Discovery of small molecule protease inhibitors by investigating a widespread human gut bacterial biosynthetic pathway

Article abstract of DOI:10.1016/j.tet.2018.03.043

Natural products from the human microbiota may mediate host health and disease. However, discovery of the biosynthetic gene clusters that generate these metabolites has far outpaced identification of the molecules themselves. Here, we used an isolation-independent approach to access the probable products of a nonribosomal peptide synthetase-encoding gene cluster from Ruminococcus bromii, an abundant gut commensal bacterium. By combining bioinformatics with in vitro biochemical characterization of biosynthetic enzymes, we predicted that this pathway likely generates an N-acylated dipeptide aldehyde (ruminopeptin). We then used chemical synthesis to access putative ruminopeptin scaffolds. Several of these compounds inhibited Staphylococcus aureus endoproteinase GluC (SspA/V8 protease). Homologs of this protease are found in gut commensals and opportunistic pathogens as well as human gut metagenomes. Overall, this work reveals the utility of isolation-independent approaches for rapidly accessing bioactive compounds and highlights a potential role for gut microbial natural products in targeting gut microbial proteases.

Full text of DOI:10.1016/j.tet.2018.03.043

Accepted Manuscript
Discovery of small molecule protease inhibitors by investigating a widespread human
gut bacterial biosynthetic pathway
Benjamin A. Schneider, Emily P. Balskus
PII:
S0040-4020(18)30308-9
10.1016/j.tet.2018.03.043
TET 29383
DOI:
Reference:
To appear in: Tetrahedron
Received Date: 18 January 2018
Revised Date: 15 March 2018
Accepted Date: 20 March 2018
Please cite this article as: Schneider BA, Balskus EP, Discovery of small molecule protease inhibitors by
investigating a widespread human gut bacterial biosynthetic pathway, Tetrahedron (2018), doi: 10.1016/
j.tet.2018.03.043.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT
Title: Discovery of small molecule protease inhibitors by investigating a widespread
human gut bacterial biosynthetic pathway
Authors: Benjamin A. Schneider and Emily P. Balskus*
Affiliations: Department of Chemistry & Chemical Biology, Harvard University,
Cambridge, Massachusetts 02138, United States
*
Corresponding author. Tel.: +1 617 496 9921; fax +1 617 496 5866; e-mail address:
balskus@chemistry.harvard.edu
Abstract: Natural products from the human microbiota may mediate host health and
disease. However, discovery of the biosynthetic gene clusters that generate these
metabolites has far outpaced identification of the molecules themselves. Here, we used an
isolation-independent approach to access the probable products of a nonribosomal
peptide synthetase-encoding gene cluster from Ruminococcus bromii, an abundant gut
commensal bacterium. By combining bioinformatics with in vitro biochemical
characterization of biosynthetic enzymes, we predicted that this pathway likely generates
an N-acylated dipeptide aldehyde (ruminopeptin). We then used chemical synthesis to
access putative ruminopeptin scaffolds. Several of these compounds inhibited
Staphylococcus aureus endoproteinase GluC (SspA/V8 protease). Homologs of this
protease are found in gut commensals and opportunistic pathogens as well as human gut
metagenomes. Overall, this work reveals the utility of isolation-independent approaches
1

ACCEPTED MANUSCRIPT
for rapidly accessing bioactive compounds and highlights a potential role for gut
microbial natural products in targeting gut microbial proteases.
Keywords: gut microbiota; natural product biosynthesis; nonribosomal peptide
synthetase; peptide aldehyde; protease inhibitor; protease
1
. Introduction
Small molecules produced by the human gut microbiota are potential mediators of this
1
microbial community’s effects on host health and disease. However, the major
inhabitants of the gut have not been extensively investigated as natural product
producers. Though genome and metagenome sequencing continues to reveal that human
gut microbes have a rich biosynthetic potential, discovering natural products from these
organisms has proven challenging, in part because many cannot be cultivated in the
laboratory. Moreover, investigations to date have found that gut microbial natural
products are often difficult or impossible to isolate or are not produced under standard
1
–3
laboratory conditions. Though there are limited examples of isolating small molecules
4
produced by gut microbes in pure culture (e.g., ruminococcin A), alternative strategies
5
such as functional metagenomics and expression of biosynthetic gene clusters in
6
heterologous hosts have also revealed gut microbial natural products (Fig. 1A). Overall,
there is clearly a continued need for new approaches that will provide more rapid access
to products of gut microbial gene clusters.
2

ACCEPTED MANUSCRIPT
Fig. 1: Isolation-independent approaches may accelerate identification of bioactive natural
products from the human gut microbiota. (A) Selected natural products from human gut bacteria,
including the proposed structure of the lanthionine-containing bacteriocin (lantibiotic)
ruminococcin A, PZN10, and the antibiotic humimycin. (B) Our isolation-independent workflow
for characterizing small molecules produced by important gut commensals involves first selecting
nonribosomal peptide synthetase (NRPS)-encoding biosynthetic gene clusters of interest based on
abundance in metagenomic sequencing data and microbial ecology. Bioinformatic predictions
and in vitro biochemical assays then provide structural information that informs the chemical
synthesis of candidate natural product structures. These focused small molecule libraries can then
be evaluated for bioactivity.
We envisioned a strategy for accessing gut microbial secondary metabolites that would
combine in vitro characterization of biosynthetic enzymes with chemical synthesis (Fig.
1
B). By mining human gut metagenomic sequence data, we could identify small
nonribosomal peptide synthetase (NRPS) biosynthetic gene clusters of interest based on
metagenomic sequencing data and microbial ecology. These enzymes share a conserved
3

ACCEPTED MANUSCRIPT
chemical logic and would therefore be amenable to bioinformatic analyses and prediction
of their natural product structures. We could then test our predictions and identify key
biosynthetic building blocks using in vitro biochemical assays with purified biosynthetic
enzymes. Finally, we would access the candidate natural product structures using
chemical synthesis and evaluate these focused small molecule libraries for bioactivity. A
key advantage of this approach is that it could provide a more rapid way to access
bioactive small molecules compared to traditional isolation- or heterologous expression-
based natural product discovery. Indeed, Brady and coworkers recently demonstrated the
utility of a related strategy (the “synthetic-bioinformatic natural products”, or syn-BNPs,
7
approach) in their discovery of humimycin A (Fig. 1A). By mining sequenced genomes
from the human microbiota for NRPS gene clusters, predicting the structures of the likely
gene cluster products using bioinformatics, and synthesizing the predicted nonribosomal
peptides, they accessed a new antibiotic that is active against methicillin-resistant
Staphylococcus aureus clinical isolates.
Here, we have used our isolation-independent approach to access the putative products of
an NRPS gene cluster from Ruminococcus bromii, one of the most abundant commensal
microbes in the human gut. We first employed bioinformatic analyses to predict the
product of this conserved and widely distributed gene cluster (the rup gene cluster) as a
reactive, N-acylated dipeptide aldehyde (ruminopeptin). We then used in vitro
biochemical characterization of the NRPS assembly line enzymes to identify the building
blocks of ruminopeptin. Using a short, solution phase synthesis, we accessed a library of
ruminopeptin analogues and evaluated their bioactivities. We found these molecules
4

ACCEPTED MANUSCRIPT
inhibit S. aureus endoproteinase GluC (SspA/V8 protease), which has been implicated in
8
virulence in a mouse abscess model. The human gut microbe and opportunistic pathogen
9
Enterococcus faecalis also produces a virulence-related glutamyl endopeptidase, and
further bioinformatics analyses revealed additional homologs of this enzyme in gut
microbial genomes and metagenomes. We hypothesize that protease inhibitors of this
family may be important for mediating microbe-microbe interactions in the human gut.
2
. Results and discussion
2
.1 The prominent human gut microbe Ruminococcus bromii possesses an abundant
and conserved biosynthetic gene cluster
With the goal of discovering bioactive secondary metabolites from the human gut
microbiota, we initially focused on the prominent gut commensal R. bromii. This
organism is one of the most abundant microbes in the human gut across a diversity of
1
0–13
environments and diets,
keystone species in the degradation of resistant starch.
Clostridium cluster IV, which is significantly less abundant in patients with inflammatory
and it has an important ecological role in the colon as a
1
4,15
R. bromii is a member of
1
6
bowel disease (IBD) as compared with healthy subjects. This phylogenetic group of
Clostridia contains organisms that are generally considered to be beneficial in the gut
environment and includes Faecalibacterium prausnitzii, which has a well-studied anti-
1
7
inflammatory role. Though to our knowledge R. bromii has not yet been reported to
5

ACCEPTED MANUSCRIPT
produce natural products, we hypothesized that this could be a mechanism by which this
organism exerts its beneficial effects or maintains its ecological niche in the human gut.
Fig. 2: A biosynthetic gene cluster from the abundant gut commensal Ruminococcus bromii
encodes a putative peptide aldehyde natural product. (A) The rup gene cluster from R. bromii.
The gene cluster encodes a single multi-modular NRPS, a transporter, two regulatory elements,
and two hypothetical proteins. (B) The RupA NRPS contains a condensation-starter (C-starter)
domain and a terminal reductase (R) domain (A = adenylation domain, T = thiolation domain).
(
C) Biosynthetic hypothesis for the production of ruminopeptin by the rup gene cluster.
R. bromii encodes a 10.9 kb biosynthetic gene cluster that encodes a single di-modular
NRPS, an efflux pump (ABC transporter), two regulatory elements, and two hypothetical
proteins (Fig. 2A, Table S1). The rup gene cluster (also known as bgc45) has been
identified previously by Fischbach and co-workers in a large survey of biosynthetic gene
clusters from the human microbiome and is part of a larger family of NRPS gene clusters
6

ACCEPTED MANUSCRIPT
1
8
found in gut microbial genomes and metagenomes. This study also revealed the rup
gene cluster to be one of the most abundant gene clusters found in human microbiome
project (HMP) stool metagenomes. Moreover, a highly similar gene cluster (bgc71,
9
7.2% nucleotide sequence identity) from a closely related, unisolated Ruminococcus
species was identified in several RNAseq datasets from stool samples of healthy subjects,
indicating that this biosynthetic pathway is likely expressed under physiological
6
conditions. Overall, these findings suggest the product of the rup gene cluster is likely
produced under physiological conditions. Coupled with the established importance of R.
bromii, this may indicate a particularly important role for this metabolite in the human
gut microbiota.
Based on gene content and NRPS biosynthetic logic, we predicted that the rup gene
cluster would produce a peptide aldehyde natural product. The NRPS (RupA) features a
condensation-starter (C-starter) domain, indicating that the N-terminus of the product
1
9
non-ribosomal peptide is likely N-acylated, one complete NRPS module, and a terminal
reductase (R) domain (Fig. 2B). This final domain should catalyze release of a nascent
thioester intermediate from the NRPS enzyme, generating either an aldehyde or a primary
2
0
alcohol-containing product. A peptide aldehyde product would likely able to act as an
inhibitor of serine, cysteine, or threonine proteases as has been demonstrated for NRPS-
2
1
derived peptide aldehydes produced by soil microbes (e.g. fellutamide B and the
2
2
flavopeptins ). Notably, Ruminococceae are negatively correlated with protease activity
2
3
in the colon, and production of small molecule protease inhibitors by these organisms is
a potential mechanism by which this association could arise.
7

ACCEPTED MANUSCRIPT
If the product of the rup gene cluster does play a crucial role in R. bromii’s ecology and
evolutionary history, we might expect it to be highly conserved in this species. To assess
the presence of this gene cluster across R. bromii strains, we used PCR with specific
primers to amplify a fragment of the first NRPS adenylation domain (RupA ) in three
A1
available human-derived R. bromii isolates (R. bromii L2-63, R. bromii ATCC 27255,
1
4,24
and R. bromii 22-5-S 6 FAA NB).
We observed amplification in each strain (Fig. S1).
We then subsequently PCR-amplified and sequenced the full gene clusters to reveal
greater than 96% identity on the nucleotide level (Table S2). Conservation of this
biosynthetic gene cluster across all known human-derived R. bromii isolates provides
evidence that this pathway may be important for the organism’s native biological role.
2
.2 Bioinformatic analysis of the rup gene cluster predicts production of an N-
acylated dipeptide aldehyde
In order to gain information about the product of the rup gene cluster, we first used
bioinformatic analyses to predict the activities and substrate specificities of each of the
domains in this two-module NRPS assembly line. RupA lacks an adenylation-thiolation
(A-T) didomain loading module and instead contains a predicted C-starter domain. C-
starter domains catalyze N-acylation of an initially loaded, assembly-line tethered amino
acid with a fatty acyl-CoA. Multiple sequence alignments with biochemically and
genetically characterized C-starter domains (ClbN, XcnA, and GlbF) revealed the RupA
C-starter domain contains key conserved residues indicative of N-acylation activity (Fig.
8

ACCEPTED MANUSCRIPT
2
5–27
28
S2).
We then used the Maryland PKS/NRPS server to predict the substrate
specificities of the two A domains of RupA. We found that the first NRPS module likely
preferred -leucine and the second NRPS module likely used either -aspartate or
L
L
L-
glutamate (Fig. S3). Finally, we generated a structure-based multiple sequence alignment
of the final RupA domain with other characterized NRPS terminal R domains using
R
2
9
PROMALS3d. From this alignment, we could identify all of the key conserved active
site residues involved in NAD(P)H binding as well as the Thr/Tyr/Lys catalytic triad
required for thioester reduction (Fig. S4).
Together, these analyses allowed us to propose a biosynthetic hypothesis for the rup
pathway and predict the structure of the final peptide aldehyde product(s), which we
named ruminopeptin (Fig. 2C). After post-translational modification of the RupA T
domains by a phosphopantetheine (ppant) transferase, initiation of biosynthesis occurs
with the activation of
onto the ppant arm of the first T domain. The C-starter domain of the first module then
acylates the amino group of the tethered -leucine with a fatty acyl CoA. The resulting N-
acylated aminoacyl thioester intermediate is then elongated by amide bond formation
with the amino acid loaded by the second NRPS module, either -aspartate or
L-leucine by the A domain of the first NRPS module and loading
L
L
L-
glutamate, to generate a nascent N-acylated dipeptide thioester intermediate. Finally,
reductive offloading of this intermediate by the R domain will give a peptide aldehyde
product that we have named ruminopeptin.
9

ACCEPTED MANUSCRIPT
2
.2 The product(s) of the rup gene cluster are not readily isolated from R. bromii
cultures
Having proposed a candidate structure for the rup gene cluster product(s), we wondered
if it would be possible to isolate these secondary metabolites from R. bromii. We began
by identifying culture conditions in which the rup pathway was expressed. We cultivated
two strains of R. bromii using a variety of nutrient sources and several unusual culture
additives (rumen fluid, chopped meat broth). We extracted RNA from saturated cultures
and assessed gene cluster expression using specific primers with single-step RT-PCR. We
observed that including fructose as a carbohydrate source in growth media was necessary
for rup gene cluster expression and that inclusion of additives had no effect (Fig. S5).
However, in numerous attempts using cultures grown under conditions where the rup
genes were expressed (5 mL to 1 L scales) we could not identify candidate masses
corresponding to any predicted ruminopeptin peptide aldehyde product by LC-MS. We
3
0
also attempted comparative metabolite profiling using XCMS, but this analysis did not
reveal any candidate masses of interest.
2
.3 In vitro biochemistry reveals the building blocks of the rup gene cluster product
ruminopeptin
Since we could not readily isolate the predicted product(s) of the rup gene cluster, we
sought to reconstitute this pathway in vitro to confirm our biosynthetic hypothesis and
identify the preferred amino acid and acyl-CoA building blocks used by the NRPS
1
0

ACCEPTED MANUSCRIPT
assembly line. The individual modules of the RupA NRPS were expressed and purified in
Escherichia coli as C-His -tagged constructs (RupA
and RupAC2-A2-T2-R) (Fig. S6).
6
C1-A1-T1
We then used a set of standard biochemical assays to verify the activities of the two
NRPS modules and determine the substrate specificities of their A domains. The ATP-
3
2
31
[
P]PP exchange assay was used to assess amino acid activation by each individual
i
module of RupA. These experiments revealed that RupAC1-A1-T1 preferentially activates
leucine but can also accept -valine, while RupAC2-A2-T2-R preferentially activates
glutamate over -aspartate (Figs. 3A, S7 and S8). We then used the promiscuous
phosphopantetheinyl transferase Sfp to load BODIPY-CoA onto the T domain of each
L-
L
L-
L
3
2
module, verifying that this enzyme can posttranslationally modify the RupA NRPS
3
1
13
(
Fig. S9). Finally, T domain loading assays with C-labeled amino acids confirmed that
both -leucine and -valine were tethered onto RupAC1-A1-T1 and that both -glutamate
and -aspartate could be loaded onto RupAC2-A2-T2-R (Figs. S10 and S11).
L
L
L
L
1
1

ACCEPTED MANUSCRIPT
Fig. 3: RupA preferentially uses hexanoyl-CoA,
L-leucine, and L-glutamate to produce a T-
3
2
domain tethered N-acylated dipeptide thioester intermediate. (A) ATP-[ P]PP
i
exchange assay
for individual RupA modules. (B) LC-MS assay for C-starter domain specificity. Mass
abundances (extracted ion chromatogram intensities) are shown for hydrolyzed products obtained
from the reaction of RupAC1-A1-T1
,
L
-leucine, and equimolar amounts of C
2
−C14 fatty acyl-CoA
substrates. The mass abundance of N-hexanoyl- -leucine is highlighted in blue. Representative
L
results are shown from at least two independent experiments. (C, D) LC-MS assay for tethered N-
acylated dipeptide synthesis by RupA. Mass abundances (extracted ion chromatogram intensities)
are shown for N-acylated dipeptide products. The mass abundance of N-hexanoyl-L-leucyl-L-
glutamic acid in each experiment is highlighted in red. Representative results are shown from at
least two independent experiments. (C) The assay mixture contained RupAC1-A1-T1, RupAC2-A2-T2-R
,
ATP, hexanoyl-CoA, and equimolar amounts of
L-valine, L-leucine, L-aspartate, and L-glutamate
(
amino acid competition format). (D) The assay mixture contained RupAC1-A1-T1, RupAC2-A2-T2-R
,
1
2

ACCEPTED MANUSCRIPT
ATP,
2
L-leucine, L-glutamate, and equimolar amounts of C −C14 fatty acyl-CoA substrates (acyl-
CoA competition format).
With the amino acid building blocks established, we next set about identifying the
specific acyl-CoA(s) that would be recognized and incorporated on the N-terminus of
ruminopeptin. Though fatty acids can be incorporated into nascent polyketides and non-
2
7,33,34
ribosomal peptides by several mechanisms,
of the RupA NRPS would N-acylate -leucine using a freely diffusible fatty acyl-CoA
co-substrate. In order to determine the preferred fatty acyl-CoAs, we reconstituted the
activity of RupAC1-A1-T1. We incubated RupAC1-A1-T1 with -leucine, ATP, and a set of
short, medium and long even-chain acyl-CoAs (C to C ) in a competition assay format.
we predicted that the C-starter domain
L
L
2
14
We then hydrolyzed the resulting N-acylated aminoacyl thioester intermediates from the
2
5,27
NRPS for detection using LC-MS (Figs. 3B and S12, Table S3).
Using this assay we
identified N-hexanoyl- -leucine as the most abundant product, indicating that hexanoyl-
L
CoA and other medium-chain acyl-CoA’s are likely preferred substrates of the RupA C-
starter domain.
We subsequently modified this assay to probe the activity of both NRPS modules. As we
were unable to successfully express full-length RupA, we instead included the
individually expressed and purified NRPS modules (RupAC1-A1-T1 and RupAC2-A2-T2-R) in
the reaction mixture along with amino acids, acyl-CoA substrates, and ATP. As the
NAD(P)H cofactor required for the reductase domain was not provided in these initial
attempts, we predicted this assay should generate T-domain-tethered N-acylated dipeptide
thioesters which could be hydrolyzed from the enzyme and detected by LC-MS. We first
1
3

ACCEPTED MANUSCRIPT
performed a competition experiment to identify the preferred amino acid building blocks,
including in the reaction mixture multiple amino acids (
and -glutamate) along with a single fatty acyl-CoA substrate (hexanoyl-CoA) (Figs. 3C
and S13, Table S4). The preferred product generated in this experiment incorporated
leucine and -glutamate. To verify the identity of the preferred acyl-CoA substrate, we
next performed this assay using the preferred amino acid substrates ( -leucine and
glutamate) and a mixture of even-chain acyl-CoA’s (Fig. 3D, Table S5). In this
experiment, N-hexanoyl- -leucyl- -glutamic acid was the most abundant product. From
L-valine, L-leucine, L-aspartate,
L
L
-
L
L
L-
L
L
these results, we concluded that RupA can produce a range of nascent T-domain tethered
N-acylated dipeptide thioester intermediates, and may preferentially use hexanoyl-CoA,
L
-leucine, and L-glutamate building blocks.
Finally, we sought to completely reconstitute the RupA NRPS in vitro to access putative
peptide aldehyde products. To accomplish this, we included NAD(P)H, the cofactor
required for R domain activity, in reaction mixtures along with ATP and the preferred
substrates hexanoyl-CoA, L-leucine, and L-glutamate. We analyzed the supernatants of
reaction mixtures by LC-MS and attempted to detect the expected masses of peptide
aldehydes, primary alcohols, truncated products, or molecules that could arise from
degradation of the predicted structures. However, after extensive optimization we were
unable to detect any putative final products in this experiment. We were also unable to
identify final products in the presence of any other combinations of building blocks that
we had previously examined. We did observe formation of the hydrolysis products of
tethered N-acylated dipeptide thioester intermediates, indicating that the NRPS modules
1
4

ACCEPTED MANUSCRIPT
were functional (data not shown). We also confirmed that synthetic standards of the
predicted peptide aldehyde products (see section 2.4 for synthesis) could be detected
under these assay conditions (data not shown).
Suspecting that our RupAC2-A2-T2-R construct may have purified with an inactive R
domain, we individually expressed and purified two additional constructs (RupA single
R
domain and RupAT2-R di-domain) and evaluated their reactivity toward a synthetic N-
acetylcysteamine substrate 5 that mimics the preferred RupAT2-tethered intermediate
(
Fig. 4A). Monitoring consumption of NAD(P)H by the change in absorbance at 340 nm
A ), we could detect activity of neither RupA , RupA nor the full module RupAC2-
(
3
40
R
T2-R
toward the synthetic substrate, suggesting that we have not successfully purified an
A2-T2-R
active form of RupA (Fig. 4B).
R
Fig. 4: The RupA reductase domain was inactive toward substrate mimic N-hexanoyl-L-Leu-L-
Glu-SNAc 5 in vitro. (A) Synthesis of an N-acetylcysteamine (SNAc) substrate for RupA (DCC
R
=
N,N'-dicyclohexylcarbodiimide, NHS = N-hydroxysuccinimide). (B) By measuring decrease in
1
5

ACCEPTED MANUSCRIPT
absorbance at 340 nm, no consumption of NAD(P)H was observed when reacting 5 with purified
RupA , RupAT2-R, or RupAC2-A2-T2-R
.
R
Though we were unable to reconstitute the activity of the RupA R domain in vitro,
bioinformatic analyses suggest this domain should be active in vivo. Though RupAR
shows only low amino acid sequence identity to other biochemically characterized R
3
5
36
R R
domains (e.g., 23.5% with AusA , 18.7% with MxcG , and 24.6% with the R domain
6
from bgc35, which was previously reconstituted in vitro ), we were able to identify the
conserved catalytic triad and NAD(P)H binding motifs (Fig. S4). Among the diverse
superfamily of short-chain dehydrogenases/reductases, which includes NRPS terminal R
domains, proteins with sequences identities as low as 15-30% are reported to share
3
7
similar three dimensional folds. We generated a homology model of RupA with the
R
known two-electron reducing terminal R domain from AusA, using HHPred and
R
MODELLER. This model suggests that the RupA motif for NAD(P)H binding and
catalytic triad for reduction chemistry are properly oriented (Fig. S14). Therefore, we
propose that this R domain is likely active in vivo and involved in producing the final
product of the rup gene cluster.
In summary, our efforts to reconstitute the activity of the R. bromii NRPS RupA strongly
suggest that the peptide aldehyde N-hexanoyl-L-Leu-L-Glu-H is a likely product of this
enzymatic assembly line. This work does not rule out the possibility that RupA may
produce additional metabolites in vivo. We observed some promiscuity in the N-acylation
1
6

ACCEPTED MANUSCRIPT
activity of the C-starter domain and A domain specificities of RupA. However, this type
of promiscuity is often observed for NRPS enzymes reconstituted in vitro, and the
preferred products observed in this format typically correspond to the most abundant
natural analogues, even in cases where multiple products can also be isolated from
3
5,38
cultures.
It is also possible that there are unusual biosynthetic substrates available to
R. bromii, particularly acyl-CoA’s with unusual acyl-chain modifications, which we did
not provide in our in vitro reconstitution experiments. Though RupA may produce
additional molecules with different scaffolds, it is reasonable to propose that N-hexanoyl-
L
-Leu-L-Glu-H could be one major biosynthetic product.
Additionally, as we have never directly observed activity of the RupA R domain, we
could not determine if this assembly line produces an aldehyde or an alcohol. In previous
biosynthetic reconstitutions of NRPS terminal reductase domains, the aldehyde
intermediate has not been detected in significant quantities when the final expected
39,40
product is the peptide alcohol.
Though the activity of aldehyde-producing NRPS
3
5,41
terminal reductase domains has been reconstituted in several cases,
the natural
products generated by these pathways are cyclic imines or pyrazinones, so only trace
amounts of free aldehyde intermediates were detected in these experiments. Given the
difficulties encountered in resolving this biosynthetic step, we decided to move forward
to examine the biological activity of the putative peptide aldehydes we predict could be
produced by RupA.
2
.4 Synthesis and biological activity evaluation of ruminopeptin analogues
1
7

ACCEPTED MANUSCRIPT
We next used the information that we obtained from our bioinformatic and biochemical
analyses of the rup biosynthetic pathway to inform the chemical synthesis of a focused
library of predicted ruminopeptin structures. We designed 12 analogues of the predicted
N-acyl dipeptide aldehyde scaffold that varied in the N-acyl substituent and amino acid
components. We then accessed these compounds using a solution-phase synthetic route
4
2,43
adapted from previous syntheses of aspartyl and glutamyl peptide aldehydes.
Peptide
synthesis has previously been used as a tool to access natural product peptide aldehydes
when isolation efforts yielded insufficient quantities of pure material for activity
2
2
screening. In our case, we envisioned that accessing a small library could not only
provide compounds for assays but also enable structure-activity relationship studies.
From N-Cbz- and O-tBu-protected
L-glutamate and L-aspartate precursors, we accessed
key intermediates containing an aldehyde masked as a semicarbazone functional group
4
2,43
using the previously reported reaction sequences (6a-b, Fig. 5).
semicarbazone-protected intermediates were then coupled to N-acylated
valine derivatives (1a-i, Fig. 5) using the peptide coupling reagent HATU to yield 7a-l
Table 1). From these coupling products, deprotection of the O-tBu ester proceeded with
0% trifluoroacetic acid in dichloromethane. Finally, transfer of the semicarbazide
The resulting
L-leucine and L-
(
2
functional group to formaldehyde under acidic conditions and corresponding regeneration
of the aldehyde provided the desired peptide aldehyde products (8a-l) (Table 2). Using
this route, we accessed a small library of ruminopeptin analogues on a multi-milligram
scale (6-42% overall yield, 7-36 mg obtained).
1
8

ACCEPTED MANUSCRIPT
Fig. 5: Synthetic precursors used in this study.
Table 1: Coupling reaction between O-tBu-protected semicarbazones 6a-b and N-acyl amino
acids 1a-i to give semicarbazone intermediates 7a-l.
1
9

ACCEPTED MANUSCRIPT
Table 2: Removal of O-tBu protecting groups from 7a-l and exchange of semicarbazones with
formaldehyde to afford the desired ruminopeptin analogues 8a-l.
With access to sufficient quantities of peptide aldehydes 8a-l, we could begin to identify
potential target(s) of these molecules. Our biosynthetic reconstitution experiments
strongly suggest that ruminopeptin contains a glutamate residue in its P1 position. We
thus gained insights into potential targets by comparing the predicted structures of the
ruminopeptins to the known substrate specificities of secreted microbial serine and
cysteine proteases, as well as host proteases. As specific post-glutamyl hydrolyzing
activity is rare among microbial proteases and unknown among human proteases, this
analysis revealed only one promising candidate: the glutamyl endopeptidases. Glutamyl
endopeptidases are a class of secreted serine proteases found in several bacterial species,
4
4
45
including the human pathogens E. faecalis (SprE) and S. aureus (endoproteinase
GluC/SspA/V8 protease). These proteases are regularly found in quorum-sensing
2
0

ACCEPTED MANUSCRIPT
regulated operons alongside additional proteases, and glutamyl endopeptidases appear to
regulate the action of these enzymes (a metalloprotease in E. faecalis and a cysteine
protease in S. aureus). SspA and SprE are thought to play roles in biofilm formation in
these pathogens, though the precise details of their involvement vary substantially among
4
6,47
different strains and assay systems.
Additionally, results from several animal models
8
,48–50
implicate SspA and SprE in virulence.
We screened our library of ruminopeptin analogues (4a-l) for their ability to inhibit the
activity of SspA in vitro. In this assay, protease was pre-incubated with the peptide
aldehyde for 10 min. Protease activity was then quantified by measuring the increase in
fluorescence corresponding to the release of the AMC fluorophore from the fluorogenic
peptide substrate Z-Leu-Leu-Glu-AMC. We found that several of the synthetic
compounds inhibit SspA, with approximately 50% inhibition observed for the most
potent compounds, medium-chain acyl analogues 8h and 8l, at 10 µM (Figs. 6A and
S15). Intriguingly, in vitro reconstitution experiments suggested these compounds are
also among the mostly likely products generated by the rup gene cluster. We observed
reduced inhibition with the short-chain acyl analogues 8b-d and insignificant inhibitory
activity with the branched acyl chain analogues 8e-g or aspartyl analogues 8a, 8j and 8k.
2
1

ACCEPTED MANUSCRIPT
A
B
% inhibition
0
100
8a 8b 8c 8d 8e 8f 8g 8h 8i 8j 8k 8l
1
00 µM
50
10
Ile35
Gly136
Gly167
Gly166
Ala36
Val1
His51
aldehyde 8h
Ser169
Thr164
His184
Fig. 6: Ruminopeptin-like compounds inhibit SspA from S. aureus. A) Inhibition profile of
predicted ruminopeptins against SspA. The assays were conducted by pre-incubating 1 ng/µL
SspA with inhibitor for 10 min at room temperature followed by addition of fluorogenic peptide
substrate Z-Leu-Leu-Glu-AMC to a final concentration of 75 µM. Fluorescence (367 nm
excitation/460 nm emission) was then monitored for 20 min at 30 ºC and inhibitor efficiency
calculated by comparing the slope of the linear portion of the curve with the negative control (no
inhibitor). Reactions were performed in duplicate and inhibitor efficiency is reported as a mean of
both trials. B) Potential interaction between peptide aldehyde 8h (grey) and SspA (blue). The
structure was docked into the crystal structure of SspA (PDB: 1qy6) using the induced fit docking
algorithm in Glide.
To better understand the interaction between 8h and SspA, we performed a docking
5
1
experiment using Glide. Substrate recognition by SspA is reported to rely on an
electrostatic interaction between the negatively charged glutamate side chain in position
2
2

ACCEPTED MANUSCRIPT
5
2
P1 of the peptide substrate and the positively charged N-terminal amine of SspA. We
observed a similar interaction between this N-terminal amine and the glutamyl side chain
of 8h when we docked this inhibitor into the crystal structure of SspA (PDB: 1qy6). The
electrophilic aldehyde warhead of 8h was also located within reasonable proximity (4.6
Å) to the nucleophilic Ser residue (Fig. 6B), suggesting that this inhibitor binds the
protease similarly to a model protein substrate and an orientation that would facilitate
formation of a reversible, covalent hemiacetal linkage.
5
3,54
Though synthetic inhibitors of SspA have previously been reported,
this work
provides the first indication of endogenous inhibition of glutamyl endopeptidases by
microbial natural products and is also the first report of peptide aldehyde inhibitors of this
enzyme class. By surveying the structure-activity relationships (SAR) in the synthetic
library of closely related family members, we revealed that the presence of both
and -glutamate are important for SspA inhibition.
L-leucine
L
2
.5 Identification of glutamyl endopeptidase homologs in the human gut microbiota
The observation that putative peptide aldehydes derived from a gut commensal could
inhibit a bacterial glutamyl endopeptidase made us curious about the relevance of these
proteases within the human gut. Though S. aureus is more commonly associated with the
nasal microbiota and can be detected from nasal swabs of approximately 40% of healthy
individuals, studies have consistently detected this bacterium in the stool microbiomes of
5
5
approximately 20% of healthy individuals. Indeed, intestinal carriage of S. aureus is
2
3

ACCEPTED MANUSCRIPT
5
5
hypothesized to contribute to bacterial dissemination in the environment. In
comparison, the opportunistic pathogen E. faecalis can be detected in 47% of fecal
56
samples from healthy individuals, and its glutamyl endopeptidase SprE resembles SspA
5
7
(
49% amino acid similarity and 27% identity). Therefore, the homologous E. faecalis
glutamyl endopeptidase represents an additional possible target that may be more
relevant within the habitat of R. bromii.
It is also possible that peptide aldehydes produced by R. bromii interact with related
proteases found in other gut commensal microbes. To our knowledge, the presence and
roles of glutamyl endopeptidases in the human gut microbiota has not yet been
investigated. Glutamyl endopeptidases have been discovered in Staphylococcus, Bacillus,
and Streptomyces species, and many have been biochemically characterized, including
5
8
SspA, SprE, glutamyl peptidase BL (from Bacillus licheniformis), glutamyl peptidase
5
9
60
BS (from Bacillus subtilis), glutamyl peptidase BI (from Bacillus intermedius), and
6
1
glutamyl endopeptidase II (from Streptomyces griseus). These enzymes all belong to
6
2
the structural chymotrypsin family, and though they exhibit some differences in kinetic
parameters and specificity, they all share a preference for cleavage after glutamyl
residues and would therefore likely be inhibited by a glutamyl aldehyde. The diversity of
these biochemically characterized examples of glutamyl endopeptidases provided a broad
starting point for identifying additional potential targets of ruminopeptin in the human
gut.
2
4

ACCEPTED MANUSCRIPT
Fig. 7: Glutamyl endopeptidase homologs are found in gut microbial genomes and human gut
metagenomes. ClustalW2 alignment of biochemically characterized glutamyl endopeptidases
(
black), homologs from sequenced organisms (blue), and homologs from human gut microbes
and gut metagenomes (red). Included are the sequences of characterized glutamyl endopeptidases
52
57
59
from S. aureus (1qy6), E. faecalis (Q47809), B. licheniformis (P80057), B. subtilis
6
3
64
61
(
(
P39790), B. intermedius (1p3c), S. griseus (Q07006), Staphylococcus epidermidis
65
66
BAC24763.1), and epidermolytic toxin A from S. aureus (1agj), with additional predicted
glutamyl endopeptidases from L. monocytogenes (WP_014601768.1), E. faecium (EEV49703.1),
and F. prausnitzii (CUO15772.1). Metagenomic sequences were retrieved using the BLAST tool
67
at JGI Integrated Microbial Genomes & Microbiome Samples. Catalytic Ser195 is indicated
with a black asterisk. Positions 190 and 213, which may be involved in conferring substrate
specificity, are indicated with red asterisks.
To explore whether additional glutamyl endopeptidases are present in the human gut
microbiota, we used BLAST searches to locate members of this family in sequenced gut
microbial genomes. Queries of the non-redundant (nr) protein database of NCBI with six
representative glutamyl endopeptidase sequences identified putative hits in other
-12
prominent residents of the human gut (e-value cutoff 9 e ). Conserved residues Thr190
or Ser190) and His213 (chymotrypsin numbering) in the S1 binding pocket of
crystallized glutamyl endopeptidases have been identified as important for binding
(
2
5

ACCEPTED MANUSCRIPT
6
2
glutamate-containing substrates (Thr164 and His184 in SspA). We were able to identify
these residues in the BLAST hits from Enterococcus faecium (24.5% ID to SprE) and the
pathogen Listeria monocytogenes (25.5% ID to glutamyl endopeptidase BL) (Fig. 7).
Additionally, we identified a sequence from F. prausnitzii that is annotated as a glutamyl
endopeptidase precursor. This sequence maintains His213, but not Thr190, in the S1 site.
It remains to be determined if these putative glutamyl endopeptidases from prominent gut
commensals and human pathogens actually exhibit post-glutamyl hydrolyzing activity.
Overall, these proteins may not only represent ecologically relevant targets of the
ruminopeptins but also provide a promising starting point for investigating the biological
roles of microbial proteases in the human gut.
In order to assess the presence of these proteases in human subjects and determine the
distribution of glutamyl endopeptidases among unsequenced members of the gut
microbiota, we also performed a BLAST search of representative glutamyl endopeptidase
sequences against assembled stool metagenomes available through the JGI (268
metagenomes). After limiting the results based on an e-value cutoff (2e-10), length (188-
4
00 residues to account for the diversity among characterized members of this protease
family), and the presence of a candidate histidine residue in the S1 binding pocket, we
identified 52 glutamyl endopeptidase homologs in 51 different samples. 47 of these
sequences have ≥99% amino acid sequence ID to the putative glutamyl endopeptidase
from F. prausnitzii. The remaining sequences do not map to sequenced genomes. This
analysis suggests that these putative targets of the ruminopeptins may be present in the
human gut.
2
6

ACCEPTED MANUSCRIPT
2
.6 Potential implications of gut microbial peptide aldehyde production
A multitude of human and microbial proteases are present in the human gut environment.
While host-derived digestive proteases are not a major contributor to protease activity in
the colon, other human proteases are involved in gut barrier maintenance and facilitating
6
8
migration of immune cells within the mucosal layer. Host proteases are also involved in
regulating the immune system, and during inflammation human immune cells secrete
6
9
proteases which are responsible for degrading extracellular tissues. Gut protease
70,71
69
activity is upregulated in ulcerative colitis
and Crohn’s disease, and is also involved
7
2
in the pathogenesis of colorectal cancer. Host derived proteases are therefore potential
therapeutic targets, with the protease inhibitor camostat mesilate previously explored as a
7
3
treatment for Crohn’s disease.
Though much work has focused on the activity of human proteases in gastrointestinal
(GI) disease, microbial proteases are also active in this environment. Bacterial proteases
7
4,75
in the healthy gut are involved in metabolism and nutrient acquisition,
but they may
7
6
77
also disrupt mucosal barriers, interact with protease-activated receptors, or modulate
7
5
the host immune response. These enzymes are also hypothesized to have roles in
7
5
biofilm formation and modification of the microbial or intestinal surface. Additionally,
several microbial secreted proteases are implicated as virulence factors in the gut
7
6
context. Therefore, inhibition of microbial proteases has recently attracted interest as a
therapeutic strategy for GI disease.
2
7

ACCEPTED MANUSCRIPT
Our work has uncovered the first evidence that gut microbial natural products may be
capable of modulating microbial protease activity. We hypothesize that the peptide
aldehyde(s) produced by the rup gene cluster likely target a microbial protease found in
this environment. Aside from the restriction of glutamyl endopeptidase activity to
bacteria, the biogeography of R. bromii in the human gut and the potential instability of
the ruminopeptins also suggest these natural products likely have a microbial target. In a
study of gut microbes associated with insoluble, undigested polysaccharide particles in
fecal samples, R. bromii was one of the three most enriched species in this phase as
7
8
opposed to the soluble phase. This observation may indicate that R. bromii is located
distantly from host cells in comparison to other gut species. Moreover, a potential
explanation for our inability to identify putative rup gene cluster products in R. bromii
cultures could the instability of these peptide aldehydes. Indeed, incubation of peptide
aldehyde 8h in an R. bromii culture resulted in almost complete degradation in just 15
min when incubated at 37 ºC (data not shown). Overall, given their instability and the
localization of R. bromii, we hypothesize that peptide aldehydes produced by this
organism have evolved to target other microbial species living in close proximity.
As highlighted earlier, the rup gene cluster is part of a larger family of NRPS
biosynthetic gene clusters found in human gut bacterial genomes and metagenomes. In
recent work, Fischbach and coworkers accessed the putative products of several of these
6
gene NRPS gene clusters using a distinct workflow. Relying principally on heterologous
expression of these gene clusters in E. coli and B. subtilis, they were able to identify
2
8

ACCEPTED MANUSCRIPT
primarily cyclic pyrazinones and dihydropyrazinone compounds, along with one N-
acylated peptide aldehyde (N-octanoyl-Met-Phe-H, a product of bgc33 from Clostridium
sp. CAG:567). This result indicates that the R domains of these NRPS assembly lines can
produce aldehyde products. Interestingly, their heterologous expression strategy was not
universally effective, as products could be identified for only 7 of the 14 gene clusters
6
investigated. As evidence that these products were not simply artifacts of heterologous
expression, they also isolated one cyclic compound from its native producing organism
and reconstituted of the biosynthesis of a cyclic pyrazinone in vitro.
Fischbach and coworkers hypothesized that the cyclic compounds observed in their study
were derived from linear dipeptide aldehyde precursors and that these dipeptide
aldehydes may be the relevant, bioactive metabolites in vivo. Therefore, they synthesized
several dipeptide aldehydes and evaluated their inhibitory activity against human
proteases. They found that free-amino dipeptide aldehydes had potent activity against
cysteine protease cathepsins (cathepsins B, L, C and S), which are found in the host
lysosome, as well as calpain. Using chemoproteomics to measure the global interactions
of a representative dipeptide aldehyde (L-Phe-L-Phe-H) with the human proteome, they
concluded that the cathepsins are likely principal targets of this compound. They suggest
that inhibition of these lysosomal proteases may disrupt immune recognition of members
of the commensal microbiota, as lysosomal cathepsins are involved in antigen processing
and presentation.
2
9