Specificity Effects of Amino Acid Substitutions in Promiscuous Hydrolases: Context‐Dependence of Catalytic Residue Contributions to Local Fitness Landscapes in Nearby Sequence Space

Abstract Catalytic promiscuity can facilitate evolution of enzyme functions—a multifunctional catalyst may act as a springboard for efficient functional adaptation. We test the effect of single mutations on multiple activities in two groups of promiscuous AP superfamily members to probe this hypothesis. We quantify the effect of site‐saturating mutagenesis of an analogous, nucleophile‐flanking residue in two superfamily members: an arylsulfatase (AS) and a phosphonate monoester hydrolase (PMH). Statistical analysis suggests that no one physicochemical characteristic alone explains the mutational effects. Instead, these effects appear to be dominated by their structural context. Likewise, the effect of changing the catalytic nucleophile itself is not reaction‐type‐specific. Mapping of “fitness landscapes” of four activities onto the possible variation of a chosen sequence position revealed tremendous potential for respecialization of AP superfamily members through single‐point mutations, highlighting catalytic promiscuity as a powerful predictor of adaptive potential.


Introduction
In apparent conflict with traditional views on enzymatic catalysis, promiscuous enzymesa re not exclusively specific for single substrates, but turn over multiple substrates, even if those compounds differ substantially in their molecular recognition properties. [1] The secondary,p romiscuouss ubstrates of such catalysts can differ either only in peripherals pectatorg roups that do not directly participate in the reaction (substrate promiscuity), or in groups that participate in the reaction, sometimes even through as ubstantially different mechanism (catalytic promiscuity). [1][2][3][4][5][6][7][8][9][10] Members of the alkaline phosphatase (AP) superfamily exhibit widespread catalytic promiscuity,c atalyzing multiple, chemically distinct phosphate-and sulfate-transfer reactions. [2,[11][12][13][14][15][16][17] Crosswise promiscuity-the promiscuous activities of one enzyme are the primary activitieso fa nother-is commonplace between members of this superfamily. [2,17,18] The chemicalv ersatility of these catalysts raises fundamental questions about the molecularr ecognition mechanismst hat these enzymes exploit in order to bind and turn over substrates that differ widelyi nc harge, size, reactivity,a nd the identities of the bonds that are made and broken. [1,19] Catalytic promiscuity has been taken as an indication of a functional relationship between evolutionarily relatedenzymes, reflecting either evolutionary history or future potential. [8,10,20,21] Ap rotein-coding gene that is under selectivep ressure for a particulara ctivity can serve, after gene duplication, as starting point for the evolution of agene with anew function. The likelihood that one of these duplicated genes acquires al evel of an ew activity high enough to provide an immediate selective advantage is low.H owever,t he presence of ap romiscuousa ctivity gives a" head start" to adaptation, because there are fewer steps in sequence space to travel until an ew activity reaches as elective threshold. [10] Any functionally beneficialm utation can carry as tability cost, [11,22,23] thus dictating that the number of mutations must be minimized during as uccessful functional switch. Ideally,t he level of promiscuousa ctivity in the originalp rotein should be high enough to provide as elective advantagei mmediatelya fter gene duplication:t hat is, the enzyme must be able to accommodate considerable levels of catalysis for both substrates. Promiscuity of catalysts is thus implicit in the evolutionary model of Ohno, [24] but also in the alternative scenarios of innovation/adaptation/divergence( IAD) and plasticity/relaxation/mutation (PRM). [25][26][27] Comparatives tudies of functional divergence within promiscuous enzyme families have been used successfully to provide Catalytic promiscuity can facilitate evolution of enzyme functions-a multifunctional catalyst may act as as pringboard for efficient functional adaptation. We test the effect of single mutations on multiple activitiesi nt wo groups of promiscuous AP superfamily members to probe this hypothesis.W eq uantify the effect of site-saturating mutagenesis of an analogous, nucleophile-flanking residue in two superfamily members:a n arylsulfatase (AS) and ap hosphonatem onoesterh ydrolase (PMH).S tatistical analysis suggests that no one physicochemi-cal characteristic alone explains the mutationale ffects. Instead, these effects appear to be dominated by their structural context. Likewise, the effect of changing the catalytic nucleophile itself is not reaction-type-specific. Mapping of "fitness landscapes"o ff our activities onto the possible variation of a chosen sequence positionr evealed tremendousp otential for respecializationo fA Ps uperfamily members through singlepoint mutations, highlighting catalytic promiscuity as ap owerful predictor of adaptive potential.
insight into the evolution of function in variousp rotein families. [28][29][30][31][32] Within the AP superfamily,t wo structurally and phylogenetically related, but functionally divergentg roups of arylsulfatases( ASs) andp hosphonate monoester hydrolases (PMHs, Figure 1A), [11,14,17] provide an opportunity to explore pathways for their functional interconversion. The amino acid sequence conservation amongst the branches of PMHs and the "new" ASs [17] is moderate (31-34 %), butt heir structures align well (r.m.s.d. 1.62-1.85 ) [17] and appear highly conserved. Despite the low overall sequence homology,m ost active-site residues suggested [17] to assist in substrate binding, nucleophilic attack, or leaving group stabilization are conserved in both ASs andP MHs ( Figure 1A). Their catalytic mechanisms ( Figure 1B)i nvolven ucleophilic catalysis (by af ormylglycine (fGly) residue), presumably assisted by general base catalysis, to pass through at ransition state (TS) in whichn egative charge development on the leaving group and phosphoryl oxygen atoms is stabilized by cationic groups (that is, Lewis acid catalysis) or general acid catalysis. [33] The resulting hemiacetal intermediate needs to be cleaved to regenerate the nucleophile.I nA Ss, a histidine( His A in Figure 1B)r esidue adjacent to the fGly nucleophile has been postulated to assist in the breakdown of the intermediate through general base catalysis. [34,35] In contrast, the threonine residue present at the "His A "-position in Rhizobium leguminosarum PMH (RlPMH) is unlikely to act as ag eneral base and consequently its mutationi nto an alanine residue had no detrimental effect on the enzyme-catalyzedc onversion of phosphonate monoester 3c. [11] The substrates hydrolyzed by members of the AP superfamily are diverse in their molecular recognitionp roperties:m embers of this superfamily have been shown to convert substratesw ith one, two, or no negative charges, and the substratesh ave half-lives from severalm onths to millions of years, [2,6,[11][12][13][14][15][16][17][18] thus implying very differentc atalytic requirements.T hisc hemical diversity is further supported by the observation that the hydrolytic reactions of phosphate diesters and phosphonate monoesters proceed via concerted TSs, [36,37] whereas those of phosphate and sulfate monoesters are characterizedb ymore expanded, dissociative TSs. [37][38][39][40] Here we report the effects of active-site mutations on catalytic rates for multiple substrates for severalo ft he related ASs and PMHs mentioned above.T he resulting systematic analysis of the effects of residue substitutions in homologous positions on catalysis of the diversec hemical reactions in the AP superfamily probes the feasibility of establishing new activities duringa daptation, and illustrates how structuralc onstraints of Figure 1. Active sites in the PMH and AS branches of the AP superfamily and the reactionsc atalyzed.A)Bottom:s chematic representation of evolutionary relationships between PMHsand the new ASs that together form ag roupo fenzymes related to arylsulfatases within the AP superfamily. [17] To p: the conserved active sites of PMHsand ASs,represented by RlPMH (PDB ID:2 VQR) and SpAS1 (PDB ID:4 UPI). Both enzyme classes contain the samef Gly nucleophile, arising from co-translationalmodificationofaconserved cysteiner esidue). PMHsa nd the new ASs differ in two active-sitep ositions: K101 and H103o fSpAS1a nd Y105 and T107 in RlPMH (red labels in both structures). The residue structurally analogous to H103/T107i sac onserved histidine [90] (His A according to the nomenclatureo fH ansonetal. [35] )ina ll known AP-superfamily arylsulfatases. [35] In the majority of PMHs, the corresponding residue is an aspartatem oiety,but ag roup of PMHs with at hreonine residuea tt his position,s uch as RlPMH, are the most proficient PMHs currently known. [17] B) Mechanismofn ew ASs. [11,17] Residue identifiersfollow the revised Hanson nomenclature. [17,35] "E" is an electrophile (either PorS ), and "X" is variously oxygen (sulfate monoester and phosphatem onoester), an alkylo ra ryl group (phosphonate monoesters), or an alkoxyo rp henoxy group( phosphate diesters). LG denotesthe leaving group( here 4-nitrophenyl). Most residues are analogousi nA Ss and PMHs. For example, Asp B is ag lutamine residue and Asn A ahistidine residueb oth in PMHsa nd in "new"A Ss such as SpAS1.A sn B is not universally conserved among ASs but is 100 %c onserved in PMHsa nd new ASs. [17] Lys A is at yrosine residue in PMHs, so its possible charge-charge interactions with the substratea re not available. No unambiguous function could be assigned to the aspartate residue found in D-PMHsa nd the threonine residue foundi nT -PMHs (such as RlPMH)att he His A position.However,T107 in RlPMH appears to coordinate the fGly nucleophile through ahydrogen bond. [11] C) Structureso fp hospho-(compounds 1-3)and sulfoester (compounds 4)s ubstrates used in this study.The backgroundc olorcoding is used to identify the reactionso ft he various substrates in subsequent figures.
ChemBioChem 2017, 18,1001 -1015 www.chembiochem.org 2017 The Authors. Publishedb yWiley-VCH Verlag GmbH &Co. KGaA, Weinheim the protein framework shape the effect of mutations on primary and promiscuous activities. Previous studies of protein mutagenesis and evolution have highlighted the importance of active-site and adjacent residues for enzyme function. [41][42][43] Our approacht argets residues that form part of the active site or are in close contact with active-site residues. Kinetic data for mutantswere measured, andthe data were subjected to statistical analyses to study relationships between mutations and the observed rates as af unction of chemical properties of substrates or reactiont ypes (e.g.,s ubstrate charge, nature of TSs).

Results and Discussion
Mapping local fitness landscapes:Site-saturation mutagenesis of structurally analogous active-site positionsi nPMH and AS families We investigated the accessibility of improvements of promiscuous activities by single amino acid replacement mutagenesis of an analogous position( His A in Figure 1B)i no ne representative each of the AS (SpAS1) and PMH (RlPMH) families. This residue is located within hydrogen-bonding distance of the fGly nucleophile, [11,14,17,34] and the nature of this residue can be used to differentiate between ASs (His) and PMHs (Asp/Thr) (that is, H103 in SpAS1 and T107 in RlPMH).
Af ulls et of wild-type and all 19 standard proteinogenic amino acid substitutions was generated, resulting in two sets of variants for SpAS1 (H103X) and RlPMH (T107X), expressed in Escherichia coli and isolated from crude lysate by Strep-tag based affinity purification (for detailss ee the Experimental Sec-tion and Figure S1 in the Supporting Information). Each variant was tested for activity towards each of the four substrate classes hydrolyzed by ASs and PMHs (phosphate monoester 1, phosphate diester 2b,p hosphonate monoester 3a,a nd sulfate monoester 4,F igure1C). Activity tests were performed at substrate concentrationsa pproachingf irst-order conditions with regardt os ubstrate (i.e.,a ts ubstrate concentrations at least two to three times lower than the corresponding wild-type Michaelis constant, K M ). Under these conditions, differences in observed reactionr ate are representativeo fc hanges in catalytic efficiency (k cat /K M ;E xperimental Section, Figure S2, Ta ble S2). These enzyme activity measurements were corrected for variations in protein concentration (relative to the wild type). As ac onsequence, they can be directly interpreted as the ratio of second-order rate constants, k mutant 2 /k WT 2 .T his ratio provides ac omparison of the first irreversibles teps of the reactions equence, in this case the formation of the fGly intermediate.
In SpAS1, mutation of H103 into any other residue resulted in am ore than tenfold( fort en out of 19 substitutions, > 100fold) decreasei nt he primary activity (hydrolysis of sulfate monoester 4,F igure 2A,T able S3), thus confirming that ah istidine residue at this position is the optimal residuef or the native activity of ASs. At the same time, activity toward phosphate mono-and diestersi ncreased for all mutants, with more than tenfold improvement observed for ten out of 19 mutants ( Figure 2B,T able S3). Some of the mutants even show a % 100fold increase in those activities (observed for mutants H103S, -T,a nd -K). Changes in enzyme-catalyzed phosphonate monoester hydrolysis ranged from a3 0-foldi ncrease to at enfold decrease in catalytic efficiency,a nd mutants with strongly im- Figure 2. The effect of mutation of A) H103inS pAS1, and B) T107 in RlPMH on the enzyme-catalyzed hydrolysis of phosphate monoester 1 (purple), phosphate diester 2b (orange), phosphonate monoester 3a (red), and sulfate monoester 4 (blue). Initial rates (V 0 )were measured at substrateconcentrations at least two timesl owert han the wild-type K M ,atwhich point changes in the observedsecond-order rate constant (k 2 = v 0 /[enzyme]; i.e.,initial rates normalized for varyingprotein concentrations) translate directly into changes in k cat /K M ( Figure S2 andT able S2). Changes in enzyme activities are indicated relativet o wild type (log(k mutant 2 /k WT 2 )). The box labelled "n.d."indicates that no significanta ctivity above background was observed (on the basis of the detection limit, activity would be at least % 80 times (phosphate monoester) or % 60 times (phosphate diester) lower than wild type).MutantH 103V could not be expressed (n.e.). All ratesw eredetermined in triplicate at 30 8Ci n100 mm Tris·HCl, 500 mm NaCl,a nd 100 mm MnCl 2 at pH 7.5. Error bars represent the standard deviations of measurements. See Ta bles S3 (SpAS1 H103X)a nd S4 (RlPMH T107X)f or details. provedp hosphomono-and -diesterase activities showed decreasedp hosphonate monoesterase activity (e.g.,H 103K, % 100-fold increase vs. % tenfold decrease, respectively).
The differential effectso ft he various mutations, resulting in up to 10 5 -fold changes in specificity between substrate pairs, confirm the proposed role of the nucleophile-flanking residue H103 as as pecificity determinant between ASs and PMHs. Five out of the 19 possible mutants no longerh ave sulfatem onoester hydrolysis as their "best" activity ( Figure 3A-C). This effect is particularly prominent for mutation H103K, which switchest he enzyme from a % 10 4 -fold preference toward sulfate monoester 4 over phosphate monoester 1 to at enfold preference in the reverse direction, accompaniedb ya100-fold improvement towards phosphate monoester 1.
The resultsw ith the library SpAS1 H103X are reminiscent of those obtained in the study of as ite-saturation library of E192 in N-acetylneuraminic acid lyase, [44] for which all substitutions lead to areduction in its native retro-aldol cleavage activity,accompanied by an increase in activity toward one of its promiscuous substrates for 16 of 19 possible mutations. Ther esults for both enzymes reinforce the notiont hat promiscuous reactions, whichd on ot use the active-site functionalities in an optimizedw ay,m ight be more robust towards (or even more readily improved by) as mall number of mutations than the native activity.I nb oth cases, the new activityw as improved and the original one had decreased, thus causing al arge specificity switch, as was also observed in several other studies in which specificity changes of promiscuous enzymes were causedb yc ombinationso fb eneficial effects on one activity and detrimental effectso na nother. [7,21,45,46] The drastic changes in activity and specificity observed for the H103X mutants further add to as ubstantial body of evidence that shows that the chemicalf unctiona nd specificityo fa ne nzyme can be changed with only af ew amino acid substitutions. [41,43,47]  toward the primary substrate versus those of each of the promiscuous substrates.T he mutationsare grouped into the following categories:1)improved specialists (blue quadrant, increased specificity for the cognatea ctivity),2 )enzymesi mproved onlyi ntheir promiscuous activity (yellow), 3) generalimprovers (green,a ll activities increased), and 4) general decreasers(gray,m utationa ffects botha ctivities detrimentally). The diagonal represents the locations of hypothetical generalists, which catalyzeb othreactions at equal rates.N one of the mutations resultedi nani ncreasei nt he primary activity (in which case they would have been observed in the greeno rb lue quadrants). Catalytic efficiencies (k cat /K M )for the variousm utants were calculated from wild-typel evels (as listed in Ta bles S9 (SpAS1) and S13 (RlPMH)) by using the experimentally determined k mutant Unlike in SpAS1, substitutions in the analogousp osition T107 in RlPMH predominantly result in decreased catalytic activities toward all four substrate classes, with the exception of phosphomonoesterase activity,f or whiche ight out of 19 mutants show improvements (largesti ncrease: % tenfold in RlPMH T107R, Figure 2B and Ta ble S4). However,e ven in the absence of substantial improvements, RlPMH is robust to mutagenesis at T107:1 0o ut of 19 substitutionsa re "neutral" (< 10-fold decrease in catalytic efficiency) with respect to the two primary activities( in two of 19 cases for phosphodiesterase;i ne ight of 19 for phosphonate monoesterase). The various mutationsi n T107 in RlPMH result in up to 700-fold changes in substrate preference (e.g.,f or T107K, phosphomonoesterase vs. phosphodiesterase activities;F igures 2a nd S10). However,t he general specificity of, for example, the activity towards phosphonate monoester 3a over the promiscuous activity towards sulfate 4 is only reduced by less than tenfold by any of the mutations to T107;t his stands out in contrast with the much larger effects found for SpAS1 H103X. At least for as ingle-residue mutations tep in position 103/107, the observed readily accessible specificity change appears to be unidirectional( e.g.,f or SpAS1 H103T). In contrast, the analogous "reverse" mutation T107H would not be an accessible first step in evolutions tarting from RlPMH.

Apparent redundancyo fthe active sites opens up multiple pathways for futurea daptivee volution
Our resultss how that mutations affect promiscuous activities differentially,r esulting in significant changes in specificities. In some cases they can even change which substrate is preferred ( Figure 3). In SpAS1 this switch occurs as ar esult of the introductiono fo ne of five amino acids with very different molecular recognitionf eatures (P,W ,N ,K ,a nd R). In particularf or phosphate mono-a nd diester hydrolysis, mutation of H103 in SpAS1 is beneficial in all cases, irrespective of which residue it is replaced with. Therea re no obvious amino acid properties that correlate with the degree of improvement( Figures S3 and S5): for example, both removal of the histidine side chain (H103G,H 103A) or replacement with al arge charged residue (H103K and H103R) show similar degrees of improvement in phosphomono-and-diesterase activity (Figures 2A and 3A,B ). For phosphomonoesterase activityt he results for H103K and H103R can be explained in terms of improved interaction of the now cationic side chain with the negative chargei nt he more charged substrates (and also the TS of phosphoryl transfer). [48] However,t he beneficial effect of H103G andH 103A cannotb er eadily explained. The improvements for phosphomono-a nd -diesterase activity are positively correlated for SpAS1 H103X ( Figure S9,s ee below for details), which suggests that the differenceinnet chargebetween the primary andpromiscuous substrates is not the reason for the differential effects of the mutation on sulfatase and phosphomonoesterase activity.T hese observations imply that ah istidine residue in the conserved positioni sdisfavoring phosphoryl transfer in the SpAS1 active site, in which case its mutation into any other residue removes al imitation, rather than the chemical properties of the "new" side chain providing explicitly beneficial interactions for phosphorylt ransfer.
The site-saturation mutagenesis data obtained in this study were used to build an etwork of mutationalt ransitions that corresponds to an empirical fitness landscape. [49,50,51,52,89] Local fitness landscapes were constructed for mutations in the nucleophile flankingr esiduei nSpAS1 ( Figure 4) and RlPMH (Figure S8). We refer to these landscapes as "local" because we map only the immediate vicinity of the wild type, resulting from as ingle-residue mutation.T his approach is similart o  Table S3). The node diameter was scaled according to log [k mutant 2 /k WT 2 ] + 1, so that nodesb igger than His (WT) denote morea ctive enzymes,w hereas nodes smallert han His correspond to less-active mutants. Edges (or connections) are only allowed between aminoa cids that can be interconverted withoutp assing through ac odon codingfor at hird amino acid (i.e.,they can be converted either by as ingle nonsynonymousn ucleotide substitution, or by one or two synonymous mutations followedb y an onsynonymous one). The wild-type residueo fSpAS1 (His103) and all amino acids directly accessiblefrom any of the histidinecodons are highlighted in bold black.S ubsequents teps that would lead to threonine,the amino acid of RlPMH WT at position 107 (the "target"i nahypothetical evolutionary "trajectory") are highlighted in bold gray. Thus, the black/gray subnetwork showshypothetical trajectories of interconversionb etweenh istidine and threonine at position 103 of SpAS1. previouss tudies on constraints for in vitro protein evolution. [49,[52][53][54][55] For SpAS1 H103X, for example, the fitnessl andscapes for the primary sulfatase and the promiscuousp hosphodiesterase activities show as teep decline and as teep incline, respectively,i nm ultiple directions (Figure 4). For the primary sulfatase activity,a ll amino acids that are directly accessible from any codon coding for the wild-type residue (His) result in am ore than tenfold reduction in activity.I nf act, all amino acids other than the wild-type residue (His) exhibit lower sulfatase activity.I nevolutionary terms, SpAS1 WT is at the globalf itness peak for its primary activity withr espect to position103. For its promiscuous activities,t he situation is quite the opposite:f or example, for its promiscuousp hosphodiesterase activity, SpAS1 WT is almosta tt he global fitness minimum for position103 ( Figure 4A), with the exceptiono f the mutants H103E( detrimental to all four activities) and H103V (no expression).M ore importantly,H is103 is located directly in af itness "well": that is, any of the direct neighbors accessible from histidine-encoding codons through as inglenucleotide substitution are better phosphodiesterases than the wild type. This suggests that once SpAS1 is under selective pressure for improved phosphodiesterase activity, any non-synonymous mutation in the codon for His103i sb eneficial and can remaini nt he gene pool, thus providing multiple starting points for adaptive evolution. From three of the mutations accessible by single-nucleotide substitution from SpAS1 H103 (wild type)-arginine, proline, and asparagine-the codon for threonine is accessible through just one additional nucleotide change. This suggests that, in SpAS1, the transition from the sulfataser esidue His A to the signature residue of proficient phosphodiesterases/phosphonate monoester hydrolases, threonine, is accessible by ap athway via two subsequent nucleotide exchanges. This meanst hat an adaptivep athway from sulfataset op hosphatase that does not requirem utationse lsewhere in the protein exists.
The strongly detrimental effect of any mutationo nt he primary sulfatase activity rules out as cenarioi nw hich as trong increaseinapromiscuousactivity is accompanied by arelatively small change in the original function (i.e.,m aintained at > 10 %o ft he wild-type level),w hich hasb een reported for the "generalist intermediate scenario" characterizedb yl ow initial negative trade-off. [7,21,56] This observation is also more consistent with Ohno's rather than the IAD or PRM models. [24][25][26][27] The unavailability of this scenario suggests that mutationsa tt his positionw ould be unlikely to occur in vivo if the selection pressure for the originalf unction were maintained:t hat is, duplication of the SpAS1-encoding gene would have to precede these mutations.O nce the originals election pressure is relieved, there are multiple mutationalp aths that result in tento 100-fold improvements in, for example, phosphomonoesterase ( Figure 2A This variety of possible amino acids ubstitutions opens up diversep athways for further adaptive evolution. For example, the mutations H103G, H103T, and H103K-to ah ydrophobic, ah ydrophilic, and ap ositively charged residue, respectivelyall result in > 30-fold improvements of both phosphomonoesterase and phosphodiesterasea ctivities, but each of theses ubstitutions would be expected to affect the electrostatic environment of the active site differently.E ven the substitutions directly accessible from histidine-encodingc odons (H103Y, H103N,H 103D,a nd H103R) are vastly different with respect to the expected effect their introduction will have on the physicochemicalp roperties of the active site. The factt hat multiple pathways result in improvement of ap romiscuousa ctivity, such as described in this study and elsewhere, [44,53,54] provides evidencet hat ag roup of mutants at this positionc an act as "molecularq uasi-species": that is,agroup of variants of ap rotein sequence that are close in functional space (for the new activity), but have started to diverge in sequence space. [57] Such populations of variants have been considered to be the real subject of selection (rather than single-enzyme variants) and might, due to their higher diversity,i ncrease the success of further evolution because they would reducet he likelihood of the occurrence of evolutionary dead-ends. [57,58] The steep inclines observed in the fitnessl andscape for enzyme-catalyzed phosphate monoestera nd phosphated iester hydrolysis ( Figure 4A)c ontrast with previous findings with other enzymes,i nw hich combinations of mutations were required for substantial improvements in, for example, enantioselectivity. [59] This was attributed to av ery flat local fitness landscape( Figure 5B), in whichs ingle mutations did not result in any significant improvement of the desired functionality. When multiple mutations are required, the chance that all of them arise is low." Epistatic ratchets" can lower this likelihood even further: [60] that is, the mutations can only be acquired in as pecific order,l imiting the number of possible evolutionary paths. [50,52,61,62] Because about one in three mutationsi nap rotein is expected to be detrimental to function and/ors tability, [9,22,23] it is vital that beneficial levelso fn ew activities are accessible througha sf ew mutations as possible. Enzymep romiscuity has been suggested as ap ossible mechanism of such quick adaptation, [7,10,20] because the presenceo fapromiscuous activity close to the selectivet hreshold probably decreases the number of mutationsn eeded to provide as electiveadvantage. Indeed, for SpAS1, the observation of ap romiscuousa ctivity in the wild type would correctly predict rapid improvement in, for example, phosphomonoesterase activity through few mutations (or in fact, as observed, as ingle-point mutation).

Dependence of functionaltrade-offs on physicochemical properties of substrates in different active-site contexts
Conventional molecular recognition analysis-based on, for example,s ize or hydrophobicity of the various amino acid side chains( as in Figures S3-S6 [63] in order to detect and to visualize trends and correlationst hat are beyond the obvious,a sh as been done previously for analysis of the distributionso fp rotein familieso rm utant proteins in promiscuous "activity spaces". [57,[64][65][66][67] PCA was performedo nt he set of mutants of each enzyme individually,a nd the coordinate system was calculated for AS and PMHi ndependently.B ecause the coordinate system for each enzyme is optimized on the basis of the redundancy of the originald ataset, the coordinate systemsa re not interchangeable between SpAS1 H103X ( Figure 6A)a nd RlPMH T107X ( Figure 6B). However,t he relative positions of residue coordinates to each other,a nd their clustering, can be compared.
The distribution patternso fa mino acids ubstitutions differ considerably between the two enzymes. Physicochemically or sterically very different residues are observed close together in activity space:f or example, the clustero fC /D/F/Y/Q in SpAS1 ( Figure 6A)a nd D/M or E/H in RlPMH (Figure 6B). At the same time, residues with similarc atalytic groups or steric demands can be quite far apart:f or example, Q/N or H/K/R in SpAS1 and E/D or A/V in RlPMH. In some cases, such as S/T,t he proximity of mutations is conservedi nb oth enzymes,b ut it is evident that the structural differences between the protein structures and the direct environments of the active sites must lead to quite different effects of identical amino acid substitutions.
The patterns of possible correlations between the effects of the various mutations on the four activities in the two enzymes differ considerably ( Figure 6). For SpAS1, the projected vectors for phosphate monoester 1 and phosphodiester 2b ( Figure 6A)a re similari nb oth direction and length, thus indicating ap ositive correlation betweent he mutational effects in H103 on these two activities. This correlation was shown to be significant by direct comparison of thesee ffects (r = 0.97, p < 10 À4 ,F igures 6C and S9). The correlationb etween phosphonate monoesterase activity (substrate 3a)a nd the other two phosphohydrolase activities is weaker( r = 0.47 and 0.52), but still significant (p = 0.050 and 0.028). The effects of the mutations on sulfate monoesterase activity were weakly,b ut significantly,c orrelated with those for phosphonate monoester 3a hydrolysis (r = 0.57, p = 0.0013). They showedn oc orrelation with the effects on the other two phosphohydrolasea ctivities. The physicochemical property that at least partly predicts these correlations is the natureo ft he electrophilicr eaction center: data for all reactions involving phosphorus centersa re correlated. In contrast, the TS of the solution reactiona nd the ground state (GS) chargeo ft he substrate are poor predictors for correlation-the two activities that show the strongest correlation, phosphomonoesterase and phosphodiesterase, differ in both of these aspects (dissociative vs. associative and À2v s. À1, respectively). [37,39,40] The analysis for the effects of mutationsi nT 107 in RlPMH suggestsacorrelation that is governedb yt he nature of the TS in solution, because the projected vectors both of the phosphodiesterase/phosphonate monoesterase (concerted) and of the phosphatase/sulfatase (dissociative) pairs appear in the same quadrant of the PCA ( Figure 6B). Direct comparison of both substrate pairs confirmed this observation ( Figures 6D  and S10). The same direct comparisons between all possible substrate pairs also indicated that substrate chargea tl east partly predicts correlation,b ecause all reactions with as ubstrate chargeo fÀ1( at the experimentalp Ho f7 .5) are correlated, andt he correlation between the two dissociative reactions, which differ in GS charge, is weaker than that between the two concertedr eactions that have the same substrate charge(r = 0.48; p = 0.033 vs. r = 0.75; p = 2 10 À4 ).
Comparison of the possible physicochemical predictors for the effect of mutations in the nucleophile flanking residuef or both enzymess tudied here shows that their predictive values The sequence space is reducedt oasingle dimension (Hamming distance for the expressed amino acid sequence) for purposes of illustration. A) As cenario for rapid adaptation.T he localfitness landscape for the promiscuous phosphatase activity of SpAS1 is apparently very steep, from our results of mutating residue H103:i mprovements of up to 100-fold are already accessiblet hroughv arious single mutations, such as H103R (Figures 2and 4). This means that the promiscuous function can be improved quickly to the levelofa" generalist" that can hardly differentiate between primarya nd promiscuous activity.B)Ascenario for adaptation requiring multiple mutations. The localfitness landscape for an ew activityi sv ery flat, with the consequence that three amino acid substitutions are required to reach significant improvements, suchast he one reported by Sandstrçm et al., [59] and four mutationsare requiredt oa chieveani mprovementt hat surpassest hat illustratedf or the case of as teep gradient in afitness landscape, as displayed in (A). are mostl ikely strongly context dependent.F urthermore, even within one enzyme, none of the three properties described thus far-GSc harge, TS of the solution reaction, or reaction center-is strongly dominating,t huss uggesting that am ore complex interplay of various factors governs the trade-off between different activities.
Inclusiono fs pecificity data for all variants of both enzymes for ab ulky substrate (Figures 7, S11, and S12) suggests that the three properties discussed above are all far lessi mportant than the size of the substrate in question.A ctivity toward phosphonate monoester 3c shows no correlation with that toward the less bulky,b ut in every other way similar,p hosphonate 3a in the site-saturation libraries of both enzymes (Figures S11C and S12 C). The fact that activities toward the similarly sized sulfate monoester 4 and phosphonate 3a are correlated in both enzymes ( Figures S9 and S10), despitedifferences in reactionc enter and TS in solution, is also consistent with this observation. The only substrate for which phosphonate 3c shows significant correlation is phosphate diester 2b (only in RlPMH, r = 0.64; p = 0.0025, Figure S12 B), the bulkiest substrate out of the four possibilities. This strong apparent dominance of size over other substrate/reaction properties cannot readily be explained in terms of steric clashes, atl east not those with the residue at position103:t here is no significant correlation between the volume of the residue at position 103 in SpAS1 and the activity towards any of the tested substrates (Figures 7A and S5).
The lack of correlationb etween the activities towards the two phosphonates is most likely the result of beneficial hydrophobic interactions that are only present for the substrate with the large hydrophobic substituent (Figure S5 D). For position 107 in RlPMH, however,t he lack of correlation is due to the steric exclusion of the bulky phosphate 3c,w hich is absent in the case of phosphonate 3a.T his is evident from comparison of the effect of residue volumeo nb oth activities across mutantsa mong RlPMH T107X:t here is as ignificant inverse correlation for residue volume and activity towards phosphonate 3c,b ut none forp hosphonate 3a ( Figures 7B and S6 C,  D). Furthermore, the effects on 3a and 3c are not correlated at all in the two enzymes:t hat is, 3a and 4 (chemically different, same size) experience ag reater mutational effect than 3a and 3c (chemically identical,d ifferent in size;s ee Figures S9 and S10). These observations suggestt hat any difference in solution TS, reactionc enter,o rG Sc harge of the various sub-

Mutation of the active-sitenucleophile
The effect of multiple amino acid substitutions in the nucleophile-flanking position 103/107 in an AS and aP MH on their catalytic specificity showedn oc onsistenttrends with regardt o substrate or TS properties that might explain the effect of these mutations. We further tested whethero rn ot the effect of an inactivating mutation to the nucleophile itself might be explained in terms of properties of substrate GS or TS. The active-site functionality of five ASs and four PMHs [17] was compromised by changing the active-site nucleophile from af Gly residue (encoded in the form of ac ysteine residue embedded in ar ecognition sequence) into as erine residue. Unlikec ysteine, serine cannot be post-translationally modified to fGly in E. coli. [68][69][70] The mutationt os erine wasp referred over the more strongly deleterious mutation to alanine, in order to allow quantitative determination of the effect of the nucleophile mutation on specificity,b ecause the serine variantse xhibit significant residual activity that is still measurable for all promiscuous activities.
All nine mutant enzymes were overexpressed in E. coli and purifiedf rom cell lysate( essentially as described by van Loo et al. [17] ), and kinetic parameters were determined for substrates 1-4.F or individuals ubstrate/enzymec ombinations, the effect of the mutation varied from as lighti ncrease to a % 10 3 -10 4 -fold drop in catalytic efficiency relative to the wild type ( Figure 8A,T ables S6-S14). The magnitudes of the mutational effectsv ary considerably:e ven the activities towards imilar substrates (i.e.,phosphodiesters 2a-c and phosphonate monoesters 3a-c)a re highlyv ariable and do not show anyo bvious correlation with the size of the unreactive substituent (Figure 8A). However, despite the large differences between "outliers", the overall effect of mutating the nucleophile is well correlateda cross all enzyme/substrate combinations.T he linear fit of al ogarithmic correlationp lot between catalytic efficiencies (k cat /K M )f or mutant versus wild typefor all eight substrates had as lope near unity (r = 0.90, p < 10 À4 ,F igure 8B). A y-axis intercept of À1i ndicates that replacing the fGly nucleophile with as erine residue results on average in an approximately tenfold decrease in catalytic efficiency,i ndependent of the wild-type level of activity (despite k cat /K M valuesf or the wild-type enzymesr anging from 10 À2 to 10 7 s À1 m À1 ). This correlation was virtually identicali nA Ss and PMHs ( Figures 8C and S13 A, B), thus suggesting that the effect of mutation of the nucleophile on any reactionc atalyzed by one of these enzymes is not correlated with the apparent primary function.
As mentioned above, the ASs and PMHs catalyze reactions that proceed through different TSs in solution. Phosphate and sulfate monoester hydrolysis proceedst hrough loose or dissociative TSs, [38][39][40] whereas phosphodiesters and phosphonate monoesters are hydrolyzed via concerted TSs in solution. [36,37] The contribution of the nucleophile to catalysis is expected to be more important forc oncerted reactions, in which case changing the nucleophile would affect their catalysis more strongly than for dissociative reactions. However,t here was no significant differencei nt he decreasesi nc atalytic efficiency as ar esult of changing the nucleophile for reactions that follow either more dissociative or more concerted reaction pathways ( Figures 8C and S13 C, D). This observation suggestst hat, for the given set of hydrolytic reactions, the nature of the TS of the catalyzed reactioni ns olution does not determine the effect of mutation on the enzymatic activity.T his might be because TSs in the enzymatic reactions are more similart han in solution (as proposed, for example, by McWhirter et al. [71] ). However,ac onsiderable body of evidence suggests that such changes in the extent of transition state charge distribution and bond order are typicallyr elativelys mall. [36,37,72,73] Other aspects of catalysis-substratep ositioning for nucleophilic attack, for example-might be more important, overriding effects that the nature of the TS might have.
Even thoughthe overall fit to the mutational effects is significant and reliable, there are considerable deviations from the ) on residue volume of A) SpAS1 H103X, and B) RlPMH T107X.V olumeso famino acid side chains were taken from Zamyatin et al. [88] (See Table S5 for details). Black:4 -nitrophenyl methylphosphonate (3a). Red:4-nitrophenyl phenylphosphonate (3c). Linear regression lines are showni nthe plots. A) SpAS1 H103.C orrelation coefficient R 2 for methylphosphonate 3a (black):0 .044. R 2 for phenylphosphonate 3c:0 .027.B )RlPMH T107. Correlation coefficient R 2 for methylphosphonate 3a (black): 0.044. R 2 for phenylphosphonate 3c: 0.481. Pleaser efer to Figures S3-S6 for amoredetailed analysisoft he interplay of substituted residuev olume/hydrophobicity and activity.Steric effects are furtheranalyzedinFigures S11and S12, with the focus on how the activity of the bulkiest substrate, phosphonate 3c,isc orrelated to smaller substrates. fitted curve for individuale nzymes/activities. If the nature of the substrate is the dominant determinanto ft he effect of mutation on ap articulara ctivity (and thereby responsible for its deviation from the fitted line), the patterns of effects for the nine differente nzymes should be largely the same-that is, the activities with the weakest and strongeste ffects should be the same for all enzymes.H owever,w hen the same data shown in Figure 8B are disaggregated in terms of enzymea nd substrate (shown in Figure 8A), the patterns of mutational effects (e.g.,t he reduction of the activity towards ag iven substrate in the Cys-to-Ser mutanti nr elationt ot he wild type) show large variation. Apparently,e ach enzyme responds differently to the same mutation and the effects of mutation on each activity are not the same, leading to unique patterns (Figure 8A).
Ta ken together, these data suggest that, despite the conserved identity of the core active-site residues (nucleophile, metal-binding residues,l eaving group activation;s ee van Loo et al. [17] )int he two enzymeclasses, their contexta nd surroundings have as ubstantial effect on how mutationsa ffect their catalytic power.

Context-dependence of mutations
This comprehensive study of mutationale ffects in members of related groups of promiscuous enzymes suggestst hat the main factor shapingt he local fitnessl andscape of active-site residues is the overall structuralc ontext imposed by the enzyme,r ather than the physicochemical properties of the substrate or the TSs of the catalyzed reactions.
The effect of mutating the nucleophile is similarly independent of enzyme class (AS or PMH), as indicated by the correlations between the corresponding catalytic efficiencieso fw ild type and the fGly-to-Ser mutants ( Figure 8C). Furthermore, the linear correlation with as lope near unity suggeststhat the mutationale ffect is independent of the wild-type activity level. These two observations strongly imply that the effect of the mutation is not relatedt ot he identity of the primary activity of an enzyme. In addition, we cannot detect as ignificant differencei nt he slopes of the correlation lines between reactions that are believed to proceed via concerted or more dissociative-likeT Ss in solution. This observation stands in contrastt o the expectation that mutation of the nucleophile shoulda ffect the rate of ar eactions ignificantly more if the reaction proceeds through ac oncerted TS, because there is more involvement of the nucleophile in the TS. Along with the unique enzyme-dependent differences between the patterns of mutational effects across the panel of substrates studied, this unexpected result strongly suggests that the protein environment has as tronger influence on the effect of mutation than the nature of the TS of the catalyzed reaction. These findings are furthers upported by the comparative site-saturation mutagenesis of the nucleophile-flanking residue in one member of each of the AS and PMH subgroups. Neither the charge nor the nature of the TS observed fort he solution reaction is as trong predictoro ft he effect of amino acid substitutions when comparing the effects of the various mutationsi nt he nucleophileflankingr esidue on the substrate specificities of SpAS1 and RlPMH. Figure 8. The effect of replacing the active-site nucleophile fGly with serine for PMHs (circles) and ASs (triangles)was measured by determining Michaelis-Menten parametersw ith purified enzymes. A) Heat maps representing the changes in catalytic efficiency (k cat /K M )a saresult of the Cys-to-Ser mutation in ASs and PMHs relative to the wild type for enzyme-catalyzed hydrolysisofp hospho-and sulfoesters 1-4.See Ta bles S6-S14 for detailed kineticd ata and experimentalc onditions. B) Correlation plot betweent he catalytic efficiencies of the wild-type (fGly) and mutant( Ser) enzyme variants toward substrates 1-4. AkAS: , RpAS:~, SaAS:~, SpAS1:~), SpAS2:~), AkPMH: *, BcPMH: *, RlPMH: *,a nd SpPMH: *.W eobserveal inearc orrelationb etween log (k cat /K M ) Ser for the serine mutants (y-axis) and log (k cat /K M ) WT for the wild-type enzyme with an fGly nucleophile (originatingfrom cysteine)( x-axis).T he blackdotted line represents alinear correlation for all data (slope = 0.96 AE 0.06;intercept = 1.0 AE 0.2; r = 0.90, see also (C)). The linearity of this curve indicates that the mutation resulted in an % tenfold decrease in catalytic efficiency (k cat /K M )that is largelyi ndependent of the level of wild-type activity,across k cat /K M valuesr anging from 10 À2 to 10 7 s À1 m À1 .C)Slopesa nd intercepts for linear correlationg raphsfor the mutational effects classified either according to enzyme type (ASs vs. PMHs) or transition state type of the solution reaction catalyzed(dissociative vs. concerted). The corresponding fits are showni nF igureS13. No clear distinction could be observed between the mutational effect in PMHsa nd ASs.S imilarly, no difference was observed betweens ubstratest hat follow adissociative or concerted transition stateins olution. All fits showed p < 10 À4 .S trong deviationsfrom the correlation were associatedw ith ap articulare nzyme rathert han ap articulars ubstrate class. These deviationsw ere non-systematic( e.g., AkAS:~showed large deviations from the trendinb oth directions: 10 3 -fold (phosphate 1)and 1.5-fold (phosphonate 3b)d ecrease vs. tenfold averagedecrease for the complete data set). As imilar lack of predictability of enzymatic activity based on the nature of the catalyzed reactionsw as described forg lutathione S-transferase (GST) by Kurtovic et al.,w ho observed that clusters of functionally similarG ST variants in am ultidimensional (promiscuous) substrate-activity space still contained enzymes that catalyzed very differentr eactions as individual promiscuousa ctivities, and that different reactiont ypes did not significantly contributet of unctional clustering of mutants. [66] In another study,Z hang et al. [67] observed correlations between mutational effects on mechanistically different reactions catalyzed by GST.T his becamea pparent from examination of factor loadings for activities when projected onto "functional space" mapped by PCA, in which,f or example, transacetylation and reduction reactions are similarly affected by mutagenesis. In that particular case, the chemical properties of catalyzed reactions did not serve as ar eliable predictoro f mutationaleffects in promiscuous enzymes.
What might explain the correlations that we and others report for activities in mutant populations of promiscuouse nzymes?A ctive sites, in particulart hose of promiscuouse nzymes, can accommodate substrates in multiple orientations with regardt ot he catalytic groups. [1,[74][75][76] Subsets of active-site residues might contribute to catalysis of multiple reactions to differente xtents. [41,[77][78][79] The bound substrates are exposed to highly localized, intrinsically nonspecific medium effects, leading to al acko facorrelation with reaction type. Steric effects that would constrain the orientation of substrates relative to active-site residues are another factor that can contribute to the "masking" of chemical properties of substrates-the data reportedi nF igure 7s how evidence that steric effects due to non-reacting groups in substrates can cause ac orrelation of activity with the volumeo fs ubstituted residues in one (RlPMH), but not in another relatede nzyme (SpAS1), suggesting differents teric constraints in the two enzymes. Amutagenesis study on nucleotide pyrophosphatase/phosphodiesterase (NPP) by Wiersma-Koch et al. [80] showedt hat bulky,h ydrophobic residues in the vicinity of the active site tune the specificity in wild-type NPP towardapreference for phosphate diesters over phosphate monoester substrates by four orders of magnitude. This observation was ascribed to favorable interactions with the non-reactingRgroup of the phosphate diester and to unfavorable interactions of the groupsw ith the negatively chargedn onbridging oxygen atom of the phosphate monoester.S imilar steric or hydrophobic interactions would be sufficient to explain the different trends observedf or the site-saturation libraries SpAS1 H103X and RlPMH T107X.

Conclusions
Ahighly plastic "gatekeeper"residue allows rapid specificity switchingfrom sulfatase to phosphotransferase We have identified His103 in the sulfatase SpAS1 as ah ighly plastic, specificity-determining position at which virtually every mutations ubstantially improves both activity and specificity of the enzyme toward phosphate or phosphonate esters (vs. its originals ulfate monoester substrate). Specificity changes of up to 10 5 -fold are readily obtained by this "minimal evolutionary step" [47] and suggest that considerable respecialization can be achieved through as ingle mutation. Up to 100-fold enhancementso fp romiscuous activities are obtained in individual mutants, and in nine out of 19 substitutions in whichH is103 of the arylsulfatase SpAS1 is replaced, promiscuousa ctivities are improved moret han tenfold. The profound effects of singleresidue substitutions demonstrate that promiscuous arylsulfatases are versatile evolutionary starting points, in which a single-point mutation hasa ni mmediate, large positive effect on the "new" activity,a voiding loss of function in evolutionary intermediates. These insights chart af orward evolutionary link acrossa na ctivity landscape for functional adaptation.
Consideration of fitness landscapes provides aframework for assessing whethera ctivity transitions are feasible Despite the remarkable plasticity of amino acid position 103, not all substitutions in this residue that improve promiscuous activities in SpAS1 are directly accessible through single-nucleotide substitutions. Fitness landscapes (such as those shown in Figures4 and S8) link function and connectivity of amino acids, illustrating which amino acids can be interconverted directly through single-nucleotide mutations between any of their codons. This allowsf or "mapping" of the local fitness landscape of position103, showingt hat the histidine residue in SpAS1 WT represents af itness "well" with respectt oi ts promiscuous activities:a ll amino acids accessible from histidine codonsg ive rise to improved promiscuousa ctivities. However, for the native sulfatase activity,H is103 represents af itness maximum. This asymmetry in the mutationale ffects makes improvement of promiscuous activitiesf or this positiona lmost automatic.A tt he same time, survival of the gene directly after mutation will be dependent on partial relief from selection pressure for maintenance of sulfatase activity,t hrough,f or example,g ene duplication prior to mutation of H103 in one of the two copies. The fitness landscape also shows that threonine (found in most activeP MHs) is accessible by three continuouslyu phill pathways:t hat is, every mutational step resultsi n improvement of the promiscuousp hosphodiesterase activity relative to the enzymev ariant that directly precedes it.
Principal component analysis( PCA) is validated as at ool for visualizing and analyzing specificity effects of structurally analogousm utations, as well as trade-offs between different catalytic activities across ap anel of mutants. PCA allowed the projectiono ft he four-dimensionala ctivity space of the promiscuous enzymes studied onto a2Dp lane.I nc ontrast with, for example,c orrelation dot-plots between pairs of activities (six of which would be required per enzyme library studied), PCA condenses the information to as ingle plot. The unique pattern of locations in the new activityp lane mapped by principal components allowed for ready comparison of the specificity effects of structurally analogousm utations in the SpAS1 H103X and RlPMH T107X libraries. Combination with ap lot of the four activities as vectors into the same activity space allowed analysis of correlation patterns between activities, and facilitated detection of inconsistencies in correlation between activities. PCA is complemented by conventionala nalysis of correlation coefficients between activities.B oth analyses revealt hat the same, structurally analogous amino acid substitution can have substantially different specificity effects in two related enzymes, and that properties of substrate GS or TS do not consistently predict the effect of mutation.
Steric constraints and local structural environments of active sites appear to determine much of the variation in promiscuous enzymes pecificity in response to mutation.B oth the activity effects in the two site-saturation libraries analyzed here,a s well as those observed with nucleophile mutants (Cys/fGly! Ser) of as et of ASs and PMHs, reveal that properties of the substrates and catalyzed reactions fail to explain the effect of mutationo ns pecificity.S tructurally analogous mutationst o the same residue have distinct effects on specificity,depending on the enzyme into whicht hey are introduced. The fact that chemicalp roperties of reactions and substrates are not always useful predictors of mutational effects suggests that subtle differences in the active-site environments, leadingt os teric constraints, as well as multiple bindingm odes, might be very important in tuning the precise specificity of promiscuouse nzymes.
Active-site residues can contribute to catalysis in many ways, depending on their orientation with respectt ot he substrate. Promiscuous enzymes are opportunistic, or rather permissive, and this permissiveness can be furthere nhanced by conformational diversity, [81][82][83] which has also been shown to contribute to physiological multifunctionality, [78,79] and can be altereda s ac onsequence of evolution. [84] If two substrates, despite chemical similarities, interactw itha na ctive site in differento rientations-for example, as ac onsequence of subtle steric clashes-the effects of mutation on activities might indeed be very different.

The degreeo fchemical differentiation between reactions in promiscuous enzymes might be limited
The natureo ft he TS and its charge distribution should be of paramount importance for the highly specific interactions between an enzymea nd its substrate. However,i fm olecular recognition is tolerant rather than highly specific,s eemingly straightforward extrapolations from the solution mechanism might be invalid. The observation that specificitye ffects of mutations are more dependent on the substrate structure than on the properties of the reactions furthers upports the notion of catalytic promiscuity as the result of inherentr eactivity of the active site:p romiscuous enzymes might not be able (or need) to specialize more in terms of chemistry,n or change the nature of transition states-leaving sterics as the paramount contributor to the observed specificity. [1,9,10,43,72,77] This suggests that the specificity pattern, or rather ab asic catalytic capability, might indeed be intrinsic, and that promiscuity might exist, as stated by Copley, [85] "simply because it is impossible to exclude all potentialsubstrates".
As ar esult of this apparently limited ability to differentiate reactions even in closelyr elated enzymes,i na ny larger enzyme family, or even just a" quasi-species" made up of func-tionally similarm utants, [57,58] the chances that an enzyme sequencet hat "fits the bill" to allow for ap articular promiscuous activity will be present are substantial.

Acase for keeping template variability highduring protein engineering
The factt hat very different scenarios are reported in the literature with respectt ot he number of mutations required for functional transitions highlights the fact that generalizations answering the questions "how many mutations are required?" or "are mutationsg enerally detrimental?" are unlikely to hold for the engineering of anyo ne particular enzyme.T hey will have to be defined for each individual enzyme,b ecause the shape of the local fitnessl andscape of ar esidue is determined by its structurale nvironmenta nd its quantitativec ontribution to TS stabilization.T he observation of very different physicochemicals olutions to the same problem (in this case, improvement of phosphodiesterase activity in the SpAS1 H103Xlibrary) in combination with the context-dependence of mutational effects observed in this work, and by other researchers, [22,62] demonstrates how important it is to take as many variants as possible forward into subsequent rounds of enzyme optimization:d ifferent solutions might improveagiven activity initially, but the chosen starting point for subsequent engineering might substantially influence its long-term outcome.

Experimental Section
Materials:P hosphate monoester 1,p hosphonate monoester 3c, and sulfate monoester 4 were purchased from Sigma-Aldrich. Phosphate diester 2b was prepared as described previously. [86] All other phosphoester substrates were prepared as described in van Loo et al. (2016). [17] StrepTactin-coated spin columns (IBA) and StrepTactin-Superflow resin (IBA) for protein purification were purchased from Stratech Scientific. Pfu DNA polymerase was obtained from Agilent. DpnI was from ThermoFisher Scientific. Custom mutagenic oligonucleotides were obtained from Invitrogen.
Mutant construction:M utants RlPMH C57S and T107A were constructed previously. [11] All other site-directed mutants were created by using the QuikChange site-directed mutagenesis protocol (Agilent), with the corresponding pASK-IBA5plusPMH/AS wild-type plasmids [17] as templates and use of the primers listed in Ta ble S15-S17. The presence of the mutations was confirmed by DNA sequencing (sequencing facility of the Department of Biochemistry,University of Cambridge).
Protein production and purification:P roduction of the Cys-to-Ser mutants of the various ASs and PMHs was achieved by growing E. coli TOP10 containing the appropriate pASK-IBA5plus constructs in 2YT medium (750 mL) containing ampicillin (100 mg L À1 )a t3 78C to an OD 600 of % 0.5, at which point the culture was cooled to 25 8Ci n% 30 min. Expression of the corresponding AS/PMH Cys-to-Ser variants was achieved by addition of anhydrotetracycline (up to 200 mgL À1 )f ollowed by overnight growth at 25 8C. Harvesting of cells and subsequent protein purification were performed as described previously. [11,14,17] Production of the other SpAS1 and RlPMH variants was done by inducing expression of the appropriate pASK-IBA5plus constructs in E. coli Rosetta 2c o-expressing the formylglycine-generating enzyme (FGE) from Mycobacterium tuberculosis H37v (MtbFGE) [87] from the pRSFDuetMtbFGE plasmid. [11] Large-scale protein production of SpAS1 H103T and RlPMH T107G was achieved by growing the appropriate variants in 2YT medium (750 mL) containing ampicillin (100 mg L À1 )a nd kanamycin (50 mg L À1 )a t3 78Ct oa nO D 600 of % 0.5, at which point the culture was cooled to 25 8Ci n % 30 min. Expression of MtbFGE was induced by adding isopropyl b-d-1-thiogalactopyranoside (IPTG, up to 1mm), approximately 20 min prior to induction of expression of SpAS1 H103T or RlPMH T107G by addition of anhydrotetracycline (up to 200 mgL À1 )f ollowed by overnight growth at 25 8C. Harvesting of cells and subsequent protein purification was carried out as described previously. [11,13,17] For the small-scale activity tests, cells were typically grown in lysogeny broth (LB) medium (5 mL) containing ampicillin (100 mg L À1 )a nd kanamycin (50 mg L À1 )a t3 7 8Ct oa nO D 600 of % 0.5, at which point the culture was cooled to 25 8Ci n% 30 min. Expression of MtbFGE was induced by addition of IPTG (up to 1mm), approximately 20 min prior to induction of expression of the SpAS1/RlPMH variant by addition of anhydrotetracycline (up to 200 mgL À1 )f ollowed by overnight expression at 25 8C. Cells were harvested by centrifugation and resuspended in lysis solution (500 mL, BugBuster (Novagen, 0.5 ), Lysonase (Novagen, 1 mLp er mL lysis solution), cOmplete EDTA-free protease inhibitor (Roche, 1t ablet per 15 mL lysis solution)) and incubated at 25 8Cf or 15 min. After addition of wash buffer (1 mL, Tris·HCl (100 mm), NaCl (150 mm), MnCl 2 (100 mm), pH 8.0) and mixing, cell debris was removed by centrifugation at 4 8C. Recombinant Strep-tagged SpAS1 and RlPMH variants were purified from cleared lysates over Strep-Tactin spin columns by following the manufacturer's protocol (IBA, GmbH, www.iba-lifesciences.com) with the wash buffer mentioned above and elution buffer (3 150 mL, Tris·HCl (100 mm), NaCl (500 mm), MnCl 2 (100 mm), d-biotin (2 mm), pH 7.5). The eluates showed excellent purity (> 95 %) by SDS-PAGE analysis (Coomassie staining), with only one visible minor additional gel band that was ascribed to degradation.