Deciphering the role of recurrent FAD-dependent enzymes in bacterial phosphonate catabolism

Summary Phosphonates—compounds containing a direct C–P bond—represent an important source of phosphorus in some environments. The most common natural phosphonate is 2-aminoethylphosphonate (AEP). Many bacteria can break AEP down through specialized “hydrolytic” pathways, which start with the conversion of AEP into phosphonoacetaldehyde (PAA), catalyzed by the transaminase PhnW. However, the substrate scope of these pathways is very narrow, as PhnW cannot process other common AEP-related phosphonates, notably N-methyl AEP (M1AEP). Here, we describe a heterogeneous group of FAD-dependent oxidoreductases that efficiently oxidize M1AEP to directly generate PAA, thus expanding the versatility and usefulness of the hydrolytic AEP degradation pathways. Furthermore, some of these enzymes can also efficiently oxidize plain AEP. By doing so, they surrogate the role of PhnW in organisms that do not possess the transaminase and create novel versions of the AEP degradation pathways in which PAA is generated solely by oxidative deamination.

These enzymes serve to degrade the naturally occurring N-monomethyl-AEP Some of these enzymes also act on AEP, creating a novel AEP catabolic pathway

INTRODUCTION
For an organism, the ability to consume multiple distinct but related compounds through a single catabolic pathway represents a straightforward and economical way to exploit different nutrient sources and hence increase organismal fitness under variable nutrient conditions.2][3] Examples of this kind are widespread also in other, less central catabolic routes.For instance, in Escherichia coli a single enzyme has been described that transforms two different dehydrated forms of N-acetylneuraminate (sialic acid) into plain sialic acid, to feed the dedicated degradation pathway. 4erein, we looked for ancillary enzymes that might expand the versatility of the catabolic pathways for 2-aminoethylphosphonate (AEP; also known as ciliatine), which is the most prevalent natural phosphonate. 5,64][15] In particular, since AEP is abundant in nature, 5,6,14,16 a number of bacteria possess specialized ''hydrolytic'' pathways for the degradation of this compound (Figure 1).
In the most common case, the hydrolytic degradation pathway begins with a reaction catalyzed by the transaminase PhnW, which converts AEP into phosphonoacetaldehyde (PAA) 20 ; subsequently, PAA is cleaved into acetaldehyde and inorganic phosphate by the hydrolase PhnX (Figure 1A).In an alternative pathway, PAA is converted to phosphonoacetate by the dehydrogenase PhnY and finally to acetate and phosphate by the hydrolase PhnA 21,22 (Figure 1B).
Given the widespread diffusion of both the PhnWX and PhnWYA pathways, it is biologically plausible that accessory enzymes may have evolved to convey into these catabolic routes other naturally occurring, AEP-related phosphonates.As a case in point, we recently reported the identification and characterization of one enzyme that is encoded in over 13% of the bacterial gene clusters containing the phnWX (or phnWYA) combination. 23We termed this enzyme PbfA (for ''phosphonate breakdown factor A 00 ) and eventually showed that it is a lyase acting (B) Variant of the pathway in A where PAA is first converted to phosphonoacetate and then to acetate and phosphate by the enzymes PhnY and PhnA, respectively.][19] oxidases, indicating that they use molecular oxygen as the physiological electron acceptor and hence produce H 2 O 2 .However, for the mammalian sarcosine dehydrogenase (SarDH) the physiological electron acceptor is a flavoprotein 42 ; whereas for the D-amino acid dehydrogenase from Helicobacter pylori (DadA) the electron acceptor is presumably a quinone 40 (Figure S4).

Hypotheses on the function of the FAD-dependent oxidoreductases
Enzymes from the PbfB, PbfC, and PbfD subgroups were never co-present in the same cluster.This observation, together with the fact that the different FAD-dependent enzymes were invariably similar to amine oxidoreductases, suggested that they all might be performing essentially the same transformation in aminophosphonate degradation.The enzymes of known function are: PuuB, g-glutamyl putrescine oxidase from E. coli 35 ; MabO, 4-methylamino butyrate oxidase from Arthrobacter nicotinovorans 36 ; ThiO, glycine oxidase from Bacillus licheniformis 37 ; SolA, N-methyl-L-tryptophan oxidase from E. coli 38 ; SarDH, sarcosine dehydrogenase from Rattus norvegicus 39 ; DadA, D-amino acid dehydrogenase from Helicobacter pylori 40 ; SoxA, sarcosine oxidase from Bacillus sp.B-0618. 41The GenBank accession IDs for these enzymes are found in Table S2 and the reactions they catalyze are shown in Figure S4.Percent identities were evaluated by the pairwise global alignment of the portions of the sequences denoting the PF01266 domain, carried out with Needle (https://www.ebi.ac.uk/Tools/psa/emboss_needle/) with a Blosum30 substitution matrix.Percent identities rarely exceed 30%, indicative of only distant homologies between the oxidoreductases described in this study and the other functionally characterized PF01266 enzymes.A more refined analysis of the evolutionary relationships between the enzymes in this table is provided in Figure S5.Related to Table S2, Figures S4 and S5.The tree only reports bootstrap values for the main basal nodes and those supported by bootstrap values <50 were collapsed.Full bootstrap metrics, together with organisms' names, are reported in Figure S3, whereas the sequence accession IDs are provided in Table S1.The scale bar indicates the number of substitutions per site.Groups of sequences that we labeled PbfB, PbfC, and PbfD are highlighted in light green, light red and light blue, respectively.The monophyly of all three clades was extremely well supported (bootstrap values = 100), in line with the high sequence divergence observed between groups (i.e., median inter-group p-distances was equal to = 0.81).A subgroup of monophyletic early branching PbfD sequences, termed PbfD1, was also wellsupported (bootstrap value = 77) and is signaled by a darker shade of blue.pbfD1 genes usually clustered with phnW and phnX, whereas other pbfD genes were most commonly associated with phnX alone (see Table S1).Related to Figures S1-S3 and Table S1.
But what could be the function of these oxidoreductases?As outlined in the introduction, we assumed that these enzymes served to degrade compounds related to AEP.More detailed hypotheses about their actual activity could be formulated based on the following considerations: (i) the substrate of these enzymes should contain a primary or secondary amino group, consistent with the reactions catalyzed by the most closely related enzymes of known function (Figure S4); (ii) the product of the reaction should presumably be either AEP or PAA, in order to enter the PhnWX or PhnWYA pathway.
Based on these considerations, our initial hypothesis was that PbfB, PbfC, and PbfD enzymes could all catalyze the oxidative deamination of some AEP derivative(s) alkylated on the amino group.8][29][30][31][32][33] This compound cannot be processed by PhnW, which-like all PLP-dependent transaminases-requires a substrate with a primary amino group.We thus envisaged that the FAD-dependent oxidoreductases might carry out the reaction shown in Scheme 1.
The direct oxidation of M 1 AEP would produce methylamine and PAA, which could be then conveniently hydrolyzed by PhnX.Given the structural similarity of M 1 AEP to plain AEP, we also expected that the FAD enzymes could oxidize AEP with some efficiency, producing PAA with the release of ammonia.This reaction would be redundant with respect to the reaction catalyzed by PhnW, which also generates PAA from AEP (Figure 1); however, in those organisms that do not possess PhnW (e.g., M. fucicola-Figure S1), the oxidative deamination of AEP could be not just biologically acceptable, but wholly desirable.

The PbfC and PbfD enzymes oxidize M 1 AEP
Synthetic genes encoding the PbfB protein from Vibrio vulnificus, the PbfC protein from Azospirillum sp.B510, and the PbfD proteins from Acinetobacter baumannii (henceforth termed PbfD1) and Mariniblastus fucicola (henceforth termed PbfD2) were purchased with codon optimization for expression in E. coli.The genes were inserted in the bacterial expression vector pET28a and the recombinant His-tagged proteins were expressed and purified from E. coli cells as described in the STAR Methods.
All four recombinant proteins appeared to be soluble to some extent; however, despite our efforts, PbfB could not be purified by metalaffinity chromatography from cell extracts, which greatly limited its characterization (see in the following section).On the other hand, we obtained the recombinant PbfC, PbfD1, and PbfD2 enzymes in pure form and proceeded to characterize their catalytic activity toward a number of potential substrates, including some non-phosphonate AEP analogs.
We initially used 96-well plate assays to test the actual ability of the FAD enzymes to oxidize various amino compounds, employing the artificial dye DCPIP as the final electron acceptor and PMS as an electron mediator. 43DCPIP is routinely used in the field of FAD-dependent enzymes to measure the activity of both oxidases [44][45][46] and dehydrogenases. 43,46,47Using this assay, any oxidation reaction of the substrate results in DCPIP turning from blue to colorless, so that the color of the reaction mixture shifts from green to yellow (the yellow color being due to the presence of PMS).
A second type of colorimetric assay served to evaluate the ability of the enzymes to oxidize potential substrates with electron transfer to oxygen.The assay measured the formation of H 2 O 2 , which can be used by horseradish peroxidase to oxidize the chromogenic compound odianisidine, turning the solution to deep pink.Contrary to the DCPIP assay, this method entailed a single readout taken after a relatively long incubation, thus it could only qualitatively assess the capacity of the enzyme to reduce molecular oxygen.
The results of these microtiter assays are reported in Figure 3. Consistent with our initial hypothesis, these preliminary assays implied a strong reactivity of all three oxidoreductases with M 1 AEP, as well as some activity toward the closely related compound N-ethyl-AEP (which, however, is not known to occur in nature).In particular, PbfC did not react appreciably with other compounds: moreover, activity of this enzyme was readily observed in the DCPIP assay, but barely detectable in the peroxidase-based assay, suggesting that PbfC inefficiently uses molecular oxygen as the electron acceptor.
On the other hand, PbfD1 and PbfD2 enzymes manifested a strong oxidative activity toward M 1 AEP in both the DCPIP-based assay and the peroxidase-o-dianisidine assay.These two enzymes also showed a clear reactivity toward AEP.Furthermore, PbfD1, but not PbfD2, appeared to appreciably oxidize N,N-dimethyl-2-aminoethylphosphonate (M 2 AEP).
None of the enzymes reacted with non-phosphonate analogs of M 1 AEP (such as sarcosine) or of AEP (e.g., taurine, b-alanine).The enzymes also failed to oxidize 2-hydroxyethyl phosphonate, an analog of AEP that contains a hydroxyl group in place of the amino group (data not shown).

The PbfC and PbfD enzymes generate PAA
Although the microtiter assays indicated the ability of the FAD enzymes to oxidize M 1 AEP, other details of the hypothesized reaction had to be confirmed.In principle, the oxidation of M 1 AEP (but not AEP) could generate AEP and formaldehyde, instead of PAA and methylamine.To test for this possibility, we employed a spectrophotometric assay in which the three oxidoreductases were incubated with M 1 AEP for 30 min; afterward, NADH and alcohol dehydrogenase (ADH) were added to detect the oxidation of NADH in case formaldehyde had been produced.The observation that NADH was not appreciably consumed (data not shown) implied that formaldehyde had not been formed (PAA is not a substrate for ADH).Conversely, when the oxidoreductases were incubated for 30 min with M 1 AEP and PhnX, which hydrolyzes PAA to acetaldehyde, and then the reaction mixture was supplemented with ADH and NADH, the latter was quickly oxidized.This observation was fully consistent with the hypothesis that the PbfC and PbfD enzymes produce PAA.
We also demonstrated that the reaction of PbfD1 and PbfD2 with AEP (but not M 1 AEP) released ammonia, as revealed by a coupled assay with GDH (data not shown).Other controls established that phosphate was not released in the reaction of the oxidoreductases with either M 1 AEP or AEP, except when PhnX was also present (BIOMOL Green assay, data not shown).
Finally, the reaction of PbfC, PbfD1, or PbfD2 with M 1 AEP was monitored by 1 H-NMR, establishing beyond doubt that the reaction products are methylamine and PAA.The results pertinent to the reactions of the three enzymes are summarized in Figure S6.NMR experiments also showed that the reaction of PbfD1 with M 2 AEP generates dimethylamine and PAA (Figure S7).

The PbfC and PbfD enzymes have different propensities to reduce oxygen
The purified PbfC, PbfD1, and PbfD2 enzymes appeared yellow, with their absorption spectra exhibiting two peaks in the visible range, typical of an oxidized flavin cofactor (Figures 4A-4C).The first peak was centered at $365 nm for both PbfD1 and PbfD2, but at 387 nm for PbfC.The second peak ranged from $447 nm for PbfD1 to $462 nm for PbfC (Figures 4A-4C).The cofactor was released in solution when the enzymes were heat-precipitated (10 min at 100 C, in the dark), suggesting that the flavin was not covalently bound.Furthermore, when the proteins were denatured using SDS, the spectrum of the released cofactor was typical of FAD 48 (and not FMN; see STAR Methods).Addition of an excess of M 1 AEP to the native enzymes caused spectroscopic changes indicative of FAD reduction in all cases (Figures 4A-4C).
To assess the ability of the enzymes to transfer electrons from FAD to different acceptors, we used two continuous spectrophotometric assays.A first assay was based on the reduction of the artificial acceptor dye DCPIP (in the presence of the electron mediator PMS), as described earlier.The second assay was conducted in the absence of artificial electron acceptors and coupled the production of PAA with the consecutive reactions of PhnX (which converts PAA into acetaldehyde) and of alcohol dehydrogenases (that reduces acetaldehyde while consuming NADH). 23In this assay, molecular oxygen was assumed to serve as the electron acceptor for the oxidoreductases' reaction.
Consistent with the findings of the microtiter assays in Figure 3, PbfC showed a very weak activity in the presence of oxygen alone, but was much more active when DCPIP could serve as the electron acceptor.Conversely, PbfD1 and PbfD2 were more active in the coupled assay, suggesting a strong propensity of these enzymes to directly reduce oxygen (Figures 4D-4F).In the green/yellow wells, the potential substrates were incubated at room temperature in the presence (+) or absence (À) of the oxidoreductases (0.5 mM); the mixture also contained the redox dye DCPIP (which served as the final electron acceptor) and the redox mediator PMS.Reactions were started by adding the enzyme last and photographed at regular intervals-the images in this figure were taken 6 min after adding the enzyme.In wells where the reaction occurred, the solution turned from green to yellow.The transparent/pink wells refer instead to an endpoint microtiter assay aimed at detecting the release of hydrogen peroxide.The reactions were conducted as aforementioned, except that artificial dyes were omitted and molecular oxygen served as the electron acceptor.After 30 min, the wells were supplemented with horseradish peroxidase and o-dianisidine and (after another 5 min) sulfuric acid.Wells where hydrogen peroxide had formed turned deep pink.Abbreviations: PnAla, 3-phosphono-DL-alanine; 3APP, 3-aminopropylphosphonate; PEA, phosphoethanolamine.Related to Figures S6 and S7.

Catalytic efficiency and specificity of the PbfC and PbfD enzymes
To establish the catalytic performances of the PbfC and PbfD oxidoreductases toward naturally occurring substrates, we measured the activity of these enzymes using the continuous spectrophotometric assays that worked best for each of them (i.e., the DCPIP reduction assay for PbfC and the coupled PhnX-ADH assay for the PbfD enzymes).The pertinent titrations fitted well to the Michaelis-Menten equation (Figures 4G-4I) and the apparent catalytic parameters are reported in Table 2.
The apparent catalytic efficiencies (k cat /K M values) of the three enzymes toward M 1 AEP were consistently in the range 2310 3 -5310 4 M À1 s À1 , whereas K M values ranged from about 0.2 mM to about 5 mM.These performances, while not outstanding, are comparable to those reported for other FAD-dependent oxidoreductases acting on relatively small substrates.For example, glycine oxidase from B. licheniformis  (A-C) UV-vis absorption spectra of the three enzymes (10 mM) before (black line) and after (gray line) the addition of 20 mM M 1 AEP.Spectra were collected at room temperature under aerobic conditions in 50 mM pyrophosphate buffer pH 8.0 (PbfC) or 50 mM phosphate buffer pH 8.0 (PbfD1 and PbfD2).(D-F) Relative activities of the three enzymes in the presence of 5 mM M 1 AEP, measured at room temperature through the assay based on DCPIP-PMS (lime yellow bar) or through the coupled assay with PhnX and ADH (magenta bar).The error bars represent standard errors for triplicate experiments.Activity in the assay that gave the maximum rate was taken as 100%.Both types of assays were conducted in TEA-HCl buffer, pH 8.0.(G-I) Activity of the three enzymes toward naturally occurring substrates that had shown some reactivity in the microtiter assays (TEA-HCl buffer pH 8.0, 25 C).The activity of PbfC, was measured with the DCPIP reduction assay whereas the activities of PbfD1 and PbfD2 were measured under atmospheric oxygen through the PhnX-ADH coupled assay.Each data point represents a triplicate experiments; bars signal standard errors.Solid lines represent fittings of the data to the Michaelis-Menten equation.The fitting-derived kinetic parameters are reported in Table 2.
(ThiO, Table 1) was reported to oxidize glycine with a k cat /K M of 340 M À1 s À1 and with a K M of 0.9 mM 37 ; whereas the monomeric sarcosine oxidase from Bacillus sp.B-0618 (SoxA, Table 1) oxidized sarcosine with a k cat /K M of 10,100 M À1 s À1 and a K M of 4.5 mM. 41bfC showed the highest apparent catalytic efficiency in the oxidation of M 1 AEP and, in agreement with the semi-quantitative results of the microtiter assays, PbfC also showed the highest discrimination toward non-methylated AEP (400-fold difference in catalytic efficiency), which seemed mostly due to a weaker substrate binding (100-fold difference in K M ).In comparison, PbfD2 showed only a 4-fold difference in k cat /K M values between M 1 AEP and plain AEP, while PbfD1 specificity was somewhat intermediate, with an 18-fold discrimination toward AEP.
While the experiments earlier refer to naturally occurring aminophosphonates, we also performed assays using N-ethyl-AEP and N-propyl-AEP, which are not known to occur in nature.The results, shown in Figure S8, confirm in general a dichotomy between the rather substratespecific PbfC and the less strict PbfD enzymes.

Evidence that the PbfB enzymes also oxidize M 1 AEP
As stated previously, we could not obtain the recombinant PbfB from V. vulnificus in pure form.Therefore, to test its activity, we resorted to performing assays on the extracts of E. coli cells that expressed this enzyme.In this case, however, assays based on the formation of H 2 O 2 (Figure 3), were deemed unreliable because other enzyme activities in the extracts (catalases, for example) could curb the accumulation of the expected products.Therefore, considering that inorganic phosphate is a much more stable end-product, we opted initially for a microtiter assay based on the BIOMOL Green assay kit.
The rationale was that if the extract contained any activity capable of producing PAA from M 1 AEP, coupling this reaction with the PhnX reaction would result in the eventual accumulation of phosphate.As shown in Figure 5A, this was exactly the case.
Phosphate accumulation was dependent on the simultaneous presence in the reaction mixture of the cell extract, of PhnX and of M 1 AEP, consistent with the rationale of the assay.No accumulation of phosphate was detected when the extract came from E. coli cells not expressing PbfB, or when the extract of PbfB-expressing cells had been inactivated by boiling (Figure 5A).No accumulation of phosphate was observed when M 1 AEP was replaced by plain AEP, suggesting a certain substrate specificity of the process.Finally, when cell extracts were used in a DCPIP-reduction assay (analogous to the one depicted in Figure 3) the strongest signal occurred when both the extract from PbfB-expressing cells and M 1 AEP were present (Figure 5B).
Overall, these data were fully consistent with the hypothesis that PbfB from V. vulnificus catalyzes the oxidation of M 1 AEP, much like the other oxidoreductases described in this work.The activity detected in extracts of PbfB-expressing cells apparently took place both in the presence and in the absence of electron-accepting dyes (Figure 5B); however, the limits of the assays conducted on whole cell extracts do not allow one to conclude whether PbfB behaves as an oxidase (like the PbfD enzymes) or as a dehydrogenase (like PbfC).

FAD-dependent enzymes from AEP degradation clusters serve to oxidize M 1 AEP
Different studies have highlighted the role of phosphonate degradation from an ecological standpoint (in the frame of the global biogeochemical P-cycling) and as a tool that bacteria employ for surviving in nutrient-limited ecological niches.0][51][52] However, except for the very flexible but complex C-P-lyase system, 52,53 these pathways appear to be very narrow in substrate range, being specialized for the degradation of one particular compound.
The hydrolytic pathways for AEP degradation are a case in point.While AEP is indeed the most abundant natural phosphonate, 5,6,16 the known pathways shown in Figure 1 are by themselves unable to process other AEP-related phosphonates that abound in the environment.As The kinetic parameters were obtained by fitting the titration data reported in Figure 4 (panels G-I) to the Michaelis-Menten equation.The reported parameters (G SE of the fitting) are rounded to the first two significant figures; they are deemed 'apparent' because they were measured in the presence of a fixed concentration of the electron acceptor co-substrate.The activity of PbfC was assessed in the presence of 80 mM DCPIP and 3 mM PMS. Activities of the PbfD enzymes were determined in aerated solutions, i.e., in the presence of $0.25 mM O 2 .Related to Figure S8.
noted in the introduction, one evolutionary workaround could be the co-option of accessory enzymes able to convey additional compounds into the AEP hydrolytic pathways.This study deals with the recurrent presence of genes encoding putative (and only weakly interrelated) FAD enzymes within bacterial clusters for AEP degradation.With ''recurrent,'' we refer to genes that are repeatedly found in such clusters-but whose presence is neither constant nor limited to a specific cluster type.The optional but frequent occurrence of these genes is very consistent with them encoding ancillary enzymes, acting on compounds related to the main substrate of the pathway.Establishing the precise function of these accessory enzymes, however, is often not straightforward.For example, to convincingly establish the specificity of the oxidoreductases described in this work we had to test a variety of phosphonates 54 and this required in particular the chemical synthesis of numerous N-alkylated derivatives of AEP and (R)-HAEP, by either newly developed or adapted literature methods.M 1 AEP and M 2 AEP were obtained by reaction of diethyl 2-bromoethylphosphonate with the corresponding mono-and dialkylated amines, followed by acidic deprotection.M 3 AEP became accessible from diethyl 2-(dimethylamino)-ethylphosphonate by alkylation with methyl iodide following an adapted literature procedure. 55The trimethylated analogue (R)-M 3 HAEP was obtained from (R)-HAEP by methylation, while the (R)-M 2 HAEP was accessible by an Eschweiler-Clark approach.The monomethylated derivative (R)-M 1 HAEP was accessible from (R)-diisopropyl (2-(1,3-dioxoisoindolin-2-yl)-1-hydroxyethyl)phosphonate 18 by using a tosylated intermediate to control the degree of methylation.
The possible role of some of the FAD oxidoreductases described herein was tentatively hinted at by their genomic annotations, but such annotations, when present, remained vague and unsubstantiated.For example, PbfD1 from A. baumannii is annotated in GenBank as a member of the TIGR03364 family (see Table S1), whose description reads ''FAD dependent oxidoreductases [.] syntenically associated with a family of proposed phosphonatase-like enzymes [.]A likely role for this enzyme involves the oxidative deamination of an aminophosphonate differing slightly from AEP, possibly 1-hydroxy-2-aminoethylphosphonate'' (that is, HAEP).
We have now provided strong evidence (in the case of PbfB) or complete proof (for the PbfC and PbfD proteins) that these FAD enzymes catalyze the oxidative deamination of the N-monomethylated form of AEP (M 1 AEP), generating PAA that can be then hydrolyzed by PhnX to obtain acetaldehyde and inorganic phosphate.
This activity seems biologically well justified: M 1 AEP is frequently found in nature (where its abundance can sometimes approach that of AEP 28,31 ) but it cannot serve as a substrate for PhnW and therefore it cannot directly enter the pathways shown in Figure 1.Thus, collectively, the PbfB, PbfC, and PbfD enzymes serve to funnel M 1 AEP into the hydrolytic AEP degradation pathways, allowing the bacteria to consume different aminophosphonates and presumably improving organismal fitness under different nutrient conditions.

Functional diversity of the PbfC and PbfD enzymes
Even though all the oxidoreductases described in this work share the ability to oxidize M 1 AEP, their mechanistic features are remarkably distinct.The PbfC enzyme from Azospirillum sp. was peculiar in its strong discrimination against plain AEP, which was oxidized $400-fold less efficiently than its N-methylated counterpart (comparison is based on k cat /K M values 56,57 ; Figure 4G and Table 2).This preference even exceeds the calculated maximum discrimination against suboptimal substrates that lack a methyl group (the theoretical limit would be a $160-fold discrimination 58,59 ) but in this case the methyl group is directly attached to the nitrogen undergoing reaction and this is expected to change some properties of the amino group, like its pK a .
Notably, PbfC proved very inefficient in using O 2 as the immediate electron acceptor, whereas its reaction proceeded quickly in the presence of the artificial electron acceptor DCPIP.This suggests that PbfC should be formally classified as a dehydrogenase and we propose for this enzyme the systematic name N-methyl-2-aminoethylphosphonate dehydrogenase.Physiologically, it is not clear what the electron acceptor could be.NAD + and NADP + did not apparently accept electrons from this enzyme (data not shown), so the most obvious possibilities would be a quinone (in Azospirillum species, ubiquinone Q-10 is reportedly the major respiratory quinone 60 ) or perhaps some carrier protein, as for the mammalian sarcosine dehydrogenase (SarDH, Figure S4). 42n contrast to PbfC, the PbfD enzymes showed a similarly efficient reaction in the presence of either O 2 or DCPIP-PMS suggesting that these enzymes should be classified as oxidases (even though we cannot rigorously exclude that an electron acceptor better than oxygen may exist in vivo).Our data hint that the PbfB-type enzymes can carry out the M 1 AEP oxidation in the absence of acceptor dyes (Figure 5); however, our inability to obtain PbfB in pure form prevents any firm conclusion on the formal classification of this enzyme as an oxidase or dehydrogenase.
A further distinctive feature of PbfD1 and PbfD2 is that they exhibited a less pronounced specificity for M 1 AEP.In particular, PbfD1 was unique in showing an appreciable activity toward N,N-dimethyl-AEP (M 2 AEP), through a reaction somehow resembling that of dimethylglycine oxidase (also belonging to the PF01266 group). 61The PbfD1-catalyzed oxidation of M 2 AEP was just 6-fold less efficient than the oxidation of M 1 AEP (Table 2) and could perhaps have some physiological relevance.Small amounts of M 2 AEP have been occasionally isolated from marine organisms (e.g., ref. 28 ); furthermore, M 2 AEP may conceivably be an intermediate in the biosynthesis of M 3 AEP, which is another AEP analog frequently found in nature.Notably, M 3 AEP itself was not a substrate for any of the oxidoreductases described in this study.Indeed, there are no FAD-dependent enzymes reported to act on quaternary amines, while the only M 3 AEP degradation pathway described to date involves two iron-dependent oxygenases and produces trimethylglycine in addition to phosphate. 62ore relevant from a biological viewpoint was the ability of PbfD2 from M. fucicola to react with AEP only 4-fold slower than with M 1 AEP (Figure 4I and Table 2).Such a finding implies that in vivo this enzyme could be performing a double duty, generating PAA not only from M 1 AEP but also from AEP.Through this latter reaction, PbfD2 would in effect surrogate the function of PhnW, whose gene is absent in the M. fucicola gene cluster (Figure S1) and apparently from the whole M. fucicola genome.Altogether, these observations strongly suggest that enzymes like PbfD2 may contribute to ''novel'' versions of the hydrolytic pathways, in which PAA is generated by the oxidative deamination of AEP rather than by transamination (see in the following section).

On the evolution of M 1 AEP degrading enzymes
The heterogeneity of the FAD-dependent enzymes analyzed in this study, both in terms of sequence and mechanistic details, outlines a remarkable instance of convergent evolution.It suggests that evolution has ''invented'' an ancillary M 1 AEP oxidoreductase at least three times, co-opting different ancestor enzymes to perform the task.PbfB sequences might have emerged last, as they are the less divergent (Figure 2) and their distribution is mostly limited to Betaproteobacteria and Gammaproteobacteria (Table S1 and Figure S3).On the other hand, PbfC homologs associated with AEP degradation clusters were found in Alphaproteobacteria, Gammaproteobacteria (where they were widespread), and Betaproteobacteria, but also frequently observed in bacteria belonging to the FCB group and occasionally in Terrabacteria.The PbfD group of enzymes might have been the most ancient to emerge: this is suggested both by the large divergence between the sequences of these enzymes (the two PbfD enzymes tested in this work, for example, show only a $30% identity) and by the relatively broad distribution among bacterial phyla.PbfD homologs are encoded in AEP degradation clusters in Alphaproteobacteria, Betaproteobacteria, Gammaproteobacteria, and bacteria from the FCB and Planctomycetota, Verrucomicrobiota, and Chlamydiota (PVC) groups (mostly Planctomycetota).Notably, PbfD homologs not associated with phnWX/phnWYA clusters were also detected in many Terrabacteria (mostly cyanobacteria).
The varying presence of PbfB, PbfC, and PbfD enzymes in closely related organisms is consistent with a significant degree of horizontal gene transfer involving these pathways.For example, Rhodoferax sediminis was found to possess pbfB-phnWYA genes, while the closely related Rhodoferax koreense possesses pbfD-phnX instead (Table S1).Additionally, the similarities between genes in Figure 2 (and Figure S3) are not congruent with the phylogeny of the organisms which possess them; e.g., the PbfD1 clade of Figure 2 consists of sequences from diverse phyla, which are more similar to each other than to sequences from similar phyla in the rest of the PbfD group.The horizontal transfer of these accessory genes is consistent with the previously inferred horizontal transfer of the ''core'' genes of the hydrolytic pathways (phnW and phnX for example), and it is logical that these would move between organisms as a part of complete clusters rather than separately. 63ological significance of M 1 AEP degradation Overall, the oxidoreductases described here occur in all the major clades of PhnA and PhnX-possessing bacteria observed in earlier database searches, i.e., predominantly proteobacteria, with some representatives of the PVC group 14,64 (Figure 2 and Table S1).These organisms have very diverse habitat ranges, as they include species from predominantly aquatic (e.g., Vibrio spp. 65) and soil habitats (e.g., Azospirillum spp. 66), as well as cosmopolitan species (e.g., Acinetobacter spp. 67).9][70] ) have been taken as evidence that AEP degradation is advantageous across a wide range of habitats.Our data hint that this may also be true for M 1 AEP degradation, although in vivo studies will be required to fully establish this point.
The most obvious advantage of AEP degradation is the retrieval of inorganic phosphate.However, previous research has shown that AEP degradation (via the PhnWX or PhnWYA pathways) can sustain marine microbes as a source of nitrogen even when phosphate levels are high. 70The degradation of M 1 AEP carried out by PbfC and PbfD enzymes generates methylamine as a product (Figure S6) and in all likelihood this is also true for PbfB.Methylamine is reportedly used as a nitrogen source by a diverse range of Alpha and Gammaproteobacteria, 71 suggesting that hydrolytic pathways that include the oxidoreductases could support growth on M 1 AEP as a nitrogen source.While phosphateinsensitive expression of oxidoreductase-containing pathways has yet to be experimentally demonstrated, this would be consistent with these enzymes broadening the range of aminophosphonates which contribute to marine phosphorus, nitrogen, and possibly carbon biogeochemistry.
Finally, it is also possible to hypothesize that some of the oxidoreductase enzymes may be involved in the host-interactions of microbial species.For example, the Vibrio spp. that possess pbfB (Figure S3) are closely associated, possibly as symbionts, with marine gastropods which synthesize sphingophosphonolipids containing M 1 AEP. 29,72Alternatively, as these Vibrio spp.are also pathogens of mussels and other marine invertebrates that produce lipid-bound M 1 AEP, the PbfB enzymes may play a role in the degradation of host lipids during pathogenesis. 32,73P degradation a `la carte While the FAD-dependent oxidoreductases expand the versatility of the hydrolytic pathways for the degradation of AEP, our data also strongly suggest that some of these enzymes participate in novel versions of such pathways, in which PAA is not generated from AEP by transamination, but rather by oxidative deamination.
Figure 6 provides an updated overview of the different possible hydrolytic pathways for the degradation of AEP and related compounds.All of these pathways center on the common intermediate PAA, which can be formed or degraded through different enzymes, thus providing four different pathways for AEP catabolism (Figure 6).The first is the classic PhnWX pathway (reactions highlighted in light blue and light green, Figure 6), as found for example in Azospirillum sp.B510, where it is enriched by the activities of PbfA and PbfC (Figure S1).The second is the PhnWYA pathway (in Figure 6, reactions highlighted in light blue and yellow), as found in Burkholderia multivorans.A third option is a pathway based on PbfD2 and PhnX (upper set of reactions in Figure 6, highlighted in light red and light green), as observed in M. fucicola.The newly identified enzymes described in this work are highlighted in red.In the oxidation of M 1 AEP the electron acceptor is not indicated for simplicity; as detailed in this work, the acceptor is O 2 for PbfD enzymes, but unknown for PbfC and PbfB.The enzyme catalyzing the oxidative deamination of AEP is indicated as PbfD2 by approximation.Indeed, most of the genes of the pbfD2 subgroup belong to clusters that do not include phnW, but there are exceptions; at the same time, a minority of pbfD genes outside the pbfD2 subgroup localizes in clusters that lack phnW (see Table S1 for examples).
Finally, a fourth pathway based on PbfD2, PhnY, and PhnA is also possible, as found in Luteolibacter luteus (Figure S1).Notably, there are no enzymes in common between the second pathway and the third, nor between the first and the fourth.
The distribution of the four pathways among different bacteria may be an accident of evolution.Nevertheless, it should also be noted that certain pathways may suit better the lifestyles and metabolic setups of individual bacterial species.For example, a pathway based on PhnWYA requires three enzymes and is unable to process M 1 AEP, but is compatible with an anaerobic lifestyle and it generates only harmless products.On the other hand, a pathway based on PbfD2 and PhnX employs only two enzymes and is more versatile, but it can only function under aerobic conditions and it generates two reactive products (hydrogen peroxide and acetaldehyde) that will need to be further processed by other metabolic enzymes.

Limitations of the study
The functional characterization described herein was conducted in vitro, on recombinant proteins expressed in E.coli.It could be useful to complement our results with in vivo approaches, encompassing for example the deletion of the pbfB, pbfC, or pbfD genes in organisms that possess them.Such an approach might, however, be limited by several confounding factors and requires an understanding of the conditions under which the AEP degradation gene cluster is expressed in that organism.Furthermore, while our data strongly indicate that PbfB is involved in the processing of M 1 AEP, details about its exact reaction remain uncertain because we did not succeed in purifying the protein.In particular, the limits of assays conducted on whole cell extracts do not allow us to conclude whether PbfB behaves as an oxidase (like the PbfD enzymes) or as a dehydrogenase (like PbfC).

Procedure A
The respective monoalkylamine (12 eq.) was added at 0 C to a solution of diethyl-(2-bromoethyl)phosphonate (1, 4.7 mmol, 1.152 g, 1 eq.) in dry EtOH (10 mL), and the resulting mixture was stirred for 3 h at room temperature.The solvent was removed under reduced pressure to give diethyl 2-alkylamino-ethylphosphonates (2-4, partially as their hydrobromide), in sufficient purity for the next step.

Procedure B
The respective diethyl 2-alkylamino-ethylphosphonate (2-4) was dissolved in hydrochloric acid (6 M in H 2 O, 12 mL) and stirred at 120 C for 6 h.The crude product was concentrated in vacuo and purified by ion exchange chromatography (Dowex 50W 3 8 / H + -form) using H 2 O as eluent.The product containing fractions (ninhydrin positive) were pooled and the solvent was removed under reduced pressure to yield (2-alkylamino-ethyl phosphonates 5-7, which could be crystallized from H 2 O/EtOH to give colorless needles. 2-N-Methylamino-ethylphosphonic acid (M 1 AEP; 5, 98% over two steps)

Synthesis of Di-and Trimethylated derivatives of (R)-HAEP
Related to Scheme S3. (R)-1-Hydroxy-2-aminoethylphosphonic acid 18 [(R)-13; 141 mg, 1.00 mmol, 1 eq.] was dissolved in formic acid (conc., 0.38 mL, 10.00 mmol, 10 eq.), formaldehyde (37% in H 2 O, 0.60 mL, 6 mmol, 6 eq.) was added and the mixture was refluxed for 18 h.The solvent was removed under reduced pressure and the crude residue was purified by ion exchange chromatography (Dowex 50W 3 8 / H + -form) with H 2 O as eluent.The product containing fractions (ninhydrin positive) were pooled and the solvent was removed in vacuo to yield (R)-  Related to Scheme S3. (R)-13 (479 mg, 3.40 mmol, 1 eq.) was dissolved in MeOH (45 mL) and K 2 CO 3 (2.347g, 16.98 mmol, 5 eq.) was added.After 10 minutes, methyl iodide (1.928 g, 13.58 mmol, 0.85 mL, 4 eq.) was added and stirring was continued for 48 h.The solvent was removed in vacuo and the residue was dissolved in H 2 O (3.4 mL).Subsequently, the pH was adjusted to 1 with conc.H 2 SO 4 and the resulting mixture was stirred for 1 h.The solvent was removed under reduced pressure, the residue was dissolved in water and washed Et 2 O (3 3 4 mL).The aqueous phase was concentrated to dryness, dissolved in a minimal amount of water and applied to an ion exchange chromatography resin (Dowex Monosphere-550A / OH --form).An aqueous solution of formic acid (5% v/v) was used as eluent.The product containing fractions were pooled and concentrated.The obtained amorphous solid can be recrystallized from EtOH/H 2 O to give (R)- Synthesis of a Monomethylated Derivative of (R)-HAEP Related to Scheme S4.Hydrazine-monohydrate (4.06 mmol, 0.2 mL, 2 eq.) was added to a solution of (R)-diisopropyl (2-(1,3-dioxoisoindolin-2yl)-1-hydroxyethyl)phosphonate 5 [(R)-12, 2.03 mmol, 721 mg, 1eq.] in ethanol (10 mL) and the mixture was stirred at 80 C for 2 h.The reaction mixture was filtered, concentrated in vacuo and the resulting residue was purified by column chromatography (Al 2 O 3 ; EtOH; Note that the compound could not be visualized using the ninhydrin stain described in the general experimental section in combination with Al 2 O 3 -coated TLC plates.Thus, SiO 2 -coated TLC plates were used to identify the product-containing fractions) giving (R)-diisopropyl 2-amino-1hydroxyethyl-phosphonate [(R)-14, 349 mg, 1.6 mmol, 80%] as a colorless oil. 1

Bioinformatics analyses
An initial visual inspection of bacterial AEP degradation gene clusters was performed using tools publicly available at the Department of Energy Integrated Microbial Genomes (IMG) database 75 and at the MicrobesOnline website. 74This preliminary analysis revealed the frequent presence of genes encoding putative FAD-dependent oxidoreductases in clusters for the hydrolytic degradation of AEP.The FAD enzymes belonged to at least three subgroups, exemplified by the representative sequences from Vibrio vulnificus (PbfB, GenBank: WP_049798008), Azospirillum sp.B510 (PbfC, GenBank: WP_012976454) and Acinetobacter baumannii (PbfD, GenBank: WP_079548425).Homologs of the three oxidoreductases were identified using a BLASTp search 76 against complete bacterial genomes available at the IMG.Only sequences that shared significant similarity to PbfB, PbfC or PbfD and whose genes were found within the expected genomic context (i.e.AEP degradation gene clusters) were considered for further analysis.
The phylogenetic relationships among PbfB, PbfC and PbfD, as well as between these enzymes and other functionally characterized oxidoreductases sharing a PF01266 domain, were investigated as follows.
Based on genomic searches, we assembled an initial set encompassing all the sequences containing a PF01266 domain that we identified within AEP degradation gene clusters (260 sequences).This initial set was reduced to 64 representative proteins, primarily by removing all sequences showing a pairwise identity higher than 90% with CD-HIT. 83The list of these 64 representative sequences (dataset A) is provided in Table S1.These sequences were aligned with MUSCLE 77 and the resulting multiple sequence alignment (MSA) was trimmed according to the boundaries of the PF01266 domain.Alignment positions that were poorly informative in terms of phylogeny, (e.g., sites where >50% of sequences in the alignment showed gaps) were removed with Gblocks. 84Following this curation step, the MSA was used as an input for a Maximum Likelihood analysis with IQ-TREE v.1.6.12. 78Radial unrooted trees were graphically depicted with the FigTree software.
A second, larger dataset (dataset B) comprising a total of 80 sequences was subsequently prepared by adding a custom selection of oxidoreductases of known function, which shared with PbfA, PbfB and PbfC the presence of the PF01266 domain.In addition to the 64 sequences from dataset A, dataset B contained: (i) two monomeric sarcosine oxidases (SoxA); (ii) two D-amino acid dehydrogenases (DadA); (iii) two g-N-methyl aminobutyrate oxidases (MabO); (iv) two glycine oxidases (ThiO); (v) two g-glutamylputrescine oxidases (PuuB); (vi) two N-methyl-L-tryptophan oxidases (SolA); (vii) two mitochondrial sarcosine dehydrogenases (SarDH); (viii) two mitochondrial dimethylglycine dehydrogenases (DMGDH).The accession IDs of these additional sequences are provided in Table S2.Dataset B was processed as described for dataset A, removing additional C-terminal domains prior to the preparation of the MSA.
ModelFinder 79 identified the WAG+F+R5 and LG+R5 molecular models of evolution as the most appropriate ones for dataset A and B, respectively, based on the Bayesian Information Criterion.The reliability of the two phylogenetic trees, with particular reference to the monophyly of the PbfA, PbfB and PbfC clades, was tested with 1000 ultrafast bootstrap replicates. 85

Expression and purification of recombinant PbfC
Related to Figure S9.PbfC from Azospirillum was overexpressed in E. coli Tuner (DE3) cells.A 10-mL bacterial preculture grown overnight at 37 C was used to inoculate 1L of Luria-Bertani broth (LB) containing kanamycin (50 mg/mL) until OD 600 reached 0.7-0.8.The temperature was then lowered to 20 C and protein expression was induced by adding 0.25 mM isopropyl-b-D-1-thiogalactopyranoside (IPTG).20 hours after induction, the cells were harvested by centrifugation.The cell pellets were washed once with phosphate-buffered saline solution (PBS), resuspended, centrifuged again, and finally stored at À20 C.
After sonication, the cell lysate was cleared by centrifugation (26,200 3 g for 40 minutes at 4 C).The supernatant was recovered and continuously looped (flow rate: 1 mL/min) on a 5 mL HisTrapä Fast Flow column (Cytiva) for 2 hours at 4 C through a peristaltic pump.Afterwards, the loaded column was mounted onto an AKTA Pure System FPLC apparatus and proteins were eluted with a gradient of buffer A supplemented with 208 mM imidazole.Fractions containing PbfC were pooled and extensively dialyzed against buffer A, supplemented with 10% glycerol and 10 mM FAD, then concentrated by an Amiconâ Ultra-2 3K device (Merck) to 3 mL.The protein was further loaded into a 5 mL loop of AKTA Pure System FPLC and purified by size exclusion chromatography (SEC) with a HiPrepä 16/60 Sephacryl S-300 HR (Cytiva) column, equilibrated with the same buffer as above (buffer A plus 10% glycerol and 10 mM FAD).The protein was finally concentrated by Amicon centrifugation and subdivided into aliquots (about 0.5 mL each) to be stored at À80 C. Protein purity was assessed higher than 95% by SDS-PAGE and the final yield was z3 mg of purified PbfC per liter of bacterial culture.A SDS-PAGE of purified PbfC (side by side with purified PbfD1 and PbfD2) is shown in Figure S9.

Expression and purification of recombinant PbfD1
Related to Figure S9.PbfD1 from Acinetobacter baumannii was expressed in E. coli Tuner (DE3) cells (EMD Biosciences), which were grown in self-inducing medium (LB containing glucose 0.5 g/L and lactose 2 g/L).A 10-mL bacterial preculture grown overnight at 37 C was used to inoculate 1L of self-inducing medium containing kanamycin (50 mg/mL).The culture was kept under agitation at 20 C for up to 24 h, then the cells were harvested by centrifuging at 7,200 3 g for 10 min at 4 C.
The pellet was resuspended in buffer B (50 mM sodium phosphate pH 7.5, 200 mM NaCl) and centrifuged again.The supernatant was discarded, while the pellets were stored at À20 C.
For protein purification, the pelleted cells were thawed on ice and resuspended in buffer B supplemented with 1 mM PMSF, 1 mM benzamidine, 10% glycerol and 10 mM FAD. Lysozyme (1 mg/mL final) was added to the cell suspension, which was then incubated at 4 C under agitation for 30 minutes and finally sonicated.At the end of sonication, the lysate was centrifuged at 26,200 3 g for 40 minutes (at 4 C) and the supernatant was used for the next purification step.
PbfD1 purification exploited a 5 mL HisTrapä Fast Flow column (Cytiva).The cleared lysate was continuously looped (flow rate: 2 mL/min) on the column for 2 hours at 4 C, after which the column was transferred to an AKTA Pure System FPLC apparatus.
The protein was eluted with a gradient of buffer B supplemented with 250 mM imidazole.PbfD1 fractions were pooled and extensively dialyzed against a storage buffer (50 mM TEA-HCl pH 7.5, 200 mM NaCl, 10% glycerol).The protein was subdivided into aliquots (about 0.5 mL each) to be stored at À80 C. Protein purity was assessed higher than 95% by SDS-PAGE and the final yield was z100 mg of purified PbfD1 per liter of bacterial culture.

Expression and purification of recombinant PbfD2
Related to Figure S9.Production of PbfD2 in a partially soluble form was achieved by expressing the protein in E. coli Chaperone Competent pGro7/BL21 cells (Takara) and exploiting D-sorbitol as a carbon source in the growth medium.This reportedly promotes the activation of an osmotic stress response helpful to increase protein solubility. 86 10-mL bacterial preculture grown at 37 C overnight was used to inoculate 1 L of self-inducing medium (LB containing glucose 0.5 g/L and lactose 2 g/L) supplemented with kanamycin (50 mg/mL), chloramphenicol (20 mg/mL), arabinose (0.5 mg/mL, to allow induction of chaperonins by the pGro7/BL21 cells) and D-sorbitol (5 mg/mL).The culture was kept under agitation at 20 C for up to 24 h, then the cells were harvested by centrifugation at 7,200 3 g for 10 min (4 C).After removing the supernatant, the pellet was resuspended in buffer B, then centrifuged again, discarding the supernatant.The pellets were then stored at À20 C waiting to proceed with lysis and purification.
For protein purification, pellets were thawed and resuspended in buffer B, supplemented with 1 mM PMSF, 1 mM benzamidine, 10% glycerol, 0.05% Tritonâ X-100 and 10 mM FAD.After adding lysozyme 1 mg/mL, the cell suspension was incubated in a cold room under agitation for 30 minutes.
The suspension was frozen in liquid N 2 and thawed quickly at 37 C in a thermostated water bath.The freezing-thawing procedure was repeated two more times, after which the suspension was sonicated briefly.At the end of sonication, the lysate was centrifuged at the maximum speed (26,200 3 g for 40 minutes at 4 C); the pellet was discarded and the supernatant was retained for further processing.
PbfD2 purification exploited also in this case a 5 mL HisTrapä Fast Flow column (Cytiva).The lysate was continuously looped (flow rate: 2 mL/min) on the column for 2 hours at 4 C, after which the column was transferred to the FPLC apparatus.
The column was first washed with 30 mL of chaperone-removal buffer (50 mM Na phosphate pH 7.5, 100 mM KCl, 10% glycerol, 500 mM sucrose, 20 mM MgCl 2 , 5 mM ATP) to displace the chaperonins bound to nickel resin; subsequently, PbfD2 was eluted with a gradient of buffer B supplemented with 250 mM imidazole.PbfD2 fractions were pooled and concentrated by an Amiconâ Ultra-2 3K device (Merck), then extensively dialyzed against a buffer containing 50 mM TEA pH 7.5, 200 mM NaCl and 10% glycerol.The protein was subdivided into aliquots (about 0.5 mL each) to be stored at À80 C. Protein purity was assessed higher than 80% by SDS-PAGE and the final yield was z2.5 mg of purified PbfD2 per liter of bacterial culture.

Absorbance measurements and establishment of extinction coefficients for PbfC, PbfD1 and PbfD2
Absorption spectra were recorded with either a Cary 50 (Varian) or with a thermostated V-750 (Jasco Inc.) UV-Vis spectrophotometer and corrected for buffer contribution.
Denaturation of the proteins with sodium dodecyl sulfate, aimed at identifying the type of cofactor (FAD or Flavin mononucleotide, FMN) and at calculating the extinction coefficients of protein-bound flavin, was carried out as described by Aliverti et al. 48Briefly, spectra of the enzymes were collected in TEA-HCl 50 mM pH 8.0, then SDS was added to a final concentration of 1% (w/v).The samples were monitored spectroscopically until no further changes were observed in the $280 and $450 nm regions.In all cases, the final spectra showed an absorption maximum at 450 nm, typical of FAD (rather than the 446 nm maximum of FMN).These spectra were used to calculate the concentration of released cofactor, using ε 450nm = 11300 M À1 cm À1 as described. 48This information was confronted with the spectra of the enzymes before SDS treatment, allowing us to estimate the extinction coefficients of the protein-bound FAD and hence (assuming a 1:1 enzyme:cofactor ratio) of the active oxidoreductase species.The calculated extinction coefficients (at the respective l max ) were: ε 462nm =10,650 M -1 cm -1 for PbfC (Azospirillum sp.B510); ε 447 =13,340 M -1 cm -1 for PbfD1 (A. baumannii) and ε 456 =11,950 M -1 cm -1 for PbfD2 (M.fucicola).In subsequent experiments, enzyme concentrations were calculated based on these coefficients.

Microtiter assays for oxidoreductase activities
The oxidoreductase activity of the recombinant enzymes was first assessed through two assays conducted in 96-well microtiter format.These colorimetric assays served to rapidly screen the ability of the enzymes to oxidize a number of potential substrates (both phosphonates and non-phosphonates), transferring the electrons to either an 'artificial' electron acceptor or to molecular oxygen.
In the first assay, the redox dye 2,6-dichlorophenol indophenol (DCPIP) was the final electron acceptor, whereas phenazine methylsulfate (PMS) served as an electron mediator. 43In this assay, the oxidoreductases (0.5 mM, final concentration) were incubated in 50 mM TEA-HCl pH 8.0, 200 mM DCPIP and 3 mM PMS, in the presence or absence of the potential substrate (5 mM; 10 mM for racemic mixtures).Reactions were conducted at room temperature in a final volume of 200 mL and started with the addition of the enzyme.The plate was photographed at regular intervals over a time of 30 min.
A single-point plate assay was used to screen for the ability of the recombinant enzymes to use molecular oxygen as the electron acceptor.The plate assay detected the release of hydrogen peroxide, which was revealed by exploiting horseradish peroxidase and the chromogenic substrate o-dianisidine. 87In each well, the enzyme under examination (0.5 mM, final concentration) was incubated in 50 mM TEA-HCl pH 8.0 with 5 mM substrate (10 mM for racemic mixtures).The reaction was started by the addition of the enzyme.After a 30-min incubation at room temperature, the reaction mixture was supplemented with horseradish peroxidase (3 U) and o-dianisidine (1 mM final concentration).After another 5 min, 3.5 M sulfuric acid was added to each well (to reach a final volume of 100 mL) and the plate was photographed.

Measurement of phosphate release
Phosphate release was also determined in a microtiter assay, by employing the BIOMOLâ Green kit according to the manufacturer's instructions.Enzymes that had been stored in phosphate or pyrophosphate buffer were dialyzed against a different buffer (50 mM TEA-HCl pH 8.0, 150 mM NaCl, 10% glycerol) before using them in these assays.Likewise, cell extracts to be subjected to this assay were obtained by lysing the E. coli cells in phosphate-free buffer (50 mM sodium 4-(2-hydroxyethyl)-1-piperazineethanesulfonate, pH 7.5).
A typical reaction contained 50 mM TEA-HCl buffer pH 8.0, the oxidoreductase under examination (or, in the case of PbfB, which resisted purification, the extracts from E. coli cells expressing the recombinant enzyme), 4 mM PhnX, 2 mM MgCl 2 and the substrate.After a one-hour incubation at room temperature, 18 mL of the reaction mixture (or 36 mL in the case of cell extracts) was transferred into 200 mL of BIOMOLâ Green reagent to block the enzyme activity.Color development was assessed after another 30 min, by measuring the absorbance at 600 nm using a Cary 50 UV-Vis spectrophotometer (Varian).

Detection of ammonia release from AEP oxidation
To investigate the potential release of ammonia upon the oxidative deamination of AEP (which contains a primary amino group), we utilized GDH to convert a-ketoglutarate to L-glutamate by using ammonia and oxidizing NADH, whose disappearance can be monitored spectroscopically at 340 nm.
For this assay, the oxidoreductase under examination (0.45 mM in the case of PbfD1 and PbfD2; 1.9 mM in the case of PbfC) was incubated in a solution containing 50 mM TEA-HCl buffer pH 8.0, 2 mM AEP and 1 mM a-ketoglutarate.After 30 min, the mixture was supplemented with NADH ($200 mM final concentration) and finally with GDH (1.7 U/ml final concentration).The disappearance of NADH absorption at 340 nm was monitored by a UV-Vis Cary 50 spectrophotometer.In control experiments where either the oxidoreductase or AEP was omitted, no significant NADH oxidation was observed.

Figure 1 .
Figure 1.Hydrolytic pathways for the microbial catabolism of AEP (A) In the simplest version of this pathway, the transaminase PhnW converts AEP to PAA, which in turn is cleaved into acetaldehyde and phosphate by PAA hydrolase (PhnX).(B)Variant of the pathway in A where PAA is first converted to phosphonoacetate and then to acetate and phosphate by the enzymes PhnY and PhnA, respectively.It should be noted that AEP can also be degraded through an ''oxidative'' pathway (totally unrelated to the pathways depicted here), which proceeds through the consecutive reactions of two dioxygenases (PhnY* and PhnZ) and yields as final products glycine and phosphate.[17][18][19]

Figure 2 .
Figure 2. Simplified maximum likelihood unrooted phylogenetic tree of 64 representative FAD-dependent enzymes encoded in gene clusters for the degradation of AEP The tree only reports bootstrap values for the main basal nodes and those supported by bootstrap values <50 were collapsed.Full bootstrap metrics, together with organisms' names, are reported in Figure S3, whereas the sequence accession IDs are provided in TableS1.The scale bar indicates the number of substitutions per site.Groups of sequences that we labeled PbfB, PbfC, and PbfD are highlighted in light green, light red and light blue, respectively.The monophyly of all three clades was extremely well supported (bootstrap values = 100), in line with the high sequence divergence observed between groups (i.e., median inter-group p-distances was equal to = 0.81).A subgroup of monophyletic early branching PbfD sequences, termed PbfD1, was also wellsupported (bootstrap value = 77) and is signaled by a darker shade of blue.pbfD1 genes usually clustered with phnW and phnX, whereas other pbfD genes were most commonly associated with phnX alone (see TableS1).Related to FiguresS1-S3and TableS1.

Scheme 1 .
Scheme 1.A proposal for the reaction catalyzed by the FAD-dependent enzymes described in this study

Figure 3 .
Figure 3. Microtiter assays to test the activity of PbfC, PbfD1, and PbfD2 toward a panel of potential (amino-containing) substratesIn the green/yellow wells, the potential substrates were incubated at room temperature in the presence (+) or absence (À) of the oxidoreductases (0.5 mM); the mixture also contained the redox dye DCPIP (which served as the final electron acceptor) and the redox mediator PMS.Reactions were started by adding the enzyme last and photographed at regular intervals-the images in this figure were taken 6 min after adding the enzyme.In wells where the reaction occurred, the solution turned from green to yellow.The transparent/pink wells refer instead to an endpoint microtiter assay aimed at detecting the release of hydrogen peroxide.The reactions were conducted as aforementioned, except that artificial dyes were omitted and molecular oxygen served as the electron acceptor.After 30 min, the wells were supplemented with horseradish peroxidase and o-dianisidine and (after another 5 min) sulfuric acid.Wells where hydrogen peroxide had formed turned deep pink.Abbreviations: PnAla, 3-phosphono-DL-alanine; 3APP, 3-aminopropylphosphonate; PEA, phosphoethanolamine.Related to FiguresS6 and S7.

Figure 4 .
Figure 4. Spectroscopic and functional properties of PbfC, PbfD1, and PbfD2 (A-C) UV-vis absorption spectra of the three enzymes (10 mM) before (black line) and after (gray line) the addition of 20 mM M 1 AEP.Spectra were collected at room temperature under aerobic conditions in 50 mM pyrophosphate buffer pH 8.0 (PbfC) or 50 mM phosphate buffer pH 8.0 (PbfD1 and PbfD2).(D-F) Relative activities of the three enzymes in the presence of 5 mM M 1 AEP, measured at room temperature through the assay based on DCPIP-PMS (lime yellow bar) or through the coupled assay with PhnX and ADH (magenta bar).The error bars represent standard errors for triplicate experiments.Activity in the assay that gave the maximum rate was taken as 100%.Both types of assays were conducted in TEA-HCl buffer, pH 8.0.(G-I) Activity of the three enzymes toward naturally occurring substrates that had shown some reactivity in the microtiter assays (TEA-HCl buffer pH 8.0, 25 C).The activity of PbfC, was measured with the DCPIP reduction assay whereas the activities of PbfD1 and PbfD2 were measured under atmospheric oxygen through the PhnX-ADH coupled assay.Each data point represents a triplicate experiments; bars signal standard errors.Solid lines represent fittings of the data to the Michaelis-Menten equation.The fitting-derived kinetic parameters are reported in Table2.

Figure 5 .
Figure 5. Evidence for M 1 AEP degradation by soluble extracts of E. coli cells expressing PbfB (A) Phosphate release assay.Cultures (10 mL) of pGro7/BL21 cells containing either the pET28-PbfB plasmid or an empty pET28 vector were induced with 1 mM IPTG overnight (20 C), then harvested, resuspended in 2 mL of HEPES buffer pH 7.5 and homogenized by sonication.A part of the soluble extract from PbfBexpressing cells was heated for 10 min at 100 C, as a control.Extracts (40 mL) were subsequently incubated in 100 mL (final volume) of 50 mM TEA-HCl buffer pH 8.0 containing 2 mM MgCl 2 and (when indicated) 4 mM PhnX and 5 mM M 1 AEP or AEP.After 1 h at room temperature, 36 mL of the reaction mixture was transferred into 200 mL of BIOMOL Green reagent.Color development was read after $30 min.(B) A partial repetition of the phosphate release assay (upper row) performed on a different batch of cell extract and conducted in parallel with a DCPIP reduction assay (lower row).Wells in the lower row contained 200 mM DCPIP, 3 mM PMS, and (when indicated) 5 mM substrate (M 1 AEP or AEP) and 15 mL cell extract in a total volume of 100 mL.PhnX and MgCl 2 were consistently omitted for the assays in the lower row.

Figure 6 .
Figure 6.A schematic, updated summary of the different possible pathways for the hydrolytic degradation of AEP and related compoundsThe newly identified enzymes described in this work are highlighted in red.In the oxidation of M 1 AEP the electron acceptor is not indicated for simplicity; as detailed in this work, the acceptor is O 2 for PbfD enzymes, but unknown for PbfC and PbfB.The enzyme catalyzing the oxidative deamination of AEP is indicated as PbfD2 by approximation.Indeed, most of the genes of the pbfD2 subgroup belong to clusters that do not include phnW, but there are exceptions; at the same time, a minority of pbfD genes outside the pbfD2 subgroup localizes in clusters that lack phnW (see TableS1for examples).

Table 1 .
Percent identities between representative PbfB, PbfC, and PbfD enzymes and the most similar enzymes of known function