Intertwined Precursor Supply during Biosynthesis of the Catecholate-Hydroxamate Siderophores Qinichelins in Streptomyces sp. MBT76.

The explosive increase in genome sequencing and the advances in bioinformatic tools have revolutionized the rationale for natural product discovery from actinomycetes. In particular, this has revealed that actinomycete genomes contain numerous orphan gene clusters that have the potential to specify many yet unknown bioactive specialized metabolites, representing a huge unexploited pool of chemical diversity. Here, we describe the discovery of a novel group of catecholate-hydroxamate siderophores termed qinichelins (2-5) from Streptomyces sp. MBT76. Correlation between the metabolite levels and the protein expression profiles identified the biosynthetic gene cluster (named qch) most likely responsible for qinichelin biosynthesis. The structure of the molecules was elucidated by bioinformatics, mass spectrometry, and NMR. The genome of Streptomyces sp. MBT76 contains three gene clusters for the production of catecholate-peptide siderophores, including a separate cluster for the production of a shared catecholate precursor. In addition, an operon in the qch cluster was identified for the production of the ornithine precursor for qinichelins, independent of primary metabolism. This biosynthetic complexity provides new insights into the challenges scientists face when applying synthetic biology approaches for natural product discovery.

Actinobacteria are renowned for their ability to manufacture a diversity of bioactive small molecules.[1,2] High-throughput screening of actinomycetes has yielded many useful therapeutic agents but also turned big pharma away from NPs for drug-discovery programs due to high cost and chemical redundancy.[3,4] The increase in genome-sequence information has uncovered a vast and yet untapped biosynthetic potential and metabolic diversity, which has brought the microbial NPs back into the spotlight. However, many of the biosynthetic gene clusters (BGCs) discovered by genome mining are poorly expressed under laboratory conditions, and a major new challenge lies in finding the triggers and cues to activate their expression.[5] Such approaches include, among others, chemical triggers, microbial cocultivation, induction of antibiotic resistance, and heterologous gene expression.[6−10] In addition, the advances in genetic tools applied in synthetic biology, such as transformation-associated recombination (TAR), Red/ET recombination, and CRISPR-Cas9, had aided in the discovery of cryptic products through engineering of their biosynthetic pathways.[11]

A second bottleneck in genomics-based approaches is to establish a link between genomic and metabolomic data.[5,12] It is difficult to assign the genetic basis for specific chemical scaffolds through bioinformatics analysis alone, largely due to nature’s flexibility in catalytic enzymology, i.e., enzyme promiscuity[13] and crosstalk among different gene clusters.[14,15] The latter offers a significant hurdle in drug-discovery approaches that are based solely on heterologous expression of single gene clusters.[16] This gap can be bridged by genomics-based methodologies based on tandem MS analysis of metabolites[17,18] that allow the linkage of specific biosynthetic genes to the bioactivity of interest. As we and others have exemplified, statistical correlation between transcript or protein expression levels and the presence of bioactive molecules is equally feasible.[19−21] Subsequent bioinformatics analysis of a biosynthetic gene cluster (BGC) provides important (partial) structural information.[22] This information can guide researchers to optimize compound isolation and identification, so as to recover sufficient quantity of targeted metabolites(s) from highly complex matrices to warrant de novo structural elucidation.[23]

A specific class of natural products is the siderophores, which are synthesized by nonribosomal peptide synthases (NRPS) and act as iron scavengers.[24] Their chemical topologies and biosynthetic machineries have been studied extensively,[14,15,25] and a wide range of structures have been reported. Siderophores are generally classified into catecholates, hydroxamates, (hydroxy)-carboxylates, and mixed ligands thereof.[24] Members of the mixed catecholate-hydroxamate subfamily, including rhodochelin,[15] heterobactins,[26] rhodobactin,[27] lystabactins,[28] mirubactin,[29] and S-213L,[30] feature both a 2,3-dihydroxybenzoate(s) (2,3-DHB) moiety and (modified) δ-N-hydroxyornithine residues within the same molecule. Consequently, the biosynthesis of catecholate-hydroxamate siderophores is always initiated by loading 2,3-DHB as a starter unit into the modular NRPS assembly line, followed by successive incorporation of amino acids, including ornithine, into the growing peptide chain.[15,26,29]

Streptomyces sp. MBT76 was previously identified as a prolific producer of antibiotics,[31] such as isocoumarins, prodiginines, acetyltryptamine, and fervenulin.[32] More recently, the activation of a type II polyketide synthase (PKS) gene cluster (qin) in Streptomyces sp. MBT76 induced the production of novel glycosylated pyranonaphthoquinones (qinimycins).[33] As this metabolic spectrum was dominated by polyketides, while the BGCs for NRPS are as commonplace as PKS in bacterial genomes,[34] we anticipated that peptides were underrepresented in our studies. Here, we describe the discovery and characterization of qinichelins, new mixed-type catecholate–hydroxamate siderophores from Streptomyces sp. MBT76. The aforementioned limitations in genome-mining strategies were overcome through varying the growth conditions to fluctuate peptide production, after which quantitative proteomics allowed the connection of the NRPS gene clusters to their metabolic products.

Results and Discussion

Biosynthetic Loci for Catechol-Peptide Siderophores Are Dispersed through the Genome of Streptomyces sp. MBT76

Previous analysis of the genome of Streptomyces sp. MBT76 by AntiSMASH[35] identified 55 putative biosynthetic gene clusters (BGCs) specifying secondary metabolites.[33] A total of 16 of these contained gene(s) encoding NRPS, suggesting rich peptide metabolism. Our attention was in particular directed to three distinct NRPS BGCs, containing genes for the biosynthesis of catechol-peptide siderophores. One BGC (Figure c) matched the described BGC for griseobactin,[36] but this lacked the dhb genes required for 2,3-DHB synthesis. However, a copy of this operon was present in a second cluster, entA-C-E-B (Figure b). Together with entF and entD, this cluster contains all genes necessary for the biosynthesis of enterobactin, a catechol-peptide siderophore from E. coli, albeit in a different order than in the original cluster.[37]

Figure 1

NRPS BGCs involved in catechol-type siderophore biosynthesis in Streptomyces sp. MBT76. BGCs for a new siderophore (a), for enterobactin (b), and for griseobactin (c) could be identified. Carrier protein domains (ACP/PCP) are depicted in light blue, condensation domains in dark blue, epimerization domains in green, and thioesterase (termination) domains in yellow. Adenylation domains are shown in purple, together with their predicted substrates, and transport proteins in red. Triangles indicate the position of iron boxes likely bound by the iron repressor DmdR.

A third NRPS BGC, designated qch (Table and Figure a), also lacked the dhb genes but contained the qchG gene for a 2,3-DHB ACP homologous to the EntB-ACP domain. The starter condensation (C) domain of the NRPS QchI, which likely appends a 2,3-DHB unit to the N-terminus of the peptide, also indicated the presence of a 2,3-DHB moiety in the final structure.[37] Through phylogenetic analysis of adenylation (A) domains[38] of the two core NRPS (QchI and QchJ), a nonribosomal peptide with the sequence 2,3-DHB-Ser-Orn(ornithine)/Asp-Ser-Ser-Orn-Orn was predicted as the product specified by the BGC, whereby no clear consensus prediction could be made for the second A domain. Two epimerization (E) domains in the first and third modules of QchI probably transform the stereochemistry of l-Ser into d-Ser, while two other genes are likely involved in tailoring of Orn: qchF coding for an l-ornithine-5-monooxygenase and qchQ coding for a GCN5-related N-acetyltransferase.[39] This strongly suggests that the product of the cluster is a mixed hydroxymate–catecholate siderophore, which is further supported by the presence of the siderophore-related transporter genes qchH and qchL-P. Interestingly, the absence of an esterase (TE) domain at the terminus of QchJ indicates an unusual release of the mature peptide, potentially leading to a linear structure, in contrast to a cyclic peptide usually produced by TE domains.

Table 1

Blastp Analysis of NRPS Cluster

Qch	length (AA)	predicted function	identity (%)	alignment length	E value	bitscore
A	325	N-acetyl-gamma-glutamyl-phosphate reductase 2 ArgC	76.31	325	2.00 × 10–175	498
B	32	questionable ORF
C	383	arginine biosynthesis bifunctional protein ArgJ	85.94	384	0	652
D	311	acetylglutamate kinase ArgB	79.86	283	3.00 × 10–152	437
E	398	acetylornithine aminotransferase ArgD	76.1	385	0	566
F	474	l-ornithine 5-monooxygenase	71.4	437	0	627
G	82	isochorismatase ACP domain	55.41	74	2.00 × 10–18	77.8
H	326	iron(III) dicitrate transport permease	33.94	330	5.00 × 10–36	139
I	4295	nonribosomal peptide synthetase	45.92	3151	0	2075
J	3247	nonribosomal peptide synthetase	49.28	3253	0	2474
K	70	MbtH-like protein	74.24	66	9.00 × 10–30	106
L	670	ABC transporter related protein	51.21	537	6.00 × 10–154	465
M	570	ABC transporter related protein	52.87	592	4.00 × 10–167	497
N	268	siderophore-interacting protein	46.21	264	9.00 × 10–67	217
O	362	ferric enterobactin transport system permease	53.08	341	1.00 × 10–109	333
P	372	transport system permease protein	60.45	354	8.00 × 10–115	346
Q	203	GNAT family N-acetyltransferase	49.68	155	5.00 × 10–34	125

A search of the CAS database (American Chemical Society, http://scifinder.cas.org), using the predicted sequence of the qch-specified peptide product as a query, yielded S213L (1, Figure ),[30] an antibiotic/antifungal siderophore with the sequence DHB-Ser-Orn-Ser-Orn-hOrn-chOrn, as a closest hit. The partial sequence for the S213L BGC has been described[40] but differed from the qch BGC sequence. Moreover, the fourth residue of S213L is Orn, instead of Ser predicted for the qch-specified product. This strongly suggested that the qch cluster might not produce S213L, but a related compound 2 (Figure ). The stereochemistry of the individual qinichelin residues was further deduced using Marfey’s protocol (see the Supporting Information Data section for details).

Figure 2

Molecular structures of S213L (1), qinichelin (2), acetyl-qinichelin (3,4), and dehydroxy-qinichelin (5). Abbreviations of moieties are shown to facilitate the comparison of respective structure. DHB, dihydroxybenzoate; Ser, serine; Orn, ornithine; hOrn, δ-N-hydroxy-ornithine; chOrn, cyclized δ-N-hydroxy-ornithine hydroxamate. Acetylation of qinichelin can occur on two positions as indicated. The shown configurations of DHB–d-Ser–l-Orn–d-Ser–l-Ser–l-hOrn–l-chOrn in qinichelins were based on Marfey’s protocol (Supporting Information data and Figures S10–S15) combined with biosynthetic considerations (see Figure ).

Figure 5

Model of intertwined biosynthetic pathways of catechol-peptide siderophores in Streptomyces sp. MBT76. The functional crosstalk among four entirely separate gene clusters ensures the assembly of two types of catechol-peptide siderophores, namely qinichelins and griseobactin. The DHB is supplied by the ent gene cluster and is shared by three NRPS systems. The DHB-hexapeptide backbone in qinichelin follows an orthodox colinear extension model, while the Orn building block could arise from either qchA-E or the canonical arg gene cluster. The differently modified Orn in the dashed boxes are likely accepted by the AOrn domain of modules 2 and 5 to produce different qinichelin variants.

Interestingly, qch contains four genes (qchA and qchCDE) that are highly similar to the arg genes argC, argJ, argB, and argD, respectively, which are required for the synthesis of the precursor ornithine from glutamate.[41] In addition, a canonical arg cluster for ornithine/arginine metabolism was also found in the Streptomyces sp. MBT76 genome, including the regulatory gene argR and argE-H for the subsequent conversion of ornithine to arginine, all of which were lacking in the qch gene cluster. Taken together, bioinformatics analysis suggested that up to three different catecholate–peptide siderophores might be produced by the strain, sharing one set of the dhb operon for 2,3-DHB synthesis, while ornithine, as a precursor for compound 2, might be produced by primary metabolism or by enzymes derived from the qch BGC.

Proteomics Analysis of the qch Cluster and Identification of the Qinichelins

We previously described the natural product proteomining pipeline, which makes use of the strong correlation between the amount of a (bioactive) molecule produced and the expression level of its biosynthetic proteins.[20] This was applied to efficiently connect genes (genotype) to a given metabolite or bioactivity of interest (chemotype). The reverse analysis whereby the expression level of a targeted BGC (known genotype) is used to predict its yet uncharacterized molecule that is produced (unknown chemotype) should be equally feasible. Accordingly, this reverse proteomining could complement a genome-mining strategy to facilitate the discovery of novel compounds.

As a prerequisite, sufficient fluctuation of protein levels should be achieved as a result of varying growth conditions.[20] Accordingly, Streptomyces sp. MBT76 was grown in modified liquid minimal medium (NMMP), supplemented with (A) no additive (control), (B) 2% (w/v) NaCl, (C) 1% (w/v) starch, (D) 0.8% (w/v) peptone, or (E) 0.6% (w/v) yeast extract, as these conditions have been proven successful previously. Subsequent quantitative proteomics analysis of whole-cell lysates, using two mixtures of three samples to compare all growth conditions, yielded 1472 protein identifications, wherein relative expression levels of 1174 proteins were quantified with at least two independent events, including proteins belonging to the BGCs of interest (Table ). Cultures grown in NMMP with peptone and, remarkably, in NMMP without additives, showed strong expression of the Qch proteins, as demonstrated by the marked upregulation of QchF and QchH-J when compared to, e.g., condition B (NMMP with 2% NaCl). This may have been caused by low iron content in the growth media. The changes in expression level for QchA and QchG were not in line with the other proteins from the qch BGC. QchA and ArgC could not be differentiated due to their high sequence similarity. However, the fluctuation pattern of QchA/ArgC for the five culture conditions was in line with that of all detected Arg proteins, strongly suggesting that the observed signals for QchA/ArgC were most likely dominated by ArgC. For QchG, the data set contained only three quantification events, due to its small size, potentially leading to errors in quantifications.

Table 2

Quantitative Proteomics Analysis

	normalized expression ratio (log2)a						number of quantificationsb
proteins	D/A	B/A	B/D	D/C	E/C	E/D	D/A	B/A	B/D	D/C	E/C	E/D
qinichelin
QchA/ArgCc	–1.6	–0.8	1.0	–1.7	–2.2	–0.3	7	7	7	3	3	3
QchF	–0.6	–2.4	–2.3	0.1	0.4	0.0	8	8	8	7	7	7
QchG				1.7	1.2	–0.7	1	1	1	3	3	3
QchH	–1.2	–2.5	–1.2	–1.0	–1.5	–1.2	6	6	6	4	4	4
QchI	–0.2	–3.1	–3.3	–0.7	–1.1	–0.5	12	12	12	7	7	7
QchJ	–1.3	–3.0	–1.7	–0.2	–0.4	–1.0	22	22	22	13	13	13
enterobactin
EntA	–2.1	–0.6	1.3				3	3	3	0	0	0
EntB-IC	–0.8	0.9	1.1				3	3	3	1	1	1
griseobactin
GriG	–1.3	–1.5	0.1	–2.2	–2.2	–0.7	8	8	8	8	7	7
GriF				–1.4	–1.2	–0.1	0	0	0	6	6	6
GriE	–0.8	–1.4	–0.3				3	3	3	0	0	0
arginine biosynthesis cluster
QchA/ArgCc	–1.6	–0.8	1.0	–1.7	–2.2	–0.3	7	7	7	3	3	3
ArgJ	–0.9	–0.9	0.0	–0.6	–0.8	–0.9	7	7	7	6	5	5
ArgD	–1.0	–0.2	0.7	–1.7	–0.5	0.9	2	2	2	2	2	2
ArgR	–1.5	0.1	1.5	–2.5	–1.0	1.1	9	9	9	4	4	4
ArgG	–0.6	–0.6	–0.1	–0.6	–0.7	0.0	4	4	4	2	2	2
ArgH	–1.5	–1.1	0.6	–1.7	–0.8	0.5	8	8	8	9	9	9

Changes in protein expression levels observed for the indicated proteins when compared among growth conditions A–E.

Number of quantification events used to calculate the expression ratios. Quantifications based on less than two events (italicized) were discarded.

The sequences of QchA and ArgC were very similar, resulting in insufficient unique peptides for quantification. Instead, quantification is shown for nonunique peptides.

The proteomics analysis demonstrated the expression of the Qch proteins in, among others, culture condition D (NMMP with peptone) and thus indicated the existence of the corresponding catecholate-peptide siderophore under these growth conditions. In our previous metabolomics study of Streptomyces sp. MBT76 under the same conditions,[32] no siderophores were identified, which is most likely due to the use of ethyl acetate for the extraction, which is not suited for the isolation of the hydrophilic peptidic siderophores. Therefore, here, spent media from five culturing conditions were desalted only and directly subjected to reverse-phase LC-MS analysis (in positive mode) without any prior extraction, resulting in the detection of a signal at m/z 772.3 for NMMP (A) and NMMP with peptone (D), with the strongest signal obtained for A (Figure a). The fluctuation pattern of this molecule correlated well with the expression level of the qch gene cluster, suggesting this may be the sought-after compound 2. In addition to the molecular ion [M + H]+ at m/z 772.3 for the iron-free compound, a coeluting peak was observed at m/z 825.3 corresponding to the iron-bound [M + Fe3+ – 2H]+ species. Figure a depicts the combined signals for both species to compensate for any differences in iron(III) concentration among the different culture conditions.

Figure 3

Comparison of qinichelin production by LC-MS analysis. Spent medium samples of Streptomyces sp. MBT76 grown in conditions A–E (a) and in condition A in the absence (red line) or presence (black line) of Fe3+ (b), respectively, were compared. Shown are summed extracted ion chromatograms of [M + H]+, 772.3 m/z; [M + Fe3+ – 2H]+, 825.3 m/z ± 0.5 Da.

To confirm the structure of 2, the spent medium of condition A was reanalyzed on a high resolution LTQ-orbitrap instrument, including both MS1 and MS2 analysis. Due to the use of formic acid instead of trifluoroacetic acid in the eluent, the MS1 spectrum of 2 presented the highest intensity at m/z 386.6773 assignable to [M + 2H]2+ species, followed by the [M + H]+ peak at m/z 772.3471 (Figure S1), within 0.5 ppm accuracy from the predicted mass. Indeed, the MS2 analysis yielded almost all the expected fragmentation products of the predicted compound 2, with complete sequence coverage for both the b- and y-ion series (Figure a). Moreover, the MS2 analysis corroborated the hydroxylation of two ornithines (hOrn-5 and chOrn-6) at the C-terminus, and the cyclization of the last ornithine (chOrn-6). The most intensive signals were obtained for the b5 and y2 ions, indicating that a potential hydroxamate bond might be more susceptible to cleavage than an amide bond. However, it was noteworthy that MS/MS analysis alone was not enough to indicate the presence of a peptide or isopeptide bond between Ser-4 and hOrn-5. To clarify this, the m/z 772.3 was used as a probe to guide the separation of target compound from the spent medium of condition A on reversed phase HPLC. The obtained semipurified compound 2 was analyzed by 1H NMR (850 MHz, in D2O, Table ), COSY, HSQC, and HMBC techniques (Figures S2–S6), which indeed supported a catecholate–hexapeptide architecture comprising three serine and three ornithine residues. In particular, a key HMBC correlation from H2-5 of hOrn-5 to C-1 of Ser-4 established that the linkage between these two residues was through the δ-hydroxylated-amine rather than α-amine of hOrn-5. The free amine group at C-2 of hOrn-5 could be also reflected by the upfield shifted H-2 (δH 3.99), in contrast to the amidated H-2 of Orn-2 (δH 4.44) and chOrn-6 (δH 4.40).

Figure 4

High resolution MS/MS analysis of qinichelin. A spent medium sample of Streptomyces sp. MBT76 grown in condition A was subjected to high resolution LC-MS/MS analysis to obtain insights into the structures of unacetylated (a) and acetylated (b) qinichelin.

Table 3

NMR Data Assignment of Qinichelin[2] in D2O

		13Ca	1H
residue	position	δc	δH	intensity	multiplicity	J (Hz)	carbon correlated in HMBC
DHB	1	171.0
	2	117.8
	3	147.6
	4	145.5
	5	120.7	7.11	1	dd	8.5, 1.7	DHB (C-3, C-4, C-7)
	6	120.7	6.90	1	t	8.5	DHB (C-2, C-4)
	7	120.6	7.34	1	dd	8.5, 1.7	DHB (C-1, C-3, C-5)
Ser-1	1	173.5
	2	56.9	4.66	1	t	5.1	Ser-1 (C-1, C-3), DHB (C-1)
	3	62.1	4.00	3b	m		Ser-1 (C-1, C-2)
Orn-2	1	174.5
	2	54.5	4.44	1	dd	8.5, 5.1	Orn-2 (C-1, C-3, C-4), Ser-1 (C-1)
	3	28.5	1.96; 1.83				Orn-2 (C-1, C-2, C-4, C-5)
	4	24.2	1.76; 1.72				Orn-2 (C-2, C-3, C-5)
	5	39.7	3.02	2	td	8.5, 0.85	Orn-2 (C-3, C-4)
Ser-3	1	172.2
	2	56.5	4.53	1	t	5.1	Ser-3 (C-1, C-3), Orn-2 (C-1)
	3	62.0	3.86	2	m		Ser-3 (C-1, C-2)
Ser-4	1	171.2
	2	53.6	5.03	1	t	5.1	Ser-4 (C-1, C-3), Ser-3 (C-1)
	3	61.4	3.78	2	d	5.1	Ser-4 (C-1, C-2)
hOrn-5	1	170.1
	2	53.8	3.99	3b	m		hOrn-5 (C-1, C-3, C-4)
	3	28.8	1.87				hOrn-5 (C-1, C-2, C-5)
	4	22.2	1.76				hOrn-5 (C-5)
	5	48.6	3.65; 3.67				hOrn-5 (C-3, C-4), Ser-4 (C-1)c
chOrn-6	1	167.2
	2	51.5	4.40	1	dd	11.1, 6.0	chOrn-6 (C-1, C-3), hOrn-5 (C-1)
	3	27.1	1.84; 2.05				chOrn-6 (C-1, C-2, C-4, C-5)
	4	21.0	2.01; 1.94
	5	52.5	3.61; 3.67				chOrn-6 (C-3, C-4)

Chemical shifts of the carbon resonances are estimated from the HMBC data set.

Signals from C-3 of Ser-1 and C-2 of hOrn-5 overlapped, and no clear integral could be measured.

Key HMBC correlation confirmed the hydroxamate bond between Ser-4 and hOrn-5.

Together, these experiments confirmed the existence and the precise chemical structure of compound 2. With three iron-coordinating groups including one DHB moiety and two hydroxamates, our new compound resembles other mixed-ligand siderophores like amychelin[25] and gobichelin.[42] This strongly suggested that compound 2 was a siderophore, which was named qinichelin. The name refers to the origin of Streptomyces sp. MBT76, which was isolated from the Qinling Mountains in China.[31]

High Resolution MS/MS Analysis Reveals Production of Qinichelin Variants (3–5), Griseobactin, but Not Enterobactin

We suspected that an acetylated analogue of qinichelin could be produced by Streptomyces sp. MBT76, because acetylation by an N-acetyltransferase encoded by qchQ had not yet been found in qinichelin. Indeed, we observed an [M + H]+ species at m/z 814.3587 for acetylated qinichelin, with a slightly longer retention time than qinichelin. The high abundance of an [M + H]+ species instead of [M + 2H]2+ already indicated that one of the two free amines in qinichelin, δ-NH2 in Orn-2 or α-NH2 in hOrn-5, was acetylated, while a derivative with both acetylations was not detected. Upon fragmentation for MS/MS analysis, a surprising result was obtained because the fragmentation spectrum (Figure b) corresponded to a mixture of two different acetylated peptides 3 and 4 (Figure ). Some masses could only be assigned to acetylation at δ-NH2 in Orn-2 while other masses indicated acetylation of α-NH2 in hOrn-5. Since fragmentation of this [M + H]+ ion was less efficient than the unacetylated [M + 2H]2+ ion (Figure a), a complete sequence coverage could not be achieved for b and y ions. However, at least one b or y ion was present for each peptide/hydroxamate bond for both variants, thus providing strong evidence for the position of the post-translational modification. In addition, qinichelin variant 5 gave a [M]+ peak at m/z 755.3314, and the characteristic fragment at m/z 512.2096 indicated an Orn-5 instead of an hOrn-5 residue (Figure S7). We did not obtain sufficient amounts of compounds 3–5 for 2D NMR analysis, as they are minor relative to 2.

Since the proteomics analysis also revealed expression of the ent and gri clusters (Table ), we attempted to find their respective products, enterobactin and griseobactin, by MS/MS analysis. Indeed, griseobactin could be readily detected with highest intensity at m/z 394.1720 for the [M + 3H]3+ species, within 0.5 ppm of the expected mass. Another signal was observed for the [M + 2H]2+ species at m/z 590.7538, with an MS/MS fragmentation pattern corresponding exactly with published data.[43] Surprisingly, no enterobactin could be detected. This suggests that only the dhb operon in the ent cluster may be functional for 2,3-DHB precursor supply for griseobactin and qinichelin, but not enterobactin, production in Streptomyces sp. MBT76.

Qinichelin Production Belongs to the Iron Homeostasis Regulon

To support the iron-chelating function of qinichelin and its possible role in iron homeostasis of Streptomyces sp. MBT76, we searched for the occurrence of iron boxes within the qinichelin BGC. Iron boxes are cis-acting elements with a 19 bp palindromic consensus sequence TTAGGTTAGGCTAACCTAA that are bound by DmdR1, the global iron regulator in Streptomyces species.[44] When sufficient iron is available, the DmdR1–Fe2+ complex binds to iron boxes and represses the expression of siderophore biosynthetic and importer genes.[44] The dramatic reduction in qinichelin production under iron-rich conditions suggested that the expression of the qch cluster would also be under the negative control of DmdR1 (Figure b). Indeed, four highly conserved iron boxes were found within the BGC: (i) upstream of the predicted pentacistronic operon qchA-E involved in ornithine synthesis from glutamate, (ii) upstream of qchF coding for the l-ornithine 5-monooxygenase, (iii) upstream of the tricistronic operon (qchN-P) predicted to be involved in qinichelin transport, and (iv) upstream of qchQ that encodes the predicted qinichelin N-acetyltransferase (Figure a, and Table S1). The iron box identified 109 nt upstream of qchF that displayed the perfect palindromic sequence TTAGGTTAGGCTAACCTAA, which made it highly likely that the central NRPS genes of the qch cluster were regulated by DmdR1. Furthermore, the iron box upstream of the predicted qinichelin transporter system (qchN-P in Figure ) presents greater identity to the palindromic consensus sequence bound by DmdR1, compared to most of the iron boxes identified upstream of other siderophore uptake system genes present in the Streptomyces sp. MBT76 genome (Table S1). In addition, three iron boxes were identified in the gri cluster and one in the ent cluster (Figure b,c, and Table S1), suggesting that siderophore production in Streptomyces sp. MBT76 is indeed under control of DmdR1.

Interestingly, scanning for ARG boxes (consensus sequence CCATGCATGCCCATTGCATA) that are bound by the arginine repressor ArgR[45] revealed no reliable cis-acting sequences upstream of the qchA-E operon. Instead, the canonical argCJBDR gene cluster outside the qinichelin biosynthetic cluster displayed the putative ARG box at position −87 nt upstream of argC. This suggests differential regulation of the ornithine biosynthetic genes from primary metabolism and those involved in secondary metabolism.

Biosynthesis of Qinichelins Relies on Coordination between Multiple BGCs

The theoretical analysis and the experimental identification of griseobactin and qinichelins allowed us to postulate an intertwined model for the production of catecholate–peptide siderophores in Streptomyces sp. MBT76 (Figure ). The chorismate pathway within the ent gene cluster provides the building block 2,3-DHB to the three NRPS EntF, GriE, and QchI-QchJ, for enterobactin, griseobactin, and qinichelin, respectively. The 2,3-DHB moiety is activated by 2,3-dihydroxybenzoate-AMP ligase EntE and subsequently transferred to stand-alone aryl carrier proteins QchG or EntB2. As the necessary gene coding for the aryl carrier protein is lacking in the griseobactin BGC, this requirement could be remedied by either QchG or EntB2 to deliver the activated 2,3-DHB starter unit for GriE. The further mechanisms for NRPS assembly of enterobactin and griseobactin have been elaborated elsewhere.[36,37] The coordinated expression of multiple NRPS gene clusters for siderophore production in Streptomyces sp. MBT76 is striking but not unprecedented. Similar functional crosstalk between different NRPS BGCs was demonstrated for the assembly of the siderophores erythrochelin in Saccharopolyspora erythraea(14) and rhodochelin in Rhodococcus jostii RHA1.[15] Such crosstalk could enable structural diversity for siderophores on the basis of a limited number of biosynthetic genes and thus confer an evolutionary advantage for the producing bacteria in terms of iron acquisition. In particular, it would be advantageous for one bacterium to evolve specific siderophore(s) for their own benefit, to compete with the “siderophore pirates” that use siderophores biosynthesized by other species.[46] For example, the structurally novel amychelin produced by Amycolatopsis sp. AA4 seems to frustrate “siderophore piracy” of Streptomyces coelicolor by inhibiting its development.[25]

The assembly of the catecholate–hexapeptide backbone in qinichelin follows an orthodox linear logic of modular NRPS. Each module in QchI and QchJ contains an adenylation (A) domain for recognition of correct amino acid substrate, whereby Ser-1, Orn-2, Ser-3, Ser-4, hOrn-5, and hOrn-6 are sequentially bound and converted to aminoacyl adenylates. The two serine residues are converted from the initial L form[47] into their D stereoisomer by the epimerization (E) domain in modules 1 and 3. After QchG-mediated incorporation of 2,3-DHB, each condensation (C) domain is successively used to elongate the chain by formation of a peptide bond with the activated amino acid, except for the isopeptide bond catalyzed by C domain 5, while the growing peptide chain is tethered to peptidyl carrier proteins (PP). Finally, qinichelin is released from the last PP domain through an intramolecular nucleophilic substitution of the δ-hydroxylamino group of l-hOrn-6 to the carbonyl group of the thioester. However, it is challenging to understand the enzymology responsible for this reaction, because a usual thioesterase (TE) domain (e.g., in NRPS assembling gobichelin (42) and heterobactin[26]) required for peptide chain release is lacking in the C-terminus of QchJ. It is tempting to speculate that the C domain in module 6 catalyzes both the α-amidation of hOrn-6 to finalize the growing peptide chain and δ-amidation to self-cyclize the last hydroxyornithine (chOrn-6) to release the peptide chain from the NRPS system. A similar scenario for peptide chain release has recently been reported in the biosynthesis of scabichelin,[48] a pentapeptide siderophore containing a C-terminal cyclic hydroxyornithine residue as in qinichelin.

The ornithine building block for qinichelin assembly may originate from either the qch cluster or from the canonical arg gene cluster,[41] regulated by DmdR1 and ArgR, respectively. This would allow decoupling of qinichelin production from primary metabolism. The generated Orn precursor is further tailored, including hydroxylation at δ-NH2 by QchF, and/or acetylation at α-NH2 and δ-NH2 by QchQ. Alternatively, α-N-acetylation could arise from the bifunctional enzyme ArgJ (or its counterpart QchC) during ornithine precursor synthesis.[41] The characterization of qinichelin congeners (3–5) provides evidence for substrate flexibility of the AOrn domain in modules 2 and 5, whereby unmodified ornithine (Orn), δ-N-hydroxyl ornithine (hOrn), α-N-acetyl ornithine, δ-N-acetyl ornithine, and δ-N-hydroxyl-α-N-acetyl ornithine can be recognized and incorporated into the NRPS assembly. Still, we cannot rule out that QchQ post-translationally acetylates either free amine after construction of the final qinichelin. Indeed, it is difficult to discriminate between A domains activating Orn and/or hOrn through bioinformatics alone.[49] However, since qinichelin[2] was the major chemical output of the qch gene cluster, unmodified ornithine (Orn) and δ-N-hydroxyl ornithine (hOrn) are most likely preferred by AOrn in module 2 and module 5, respectively.

Conclusions

Actinomycetes adopt versatile strategies to biosynthesize structurally diverse secondary metabolites. This includes the production of a variety of siderophores, although it is not always clear what the advantage is in terms of the competition for iron in the environment. Functional crosstalk among multiple distantly located BGCs is not always predicted well by bioinformatics analysis. Therefore, chemical novelty may be missed if we solely rely on synthetic biology approaches, such as heterologous expression of a single BGC. The “protein-first” method, via reverse natural product proteomining, effectively identified the qch gene cluster expression in Streptomyces sp. MBT76 and further guided the characterization of qinichelins 2–5, a family of new catecholate–hydroxamate siderophores. The principles presented in this work can be exploited to discover a broader range of chemical frameworks and to elucidate other intertwined biosynthetic scenarios.

Experimental Section

Strains and Growth Conditions

Streptomyces sp. MBT76 isolation from Qinling mountain soil,[31] general growth conditions, and genome sequencing (GenBank accession number: LNBE00000000) have been described.[32,33] Here, Streptomyces sp. MBT76 was grown in a liquid NMMP medium[50] containing 1% (w/v) glycerol and 0.5% (w/v) mannitol as carbon sources, but lacking polyethylene glycol. This basic NMMP media were perturbed by using four different additives (or no additive) to create varying growth conditions: (A) no additive, (B) 2% (w/v) NaCl, (C) 1% (w/v) starch, (D) 0.8% (w/v) Bacto peptone (Difco), and (E) 0.6% (w/v) Bacto yeast extract (Difco). For the iron-starvation study, the minor element solution[50] was omitted from condition A. All the cultures of Streptomyces sp. MBT76 were incubated at 30 °C for 72 h, with constant shaking at 220 rpm.

Proteomics

Streptomyces sp. MBT76 cells were lysed using acetone/SDS as described.[51] Around 167 μg of protein was precipitated for each sample using chloroform/methanol[52] and then dissolved using RapiGest SF surfactant (Waters). The proteins were further digested with trypsin after iodoacetamide treatment,[53] and the resulting primary amines of the peptides were dimethyl labeled using three combinations of isotopomers of formaldehyde and cyanoborohydride on Sep-Pak C18 200 mg columns (Waters), via CH2O + NaBH3CN, CD2O + NaBH3CN, and 13CD2O + NaBD3CN, as described.[54] Light-, medium-, and heavy-labeled peptides with 4 Da mass differences were mixed 1:1:1 to obtain 0.5 mg for fractionation by cationic exchange (SCX) chromatography using a polysulfethyl A column (PolyLC, 100 × 2.1 mm, particle size 5 μm, average pore size 200 Å). Mobile phases for SCX chromatography consisted of solvent A (10 mM KH2PO4, 20% (v/v) acetonitrile, pH 3) and solvent B (10 mM KH2PO4, 20% (v/v) acetonitrile, 0.5 M KCl, pH 3). The running program for SCX was a gradient of 0–18% solvent B in 18 CV (column volume), 18–30% solvent B in 6 CV, and 30–100% solvent B in 5 CV, at a constant flow rate of 250 μL min–1. In total, 24 peptide fractions were collected for LC-MS/MS analysis on an LTQ-Orbitrap instrument (Thermo).[53] Data analysis was performed using MaxQuant 1.4.1.2,[55] whereby MS/MS spectra were searched against a database of translated coding sequences obtained from the genome of Streptomyces sp. MBT76. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository with the data set identifier PXD006577.

LC-MS Analysis of Metabolites

Spent medium samples were acidified with 1% (v/v) formic acid final concentration) and desalted using StageTips.[56] Next, 20 μL of samples were separated on an Finnigan Surveyor HPLC (Thermo) equipped with a Gemini C18 column (Phenomenex, 4.6 × 50 mm, particle size 3 μm, pore size 110 Å) at a flow rate of 1 mL min–1 and using a 0–50% acetonitrile gradient buffered with 0.1% (v/v) TFA in 10 CV. Mass spectrometry was performed using an Finnigan LCQ advantage (Thermo) equipped with an ESI source in the positive mode and scanning at 160–2,000 m/z.

For high resolution LC-MS/MS analysis on an LTQ-orbitrap the same setup was used as above for proteomics analysis,[53] but using different run parameters. Mobile phases were as follows: (A) 0.1% (v/v) formic acid in H2O and (B) 0.1% formic acid in acetonitrile. A 30 min 10–20% B gradient was followed by a 15 min 20–50% B gradient, both at a flow rate of 300 μL/min split to 250 nL min–1 by the LTQ divert valve. For each data-dependent cycle, one full MS scan (100–2000 m/z) acquired at a resolution of 30 000 was followed by two MS/MS scans (100–2000 m/z), again acquired in the orbitrap at a resolution of 30 000, with an ion selection threshold of 1 × 107 counts but no charge exclusions. Other fragmentations parameters were as described for the proteomics analysis.[53] After two fragmentations within 10 s, precursor ions were dynamically excluded for 120 s with an exclusion width of ±10 ppm. The data have been deposited in the GNPS repository (http://gnps.ucsd.edu/) with data set identifier MSV000081504.

Isolation of Qinichelins

A total of 5 mL of spent medium from NMMP-grown cultures was desalted on Sep-Pak SPE C18 200 mg columns (Waters). Columns were first washed with 1 mL of 80% (v/v) acetonitrile + 0.1% (v/v) formic acid and then equilibrated with 1 mL of 0.1% (v/v) formic acid. A total of 5 mL of spent medium was mixed with 1 mL of 5% (v/v) formic acid and loaded onto the column. After s wash with 1 mL of 0.1% (v/v) formic acid, the column was eluted with 600 μL of 80% (v/v) acetonitrile + 0.1% (v/v) formic acid. The resulting sample was dried in a speedvac to remove acetonitrile and resuspended in 900 μL of 3% (v/v) acetonitrile + 0.1% (v/v) formic acid. This desalted sample was separated by HPLC on an Agilent 1200 series instrument equipped with a Gemini C18 column (Phenomenex, 250 × 10 mm, particle size 5 μm, pore size 110 Å), eluting with a gradient of acetonitrile in H2O adjusted with 0.15% (v/v) trifluoroacetic acid from 6% to 12%. The HPLC run was performed in 3 CV at a flow rate of 5 mL/min, and the fractions were collected based on UV absorption at 307 nm. All fractions were analyzed by LC-MS (positive mode) to check the existence of the targeted mass at m/z 772.3. The fraction of interest was lyophilized and subsequently reconstituted in deuterated water (D2O) for NMR (850 MHz) measurement.

DmdR1 and ArgR Regulon Predictions

The putative binding sites for the iron utilization regulator DmdR1 and for the arginine biosynthesis regulator ArgR were detected on the chromosome of Streptomyces sp. MBT76 using the PREDetector software[57] and according to the method described.[58] For the generation of the DmdR1 position weight matrix (PWM), we used the sequence of the iron box which lies at position −82 nt upstream of desA (SCO2782) and was previously shown to be bound by DmdR1 in S. coelicolor.[59] In order to acquire more highly reliable iron boxes to generate the PWM, we scanned the upstream region of the orthologues of desA in five other Streptomyces species and retrieved their respective iron boxes (see Supporting Information Figure S8). A set of ARG boxes experimentally validated in S. clavuligerus(60) and S. coelicolor(45) were used to generate the ArgR PWM (see Supporting Information Figure S9).