Staunton, Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge CB2 1EW, UK Fax: +44 1223 336362 Tel: +44 1223 336300 E-mail: js24@cam.ac.uk Received 28 July 2009,
Trang 1during catalysis on a modular polyketide synthase
multienzyme revealed by ion-trap mass spectrometry
Hui Hong1,2, Peter F Leadlay2and James Staunton1
1 Department of Chemistry, University of Cambridge, UK
2 Department of Biochemistry, University of Cambridge, UK
Introduction
Polyketides are a large and diverse group of secondary
metabolites that are produced by a common
biosyn-thetic strategy in bacteria, fungi, plants and animals
The term polyketide refers to the early steps of a typical pathway, in which a starter acyl residue is extended by successive addition of acyl residues, each equivalent to
Keywords
enzyme-bound intermediate; erythromycin;
limited proteolysis; liquid
chromatography-mass spectrometry; polyketide synthase
Correspondence
H Hong, Department of Biochemistry,
University of Cambridge, 80 Tennis Court
Road, Cambridge CB2 1GA, UK
*Fax: +44 1223 766002
Tel: +44 1223 333659
†E-mail: hh230@cam.ac.uk
J Staunton, Department of Chemistry,
University of Cambridge, Lensfield Road,
Cambridge CB2 1EW, UK
Fax: +44 1223 336362
Tel: +44 1223 336300
E-mail: js24@cam.ac.uk
(Received 28 July 2009, revised
29 September 2009, accepted 30
September 2009)
doi:10.1111/j.1742-4658.2009.07418.x
A catalytically competent, homodimeric diketide synthase comprising the first extension module of the erythromycin polyketide synthase was analysed using MS, after limited proteolysis to release functional domains, to deter-mine the pattern of covalent attachment of substrates and intermediates to active sites during catalysis Using the natural substrates, the acyltransferase and acylcarrier protein of the loading module were found to be heavily loaded with propionyl starter groups, while the ketosynthase was fully prop-ionylated The acylcarrier protein of the extension module was partly occu-pied by the product diketide, and the adjacent chain-releasing thioesterase domain was vacant, implying that the rate- limiting step is transfer of the diketide from the acylcarrier protein to the thioesterase domain The data suggest an attractive model for preventing iterative chain extension by effi-cient repriming of the ketosynthase domain after condensation Use of the alternative starter unit valeryl-CoA produced an altered pattern, in which a significant proportion of the extension acylcarrier protein was loaded with methylmalonate, not diketide, consistent with the condensation step having become an additional slow step Strikingly, when NADPH was omitted, the extension acylcarrier protein contained methylmalonate and none of the expected keto diketide, in contrast to results obtained previously by mixing individual recombinant domains, showing the importance of also studying intact modules The detailed patterns of loading of the extension acylcarrier protein (of which there are two in the homodimer) also provided the first evidence for simultaneous loading of both acylcarrier proteins and for the coordination of timing between the two active centres for chain extension
Abbreviations
ACP, acyl carrier protein; AT, acyl transferase; DEBS, 6-deoxyerythronolide B synthase; DKS, diketide synthase; FAS, fatty acid synthase;
KR, ketoreductase; KS, ketosynthase; PKS, polyketide synthase; SNAC, thioester of N-acetylcysteamine; TE, thioesterase.
* [Correction added on 6 November 2009 after first online publication: The fax number is wrong, it should be 766002, not 966002].
† [Correction added on 6 November 2009 after first online publication: the email address for the first corresponding author is wrong, it should be hh230@cam.ac.uk, not js24@cam.ac.uk].
Trang 2the general structural ketene unit, RCH=C=O, until
the linear chain of carbons reaches the desired length
Subsequent diverse biosynthetic transformations
gener-ate an enormously varied set of structures [1,2]
The catalytic enzymes responsible for the
chain-exten-sion processes show a remarkable degree of structural
and mechanistic homology across the wide range of
bio-synthetic organisms There are large differences,
how-ever, in the manner in which the enzymes are housed in
multienzyme clusters At one extreme, a single set of
chain-extension enzymes carries out all the
chain-exten-sion steps (iterative operation); at the other extreme,
there are systems which have a separate enzyme for
every step of the chain-extension processes (modular
assembly line operation) A further source of variance is
found in the nature of association between enzymes in
the clusters; in some systems (Type II) the individual
enzymes are readily dissociable; others (Type I) contain
large assemblies of covalently linked catalytic sites
The work in this investigation applies exclusively to
Type I modular systems that occur largely in bacteria
and produce so-called ‘complex’ polyketides These
polyketides are a very large and diverse group of
sec-ondary metabolites, produced largely (if not exclu-sively) by certain genera of bacteria They include some of the most valuable natural products to have reached the clinic, such as the antibacterial erythromy-cin A, the antiparasitic avermectin and the immuno-suppressant rapamycin [1] They are produced on giant modular multienzyme assembly lines [polyketide synth-ases (PKSs)] in a linear chain-building process akin to that of fatty acid biosynthesis, with the chief difference being that PKSs may recruit a far greater variety of starter units and extender units, while the overall length of the product and the level of reduction of each unit, as it is incorporated, may also vary [2–5] The specificity is assured by utilizing a different mod-ule of fatty acid synthase (FAS)-related activities for each cycle of chain extension, as illustrated in Fig 1 for the erythromycin-producing PKS [6-deoxyerythron-olide B synthase (DEBS)], which catalyzes assembly of the aglycone 6-deoxyerythronolide B from one mole-cule of propionyl-CoA and six molemole-cules of methyl-malonyl-CoA [2,5–8] Figure 1 also shows the arrangement of enzymatic domains in an engineered diketide synthase (DKS) containing only one extension
O
OH
OH
O
OH
OH Me
O
OH
O
OH
OH Me
O
O
OH
OH Me
O
OH
OH Me
O
OH
OH Me
O
OH Me
O
Me
O
OH
OH
Me
O
O
AT ACP
TE AT
AT AT
AT AT
DH ER
Intermediates
6-Deoxyerythronolide B (6-DEB)
Cyclise on
TE domain
SH
KS
KR
AT
AT ACP
OH
TE ACP
O
Me
S
O
OH Me
S
Hydrolyse
on TE
OH Me HO
Diketide acid
Load Module 1
DKS
Fig 1 PKS assembly line responsible for assembling the macrolide core of 6-deoxyerythronolide B (6-DEB), as revealed by sequencing the genes Each cycle of chain assembly is carried out by a dedicated set of enzymes so that there is a separate enzyme for every step The enzymes are organized in sets (modules), one for each cycle Each module has the correct set of enzymes for the extent of keto group modification (hydroxyl, enoyl, saturated methylene, colour coded), which ensures that the fatty acid chain-extension cycle is appropriately truncated and the correct transfer path is followed At the terminus there is a TE domain that carries out cyclisation and release of the first enzyme-free product, 6-DEB, as a macrolactone The truncated model DKS is shown below It carries out the operations of module 1 and then releases the diketide product as a fatty acid.
Trang 3module [9], from which the diketide product is released
by the action of the C-terminal thioesterase (TE) The
purified DKS, generated from DEBS by deleting
exten-sion modules 2–6, has been shown to hydrolyze and
release the predicted diketide product in vitro [9], albeit
with a turnover that is low compared with the
better-studied triketide synthase, DEBS1-TE, where the TE
can operate its preferred mode of chain release via
cyclization [10]
At first sight, the modular paradigm for enzymatic
catalysis in these modular PKSs, one of which is shared
with nonribosomal peptide synthetases [11], appears
simple because of the apparently direct correspondence
between the enzyme activities of a given module and
the chemical structure produced during that cycle of
extension However, despite extensive study, it is still
not understood how these giant enzymatic assemblies
are controlled and orchestrated In fact, under certain
conditions, modular PKSs have been shown to give
aberrant products in vivo in which individual
enzyme-catalyzed steps [12], or even whole modules, are
‘skipped’ [13,14], while in other cases extension
mod-ules operate more than once (iteration or ‘stuttering’)
[15–17] In a few examples, such skipping or iteration is
actually required in order to produce the natural
prod-uct [18–21] Recent strprod-uctural studies on the intact
ani-mal FAS multienzyme [22,23] and on modular PKS
domains [24–27] have also given a fresh impetus to the
question of control and orchestration of the individual
steps involved in chain extension The work on FAS
has revealed a high mobility of certain domains and
potentially a key role for major conformational
changes during catalysis [22,23] Both animal FAS and
modular PKS are functional homodimers, which raises
additional questions about the interactions between the
active sites of an identical pair of modules
We report here the use of ion-trap MS and the DKS
model system [9] to study the identity of
multienzyme-bound intermediates and to establish the pattern and
level of covalent attachment of substrates and
interme-diates to individual active sites, during catalysis in vitro
on a modular PKS For dissociated (Type II) FAS and
PKS, monitoring the nature of acyl chains attached to
the acylcarrier protein (ACP) followed quickly upon
the introduction of electrospray MS [28,29] and it
continues to give valuable mechanistic insights [30,31]
into such systems
Unfortunately the sheer size of modular PKS
mult-ienzymes has hampered their analysis in this way One
successful approach [32,33] was to degrade the
multi-enzymes to short peptide fragments, fractionate the
complex mixture and use high-resolution
electro-spray ionization Fourier-Transform mass spectrometry
(ESI⁄ FTMS) to pick out the active-site peptides and determine the nature of the covalently attached group [34,35] In addition, specific ejection of the phospho-pantetheinyl prosthetic group from the ACP can be induced in the mass spectrometer, allowing highly accu-rate determination of the mass of the attached species [35] A convenient ion-trap MS-based approach may also be used in which limited proteolysis [36] is used to generate domain-sized fragments for liquid chromatog-raphy (LC)⁄ MS analysis We have previously validated this technology for the DKS (Fig 2A) [37] and used it
to demonstrate, for the first time, that during the load-ing of the synthase with the natural starter unit propio-nyl-CoA (as well as from the alternative starter units acetyl-CoA, butyryl-CoA and valeryl-CoA), three dif-ferent active sites in both DKS subunits become almost completely acylated, namely the acyltransferase (AT) and ACP domains of the loading module and the keto-synthase domain (KS1) (Fig 2B) Here, we have used the method to establish the pattern of covalent interme-diates attached to various active sites of an intact PKS module, during catalysis of overall diketide formation This has revealed previously unsuspected features of chain elongation on such enzymes
Results The same methodology as used in our previous study [37] was first applied to investigate the loading of the natural methylmalonyl extender unit, derived from methylmalonyl-CoA, onto the extension module ACP (ACP1) of DKS After incubation with commercial methylmalonyl-CoA for 10 min, the DKS protein was digested and subjected to analysis by HPLC⁄ MS The extension module AT domain from DKS (AT1) appears as two fragments corresponding to alternative sites of proteolysis; one fragment has a molecular mass
of 32582 Da and the second has a molecular mass of
32739 Da As expected, after incubation with methyl-malonyl-CoA, both fragments showed two extra peaks
at 32684 and 32839 Da, respectively, corresponding to the addition of a methylmalonyl moiety (see Fig 3) The ratios of the intensities of the peaks for the loaded and unloaded forms of each AT1 fragment were almost identical, at 60% and 40%, respectively (Table 1) Similarly, the fragment for the ACP-TE di-domain showed two peaks, one with a molecular mass of 39506 Da corresponding to the unloaded form and the other with a molecular mass of 39604 Da, cor-responding to the form loaded with methylmalonate (see Fig 3) In this case, a ratio of approximately 55% loaded to 45% unloaded forms was observed (Table 1)
Trang 4Surprisingly, analysis of the fragments derived from
the loading module in this initial experiment
pro-duced evidence that both the AT and ACP domains
of the loading module, and the KS domain, were
loaded with propionate The purity of the commercial
methylmalonate was therefore checked by HPLC,
which revealed contamination with propionyl-CoA
Although the level of contamination was low, it was
sufficient to explain the unexpected propionate
load-ing Purification by HPLC gave propionyl-CoA free
of the methylmalonyl analogue Repeating the
experi-ment gave the same results for loading of AT1 and
ACP-TE di-domains, but the domains of the loading
module and the KS were completely unloaded, as indicated in Fig 2C Apart from demonstrating the importance of using pure materials, the results with the pure methylmalonate were significant in removing any possibility that methylmalonate decarboxylation
by the KS provides an alternative source of the pro-pionate building block used in the first condensation step The DKS was also incubated with malonyl-CoA under the same conditions used for methylmalonyl-CoA No loaded residues were detected for any of the domains, showing that the AT1 domain is highly substrate specific, unlike the AT domain of the load-ing module (data not shown)
SH KS
KR AT
AT ACP
OH
TE ACP
SH OH
TE ACP
OH
TE KR
AT OH SH KS
SH
AT ACP OH
S
SH
KS
KR AT
AT ACP
OH
TE ACP
OH
SH OH
TE ACP
OH
TE KR
AT OH
O
O
O
S
O
S
AT ACP
O
O O
KS S O
SH
SH
KS
KR AT
AT ACP
OH
TE ACP
TE ACP
OH
TE KR
AT SH KS
SH
AT ACP OH Proteolysis
Proteolysis Proteolysis
O O
CO2H
S O
CO2H After incubation with purified methylmalonyl CoA
After incubation with propionyl CoA
Control experiment; no incubation prior to analysis
O O
HO2C
S O
CO2H
Load Module 1
Load Module 1
Load Module 1
A
B
C
Fig 2 Results of proteolysis of the DKS followed by HPLC coupled to electrospray MS Experiments (A) and (B) were reported in a previ-ous publication [37]; experiment (C) is part of the current results.
Trang 5The DKS was then incubated with various
combina-tions of substrates that should allow synthesis of a
diketide product and then analyzed as before to
deter-mine the nature and extent of occupancy of the KS,
AT and ACP chain-extension domains of module 1
[the ketoreductase (KR) domain does not have a
sub-strate covalently bound] and of the adjacent TE
domain Various sets of incubation conditions, listed
in Table 1, were explored The possible adducts on the
various domains are shown in Fig 4 First, the DKS
was incubated with propionyl-CoA to supply the
native starter unit, methylmalonyl-CoA as the source
of extender unit, and, with NADPH, to carry out the
keto-group reduction step catalysed by the KR
domain After a sufficient incubation period to
estab-lish steady turnover (10 min), the mixture was analysed
using the standard protocol to determine the extent of
loading on the domains of module 1 and its attached
TE domain
MS analysis of the KS1 fraction showed that the
active site was fully loaded with propionate The
absence of the free form of the KS shows that loading
of propionate onto the KS via the loading module is
not rate-limiting under these conditions The mass spectrum for the fraction containing the TE domain and the key ACP-TE domain is shown in Fig 5B Only one peak was observed for the TE domain, with
a mass corresponding to the unloaded form From this
it can be concluded that release of the diketide inter-mediate from this domain is faster than its acylation
by diketide transfer from ACP1 Therefore any cova-lently attached species detected on the ACP-TE di-domain is resident only on the ACP1 thiol Two peaks are seen, one corresponding to the unloaded form and the other to the form loaded with diketide (see Table 1) From the increased mass this could have been the keto diketide form or the hydroxy diketide form, or a mixture of the two Surprisingly, given the results of incubation with methylmalonyl-CoA alone, there was no evidence for the methylmalonyl derivative
of the chain-extension ACP, despite the presence of the free thiol form of the domain
The nature of the diketide adduct in this experiment was determined by treatment of a sample of the loaded ACP-TE di-domain with hydrazine to remove the added diketide ligand as the hydrazide derivative
Methyl-malonyl adduct
Unloaded forms
Methyl-malonyl adducts
Unloaded form
adduct
33 200
32 800
Mass (Da) Mass (Da)
Fig 3 The electrospray mass spectra
pro-duced by the AT and ACP-TE fractions
derived from the extension module 1 after
incubation with methylmalonyl-CoA The AT
domain shows two sets of peaks that arise
from alternative sites of proteolysis in the
downstream linker to KR.
Table 1 Occupancy levels of intermediates on chain extension ACP1 under various assay conditions.
Percentages of derivatized forms of the chain-extension ACP domain
a Absence of keto-diketide assumed by analogy with experiment 3.
Trang 6First, suitable conditions for the reaction were
estab-lished by a control study with a synthetic sample of
the N-acetylcysteamine analogue (Fig 6) The extract
of the reaction mixture showed a product which was
identified by high-resolution MS (calculated for
expected product [M+H]+ 147.1133, found [M+H]+
147.1150) Further support for the structure of the
hydrazide was obtained from an MS⁄ MS spectrum,
which produced a fragment ion for loss of H2O at m⁄ z
129.1 Repeating this experiment with the loaded
ACP-TE di-domain gave the same product, as judged by
MS analysis Careful examination of the LC-MS trace
failed to show evidence for any of the possible
hydra-zine derivatives of the keto analogue of the diketide
(m⁄ z 145 or 127), and so a confirmatory control
exper-iment with the keto derivative was not necessary
Analysis of the ACP-TE fraction recovered from the
hydrazine treatment showed it to be in the free form,
as expected These experiments confirmed that the
DKS is active in diketide synthesis under these
condi-tions, and that the hydroxy diketide intermediate
accu-mulates on the synthase in a significant quantity It
appears that very little, if any, of the accumulated
diketide intermediate is in the unreduced keto-form
When the natural starter unit was replaced with
butyrate (an increase in size of 14 mass units), analysis
of the DKS showed that the KS domain was fully
loaded by butyrate As with propionate, the
chain-extension ACP (ACP1) showed a peak for the
unloaded form and a peak for the form loaded with
the heavier diketide analogue (Fig 5C) The two
species were now present in approximately equal amounts (Table 1) Again, the isolated TE domain was free of diketide derivative, and there was no evidence for the methylmalonyl derivative of the ACP Incuba-tion of DKS with valeryl-CoA likewise led to the com-plete loading of KS with the valeryl group, but here there was a marked change in the pattern of loading
of ACP1 This now showed three peaks: free thiol group (37%), the valeryl diketide derivative (28%) and, in addition, the methylmalonyl derivative (35%) (Fig 5D) not seen when either propionyl-CoA or butyryl-CoA was used (Table 1) As in the previous two experiments, the TE was not found to be acylated
ACP KS
KR
TE AT
OH SH
SH SH
R
O O
S
CO2H O S
R
O OH O
R
O OH S
R O S R
O
S
R
O
O
CO2H O S
Fig 4 The range of biosynthetic intermediates predicted to be
covalently bound to the various enzyme active sites in the course
of a chain-extension cycle and product release on the DKS. Free 39 509
Methylmalonate
39 611
Diketide
39 644
39 300 39 500 39 700 39 900
Mass (Da)
Mass (Da)
A
B
C
D
Fig 5 MS results from experiments 3, 4 and 5 (A) Control experi-ment without added precursors (B) Incubation with propionate as the starter in experiment 3 (C) Incubation with butyrate as the star-ter in experiment 4 (D) Incubation with valerate as the starstar-ter in experiment 5, showing the expanded version of the ACP-TE region.
Trang 7Finally, an experiment was carried out in which the
natural substrates propionyl-CoA and
methylmalonyl-CoA were supplied, but in which NADPH was not
supplied The aim was to see if the keto-ester
interme-diate accumulated, and, if so, which stereoisomer
dom-inated in the keto-ester product As expected, the KS
domain was found to be fully loaded with a propionyl
unit, and the two domains of the loading module were
also substantially loaded with the starter acyl residue
units However, the omission of NADPH had a
marked effect on the pattern of loading on the
chain-extension ACP (ACP1) in this experiment There was a
peak for the methylmalonyl derivative, as well as for
the free thiol form, but there was no detectable peak
for any diketide intermediate Interestingly, the relative
proportions of the free thiol form of ACP1 (45%) and
the methylmalonyl form (55%) were identical to those
observed in the experiment in which no starter unit
was supplied, and the TE domain was again unloaded
Discussion
Incubation of DKS with methylmalonyl-CoA gives
incomplete acylation of AT1 and ACP1 domains
In our previous experiments [37] we showed that
incubation of the DKS in vitro with saturating
con-centrations of starter substrates led to complete
acyla-tion of the KS domain and nearly complete acylaacyla-tion
of both domains in the loading module By contrast,
in the present study we found that the level of
load-ing of methylmalonate on both the chain-extension
AT and ACP domains was considerably less than
100% As the AT-catalyzed reaction is readily
revers-ible, the extent of loading of methylmalonate on the
AT and ACP domains is probably determined (at
least in part) by the relative stabilities of free
methyl-malonyl-CoA ester and the loaded forms of the
domains, and thus by the concentrations of
methyl-malonyl-CoA and protein used in this study We have
also previously demonstrated that the AT domains of
purified DEBS possess a slow methylmalonyl-CoA
hydrolase activity [38], which may also influence the steady-state level of acylation of AT1 and ACP1 domains
Identification of rate-limiting steps, and a model for suppression of iteration and the maintenance
of fidelity of reduction
In the presence of all the (natural) substrates required for diketide synthesis on the DKS, the relatively high level of loading of multiple sites (> 50%) persists, apart from the TE domain It would appear that under these conditions, two chains can be elongated at the same time, and that the rate-limiting step is the trans-fer of the diketide intermediate from the chain-exten-sion ACP to the TE, not the subsequent release of the diketide acid from the TE This bottleneck at the exit stage causes a backlog of intermediates to build up at previous steps
When the alternative (progressively poorer) starter substrates butyryl-CoA and valeryl-CoA were used, the KS remained fully loaded and the extent of methylmalonate loading on the chain-extension AT was not significantly changed By contrast, there were dramatic changes in ACP1 occupancy With butyrate, the proportion of ACP1 loaded with dike-tide fell significantly, consistent with a slowing of the rate of the condensation step relative to the offload-ing step (we assumed that all the diketide intermedi-ate was in the hydroxy form in experiments 4 and 5) In the valerate experiment there was a more dra-matic change The proportion of the diketide inter-mediate fell even further (below 50%) and a detectable amount of ACP1 was found to be loaded with methylmalonate It would appear that the con-densation step, under these conditions, provides a significant additional bottleneck in the chain-assem-bly process However, the key point to emerge from these experiments is that by using the normal sub-strates, all the active sites are found to be heavily loaded This situation, if it holds for PKSs in vivo, would contrast sharply with a conventional metabolic pathway where overall rate control is dominated by early enzymes to avoid the accumulation of large pools of enzyme-free intermediates
Regulation of productivity at the release step could also provide important advantages for the control of fidelity in natural modular PKS assembly lines In the case of the complete DEBS assembly line, for example, regulation at the stage of product release would cause all the ACP and KS domains to be loaded by appro-priate intermediates and all the AT domains to be primed with the relevant building blocks This would
OH TE ACP
O OH
S O
OH
SNAC
O OH
H 2 NHN
Fig 6 Chemical treatment of the diketide N-acetylcysteamine
(NAC) derivative and the ACP-TE derivative with hydrazine to
release the hydrazide product for analysis by MS SNAC, thioester
of N-acetylcysteamine.
Trang 8lead to an intermittent mode of operation, in which
successive rounds of chain-extension cycles would be
triggered by release of the heptaketide intermediate
from the ACP domain in the last module (Fig 1),
rather like the operation of an automatic drinks-can
dispenser In each module, when the KS domain has
become free, it would be immediately reloaded with
substrate from the upstream ACP This would prevent
back-transfer of biosynthetic intermediates
subse-quently generated on the ACP domain within that
module, and suppress iterative use of the module The
term ‘congestion control’ has been suggested for this
effect [5] It is consistent with this hypothesis that
aberrant iteration in a PKS has been seen when the
levels of a PKS were increased without also increasing
the levels of intracellular precursors [15]
The throttling back of the process of product
release could also contribute to the maintenance of
fidelity in the reductive steps of the DEBS synthetic
operations It is vital that in every cycle all the
pro-grammed reductive steps are completed before
down-stream transfer of the fully modified intermediate
from the ACP to the KS domain of the next module
Molecular recognition cannot be the sole factor in
this, as shown most clearly for the mycolactone PKS,
where all 16 extension KS domains have an
essen-tially identical sequence [39] and cannot therefore be
expected to discriminate between the various
interme-diates sequentially generated in the upstream module
An alternative and simpler mechanism for the control
of fidelity is that all the steps of keto group
modifica-tion go effectively to complemodifica-tion, under the condimodifica-tions
in which PKSs operate; and that in every module the
downstream transfer of product to the KS of the next
module is delayed because it remains loaded until the
modification reactions have had time to reach
com-pletion The term ‘retardation control’ has been
sug-gested for this proposed effect [5] The implied high
level of loading in all the modules of modular PKS
multienzymes is consistent with the suggested ‘leaky
hose’ explanation for the release of biosynthetic
inter-mediates from the mupirocin PKS when downstream
catalytic sites are blocked [40]
Maintenance of fidelity in the reductive steps relies
not just on suppression of iteration, and of premature
transfer of incompletely reduced polyketide chains to
the next extension module, but also on precise control
of reaction stereospecificity by way of molecular
recog-nition between substrates and individual ketoreductase,
dehydratase and enoylreductase domains [10,41–44]
Perturbation of these interactions leads to inactivation,
or to the generation of aberrant products, both in vivo
[12,42] and in vitro [43]
Evidence for coordination of condensation and ketoreduction
In previous work [44] the operation of a single exten-sion module from DEBS was studied in vitro by mix-ing individually expressed and purified domains (ACP, KR) and di-domains (KS-AT) This flexible approach allowed various combinations of each type of domain
to be assayed and easily analyzed, and for individual steps to be deconvoluted For example, when a KS-AT di-domain was incubated with a diketide thioester substrate, methylmalonyl-CoA and ACP, keto triketide attached to the ACP was efficiently formed We there-fore expected that when DKS was incubated with substrates in the absence of NADPH, the ACP1 would
be found to carry the keto diketide However, under the conditions used, we found only the building blocks loaded on the KS and ACP1 domains respectively, and
no diketide intermediate Given the surprising nature
of this result, the experiment was repeated many times, always with the same outcome
It appears that either the keto diketide intermediate
is subject to very rapid release by the TE, or the con-densation step is seriously inhibited in the absence of NADPH The former explanation can be ruled out because no evidence for a ketoacid by-product was found in any of our careful searches for by-products in early investigations of the diketide synthase in vivo or
in vitro [9,45] The possible existence of an unexpected allosteric effect that inhibits the condensation step therefore deserves consideration The missing ingredi-ent, NADPH, would be expected to bind to the KR domain, not to either of the domains involved in the condensation step Any effect is therefore remote and must depend on the quaternary structure A similar inhibition of loading of methylmalonate onto an ACP was also reported in a study of the epothilone syn-thase, and again the effect was attributed to the qua-ternary structure [33] Figure 7 shows a schematic representation of the arrangement of domains within the extension module of the DKS, based originally on modelling and detailed proteolytic studies of DEBS that established the homodimeric nature of PKSs [46], and incorporating subsequent evidence from func-tional complementation [47], NMR studies [26] and X-ray crystal structures of intact animal FAS and of DEBS domains and di-domains [8,22] There is now a consensus that the two polypeptide chains in a ho-modimeric PKS are aligned ‘tail to tail’, as well as
‘head to head’, in each module, as predicted [46] In the absence of a crystal structure for an intact PKS module, it remains unclear how the two KS domains and the two ACP domains are able to approach
Trang 9closely enough to co-operate in the condensation, but
one suggestion is shown in Fig 7 The two ACP
domains move, perhaps in unison, along the axis
through a central passage, and so make contact with
the upstream pair of KS domains The double-helical,
rope-like twist of the two chains provides evidence
that the ACP of one chain makes contact with the
KS of the other [46,47] The cartoon shown here is
equivalent to the earlier ribbon representation [46],
except that the two-fold axis of symmetry runs
hori-zontally rather than vertically, and the alternative
direction of the helical twist is adopted in accordance
with that established from the recent X-ray structure
produced for the KS-AT di-domain [24]
In this working model of the DKS, the AT and KR
domains form a ‘collar’ surrounding the backwards and
forwards path of travel of the two ACP domains, as
they interact with their various catalytic partners The
collar shelters the central region and protects the
bio-synthetic intermediates from the surrounding aqueous
medium If the collar can expand and contract to
con-trol the lateral passage of the ACP domains, there
exists a possible mechanism by which the presence of
NADPH might enable the condensation step by
bind-ing to the KR and inducbind-ing a conformational change
that opens the collar The AT domain may also be
involved in such movements, as although the KS-AT
linker exists with a tightly folded tertiary structure in
the X-ray structure of the di-domain [24], this
interdo-main region is readily cleaved by the mild conditions of
proteolysis conditions used in the present study It is
tempting to speculate that the absence of
methylmalo-nyl loading of ACP1 when a readily converted substrate
(propionyl-CoA) was used, compared with the
signifi-cant level of methylmalonyl loading of ACP1 with a
KS1
AT1
ACP
ACP
KR1
TE
Fig 7 Proposed quaternary structure of the DKS chain-extension module based on the topology of the Cambridge Double Helical Model The two identical chains of the homodimeric structure are differentiated by red and blue colouring The KS domains have strong homodimeric interfaces (purple blocks) and are placed in contact at the ‘head’ of the structure The strongly homodimeric
TE domains are also placed in contact with each other at the
‘tail’ The remaining three domains are not homodimeric and so can move away from the common axis running through the pairs
of KS and TE domains The pair of ACP domains, however, are held close to the TE domains by short linker regions and so must remain in close proximity to each other and to the axis To aid visualization, the two domains, KR and AT, in the mid-section of the homodimer, are shown as small black blobs rather than as coloured spheres of appropriate size These domains are sited away from the axis of the proposed structure to free up a central passage The pair of ACP domains can now make contact with the pair of KS domains by moving parallel to the axis with the TE domains in tow The structure is also given a helical twist of 180 degrees in accordance with evidence that the ACP of one chain interacts with the KS of the opposite chain In the resulting qua-ternary structure, each ACP domain can access the appropriate
KS domain for the condensation step (and other domains in suc-cession through the chain extension cycle) by moving backwards and forwards (dashed green arrows) along the axis within the cen-tral core of the structure Because of the restricting effect of the short ACP to the TE linker, and the anticipated need for co-ordi-nated movements of the two looped-out domains, it is likely that the pair of ACP domains move backwards and forwards in tan-dem, rather than independently As a result, the successive reac-tions of the chain-extension cycle on the two chains might also
be constrained to operate in tandem At the start of each step of the chain-extension cycle, the pair of ACP domains would be loaded with identical intermediates and both would bind with the appropriate domain for the next step The first intermediate to complete the reaction would be free to leave its catalytic domain, but its ACP would stay put until the second intermediate had completed the same operation Both intermediates are now free
of the catalytic sites and the two ACP domains would then move
in tandem to co-operate with the next pair of catalytic domains.
In the Figure, the relative positioning of the KS and AT domains conforms to that shown in the X-ray structure of an isolated
KS-AT fragment [24] with the KS-AT domains in an outer position, remote from the axis However, the KS–AT linker undergoes facile proteolysis, so it must be in equilibrium with an unfolded form, not revealed in the X-ray image With the linker unfolded, the two
AT domains would be free to move to inner positions closer to the axis In the outer position, they would reload with methylmal-onate The subsequent move to an inner position (the original cartoon illustrating the structural principles of the Double Helical topology showed this quaternary conformation [46]) would facili-tate transfer of the methylmalonate to the ACP, and, in addition, would block premature access of the unloaded ACP thiol to the
KS active site, a situation that would lead to skipping of the mod-ule These predictions of co-ordinated movements of domains in the Cambridge Double Helical Model are consistent with the intriguing patterns of occupancy of ACP1 that are revealed in this investigation.
Trang 10less readily converted substrate (valeryl-CoA), is
preli-minary evidence of a threshold level (50%) of diketide
attachment to one of a pair of ACP1 domains being
sufficient to suppress the movement of the unloaded
partner ACP1 towards the AT1 domain for re-acylation
to occur A requirement for the ACP domains to
move in tandem could neatly account for such effects
Unfortunately, in cutting up a homodimeric
mega-synthase to facilitate MS analysis, a crucial aspect of
the loading information is destroyed: the analysis
reveals the average level of loading across all the
indi-vidual domains in the homodimeric species, but, in
modules that are only partly loaded, it does not reveal
how vacant and loaded sites are distributed within
individual multi-enzymes There is a pressing need to
develop MS protocols for analyzing larger proteins It
may then be possible to study loading patterns in
intact dimeric multi-enzymes, or at least in fragments
that retain the homodimeric bonding that exists in the
intact systems
Concluding remarks
Limited proteolysis followed by LC⁄ ion-trap MS is a
powerful and convenient technique for establishing
fea-tures of PKS catalysis that are not readily accessible
by other means The discovery that the first
condensa-tion reaccondensa-tion on the DEBS becomes an addicondensa-tional
bottleneck with an unnatural starter acid provides a
rational basis for efforts to improve productivity
Thus, alteration of the AT domain of the loading
module would be unlikely to remedy the limitation,
whereas replacement of the KS by one normally
oper-ating with longer acyl chains might do so Turnover
on the DKS with its normal starter acyl unit is clearly
regulated by the rate of release of the diketide product
by the TE domain Studies of the extent of loading of
multienzymes with more than one module are needed
to establish if product release is indeed a general basis
of regulation in PKS operations, especially in more
fully evolved systems, such as the DEBS The proposal
that there are two forms of control resulting from high
levels of occupancy of active sites by intermediates,
congestion control and retardation control, is based on
the direct evidence that the KS domain is fully
occu-pied under conditions of synthesis The evidence for
the proposed quaternary effects, suppression of
con-densation and tandemization, is more circumstantial
but deserves further study
MS also has the potential to play a major
support-ing role in structural studies of modular PKS
multi-enzymes It provides a method of quality control in
preparing samples for structural studies by NMR,
elec-tron microscopy, or X-ray crystallography This check
on structure will be particularly desirable in the prepa-ration of derivatized forms of multienzymes that have natural or unnatural ligands attached to the active sites
of selected domains
Experimental procedures Purification of the DEBS-derived DKS
The expression (in Escherichia coli) and purification of the engineered DKS (comprising the first extension module of DEBS1, covalently linked to the C-terminal TE domain from DEBS 3) have been previously described [9,37] In this construct the extension module ACP (referred to here
as ACP1) is a hybrid of ACP1 and ACP6, in order to preserve the native linker between the TE domain and the ACP
Limited proteolysis and LC⁄ MS analysis
Limited proteolysis was performed at a protein⁄ trypsin ratio of 80:1 (w⁄ w) at 30 C for 5 min After digestion, the mixture was immediately injected onto a pre-equili-brated C4 column (4.6· 250 mm, 300A˚; Vydac, Hesperia,
CA, USA) and proteins were eluted with a linear gradi-ent of 35–55% acetonitrile (containing 0.1% trifluoroace-tic acid) over 40 min The analysis was performed using online LC⁄ MS on an ion-trap instrument (LCQ Classic; ThermoFinnigan, San Jose, CA, USA) xcalibur 1.0 (ThermoFinnigan) software was used to operate the sys-tem, and bioworks 1.0 software (ThermoFinnigan) was used for mass deconvolution The detailed conditions for limited proteolysis and LC⁄ MS analysis have been described previously [37]
Substrate specificity for the chain-extender unit
A 6 mm concentration of malonyl-CoA or (RS)-methyl-malonyl-CoA was incubated with 6 lm DKS in a total volume of 30 lL, containing 400 mm potassium phos-phate (pH 7.4), 1 mm EDTA, 1 mm dithiothreitol and 20% glycerol The reactions were carried out at 30C for 10 min After the incubation, samples were immedi-ately subjected to tryptic digestion and analysed using
LC⁄ MS
Purification of commercial methylmalonyl-CoA
Commercial methylmalonyl-CoA was dissolved in distilled deionized (MQ) water (Millipore, Billerica, MA, USA), and loaded onto a reverse-phase C18 column (Prodigy C18, 4.6· 250 mm, 5 l; Phenomenex, Torrance, CA, USA) Methylmalonyl-CoA and propionyl-CoA were separated by