ARTICLE OPEN DOI:10.1038/s41467-017-02362-x Pathway design using de novo steps through uncharted biochemical spaces Akhil Kumar1, Lin Wang2, Chiam Yu Ng2 & Costas D. Maranas2 Existing retrosynthesis tools generally traverse production routes from a source to a sink metaboliteusingknownenzymesordenovosteps.Generally,importantconsiderationssuch 0 123456789 amsasbslecnodninsgervkantoiownn, ctorafancstfoorrmbaaltaionncse,wthitehrmpoudtaytnivaemsictefpesa,sicboilmityp,lemxiictyroobfialpacthhawssaiys stoepleocltoiogny,, and cost are largely dealt with in a posteriori fashion. The computational procedure we presentheredesignsbioconversionrouteswhilesimultaneouslyconsideringanycombination oftheaforementioned design criteria.First, we track andcodify asrules allreaction centers using a prime factorization-based encoding technique (rePrime). Reaction rules and known biotransformations are then simultaneously used by the pathway design algorithm (novoS- toic) to trace both metabolites and molecular moieties through balanced bio-conversion strategies. We demonstrate the use of novoStoic in bypassing steps in existing pathways through putative transformations, assembling complex pathways blending both known and putative steps toward pharmaceuticals, and postulating ways to biodegrade xenobiotics. 1TheHuckInstitutesoftheLifeSciences,ThePennsylvaniaStateUniversity,UniversityPark,PA16802,USA.2DepartmentofChemicalEngineering,The PennsylvaniaStateUniversity,UniversityPark,PA16802,USA.AkhilKumar,LinWangandChiamYuNgcontributedequallytothiswork. Correspondence andrequestsformaterialsshouldbeaddressedtoC.D.M.(email:[email protected]) NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications 1 ARTICLE NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02362-x A dvancesingeneticengineeringcapabilitieshaveexpanded existing reactions and novel reaction rules. First, we augmented therangeofchemicalssynthesizedbymicrobialplatforms the MetRxn repository19 with a new data set of elementally to non-natural synthetic molecules such as drugs and balancedreactionoperatorsusingtheautomatedCLCA-34based variouspharmaceuticalprecursors.Synthesisofthesenon-natural reaction rule extraction procedure termed rePrime. For each molecules often relies on enzymes with broad-substrate range1,2 reaction, the rePrime procedure identifies and captures as as well as enzymes with promiscuous activity on novel sub- a reaction rule the molecular graph topological changes under- strates3.ArecentsurveyofEscherichiacolienzymesrevealedthat pinning the substrate to product graph conversion. A reaction 37% of them could act on multiple substrates4. Protein engi- ruleisavectorthatcapturesthelocationofactivereactioncenters neering techniques have already demonstrated the feasibility of affectedbytheconversionofsubstratestoproducts.Thereaction expanding the substrate range of existing enzymes5–8. These rules and known reactions are then operated upon a mixed examples allude to the increasingly important role of in silico integer linear programming (MILP) procedure, novoStoic that protein modeling tools such as IPRO9 and Rosetta10 to system- identifies a mass-balanced biochemical network that converts a atically change enzyme substrate specificity. In addition, recent source metabolite to a target while satisfying a multitude of successes in the de novo enzyme design11,12 have expanded the constraints and optimizing the objective function. Both pre- scope of enzymatic functions that can be called upon in the cursors and possibly co-substrates and co-products required to construction of synthetic de novo metabolic pathways. Never- satisfy the mass balance constraints serve as optimization vari- theless, the systematic redesign of enzymes for new and novel ables (see Fig. 1 for a schematic overview). Overall, novoStoic activities remains a daunting challenge implying that novel identifies“bydesign”mass-balanced,highyield,andeconomically transformationsshouldonlybesparinglyusedonlyiftheyconfer favorable biotransformations with a negative overall standard pathway length, carbon yield, and/or redox benefits. Gibbs free energy change from a substrate(s) to natural and Computationalpathwaydesignandretrosynthetictoolscanbe synthetic product(s) by combining both existing and novel used to guide the systematic recruitment of either native or de reactions. novoStoic can also be used to elucidate strategies to novo enzymatic functions to assemble pathways toward targeted biodegrade xenobiotics by targeting a broad activity toward a chemicals.Thisisanareaofresearchwithsignificantpriorwork. topologically diverseset of substrates. In theResults sections, we Network-based path finding methods such as PathComp13, introduce four case studies of different complexities, i.e., an Pathway Hunter Tool14, and MetaRoute15 identify linear path- illustrative example of rePrime/novoStoic procedure, the bio- waysfromasinglesourcetoonetargetmolecule.Thesemethods synthesisof1,4-butanediol(bdo),thesynthesisofphenylephrine, rely on heuristics such as substrate-product similarity, atom and the biodegradation of benzo[a]pyrene. transitions, or substrate-product reaction co-occurrence fre- quency to reduce carbon loss while designing the path from source to target. In contrast to linear pathfinding methods, net- Results work optimization-based approaches such as CFP16, k-shortest An illustrative example for rePrime and novoStoic. We first EFM17, and optStoic18 can incorporate non-injective (i.e., not clarify the rePrime and novoStoic procedures using only two necessarilyaone-to-onemappingbetweenreactantstoproducts) decarboxylase reactions (i) 2-hydroxyisophthalate decarboxylase stoichiometry and directlymodel carbon flow aswell as cofactor (2HIPD) and (ii) salicylate decarboxylase (SLD), and four parti- balancing. All network-based path finding methods require lit- cipating metabolites. The first step (rePrime) extracts reaction erature extracted information available in biochemical databases rules from a database of known reactions (Fig. 1). To this end, such as MetRxn19, KEGG20, BRENDA21, and MetaCyc22. Pre- each metabolite i within the database is first encoded using a diction methods, on the other hand, expand upon uncovered molecular signature (see Supplementary Methods for d(cid:1)etai(cid:3)l pro- knowledge space by suggesting putative reactant combinations cedure). For a moiety size λ, a molecular signature Cλ is a mi plausibleunderthetenetsoforganicchemistry.Bydrawingfrom vector that concatenates into a single numeric value the number biochemistry principles, atom connectivity changes, or molecule of attributes for every moiety m in a metabolite i, described as a fingerprint changes between substrates and products have been collection of prime numbers centered at node n. For λ=1, the encoded as reaction rule operators in formats such as BEM23, moietyissimplytheatomatnoden,whereasatλ=2themoiety RDM24, and SMIRKS25. Pathway prediction techniques such as is composed of all the atoms bonded to the atom at node n BNICE26, XTMS27, UM-PPS28, PathPred29, Route Designer30, (Fig. 2a). Figure 2b, c shows the iterative rePrime procedure to and GEM-Path31 employ reaction rule operators for a single generate molecular signature at moiety size λ=1 and 2. In brief, molecular target iteratively in a retrosynthetic fashion so as to each iteration involves the assignment of a canonical label fol- identify a bioconversion from a single source. This traversal lowed by a prime number assignment to each node. The unique strategy invoked by such retrosynthesis algorithms prunes the prime numbers are then counted and stored in the molecular vast combinatorial space of putative transformations by evaluat- signaturevector.WeapplytherePrimealgorithmtogeneratethe ing free-energy change and substrate similarity metrics at each molecular signatures for metabolites 2-hydroxyisopthalate iteration. However, after each step, the trade-off between carbon (2hipa), salicylate (sal), carbon dioxide (CO ), and phenol 2 yield,energy(ATPrequirements),andthermodynamicfeasibility (phnl) at different moiety sizes λ (see Fig(cid:4). 2b,(cid:5)c and Supple- of the main carbon conversion path often remain unexplored. mentaryTables1,2,and3).Areactionrule Tλ isdefinedasa mj Instead,toolssuchasTMFA32andEFM33areusedinaposteriori vector that captures the changes in all moieties m of the parti- stepstoassessenergyandcarbonefficiency.Computationaltools cipating metabolites upon reaction j. It is derived based on the such as optStoic18 can be used to design mass and energy reaction stoichiometry and the corresponding molecular sig- balanced pathways. However, they are limited to using reactions natures of the reactants and products. A reaction rule uniquely present in existing biochemical databases and metabolic models. captures the eliminated and newly formed moieties around the Therefore, invoking novel molecules as intermediates through reaction center. In Fig. 2d, we generated the reaction rules for hypothetical reactions while maintaining a mass and energy reactions 2HIPD and SLD for moiety size λ=1. Upon removing balanced pathway remains elusive. repetitive rules, a unique set of reaction rules are indexed by set To address the problems in current computational methods, Rλ(i.e.,Tλ ).Hence,T1 andT1 arenowstoredasT1 . mr m;2HIPD m;SLD m;1 hereinwepresentrePrimeandnovoStoic,anoptimization-based The second step (novoStoic) uses an MILP representation to de novo pathway design framework that seamlessly integrates pose the task of identifying a component and moiety balanced 2 NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications ARTICLE NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02362-x MetRxn rePrime Extract unique reaction rules from a database of reactions Reaction n Name Reaction rule r MetRxn PhosRpehaocetinoonl p2yruvate Reaction 44,784 pRheoasNcptaihomantease rules (T) Tm1,r =Rea1c,tio−n ru1le, 1 PhosSpahlioRceyelnaaotceltpi odynreu c1vaartbeo +xOy H →lase −1 2v 3 Reactions PyruRveaateNc ta+iom Pnehosphate 1 =Tm13,1 =−Cm12,2 +Cm1,3 −Cm1,1 Sali2cy-Hlayte→d ro Pxhyeisnoopl t+h aClaatreb on −1 2 0 3 dioxdideecarboxylase 0 10 01 2 2 Reaction = + − 2-Hydroxyisophthalate → −2 3 0 5 Salicylate + Carbon dioxide novoStoic Pathways Identify biochemical network that Pathway A Pathway B converts source molecule to Rxn 1 Rxn 2 Rxn 1 Rule 1 target molecule ? Formulation Reaction 2 Reaction rule 1 Maximize profit Name or Salicylate decarboxylase Tm1,1 =Cm1,2 +Cm1,3 −Cm1,1 Minimize number of reactions Reaction 1 −Re1action2 1 0 3 Reaction Subject to: SaNlicaymlaete→ Phenol + Carbon N1ame= 1 + 2 − 2 (cid:129) Uptake of source molecule(s) dio2dx-eidHceaydrbrooxxyyilsaospethalate −2d-e2Hcyadrbrooxx3yyilsaospet0halate5 (cid:129) Export of target molecule(s) Reaction Reaction (cid:129) Network stoichiometry 2-Hydroxyisophthalate → 2-Hydroxyisophthalate → (cid:129) Number of reaction rules ≤ limit Salicylate + Carbon dioxide Salicylate + Carbon dioxide (cid:129) Thermodynamic feasibility Fig.1SchematicoverviewoftherePrime/novoStoicprocedure.First,therePrimeprocedureisusedtopre-processtheMetRxndatabaseofreactions(blue boxes)toextractauniquesetofreactionrules,Rλ(redboxes)atmoietysizeλ.ThereactionrulesarederivedfromthemolecularsignatureCλ ofeach mi participatingmetabolitesandarecapturedbyTλ .ThenovoStoicprocedureisthenusedtoidentifyaseriesofinterveningreactionsandreactionrulesthat mr convertsourcemolecule(s)(greencircle)intotargetmolecule(s)(orangecircle)suchthattheprofitcanbemaximizedorthenumberofreactionsinthe pathwaycanbeminimized.Othercriteriaincludingthenumberofreactionrulesandthermodynamicfeasibilityofthepathwaycanbeflexiblyincorporated asconstraints.Bycontrollingthenumberofreactionrules,twopossiblepathwaysdesignedbynovoStoicareshowninthebottomrightpanel.PathwayA usestwoknownreactions(blueboxes)thatpresentintheMetRxndatabase,whereaspathwayBusesacombinationofaknownreaction(bluebox)anda reactionrule(redbox)toperformthesameconversion biochemical pathway that converts a source to a target The toy example h(cid:1)elps(cid:3)explain how the tw(cid:1)o k(cid:3)ey quantities metabolite as an optimization problem. It searches the database molecular signatures Cλ and reaction rules Tλ are used to mi mr of both known reactions and reaction rules to complete the imposeconstraintsformassandmoietybalance.Inthefollowing pathway (Fig. 3) by introducing moiety balances along with case studies, we will explore much more complex conversion componentbalancesaskeyconstraintsintheMILPformulation. strategies that make use of multiple novel reactions absent from Thetwobalancesarelinkedthroughtheimbalancemetabolicflux the reaction databases. Each reaction rule identified is then vimb,whichquantifiesthesurplusordeficitforanymetabolitesin matched to a corresponding enzyme homolog. Supplementary themetabolicnetworkthatnecessitatestheinvolvementofnovel Methods provide a thorough description of the algorithmic reaction steps (Fig. 3a) to b(cid:4)alance the ove(cid:5)rall conversion. The details of rePrime and novoStoic along with a pseudo-code P overallchangeformoietym Cλ vEX issettobeequalto description. i2Iex mi i (cid:1)thPe moiety ch(cid:3)anges incurred in both the novel reaction network λ 1,4-butanediol synthesis. 1,4-butanediol (bdo) is a commodity r2RλTmrvr and(cid:1)tPhe ones im(cid:3)plied by the known metabolic chemical that is used as an industrial solvent and as a monomer reaction network i2ICmλiviimb (Fig. 3b). Herein, we used in polymer synthesis. Efforts are under way aimed at replacing novoStoic to design biosynthesis pathways from 2hipa to phnl petroleum-derived production with more sustainable processes and obtained four separate solutions. novoStoic not only using bio-based production based on genetically engineered identifiedapathwaywhichinvolvestwoknownreactions2HIPD microbial strains35. By leveraging computational design and a(cid:4)nd S(cid:5)LD but also pathways which involve a single reaction rule scoring of a large number of bdo biosynthetic routes, Yim et al. T1 implying a novel conversion (see Supplementary Fig. 1 reconstituted a pathway in E. coli with a titer of 18g/L[35]. m;1 Herein, we demonstrate three pathways designed using KEGG and Supplementary Fig. 2 for the details). NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications 3 ARTICLE NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02362-x a Distance ((cid:2)) 1 2 3 Moiety C # unique 50 298 1100 moieties 2-hydroxy Atom n Isophthalate Neighbor(s) of atom n at (cid:2) = 2 # reaction 826 1929 6043 (2-hipa) Neighbor(s) of atom n at (cid:2) = 3 rules b (cid:2) = 1 Assign prime to Generate molecular Define atom feature each atom feature signature 1-2-08-0 1-1-08-1 3 2 Atom feature: 3-4-06-0 3-4-06-0 11 Count Moiety (m) 3: # of non-hydrogen adjacent atoms 1-1-08-1 3-4-06-0 2 11 3 2 4: # of non-hydrogen bonds 3-4-06-0 2-3-06-1 11 5 C1m, 2hipa=23 35 06: Atomic number 1-2-08-0 3-4-06-0 2-3-06-1 3 11 5 0 7 0: # of hydrogen bonds 5 11 3-4-06-0 2-3-06-1 11 5 1-1-08-1 2 c (cid:2) = 2 Assign prime to each Generate molecular Calculate prime-product prime-product signature Prime-product CountMoiety (m) 3 2 99 44 5 2 3 2 112 · 3 · 2 · 11 211 11115 = 7986 294,4282 737,9280165375 2 29 311913 C2m, 2hipa= 0201 13571 5 2 13 3 11 5 99 73,250 625 31 11 0 17 11 5 7986 1375 19 13 2 19 0 23 2 44 2 1 29 2 31 d Inference of reaction rule ((cid:2) = 1) Rxn 1: 2-hydroxyisopthalate decarboxylase (2HIPD) Rxn 2: salicylate decarboxylase (SLD) 3 112 2 3 2 3 Reaction 3 2111111 151 55 2 511 111155 + 3 C7 3 2 511 111155 2 151 55 55 + 3 7 3 2-hydroxy 5 2 isophthalate Salicylate CO2 Salic5ylate Phenol CO2 Moieties C1m, 2hipa C1m, sal C1m, co2 C1m, sal C1m, phnl C1m, co2 3 2 2 2 0 2 2 2 1 2 0 2 2 3 1 3 2 3 1 3 0 3 2 3 3 5 4 5 + 0 5 4 5 5 5 + 0 5 0 7 0 7 1 7 0 7 0 7 1 7 5 11 3 11 0 11 3 11 1 11 0 11 Reaction rule T1 = C1 + C1 − C1 = [−1 1 1 1 −2]T T1 = C1 + C1 − C1 = [−1 1 1 1 −2]T m, 2HIPD m, sal m, co2 m, 2hipa m, SLD m, phnl m, co2 m, sal Note: At (cid:2) = 1, T1 = T1 = T1 ;At (cid:2) = 2, T2 ≠ T2 m, 2HIPD m, SLD m, 1 m, 2HIPD m, SLD Fig.2TherePrimeprocedureisdemonstratedforthemolecule2-hydroxyisophthalate(2hipa).aThemoietysize(λ)indicatesthedistancebetweennodes inamoleculargraph.Atλ=1,themoietyatnoden(thecarbonatomshadedinorange)istheatomitself.Themoietycenteredatnoden,thenumberof uniquemoietiesandreactionrulesextractedfromMetRxnatλ¼f1;2;3gareprovidedonthetable.b,cTherePrimeprocedureiterativelyassignsprime numbertoeachnodeatλ=1,2(seeSupplementaryMethodsfordetails).Themolecularsignatureof2hipa,Cλ containsthecountofuniquemoieties m;2hipa (i.e.,primenumber).TheresultingCλ increasesinspecificityatlargermoietysize(λ)asneighboringatomsarealsoinvolvedindeterminingthe m;2hipa canonicallabelofthenode.dThederivationofreactionrulefortwodecarboxylasereactions(i)2-hydroxyisophtalatedecarboxylase(2HIPD)and(ii) salicylatedecarboxylase(SLD)aredemonstratedatλ=1.TheresultingreactionrulesT1 andT1 areidenticalatthismoietysize,hencetheyare m;2HIPD m;SLD storedasT1 .Thespecificityofthereactionrulesincreaseswithincreasingmoietysizeλ,thereforeT2 isnotequaltoT2 (SupplementaryTable3) m;1 m;2HIPD m;SLD 4 NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications ARTICLE NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02362-x a Source metabolite Target metabolite V j ? Vimb vr i Rules b Moiety Tm(cid:2)rvr + Cm(cid:2)iViimb = Cm(cid:2)iVi E X ,∀m ∈ M balance r∈R i∈I i∈I ex Moiety changes by Moiety changes by Overall moiety changes reaction rules reactions −1 2 Reaction −1 3 3 2 rules (T) CountMoiety (m) 1 5 11 3 2 2 2 0 7 1 2 2 11 11 5 2 3 1 3 −2 11 0 3 2 5 3 5 4 5 Tm1r 5 5 11 5 3 11 5 0 7 0 7 0 7 11 5 vsEoXurce 5 11 3 11 vr 1 11 VtEaXrget 5 5 2 is2o-phhytdhraolxayte Cm1, 2hipa Cm1, sal Cm1, phnl Phenol5 V2imhipba Vsimalb MetRxn 3 112 2 3 2 44,784 11 11 11 Reactions 5 Vj 2 11 11 3 11 5 5 121-hyd5roxy 5 5 2 isophthalate 5 Salicylate 2hipa: 2-hydroxyisophthalate; sal: salicylate; phnl: phenol Fig.3ThenovoStoicprocedure.novoStoicseamlesslyblendexistingreactionsandreactionruleswhiledesigningapathwayfromsourcemetaboliteto targetmetabolite.aThevariablevimbdefinestheconnection(greenarrows)betweentheknownreactionsnetwork(blackarrows)andthereactionrules i network(redarrowsandreddottedbox).Itindicatesthesurplusordeficitofmetaboliteiintheknownmetabolicnetwork.Thereactionrulesnetworkis invokedtocompleteapathwaywhennoknownreactioncancatalyzetheconversion.bIllustrationofthecomponentbalanceandmoietybalancedefinedin novoStoicconstraints(2)and(3)(SupplementaryMethods).Thebottomhalf(graydottedbox)istheknownnetworkoperatingonmetabolites,whilethe upperhalf(reddottedbox)isthereactionrulesnetworkoperatingonmoieties.Thesectionoftheentiresystemthatisinvolvedinthemoietybalance constraintiscolor-coded:(red)moietychangesbyreactionrules,(blue)overallmoietychanges,and(green)moietychangesbyknownreactions.The pathwayhereconverts2hipatophnlusingaknownreaction(i.e.,2hipa→sal,blackarrow)andareactionrule(i.e.,sal→phnl,redarrow).Metabolite 2hipafirstentersthereactionrulenetworkusingexchangereaction(bluearrow,vEX )andsubsequentlyexportedtotheknownreactionnetwork(green source arrow,vimb ¼(cid:2)1).Itisconvertedbyaknownreaction(blackarrow)intosal,whichisthentransferredtoth(cid:4)ereact(cid:5)ionrulenetwork(greenarrow, 2hipa vismalb¼1).ThereactionruleTm1r (redarrow)convertssalintophnl,whichisexportedoutoftheentiresystem vEtaXrget .Notethatforclaritypurpose,the involvementofcarbondioxideintheconversionof2hipatosalandsaltophnlisomitted NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications 5 ARTICLE NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02362-x OH OH OH CoA C C O O C C 1 O C C C C O 2 O C C C C OH 3 O C C C C OH CoA Succinyl-CoA Succinic semialdehyde 4-hydroxybutanoate 4-hydroxybutyryl-CoA h+ h+ coa nadh nad nadph nadp AccoaAcetate h+ 7 nahd+ph 6 nnaaddpph 4 h+ nadp R1 nadph R1 h2o coa nadp h2o 8 5 Pathway A: 1–5 Succinic aldehyde 4-hydroxybutanal 1,4-butanediol Pathway B: 1, 2, 6, and 5 C C Oh+ C C OHh+ C C OH Pathway C: 1, 7, 8, and 5 O C C nadph nadpO C C nadph nadp HO C C R2 Fig.41,4-butanediolbiosynthesis.Thefigureshowsthreeroutesfromsuccinyl-CoAto1,4-butanediol.Knownreactionsareindicatedbysolidlineswhile reactionrulesaredenotedwithdashedlines.Everyhypotheticalreactionstepisindexedwithareactionruleidshowninred(seeTable1fordescription) database and our novoStoic algorithm, which seamlessly com- on only increasing NADPH availability without having to worry bines known and novel steps starting from the precursor about the balance of the stoichiometric ratio NADH:NADPH succinyl-CoA (succoa) toward bdo. novoStoic provides possibi- while improving both their availabilities. litiesforloweringthenumberofstepsinthepathwaybyinvoking noPvealthswteapys.4A identified by novoStoic is a five-step pathway of succoaþ4NADPHþ4Hþ(cid:2)52(cid:2)kc!al=molbdoþ4NADP bdosynthesisthatrecapitulatesthedownstreampathwayofYim þcoaþH2O: etal.35(Fig.4,steps1–5).Thispathwayconvertssuccoaintobdo with a concomitant oxidation of four NAD(P)H and production of two co-enzyme A (coa) molecules. Co-substrate acetyl-coA is This exact pathway was identified before by GEM-path31 and convertedintoacetateinthethirdstep.Theoverallconversionof foundtobeexhibitinghighertheoreticalproductyieldusingFBA. this pathway is given by In contrast to pathway 4B, two NADPH-dependent novel steps succoaþ3NADPHþNADHþ4Hþþaccoa(cid:2)49(cid:2)kc!al=molbdo are required in this pathway, in order to bypass the NADH- dependent4-hydroxybutyratedehydrogenase(step2).Thedirect þacþ3NADPþNADþ2coa: production of succinic aldehyde from succinic semialdehyde (product of step 1) was proposed by novoStoic by invoking a homolog of succinate-semialdehyde dehydrogenase (R1). The Pathway 4A can be further improved upon by preventing homolog of 4-hydroxybutryate dehydrogenase (R2) was next carbon flux from draining toward undesirable co-substrates/co- suggested for the conversion of succinic aldehyde to 4- products (i.e., accoa and acetate) when glucose is used as hydroxybutanal. In all three pathways, the intermediate 4- feedstock thereby leading to a higher theoretical carbon yield. hydroxybutanal is produced in the second to last step, which is Furthermore, we explored if a shorter pathway is feasible upon then converted into bdo using alcohol dehydrogenase (step 5). invoking one or two novel transformations. Therefore, we Despite requiring the same numbers of redox cofactors, imposed three additional design criteria: (1) the fewest number pathways 4B and 4C designed by novoStoic bypass the ofreactionrules,(2)themaximalnumberofstepsfromsuccoato involvement of acetyl-CoA and generation of the by-product bdo is less or equal to four, and (3) no acetyl-CoA as a co- acetate. We caution the reader that computational pathway substrate. Criterion (1) was set as the objective function and design using novoStoic requires careful scrutiny of the cofactor criteria (2) and (3) were defined as constraints in the novoStoic systems as there are often inconsistencies between current MILP formulation. As a result, novoStoic identified pathway 4B databases and experimental studies. For example, reaction rule thatrequiresonlyfourreactionsteps(Fig.4).Pathway4Bshares R1 was reported to utilize ATP as a cofactor36, whereas the threereactionswithpathway4A(steps1,2,and5).Inadeparture manually approved Rhea database37 and KEGG20 denote it as a frompathway4A,intermediate4-hydroxybutanoategeneratedby reversiblereactionwithoututilizationofATP.Whenweenforced step2isdirectlyconvertedinto4-hydroxybutanal,thusbypassing rule R1 to use ATP (Supplementary Tables 5, 6), novoStoic steps3and4,usingreactionruleR1(Fig.4;Table1).Aputative designed a pathway with a different overall stoichiometry succinate-semialdehyde dehydrogenase is suggested as the including ATP as one of the cofactors: enzyme homolog for this step. The overall conversion can be described as followed: ATPþsuccoaþ4NADPHþ4Hþ(cid:2)58(cid:2)kc!al=molbdo succoaþ3NADPHþNADHþ4Hþ(cid:2)52(cid:2)kc!al=molbdo þ4NADPþcoaþADPþpi: þ3NADPþNADþcoaþH O: 2 This small molecule focused example demonstrates the Pathway 4C was identified by novoStoic upon appending the potentialofnovoStoictohandlemultipleconstraintsonpathway constraintofusingasinglecofactor(i.e.,NADPH).Reducingthe design and bypass multiple known steps by invoking novel numberofcofactorscanhelpfocusthestrainengineeringefforts conversions only when needed. 6 NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications ARTICLE NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02362-x Table1Reaction templatesfor1,4-butandiol (bdo)synthesis OH OH O O O h+ O OH nadph nadp Succinate Succinate semialdehyde OH OH O OH O h+ O nadph nadp Succinate semialdehyde 4-Hydroxybutanoic acid Thereaction-ruletemplatecolumnpresentsthereactionsthatcorrespondtothereactionrulesidentifiedbythealphanumericscheme(i.e.,R1andR2)for14bdosynthesis.Stepididentifiesthereaction thatwaspredictedfromthecorrespondingreactionrule Phenylephrine synthesis. The non-natural molecule pheny- steps 7 and 9, thereby maintaining the NADH/NAD+ cofactor lephrine(phep)isamemberofthephenylethanolaminesclass.It balance. mimics the action of stimulants such as adrenaline and dopa- Inthesecondhalfofthenetwork,thephaltophepconversion mine.Traditionally,phenylephrineisproducedthroughchemical starts with the reduction of phenylalanine by a homolog of synthesis involving the reduction of the aromatic substrate phenylalaninase(step10).Atypicalphenylalaninasehydroxylates m-hydroxybenzaldehyde38. Figure 5 illustrates three pathways phal at the 4th carbon position of the phenyl ring to produce p- each starting with a different aromatic precursor with B and C tyrosine(Table2,R6).However,toproducem-tyrosine(mtro),a identified by novoStoic. The pathways involve (i) homovanillate hydroxylation at the 3rd carbon position is suggested (step 10). (hmv) as the only substrate, (ii) co-utilization of phenylalanine Thisreactionisnotpresentinanyofthereactiondatabasesandis (phal) and methyl salicylate (msal), and (iii) catechol along with predicted de novo by novoStoic. Nevertheless, pacidamycin methyl salicylate. studies focusing on bacterial phenylalaninase have indicated the Pathway5Aisanexampleofadesignwithapositivestandard presence of the phenylalaninase homolog in many Streptomyces Gibbs free energy of change (Fig. 5a). It involves three existing specieswithregiospecifichydroxylationactivityatthe3rdcarbon reactions (steps1–3)and threereactionrules(R3,R4,and R5in atom of the phenyl ring39. For example, the phal to m-tyrosine steps 4–9) to perform the overall conversion as followed: reaction suggested by novoStoic is predicted as a secondary activity by the phenylalanine hydroxylase homolog encoded by hmvþNH3þ1(cid:2)2!7kcalphepþO2: pacX from Streptomyces coeruleorubidus39. Step11involvesthedecarboxylationofm-tyrosinetoformm- tyramine catalyzed by m-tyrosine decarboxylase (EC 4.1.1.28) Itisasolutiontypicallypredictedbysubstratesimilarity-based (Fig.5b).Theconversionofm-tyraminetophep(steps12,13,15, retrosynthesistools.Suchalgorithmsfirstchoosealikelysubstrate and 16) in the remainder of the network is identical to the as the target feedstock and then proceed to prune the retro- subnetwork of pathway 5A with the same N-methyltransferase synthesis reaction network using similarity-based filters on each (R4) and oxidoreductase (R5) reactions. With msal contributing backward reaction step until the shortest (one substrate to one only one carbon as a methyl group toward phep synthesis, we product) linear route is found. The free energy change in the can,therefore,consideronlythesecondhalf(steps10–14)ofthe direction of bioconversion toward phep is positive and thus network for phep if endogenous adm is supplied. thermodynamically infeasible. Pathway 5C shown in green in Fig. 5c combines a number of Pathway 5B involves the conversion of co-substrates phal and fungalreactionsfromthetyrosineandphenylalaninemetabolism msaltophepwithaconcomitantproductionofL-serine(ser)and pathways for the conversion of catechol and methyl salicylate to acetate (Fig. 5b). The overall conversion for pathway 5B is: acetaldehyde and phenylephrine. This pathway is similar to the (cid:2)323kcal phytochemical synthesis pathway of pseudoephedrine, wherein phalþmsalþ2H Oþ4O þNH (cid:2)! serþphep 2 2 3 the carboligation product of the benzoate derivate and pyruvate þacetateþ3CO þH O : undergoes transamination and subsequently N-methylation40. 2 2 2 The overall conversion for pathway 5C is: In the first half of the 14 reaction cascade (steps 1–9), the catecholþmsalþ2O þH OþNH (cid:2)(cid:2)60!kcalphep 2 2 3 salicylate produced uponthe demethylation of msal is converted þacetaldehydeþ2CO : to acetate and pyruvate via the benzoate degradation pathway. 2 Pyruvate is then aminated by serine hydratase to produce L- serine.ThedegradationofmsaltoL-serineandacetaldehydealso produces an S-adenosyl-L-methionine (adm), which acts as a Two possible routes with 17 reactions each were suggested methyldonorinthesecondhalfofthenetwork(steps10–16).In (Fig. 5c). novoStoic predicted a novel reaction (R3) to convert addition,theNADHconsumedinsteps2and10isregeneratedin 23hpatom-tyramineasthenextstep(step14),whichrequiresa NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications 7 ARTICLE NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02362-x a C h2o2 2 h2o N C 7 C C admadh C C ascdhasc C C O C O O C O O C O C O C N 5 OC 6 C C C C 1 C C C C 2 C C C C 3 C C C C nh 4 o C C C C R4 mep R5 C CChOmCvC O dChOCpaC Oh2o`o2hC3+hOCpphC+O `h2o 2C3OChpCa 3 R32 CmOtCyrC ha2soc2dh2a7 shc2OoCC CN C C adom2 ahd2oh pheOpCC CNC CC CC nad nadh nadh nad h2o2 ho C C C O adh adm 2 8 nrOfn R5 R4 9 o ho 2 2 b O OCC CCC OCC adh adm OCCsCCClaCCO O oh22o` + coC2OaCCtecOCChCCol o2 O OCCC2hCCCsOO 4 2OhpOCCCaCChO2o5 OO4oOCCpOCCaC 6acalhdp2oyrnad7nahd+hAsceertaCteO msalC 1 hn+adhnad 3 h2o fmha+ co2 8 O CC N 2 CN h2o2 2 h2o n9ad nadhh+ nh3 O nad nadh C 14 C C h+ admadh C C ascdhasc O C O O C O C N 12 C C C N pCCChCCOalCC thb dhb NO CCCmtOCCroCC co2 C CCmOCCtyrCC h2o2R241 4h2oCmNeOp R5 o2 h2o13 O CC CCN C C 11 ascdhascO C C C C adm adh phep C C C C C 10 o2R6h2o 15 nrOCfn O R5 R4 16 c co 2 nadh h+20 o2 h2o O nad O O C C O OC O O C O O CC CC CC o21 CC CC Oh2o2fmaO CC CC h2o3 O OC CC C 4acald O O O C O Catechol O OCC C2CChOCCs 5 OCC2CCChpOCCaOo26h2oO OC4CoCCCpaCCO O 7h2OoOCCpCCCyrCCO O C OC OCCC OCCC OCC 9O COC CCCCCOOCCC O10 h+O CCC OCCC CCo211cOo2CC COC OOCC CC msal C sla 25hb 3hb chrm prph 34hpp hmg h+ h+ 8 co adh adm nadh nad nadh nad nad nadh 2 C h2o2 2 h2o N C 17 C C admadh C C ascdhasc C C O C O O O C O C O C N 15 OC C CChOCCmgCCh2o12o2C 3CChOCCppCC 13h2oC 2CC3OCChpCCanh314 o2 C CCmOtCCyrCC h2o2R241 7h2oCmeNp R5 o2 h2o16 O CC CCN C C h+ h+ ` R3 ascdhascO C C C C adm adh C C C nad nadh nadh nad h2o2 h2o 18 C C C phep O O nrfn R5 R4 19 o ho 2 2 Fig.5Phenylephrinesynthesis.a–cThreeroutesaredepictedforthesynthesisofthephenylephrinefrombenzoatederivatives.Eachrouteiscolorcodedto differentiatetheroutebychoiceofprecursorandoverallconversion.Inaddition,theprecursorandtargetmoleculesarecoloredtodepictmoiety translocation.Knownreactionsareindicatedbysolidlineswhilereactionssuggestedusingreactionrulesaredenotedwithdashedlines.Everyhypothetical reactionstepisindexedwithareactionruleid(seeTable2fordescription).Thestructureforpyruvate(pyr),acetaldehyde(acald),acetate,carbondioxide (CO2),oxygen(O2),hydrogenperoxide(H2O2),formate(fma),S-adenosyl-L-methionine(adm),S-adenosylhomocysteine(adh),ascorbicacid(asc), dehydroascorbicacid(dhasc),tetrahydrobiopterin(tdh),and4a-hydroxytetrahydrobiopterin(dhb)arenotshownforclaritypurpose 8 NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications ARTICLE NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02362-x Table2 Reactiontemplatesforphenylephrine synthesis Rule Reaction-rule template Enzyme/homolog name Step id id R3 TYRAMINASE 4a, 14c 4-Hydroxyphenylacetaldehyde p-Tyramine 5a, 9a, 12b, R4 TYRAMINE METHYLPHERASE 16b, 15c, p-Tyramine n-Methyltyramine 19c 6a, 8a, 13b, 15b, R5 DOPAMINE HYDROXYLASE 16c, Dopamine L-Noradrenaline 18c R6 PHENYLALANINASE 10b L-Phenylalanine L-Tyrosine S-adenosyl-L-methionine(adm),S-adenosylhomocysteine(adh),ascorbicacid(asc),dehydroascorbicacid(dhasc),tetrahydrobiopterin(tdh)and4a-hydroxytetrahydrobiopterin(dhb) homolog of monoamine oxidase. The monoamine oxidase updatedoverallstoichiometrywithATPasacofactorforstep13 typically deaminates p-tyramine, with trace deamination activity becomes: reported for m-tyramine41,42. Similar with pathway 5A and 5B, the conversion of m-tyramine to phenylephrine in the last two ATPþcatecholþmsalþ2O þ2H OþNH (cid:2)(cid:2)66!kcalphep 2 2 3 steps proceeds further by the action of an N-methyltransferase þacetaldehydeþ2CO þADPþpi: (R4) and oxidoreductase (R5). The de novo steps suggested by 2 novoStoic provide the reaction templates for the development of protein engineered dopamine hydroxylase (R5, steps 16, 18) and phenylethanolaminemethyltransferase(R4,steps15,19)thatcan Thetotalnumberofstepsofthedesignedpathwaysisgenerally act upon new substrates. larger than the synthetic chemistry-based approaches43–45. The co-substrate S-adenosyl-L-methionine (adm) required by However, if the methyl donor adm can be provided by the host the N-methylation reaction is generated in the O-methylation cell (e.g., engineered Saccharomyces cerevisiae46 and Pichia reactionbytheactionofsalicylate1-O-methyltransferase(step5). pastoris47), then pathway 5B can be shortened into four steps Therefore, msal provides the carbons in the phenolic and the therebybecomingonparwiththesyntheticchemistrypathways. methylamino moieties, whereas catechol provides the carbons in Itisnoteworthythattheidentifiedpathwaysuseamoreappealing the ethyl moiety of phenylephrine through its degradation to setofsubstrates(e.g.,catecholandmethylsalicylate)withalower pyruvate (Fig. 5c). Steps 1–4 can be bypassed if pyruvate is costthanthesubstrate(i.e.,m-hydroxybenzaldehyde)commonly directly provided. Note that the reaction in step 13 is treated as used for chemical synthesis. In all the three pathways designed, reversible in accordance with the reaction designation in the the conversion of theintermediate m-tyramine tophenylephrine source database MetRxn without requiring ATP. However, involves reaction mechanisms that can be derived from natural literature evidence36 suggests that step 13 requires ATP. The enzymes thus suggesting potential candidates for protein engineering (Table 2). In the next case study, we focus on NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications 9 ARTICLE NATURECOMMUNICATIONS|DOI:10.1038/s41467-017-02362-x Table3 Reactiontemplatesforbenzo[a]pyrene Rule Reaction-rule template Enzyme/homolog name Step id id 1, 34, R7 NAPHTHALENE DIOXYGENASE 38, 42 Anthracene Anthracene 1,2-dihydrodiol NAPHTHALENE DIHYDRODIOL 2, 35, R8 DEHYDROGENASE 39, 43 Pyrene 1,2-dihydrodiol Pyrene 1,2-diol 3, 36, R9 PROTOCATECHUATE OXYGENASE 40, 44 Protocatechuate 3-Carboxy-cis,cis-muconate 4, 22, 29, 37, 4-HYDROXYPHTHALATE CARBOXY- R10 41, 45, LYASE 4-Hydroxyphthalate 3-Hydroxybenzoate 46, 47, 48, 49 5, 23, R11 TEREPHTHALATE 1,2-DIOXYGENASE Terephthalate (3R,4S)-3,4-Dihydroxyc 25,30 yclohexa-1,5-diene-1,4- dicarboxylic acid (3S,4R)-3,4-DIHYDROXYCYCLOHEXA- 6, 24, R12 1,5-DIENE-1,4-DICARBOXYLATE 26, 31 (3R,4S)-3,4-Dihydroxyc 3,4-Dihydroxybenzoate DEHYDROGENASE cyclohexa-1,5-diene-1, 4-dicarboxylic acid R18 BIPHENYL-2,3-DIOL DIOXYGENASE 27 Biphenyl-2,3-diol 2,6-Dioxo-6-phenylhexa-3-enoate 2,6-DIOXO-6-PHENYLHEXA-3-ENOATE R19 28 Biphenyl-2,3-diol Benzoate 2-Hydroxy-2,4- BENZOYLHYDROLASE pentadienoate NotethatreactionrulesR13toR17arelistedasstep17to21inFig.6 identifying biodegradation pathways using reaction rules from a properties, are unfortunately ubiquitous in the environment and particular organism or species (see Supplementary Methods, have both natural and anthropogenic origins48. Biodegradation novoStoic design criteria iii/iv). studies around industrial effluent treatment plants and hydro- carbondrillingsiteshaveimplicatedPAHsasthesolecarbonand Oxidativedegradationofbenzo[a]pyrenetocatechol.Polycyclic energy source for many soil dwelling organisms49. Metabolic aromatichydrocarbons(PAHs),withmutagenicandcarcinogenic assays on a number of terrestrial bacterial species indicate the 10 NATURECOMMUNICATIONS| (2018) 9:184 |DOI:10.1038/s41467-017-02362-x|www.nature.com/naturecommunications
Description: