Virtual Screening Assisted Search for Inhibitors of the Translocated Intimin Receptor of Enteropathogenic Escherichia Coli

Graphical Abstract

Virtual screening methods were used to search for inhibitors of the bacterial virulence molecule intimin and its receptor Tir. Compounds were selected for biological study using an imaging-based method.

Abstract

This study aimed to identify inhibitors of the translocated intimin receptor (Tir) of enteropathogenic Escherichia coli (EPEC). EPEC is an intestinal pathogen that causes diarrhea and is a major health concern worldwide. Because Tir is a key virulence factor involved in EPEC pathogenesis, inhibiting its function is a potential strategy for controlling EPEC infections. Virtual screening was applied to chemical libraries to search for compounds that inhibit Tir-mediated bacterial adherence to host cells. Three sites were targeted using the cocrystal structure published earlier. A selection of compounds was then assessed in a cell-based infection model and fluorescence microscopy assay. The results of this study provide a basis for further optimization and testing of Tir inhibitors as potential therapeutic agents for EPEC infections.

Introduction

It is widely accepted that the current collection of antimicrobial therapies is insufficient to counter the effects of pathogenic bacteria in the future. Classical antibiotic strategies have proven to be extremely successful and have transformed the world we live in.¹ However, they have drawbacks, for example, their non-specific nature causes unwanted effects by eliminating health-promoting bacteria. Furthermore, pathogens have managed to evolve many mechanisms of resistance, and these severely hinder the usefulness of such drugs. In fact, it has been shown that resistance to new types of antibiotics can be detected within a few years from their deployment in treatment.² One way to solve this problem would be by developing drugs capable of specific inhibition of pathogenic physiological reactions, without decreasing the viability of the pathogens. While such drugs would not be entirely immune to resistance-driving forces, the selection pressure would likely be much lower as the targeted pathogens could survive by switching to alternative lifestyles.

In practice, this could be achieved by pharmaceutical intervention directed towards virulence inducing molecules utilized by pathogens.³ Most gram-negative pathogens use injected virulence molecules to infect host cells. Enteropathogenic Escherichia coli (EPEC) belongs to a subclass of such pathogens, often called attachment and effacing (A/E) pathogens.⁴ These pathogens cause the effacement of intestinal microvilli and characteristic actin rich protrusions called pedestals, upon which EPEC stands on the plasma membrane.⁴ This virulence cascade is not exclusive to EPEC; a highly similar process occurs in other A/E pathogens such as Enterohemorrhagic Escherichia coli (EHEC) and the mouse specific pathogen Citrobacter rodentium. These pedestals are known to be caused by virulence molecules transferred from the periplasm to the cytoplasm of the host cell via the type three secretion system (T3SS), a needle-like appendix that employs a molecular pump to do so.⁴ Interestingly, the T3SS and the mentioned virulence molecules are encoded on a conserved pathogenicity island known as the locus of enterocyte effacement, prophages, or other insertion elements, meaning that many variations of the theme exist among A/E pathogens in the wild.

Among the proteins injected into the host‘s cytosol is a receptor known as the translocated intimin receptor (Tir).⁴ This receptor is responsible for the aberrant actin condensation associated with the lesions caused by A/E pathogens. Tir integrates into the plasma membrane of the host cell, and works as a receptor for intimin, a ligand protein that is autotransported from the periplasm.⁴ The structures of the translocated intimin receptor bound to intimin have been described using x-ray diffraction crystallography, shedding light on how the Tir interacts with intimin.⁵ Ultimately, upon activation by the ligand, the receptor clusters⁶ and interacts with downstream host adaptors and kinases responsible for actin nucleation, leading to the emergence of the pedestals and the destruction of microvilli.

Interfering with this process could attenuate the virulence of EPEC, given that experiments have shown that Tir-intimin interaction is necessary but not sufficient to display pathogenic-like phenotypes in cell culture models and ex vivo transplants.⁷ However, no orthosteric small molecule inhibitors of Tir have been found to date, despite the extensive research on the binding interaction. Ross and Miller demonstrated the existence of a binding hotspot consisting of three amino acid residues at the interface between Tir and intimin.⁷ They then created short linear peptides containing these central Tir residues and tested their ability to bind the extracellular domain of intimin, but these did not demonstrate binding activity. This suggests that linear peptide mimetic inhibitors are insufficient for binding intimin, and other structural properties perhaps provided by buried residues are required for efficient binding. The authors speculate that this may be the structural preorganization of the β-hairpin residues of the receptor. The design of Tir binding antibodies and nanobodies has also been described (CA2479270C).⁸ However, the nanobody did not bind to EPEC Tir. Additionally, as orally administered treatments, these would need to be able to resist digestion in the human gastrointestinal tract.

We hypothesized that small molecule inhibitors of the Tir-intimin interface would effectively limit EPEC-induced lesions and thus virulence without compromising EPEC viability. Such molecules could possibly be administered orally avoiding many of the typical problem related to the development of orally administered pharmaceutical, such as reaching a high plasma and tissue concentration and avoiding efflux from the bacteria. Furthermore, since the sequences of EHEC Tir and EPEC Tir hairpin share 85 % identity, in the best case, inhibitors of this target could also apply to other A/E pathogens.⁹

To answer this research question and discover such compounds, we applied virtual screening methods to three putative binding sites to select compounds from compound libraries and studied the compounds by using an actin pedestal detecting imaging assay.

Our aim was to provide insight into how this receptor could be inhibited and used as a target for drug discovery and for designing molecules that can prevent the virulence of EPEC. Preventing the virulence of EPEC would be an alternative to the traditional antibiotic-based approach to treating the illness, and in the long run, this could help to alleviate the problem of antimicrobial resistance.

Results and Discussion

The interaction between Tir and intimin is critical to EPEC pathogenicity, and the resulting pathological phenotypes may be analyzed using an imaging-based phenotypic assay. We hypothesized that small molecule compounds could prevent EPEC virulence by interfering with the normal binding of the ligand to the receptor. Using structural data from the crystal structure of the receptor and ligand complex (PDB entry 1F02), we first estimated the capacity of small molecules from large-scale virtual libraries to bind to three selected locations on these structures using docking studies. Following that, the most promising compounds were empirically tested using an actin pedestal counting assay with high-content imaging.

The Crystal Structure of Tir-Intimin Suggests Three Potential Sites for Binding of Inhibitors

The interface between intimin and its receptor, Tir, is a broad (1335 Ų) and shallow protein-protein interface that is mainly hydrophobic (Figure 1).⁵ The positively charged intimin tip, on domain 3, is formed by a short α-helix (residues 904–909). The helix is hydrogen-bonded to the main chain amides of Tir residues Asp 315 and Asp 316 via the main chain carbonyl oxygens of Ser 909 and Lys 908. Adjacent to these residues there is a β-sheet with similar characteristics and topology as the carbohydrate recognition site of C-type lectin domains. This area is known to participate in binding Tir.⁹

Figure 1

Intimin, a bacterial adhesin binds to a receptor that has translocated to the plasma membrane; A) A schematic representation of intimin and its receptor Tir. Both the adhesin and its receptor are believed to form dimers and bind to each other in a reticulating manner. Intimin consists of several domains (D00 to D3). The passenger domains of the adhesin are exported through the outer membrane and the extracellular lectin-like domain of intimin (D3, blue), interacts with the receptor (Tir, green), and the receptor clusters into a dimer. B) The interface between intimin (blue surface) and Tir (green surface), secondary structure as ribbon; C) The Tir-Tir interface, two instances of the Tir monomer in red and in green. The ribbon depicting an intimin unit; D) Compound 7 docked in Model 1 (Tir); E) Compound 39 in Model 2 (intimin D3 domain); F) Compound 48 in Tir-Tir interface (Model 3).

Monomers of the Tir intimin binding domain are hairpin-shaped molecules with two long α-helices connected via a β-hairpin. This hairpin structure is the main binding partner with intimin D3. Additionally, the two α-helices mediate dimerization by forming an antiparallel four-helix bundle similar to ColE1 Rop protein, a biomolecule responsible for maintaining E. coli plasmid copy number. This interface is also hydrophobic except for a pair of hydrogen-bonded Thr 282 side chain residues.

We selected three different sites in the complex as starting points for interfering with the interaction between the receptor and ligand. Firstly, 1) the intimin binding site on Tir, as this contains the hotspot for the PPI and previous studies suggest it to be required for function, second 2) the extracellular domain of intimin itself, as this is the opposing binding partner to the above mentioned binding site on the receptor and lastly 3) the receptor dimer interface (Figure 1B), as clustering of the receptors mediates their activity. Computational models for docking were developed for these sites as follows (Figure 1D–F).

Model 1. Intimin binding site on Tir

Tir acts as the receptor on the plasma membrane of the host cell. We ran virtual screening on a library of commercially available compounds to identify small molecules capable of disrupting the interaction between intimin and Tir. These compounds were docked into the binding site around the Asn 300 of Tir, which scanning alanine mutagenesis has shown to be a part of the binding hotspot, and thus important for interaction with intimin.⁷ The model binding site was composed of residues Phe 291 and Asn 296 – Val 318. The Asn 300 side chain was set as a hydrogen bond donor and acceptor constraint in the docking calculations, therefore all hit compounds could form two hydrogen bonds with Asn 300 (Figure 1D).

Model 2. Extracellular domain of intimin

As mentioned earlier, lectin-like intimin D3 domain acts as the natural ligand to Tir. We carried out virtual screening utilizing a library of compounds available from the Institute for Molecular Medicine Finland (FIMM) to identify small molecules capable of disrupting the interaction between intimin-Tir via binding to the extracellular domain of intimin. The compounds were docked to a site near the 890–930-residues, interacting with the short alpha helix or β-strands C, D, and E, relevant for binding Tir (Figure 1E) according to mutagenesis and yeast two-hybrid studies.⁹

Model 3. Receptor dimer interface

Because the crystal structure of Tir shows a dimer of the receptor molecules in an antiparallel orientation⁵ and empirical research has shown that the receptors dimerize upon activation by intimin,⁶ we hypothesized that a small molecule compound binding at the dimerization interface might inhibit the dimerization and thus prevent the pathological changes in actin dynamics. SiteMap was used to uncover a site on the receptor at the dimerization interface. To identify small molecules able to disrupt Tir-Tir clustering, we performed virtual screening using a library of compounds available from FIMM (Figure 1F).

Compounds from these models were selected for biological evaluation based on the scoring function, interactions formed at the binding site, binding site shape complementarity, and chemical structure (Table S1).

In Vitro Evaluation of the Selected Compounds

We employed an imaging assay based on the fluorescent actin staining assay adapted for high-content screening (FAS-HCS) to analyze the compounds identified via virtual screening.¹⁰ In this assay, the actin binding molecule phalloidin is utilized to visualize EPEC infection in vitro, which is characterized by aberrant actin-rich structures on host cells known as pedestals. To visualize the bacterial colonies, an EPEC strain expressing mCherry is employed and, aurodox, a Tir translation inhibitor, is used as a control.

None of the compounds targeting the intimin binding site on Tir (Model 1) inhibited pedestals (Figure S2). One of the compounds (19) targeting the intimin lectin domain (Model 2) decreased the readout (Figure S3). However, closer examination revealed that the compound is fluorescent at the detection wavelength, interfering with the assay detection signal (Figure S4). Such interference at lower wavelengths is typical and should always be controlled for. Compounds targeting the receptor dimer interface (Model 3) were initially tested at two concentrations. At 50 μM, one compound (48) considerably and five (60, 66, 73, 74, and 75) marginally lowered the assay readout (Figure 2, Table S1), but did not do so at the lower concentration of 25 μM (Figure S5, Table S1).

Figure 2

Six of the compounds (48, 73, 66, 75, 60, and 74) selected via the virtual screening model 3 targeting the receptor dimer interface decreased pedestals (50 μM). A) The FAS-HCS assay was run once with four technical replicates (n=1). Aurodox was used as the positive control and 1 % DMSO in MEM as the no treatment condition with 10 replicate wells each. Compounds are plotted in descending order based on activity. Three compounds with respective concentration response data from the FAS assay with growth inhibition at the same concentrations at 6 h; B) Compound 48; C) Compound 60; D) Compound 66. Measurements based on two technical replicates, the average of which indicated by an X for each three independent experiments (n=3). The grand mean for these three is indicated by the purple circle. The concentrations range from 50 μM down in ten two-fold serial dilutions.

Compounds 48, 60, 66, 74, and 75 were repurchased from commercial vendors for additional testing. Compound 73 was no longer available from the suppliers. Furthermore, compound 75 did not show a reproducible concentration-dependent impact, and compound 74 was a known cytotoxin (lanatoside C), so the two were excluded from further testing (Figure 3).

Figure 3

Structures of the three compounds (48, 60, and 66) selected from primary screening for evaluation in further experiments.

Known Antimicrobial Compounds at Subinhibitory Concentrations Do Not Prevent Pedestals

Some of the compounds selected by the virtual screening in model 3 were known antimicrobial compounds (79, 80, 81, 82, and 83) (Table S1). At these high concentrations antimicrobials abolish microcolonies required for image analysis. Therefore, we decided to test the capacity of these compounds to inhibit actin pedestals at multiple lower concentrations. Antimicrobials typically display polypharmacy, and previously described T3SS-inhibitors, such as aurodox, act as growth inhibiting antibiotics at higher concentrations.¹¹

However, only the lowest concentration (95 nM) of compound 81 (ciprofloxacin) did not inhibit growth (Figure S7), although it did lower colony forming units significantly (Table S2). This lowest concentration showed no activity in inhibiting actin pedestals (Figure S7). Compound 80 (tomatine) also decreased pedestals significantly at concentrations only minorly affecting growth (Figure S7). Compounds 79, 82 and 83 fully inhibited growth even at the lowest tested concentration (95 nM), but did not show inhibition in the FAS-assay at this concentration.

Hit Compounds Inhibit the Growth of Bacteria or the Viability of Host Cells

Growth inhibition affects actin pedestal counts in the FAS-HCS assay, leading to false positive results. For example, antibiotic treated EPEC typically still adhere to cell monolayers to some extent, but their ability to establish microcolonies or produce actin pedestals is impaired. Microcolonies are also required to properly segment the image.

Therefore, we investigated whether the three hit compounds influenced bacterial growth with some non-specific toxic mechanism. At the primary screening concentrations (50 and 25 μM), all the hit compounds inhibited growth (Figure 2). These findings were further supported by colony plating experiments. The 50 μM treated plates had no colonies, but the 25 μM treated plates had low colony counts, indicating toxicity (Tables S2A and S2B).

Pedestal inhibition, just as the prevention of any other cellular phenotype, may also result from a decrease in host cell viability, especially at high concentrations of compounds more likely to display cytotoxicity. As a result, tests were conducted to investigate the cytotoxicity of compounds that decreased the readout in the primary screening at high concentrations (50 μM). Except for compound 80 (tomatine) at the maximum 50 μM concentration, none of the compounds reduced the viability of the host cells (Figure S6).

Conclusions

In this study we searched for inhibitors of the translocated intimin receptor of EPEC. We utilized computational chemistry methods to select compounds potentially capable of interfering with the receptor complex and then tested these using a robust screening assay based on widely used FAS-assay and automated image analysis. Selective inhibition of pedestals via inhibiting Tir could prove to be a medically efficient strategy to treat EPEC infections, but there are currently no known inhibitors of this receptor.

Our work identified some challenges in interfering with this pathway. The crystal structure of the protein complexes does not appear to have a deep pocket or clefts found on many drug targets such as on G-protein coupled receptors, kinases, other enzymes. We demonstrated that the used methodology is not sufficient to uncover small molecule inhibitors of the translocated intimin receptor. Without any good structure-based information to go on, the best way forward would be to use a very comprehensive search, e. g., by screening large biogenic compound collections.

Small molecule inhibitors of Tir would be a new class of therapy for infections of A/E pathogens. Currently there are no A/E pathogen infection specific therapies.

Overall, antivirulence therapies could avoid the issues in traditional antibiotic treatments, both the following adverse effects and the increase of antimicrobial resistance, thus giving new opportunities to tackle the problem of antimicrobial resistance.

Experimental Section

Virtual Screening and Selection of Compounds

A 2.9 Å resolution crystal structure of the C-terminal 282-residue fragment of intimin in complex with the intimin-binding domain of translocated intimin receptor (PDB code 1F02) was used for all the models.

Model 1. Intimin binding site on Tir

Compound library

Drug-like small molecule libraries from Asinex, ChemBridge, Enamine, Life chemicals, Key Organics, and Vitas-M were downloaded in SDF format. These libraries were merged, duplicates were removed, and the correct protonation and tautomeric states were assigned which resulted in a library containing 741 778 compounds. For these compounds a library of conformers was generated using the OMEGA software (Release 4.1.1.1, OpenEye Scientific Software, Inc., Santa Fe, NM, USA; www.eyesopen.com)¹¹ using default settings, which resulted in a maximum of 200 conformers per ligand.

A collection of compounds available at the Institute for Molecular Medicine Finland (FIMM), University of Helsinki, in a database of 132 566 compounds was prepared using LigPrep by including options such as adding hydrogens, generating ionization states at pH 7.2±2.0, generation of tautomers, and multiple conformers. This library has been generated from multiple sources, from commercial libraries (Microsource, SPECTRUM, NIH clinical collection, diversity sets from ChemDiv and ChemBridge, peptidomimetic compounds, and FDA-approved drugs). The protein preparation workflow of Maestro Suite was used to prepare the intimin protein domain, by separating it from the cocrystal structure (1F02), correcting errors in the structure, such as adding missing hydrogens and minimization of hydrogen atoms, filling missing side chains.

Structure-based virtual screening

For preparing the binding site, the above-mentioned crystal structure was used, and the binding site was prepared using MAKE RECEPTOR (Release 4.1.0.1, OpenEye Scientific Software, Inc., Santa Fe, NM, USA; www.eyesopen.com). The grid box around the Asn 300 residue of Tir was generated and adjusted to get a box with the following dimensions: 17.33 Å×26.33 Å×14.67 Å and the volume of 6694 ų. For “Cavity detection” the slow and effective “Molecular” method was used for detection of binding sites. Inner and outer contours of the grid box were also calculated automatically using “Balanced” settings for “Site Shape Potential” calculation. The inner contours were disabled. Asn 300 was defined as a hydrogen bond donor and acceptor constraint for the docking calculations. The small molecule library, prepared by OMEGA, was then docked to the prepared binding site on intimin (PDB entry: 1F02)⁵ using FRED (Release 4.1.0.1. OpenEye Scientific Software, Inc., Santa Fe, NM, USA; www.eyesopen.com)^{12, 13} Docking resolution was set to high, other settings were set as default. A hit list of the top 1 000 ranked molecules was retrieved and the best ranked FRED-calculated pose for each compound was inspected visually and used for analysis and representation.

Model 2. Extracellular domain of intimin

Compound library and protein preparation

The same collection of compounds available at FIMM was used similarly as above (Model 1). The protein preparation workflow of Maestro Suite (version 2018-1, Schrödinger) was used to prepare the Tir-binding domain of intimin, by separating it from the cocrystal structure, correcting errors in the structure, such as adding missing hydrogens, and minimization.

Structure-based virtual screening

For the binding site preparation, the above-mentioned crystal structure was used, and a grid box was created by Glide¹⁴ grid generation around the alpha helix (residues 904–909) of intimin with dimensions 20 Å×20 Å×20 Å.

Docking studies were carried out to identify possible interacting ligands using the Glide virtual screening workflow implemented in Schrödinger. Virtual screening workflow of Glide includes various stages with increasing docking precisions such as high-throughput virtual screening (HTVS), standard prediction (SP), and extra precision (XP) mode. All molecules were processed initially in HTVS mode and the top 10 % of HTVS output was processed with SP docking mode. Then, the top 10 % molecules of SP docking were processed through XP docking.

Model 3. Receptor dimer interface

Compound library and protein preparation

The same collection of compounds available at FIMM was used similarly as above (Model 1). The protein preparation workflow of Maestro Suite was used to prepare the extracellular domain of Tir, by separating it from the cocrystal structure, correcting errors in the structure, such as adding missing hydrogens, and minimization.

Structure-based virtual screening

Schrödinger Sitemap¹⁵ was used to discover a site with Dscore 0.47 using the same PDB entry as in Model 2. Using the default settings for SiteMap, only this one site was discovered on the receptor. Using the shallow setting, a site with a Dscore of 0.97 was in the same cavity, so we decided to use this. However, the Glide grid was based on the site located using the default settings. This site is located next to the Trp 285 residues that form a hydrogen bond in the dimer interface.

Using the Glide virtual screening workflow built in Maestro, docking tests were performed to identify potential interacting ligands. All molecules were originally processed in HTVS mode, with the top 10 % of HTVS output being processed in SP docking mode. The top 10 % of SP docking molecules were then processed through XP docking.

High-content Imaging and Screening

Primary screening and concentration-response experiments were performed using a modification of the widely used FAS assay published earlier.¹⁰ In short, this is an infection assay with fluorescence imaging readout. A cell monolayer is infected with an EPEC strain emitting fluorescence and host cell actin is stained with phalloidin. The data is processed using a custom data reduction pipeline which produces as a readout the proportion of all bacterial microcolonies with actin pedestals. This is achieved by segmentation algorithms and a colocalization analysis. The code used for analysis is available at https://github.com/tpylkko/FAS-HCS.

Compound Sources

For testing compounds from docking Model 1 the compounds were purchased from Molport, and 10 mM stocks in DMSO were prepared and stored at −20 °C until use. The stocks were serially diluted to source plates (384-well, Nunc) using Biomek i7 liquid handling workstation (Beckman Coulter) in a volume of 37.5 μL.

For testing compounds from Models 2 and 3, compounds were plated to 384-well plates by using an Echo liquid handler (Labcyte) at the Institute for Molecular Medicine Finland (FIMM), High Throughput Biomedicine unit. Before use, 37.5 μL of MEM was dispensed to these plates and they were shaken at RT for 30 min (500 rpm).

For retesting compounds for concentration-response, compounds were reacquired from Molport and stocks of 2.5 mM or 10 mM were prepared in DMSO. The stocks were serially diluted to source plates (384-well, Nunc) using Biomek i7 in a volume of 37.5 μL. Two dimensional representations of all these compounds are available in Scheme S1.

Screening Procedure

To screen the compounds, 37.5 μL of preincubated EPEC (2348/69) suspension was added to the above-mentioned source plates, containing the compounds in 37.5 μL of MEM with a dispenser (Mantis, Formulatrix) in such a way that in the end the source plate contained 75 μL of a mixture with the correct concentration of the compounds and a bacterial suspension with a multiplicity of infection of 1 : 15. A volume of 60 μL of this mixture was moved to the screening plates containing Caco-2 (HTB-37 ATCC) cells and the plates centrifuged at 1000 g for four minutes and moved to 37 °C, 5 % CO₂ for two hours (Biospa, Biotek). After this, an equal volume (60 μL) of the 2X staining solution was applied to the plate using a dispenser (Mantis, Formulatrix) and incubated at RT for 20 minutes. Biomek i7 was used to wash the plate three times with HBSS and the imaging was then done on the Cytation 5 (Biotek). More details are provided in the previously published protocol.¹⁰

Bacterial and Cellular Viability Assays

EPEC 2348/69 was grown overnight in LB liquid culture. Compounds were plated onto a 384-well plate in LB or Miller Hinton broth and shaken (500 rpm, 30 min) using a Biomek i7 protocol. The bacterial suspension was then diluted to the appropriate range (2 million CFU ml⁻¹) by measuring the optical density using a densitometer. This suspension was dispensed using a dispenser (Mantis, Formulatrix) into the compound plate to reach a total volume of 75 μL. After this, the plates were incubated using Biospa (Biotek) for 24 hours and their OD₆₀₀ measured once an hour using Cytation 5. This procedure was repeated three times using a different colony and newly prepared media and other reagents, and the well location of a specific compound series was varied among the repeats randomly to diminish any possible plate area effects.

To measure CFU counts, a similar procedure was used, except that after treatment of two hours, a 3 μL sample of bacteria was taken from each well and diluted by a factor of 30 into a new 96-well plate containing PBS. This step was repeated to reach a total dilution factor of 900. From this suspension, 50 μL samples were spread on LB agar plates and incubated at 37 °C overnight. Images were captured of the plate using a digital camera from a phone. The number of colonies on plate was then calculated using OpenCFU 3.9. Three independent experiments were performed on distinct days using distinct colonies, and newly prepared media and reagents.

Cellular viability was assessed using the CellTiter-Glo (Promega) kit according to the manufacturer‘s instructions. Assay conditions were identical to the screening procedure described above except that 0.5 % Triton-X was used as the positive control and PBS as the negative. Three independent experiments were performed on distinct days using newly acquired media and reagents.

Author Contributions

Tuomas Pylkkö: Conceptualization, Methodology, Software, Data curation, Writing – Original draft preparation, Visualization, Investigation, Validation, Writing – Review & Editing; Tihomir Tomašič: Investigation, Methodology, Writing – Review & Editing; Antti Poso: Supervision, Writing – Review & Editing; Päivi Tammela: Conceptualization, Supervision, Writing – Reviewing and Editing, Funding acquisition.

Acknowledgments

The authors wish to acknowledge CSC – IT Center for Science, Finland, for computational resources and OpenEye Scientific Software (Santa Fe, NM, USA) for free academic licenses for the use of their software. The facilities and expertise of the FIMM High Throughput Biomedicine Unit and the DDCB unit at the Faculty of Pharmacy, supported by HiLIFE and Biocenter Finland, are gratefully acknowledged. Figure 1A and Table of Contents image created with BioRender.com.

Conflict of interest

The authors declare no conflict of interest.

Open Research

Data Availability Statement

The data that support the findings of this study are available in the supplementary material of this article.