A high-content image-based drug screen of clinical compounds against cell transmission of adenovirus

Human adenoviruses (HAdVs) are fatal to immuno-suppressed individuals, but no effective anti-HAdV therapy is available. Here, we present a novel image-based high-throughput screening (HTS) platform, which scores the full viral replication cycle from virus entry to dissemination of progeny and second-round infections. We analysed 1,280 small molecular weight compounds of the Prestwick Chemical Library (PCL) for interference with HAdV-C2 infection in a quadruplicate, blinded format, and performed robust image analyses and hit filtering. We present the entire set of the screening data including all images, image analyses and data processing pipelines. The data are made available at the Image Data Resource (IDR, idr0081). Our screen identified Nelfinavir mesylate as an inhibitor of HAdV-C2 multi-round plaque formation, but not single round infection. Nelfinavir has been FDA-approved for anti-retroviral therapy in humans. Our results underscore the power of image-based full cycle infection assays in identifying viral inhibitors with clinical potential.


Methods
Virus. HAdV-C2-dE3B-GFP was produced as described 24 and fully sequenced 35 . In brief, the virus was generated by exchange of the viral E3B genome region with a reporter cassette harbouring the enhanced green fluorescent protein (GFP) under the immediate early Cytomegalovirus (CMV) promoter 24 . The virus was grown in A549 cells and purified by double caesium chloride gradient centrifugation 36 . Aliquots supplemented with 10% glycerol (v/v) were stored at −80 °C. HAdV-C2-dE3B-GFP was found to be homogeneous by SDS-PAGE and negative-stain analyses by transmission electron microscopy. Preparation of pre-plates. Ten µl 0.0125% DMSO in PBS was spotted on all 384 wells each of imaging-compatible 384-well plates (Matrix plates #4332, Thermo Fisher Scientific, Waltham, USA) using a Matrix WellMate dispenser and normal bore Matrix WellMate tubing cartridges (Thermo Fisher Scientific, Waltham, USA). Plates were sealed and stored at −20 °C.
Blinding. The PCL compound arrangement as dispensed by EPFL in four subset plates A -D comprising each screening set replicate 1-4 was blinded and replaced by UZH with internal identifier (Raw Plaque-2.0 infection scores of the PCL screen, imaged and analysed at UZH and Processed Plaque-2.0 infection scores of the PCL screen, imaged and analysed at UZH 29 , compoundIdentifier 1 to 1280). The identity of the compounds was only disclosed after the screening process and hit filtering (Raw Plaque-2.0 infection scores of the PCL screen, imaged and analysed at UZH and Processed Plaque-2.0 infection scores of the PCL screen, imaged and analysed at UZH 29 and Table 1, PCL_ID Prestw-1 to Prestw-1804 and compoundName).
Compounds. The PCL was obtained from Prestwick Chemical (Illkirch, France). 3′-deoxy-3′-fluorothymidine (DFT, CAS number 25526-93-6) was obtained from Toronto Research Chemical, North York, Canada. All compounds were dissolved in dimethyl sulfoxide (DMSO, Sigma-Aldrich, St. Louis, USA) at a final stock concentration of 10 mM and stored at −20 °C. PrestoBlue toxicity assay. Toxicity of the PCL compounds on uninfected A549 cells was assessed using the PrestoBlue Cell Viability reagent (Thermo Fisher Scientific, Waltham, USA). PrestoBlue 37,38 is a cell viability and cytotoxicity indicator based on resazurin. Resazurin is reduced to resorufin in cellular respiration by accepting electrons from NAPDH, FADH, FMNH, NADH and cytochromes. This reduction causes PrestoBlue to change from a non-fluorescent to a strongly fluorescent form. The conversion of PrestoBlue is proportional to the number of metabolically active cells and can be evaluated quantitatively using fluorescence or absorbance measurements. Briefly, following 3.5-day continuous treatment of A549 cells with compounds at concentrations and cell densities as in the screening protocol, 10% final PrestoBlue was added to each well and incubated for 1 h at standard cell incubation conditions. Fluorescence intensity (bottom-read) was measured using a multi-well plate reader (Tecan www.nature.com/scientificdata www.nature.com/scientificdata/ Fig. 1 The compound screening procedure. (a) Following assay development, stability and quality testing, the screening of the PCL against HAdV infection was performed. Imaging, image analysis and data processing were independently carried out at UZH and EPFL, before hit ranking. (b) Schematic overview of the wet-lab pipeline. PCL compounds and DFT positive control in DMSO as well as DMSO alone as negative control were pre-spotted onto 384-well imaging plates by Echo acoustic liquid handling at 10 nl corresponding to a final concentration of 1.25 µM in 80 µl assay volume/well and stored at −20 °C. Compound-blinded plates are thawed and 4,000 A549 cells/wells seeded. The following day, the cells were inoculated with HAdV-C2-dE3B at 1.77*10 5 genome equivalents/well. Allowing for multiple viral replication rounds, the cells were PFA-fixed at 72 hpi and the nuclei stained with Hoechst 33342. The infection phenotypes were imaged using an epifluorescence HT microscope and scored using Plaque2.0. The data of the four technical replicates were further processed in R or through EPFL-BSF LIMS. (c) Exemplary epifluorescence microscopy images of cells in 384-wells stitched to a screening plate overview of 32 replicates of negative (two most left columns) and positive control (two most Infinite F500, Tecan, Männedorf, Switzerland) with excitation at 560/10 nm, emission at 590/10 nm at a fixed gain. Doxorubicin hydrochloride (Prestw-438, Prestwick Chemical, Illkirch, France) was used as a positive control for cytotoxicity, at a final concentration of 10 µM, and the corresponding concentration of the drug solvent DMSO was used as a negative control. The full PCL library was tested on duplicated plates. The EPFL-BSF in-house Laboratory Information Management System (LIMS) was used for data processing and statistical validation. First, raw PrestoBlue readings were normalized per plate to negative control values at 0 and positive controls at 1. Then, the normalized values of both duplicates were averaged. Assay quality was assessed for each plate through the Z'-factor calculation. Compounds were considered toxic, when the normalized value for all replicates was higher than the average + 3σ (standard deviation, SD) of the DMSO negative control for the corresponding plate. Scores and score SD were then calculated for hit compounds by averaging normalized value for all replicates.
Preparation of plates for Z'-factor and drug screening. Ten nl of 10 mM PCL compounds, the nucleoside analogue DFT positive control (all dissolved in DMSO) and DMSO only as negative control were pre-spotted on imaging-compatible 384-well plates (Falcon plates, Corning Inc., New York, USA) using an Echo acoustic liquid handling system (Labcyte, San Jose, USA) by the EPFL-BSF, sealed and stored at −20 °C. Each Z'-factor 384-well plate consisted of 192 technical replicates of positive and negative controls, each. Each screening plate set consisted of four subset plates A to D. Each screening plate comprised 32 technical replicates of positive and negative controls, each, and 320 PCL compounds. Neg. ctr. refers to solvent control (DMSO), pos. ctr. to DFT-treated wells. Normalized indicates the mean readouts of each compound relative to the mean of the positive control over all replicates. Toxicity was accessed by PrestoBlue assay of 3.5-day treatment of uninfected A549 cells as well as by the nuclei Z'-factor in the screen. Hits were selected for low toxicity and high inhibitory effects compared to solvent control samples. Note that compounds were scored toxic, if they showed significant toxicity in either of the assays.
right columns) and 320 blinded PCL compounds (centre 20 columns). Hoechst-stained nuclei are shown in blue, viral GFP in green. (d) Representative 384-well epifluorescence microscopy images of the DMSO negative control (most left), the DFT positive control (most right) and the top hit Nelfinavir mesylate (centre). Empty black triangle indicates a plaque (infection focus) from a productively infected cell. White arrows point out infected cells that did not form a plaque. Hoechst-stained nuclei are shown in blue, infected cells expressing GFP in green. Scale bar is 5 mm.
www.nature.com/scientificdata www.nature.com/scientificdata/ Wet-lab screening pipeline. The screening was performed in four independent biological replicates 1-4. Liquid handling was performed using a Matrix WellMate dispenser and Matrix WellMate tubing cartridges (Thermo Fisher Scientific, Waltham, USA). Prior to usage, tubings were rinsed with 125 ml autoclaved double-distilled (dd) H 2 O followed by 125 ml autoclaved PBS. Pre-spotted compound plates were thawed at room temperature (RT) for 30 min, briefly centrifuged before compounds were dissolved in 10 µl/well of PBS. 4,000 A549 cells/well in 60 µl full medium were seeded onto the compounds using standard bore tubing cartridges. Following cell adhesion over night, the cells were inoculated with 1.77*10 5 genome equivalents per well of HAdV-C2-dE3B-GFP in 10 µl of full medium using bovine serum albumin (BSA, cell-culture grade, Sigma-Aldrich, St. Louis, USA)-blocked small bore tubing cartridges. The final compound concentration was 1.25 µM at 0.0125% DMSO. Infection was allowed to progress over multiple infection rounds for 72 h giving rise to foci of infected cells originating from a single first round infected cell. Cells were fixed for 1 h at RT by addition of 26.6 µL 16% PFA and 4 µg/ml Hoechst 33342 (Sigma-Aldrich, St. Louis, USA) in PBS using standard bore tubing cartridges. Cells were washed three times with PBS before PBS supplemented with 0.02% N 3 was added and plates were sealed for long-term storage at 4 °C. Following usage, tubings were rinsed with 125 ml autoclaved ddH 2 O followed by 125 ml autoclaved PBS and autoclaved for re-usage.
Imaging. Nuclei stained with Hoechst 33342 (DAPI channel) and viral GFP (FITC channel) were imaged on two devices. At UZH, plates were imaged on an IXM-C automated high-throughput fluorescence microscope (Molecular Devices, San Jose, USA) using MetaXpress (version 6.2, Molecular Devices, San Jose, USA) and a 4x air objective (Nikon S Fluor, 0.20 NA, 15.5 mm WD, Nikon Instruments, Minato, Japan) at widefield mode. Images of 2,048 2 px at 1.72 µm/px resolution were acquired on an Andor sCMOS camera (Oxford Instruments, Abingdon, UK). Exposure times: DAPI 150 ms, FITC 20 ms. At EPFL, images were acquired on a IN Cell 2200 automated high-throughput fluorescence microscope (GE Healthcare, Chicago, USA) using IN Cell Analyzer (version 6.2, GE Healthcare, Chicago, USA) and a 4x air objective (Nikon Plan Apo, 0.20 NA, 15.7 mm WD, Nikon Instruments, Minato, Japan) at widefield mode. Image size 2,048 2 px at 1.625 µm/px resolution acquired on an Andor sCMOS camera. Exposure times: DAPI 300 ms, FITC 40 ms.
Image analysis. The infection phenotype for each well was quantified by Plaque2.0 32 (https://github.com/ plaque2/matlab/tree/antivir) via five main read-outs: number of nuclei, number of infected nuclei, the ratio between infected and total nuclei referred to as infection index, number of multi-round infection foci termed plaques (plaque forming unit(s), pfu) and the integrated viral transgenic GFP intensity. Plaque2.0 parameters were optimized independently at UZH and EPFL for the data acquired at the respective institution. Well-and object-based read-outs are provided in the Plaque2.0 output files. www.nature.com/scientificdata www.nature.com/scientificdata/ Z'-factor calculation. The Z'-factor was computed using R version 3.3.2 39 according to Eq. (1) (1) where σ + is the SD of the positive control, σis the SD of the negative control, μ + the mean of the positive control and μthe mean of the negative control.
Screening data processing. Plaque2 Each compound's scores were normalized by the mean score of the DMSO negative control of the respective plate. Only non-toxic, effective PCL compounds were considered as HAdV inhibitor candidates. Non-toxic compounds were filtered by applying an inclusive μ + (mean of the negative control) ± 2σ (SD of the negative control) threshold for number of nuclei. Efficacy was filtered by applying an excluding μ + ± 3σ threshold for the infection scores (number of infected nuclei, infection index, number of plaques or integrated GFP intensity). Subsequently, compounds exhibiting significant toxicity to noninfected cells were excluded.

Data Records
Data structure and repositories. The screening data comprise the information collected during assay development, including stability, quality and screening of the PCL itself. The latter two were imaged on two different microscopes. We provide the parameters used for Plaque2.0 image analysis, and the code for the subsequent hit filtering in R. The data available for download at the IDR, accession number idr0081 33 , are structured as outlined in Fig. 2a. For download instructions, see idr.openmicroscopy.org/about/download. Moreover, the data can be viewed conveniently on the IDR web client (idr.openmicroscopy.org/webclient), where it is structured as depicted in Fig. 2b. Additionally, an annotated list of the PCL compounds as well as raw and scored screening data are available on figshare 29 as.txt files.
Data sets and file types. The data provided for download consists of three data sets 1 to 3 (see Fig. 2a).
-1-prePlates contains layouts (.csv), images (.tif), Plaque2.0 image analysis parameters (.mat) and results (.csv) for the assay stability test plates performed at UZH prior to Z'-factor plates (preZ) and the screen (preScreen). Analysis also contains the PrestoBlue raw results (.csv) for toxicity in absence of infection. The data provided for browsing via the IDR web client are assembled into five screens A to E (see Fig. 2b).
-idr0081-study.txt summarizes the overall study and the five screens that were performed.
-screenA contains the assay stability test plates performed at UZH prior to Z'-factor plates (preZ) and the screen (preScreen  www.nature.com/scientificdata www.nature.com/scientificdata/ -screenB contains the assay quality test plates (Z'-factor plates a and b) performed at UZH. idr0081-screenB-library. txt provides thorough information on the tested compounds including PubChem identifiers and their plate layout. idr0081-screenB-processed.txt presents the results of the Plaque2.0-based image analysis. idr0081-screenB-mean. txt summarises the infection scores per Z'-factor plate.
-screenC contains the assay quality test plates (Z'-factor plates a and b) performed at EPFL. idr0081-screenC-library.txt provides thorough information on the tested compounds including PubChem identifiers and their plate layout. idr0081-screenC-processed.txt presents the results of the Plaque2.0-based image analysis. idr0081-screenC-mean.txt summarises the infection scores per Z'-factor plate.
-screenD contains the PCL screening plates (in replicates 1 to 4, consisting of subset plates A to D) performed at UZH. idr0081-screenD-library.txt provides thorough information on the tested compounds including PubChem identifiers and their plate layout. idr0081-screenD-processed.txt presents the results of the Plaque2.0-based image analysis. idr0081-screenB-filtered.txt summarises the infection scores per compound and indicates if it was identified as hit.
-screenE contains the PCL screening plates (in replicates 1 to 4, consisting of subsets A to D) performed at EPFL. idr0081-screenE-library.txt provides thorough information on the tested compounds including PubChem identifiers and their plate layout. idr0081-screenE-processed.txt presents the results of the Plaque2.0-based image analysis. idr0081-screenE-filtered.txt summarises the infection scores per compound and indicates, if it was identified as hit. www.nature.com/scientificdata www.nature.com/scientificdata/ technical Validation assay stability. The wet-lab screening pipeline was optimized regarding liquid handling, cell seeding, virus inoculum, positive and negative controls, infection time, as well as imaging and image analysis. This ensured a high assay stability and reproducibility. Furthermore, all compounds, especially media and supplements, the BSA for tubing saturation, PFA-and Hoechst-supplemented fixative were prepared as large batch from a single lot and stored as single-use aliquots. Prior to every experiment, assay stability with respect to cell and infection phenotype was tested on pre-plates according to the established wet-lab, imaging and image analysis pipeline. Since the solvent control had already been spotted in 10 µl PBS, no further PBS was added prior to cell seeding. Periodically, the virus stock dilution was tested and adjusted for experiments if necessary. Assay quality determination: Z'-factor. The accuracy of the wet-lab, imaging and image analysis pipeline was assessed by two independently imaged and analysed Z'-factor plates ( Table 2 and Fig. 3). 3σ Z'-factors of numberOfInfectedNuclei, infectionIndex and numberOfPlaques were in the range of 0.30 to 0.57, scoring good to excellent. totalVirusIntensity (Z'-factors between −0.07 to 0.08) were not suitable to identify HAdV infection inhibitors, while numberOfNuclei (Z'-factors between −1.11 to −8.10) was not a useable readout either. Additionally, the Z'-factors were determined for each of the 16 screening plates (Table 3 and Fig. 4). 3σ Z'-factors of numberOf-InfectedNuclei, infectionIndex and numberOfPlaques were in the range of 0.27 to 0.57, scoring good to excellent. Independent analysis and filtering. Imaging, image analysis and screening data processing were performed by two independent research teams at UZH and EPFL, as depicted in Fig. 1. Raw and scored infection phenotypes are shown for UZH and EPFL analyses (Raw Plaque-2.0 infection scores of the PCL screen, imaged and analysed at UZH, Processed Plaque-2.0 infection scores of the PCL screen, imaged and analysed at UZH and Raw Plaque-2.0 infection scores of the PCL screen, imaged and analysed at EPFL, Processed Plaque-2.0 infection scores of the PCL screen, imaged and analysed at EPFL, respectively 29 ). Both dry-lab pipelines confirmed the high assay quality (Tables 2 and 3). During hit filtering, PCL compounds that gave significant toxicity in uninfected cells were excluded during hit filtering (Fig. 5, PCL compounds excluded due to toxicity in uninfected cells 29 ). As summarized in Fig. 6 left panel, both scores are strongly correlated with R 2 between 0.6870 and 0.9870. Both approaches yielded identical top scored compounds (Fig. 6, right panel), of which Prestw-1764, Nelfinavir mesylate, was the top hit.

Usage Notes
Five parameters were used to score the infection phenotype of each well: the number of nuclei (numberOfNuclei), number of infected nuclei (numberOfInfectedNuclei), the ratio between number of infected and total nuclei (infec-tionIndex), the number of multi-round infection foci termed plaques (numberOfPlaques) and the extend of viral GFP reporter expression as integrated GFP intensity totalVirusIntensity).  www.nature.com/scientificdata www.nature.com/scientificdata/  Of the 1,278 PCL compounds tested, 126 PCL compounds are found to be toxic, as shown in red, and listed in PCL compounds excluded due to toxicity in uninfected cells 29 . A549 cells were treated with PCL compounds in duplicates according to the screening wet-lab protocol, however, in absence of HAdV infection for 3.5 days. Doxorubicin hydrochloride (Prestw-438) was used as a positive control for cytotoxicity, at a final concentration of 10 µM, and the corresponding concentration of the drug solvent DMSO was used as a negative control. Cell viability was determined by PrestoBlue assay. PrestoBlue fluorescence intensities of each well were normalized per plate to negative control values at 0 and positive controls at 1. Compounds were considered toxic, when the normalized value for all replicates was higher than the average +3σ (standard deviation, SD) of the DMSO negative control for the corresponding plate. X-axis indicates compounds by their PCL identifier (PCL ID, see PCL compounds tested in the screening procedure 29 ). Normalized average PrestoBlue read-outs are depicted on the y-axis.
www.nature.com/scientificdata www.nature.com/scientificdata/ Fig. 6 Infection scores from independent dry-lab pipelines. Imaging, image analysis and data processing were performed independently at UZH and EPFL. Infection phenotypes in PCL-treated cells of four biological replicates were averaged and normalized against the DMSO solvent control. Linear regression plots of UZH and EPFL data are shown for (a) numberOfNuclei, (b) numberOfInfectedNuclei, (c) infectionIndex, (d) numberOfPlaques and (e) totalVirusIntensity of the 1,278 tested PCL compounds from are (green line). Red dots indicate toxicity in the absence of infection. Non-toxic compounds are shown as green dots. R 2 was calculated using GraphPad Prism 8.2.1. Highest scoring compounds are shown on the right, including the PCL_ID of some non-toxic compounds.