to 96-well plates, prepared with beds of low-melting agarose using a 3D-printed mold and imaged using a High-content Smart Imaging Fluorescent Microscope. This was followed by an image analysis pipeline (Fig. 2b) and metaanalysis and visualization (Fig. 2c).
On the other hand, we evaluated swimming behavior, an established method for testing neuroactive compounds (Basnet et al.), using DanioVision™ recording chamber. Tracking the behavior of zebrafish larvae is based on switching between bright and dark phases and can reveal anxiety, vision impairment, muscular weakness and reactivity18 (Fig. 2, Supplementary Fig. S1). These data were also integrated into the metaanalysis and web-based visualization platform.
A third of tested compounds altered cardiovascular development, embryonic growth, or behavior in the zebrafish model
First, we evaluated the survival and pericardial effusion of the embryos during treatment for both assays (Fig. 3a and Supplementary Fig. S2). The mortality rate (MR) was calculated as the average percentage of the larvae (n = 20–60 larvae per drug) that died before imaging in the respective assay. We observed that 7 out of 162 drugs (4.3%) caused a MR over 10% at 1 µM, including Apilimod, Astemizole, Ivermectin, Manidipine, Midostaurin, Niclosamide and Pimozide. Moreover, 18 drugs (about 12.5%) caused pericardial effusion in a quarter or more of the treated larvae. Among them, Astemizole (94.4%), Pimozide (100%) and Ponatinib (100%) were the compounds most frequently leading to pericardial effusion.
Summary of results obtained for the compounds screening. (a) Bar plot representing the percentage of mortality and pericardial effusion of embryos tested in morphological assay. (b) Bar plot representing percentage of compounds leading to statistically significant alterations in the analyzed parameters. (c) Heat map of the results obtained for the tested compounds. Compound names are listed on the right of the heat map and are allocated to groups: ΔHeart, ΔBody length, ΔVasculature, ΔActivity or “No significant effect”, depending on the results. E.g. a compound that led to high reduction of ejection fraction compared to controls would appear in the group ΔHeart. The formula used to calculate the scores in the heatmap is given at the bottom. The p-value is derived from the Mann–Whitney U test treatment vs. control, the lower the p-value the higher the –log10(p-value). The hedge’s effect size gives an estimation of how strong an effect is. The fold change (FC) shows whether the values are higher or lower than the control and how many times compared to the control. The scores are shown in a relative color scale: green, most positive score; magenta, most negative score; grey, no significance using a Mann–Whitney U test. Shown are the mean values from at least two technical replicates, each replicate is composed of at least 10 embryos per condition. o symbol indicates novel statistically significant observations in behavioral assay. All the compounds are used at 1 μM except when it is mentioned otherwise. Asterisk in Ivermectin refers to the fact that it was used at 1 μM for the morphological assay and at 0.5 μM for the behavioral assay. Compounds mentioned throughout the manuscript are highlighted in bold.
We measured the body length and several parameters to assess cardiovascular development and behavior of larvae and clustered them according to their effect on body length, heart function (rate and ejection fraction), vasculature formation (number and area of ISV) and effect on activity (moving during accommodation, velocity under bright light and dark conditions as well as the ratio of velocity between dark and bright phases). ∼20% (33/162) of the drugs significantly altered the size of the larvae, 56 drugs (34.5%) altered the heart rate and 33 drugs (∼20%) affected the ejection fraction (Fig. 3b). For assessment of behavior in the presence of the compounds, we scored three phases of the bright/dark locomotion test—accommodation, bright and dark phases—and calculated the ratio between swimming velocity in the dark and bright. 27 drugs affected velocity in the dark, 30 in the light phase and 27 drug treatments showed alterations in activity during accommodation. To understand whether the difference between activity in the bright and dark phases was proportional to the control, we calculated the logarithmic bright/dark ratio. The bright/dark ratio was altered in 10.5% (17/162) of the drug treatments.
We clustered compounds according to the parameter affects and also generated a final cluster by those compounds that at the tested concentrations did not produce any deviation in the analyzed parameters when compared to controls (Fig. 3c).
We looked closer to those compounds leading to more pronounced defects on development (Figs. 4 and 5). Astemizole, Pimozide, Ponatinib, and Ivermectin led to the highest reduction in body length. While these compounds had been studied in the zebrafish27,28,29,30, size reduction had not been documented. Furthermore, 17α-Hydroxyprogesterone caused a previously not described larval overgrowth. Some of the observed effects on heart rate and ejection fraction had previously been reported, as was the case for Manidipine and Astemizole31,32. However, neither the increase in ejection fraction by Tacrolimus nor the effects of Ravuconazole (an antifungal agent), GSK-369796 (an antimalarial compound), Amuvatinib and Regorafenib have to our knowledge not been reported before. The Tyrosine kinase inhibitors Regorafenib, Cabozantinib, and Sorafenib revealed a strong inhibition of ISV formation, as reported before, and thus served as a positive control for the experimental set up33,34,35. However, the negative effects on ISV area by the compounds Astemizole and Pimozide had not been reported before.
Combined boxplots and swarmplots showing the morphology and activity measurements for each larva. (a) Results from treatments using drugs studied in the context of Covid-19 research. (b) Results from treatments leading to strongest phenotypic alterations. (c) Results from treatments using COVID-19 drugs used at Cmax. The asterisks represent the p-values from the Mann–Whitney U test * < 0.05; ** < 0.01;*** < 0.001. Results from control larvae highlighted with red fill color. The red line indicates the median of the control as reference for treatment conditions. The plots are shown for all performed measurements.
Selection of specific compounds eliciting effects on zebrafish embryonic development. (a) Results of 11 drug treatments leading to strong phenotypic alterations. Heatmap showing the highest positive score in dark green and the lowest negative score in dark magenta. Each measurement is individually scaled. Grey color indicates no significance according to the Mann–Whitney U test. Shown are the mean values from at least two technical replicates, each replicate is composed of at least 10 embryos per condition. Boxplots with individual data points for the morphology assay can be found in Fig. 4. (b) Example images of the control and treatments with visible effects are shown in brightfield, vasculature Tg(fli1:EGFP) and heart Tg(myl7:mRFP). Brightfield and GFP images are shown as sobel-projections. The dynamic range of GFP images was homogenized using the DEVILS Fiji-plugin. Scale bars: 250 μm. (c) Selected compounds at clinically relevant concentrations result in mild modulation of embryonic. A heatmap (left) showing percentage of pericardial effusion and mortality of the selected drugs at their clinically relevant concentrations. Mortality rates for morphological assay (with PTU) and behavioral assay (without PTU) are plotted separately followed by a column of mean mortality rate. Dark magenta shows the highest effect. The relative percentage effect of these selected drugs at their clinically relevant concentrations are shown in the heatmap (right). The highest positive effect is shown in dark green and the lowest negative effect is dark magenta. Grey color indicates no significance according to the Mann–Whitney U test. Shown are the percentage from median values from at three technical replicates for drugs except for Remdesivir, that was tested in two replicates, each replicate is composed of at least 10 embryos per condition during drug treatment.
Of all compounds tested, Ivermectin treatment also caused the most severe effects on swimming behavior, consistent with previous findings26. The effect on swimming behavior of Ponatinib, Regorafenib, Cabozantinib, Ravuconazole, Delanzomib, Oxyclozanide, Anidulafungin, Camostat, Amuvatinib, Hanfangchin B, Bemcentinib and Apilimod and Molnupiravir had to our knowledge not been reported before.
Effect of compounds studied in the context of COVID-19
Next, we focused at the impact of those compounds with the highest interest for COVID-19 treatment on cardiovascular development and larval activity in zebrafish.
The compounds Ivermectin and Hydroxychloroquine were repurposed for COVID-19 treatment in the initial stage of the pandemic but their use was discontinued36,37,38. Consistent with previous reports in zebrafish and the reported side effects in human patients30,36. Ivermectin was highly toxic also in this study. At 1 µM embryos died between 4.5 and 5 dpf and at 0.5 µM mortality disappeared, we still observed effects on larval behavior (Figs. 3b,4 and 5). (Figs. 3c, 4, 5a and Supplementary Figs. S2). Hydroxychloroquine treatment showed significant reductions of ejection fraction and ISV area (Figs. 3c, 4 and 5). The effects of Hydroxychloroquine are in line with mild cardiac phenotypes, observed in neonates after treatment during pregnancy and increased cardiovascular mortality39.
Clinical trials on Favipavir, Ribavirin, Umifenovir and Lopinavir have been started40 but are not FDA approved for COVID-19 treatment. Favipiravir had a mild effect on body length and Lopinavir reduced ejection fraction (Figs. 3c, 4). We found that Ribavirin and Umifenovir treatment mildly reduced the ISV area (Figs. 3c, 4). While the effects observed for these compounds showed statistical significance, the amplitude of these effects was very low.
Remdesivir, Molnupiravir19, Paxlovid (Nirmatrelvir and Ritonavir)41, and Baricitinib are all used in the clinics and were approved by the FDA for COVID-19 treatment42,43. Sabizabulin has also been recommended for COVID-19 treatment20. For these compounds we carried out a further experimental round at concentrations with higher clinical relevance, i.e. at the highest concentration detected in blood plasma in humans (Cmax)44,45,46,47 (Figs. 4c and 5c). Sabizabulin treatment still resulted in a 20% mortality at its C max (0.2 μM) (Fig. 5c). Those larvae that survived, did however not present any cardiovascular or behavioral defects (Fig. 5c). At Cmax (4.5 μM) Remdesivir led to decrease in heart rate, ejection fraction and in body size of the embryo (Fig. 5c). For Ritonavir, Cmax was considerably higher than the tested concentrations (15.25 μM). We found that this altered the ejection fraction and heart rate (Fig. 4b). Baricitinib at C max (0.144 μM) led to 18.3% mortality and in those embryos that survived caused alterations in heart rate and motility (Fig. 5c). Despite a Cmax of only 0.04 μM, at this concentration Molnupiravir still altered swimming behavior and a decreased body size.
In sum, at clinically relevant concentrations, all five compounds used in the clinical context led to alterations in embryonic development.
Drug candidates modulate zebrafish immune response to SARS-CoV-2 spike protein treatment
Zebrafish possess many similarities in the innate immune response to that of mammals48. While successful virus amplification was not observed in wild type strains, SARS-CoV-2 spike treatment causes temporal immune response in zebrafish embryos and adult fish49,50,51. We decided to explore the effect of drugs selected for Cmax analysis on the modulation of the inflammatory and immune response gene expression profile (Fig. 6a). First, we subjected 5 dpf larvae to recombinant SARS-CoV-2 spike protein and extracted RNA from the larvae and performed qRT-PCR using a marker panel of genes involved in spike protein response. We found that Spike protein treatment altered expression of several but not all of the selected marker genes (Fig. 6b). Next, we repeated the treatments but now included a group to which we added either Remdesivir, Ritonavir, Molnupiravir, Baricitinib, Sabizabulin or DMSO for 48 h (Fig. 6c).
qPCR results for inflammation associated gene expression in zebrafish larvae (a) Scheme of larvae treatment with SARS-CoV-2 spike protein and selected drugs. (b) qPCR results for zebrafish larvae treated with SPIKE protein. (c) qPCR results represented as a heatmap representing fold change with respect to treatment with DMSO without spike (− SPIKE + DMSO). The upregulation is shown in dark green and the downregulation in dark magenta. Significance calculated between DMSO with SPIKE-protein treated group (+ SPIKE + DMSO) and the rest. Each color box represents a mean of 6 biological replicates of 5 embryo pools (n = 6, Mann–Whitney U-test treatment vs. + spike + DMSO; *p < 0.05, **p < 0.01, ***p < 0.001).
All compounds successfully reduced mxa, infΦ1, and ccl19 expression to the levels equivalent or even lower than the one of untreated controls. Baricitinib and Molnupiravir treatments also led to significantly reduced ccl20a.3 levels. The drug treatments also affected inflammasome pathway genes; ptg was upregulated to control levels by Remdesivir and Molnupiravir, while other drugs led to four-fold upregulation. Molnupiravir and Sabizabulin also successfully downregulated il4 expression. Of note, we also observed that in the absence of spike protein, drug treatments were already able to alter the expression of some of the immune response genes (Fig. 7). Of all five drugs, Remdesivir had the broadest effect on inflammatory genes in the absence of spike protein treatment.
Remdesivir, Ritonavir, Molnupiravir, Baricitinib or Sabizabulin alter the inflammatory gene response in the absence of spike protein. qPCR results for inflammation associated gene expression in zebrafish larvae treaded with selected compounds. Remdesivir, Ritonavir, Molnupiravir, Baricitinib or Sabizabulin treatment is compared to larvae treated with DMSO. Bars represent mean and SD of 6 biological replicates. Each replicate consists of a pool of 5 larvae. Asterisks represent p-values from the Mann–Whitney U test * < 0.05; ** < 0.01;*** < 0.001. Note that in all graphs the same DMSO control samples were used.
Overall, these results indicate that the zebrafish model is useful not only for screening phenotypic and behavioral alterations mediated by drug treatment, but also to assess compound effectiveness in attenuating the immune response after SARS-CoV-2 spike protein exposure. It also highlights that at Cmax, clinically relevant compounds can lead to alterations of embryonic development in the zebrafish.
Data access via an online data app
In order to facilitate mining of this large dataset we created an online data app using the open-source Streamlit app framework (Fig. 8 and accessible via this link: https://share.streamlit.io/alernst/covasc_dataapp/main/CoVasc_DataApp.py. The online data application allows to access raw measurements as well as batch corrected data and provides access example images for each treatment according to one’s individual interests.
Generation of a Web-based Data App. (a) The data app overview window shows which information can be found in the app. A heatmap with the previously described scores, results of the literature search, survival data, morphological analysis, activity analysis and images of the larvae. (b) Example tab opened, showing interactive violin plots for the larval length, which allows to see the distribution and individual data points. The checkboxes allow to view the corrected or raw data batch, and replicates can be grouped or shown individually. https://share.streamlit.io/alernst/covasc_dataapp/main/CoVasc_DataApp.py.
Here an overview of the functionality (Fig. 8a): on the left side, we allocated a side bar to select a specific tested compound or group of compounds. In the main window, tabs were created to allow visualization of an overview heat map, literature data, mortality of the treatments, morphological analysis and behavioral analysis. Additionally, we chose representative example images for each drug treatment group in each experimental replicate. An overview in brightfield and GFP fluorescence was provided in the last tab. Within the tabs we gave options to visualize the data (Fig. 8b). We provided options to show batch corrected data (check box: “Standardize to global median”) and to visualize individual experiments with the respective control or to group replicates by the treatment.
By providing the data app, we enable better insight into the data and facilitate visualization of all screening results beyond the ones shown in the figures. Furthermore, this app could be considered as a layout to be implemented for other screening projects.
