Directed Evolution of the Nonribosomal Peptide Synthetase AdmK Generates New Andrimid Derivatives in vivo

Summary

Many lead compounds in the search for new drugs derive from peptides and polyketides whose similar biosynthetic enzymes have been difficult to engineer for production of new derivatives. Problems with generating multiple analogs in a single experiment along with lack of high-throughput methods for structure-based screening have slowed progress in this area. Here, we use directed evolution and a multiplexed assay to screen a library of >14,000 members to generate three new derivatives of the antibacterial compound, andrimid. Another limiting factor in reengineering these mega-enzymes of secondary metabolism has been that commonly used hosts such as Escherichia coli often give lower product titers, so our reengineering was performed in the native producer, Pantoea agglomerans. This integrated “in vivo” approach can be extended to larger enzymes to create analogues of natural products for bioactivity testing.

Introduction

Nonribosomal peptides are a proven source of antibacterial (Recktenwald, et al., 2002), antifungal (Denning, 1997) and anticancer drugs (Du, et al., 2000). These peptides often contain non-proteogenic amino acids as well as fatty acid, polyketide, carbohydrate, and isoprenoid appendages. The enzymes that build these modified peptides, nonribosomal peptide synthetases (NRPSs), are often large enzymes with multiple active sites and are characterized by keeping intermediates covalently tethered by a series of phosphopantetheine (PPant) linked thioester bonds at the thiolation (T) domains during biosynthesis. The gatekeepers of these molecular “assembly lines” are the adenylation (A) domains which select an amino acid building block from the pool available within producing cells. The modular structure of NRPSs make them attractive targets for protein engineering, but realization of this goal over the last few decades has been far more difficult than many had hoped. Here we screen a library of ~14,330 clones for derivatives of the antibiotic andrimid (1). Focusing on the site of valine incorporation, three new chemical derivatives have been created with similar or improved biological activity.

Even though amino acid substrates for adenylation domains may be predicted with some confidence via the “NRPS Code” (Challis, et al., 2000, Rausch, et al., 2005, Stachelhaus, et al., 1999), engineering of these domains to accept non-cognate substrates has been elusive. Often referred to as combinatorial biosynthesis (Khosla, et al., 1999, Khosla and Zawada, 1996, Tsoi and Khosla, 1995, Walsh, 2002, Walsh, et al., 2003), a main approach has employed chimeric synthetases where an adenylation domain of one NRPS is replaced by an adenylation domain from another NRPS which activates a different substrate. This so-called “domain swapping” strategy has proven successful in terms of generating the desired product, but has been plagued by decreased product yields and outright dysfunction presumably due to interruption of protein-protein interactions caused by this cut and paste approach (Baltz, et al., 2006, Doekel and Marahiel, 2000, Mootz, et al., 2000, Mootz, et al., 2002, Nguyen, et al., 2006, Schneider, et al., 1998, Stachelhaus and Marahiel, 1995, Stachelhaus and Marahiel, 1995). The activity of chimeric synthetases could be restored in the first reported use of directed evolution for reengineering NRPS systems (Fischbach, et al., 2007). Domain swapping, however, is still not ideal for true combinatorial biosynthesis in that at most one variant of an NRP can be generated per swap. Here we present the second example of using directed evolution to alter thiotemplated biosynthetic systems. Instead of rescuing activity of a chimeric synthetase using error-prone PCR, we directly targeted active site residues for evolution using saturation mutagenesis (Chockalingam, et al., 2005). This orthogonal approach to NRPS reengineering focuses on modifying the sites of substrate specificity, rather than modifying the protein-protein interactions of the A- and T-domains as in a previous study (Fischbach, et al., 2007). By targeting mutations to the substrate binding site and generating hundreds of enzyme variants in parallel, we created a large library of mutants, each with the potential to generate a natural product derivative. Further, we used mass spectrometry as a structure-based readout for the library. Ultimately, four mutants that produce robust levels of new derivatives were found. The workflow for this study is illustrated in Figure 1.

Figure 1. Work Flow for Directed Evolution of AdmK.

After deletion of admK from the P. agglomerans chromosome and transformation with a library of admK mutants, the library is screened by high-throughput LC-MS/MS. New compounds are identified and assayed for bioactivity. See also Figure S1.

The system under study here is the andrimid biosynthetic pathway (Figure 2). Andrimid (1) is a hybrid NRP/PK molecule produced by P. agglomerans (Fredenhagen, et al., 1987, Jin, et al., 2006, Singh, et al., 1997). Andrimid is a broad spectrum antibiotic and acts by inhibiting fatty acid biosynthesis at the acetyl CoA carboxylase step, preventing formation of malonyl CoA, the precursor for fatty acids and some polyketides (Freiberg, et al., 2004). Previous structure-activity relationship (SAR) studies identified a portion of andrimid, the valine subunit, that could be exchanged to create more potent and specific compounds (Freiberg, et al., 2006). Domain swapping experiments on the corresponding NRPS gene, admK, showed that the valine subunit could be substituted for isoleucine or aminobutyrate and that there was no apparent editing by downstream condensing enzymes (Belshaw, et al., 1999), allowing generation of andrimid derivatives in vivo (Fischbach, et al., 2007).

Figure 2. Andrimid Biosynthetic Pathway.

Andrimid is biosynthesized by a highly disconnected and hybrid NRPS/PKS pathway. Adenylation domain substrate specificity is indicated by single letter amino acid abbreviation subscript. AdmK incorporates the valyl subunit (highlighted) of andrimid and has been targeted for mutagenesis. KS- ketosynthase, CLF- chain length factor, DH-dehydrogenase, KR- ketoreductase, T-thiolation, TG-transglutaminase, A-adenylation, Mut- aminomutase, C- condensation, TE-thioesterase.

Results

Library Design and Construction

We sought to express a library of mutant enzymes and monitor the effectiveness of the mutations by screening for end products produced by the pathway in vivo. To accomplish this, removal of the wild type (WT) enzyme activity was necessary. We employed an allele replacement strategy based on homologous recombination that has proven successful in a number of organisms, namely the streptomycin resistance counterselection (Russell and Dahlquist, 1989). Using this strategy, we made a scarless deletion of the WT admK gene in P. agglomerans. The resulting strain served as a background strain for expression of an admK mutant library (Figure S1A). This result was validated by PCR (Figure S1B). Also, we found that the E. coli expression plasmid pQE60 was capable of supplying admK and restoring andrimid production levels comparable to WT in P. agglomerans (Figure S1C).

In order to select residues in admK for mutagenesis, we turned to sequence alignments of AdmK with A-domains known to activate substrates different than the valine found in andrimid (Figure S3). We chose A-domains that activate nonpolar amino acids because SAR studies (Freiberg, et al., 2006) showed that this type of residue could substitute for valine and increase the potency and selectivity of the andrimid scaffold. From the sequence alignment with the ten residues used for substrate prediction of NRPS A-domains (Stachelhaus, et al., 1999), we chose the three most highly variant of the ten for saturation mutagenesis (Miyazaki and Arnold, 1999) (Figure 3). Our three-site saturation mutagenesis library was designed to create 1,404 distinct mutants, and was limited by selection of codons for synthetase residues likely to select nonpolar substrates (Figure 3) for incorporation into the andrimid backbone. The mutant library was constructed by utilizing sequential megaprimer and overlap extension PCRs (Horton, 1997, Sarkar and Sommer, 1990). In order to make a library that maximized the amino acid substitutions we desired, while minimizing unwanted substitutions truncations, PCR primers with limited degeneracy were used. Position 240 of AdmK was mutagenized using a WHK codon in place of the native TGG codon substituting the native Trp with Met, Thr, Ser, Tyr, Leu, Phe, Ile, Asn and Lys and only one possible stop codon. For position 265 of AdmK, a WHW codon replaced the native ATC codon thereby substituting Leu, Phe, Thr, Tyr, Asn, Lys and one possible stop codon for Ile. Position 291 of AdmK was mutagenized by using an RHK codon in place of the WT GGT codon in order to substitute Val, Ile, Asn, Ala, Glu, Met, Thr, Lys and Asp for Gly with no possibility for a truncation at this position. Together, these three degenerate codons allow for 1,404 codon combinations. To insure 95% coverage of 1,404 mutants, we screened 14,330 clones. Statistical analysis using the program GLUE (Patrick, et al., 2003) estimated that the library was 99.99% complete with a 94.9% probability that the library contained all possible variants.

Figure 3. Extracted Nonribosomal Code Residues.

Nonribosomal code residues extracted from adenylation domains from GrsA, AdmK and other systems with consensus sequences for nonpolar amino acids (shown in parentheses). Strictly conserved residues are highlighted in blue, highly conserved residues are highlighted in green and highly variant residues targeted for limited saturation mutagenesis are highlighted in red. See also Figure S3.

Assay Development and Library Screening

To screen our library of andrimid producing clones, we employed a structure based assay. We reasoned that using an antibiotic bioassay we would not be able to distinguish between andrimid and derivatives of andrimid. Also, we would be incapable of identifying which precise andrimid derivative had been produced because most derivatives would likely be active antibiotics (Freiberg, et al., 2006). During evaluation, mass spectrometric detection of andrimid showed ~23,000 fold greater sensitivity than a bioassay, and we noticed a fragmentation pattern that cleanly dissected the molecule into regions to be conserved or variant (Figure 1 lower right panel). This calculation of relative sensitivity for MS vs. classic halo assays is based on the limit of detection (LOD) during MS of andrimid, which is 20 fmol. Given that 96 samples are mixed per LC-MS injection, this translates to a LOD of 1.9 pmol for the pooled assay. The MIC data for andrimid against E. coli shows that 0.47 nmol can be detected in a 10 μL spot in the agar overlay assay. If this assay were diluted 96 fold as in the LC-MS assay this translates to a 45.1 nmol LOD. These attributes of the MS-based screen allowed us to identify which andrimid derivative was produced and conduct the screening of 96 clones in a single LC-MS run. Single clones arrayed into 96-well plates, were inoculated into 2xYT microbial medium supplemented only with antibiotic and grown for 24 hours prior to pooled sampling of each 96-well plate. Each pooled sample representing 96 individual clones constituted one LC-MS/MS injection. In this way, over 14,330 clones from 150 individual 96-well plates of the three-site library were screened and four clones producing andrimid derivatives were isolated. Twenty-four percent of assay plates (i.e., 96 samples combined into one) still showed production of andrimid while 41% of plates showed no production beyond the octatrienyl-β-phenylalanine precursor to andrimid (a truncated product, 2). Curiously, almost all of the plates assayed showed production of a species at m/z 433.21 that was determined to be octatrienyl-β-phenylalanyl-β-phenylalanine (3), a result of apparent “stuttering” by the transglutaminase AdmF, the enzyme responsible for incorporating β-phenylalanine into the growing ketide-peptide chain.

Characterization of Mutants and Derivative Compounds

The four mutant clones were identified subsequently (by pooled -row and -column searching of assay plates), sequenced and designated A2 (W240L, G291V), A7 (W240S, I265T, G291E), A9 (W240T, I265T, G291E) and B12 (R235K, W240L, I265L, G291E). Together, these clones produced three new andrimid derivatives and another that has been described previously (Figure 4 and Figure S4A i-iv). Clones A2 and B12 were found to produce an andrimid related peak at m/z 494.26, corresponding to either isoleucine- or leucine-substituted andrimid (4, 5) (Figure S4A i, ii). Clones A7 and A9 produced a mixture of alanine- and phenylalanine-substituted andrimid (6, 7) (Figure S4A iii, iv). In order to resolve the L-Ile vs. L-Leu ambiguity for clones A2 and B12, an isotope feeding study was conducted. Growing clones A2 and B12 in media supplemented with a 50:50 mixture of ¹³C₂-Leu and ¹³C₆-Ile demonstrated that both clones A2 and B12 have preference for Ile over Leu, but at different levels of discrimination (Figure S4B). Clone A2 showed a 4:1 preference for Ile to Leu, whereas clone B12 showed a 20:1 preference for Ile to Leu. Control experiments where only one isotope labeled amino acid was supplemented clearly showed that when one amino acid is supplied in 50 mM excess, incorporation of that amino acid is efficient (Figure S4C). For that reason, we employed a mutasynthesis strategy where the desired amino acid to be incorporated was supplied in excess for scale up of fermentations in order to determine yield and bioactivity. We found that applying this mutasynthesis strategy also improved the yield for clones A7 and A9. Compounds 1, 4, 5 and 7 were all found to be bioactive in disk diffusion assays (Figure 5). When using the mutasynthesis, we were able to significantly improve the yields, in some cases approaching WT levels (Table 1 and Figure S5). MICs were determined for several representative bacterial strains (Table 2).

Figure 4. Structures of Andrimid and Derivatives Produced in This Study.

1 andrimid, 2 octatrienyl-β-phenylalnine, 3 octatrienyl-β-phenylalanyl-β-phenylalanine, 4 isoleucine-andrimid, 5 leucine-andrimid, 6 alanine-andrimid, 7 phenylalanine-andrimid. See also Figure S4.

Figure 5. Andrimid and Derivatives Bioassay.

Agar-overlay bioassay of andrimid and analogues generated against E. coli imp ASR. Aliquots of each compound (numbered) and a solvent negative control (−) were spotted onto paper disks atop a layer of soft agar seeded with E. coli.

Table 1.

Mutations in admK and corresponding products compared with WT production levels

Clone	Mutations	Compound Produced	Medium	Production per Liter (mg)
WT	N/A	1	2xYT	378.16
A2	W240L, G291V	4	2xYT + Ile	4.94
		5	2xYT + Leu	0.99
B12	R235K, W240L, I265L, G291E	4	2xYT + Ile	466.39
		5	2xYT + Leu	93.83
A7	W240S, I265T, G291E	6	2xYT + Ala	0.19
		7	2xYT + Phe	0.36
A9	W240T, I265T, G291E	6	2xYT + Ala	0.20
		7	2xYT + Phe	0.49

Table 2.

MIC data for andrimid and derivatives against representative Gram negative and Gram positive bacterial strains

	MIC (nmol)
	1	4	5
E. coli imp ASR	0.47	2.72	0.73
K. pneumoniae ATCC 13883	9.05	2.93	NT*
S. aureus ATCC 25923	7.71	0.39	0.28
E. faecalis ATCC 19433	36.35	42.76	NT*

^*Not Tested

Discussion

This work takes on the challenges associated with engineering enzymes of secondary metabolism with a technological approach that applies generally to metabolites detectable by tandem mass spectrometry. When combined with high level expression in the native host, P. agglomerans, the reduced metabolic production from active site mutation of admK (largely from poor solubility of variants), did not prevent four successes from being found in a library pool where ~80% were unaffected, ~20% were dead, and ~0.03% had altered production.

Creating a mutant library of admK rather than performing several domain swapping experiments allowed us to generate four andrimid derivatives in the same overall experiment while examining many solutions for each derivative simultaneously. The key aspect of our work that allowed for this combinatorial biosynthesis experiment is the structure-based assay that provided a direct readout of not just the AdmK variants’ activity, but also the combined output of downstream enzymes in the assembly line. Unlike many indirect assays of enzyme activity (bioactivity or coupled assays), direct structural readout of mutant activity in the integrated pathway tightly connects the overall impact of the mutagenesis with a main goal of the exercise – production of new derivatives.

We found four mutants that exhibited altered substrate specificity, A2, A7, A9, and B12. All showed a broadening of substrate specificity relative to the WT, with increased preference to substrates other than the cognate substrate, valine. To our surprise, the overall best performing mutant was actually a serendipitous quadruple mutant. The fourth mutation in B12, R235K is likely an artifact of PCR; however, it is unknown if this stray mutation is required for robust production of the Ile/Leu analogs. Structures for NRPS A-domains available exhibit low sequence identity to AdmK, 25% for 1AMU and 24% for 2VSQ. However, based on alignment of AdmK to the structures above, R235 corresponds to N697 of 2VSQ and E272 of 1AMU. These positions in the respective crystal structures are remote from the active site and on the surface of the protein. We can only speculate that R235K may change a protein-protein interaction surface or induce a structural change to the active site through an outer sphere effect. The relatively higher activity and solubility of the B12 mutant over the others suggests that mutations outside the selectivity conferring code should be considered for future A-domain mutant libraries. A second round of mutagenesis and screening was performed on the four isolated clones using an error-prone PCR approach in order to increase the activity of the first round mutants, but no improvement was observed. MIC values for the derivative compounds 4 and 5 showed increased bioactivity towards Staphylococcus aureus relative to 1, and unchanged or decreased bioactivity towards E. coli consistent with previous SAR studies (Freiberg, et al., 2006) (Table 2). Prediction of A-domain substrates is fairly robust (Rausch, et al., 2005). Despite this, prescription of what active site mutations to make for a particular metabolite structure is far less developed. This study provides a large scale survey of what amino acid substitutions could be tolerated in the A-domain active site. While many of the triple mutants that still yielded andrimid were not sequenced here, one can envisage a large scale study specifically aimed at determining empirically what sites are stable to mutation. This would effectively map functional outcomes for many positions and sharply increase our predictive ability for creation of new NRPS derivatives.

One reason the andrimid system was chosen is because its biosynthetic enzymes are mostly ≪100 kDa (i.e., relatively small for NRPS or PKS proteins). There is no reason why our methods cannot be extended to larger multi-module NRPSs, provided the host is genetically tractable, or the pathway is amenable to heterologous expression with robust yields. We envision that the general approach used here could be extended to other domains within NRPS systems to probe substrate specificity of condensation domains. We estimate that a ~100 fold improvement in screening speed could be achieved, leading to an ability to screen 4- or even 5-site mutant libraries with similar coverage within the same approximate scope of time and resources used in this work.

Peptide natural products from NRPS systems draw high interest because of their medical uses, but harnessing their rather predictable correlation between enzyme primary structure and metabolite structure has proven elusive over the past two decades (Khosla, et al., 1999). Reengineering strategies to achieve combinatorial biosynthesis (Walsh, 2002) compete conceptually with organic synthesis as a strategy for enhancing the diversity of natural molecular scaffolds. Challenges to attaining the goal of true combinatorial biosynthesis include not only the general challenges of enzyme engineering, but also the methods used to evaluate the modified enzymes and pathways. The field of NRPS reengineering is making strides through better understanding of protein-protein recognition motifs (Li and Vederas, 2009) and the use of directed evolution (Fischbach, et al., 2007) to generate libraries of A-domains capable of activating a variety of substrates. Alternative approaches to making mutant libraries such as family shuffling (Crameri, et al., 1998) could be employed in the future to generate even larger libraries of large active site fragments, perhaps leading to more stable and active mutant enzymes.

Significance

Combinatorial biosynthesis is an attractive alternative to chemical synthesis for creating analogues of complex natural products. A major hurdle to realizing the potential of combinatorial biosynthesis includes the lack of selections or screens to process mutant libraries of enzymes in the context of their entire biosynthetic pathway. This work demonstrates a general method for rapidly generating and screening mutant libraries of nonribosomal peptide synthetases. The unbiased nature of structure-based screening allows for evaluation of the output from entire mutant pathways on a basis largely disconnected from total yield or specific bioactivity. Using the high-throughput multiplexed LC-MS/MS screening method outlined here, four derivatives of the antibiotic andrimid were generated, three of them not previously reported. The relative bioactivities of the derivatives were measured and it was found that two of the derivatives exhibited a shift in antibacterial specificity with concomitant decreases in their MIC values, consistent with previous synthetic chemistry driven SAR studies. Mutagenesis combined with LC-MS/MS for a structure based readout is a general approach that can be readily extended to other nonribosomal peptide synthetase and polyketide synthase pathways as well as to other biosynthetic pathway paradigms.

Experimental Procedures

Chemicals and Reagents

Restriction enzymes were from Invitrogen (Carlsbad, CA), T4 DNA ligase was from New England Biolabs (Ipswich, MA), Phusion polymerase from Finnzymes (Woburn, MA) was used for PCR. Growth medium was from BD Biosciences (San Jose, CA). Oligonucleotides were purchased from Integrated DNA Technologies (Coralville, IA) and are listed in Supplemental Experimental Procedures. HPLC solvents were from Fisher (Pittsburgh, PA). Formic Acid was from Acros Organics (Geel, Belgium). PCR cleanup, plasmid miniprep and gel extraction kits were from Qiagen (Valencia, CA). Stable-isotope labeled amino acids were from Cambridge Isotopes (Andover, MA). All other reagents, chemical and consumables were from Sigma unless otherwise specified.

Strains and Plasmids

E. coli DH5α γ pir was used for all cloning steps for plasmids containing oriR6K. E. coli DH5α and BL21(DE3) were obtained from the UIUC Cell Media Facility. E. coli imp ASR, Klebsiella pneumoniae ATCC 13883, Staphylococcus aureus ATCC 25923, and Enterococcus faecalis ATCC 19433 were used as the indicator strains for agar overlay assays. The andrimid producer, P. agglomerans Eh 335 and E. coli imp ASR were a gift from Christopher T. Walsh. K. pneumoniae ATCC 13883, S. aureus ATCC 25923, and E. faecalis ATCC 19433 were obtained from ATCC (Manassas, VA). All strains were cultured in LB or 2xYT at 37 °C (E. coli, K. pneumoniae, S. aureus, and E. faecalis) or 28 °C (P. agglomerans). Plasmid pQE60 (Qiagen) was used for expression of admK in P. agglomerans. Plasmids pACYC-Duet, pET-Duet, pET28a, and pUC-19 were obtained from Novagen (Darmstadt, Germany). The suicide plasmid pMQ118 was obtained from Presque Isle Cultures (Erie, PA). All strains and plasmids are listed with relevant characteristics in Supplemental Experimental Procedures.

Antibiotic Bioassays

100 mm diameter LB agar plates were overlaid with 4 mL molten LB agar inoculated with 75 μL of an overnight culture of the indicator strain. The inoculated top agar was allowed to solidify before sterile paper disks were laid onto the assay plates. Onto each disk 10–20 μL of sterile filtered culture supernatant or control was spotted before incubation at 37 ºC overnight or until zones of inhibition were visible.

DNA Sequencing

DNA sequencing was carried out at the University of Illinois core sequencing facility on an ABI 3730XL capillary sequencer on reactions using BigDye Version 3.0 terminator/enzyme mix and standard protocols. Sequence outputs were assembled and analyzed using Sequencher 4.6 software (Gene Codes, Ann Arbor, MI).

Mass Spectrometry

A 7 Tesla LTQ-FT equipped with a Surveyor autosampler and a MS pump was used for all mass spectrometric analysis (Thermo-Fisher Scientific). Andrimid and its derivatives were analyzed using a Jupiter 4.6 X 150 mm C18 HPLC column (Phenomenex) and a gradient of 0–75% ACN over 35 minutes. LC solvents (water and acetonitrile) contained 0.1% formic acid. Initial screening was performed at low (ion-trap) resolution with a selected ion monitoring (SIM) window from m/z 435–535 as the full scan and data-dependant fragmentation of the top 10 peaks from the full scan. Confirmation of derivative production was carried out using the same scans but in high (FT, 100,000) resolution. Mass spectrometric data was analyzed using Qual Browser software (Thermo-Fisher Scientific).

Molecular Genetics

Please refer to Supplemental Experimental Procedures for detailed methods for genetic manipulation of P. agglomerans.

Alignment of AdmK to Other Adenylation Domains and Extraction of Binding Pocket Residues

Several adenylation domains with specificity for the desired activity in the AdmK (Genbank accession number AAO39105.1) were aligned: GrsA (Genbank accession number CAA33603.1, phenylalanine specific), TycB M3 (Genbank accession number AAC45929.1, phenylalanine specific), McyA M2 (Genbank accession number CAO90227.1, alanine specific), JamO_( Genbank accession number AAS98786.1, alanine specific), BacA M1 (Genbank accession number AAC06346.1, isoleucine specific), FenB (Genbank accession number AAB00093.1, isoleucine specific), SrfAB M3 (Genbank accession number BAA08983.1, leucine specific), GrsB M4 (Genbank accession number CAA43838.1, leucine specific) and LicB M1 (Genbank accession number AAD04758.1, valine specific). Adenylation domains were aligned using the program CLUSTALX (Thompson, et al., 1997) (Figure S3). Positions aligning to D235, A236, W239, T278, I299, A301, A322, I330, C331 and K517 of GrsA were extracted as the putative binding pocket residues (Stachelhaus, et al., 1999). The binding pocket residues were inspected for their variation across the aligned sequences and three positions corresponding to T278, I299, and A322 in GrsA were chosen for mutagenesis.

Construction and Screening of admK Mutant Library

Please refer to Supplemental Experimental Procedures for detailed methods for construction and screening of the admK mutant library.

Isolation and Quantification of Andrimid and Analogues

Andrimid was produced from liquid culture of P. agglomerans in 2xYT supplemented with antibiotics and amino acids as required for 28 ºC for 24 hours. Cells were removed by centrifugation followed by filtration through a 0.45 μm membrane (Millipore, Billerica, MA). Filtered culture supernatant was extracted three times with one-half volume ethyl acetate. Ethyl acetate fractions were combined and dried under vacuum. The crude extract was dissolved in 50:50 water: methanol and extracted with 9 volumes of hexane. The water:methanol fraction was then extracted with one half volume methylene chloride. The methylene chloride fraction was dried under vacuum and then dissolved in 10 mL 20% acetonitrile and applied to a 12 cc C₁₈ SPE column (Whatman, Piscataway, NJ) equilibrated with 20% acetonitrile. The column was washed with 2 volumes 40% acetonitrile and eluted with one volume 80% acetonitrile. The eluate was dried and dissolved in 20% acetonitrile then loaded onto a 10 × 150 mm Eclipse C₁₈ column (Agilent, Santa Clara, CA). Andrimid and analogues were eluted from the column using a linear gradient 0–70% acetonitrile over 30 minutes. Andrimid and analogues eluted as pure peaks from 25–32 minutes. Peak purity and yield were determined by a LC-MS assay using the same column, gradient and mass spectrometer used in screening the mutant library calibrated with standards at known concentrations.

Stable Isotope Feeding Study

Mutants A2 and B12 were grown in M9 medium supplemented with 0.1% yeast extract, 0.1% casamino acids and 50 mM each of ¹³C₂ leucine and/or ¹³C₆ isoleucine (Cambridge Isotopes) for 24 hours at 30 ºC. The culture supernatants were syringe filtered and analyzed using the LC-MS/MS method outlined above.

MIC Determination

140 mm diameter LB agar plates were overlaid with 12 mL molten LB agar inoculated with 225 μL of OD₆₀₀ = 0.8 indicator strain. The inoculated top agar was allowed to solidify before sterile paper disks were laid onto the assay plates. Ten μL of antibiotic was spotted onto the disks and plates were incubated 8 hours at 37 ºC. Zones of inhibition were defined as the radius and measured using calipers. Zones of inhibition were plotted against Log₁₀ of the concentration of spotted antibiotic. MICs were defined as the minimum concentration required to yield a 1 mm zone of inhibition and calculated by nonlinear regression using the software package GraphPad Prism (La Jolla, CA).