The Development of a Pipeline for the Identification and Validation of Small-Molecule RelA Inhibitors for Use as Anti-Biofilm Drugs

Biofilm infections have no approved effective medical treatments and can only be disrupted via physical means. This means that any biofilm infection that is not addressable surgically can never be eliminated and can only be managed as a chronic disease. Therefore, there is an urgent need for the development of new classes of drugs that can target the metabolic mechanisms within biofilms which render them recalcitrant to traditional antibiotics. Persister cells within the biofilm structure may play a large role in the enhanced antibiotic recalcitrance of bacteria biofilms. Biofilm persister cells can be resistant to up to 1000 times the minimal inhibitory concentrations of many antibiotics, as compared to their planktonic envirovars; they are thought to be the prokaryotic equivalent of metazoan stem cells. Their metabolic resistance has been demonstrated to be an active process induced by the stringent response that is triggered by the ribosomally-associated enzyme RelA in response to amino acid starvation. This 84-kD pyrophosphokinase produces the “magic spot” alarmones, collectively called (p)ppGpp. These alarmones act by directly regulating transcription by binding to RNA polymerase. These transcriptional changes lead to a major shift in cellular function to both upregulate oxidative stress-combating enzymes and down regulate major cellular functions associated with growth and replication. These changes in gene expression produce the quiescent persister cells. In this work, we describe a hybrid in silico laboratory pipeline for identifying and validating small-molecule inhibitors of RelA for use in the combinatorial treatment of bacterial biofilms as re-potentiators of classical antibiotics.

conformation, RelA continually produces (p)ppGpp(21-23). During this time, the intracellular 48 concentrations of (p)ppGpp increase dramatically. The increased concentration of (p)ppGpp modulates 49 multiple downstream cellular signaling pathways including interacting with the RNA polymerase's 50 promoter binding region, thereby interfering with the cell's ability to produce additional ribosomes(24). 51 Currently, there are only a very limited number of inhibitors known for RelA and (p)ppGpp that have 52 been identified principally through traditional drug discovery methods, such as substrate analog design 53 and high-throughput compound screening, none of which are candidates for clinical trials for the control 54 of biofilm infections. The first of these inhibitors were analogs to ppGpp itself, such as Relacin and its 55 analogs(25). These compounds, while mildly effective, suffer from off-target effects and low binding 56 affinities (25)(26)(27)(28)(29). 57 The next compound discovered to reduce the intracellular concentrations of ppGpp was the cationic 58 peptide known as IDR1018. This peptide is an analog to bactenecin(30-33) and it was reported to directly 59 sequester and break down (p)ppGpp, thus lowering its intracellular concentration(30). It is now thought 60 that IDR1018 does not specifically target (p)ppGpp, but simply acts as an antimicrobial agent by means 61

Validation of the RelA Activity Assays 92
Several methods to study the RelA enzymatic activity in vitro and in vivo have been published (25, 26, 28, 93 44), and our methods were adapted from these sources. We performed two kinds of RelA activity tests: a 94 ppGpp-dependent fluorescent reporter in vivo assay and direct (p)ppGpp detection assays in vivo and in 95 vitro. The first method was based on the ability of ppGpp to affect expression of different genes(45). One 96 of these genes rpsJ, encodes the 30S ribosomal protein S10 (46,47). Its promoter, PrpsJ, belongs to the r-97 protein family of promoters, which are strongly inhibited by ppGpp and the DksA transcriptional 98 factors (48,49). Recently, a plasmid construct carrying a yfp (yellow fluorescent protein) gene driven by 99 the PrpsJ was published(50). The reporter plasmid contains the broad host range RK2 minimal replicon 100 and is compatible with many other plasmid vectors. Comparison of the yellow fluorescent protein (YFP) 101 activity between wild-type (WT) E. coli K12 and its relAmutant confirmed the effect of ppGpp 102 production on PrpsJ activity and served as validation of this method. 103 The direct (p)ppGpp detection in vivo and in vitro assays relied on different 32 P radioactive nucleotides (γ-104 32 P-ATP,α-32 P-GTP) for use as substrates, and thin-layer chromatography (TLC) to separate the reaction 105 products. Several methods were tested and optimized to give the best results for assessing the production 106 of (p)ppGpp. It was found that the in vitro buffer system did not need to be phosphate free as previously 107 indicated(51). It was also found that the concentration of magnesium needed to be above 5 mM for 108 optimal synthesis of (p)ppGpp. Previous work had indicated that the 70S ribosome was needed for RelA 109 to produce (p)ppGpp in vitro (52); however, we found this not to be the case. There was no difference 110 observed with 5 mM MgCl 2 with and without 70S (Fig. 2); therefore, it was not used in the in vitro 111 In the case of the in vitro assay, it was found that using γ-32 P-ATP was optimal to study the production of 116 both ppGpp and pppGpp, while α-32 P-GTP was optimal for studying only pppGpp. In the case of the in 117 vivo studies, [ 32 P]-orthophosphate was used as the radiation source, and the cells then incorporated the 32 P 118 into (p)ppGpp. Both methods required TLC with a stationary phase of a polyethyleneimine (PEI)-119 cellulose plate and a mobile phase of 1 M potassium phosphate monobasic. 120

Homology Studies 121
The active domain of the E. coli RelA cryo-EM (PDB: 5IQR) structure was determined using homology 122 studies (Fig. S1). Because there was no substrate bound to the RelA enzyme in the cryo-EM structure(36), 123 we utilized two methods to determine the active site for molecular docking. The first method was a 124 genomic-based homology method, where the known RelA protein sequences were compared, and the 125 conserved residues were evaluated (Fig. S1D). 126 The second method was a structural homology method in which we used crystallographic data obtained 127 from the S. aureus RSH-RelP that had been co-crystallized with its nucleotide substrates to identify both 128 the pre-and post-catalytic active sites(37). Alignment of the RelA and RelP predicted active site residues 129 showed that they are structurally highly similar; this allowed identification and characterization of the 130 active domain for targeting via ligand docking studies (Fig. S1A). Using this information, we were able to 131 determine two key amino acids involved in the binding of the first substrate in the catalytic process of 132

RelA Active Site Mutation Studies 134
To determine the accuracy of the in silico homology alignments and binding site determinations, two 135 amino acid residues were identified as key to the catalytic activity of RelA, and then tested in the 136 laboratory to ensure their assignment was correct (Fig. S1). Tyrosine residues Y-324 and Y-332 (from the 137 alignment) (Fig. S1D) had been determined to act as one of the largest contributors to the initial binding 138 of GDP or GTP(37). Y-324 was predicted to stabilize the phosphate of GDP/GTP by hydrogen bonding 139 though means of its hydroxyl group; and Y-332 was predicted to be involved in π-stacking with the 140 guanine's aromatic ring. These stabilizations were predicted to allow for the initial binding of GDP/GTP 141 within the active site. Y-324 and Y-332 residues correspond to the Y-310 and Y-319 residues of the E. 142 coli RelA enzyme. We hypothesized that if these residues were mutated to alanines (A-310 and A-319) 143 this should bring about a decrease in the catalytic transfer of the pyrophosphate from ATP to form 144 (p)ppGpp. Figure S2 shows the interactions of RelA with the native residues, as well as the lack of 145 interactions when mutated to an alanine residue. To obtain the Y/A-310 and Y/A-319 substitutions of the 146 E. coli RelA, we used two synthetic DNA cassettes to replace the 5' end of the gene in the pJW2755-AM 147 plasmid. The first PsiI/NsiI (1144bp) cassette contained a silent XbaI mutation(53, 54)and the Y/A-310 148 substitution. The second 365bp XbaI/NsiI cassette introduced the Y/A-319 mutation (Fig. S3). The 365-149 bp region between the XbaI (769) and NsiI (1144) restriction sites contain the predicted RelA active 150 center and can be easily exchanged with a synthetic construct to replace any of the tested amino acids. 151 Two assays were conducted to evaluate the activity of the mutant RelA enzymes: an in vivo (p)ppGpp 152 fluorescent reporter and in vitro (p)ppGpp production assay. The ASKA plasmid pJW2755-AM with the 153 WT RelA protein and its Y/A-310 and Y/A319 versions were transformed into the E. coli AG1 strain 154 containing a pAG001 plasmid, this plasmid contains a YFP gene expressed under a stringent response 155 regulated promoter PrpsJ(50). The E. coli AG1 strain contains a relA1 mutation caused by an insertion of 156 an IS2 insertion sequence between 85 th and 86 th codons of the relA gene. These mutants retain a low level 157 of (p)ppGpp synthesis activity (19). Plasmids pJW2755AM and its derivatives, and pAG001 belong to 158 different incompatibility groups and therefore can co-reside in a single cell. When plasmid encoded RelA 159 expression is induced with isopropyl β-D-1-thiogalactopyranoside (IPTG), the cells produced (p)ppGpp. Control is [γ-32P] ATP without enzyme. 174

In silico Screening for Hit Compounds 175
Non-RelA components of the E. coli RelA cryo-EM (PDB: 5IQR) model including RNA and ribosome 176 were stripped away from the file leaving only the RelA structure (Fig. S4). The RelA structure was then 177 optimized using the Schrödinger Maestro protein preparation tools including the package Prime, which 178 allows Maestro to fill in missing side chains and determine optimal amino acid orientations. The RelA 179 enzyme was then structurally minimized using the force field OPLS3e(55) (Fig. S4B). The enzyme 180 binding pocket was determined using homology ( Fig. S2) studies, as well as a general understanding of 181 RelA's function, and a docking grid box was developed for protein ligand docking calculations. 182 Schrödinger Maestro Molecular Modeling Glide(56, 57) was utilized to determine hit compounds, which 183 were then validated using the laboratory assays described above to probe their ability to inhibit RelA 184 activity. Schrödinger Glide-HTVS mode was first used to screen the entire University of California, San 185 Francisco Zinc 12 Database of commercially available compounds. This database contains over 4 million 186 compounds. The top 10% from the HTVS docking scan was then filtered into Glide-SP mode (standard 187 precision). This output was then further refined and run in Glide-XP(57) mode (extra precision). These 188 molecular docking studies resulted in 2 compounds showing a binding score that passed our threshold for 189 binding affinity (Table 1) and were higher than those of the natural substrates ATP and GTP. These two 190 compounds also fit both the Lipinski's rule of 5(58-60) for orally administered drugs, and the quantitative 191 estimate of drug likeness(61, 62). 192 Table 1. Hit compounds for the inhibition of RelA binding score. Binding score compared to the initial 193 binding compound GTP. 194 195

Effect of S3-G1A and S3-G1B on (p)ppGpp Production via In Vitro and In Vivo RelA Assays 196
After computational hit compounds were determined, the next step was to evaluate the effect of these 197 small molecules on RelA activity in the in vitro and in vivo assays established above for the production of 198 ppGpp. The results of the in vitro assay showed that both compounds S3-G1A (20 µM) and S3-G1B (20 199 µM) reduced the ppGpp production when compared to an untreated sample by 71.7% (p< 0.0001) and 200 79.7% (p< 0001), respectively (Fig. 4A). Both compounds showed higher reduction of activity than 201 Relacin (20 µM) (45.4%, p = 0.0084). The in vivo assay showed a reduction in ppGpp production in 202 samples treated with both compounds 31.4% (p = 0.0006) in S3-G1A and 17.75% (p = 0.0295) in S3-203 G1B. In this assay, no effect of Relacin on ppGpp production was observed (Fig 4B). We hypothesize 204 that Relacin is not cell permeable and therefore does not influence the in vivo ppGpp production. These

Effect of Hit Compounds on Bacterial Growth 214
Bacterial growth rate under conditions unrestricted by substrate availability is an indicator of cell health 215 and viability. Despite great efforts to determine the role of the stringent response on control of cell growth 216 rate, general conclusions have not been able to be drawn(63-65). However, all reports have shown that 217 mutants unable to produce ppGpp grow slightly more slowly (up to 30%) than their cognate WT on all 218 media tested(63-65). We found that the initial growth rates for the WT strain and CF1652 (relA::Km) 219 were the same (Fig. S5). However, growth of the WT strain started to slow down first after reaching 220 OD 600 =0.6. The WT strain was expected to sense small changes in nutrient concentrations and react to it, 221 reducing the growth rate. The relA mutant reached a higher cell density than that of the WT. After 18h of 222 growth, both strains reached their highest cell densities and thereafter we observed varying decreases in 223 OD 600 values. We found that compounds S3-G1A and S3-G1B had no effect on planktonic growth rate. 224 The maximal cell densities of the cultures with compounds were slightly lower than the control (Fig. S6). 225

Effect of Hit Compounds on Biofilm Inhibition and Dispersal 226
We have previously reported that E. coli strain C(43) is the only one of the five major "laboratory strains" 227 of E. coli that is a superior biofilm former; therefore, this strain was used in our biofilm assays. Studies 228 were conducted in 96-well high-throughput assays. In the biofilm inhibition assay, compounds were 229 added to the wells at the beginning of the experiment. For the biofilm dispersal assay, the biofilm was 230 allowed to grow for 24 and 48h, the wells were washed with sterile phosphate-buffered saline (PBS), and 231 fresh medium supplemented with the compounds was added to the wells. The amount of biofilm was 232 measured after 24h. There was no observed effect on the inhibition (Fig. S7) or dispersal (data not shown) 233 of biofilms with compounds alone. 234

Effect of Compound on Biofilm Persistence and Biofilm Viability 235
Biofilm persistence and viability were assessed with the hit compounds in combination with an antibiotic. 236 It has been determined that sub-MICs of antibiotic result in increased biofilm formation(66). Ampicillin 237 was used in all of our assays due to its bactericidal effect. Sub-MIC concentrations of ampicillin were 238 determined by growth measurements (OD 600 ). We found that the biggest change in the culture cell density 239 was observed between 40 and 60 µg·mL -1 ampicillin (Fig. S8A). Analyzing the effect of ampicillin on 240 biofilm formation, we observed that the presence of the antibiotic significantly increased the amount of 241 biofilm with the highest biomass observed at relatively high ampicillin concentrations (80 µg·mL -1 ) (Fig.  242 S8B). To analyze the effect of our hit compounds in combination with antibiotics, a range of ampicillin 243 concentrations from 30 to 50 µg·mL -1 was utilized. 244 The amount of biofilm biomass was determined in the combined presence of antibiotics and either 245 compound A & B. This combination therapy led to a highly significant reduction in biofilm mass 246 compared to the ampicillin-only-treated control (Fig. 5A). As a reference control, we used IDR 1018, an 247 antimicrobial peptide that was reported to target (p)ppGpp directly and degrade ppGpp in vitro (30). 248 Addition of the hit compounds to ampicillin concentrations of 40 µg·mL -1 (Amp40) and 50 µg·mL -1 249 (Amp50) resulted in a highly significant decreases in biofilm volume compared with their cognate 250 antibiotic only control (Fig. 5A). At Amp40 the biofilm biomass was reduced by 97.9% (p = 0.0009) for 251 S3-G1A (50 µM), by 92.4% (p = 0.0014) for S3-G1B (50 µM), and by 75.4% (p = 0.006) for IDR1018 (6 252 µM). Amp50 showed reductions in biofilm biomass of 67.9% (p = 0.0044) for S3-G1A, of 72.9% (p = 253 0.0042) for S3-G1B, and 65.2% (p = 0.0054) for IDR1018The difference between Amp40 and Amp50 254 can be attributed to the larger increase in biofilm volume from the higher concentration of antibiotic. 255 An AlamarBlue cell viability assay also showed that ampicillin killed more bacterial cells in combination 256 with the tested hit compounds (Fig. 5B). In the case of ampicillin at 30 µg·mL -1 (Amp30), the reduction 257 was 55.4% (p = 0.0024) and 54.2% (p = 0.0027) for S3-G1A (50 µM) and S3-61B (50 µM), respectively. 258 When higher concentrations of antibiotic were used, the synergetic effects of compounds S3-G1A and S3-259 G1B were less noticeable, with the decreases being only 29.2% (p = 0.0278) and 6.5% (p = 0.6), 260 respectively. This effect we attributed to the greater volume of the biofilm contained in these samples 261 (Fig. 5B).

Effect of Hit Compounds on Biofilm Structure 270
Scanning electron microscopy (SEM) allowed us to probe the structure of the biofilms treated with the hit 271 compounds. Biofilms were grown on metal pins for 3 days that were transferred daily to fresh LB 272 medium using the JEKMag technique(67). We found that while there was not a large reduction in biofilm 273 mass by the compounds alone, there was a very substantial change to the structure of the extracellular 274 matrix of the biofilms. Biofilms treated with compounds 40 µg·mL -1 S3-G1A and 40 µg·mL -1 S3-G1B 275 exhibited a greatly reduced amount of matrix compared to untreated WT E. coli C (

Computational Docking 304
High-throughput in Silico Docking Studies: The RelA enzyme (PDB: 5IQR) was prepared and optimized 305 using Maestro Protein Preparation (Schrödinger Maestro, New York, NY, USA; Version 11.9.011, 306 MMshare Version 4.5.011, Release 2019-1, Platform Windows-x64). The 5IQR PDB file contained 307 extraneous portions of the ribosome, as the structure was determined as a RelA dimer with the ribosome. 308 The ribosome and RNA subunits were removed and RelA was isolated in a separate file. The dockable 309 RelA structure was prepared and minimized using Schrödinger's protein preparation application(69). This 310 application was utilized to add hydrogens, create missing disulfide bonds, and determine lowest-energy 311 residue orientations. Geometry minimization was carried out using the force field OPLS3e (55) contains the RelA active center and can be easily swapped with a synthetic construct to replace any of the 323 tested amino acids. This method was applied to introduce the Y/A-319 mutation. A 365-bp XbaI/NsiI 324 fragment was replaced by a synthetic fragment with TAT-Y319 (position 1053) replaced with GCC-A319 325 codon as described(70). All mutations were confirmed by Sanger DNA sequencing. 326 RelA Protein Purification: the functional RelA enzyme and its Y/A-319 and Y/A-310 mutants were 327 purified from host cell AG1 strains carrying the pJW2755-AM, pJEK2020-43, and pJEK2020-20 328 plasmids, respectively. One liter of LB broth was inoculated with 20 mL of overnight culture (OD 600 = 329 0.9) and grown for 4 h (OD 600 = 0.8) before induction with 1.5 mM IPTG for 4 h. Cultures were spun 330 down, washed with phosphate-buffered saline (PBS), and resuspended in lysis buffer (50 mM NaH 2 PO 4 , 331 300 mM NaCl, 10 mM imidazole, pH 8.0) for lysing. To that resuspension, 1 µL·mL -1 ThermoFisher 332 Halt™ Protease Inhibitor Cocktail (100X) was added without EDTA and cells were lysed with sonication 333 on ice (cycles 10 s on 10 s off for a total of 3 min of sonication, 2X). Lysates were spun down to remove 334 cellular debris. Millipore Sigma PureProteome™ Nickel Magnetic Beads were used according to 335 modified manufacturer's instructions. Supernatant was placed in 200 µL of nickel affinity beads for a 336 period of 30 mins. Beads were captured on a magnetic rack and the supernatant was removed. Beads were 337 then washed 4X with wash buffer (50 mM NaH 2 PO 4 , 300 mM NaCl, 10 mM imidazole, pH 8.0). RelA 338 was eluted twice using 300 mM imidazole elution buffer (50 mM NaH 2 PO 4 , 300 mM NaCl, 300 mM 339 imidazole, pH 8.0) and a final elution using 500 mM imidazole elution buffer (50 mM NaH 2 PO 4 , 300 mM 340 NaCl, 500 mM imidazole, pH 8.0). An SDS-page gel was run to confirm presence and purity of RelA. Reader with a programmed growth cycle (18 h, 37 °C, orbital rotation 3.5). Cell density was measured at 356 OD 600 and YFP fluorescence activity was detected with 505 nm/535 nm (excitation/emission). Enzymatic 357 activity was measured as Relative Fluorescence Units (RFU -YFP/OD 600 ). 358 In vitro (p)ppGpp quantification: In vitro (p)ppGpp quantification was carried out using techniques 359 similar to those previously reported in the literature (25,26,28,44). RelA enzyme was purified as 360 described above. Roughly 0.4 µg of RelA protein was added to a 1.5-mL microcentrifuge tube containing 361 a reaction mix composed of 1X PBS, 5 mM MgCl 2 , 0.5 mM ATP, 0.5 mM GTP, 0.5 mM GDP, and 20 362 μCi[γ-32P]ATP (3,000 Ci mmol−1; PerkinElmer) and varying concentrations of the compound of 363 interest. These reactions were incubated at 37°C for 1 h. The reactions were stopped by addition of 5 μL 364 formic acid (88%). The reaction mixtures were then spotted on a stationary-phase polyethyleneimine 365 (PEI)-cellulose TLC plate using potassium phosphate monobasic (1.5 M) as the mobile-phase. The plates 366 were then dried, and the radiation levels were read using a Molecular Dynamics Storage Phosphor Screen. 367 A Molecular Dynamics Storm 840 Phosphor imager Scanner was used to read the phosphor screen and 368 ImageJ was used to process the images. 369 In vivo (p)ppGpp Quantification: In vivo (p)ppGpp quantification was carried out using techniques similar 370 to those previously reported in the literature (25,26,28,44). One milliliter of overnight cell culture of E. 371 coli C was placed in 1.5-mL microcentrifuge tubes and pelleted. To this pellet was added 50 μL of a 372 reaction mixture containing 20 μCi orthophosphoric acid and 40 μM serine hydroxamate in 1X MOPS 373 minimal medium. The cell pellet was resuspended by gentle vortexing and placed in an incubator for 1 h. 374 Cell growth arrest and cell lysis were completed by addition of 15 μL formic acid (88%). The lysate was 375 then centrifuged to remove any insoluble components and the supernatant was spotted on a stationary-376 phase PEI-cellulose TLC plate. Plates were processed and analyzed as described above. culture; approximately 10 7 cells) in fresh LB medium was added to each well. These were allowed to 380 grow for 24 h. The planktonic cells and medium were then aspirated, and the plates were washed twice 381 with 1X PBS. Fresh LB with hit compounds were added to the biofilm wells. These cultures were then 382 allowed to incubate at 37 °C overnight. Cell density were measured (OD 600 ) using a Multiscan Go plate 383 reader (ThermoFisher), and 30 μL Gram crystal violet (CV) (Remel; 3 g crystal violet, 50 mL 384 isopropanol, 50 mL ethanol, 900 mL purified water) was applied for staining for 1 h. Plates were washed 385 with water and air dried, and CV was solubilized with an ethanol:acetone (4:1) solution. The OD 570 was 386 determined from this solution, and the biofilm volume was calculated as the ratio of OD 570 to OD 600 (43, 387

72). 388
Biofilm Inhibition Assays: For biofilm formation on polystyrene surfaces, flat-bottom 96-well microtiter 389 plates (Corning Inc.) were used. The effect of different compounds on biofilm formation was tested by 390 adding compounds at different concentrations to the bacterial culture (100X diluted overnight culture; 391 approximately 10 7 cells) in fresh LB medium. Two hundred microliter aliquots were pipetted into 96-well 392 plates and placed for 24 or 48 h into a 37 °C incubator. The biofilm mass was measured by the CV 393 staining assay described above. 394 Biofilm Persistence Assays with Ampicillin: Biofilms were grown for 24 or 48 h as described above. 395 Planktonic cells were removed, and the biofilms were washed twice with 250 µL sterile PBS solution. 396 Two hundred microliters of fresh LB medium with various concentrations of ampicillin were dispensed 397 into the wells. After 18h of incubation at 37 °C, the volume of biofilm was measured by CV staining as 398 described above. 399 Synergistic Effects of Compounds and Antibiotics: Biofilms were grown for 24, 48, or 72 h as described 400 above. Planktonic cells were then removed, and biofilms were washed twice with 250 µL sterile PBS 401 solution. Two hundred microliter aliquots of fresh LB medium with multiple concentrations of the tested 402 compounds and ampicillin were dispensed into the wells. After 18h of incubation at 37 °C, the biofilm 403 mass was measured as described above. For the AlamarBlue viability test, 4 μL of AlamarBlue 404 (Invitrogen) was added and plates were incubated in a Biotek HT plate reader at 37 °C for 4 h. Cell 405 viability was measured as fluorescence at 530/590nm (excitation/emission) versus compound 406 concentration or initial cell density 407 Cell Growth Curves: The effect of the hit compounds on bacterial growth was tested by adding 408 compounds at multiple concentrations to the bacterial culture (100X diluted overnight culture; 409 approximately 10 7 cells) in fresh LB medium. Two hundred microliters aliquots were pipetted into 96-410 well plates and placed into a Biotek HT or Tecan Infinite M200 Pro plate reader for 18h at 37 °C. Plates 411 were shaken during incubation and the optical density (OD 630 or OD 600 ) was measured every 15 min. 412 Antibiotic Susceptibility Assays: For liquid cultures, the minimal inhibitory concentrations (MICs) of the 413 antimicrobial drugs were determined using 96-well plates and the broth dilution method. Suspensions 414 were then incubated at 37°C for 18 h in a Biotek HT plate reader (see bacterial growth). Biofilm 415 destruction experiments were performed with different antibiotic concentrations, and cell densities were 416 measured after 18h. Bacterial concentrations were calculated via optical density (OD 630 ), and the lowest 417 concentration causing 80% growth inhibition relative to the growth of the control was deemed to be the 418

MIC. 419
Scanning Electron Microscopy (SEM) of Biofilm: E. coli biofilms were grown in LB with multiple 420 concentrations of the hit compounds on metal pins(67). These metal pins were then washed twice in 1X 421 PBS. The biofilm-containing metal pins were then placed in a 5% glutaraldehyde solution for 1 h. Metal 422 pins were then dried using a gradient of ethanol from 50% to 100%, 5 min in each solution. The pins were 423 sputter coated with gold at a thickness of 60 Å. SEM images were taken on a Zeiss Supra 50VP Scanning 424 Electron Microscope 5 kV beam acceleration. 425 Statistical Analysis: Statistical analyses were performed using OriginPro 8.5. Relevant statistical data is 426 included in results and discussion for each experiment. Error bars indicated standard deviation from the 427 mean. Asterisks represent statistical significance of at least p < 0.05. 428