Not Every Hit‐Identification Technique Works on 1‐Deoxy‐d‐Xylulose 5‐Phosphate Synthase (DXPS): Making the Most of a Virtual Screening Campaign

In this work, we demonstrate how important it is to investigate not only on‐target activity but to keep antibiotic activity against critical pathogens in mind. Since antimicrobial resistance is spreading in bacteria such as Mycobacterium tuberculosis, investigations into new targets are urgently needed. One promising new target is 1‐deoxy‐d‐xylulose 5‐phosphate synthase (DXPS) of the 2‐C‐methyl‐d‐erythritol 4‐phosphate (MEP) pathway. We have recently solved the crystal structure of truncated M. tuberculosis DXPS and used it to perform a virtual screening in collaboration with Atomwise Inc. using their deep convolutional neural network‐based AtomNet® platform. Of 94 virtual hit compounds only one showed interesting results in binding and activity studies. We synthesized 30 close derivatives using a straightforward synthetic route that allowed for easy derivatization. However, no improvement in activity was observed for any of the derivatives. Therefore, we tested them against a variety of pathogens and found them to be good inhibitors against Escherichia coli.


Introduction
Antimicrobial resistance (AMR) is an increasing threat worldwide. The Covid-19 pandemic has raised awareness for infectious diseases, but while a lot of energy and money has been invested into vaccine and drug development against SARS-CoV-2, the same is not true for antibiotic research. Here, a lot more effort is needed to prevent the estimated 10 million deaths from AMR in 2050. [1] Bacteria have a variety of mechanisms to develop resistance against an antibiotic. Especially problematic are the ESKAPE pathogens (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter spp.), which have the highest priority for finding new antibiotics due to resistance development according to the World Health Organisation. [2] Another critical organism with rising AMR is Mycobacterium tuberculosis. [3] Its standard treatment takes six months and adherence to the protocol is crucial to prevent resistance development. However, infections with multidrugresistant M. tuberculosis (MDR-TB) (resistant to the most effective tuberculosis drugs rifampicin and isoniazid) have continued to increase. [4] To be able to treat patients in the future, it is important to find new drugs with a new mode of action and addressing novel targets. A promising pool of targets is the 2-C-methyl-d-erythritol 4-phosphate (MEP) pathway. [5] It is utilized by many of the ESKAPE pathogens and by M. tuberculosis to synthesize the essential isoprenoid precursors dimethylallyl diphosphate (DMADP) and isopentenyl diphosphate (IDP). As this pathway is absent in humans, fewer side effects for the patient are expected. The first enzyme in the pathway is 1-deoxy-d-xylulose 5-phosphate synthase (DXPS). It catalyzes the condensation of pyruvate and d-glyceraldehyde-3-phosphate (d-GAP) to 1-deoxy-d-xylulose 5-phosphate (DOXP). [6] The product is not only used in the MEP pathway but is also involved in the biosynthesis of vitamins B 1 and B 6 . DXPS is also the rate-limiting step and it has the highest flux-control coefficient, which means its activity influences the production of IDP and DMADP the most of all MEP enzymes. Hence, its inhibition will lower isoprenoid production most effectively. [7] We have recently solved the first crystal structure of truncated M. tuberculosis DXPS (ΔMtDXPS, protein data bank (PDB): 7A9H). [8] MtDXPS is usually difficult to purify and crystallize. By removing a flexible loop from the native structure, we were able to crystallize the truncated enzyme at 1.85 Å resolution. Before, there was only a homology model available to perform structure-based virtual screening. [9] Deep convolutional neural networks in drug discovery Atomwise developed the first deep convolutional neural network for molecular binding-affinity predication. Their Artificial Intelligence Molecular Screen (AIMS) Awards program ran from 2017 to 2022 with two major goals: to support researchers with resources to help advance their work and to demonstrate that the AI technology developed at Atomwise can indeed identify hits and add value to drug-discovery programs across all protein families. Through the AIMS program, Atomwise contributed AI technology and physical molecules to labs across the world while the academic partners contributed their biological knowledge and expertise in physically assessing compound activity. A manuscript summarizing the results of the program is currently in preparation.
The AtomNet platform is the first system that uses structural information of the target protein to make predictions. [10] The model takes millions of experimentally determined bioactivity values and thousands of protein structures from different families into account. Based on this information, it is even possible to perform AtomNet screens on targets for which no binding partners are known.

Virtual screening against truncated M. tuberculosis DXPS
Various approaches are available for structure-based virtual high-throughput screening (vHTS) of small-molecule databases. An overview and comparison of the different scoring methodsgenerally classified as physics-based, empirical, knowledgebased, and machine learning-based, the latter of which includes Atomwise's Deep Neural Network-based approach -is given by Li et al. [11] vHTS approaches can help identify bioactive molecules from an ever-increasing supply of available (either "off the shelf" or "make-on-demand") small drug-like molecules that can no longer be explored by traditional high-throughput screening methods. [12] Screening of such ultra-large spaces usually relies on accelerated GPU-based code, which has also been implemented into the AtomNet platform. [13] After the crystal structure of ΔMtDXPS (PDB: 7A9H) had been solved, Atomwise performed an AtomNet model virtual screen (VS) using the Enamine small-molecule library of several million compounds, as described previously. [8,10,14] The screen focused on the binding site of the second substrate d-GAP (gray mesh Figure 1). [15] Below the substrate pocket, thiamine diphosphate (ThDP) is bound to the protein (yellow and red sticks) and coordinates magnesium (green sphere). Pyruvate binds first to ThDP and they form a stable enamine intermediate that can be crystallized in the absence of d-GAP. Finding inhibitors for the d-GAP binding site has the benefit of not interfering with the stable and tightly bound enamine but preventing d-GAP from binding. In addition, this is the pocket that differs most from other ThDP-dependent enzymes, so the development of a DXPS-selective inhibitor should be possible. In total, 94 virtually binding compounds were identified and purchased from Enamine for testing.

Hit-identification and validation attempts
Both MtDXPS enzymes (native and truncated) were not suitable for activity assays as their substrate turnover in comparison to the model enzyme Deinococcus radiodurans (Dr)DXPS was very slow and a change in reaction velocity in the presence of inhibitors could not be observed reliably. [8,16] Hence, we used the two Mt-homologues in SPR binding studies, and DrDXPS for the enzyme-activity assays.
To verify that the truncation of MtDXPS not only provides a more stable and easier to crystallize protein, but that both homologues can be used interchangeably in binding studies, we immobilized both proteins on the surface of a surface plasmon resonance (SPR) chip. In both cases, the proteins gave a steady response signal (ΔMtDXPS: Figure S1, MtDXPS: Figure S2) over a period of four weeks. All 94 compounds and two dimethyl sulfoxide (DMSO) blanks were tested at 100 μm ( Figure S3, Table S1), and for each protein the four compounds with the highest response unit (RU) were defined as hits (1, 2, 3, 4 and 1, 3, 5, 6). There is an overlap of two compounds (1 and 3), and a third compound (4) is also binding well to both although it is not among the best four against MtDXPS. Dissociation constant (K D ) determinations were not possible as none of the compounds is soluble enough to reach saturation of the binding site. After validating the binding of all 94 compounds, their inhibitory activity against DrDXPS was measured in a DXPS-1deoxy-d-xylulose 5-phosphate reductoisomerase (IspC)-coupled assay ( Table S2). None of the six SPR hits show inhibition of over 50 % at 100 μm ( Table 1). The highest inhibition of DrDXPS is seen using compound 1 at 39 % inhibition that is also binding in SPR while the other SPR hits 2-6 inhibit below 20 %. The most interesting compound 1 (Scheme 1) and five others inhibit DrDXPS more than 23 % (7-12).
Despite the low activity against DrDXPS, the SPR results against MtDXPS motivated us to synthesize a compound library with closely related derivatives of 1 to investigate whether this can be developed into a hit class for DXPS.

Compound re-synthesis
First, we developed a synthetic pathway that allowed easy derivatization on both sides of the molecule (Scheme 1).

Synthesis of derivatives
The synthesis of twelve derivatives with different substituents on the Eastern ring was performed using a variety of different thiols (Scheme 2). We based the selection of derivatives on the premise of making only small changes to get a clear picture of the structure-activity relationship. Therefore, the methyl group was moved to positions two and three of the ring (22 a, 22 b) or removed completely (22 c). A variety of substituents in 4position (22 d-i) was chosen to explore this position further before looking into substituents in ortho or meta position. To increase solubility, morpholine 22 j was synthesized starting by Boc-protection of 22 e (compound 23), subsequent Buchwald Hartwig coupling (compound 24) and final deprotection to amine 21 j that was then used for amide coupling. We also tested the effect of having two substituents on the Eastern ring (22 k and 22 l).
The synthesis of the Western part was done following the same procedure as for 1 with slight modifications (Scheme 2). Hydrazides (25 m-s) decomposed in concentrated hydrochloric acid and a 1 m aqueous solution had to be used instead. In addition, the thioamide precursors 26 m and 26 n were only found for triazoles 27 m and 27 n while all other reactions led to the cyclized compounds 27 o-s directly. This was unexpected in acidic solution and has not been reported before. If the thiadiazole is the intended isomer, the thioamide is normally treated with concentrated sulfuric acid. [17] To confirm that the correct cycle was formed and not its thiadiazole isomer 30, we compared the 1 H NMR and 13 C NMR chemical shifts of compounds 27 s and 30 (Table S4). The differences especially of the proton chemical shifts confirmed that the correct isomer was formed.
Next, ethyl-2-bromoacetate was used to install the desired ester and subsequent saponification resulted in the free acids (28 m, p-r). The amide coupling to the final product was performed as described above (29 m, p-r).
As an alternative strategy, we tried to perform the amide coupling between 20 and 2-chloroacetic acid first to form chloride 31 and then add different triazoles directly (Scheme 2). which reacts with ethyl-2-bromoacetate but in a low yield and the ester intermediate is difficult to purify. For the 4-methoxy derivative 27 n, the reaction also resulted in the desired final compound 29 n. In summary, the alternative route made it possible to circumvent saponification and more challenging purification steps and resulted in higher overall yield in fewer steps. Then, compound 29 n was mixed with BBr 3 to deprotect the hydroxyl group, but the compound decomposed.
One modification that would also allow for easy derivatization, is the replacement of the 1,2,4-triazole with a 1,2,3-triazole ring. The 1,2,3-triazole 35 can be synthesized by click-chemistry between an alkyne and an azide functional group, which reduces the number of steps from five to three. In the synthesis of 1, most reactions need stirring over night while the synthesis of 36 comprises two quick steps. The azide 33 is formed from 32 within two hours and the click reaction with 34 is also quick if 1.0 equivalent of copper sulfate is added from the beginning. Only the amide coupling needs up to 72 h.
Two compounds with different linkers between the amide and the Eastern ring system were synthesized and three derivatives without the Eastern ring following the same amide coupling reaction as before (Scheme 2). First, the sulfur atom was replaced with a nitrogen atom (compound 38 t). For this, amine 37 t was synthesized following a literature procedure. [21] Removing the aliphatic linker led to derivative 38 u with the aromatic ring directly connected to the amide. The whole Eastern part of the molecule was replaced by just a methyl group (compound 38 v). In an attempt to improve the Gramnegative activity, we introduced a primary amine by amide coupling between 17 and N-Boc-ethylenediamine (compound 38 w) and subsequent deprotection (compound 38 x). [22]

Activity of derivatives against DrDXPS
After the synthesis was completed, all compounds were tested for their inhibition of DrDXPS. In order to see if the acid precursors are also active, they were included in all biological assays. First, the solubility limit for all compounds was determined (Table S3) and based on the results, the assay was performed at 200 μm, 100 μm or 50 μm (Table 2). However, none of the derivatives are active (more than 50 % inhibition) at the respective highest soluble concentrations including compound 1 . Amides 22 d, 22 j, 38 u and acid 17 are the only compounds that show an inhibition higher than 30 % at a concentration of 200 μm. The standard deviations for these measurements are high and the data difficult to rely on and to interpret.
Overall, these results suggest that the compounds are not DXPS inhibitors. When looking at the three different activity tests of compound 1, we see low reproducibility. The first test of compound 1 suggested interesting inhibition (%inhi. = 39 � 16). When the IC 50 value of the resynthesized hit 1 was determined, 50 % inhibition was detected at 100 μm, which was still promising, but when testing all synthesized compounds including 1 at 100 μm again %inhibition of 1 was only 27 � 12. This could be explained by the fact that the solubility limit of 1 is around 100 μm, which can affect the assay results.

Discussion of VS results
Despite the recent progress in the areas of purchasable chemical space, scoring methods, and computational throughput, vHTS does not always manage to find bioactive molecules. In fact, bigger libraries pose the unique challenge that while increasing library size can provide better-fitting and higherscoring, rare molecules that rank artifactually well likewise get enriched amongst the top-scoring molecules. [23] A recent paper by Céron-Carrasco discusses the results of two separate virtual screening campaigns for inhibitors of the main protease (Mpro) of SARS-Cov-2, both of which failed to provide hits that were active enough to be further optimized. [24] Since the 1000 highest-ranking compounds from the previous campaigns covered relatively narrow score ranges, the author attempted to recover active molecules from these groups by rescoring the molecules with a Molecular Mechanics/Generalized Born Surface Area (MM/GBSA) approach. This method too provided limited success, finding only one compound with an IC 50 = 0.8 mm against Mpro. However, combinations of high-throughput/lower precision initial screens with more refined (but lower throughput) methods seems like a reasonable approach to improve hit recovery, and improvement of the above-mentioned MM/GBSA and related MM/PBSA method is still ongoing. [25] It should also be pointed out that even for smaller libraries where physical high-throughput screening is an option, both physical and virtual screening approaches can lead to false positives (as well as false-negatives). However, combination of the two methods can ultimately lead to hits that overcome the shortcomings of both approaches, creating higher confidence in the results. [26] The Atomwise AIMS projects intended to prospectively use and evaluate their proprietary AtomNet vHTS platform, and thus, we chose not to employ any complementary alternative prediction methods as additional filters. The selection of the final compounds purchased for testing uses a diversity-based clustering approach, to address the above-mentioned challenges that a number of top-scoring molecules are basically identical in their predicted score and may be enriched for problematic scaffolds, while also allowing to keep human interference to a minimum (i. e. no manual "hit-picking"). For DXPS this method proved to be insufficient and future virtual screening campaigns should take this into consideration.
DXPS is a difficult target to address virtually due to its large and flexible active site. However, during previous screenings, we have successfully used known binders to identify inhibitors against MtDXPS that bind to the ThDP-binding site. [15] Freel-Meyers et al. found several compounds binding covalently to ThDP and extending into the d-GAP binding site. [27] The knowledge of their binding mode could be taken into consideration for future screenings when looking for inhibitors addressing the d-GAP binding site to improve hit quality.

Biological data
With 30 derivatives at hand, we decided to not give up on this compound class but to continue our investigations. AMR is especially problematic in Gram-negative bacteria. Therefore, we wanted to check if our VS hits are active in E. coli as a Gramnegative model organism that utilizes the MEP pathway. The twelve compounds that performed best in the SPR and DrDXPS activity assay were tested against E. coli K12 and E. coli ΔTolC. The latter is a mutant of the former in which the efflux channel TolC is removed. [28] This mutant lacks the ability to pump the molecules out of the bacterial cells efficiently. Therefore, a compound that inhibits the growth of E. coli ΔTolC but not the wild-type strain K12 is probably eliminated by the efflux TolC system. With this knowledge good inhibitors are not excluded by accident but can be optimized to also inhibit E. coli K12.
From the twelve compounds, compound 1 has a MIC against E. coli ΔTolC of 28 μm and shows a 11 % growth inhibition at 50 μm of E. coli K12 (Table 2). It was not possible to determine a MIC for any of the other compounds, although seven show some activity against E. coli ΔTolC (Table S5) The linker derivatives are more diverse, but it can be seen that the sulfur atom is essential as compound 38 t (nitrogen instead of sulfur) loses activity against both strains in comparison to parent compound 1. Removing the Eastern aromatic ring leads to complete loss of activity (38 v-x). This was unexpected because we specifically installed a primary amine to improve Gram-negative activity. In 2017, Richter et al. coined the term eNTRy rules after analyzing the ability of over 180 diverse compounds to accumulate in E. coli. [22] They found that an ionizable, sterically non-hindered primary amine, a globularity (three-dimensionality) below 0.25 and fewer than five rotatable bonds help with accumulation in Gram-negative bacteria. Compound 38 x was designed to fulfill these criteria (primary amine, globularity = 0.063) except for the number of rotatable bonds which is six. However, from the analysis of the activities of the other compounds, the primary amine should have been installed at a different position as the removal of the Eastern rings system led to a loss of activity. Even if compound 38 x is now accumulating in the cell, it is not active anymore. This has to be taken into consideration when making more derivatives.
Surprisingly, compound 38 u with the aromatic ring directly attached to the amide has the highest inhibition of all derivatives against E. coli K12 and a moderate MIC of 58 μm against E. coli ΔTolC. Similar compounds have been reported to be New Delhi metallo-β-lactamase-1 (NDM-1) inhibitors (Figure 2). [29] NDM-1 is present in several bacteria such as E. coli and K. pneumoniae, and makes them resistant to β-lactam antibiotics. The only structural difference between the reported compound 39 and compound 38 u is the position of the methyl-group on the Eastern ring. The compounds in that study have only been tested on NDM-1 directly and their intrinsic inhibitory activity against bacteria has not been investigated yet. However, another very similar compound class (representative 40 as the closest structural analogue to 38 u) has been tested against a variety of clinically relevant strains and its activity was promising in E. coli, although the target has not been determined. [30] This could lead to the development of a dual inhibitor that can be used in combination with a β-lactam antibiotic. On the one hand, it can prevent the resistance against the antibiotic and on the other hand it kills the bacteria through a different mechanism of action. It might be interesting to investigate this class of compounds further by installing a primary amine at the methyl-substituent of the aromatic ring to see if accumulation can be improved.
Finally, the acid precursors have been tested but they have no interesting inhibitory activities.
We were also curious to see how the representative compounds 1, 22 e and 22 i performed in other clinically relevant pathogens. However, the inhibition against A. baumannii and two different strains of P. aeruginosa (Table S6) was below 15 % for all compounds making them selective E. coli inhibitors.
Compound 1 was also tested against human Hep G2 cells to determine its cytotoxicity. It does not inhibit cell growth at 25 μm, but at 100 μm, 49 � 1 % inhibition was measured. This value is acceptable as its MIC in E. coli ΔTolC is 28 μm. It can be assumed that the toxicity is comparable in all derivatives of the class, but this has to be confirmed if more diverse compounds are synthesized.
Lastly, 1 was also tested against M. tuberculosis H37Rv, but no inhibition was seen at 64 μm (data not shown).
In summary, the Western part of the molecule is essential for its activity in E. coli if the Eastern part contains a sulfur atom and a phenyl ring, preferably with a 4-CF 3 substituent. The 1,3,4-triazole cannot be replaced by a 1,2,3-triazole and the 2hydroxy-substituent on the Western ring can only be replaced with a NO 2 -group under loss of some activity. Derivatives of compound 38 u should be investigated further as potential dual inhibitors.

Conclusions
We have performed the first virtual screening on MtDXPS based on its recently solved truncated crystal structure. For the first time, we used a program based on a convolutional neural network to identify a set of structurally diverse, potential inhibitors of a MEP pathway enzyme. The initial results from SPR and on-target inhibition data were ambiguous, but we wanted to give the most promising compound a chance because its estimated IC 50 value was close to its solubility limit. We established a robust synthetic route that allowed for easy derivatization, and we synthesized 30 derivatives for further testing. Although, we improved the solubility in some of the derivatives, we were not able to improve the on-target activity against DXPS. Investigating antimicrobial activity of all derivatives revived our interest in the compound class as compound 1 showed a promising MIC of 28 μm against E. coli ΔTolC. Most of the Western-ring derivatives showed comparable MIC values, but most other structural changes were detrimental to the activity. Testing compound 1 in other pathogens only confirmed its selective inhibition of E. coli. Compound 38 u stood out by showing the best activity against E. coli K12 and a look into the literature revealed that it is a very close homologue of previously published β-lactamase NDM-1 inhibitors. We are now investigating compound 1 and its derivatives as potential dual inhibitors of NDM-1 and (an)other unknown target(s) in E. coli.
This long and difficult journey in the realm of drug discovery has not ended yet. We have learned a lot about the behavior of DXPS in different assay set-ups, how important it is to have a broad variety of assays available and that you should not give up on a compound class especially if you have established a robust synthetic route.

Experimental Section
Gene expression and protein purification of DXPS and IspC: Gene expression and protein purification of D. radiodurans, native and truncated M. tuberculosis DXPS and E. coli IspC followed previously reported protocols. [31,32] Binding affinity determination by SPR: The SPR binding studies were performed using a Reichert SR7500DC surface plasmon resonance spectrometer (Reichert Technologies, Depew, NY, USA), and medium density carboxymethyl dextran hydrogel CMD500 M sensor chips (XanTec Bioanalytics, Düsseldorf, Germany). Milli-Q water was used as the running buffer for immobilization. Truncated M. tuberculosis DXPS (ΔMtDXPS, 66.61 kDa) and native MtDXPS (67.88 kDa) were immobilized in one of the two flow cells according to reported amine-coupling protocols. [33] The other flow cell was left blank to serve as a reference. The system was initially primed with borate buffer 100 mm (pH 9.0), then the carboxymethyldextran matrix was activated by a 1 : 1 mixture of 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride (EDC·HCl) 100 mm and N-hydroxysuccinimide (NHS) 100 mm at a flow rate of 10 μL min À 1 for 7 min. The ΔMtDXPS or MtDXPS was diluted to a final concentration of 1.1 μm and 1.4 μm, respectively, in 10 mm sodium acetate buffer (pH 4.5), and was injected at a flow rate of 5 μL min À 1 for 10 min. Non-reacted surface was quenched by 1 m ethanolamine hydrochloride (pH 8.5) at a flow rate of 25 μL min À 1 for 3 min. A series of ten buffer injections was run initially on both reference and active surfaces to equilibrate the system resulting in a stable immobilization level of approximately 5,000 (9,000 for MtDXPS) μRIU ( Figure S1, Figure S2), respectively. 4-(2-Hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) buffer (50 mm HEPES, 150 mm NaCl, 0.05 % v/v Tween 20, 1 mm MgCl 2 , pH^8.0) containing 5 % v/v DMSO was used as the running buffer for binding studies. The running buffer was filtered and degassed prior to use. Binding experiments were performed at 20°C. Compounds dissolved in DMSO were diluted with the running buffer (final DMSO concentration of 5 % v/v) and were injected at a flow rate of 30 μL min À 1 . The association time was set to 60 s, and the dissociation phase was recorded for 120 s. Ethylene glycol 80 % in the running buffer was used for regeneration of the gold-chip surface. Differences in the bulk refractive index due to DMSO were corrected by a calibration curve (nine concentrations: 3-7 % v/v DMSO in the running buffer). Data processing and analysis were performed by Scrubber software (Version 2.0c, 2008, BioLogic Software). Sensorgrams were calculated by sequential subtractions of the corresponding curves obtained from the reference flow cell and the running buffer (blank). SPR responses are expressed in response units (RU).

DXPS-IspC coupled assay:
The experiments to determine the inhibitory activity of compounds in the DXPS-IspC-coupled assay have been carried out as reported previously with minor adjustments. [32] The assay was conducted in transparent, roundbottom 96-well plates (Greiner BioOne), and the absorbance was measured using a microplate reader (PHERAstar, BMG Labtech). The assay mixture contained 100 mm HEPES with a pH of 7.0, 2 mm βmercaptoethanol (BME), 100 mm NaCl, 0.5 mm ThDP, 1.0 mm MgCl 2 , 0.05 mm nicotinamide adenine dinucleotide phosphate (NADPH), 1.0 mm pyruvate, 1.0 mm d/l GAP and 1.0 μm EcIspC. The amount of DXPS used in the assays was determined experimentally by a dilution series of the enzyme. The concentration, which showed a reaction velocity in the range of À 0.1 to À 0.2 OD over a time range of 10 min was chosen for further experiments. The assay was prepared using two buffers, buffer A containing HEPES, NaCl, BME and all components of the reaction (except substrates) in 2x the final concentration and buffer B, consisting of only HEPES, NaCl, BME and the substrates, 2.0 mm pyruvate and 2.0 mm d/l-GAP. After the addition of 60 μL of buffer A to the plate with a pre-made inhibitor dilution series in 6 μL DMSO the reaction was started by the addition of 60 μL buffer B. The plate was centrifuged for 1 min at 2000 revolutions per minute (rpm) and 25°C to remove possible air bubbles. The plate was then immediately supplied to the microplate reader, and the absorbance measured at 340 nm using the mode slow kinetics with a cycle time of 30 seconds and 60 cycles at 25°C. Blank correction and linear fitting of the raw data were performed using the program Origin2021. The obtained initial velocities were converted to percent inhibition and plotted against the inhibitor concentration. The IC 50 values were determined by nonlinear curve fitting using Origin2021.

Kinetic-solubility determination:
The compounds were sequentially diluted in DMSO in a round-bottom 96-well plate. 3 μL of each well were transferred into another 96-well plate and mixed with 147 μL of phosphate-buffered saline (PBS) buffer. Plates were shaken at room temperature (r.t.) for 5 min at 600 rpm, and the absorbance at 620 nm was measured after 72 h. Absorbance values were normalized by blank subtraction and plotted using GraphPad Prism 8.4.2 (GraphPad Software, San Diego, CA, USA).
Antibacterial activity: Assays regarding the determination of the minimum inhibitory concentration (MIC) were performed as described previously. [34] The experiments were based on two E. coli strains/mutants (K12, ΔTolC) as well as P. aeruginosa (PA14, ΔmexA) and A. baumannii (DSM30007). In the case that no MIC value could be determined due to activity reasons, the percentage (%) of inhibition at 100 μM (or lower, depending on the solubility of the compounds) was determined. All samples were run in duplicate for each condition, and experiments were performed independently at least twice.
Determination of cytotoxicity of compound 1: HepG2 cells (2 x 105 cells per well) were seeded in 24-well, flat-bottomed plates.
Culturing of cells, incubations and OD measurements were performed as described previously with small modifications. [35] 24 h after seeding the cells, the incubation was started by the addition of compounds in a final DMSO concentration of 1 %. The metabolic activity of the living cell mass was determined after 48 h. Three independent measurements were performed. The IC 50 values were determined during logarithmic growth using GraphPad Prism software and data reported represent the mean � SD. Antitubercular activity: Compound 1 was tested for its Mtb growth inhibitory capacity in liquid culture as previously described. [36] Tests were performed in 2-fold serial dilutions starting at 64 μm in triplicates (2 × 10 6 mCherry-Mtb H37Rv bacteria, volume 100 μL). Bacterial growth was measured after 7 days of culture as described. Obtained values were normalized to solvent control (DMSO-treated bacteria set to 100 %).
Chemical synthesis: All procedures, characterization data and spectra are given in the Supporting Information.