Genome characteristics of T. moblis KA081020-065

The complete genome sequence of T. mobilis strain KA081020-065 totaling 6,513,401 bp was established by 454 pyrosequencing to reveal five replicons comprising a 3,919,492 bp circular chromosome and four circular plasmids ranging in size from 83,885 bp (pTM4) to 1,126,962 bp (pTM3) (Fig. 3). Its genome organization with three megaplasmids greater than 600 kb each is reminiscent of several other α-proteobacteria members such as the rice plant endophyte Azospirillum sp. B510 that harbors six plasmids, five of which in excess of 500 kb.²⁴ General features of the T. mobilis genome, which has a high average G+C content of 68%, are summarized in Table S1.

Open in a separate window

Figure 3

Genome maps of T. mobilis KA081020-065. From the outside inward, the first and second circles show predicted coding regions transcribed on the forward and reverse strand, respectively, colored by functional categories according to COG classification. The third circle shows RNAs (yellow for tRNA and red for rRNA). The fourth and fifth circles show G+C content and GC skew (G-C)/(G+C) calculated using a 5 kb window (purple for values >1, orange for values <1), respectively.

Based on the chemical structure of the didemnin depsipeptides that contains the two ketide-extended amino acid residues isostatine (Ist) and α-(α-hydroxyisovaleryl)-propionic acid (Hip),¹⁷ we hypothesized their biosynthesis by a modular hybrid nonribosomal peptide synthetase-polyketide synthase (NRPS-PKS) pathway reminiscent of nostopeptolide biosynthesis.²⁵ Bioinformatic analysis of the Tistrella genome revealed just one large NRPS-PKS biosynthetic gene cluster (did) located on pTM3 (TMO_c0602–c0611) that was consistent with the proposed biosynthesis of the didemnin family of cyclic depsipeptides. We detected only one other multimodular NRPS-based gene cluster in the genome, a 34,665 bp locus on pTM2 (TMO_b0323–b0325), which rather encodes a putative eight-residue peptide product inconsistent with didemnin assembly (Fig. S3).

Bioinformatic analysis of the didemnin biosynthetic gene cluster reveals an unexpected post-assembly processing mechanism

In silico analysis of the putative didemnin biosynthetic gene cluster (did) revealed eight NRPS and two PKS-encoding genes spanning from didA to didJ (Fig. S4), comprising 13 modules in total (Fig. 4). Based on the enzymatic logic of thiotemplate-mediated assembly line biosynthesis, the projected product of the DidB–J megasynthetase perfectly correlates with the co-linear synthesis of didemnin B. The DidB adenylation (A) domain A3 is predicted to load pyruvate and reduce it in cis via the adjacent ketoreductase (KR) domain to give lactate bound to the DidB thiolation (T) domain in a reaction reminiscent to that proposed in valinomycin assembly.²⁶ Further processing by the monomodular DidC and the trimodular DidD NRPSs are next proposed to assemble the Pro-DMeLeu-Thr-Ile tetrapeptide fragment. The N-methylated leucine residue is the lone d-amino acid in the didemnin B molecule, which correlates nicely with the epimerase (E) and methyltransferase (MT) domains that reside in the Leu-specific module 5 of DidD. Malonate extension of the DidD product by the PKS DidE, which lacks an integrated acyltransferase (AT) domain reminiscent of trans-AT PKSs,²⁷ yields the ketide-extended β-hydroxy γ-amino acid residue Ist. Analysis of the genes bordering didA–J and elsewhere in the T. mobilis genome, however, did not reveal a discrete AT from trans-AT systems, suggesting that the didemnin polyketide synthase may co-opt the fatty acid synthase AT FabD as reported recently for the biosynthesis of FK228.²⁸

Open in a separate window

Figure 4

Proposed biosynthetic pathway of didemnins and domain organization of the biosynthetic did gene cluster in T. mobilis KA081020-065. Domain notation: C, condensation; A, adenylation (substrate included); E, epimerase; KR, ketoreductase; KS, b-ketoacyl synthase; MT, methyltransferase; T, thiolation; TE, thioesterase.

Subsequent activation of 2-oxoisovaleric acid by the DidF A domain, whose sequence is similar to the N-terminal HetE A domain (43% identity at amino acid level) that is also responsible for its selection in hectochlorin biosynthesis,²⁹ sets up the incorporation of the α-hydroxy acid 2-hydroxyisovaleric acid and a second round of malonate extension by the DidG PKS. The DidG MT domain putatively incorporates the α-methyl group in the Hip residue. Further chain elongation of the DidG product by the sequential addition of the residues Leu, Pro, and Me₂Tyr via the monomodular DidH–J NRPSs completes the assembly of the linear didemnin B molecule. The C-terminal thioesterase (TE) domain of DidJ is hypothesized to offload didemnin B as the cyclic depsipeptide product. In the case of co-produced nordidemnin B, the DidD A7 domain is anticipated to exhibit substrate flexibility toward aliphatic residues to also activate and incorporate Val. [Hysp]didemnin B³⁰ is an additionally characterized didemnin B derivative from T. mobilis extract (Fig. S5) which further indicates substrate promiscuity in DidF domain A9 for α-hydroxy acids 2-hydroxyisovaleric acid (didemnin B) and 2-hydroxy-3-methylvaleric acid ([Hysp]didemnin B. While the biosynthetic model described above is clearly consistent with the observed didemnin B and nordidemnin B chemistry of T. mobilis, this proposal ignores the DidA bimodular NRPS that should function as the priming synthetase prior to DidB catalysis. Inspection of both DidA A domains revealed their preference for Gln. Furthermore, sequence analysis of the DidA N-terminal condensation (C) domain suggested its enzymatic role in the fatty acylation of the first amino acid residue to yield a lipopeptide product (Fig. 4) consistent with other important nonribosomal peptides such as surfactin.³¹ Among the natural didemnin congeners previously identified from the tunicate T. solidum, only didemnins X and Y have both Gln residues and a lipid component (Fig. 1).³² Interestingly, although these two didemnin derivatives contain Gln residues, they each have more than two residues as predicted by the bimodular DidA if each module were to function just once in peptide elongation. We thus explored the possibility that the immediate product of the didemnin biosynthetic pathway is not (nor)didemnin B, but rather an acylglutamine ester such as (nor)didemnin X or Y in which the ester side chain facilitates the cellular export of the cyclic depsipeptide from the Gram-negative bacterium for extracellular unmasking of the didemnin B byproduct.

Conversion of didemnins X and Y to didemnin B captured by imaging mass spectrometry

Repeated attempts were unsuccessful to isolate additional didemnin congeners by altering the culture medium in order to evaluate the activation hypothesis. We thus turned to MALDI-TOF MS to analyze the bacterial culture extracts, whereupon we successfully detected low abundance masses consistent with didemnins X and Y in addition to two novel derivatives, nordidemnins X and Y (Fig. S6). We confirmed their structures by MSⁿ analysis and high resolution Fourier-transform ion cyclotron resonance MS as reported in the Supporting Information (Fig. S7). The production of didemnins X and Y not only suggests that they are the fully elaborated biosynthetic product of the didemnin synthetase but also indicates that the DidA bimodular NRPS functions iteratively to install the three to four Gln residues in didemnins X and Y, respectively (Fig. 4).

In order to explore the biosynthetic connection between these molecules, we next used MALDI-imaging MS^33,34 to study the temporal and spatial distribution of the didemnin compounds. We interrogated T. mobilis colonies over a time course ranging from 6 hours to 6 days (Fig. 5), which strikingly revealed that didemnins X and Y are produced and excreted from the colony during the early growth phase of the bacterium peaking at 24 hours. Concomitant with their disappearance by day 4 was the emergence of didemnin B as the primary didemnin metabolite that steadily increased over time. This time-lag difference between didemnins X and Y and didemnin B was similarly observed in the nordidemnin series (Fig. 5) and is consistent with the proposed model in which the lipoglutamine side chain facilitates the cellular export of the mature didemnin molecule for extracellular processing.

Open in a separate window

Figure 5

Time-course production of didemnins by T. mobilis KA081020-065 detected by MALDI-imaging mass spectrometry.

We next set out to determine whether the maturation of didemnins X and Y involved a secreted factor. To this end, we performed a hydrolysis assay of purified didemnin precursors and filtered growth media cultured with and without the bacterium. The assay clearly showed that didemnin X/Y turnover to didemnin B was dependent on the addition of a protein extract from a 1 day T. mobilis culture supernatant (Fig. S8). While studies are ongoing to identify the secreted hydrolytic factor involved in the peptide activation mechanism, sequence analysis of the open reading frames adjacent to the didA–J operon revealed a number of genes encoding membrane-associated transport proteins (orf3, orf6) and hydrolytic enzymes (orf2, orf14) (Fig. S4), further supporting the export-hydrolysis model. Recently an activation mechanism was established in the nonribosomal peptide synthesis of the antibiotic xenocoumacin from the γ-proteobacterium Xenorhabdus nematophila in which a precursor molecule is processed by a membrane-bound and d-Asn-specific peptidase.³⁵ Although the characterized pathway in xenocoumacin processing parallels the proposed pathway for didemnin maturation, the did cluster does not harbor a homolog of the xenocoumacin ABC transporter-transpeptidase XcnG. In addition, the maturation of xenocoumacin takes place in the periplasmic space rather than the extracellular space for didemnins, which may suggest an alternative mechanism in T. mobilis.

MS sample preparation and MS analysis of didemnins

For MALDI-TOF MS analysis, extract or peptide samples were dissolved in MeOH, mixed 1:1 with saturated Universal MALDI Matrix and spotted on a Bruker MSP96 anchor plate. MALDI-TOF MS analysis was performed on a Microflex Bruker Daltonics mass spectrometer outfitted with Compass 1.3 software suite. For MSⁿ and FTMS analysis, HPLC-purified samples were redissolved in 50 μl MeOH/water/formic acid (50:49:1) and injected by a nanomate-electrospray ionization robot (Advion) for consecutive electrospray into the MS inlet of a LTQ 6.4T Fourier transform ion cyclotron resonance (FT-ICR) mass spectrometer (Thermo Finnigan). MS and MSⁿ data were acquired in the positive ion mode. FTMS data were acquired in 400–2,000 m/z scans. Selected peptide mass signals were manually isolated and fragmented by CID. MSⁿ data were collected either in ion trap or FT detection mode. All data were analyzed using QualBrowser, which is part of the Xcalibur LTQ-FT software package (Thermo Fisher).

Genome sequencing, annotation, and analysis

We used a massively parallel pyrosequencing technology (Roche 454 GS FLX)⁴⁰ to sequence the T. mobilis KA081020-065 genome. We obtained 315,496 reads with an average length of 334 bp, which provided a 16.2-fold coverage of the 6.4 Mb genome. From these reads, we assembled 112 contigs (>500 bp) using the Newbler software of the 454 suite package. To enhance sequence quality and to construct scaffolds, we sequenced a paired-end Illumina library of an additional 2,625,640 bp with average length of 115 bp, which were then mapped to the genome sequence. All contig relationships within scaffolds were then validated by PCR, while the relationships among scaffolds were determined by multiplex PCR.⁴¹ Gaps were filled by sequencing PCR products. The final sequence assembly was carried out using the phred/phrap/consed package (http://www.genome.washington.edu) in which we resequenced the low sequence quality regions. The final error rate of the genome sequence was 0.28 per 100,000 bases. The sequences of the five replicons were deposited in the GenBank database as accession numbers CP003236–CP003240.

Putative protein-coding sequences (CDS) were determined by combining the prediction results of the Glimmer 3.02⁴² and Z-Curve programs.⁴³ Functional annotations of CDS were performed by searching the KEGG and the NCBI non-redundant protein databases. tRNA genes were predicted with tRNAScan-SE (v1.23).⁴⁴ Protein domain predictions and COG assignments⁴⁵ were performed by RPS-BLAST using the NCBI CDD library.⁴⁶

In silico analysis of the didemnin biosynthetic gene cluster and other NRPSs in T. mobilis KA081020-065

The putative roles of the proteins in the didemnin gene cluster were assigned using protein–protein BLAST and Pfam analyses. NRPS A domain specificities were predicted using the online program NRPSpredictor2 and antiSMASH.^47–49

Time-course study of didemnin production in T. mobilis by MALDI-imaging MS

One μl of fresh overnight T. mobilis culture was spotted onto GYP agar plates and allowed to grow for 6 hours to 6 days. Colonies with different growing times were transferred to Bruker MSP 96 anchor plates and analyzed concurrently using the IMS technique as previously described.^33,50

We thank Z. Shao for providing the Pacific Ocean Tistrella strains and N. Ziemert, J. Yang and P. Lai for assistance with data analysis. Financial support was provided by King Abdullah University of Science and Technology (SA-C0040/UK-C0016 to P.Y.Q.), China Ocean Mineral Resources Research and Development Association (DY125-15-T-02 to P.Y.Q.), and the National Institutes of Health (GM085770 to B.S.M., CA044848 to W.F., GM086283 and S10RR029121 to P.C.D.).