Targeted Discovery of β‑Branched Conjugated Polyketides from Bacteria Based on Genomic and Metabolomic Hallmarks.

Introduction
Polyketides are natural products that
are biosynthesized using
simple C-2 building blocks to manufacture diverse bioactive small
molecules with complex structures. Among
polyketides produced by microorganisms in nature, the major group
consists of α-branched polyketides originating from side chain-bearing
building blocks loaded by the acyltransferase (AT) domain. β-branched polyketides are much rarer because
of the difficulty of introducing alkyl branches at electrophilic β-carbon. However, bacteria have developed a unique β-branching
mechanism involving the aldol addition of acetate to the electrophilic
β-carbon of the keto-thioester, which is catalyzed by a key
enzyme, hydroxymethylglutaryl-CoA synthase (HCS) homolog. Subsequent
dehydration by enoyl-CoA hydratase, which is always found together
with HCS homolog in β-branched polyketide biosynthetic pathways,
generates a conjugated double bond in most cases and often produces
β-branched conjugated polyketides.
Even though β-branched
conjugated polyketides have been occasionally
discovered from bacteria, they often display significant bioactivity.
One of the well-known bioactive examples is mupirocin, which is now
being used as a topical ointment antibiotic drug (Bactroban) against
bacterial skin infections. The mechanism
of action of mupirocin involves the inhibition of bacterial isoleucyl-tRNA
synthetase. Bryostatin-1, which bears
a β-branching methyl acrylate moiety conjugated with a double
bond, was initially isolated from the marine bryozoan Bugula neritina, and
its actual origin was elucidated as a bacterial symbiont. This β-branched conjugated polyketide was
designated as an orphan drug by the Food and Drug Administration for
the treatment of Fragile X syndrome in 2015 (Figure
a). Despite their
significant pharmaceutical potential, β-branched conjugated
polyketides have been randomly reported without a systematic search
method, hampering their logical structural diversification. Occasional
discoveries since the 1950s include the following examples. Pulvomycin,
a 22-membered macrocyclic lactone that bears a triene and two trienone
conjugated systems, was discovered in Streptomyces sp. in 1957 and structurally revised in 1985. It exhibits significant
antibacterial activity through a unique mode of action by preventing
bacterial translational elongation factor Tu·GTP from complexing
with aminoacyl-tRNA.
−

Leinamycin, an 18-membered macrolactam bearing conjugated
and β-branched moieties discovered in Streptomyces
atroolivaceus, displays both antibacterial and antitumor
activities. Kalimantacin, a β-branched
conjugated polyketide antibiotic hybrid with a non-ribosomal peptide
produced by Alcalgenes sp., exhibits cytotoxicity
against two cancer cell lines, P388 mouse leukemic cells and HeLa
S3 cells.
,
Virginiamycin M1, a β-branched conjugated
macrolide antibiotic produced by Streptomyces pristinaespiralis, shows bioactivity against Micrococcus aureus (Figure
a).

Because of the high pharmaceutical potential of
β-branched
conjugated polyketides but the limited logical exploration of their
structural diversity, we desired to develop a metabologenomic targeting
discovery method for new β-branched conjugated polyketides.
We screened our in-house bacterial DNA library (over 1,600 strains
mostly without full genome data) with a degenerate PCR primer set
targeting the hallmark gene encoding HCS homolog, which plays a key
role in biosynthesizing β-branched conjugated polyketides. In addition, vibronic fine structures in ultraviolet
(UV) spectra were utilized as metabolomic
hallmarks for the first time in natural product discovery, which substantially
accelerates the discovery of new β-branched conjugated polyketides.
Here, we report the method development and applications demonstrating
the discovery of seven new bioactive β-branched conjugated polyketides,
paenillaene (1), paeniformicins A–D (2–5), and pulvomycins E–F (6–7).

Results and Discussion

Design of a Degenerate
PCR Primer Set Targeting HCS Homolog-Encoding
Gene
β-branched modification in polyketides is an ingenious
mechanism involving the aldol addition of acetate to the electrophilic
β-carbon of the keto-thioester, which is catalyzed by a key
enzyme, HCS-homolog. A freestanding T-domain and ketosynthase domain
are also involved. Subsequent dehydration and decarboxylation proceed
are followed by two enoyl-CoA hydratase (ECH)-mediating steps to generate
a single α, β- or β, γ-unsaturated β-branched
alkyl group (Figure
b), which contributes to conjugated double bond formation. The HCS homolog-encoding gene, which is indispensable
in the β-branching cassettes, was selected as a genomic hallmark
to mine β-branched polyketide-producing bacterial strains.
Analysis of the four representative HCS homolog-encoding genes from
β-branched conjugated polyketide biosynthetic gene clusters
in kalimantacin-producing Streptomyces sp. SLBN-134, leinamycin-producing S. atroolivaceus S-140, pulvomycin-producing S. sp. HRS33, and the virginiamycin
M1-producing strain S. virginiae
enabled the design of a degenerate PCR primer
set targeting the HCS homolog-encoding gene (Figure S1). The HCS homology gene primer set was validated using genomic
DNA of S. sp. HRS33 (pulvomycin), S. sp. MBP16 (hamuramicin C),
S. sp. DM28 (dumulmycin), and S. sp. BA01 (formicolide).

Genomic Hallmark-Based DNA Library Screening and Phylogenetic
Analysis of PCR Amplicons
We applied the HCS-like enzyme-encoding
gene primer set for screening our in-house DNA library (1,638 bacterial
gDNA samples from various natural sources, including intertidal mudflats,
volcanic islands, Siberian ice wedges, salterns, deep-sea sediment,
and insects) and identified 67 hit strains possessing HCS homologs
(hit rate: 4.1%) (Figure
c). A phylogenetic tree was generated to analyze the relationships
among the HCS homolog gene amplicon sequences (∼680 bp) of
the 67 hit strains, along with 27 reference sequences extracted from
reported β-branched conjugated polyketides (Figure
and Figure S2). The degenerate primers used in this study were designed
to target conserved motifs of HCS-like enzymes associated with canonical
β-branching cassettes, which have been predominantly characterized
in actinobacterial trans-AT PKS pathways. Accordingly,
the phylogenetic analysis presented in Figure
reflects a representative landscape of HCS
homolog-encoding genes accessible through this targeted strategy,
rather than the full natural diversity of HCS-like enzymes. Despite
the actinobacterial gene-based design, the primer set successfully
identified HCS homolog-encoding genes from phylogenetically diverse
bacteria, including low-G+C Gram-positive genera (Bacillus and Paenibacillus: 9 strains) and Gram-negative
proteobacteria (Pseudomonas and Alcaligenes: 3 strains) among the 67 hits, supporting the utility of this approach
for uncovering β-branched polyketides from diverse phylogenetic
taxa.
The amplicons were sorted into seven monophyletic
clades. Clade
1 included the reference strain producing virginiamycin M1 and 26
hit strains. Most of these strains were Streptomyces, but two strains were identified as Bacillus. The
reference strains whose metabolites were the exo-β-methylene
or β-carboxymethyl group-bearing compounds guangnanmycin, weishanmycin, and
largimycin,
endo-β-methoxymethyl
compound myxovirescin A, leinamycin with
an unusual 1,3-dioxo-1,2-dithiolane moiety, ripostatin with a carboxymethyl-β-branch installed by a trans-acting HCS homolog were assigned to clade 2, together
with 19 hit strains.

Clade 3 had
only one strain, Marinospora sp.,
which yielded a unique dimeric macrolide with β-methylation,
marinomycin. Clade 4 had a single hit
strain of Bacillus sp. WA1CD. Clade 5 incorporated
various β-branched polyketide-producing Streptomyces strains of kalimantacin, pulvomycins, hamuramicin C, dumulmycin, and formicolides which were mainly macrocyclic lactones with β-methylation,
together with 15 hit strains. Clade 6 was composed of a Burkholderia sp., which produced spliceostatin A, and two Burkholderia gladioli strains
that produced a macrocyclic lactone gladiolin and bongkrekic acid with a carboxymethyl-β-branch.

Clade 7 was massive with ten reference
strains, but only six hit
strains. First, two single hit strains of Paenibacillus sp. GAI108 and Pseudomonas sp. GIN12 were represented.
Thereafter, in this large clade, one or two reference strains were
identified at each descending taxonomic tier, including the reference
strains of mupirocin, lasonolide A, β-methyl and β-methylene bearing
leptolyngbyalide and phormidolide, oocydin (a chlorinated-branched polyketide natural
product), curacin A, jamaicamide A, bryostatin, difficidin, and
bacillaene.
,
Specifically, one hit strain, Paenibacillus sp. BYK1458, formed a small subclade with
three more Paenibacillus strains in the core area
of clade 7 (Figure
).
The phylogenetic tree of the HCS homolog gene amplicons
(∼680
bp) enabled the prioritization of strains for further chemical analysis
to discover structurally new β-branched polyketides. In the
phylogenetic tree, most of the hit strains were actinobacteria; however,
distinctive Paenibacillus strains existed (GAI108,
BYK1456, BYK1457, BYK1458, and BYK1468), which were classified only
in clade 7. Because the chemistry of Paenibacillus had not been extensively studied with respect to secondary metabolites,
in contrast to actinobacteria, we primarily investigated the metabolites
of these Paenibacillus strains along with the other
hit strains.

Detection of β-Branched Polyene Polyketides
Based on Vibronic
Fine Structures in UV Spectra as Metabolomic Hallmarks
Conjugated
systems in β-branched polyketides confer chromophores, providing
UV absorbance usually in the range of 230–340 nm. The extended
conjugations of polyenes exhibit strong π → π*
electronic transitions at designated wavelengths (λmax) depending on the numbers of conjugations. Moreover, the UV spectra of polyenes display characteristic well-resolved
vibronic fine structures, which are observed as line graphs with breaks
or sharp multiple peaks in their UV spectra (Figure S3) because of the relatively high symmetry, minimal orbital
mixing with heteroatoms, and clear Franck-Condon progressions coupled
with an electron transition with a specific vibrational mode.

The vibronic fine structures in the UV
spectra enable the unequivocal detection of conjugated double bond-bearing
polyketides because they are distinct from those of aromatic chromophores,
the general features of polyketide type II natural products, which
are not well resolved owing to degenerate electronic states, ring
conjugation effects, and more complicated vibrational modes (Figure S4). In addition,
the olefinic double bonds in β-branched polyketides can be conjugated
with carbonyl groups. The UV spectra of compounds with conjugated
carbonyl systems can be easily identified from those of aromatic compounds
because they typically show broadened vibronic fine structures modified
from sharp multiple polyene features owing to electronic perturbation
by the n → π* transition. Therefore, vibronic fine structures can be valuable metabolomic
hallmarks of β-branched conjugated polyketides potentially produced
from hit strains selected by the genomic hallmark of the HCS homolog-encoding
gene.
Vibronic fine structure in UV spectra is not exclusive
to β-branched
polyketides and can also be observed in other conjugated metabolites,
including generic polyenes. Accordingly, in the present workflow,
vibronic fine structure is not used as a standalone diagnostic criterion
but rather as an initial metabolomic prioritization feature. Selectivity
for β-branched polyketides is achieved through orthogonal integration
of this UV-based hallmark with genomic hallmark-based PCR screening
targeting HCS homolog-dependent β-branching cassettes, such
that only strains harboring the canonical β-branching machinery
are advanced to LC-MS-based metabolomic analysis and UV spectral inspection.
To further evaluate the possibility of false positives, we surveyed
the peer-reviewed literature for reports describing a single bacterial
strain that produces both a β-branched conjugated polyene and
a non-β-branched conjugated polyene. To the best of our knowledge,
documented examples of such coincident production are rare or absent
in the current literature. Consistent with this observation, our chemical
analyses of the 67 hit strains and reference strains did not reveal
concurrent production of β-branched conjugated polyenes and
non-β-branched conjugated polyenes. Collectively, these results
suggest that, although vibronic fine structure itself is not exclusive,
coincident detection of generic conjugated polyenes within the present
metabologenomic targeting framework is likely to be rare.
We
cultivated all 67 hit strains in 50 mL liquid medium and analyzed
the chemical profiles of the extracts by liquid chromatography-mass
spectrometry (LC-MS) coupled with a diode array UV detector (see Supporting Information). LC-MS was used to acquire
the full UV (200–700 nm) and MS (100–2000 Da) spectra
of the chemical constituents in the extracts. We then searched for
UV spectra displaying the hallmark vibronic fine structures originating
from polyenes and identified the production of β-branched conjugated
polyketides. This process also rapidly dereplicated the detected polyketides
based on our in-house UV library with 1,803 microbial natural products.
Based on the phylogenetic tree showing the seven clades, we logically
searched for new β-branched conjugated polyketides by utilizing
the UV metabolomic hallmark. In clade 1, which contains the virginiamycin
producer, three Streptomyces strains (ASAS0305, AMD38,
and PSN02) were found to produce mainly virginiamycin M1 (Figure S5) by strong absorption at 230 nm in
its UV spectrum and a positive molecular ion [M + H]+ at m/z 526 along with the NMR spectra of the
compound isolated from ASAS0305 (Table S13 and Figures S88 and S89). These strains
(ASAS0305, AMD38, and PSN02) originate from different niches such
as mountains and intertidal mudflats. However, they are closely related
in the phylogenetic tree of the HCS homolog gene amplicons (Figure
) and produce the
identical compound, rationalizing their homologous chemistry and demonstrating
the validity of our metabologenomic targeting method.
One strain
in clade 2 (Streptomyces sp. C2-2_21)
produced weishanmycin A2 (Figure S5), which
belongs to the leinamycin family, the major family of clade 2. Clades
3 and 4 both contained only one strain. For clades represented by
a single strain, the current sample size is insufficient to infer
statistically meaningful trends. In clade 5, the vibronic fine structures
in UV were detected in the extracts of the reference strains, confirming
the production of pulvomycins, formicolides, dumulmycin, and hamuramicin
C in Streptomyces sp. HRS33, Streptomyces sp. BA01, Streptomyces sp. DM28, and Streptomyces sp. MBP16, respectively (Figure S5).
Furthermore, the extract of Streptomyces sp. YJD131,
which is most closely related to the reference pulvomycin producer,
was detected to produce metabolites with the UV vibronic fine structures
analogous to those of pulvomycins (Figure S13) but different molecular ions in the MS spectra, indicating new
members of the pulvomycin family and thus requiring full structure
elucidation. Clade 6 showed only three reference strains without a
hit, not allowing for further discovery of β-branched conjugated
polyketides.
Interestingly, two Paenibacillus strains (GAI108
and BYK1458) in chemically diverse clade 7 produced metabolites displaying
the hallmark vibronic fine structures of polyenes (Figure S13). Because the polyene compounds were not in our
UV library and the genus Paenibacillus was chemically
underexplored, these strains were prioritized for further scaling-up
of the cultures and subsequent chemical analysis.

Structural
Elucidation and Bioinformatic Analysis of Biosynthesis
of New β-Branched Polyketides
The cultivation of Paenibacillus sp. BYK1458 yielded 1, which
was named paenillaene (Figure
). Analysis of its HR-ESI-MS spectrum revealed a negative
ion [M-H]− at m/z 608.3223; thus, we determined the chemical formula of 1 to be C35H47NO8, possessing 13
double bond equivalents. One-bond 1H–13C correlations were assigned based on 1H and HSQC NMR
data (Table S4 and Figures S14–S19). Five six-membered aromatic ring protons [δH 7.82
(2H), 7.52, and 7.44 (2H)] and seven olefinic protons (δH 6.63, 6.58, 6.52, 5.93, 5.92, 5.65, and 5.22) were observed
in the downfield region. One amide proton was identified at 8.30 ppm.
Three oxygenated methine protons were detected at 4.75, 3.87, and
3.78 ppm, and nitrogenous methylene protons were identified at 3.52
and 3.44 ppm. Further interpretation revealed the existence of 25
aliphatic protons, including 12 methyl protons [δH 1.78 (3H), 1.70 (3H), 1.64 (3H), and 0.86 (3H)], 12 methylene protons
[δH 2.60, 2.35, 2.28, 2.23, 2.22, 2.16, 2.14 (2H),
2.04 (2H), 1.63, and 1.48], and one methine proton (δH 2.25). Analysis of the 13C NMR data along with the 1H and HSQC spectra assigned one ketone (δC 210.9), two carbonyl carbons (δC 173.5 and 167.0),
16 double-bond carbons (δC 140.9–126.4), four
oxygenated carbons (δC 98.6, 73.5, 65.9, and 65.3),
one nitrogen-bearing carbon (δC 48.3), and 11 aliphatic
carbons (δC 48.6–9.2). The partial structures
of 1 were assembled using COSY and HMBC correlations
(Figure
). First,
the spin system from C-2 to C-5 was assigned by COSY correlations
from H2-2 (δH 2.22 and 2.35) to H-5 (δH 4.75), and the second spin system from C-7 to C-10 was elucidated
based on the consecutive COSY correlations from H-7 (δH 5.92) to H-10 (δH 5.93). The C-12 to C-16 fragment
was constructed by a series of COSY cross-peaks from H-12 (δH 6.63) to H-16 (δH 5.22). A small partial
structure of C-18–C-19–C-20–C-35 was confirmed
by the COSY correlations of H2-18 (δH 2.23
and 2.16), H-19 (δH 3.87), H-20 (δH 2.25), and H3-35 (δH 0.86). The last
spin system belonging to a six-membered aromatic ring was constructed
by the COSY correlations among H-27, H-28, H-29 (δH 7.52), H-30
(δH 7.44), and H-31 (δH 7.82).
The carbonyl carbon C-1 was assigned next to C-2
based on the H2-2 (δH 2.35 and 2.22)/C-1
(δC 173.5) two-bond 1H–13C correlations.
The HMBC correlation of H3-32 (δH 1.70)
to C-5 (δC 65.9), C-6 (δC 140.9),
and C-7 (δC 126.6) installed the C-32 methyl group
at C-6, connecting the two spin systems to make the C-2 to C-10 chain.
This chain structure was further extended to C-16 by the HMBC correlations
from H3-33 (δH 1.78) to C-10 (δC 128.5), C-11 (δC 132.5), and C-12 (δC 127.1). The long-range heteronuclear correlations from H3-34 (δH 1.64) to C-16 (δC 126.4), C-17 (δC 132.3), and C-18 (δC 42.7) constructed the C-2 to C-35 linear structure. On the
opposite side, a benzamide moiety was assembled from the aromatic
spin system from C-27 to C-31 with the H-28 (δH 7.44)/C-26
(δC 133.6), H-27 (δH 7.82)/C-25­(δC 167.0), and 24-NH (δH 8.30)/C-25 (δC 167.0) HMBC correlations (Figure
).
H2-24 (δH 3.52 and 3.44) displayed
COSY correlation with 24-NH (δH 8.30), and HMBC correlations
to the amide carbonyl carbon C-25 (δC 167.0) and
the dioxygenated carbon C-23 (δC 98.6), thus assigning
this methylene between C-23 and 24-NH. C-23 was then connected to
C-22 and C-24 based on the H2-22 (δH 2.60
and 2.28)/C-23 (δC 98.6), H2-24 (δH 3.52 and 3.44)/C-23 (δC 98.6) and 24-NH
(δH 8.30)/C-23 (δC 98.6) HMBC correlations.
H3-35 (δH 0.86) and H2-22 (δH 2.60 and 2.28) showed an HMBC correlation to the ketone carbon
C-21 (δC 210.9), extending the chain from C-20 to
the terminal benzamide.
The constructed structure contained
all 35 carbons in the molecular
formula, including five olefinic double bonds, one benzene ring, and
three carbonyl groups, accounting for 12 out of 13 unsaturations.
Therefore, paenillaene (1) must contain an additional
ring. An HMBC correlation from H-19 to C-23 established the connectivity
between C-19 and C-23 (Figure S21) through
an ether bond and constructed the tetrahydropyranone moiety. The ROESY
correlations of H-19 (δH 3.85)/H2-24 (δH 3.52 and 3.44) and H2-22 (δH 2.60,
2.28)/H2-24 further supported the formation of tetrahydropyranone,
satisfying the last double bond equivalent. Subsequently, the four
invisible −OHs were assigned to a carboxylic acid group at
C-1 and three hydroxy groups at C-3, C-5, and C-23. The ROESY correlations
shown in Figure
determined
the geometries of the double bonds and the relative configuration
of the tetrahydropyranone. Therefore, paenillaene (1)
was elucidated as a structurally novel polyketide incorporating two
β-methyl groups at C-11 and C-17 (C-33 and C-35 methyl groups)
and a tetraene showing the polyene UV hallmark, along with a benzamide
and a tetrahydropyranone.
Whole-genome sequencing of Paenibacillus sp. BYK1458
yielded a draft genome approximately 6.09 Mb in size. We utilized
Anti-SMASH 6.0.0 to identify BGCs in the genome (Table S5 and Figure S23). A trans-AT PKS
and NRPS hybrid BGC containing 13 modules matched the structure of
paenillaene (1). A detailed analysis of this gene cluster
identified the β-branch cassette of PnlN–R, which was
consistent with the elucidated structure. Furthermore, the absolute
configuration of the hydroxy group-bearing stereogenic centers, which
could not be determined by the modified Mosher’s method because
of the instability of 1 during derivatization under basic
conditions, could be predicted by analyzing the presence of a diagnostic
Asp residue at the third position within an LDD signature
motif in the KR domains. The B1-type
KR domains (KR2, KR7, and KR8) supported 3R, 5S, and 19S configurations (Table S6). Thus, we proposed 3R, 5S, 19S, 20S, and 23S configurations for paenillaene (1). The double-bond geometry was also confirmed using the
KR-type principle.

Chemical investigation
of the culture of Paenibacillus sp. GAI108 led to
the discovery of paeniformicins A–D (2–5). Paeniformicin A (2),
a predicted β-branched conjugated polyketide (Figure
), was purified as a yellow
powder. Based on HR-ESI-MS analysis, its molecular formula was determined
to be C33H50O6, with an unsaturated
number of 9. Combined analysis of 1H and HSQC NMR data
revealed one-bond carbon–proton correlations (Table S7).
The structural fragments were connected using
COSY and HMBC correlations.
First, the spin system from C-2 to C-18 was assigned by the consecutive
COSY correlations from H2-2 (δH 2.36 and
2.27) to H-18 (δH 5.27), and the second spin system
from C-19 to C-23 was constructed by the corresponding COSY correlations
from H-19 (δH 5.28) to H-23 (δH 5.29).
The last spin system in the tail was identified by the COSY correlations
of H-25 (δH 5.99) through H2-30 (δH 5.29 and 5.14). The linkage between C-18 and C-19 was confirmed
by the H-17 (δH 3.80)/C-19 and H-20 (δH 2.25)/C-18 HMBC correlations. The four hydroxy groups were
assigned to C-3 (δC 64.8), C-5 (δC 68.9), C-11 (δC 66.9), and C-17 (δC 71.5) by their corresponding HMBC correlations of 3-OH (δH 4.71)/C-3, 5-OH (δH 4.62)/C-5, 11-OH (δH 4.22)/C-11, and 17-OH (δH 4.64)/C-17. The
methyl group C-33 (δC 24.1) was located at the fully
substituted olefinic C-24 (δC 134.1), and the two
fragments were merged into one by the key HMBC correlations of H3-33 (δH 1.86 (3H))/C-24, H-26 (δH 6.09)/C-24 (δC 134.1), and H2-22 (δH 2.23/2.13)/C-24 (δC 134.1)
(Figure
).
The ester linkage was secured by the H-21 (δH 4.71)/C-1
(δC 170.9) HMBC correlation to elucidate the planar
structure of paeniformicin A (2) as a new 22-membered
macrocyclic lactone with one β-branched methyl group and a tetraene
moiety displaying the characteristic vibronic fine structure in the
UV spectrum (λmax = 310 nm) as a metabolomic hallmark
(Figure S13c). The double-bond geometries
were assigned as 6E, 14E, 18E, 23Z, 25Z, and 27E based on J-coupling constants and ROESY
correlations of 2 (Table S7 and Figure
).
Chiral centers C-9/C-11 and C-20/C-21 were assigned as 9S*/11R* and 20R*/21S* by J-based configuration analysis with
vicinal 1H–1H and two- or three-bond 1H–13C coupling constants, along with ROESY
correlations (Figures S30 and S33).
,
The absolute configurations of alcoholic chiral centers C-3, C-5,
C-11, and C-17 were determined as 3S, 5R, 11R, and 17R by the modified
Mosher’s method (Figures S34–S40). For the stereochemistry of C-20 and C-21, conformational search
and DP4 computational calculation were applied to propose 20R and 21S with 100% probability (Figure S41).

The UV spectrum of paeniformicin B (3) indicated the
modification of the tetraene moiety in 2 to a triene
moiety because it displayed a hypsochromic shift from 310 to 280 nm
(Figures S7 and S8). HR-ESI-MS analysis
revealed its molecular formula to be C37H54O9, possessing 11 double bond equivalents. Interpretation of
the 1H, 13C, and 2D NMR data (Table S7) illustrated that paeniformicin B (3) shares the same macrocyclic skeleton from C-1 to C-21 as 2. Furthermore, based on the COSY correlations of methylene
protons H2-28 (δH 2.94) with H-29 (δH 5.81) and H-27 (δH 5.48), we assigned aliphatic
methylene carbon C-28 between C-27 and C-29. C-29 was then connected
to the terminal olefinic methylene group, C-30, whereas the C-27 end
was expanded to C-26 and C-25 via H-27/H-26 and H-26/H-25 homonuclear
coupling. The side chain bearing a triene instead of a tetraene was
identified using the H-21/H-22 and H-22/H-23 COSY correlations, along
with the HMBC correlations from H3-33 to C-23, C-24, and
C-25. An additional ester bond between C-11 and C-34 was identified
by the key HMBC correlation H-11 (δH 4.87)/C-34 (δC 171.9). The last partial structure was determined to be a
succinate side chain using COSY and HMBC correlations, completing
the structure of 3 as a new β-branched polyketide
bearing a triene (Figure
).
The molecular formula of paeniformicin C (4) was confirmed
as C33H52O7 by the molecular ion
in the HR-ESI-MS spectrum. The NMR spectroscopic analysis (Table S7) revealed that its carbon skeleton is
very similar to that of 3 except for the absence of a
succinic side chain. However, paeniformicin C (4) did
not form a macrocycle (Figure
).
Paeniformicin D (5) possessed the molecular
formula
C37H56O10, and its carbon chain was
the same as that of 3 according to the NMR spectroscopic
analysis (Table S7). Further NMR analysis
revealed that 5 has a linear structure without macrolactone
formation based on the relatively upfield signal of H-21 (δH 3.99) along with additional H2O in the molecular
formula compared to 3 (Figure
). Detailed structural elucidation of compounds 3–5 is provided in the Supporting Information.
Bacteria culture, extraction,
and isolation of Streptomyces sp. YJD131 led to the
discovery of pulvomycins E and F (6 and 7). Pulvomycin E (6), a yellow powder
with the molecular formula C48H68O13 (15 degrees of unsaturation), was characterized using HR-ESI-MS
and NMR (1H, 13C, COSY, HSQC, HMBC and ROESY)
data. Its planar structure features two ketones, an ester, ten olefins,
oxygenated carbons, and three methoxy groups. Key spin systems and
HMBC correlations established connectivity, while ROESY and J-coupling constants confirmed geometry for most double
bonds and relative configuration in the sugar moiety. J-based configuration analysis and ROESY correlations defined relative
configurations. The CD spectrum of 6 matched that of
pulvomycin A, confirming its absolute configuration (Figure S82). Pulvomycin F (7), with the same
molecular formula of C48H68O13 as 6, differed in double-bond position and geometry (4Z instead of 4E), as supported by NMR and
CD comparison with pulvomycin B (Figure S83). The methoxy group was also shifted from C-5 to C-3 compared with 6. (Figure
)

Biological Activity
The antimicrobial activities of
the new β-branched conjugated polyketides 1–7 were evaluated against human pathogenic bacterial and fungal
strains (Staphylococcus aureus, Enterococcus faecalis, Enterococcus
faecium, Klebsiella pneumoniae, Salmonella
enterica, Escherichia coli, Candida albicans, Aspergillus fumigatus, Trichophyton rubrum, and Trichophyton mentagrophytes) (Tables S14 and S15). Paeniformicin B (3) showed
high antibacterial activity against S. aureus (MIC = 0.5 μg/mL) and K. pneumoniae (MIC
= 8.0 μg/mL), and paeniformicin A (2) displayed
moderate antibacterial activity against S. aureus (MIC = 32 μg/mL). Paenillaene (1) showed mild
bioactivity against several strains [S. aureus (MIC = 64.0 μg/mL), E. faecalis (MIC = 64.0
μg/mL), and E. faecium (MIC =
64.0 μg/mL)]. Based on the structural features of 1-5, the macrolactone scaffold bearing a limited number
of hydroxyl groups, together with the succinate moiety, might contribute
to the enhanced antibacterial activity of compound 3.
Further investigation of their S. aureus sortase A (SrtA) inhibition assay was conducted, and paeniformicin
A (2) showed moderate bioactivity (IC50 =
34.2 μg/mL) (Table S16). Especially paeniformicin D (5) exhibited significant
bioactivity against Mycobacterium tuberculosis mc2 6230 (MIC50 = 0.27 μM), whereas
the other compounds showed no anti-tuberculosis activity (Table S17).
The cytotoxicity of compounds 1–7 was evaluated against several human
cancer cell lines. 1 showed mild cytotoxicity against
colon cancer (HCT116) and stomach cancer (SNU638) cell lines (IC50 of 38.0 and 37.2 μM, respectively), whereas paeniformicins
showed no cytotoxicity (Table S18). Pulvomycin
E (6) showed high cytotoxic effects against all the tested
cancer cell lines with IC50 values of 7.58, 1.28, 6.58,
4.17, 4.61, and 3.11 μM for A549, HCT116, SNU-638, SK-HEP-1,
MDA-MB-231, and docetaxel-resistant MDA-MB-231 (MDA-DTX) cell lines,
respectively. 7 also presented mild cytotoxicity against
these human cancer cell lines (Table S19).
We further examined the biological effects of structurally
novel
paenillaene (1) against amyloid-β (Aβ) oligomers,
which are the major pathological and biological hallmarks in the brains
of patients with Alzheimer’s disease (AD). Because the most promising therapeutic strategy for AD
is to remove preformed aggregates of Aβ, we utilized an Aβ42
aggregate dissociation assay to evaluate the dissociation ability
of 1. We observed a significant
and dose-dependent reduction of the fluorescence signal, indicating
that 1 induced dissociation of Aβ oligomers, with
a 29% reduction at 5 μM and 57% at 50 μM, suggesting that 1 may reduce preformed Aβ aggregates in the AD brain.
Additional animal and clinical studies are warranted to determine
whether 1 may translate into a therapeutic agent (Figure S90).

Conclusions
The
genomic hallmark-based PCR screening,
which specifically targets
the homologous DNA sequence encoding HCS homolog, the essential enzyme
biosynthesizing β-branched conjugated polyketides, enabled the
selection of bacterial strains potentially producing β-branched
conjugated polyketides without full genome analysis. By employing
phylogenetic analysis of the partial HCS homolog-encoding gene sequences
obtained through PCR amplification (∼680 bp), we successfully
sorted various types of β-branched conjugated polyketides into
seven clades to construct their clade-structure atlas in nature, which
guided the selection of strains to be investigated further for the
discovery of structurally novel β-branched polyketides. Vibronic
fine structures in UV spectra were utilized as metabolomic hallmarks
in natural product discovery and facilitated the rapid identification
of β-branched conjugated polyketides in small-scale cultures.
Further cultivation and chemical analyses of the hit strains led
to the identification of β-branched conjugated polyketides from
four clades except for three clades containing only one strain or
reference strains. Such clade-based chemical investigation efficiently
discovered seven new β-branched polyketides: paenillaene (1), paeniformicins A–D (2–5), and pulvomycins E–F (6–7) from chemically prolific clades 5 and 7 and identified
previously reported virginiamycin M1 and weishanmycin A1 from clades
1 and 2 as predicted by the phylogenetic clade-structure atlas (Figure
). The structures
of the new compounds, including their stereochemistry, were elucidated
using spectroscopic, genomic, and computational analyses. In particular,
paenillaene (1), bearing a benzamide, tetrahydropyran,
and tetraene chain, showed high structural novelty, possibly because
of its unique starting unit, benzoic acid, which was subsequently
connected to glycine by the trans-AT PKS and NRPS
hybrid pathway.
Evaluation of the new β-branched polyketides
revealed significant
anti-tuberculosis activity of paeniformicin D (5) against Mycobacterium tuberculosis, accentuating that the
uncyclized and succinylated form is important for antitubercular activity.
In addition, paenillaene (1) dissociated amyloid-β
(Aβ) aggregates, indicating potential related to AD. Pulvomycins
E–F (6–7) showed anti-proliferative
activity with several cancer cells, especially against the MDA-DTX
cell line.
The discovery of these novel β-branched polyketides,
along
with our recent systematically-targeting discovery of natural products,
−

underscores the efficacy of the metabologenomic targeting method
as a logical approach for discovering new bioactive compounds with
targeted structural motifs.

Supplementary Material