Thioproline formation as a driver of formaldehyde toxicity in Escherichia coli

Formaldehyde (HCHO) is a reactive carbonyl compound that formylates and cross-links proteins, DNA, and small molecules. It is of specific concern as a toxic intermediate in the design of engineered pathways involving methanol oxidation or formate reduction. The high interest in engineering these pathways is not, however, matched by engineering-relevant information on precisely why HCHO is toxic or on what damage-control mechanisms cells deploy to manage HCHO toxicity. The only well-defined mechanism for managing HCHO toxicity is formaldehyde dehydrogenase-mediated oxidation to formate, which is counterproductive if HCHO is a desired pathway intermediate. We therefore sought alternative HCHO damage-control mechanisms via comparative genomic analysis. This analysis associated homologs of the Escherichia coli pepP gene with HCHO-related one-carbon metabolism. Furthermore, deleting pepP increased the sensitivity of E. coli to supplied HCHO but not other carbonyl compounds. PepP is a proline aminopeptidase that cleaves peptides of the general formula X-Pro-Y, yielding X + Pro-Y. HCHO is known to react spontaneously with cysteine to form the close proline analog thioproline (thiazolidine-4-carboxylate), which is incorporated into proteins and hence into proteolytic peptides. We therefore hypothesized that thioproline-containing peptides are toxic and that PepP cleaves these aberrant peptides. Supporting this hypothesis, PepP cleaved the model peptide Ala-thioproline-Ala as efficiently as Ala-Pro-Ala in vitro and in vivo, and deleting pepP increased sensitivity to supplied thioproline. Our data thus (i) provide biochemical genetic evidence that thioproline formation contributes substantially to HCHO toxicity and (ii) make PepP a candidate damage-control enzyme for engineered pathways having HCHO as an intermediate.

The toxicity and mutagenicity of HCHO lie in its ability to react spontaneously with amino and thiol groups. With amines such as the 6-amino group of peptidyl lysine, HCHO forms a carbinolamine that can either be oxidized to an N-formyl derivative or undergo dehydration, yielding a labile Schiff base that can form cross-links (methylene bridges) with various amino acids and nucleobases [8,9] (Figure   1). With thiols, notably free and peptidyl cysteine, HCHO forms the cyclic derivative thioproline (thiazolidine-4-carboxylate) [10,11] (Figure 1). All of these reactions except oxidation and cross-linking are reversible, with the HCHO condensation direction being thermodynamically preferred. Although it is clear that the above general types of damage reactions underlie the toxic and mutagenic effects of HCHO, the specific reactions that actually drive these effects and their relative importance in vivo remain poorly defined [10]. The least poorly defined toxic effect is probably the N 6 -formylation of lysine residues in certain histones, which is inferred to impair histone function [12].
The archetypal defense against HCHO toxicity is to oxidize HCHO to non-toxic formate via glutathionedependent or -independent dehydrogenases [1][2][3] (Figure 1). This detoxification strategy is counterproductive if HCHO is a pathway substrate, as in methylotrophic or formatotrophic pathways, because it leads to loss of the substrate. Cells operating methylotrophic or formatotrophic pathways must therefore cope with HCHO because they cannot simply detoxify it by dehydrogenation [13]. A major coping mechanism is fine-tuning the expression of HCHO-producing and -consuming enzymes [13,14], but this delicate flux-balancing act entails non-negligible steady-state HCHO levels as well as possible HCHO spikes due to upshifts in methanol availability. It is consequently worth identifying other mechanisms that cells use to control HCHO damage. The only known damage-control mechanism of this type appears to be deformylation of N 6 -formyl lysine residues by bacterial and mammalian sirtuins [15].
The research we report here was based on two ideas: that there are HCHO damage-control enzymes left to discover [16] and that such enzymes are valuable components to have in hand when designing and building metabolic pathways that involve HCHO [17]. After applying comparative genomics to identify a candidate HCHO damage-control enzyme, we obtained genetic and biochemical evidence that Escherichia coli PepP, an Xaa-Pro aminopeptidase, mitigates downstream consequences of the presence of HCHO-generated thioproline residues in proteins. That deleting pepP increases HCHO sensitivity provides genetic evidence that thioproline formation is one of the drivers of HCHO toxicity.

Bioinformatics
Comparative genomics analyses used STRING [18] (https://string-db.org/) and SEED [19] (http://pubseed.theseed.org/) databases and tools. The guide genes used to search for candidate HCHO damage-control genes were those mediating reactions near HCHO in KEGG maps 00680 (methane metabolism) and 00670 (one carbon pool by folate). Full results of the SEED analysis are encoded in the SEED subsystem named 'PepP-C1'. Phylogenetic trees were drawn with MEGA6 [20]. F0136). Formylated Boc-L-Lys-AMC was synthesized from Boc-L-Lys-AMC (Peptide Solutions LLC, Tucson, AZ, Cat. No. BAA3650.0100) using a modification of a published procedure [21]. To a solution of Boc-L-Lys-AMC (50 mg, 0.12 mmol, 1.0 equiv.) and a catalytic amount of DIPEA (0.01 mmol) in dry dimethylformamide (2 ml) was added 2,2,2-trifluoroethyl formate (61 mg, 0.48 mmol, 4.0 equiv.) and the reaction mixture was stirred at 22°C for 12 h. The mixture was concentrated in vacuo and purified by silica gel column chromatography (CH2Cl2:MeOH, 20/1 v/v) to give the product (36 mg, 67%). 1  glucose. The precultures were harvested and washed three times in M9 medium, then inoculated in M9 medium containing 10 mM glucose plus specified additions, with a starting OD600 of 0.005. One hundred and fifty μl of culture was added to each well of 96-well microplates (Nunclon Delta Surface, Thermo Scientific), then 50 μl mineral oil (Sigma-Aldrich) was added to each well to avoid evaporation (while enabling gas diffusion). The 96-well microplates were incubated at 37°C in a microplate reader (BioTek EPOCH 2). The shaking program cycle (controlled by Gen5 v3) had four shaking phases, lasting 60 s each: linear shaking followed by orbital shaking, both at an amplitude of 3 mm, then linear shaking followed by orbital shaking both at an amplitude of 2 mm. The OD600 in each well was monitored and recorded after every three shaking cycles (~16.5 min). Raw data from the plate reader were calibrated to normal cuvette-measured OD600 values according to ODcuvette = ODplate/0.23.

Chemicals
For HCHO insult experiments, 50 ml of M9 medium was inoculated with BW25113 and ΔpepP overnight precultures to a starting OD600 of 0.01 and incubated with shaking at 37°C until OD600 reached 0.2. Pilot experiments showed that adding HCHO to cultures to a final concentration of 1 mM caused a cessation of growth after which BW25113 and ΔpepP strains resumed growth after 13 h and 20 h, respectively. For untreated samples, cells were collected by centrifugation at OD600 of 0.2, flash frozen and stored at -80°C; HCHO-treated samples were similarly collected 2 h after the insult began.
M9 medium plates containing 1% agar were used to test growth on solid medium. When applicable, 400 μM formaldehyde (or 60 mM acetaldehyde, 5 mM glyoxal, or 0.5 mM methylglyoxal) were added to the medium after cooling to <45°C. Overnight M9 medium precultures were used to inoculate (1:100) 2-ml M9 medium cultures, which were grown at 37°C until OD600 of 1.0 before ten-fold serially diluting and spotting 3.5-μl aliquots on the plates. Plates were incubated at 37C for 16-20 h before imaging.
To test Ala-Pro-Al or Ala-thioproline-Ala as sole nitrogen source, M9 plates were prepared as above except that NH4Cl was omitted. Overnight M9 medium precultures of BW25113 and Keio ΔpepP strains were used to inoculate (1:100) 10-ml M9 medium cultures, which were grown at 37°C until OD600 of ~1.0. Cells from each strain were 10-fold diluted into water agar solution (0.75% w/v) cooled to ~40°C, and 5 ml was overlaid on plates prewarmed to 37°C. Once cooled, 10-mm disks of sterile No. 2 Whatman filter paper were imbibed with 10 μl of 25 mg/ml alanine or 50 mg/ml tripeptide solution, then applied to the agar-overlaid plates. The plates were incubated at 37°C for 24 h before imaging.

Production and purification of recombinant PepP
For PepP protein expression, the coding sequence was PCR-amplified (primers 3 and 4) from wild type BW25113 genomic DNA, adding an N-terminal His6-tag as well as 3ʹ and 5ʹ sequences complementary to the pET15b multiple cloning site. Following pET15b PCR amplification (primers 5 and 6), the vector and insert amplicons were gel-purified (GeneJet Gel Extraction Kit, Thermo Scientific Cat No. K0691) according to the manufacturer's instructions and inserted using circular polymerase extension cloning [24]. To produce recombinant PepP protein, E. coli BL21 (DE3) RIPL harboring the expression plasmid was grown in 100 ml LB medium at 37°C until an OD600 of 0.8, after which the cultures were supplemented with 1.0 mM isopropyl-β-D-thiogalactoside and incubated a further 2 h at 22°C. Cells were collected by centrifugation, frozen in liquid N2, and stored at -80°C. Pellets were resuspended in 1 ml ice-cold lysis buffer (50 mM potassium phosphate, pH 8.0, 300 mM NaCl, and 10 mM imidazole), and sonicated (Fisher Scientific Ultrasonic Dismembrator, model 150E) using five 12s pulses at 70% power, cooling on ice for 30 s between pulses. Following centrifugation (14 000 g, 5 min, 4°C), the supernatant was applied to a 0.1 ml column of Ni 2+ -NTA Superflow resin (Qiagen Cat No. 30410) and protein was purified using the manufacturer's protocol. The purified protein was desalted using a PD-25 column (GE Healthcare), eluting with 50 mM potassium phosphate, pH 7.5, containing 10% (v/v) glycerol; aliquots (50 μl) were frozen in liquid N2 and stored at -80°C. Protein concentration was determined by Bradford dye-binding assays [25].

Enzyme assays
For all activity assays, PepP protein was pre-incubated with 0.

LC-MS analysis of protein-bound N-formyl lysine
Cell pellets were resuspended in 1 ml PBS buffer (10 mM sodium phosphate, pH 7.4, 137 mM NaCl, 2.7 mM KCl), sonicated as above, and centrifuged to clear. Soluble proteins were precipitated in 80% (v/v) acetone and harvested by centrifugation (5000 g, 3 min, 4C); the pellets were air-dried overnight at 22C and stored at -20C. The pelleted proteins were mixed with 200 μl 1 M NaOH and heated at 50C until dissolved (briefly vortexing as necessary) before adding 400 μl 100 mM ammonium bicarbonate buffer, pH 8.5 (brought to pH 8.5 with 5 M HCl). Samples were then desalted on PD-25 columns, eluting with 100 mM ammonium bicarbonate buffer, pH 8.5, and protein content was determined as above. To digest proteins to free amino acids, Streptomyces griseus protease [11] was added (5 μg  Table S2). To resuspend, solvent was added, samples were vortexed for 30 s, sonicated for 5 m, vortexed again for 30 s, centrifuged (16000 g, 2 min, room temperature), and transferred to anLC-MS vial with glass insert. Samples were stored at 4°C in an autosampler and analyzed using a Vanquish LC system (Thermo Fisher Scientific), coupled to a 6500+ QTrap mass spectrometer (SCIEX) operated in multiple reaction monitoring (MRM) mode. Standard curves of analytes were diluted in LC-MS grade 80:20 acetonitrile/water (v/v), dried under vacuum, and resuspended in the same run solvent as samples. A Waters Acquity UPLC BEH Amide column (150 mm × 2.1 mm, 1.7 μm particle size) with an Acquity VanGuard BEH Amide pre-column (5 mm × 2.1 mm, 1.7 μm particle size) was held at 45°C throughout analysis. Mobile phase (A) was LC-MS grade water and (B) was 95:5 LC-MS grade acetonitrile/ water (v/v). Both mobile phases were modified to 10 mM ammonium formate and 0.125% formic acid. The mobile phase flow was 0.4 ml/min and the gradient was 100% B from 0-2 min, brought to 70% B between 2 and 7.7 min, brought to 40% B between 7.7 and 9.5 min, returned to 100% B between 9.5 and 12.75 min and maintained at 100% B until 17 min. Five µl of sample was injected and needle wash solution was 1:1 acetonitrile/water (v/v).
Collision gas was set to "Medium", IonSpray voltage was 5500 V and temperature was 250°C. Linear 5-7 point calibration curves of the ratio of analyte to IS with 1/x weighting were fit for lysine (IS: D8 lysine), N 6 -formyl lysine (IS: D5 threonine), and cystine (IS: D8 lysine) in MultiQuant version 3.0.3 (SCIEX). The background rate of N 6 -formylation of lysine during analysis was 0.03% determined from a calibration curve of pure lysine standard spiked into the matrix. Background formylation was corrected for by subtracting 0.0003*[lysine] from the measured concentrations of N 6 -formyl lysine in samples.

Comparative genomics links PepP with one-carbon metabolism
We first used comparative genomics analysis [26] with the STRING [18] and SEED [19] databases to find candidate genes that could play a role in controlling HCHO damage. We searched for genes that are consistently clustered on the chromosome with genes of one-carbon metabolism, in which HCHO is central. These one-carbon metabolism 'guide genes' included genes encoding enzymes that produce or consume HCHO (e.g. methanol and formaldehyde dehydrogenases) and genes of tetrahydrofolate (THF)-mediated one-carbon metabolism, which is both a disposal route for HCHO and a potential HCHO source via spontaneous dissociation of 5,10-methylene-THF [27]. The analysis detected genes encoding homologs of E. coli aminopeptidase PepP clustered with genes encoding subunits of the glycine cleavage complex, which generates a THF-bound HCHO unit [28], and genes for 5-formyl-THF cyclo-ligase, which recycles a one-carbon folate byproduct of serine hydroxymethyltransferase, another enzyme that generates a THF-bound HCHO unit [29] (Figure 2A). These clustering arrangements are unlikely to be due to chance alone because they involve different HCHO-related genes and occur in bacteria from four different phyla. Moreover, the clustering specifically involves the PepP aminopeptidase family, but not related aminopeptidase families such as YpfF ( Figure 2B).

PepP contributes to formaldehyde resistance in E. coli
The above genomic evidence linking PepP with one-carbon metabolism led us to compare the HCHO resistance of an E. coli pepP deletion strain with that of the parent BW25113 strain (henceforth referred to as wild type). The ΔpepP strain was markedly more sensitive to HCHO at high concentrations (400-500 μM) when cultured in liquid or solid medium ( Figure 3A,B). To confirm that the pepP deletion causes the HCHO-sensitive phenotype (i.e., that no mutations elsewhere in the genome are involved) we complemented the ΔpepP strain with a plasmid-borne E. coli pepP gene (Fig. 3A). The methanol stabilizer present in the HCHO reagent (37% HCHO, 10-15% methanol) did not detectably affect the growth of the wild type or ΔpepP strains when added to plates at the concentration that would have accompanied the HCHO.
To determine whether PepP impacts resistance to carbonyl compounds besides HCHO, we compared the growth of the parent and ΔpepP strains in the presence of inhibitory concentrations of acetaldehyde, glyoxal, or methylglyoxal. The ΔpepP deletant showed little or no increase in sensitivity to these carbonyls (Supplementary Figure S1), showing that PepP action is essentially specific to HCHO. We also tested whether the HCHO-sensitivity of the ΔpepP strain depends on the nature or concentration of the carbon source. Similar sensitivity was observed on eight diverse carbon sources, which included fermentable and non-fermentable substrates taken up by the phosphotransferase system or other carriers (Supplementary Figure S2). Sensitivity was also similar at various concentrations of glucose (Supplementary Figure S3). The HCHO-sensitivity of the ΔpepP strain is thus a robust phenotype.

PepP does not deformylate N 6 -formyl lysine in vitro or in vivo
E. coli PepP (Xaa-Pro aminopeptidase, EC 3.4.11.9) cleaves the N-terminal amino acid from a wide range of peptides whose second residue is proline; the reaction catalyzed is thus X-Pro-Yn → X + Pro-Yn (where n can be up to ~10) [30]. Because N 6 -formylation of lysine residues is a known HCHOdriven protein damage reaction [11,12] and because some peptidases have amidase activity [31,32], we tested whether PepP cleaves the amide bond in N 6 -formyl lysine (i.e. deformylates lysine) in vitro or in vivo. For in vitro tests, N 6 -formyl lysine and its N-and C-terminally blocked derivative Boc-N 6formyl Lys-AMC were incubated with recombinant E. coli PepP and a coupled spectrophotometric assay was used to measure formate release. No activity was detected with either substrate. The activity detection limit (0.6 nmol mg -1 protein min -1 ) was <0.01% of the activity of PepP against the canonical peptide substrate Ala-Pro-Ala (see below). To extend the study from model substrates to proteins in vivo we measured N 6 -formyl lysine levels in proteins extracted from wild type and ΔpepP cells before and after treatment with a nonlethal, bacteriostatic concentration of HCHO (Figure 4). HCHO-treated wild type and ΔpepP cells had modestly higher N 6 -formyl lysine levels than untreated cells, consistent with the known formylating activity of HCHO [11,12]. However, N 6 -formyl lysine levels in ΔpepP cells did not differ significantly from those in wild type cells either before HCHO treatment (when cells were growing) or after HCHO treatment (when cells were not growing). The lack of effect of deleting pepP agrees with the in vitro evidence that PepP does not have lysine-deformylating activity and hence confirms that this activity cannot explain the role of PepP in HCHO resistance.

A deductive hypothesis connecting PepP to formaldehyde toxicity
It is not obvious how an aminopeptidase like PepP could impact HCHO toxicity unless it has substantial deformylase activity, which the above evidence indicates it does not. However, a plausible multi-part hypothesis can be proposed, based on five facts: (i) thioproline is a close analog of proline, with similar bond lengths and angles [33]; (ii) thioproline forms spontaneously in vivo from the condensation of HCHO and free or protein-bound cysteine [10,11]; (iii) thioproline is charged on tRNA Pro and enters proteins [33,34]; (iv) PepP is specific for proline as the second residue in its peptide substrates [30]; and (v) free thioproline is detoxified via proline dehydrogenase (PutA)-mediated oxidation to 2,3thiazoline-4-carboxylate followed by hydrolysis to cysteine plus formate [35,36]. Our hypothesis is as follows: HCHO buildup causes formation of thioproline-containing proteins whose degradation produc-es thioproline-containing peptides; these peptides are toxic; PepP is necessary for hydrolysis of these peptides, enabling release of free thioproline that is safely metabolized via PutA to cysteine and formate ( Figure 5A). The core of the hypothesis ( Figure 5A, box) is that thioproline peptides are toxic and that PepP is essential for their removal.
This hypothesis explains why deleting PepP does not increase sensitivity to acetaldehyde, glyoxal, or methylglyoxal (Supplementary Figure S1). Acetaldehyde reacts with free cysteine to form the thioproline analog 2-methylthioproline [37] but this amino acid is almost surely not charged to tRNA and incorporated into proteins [38], and acetaldehyde has not been reported to react with peptidyl-cysteine; hence there would be no 2-methylthioproline-containing peptides for PepP to cleave. Glyoxal and methylglyoxal cannot react with cysteine to form thiazolidine adducts (i.e. thioproline-type derivatives) [39], so again there would be no thioproline-like peptides for PepP to act on.
The hypothesis also makes two testable predictions, namely that: (i) PepP readily hydrolyzes peptides with the general formula X-thioproline-Y to X + thioproline-Y (X and Y being any amino acid); and (ii) deleting pepP increases sensitivity to supplied thioproline because this is incorporated into proteins [34] that are cleaved to peptides that are toxic. Our next step was to test these predictions.

Experimental tests of hypothesis predictions
We first tested the activity of purified recombinant PepP against the thioproline-containing peptide Alathioproline-Ala using its natural counterpart Ala-Pro-Ala as a benchmark (Ala-Pro-Ala is known to be a good PepP substrate [30,40]). Activity was assayed via a coupled spectrophotometric assay that measured release of alanine. As pilot experiments showed similar activity against both peptides, we proceeded to a full kinetic characterization (Table 1 and Supplementary Figure S4). Our Km and kcat values for the natural substrate Ala-Pro-Ala were consistent with those previously reported for this peptide [40]. As predicted, Ala-thioproline-Ala was a good substrate, giving Km and kcat values that did not differ statistically from those for Ala-Pro-Ala. More generally, the Km and kcat values for Ala-thioproline-Ala were within the range reported for PepP acting on various peptides [30]. PepP thus acted on a model thioproline-containing peptide just as if it were a regular proline-containing peptide.
To confirm that PepP cleaves Ala-thioproline-Ala in vivo as well as in vitro we exploited the ability of E. coli to use alanine as sole nitrogen source [41]. We reasoned that if PepP is the sole or main peptidase that cleaves the N-terminal amino acid from X-thioproline-Y and X-proline-Y peptides, and if such peptides are not hydrolyzed by carboxypeptidase activity [42], then wild type E. coli should be able to use Ala-thioproline-Ala or Ala-Pro-Ala as sole nitrogen source and the ΔpepP strain should not. This proved to be the case ( Figure 5B). Plates containing a lawn of cells and a central disc charged with either tripeptide gave growth halos around the disc with wild type cells but not with ΔpepP cells.
Comparable results were obtained in liquid culture, i.e. the ΔpepP deletion strain lost most of its cap-acity to use Ala-thioproline-Ala or Ala-Pro-Ala as sole nitrogen source ( Figure 5C). The residual growth of the ΔpepP deletant was presumably due to a low level of alanine release by another peptidase(s).
Note that the similar growth patterns of wild type and ΔpepP cells on Ala-Pro-Ala or Ala-thioproline-Ala show that Ala-thioproline-Ala was not toxic at the concentration used (2.5 mM). Finally, we compared the responses of the wild type and ΔpepP strains to supplied thioproline. As predicted, the ΔpepP strain was more sensitive, becoming progressively more inhibited than wild type as the thioproline concentration was increased from 2 mM to 4 mM ( Figure 6).

Ancillary experimental tests
Given the role of PutA in detoxifying thioproline ( Figure 5A) [35], we compared the effect of deleting putA on sensitivity to HCHO to that of deleting pepP, using a concentration of HCHO (400 µM) that inhibits growth of the ΔpepP strain (Supplementary Figure S5). At this HCHO concentration, the ΔputA strain showed no growth defect. This result implies that the thioproline-containing peptides that PepP hydrolyzes are more toxic than thioproline itself. PepP thus appears to be more critical than PutA in the chain of reactions that cope with protein-bound thioproline ( Figure 5A).
Because glutathione is ~100-fold more abundant in E. coli than free cysteine [43,44] and reacts readily with HCHO to form S-(hydroxymethyl)glutathione and other adducts [45] we tested whether PepP can hydrolyze the adduct γ-Glu-thioproline-Gly. This compound could conceivably form by spontaneous reaction of HCHO with glutathione's cysteine residue, although it is not one of the reported HCHO adducts of glutathione [45]. γ-Glu-thioproline-Gly was incubated with PepP and glutamate release was assayed by a coupled spectrophotometric procedure. No activity was detected; the detection limit (2 nmol mg -1 protein min -1 ) was <0.01% of the activity of PepP against Ala-Pro-Ala.

The Keio collection ΔyhbO strain carries a pepP nonsense mutation
In the course of this work we found by genome sequencing that the Keio collection ΔyhbO strain carries a nonsense mutation in the pepP ORF that changes codon 285 from CAG (glutamine) to TAG (stop). The resulting truncated PepP protein is evidently non-functional because the Keio ΔyhbO strain had the same HCHO sensitivity as the ΔpepP strain (Supplementary Figure S6). A ΔyhbO strain lacking the pepP nonsense mutation (constructed by P1 phage transduction from the BW25113 Keio parent strain) showed no increase in HCHO sensitivity (Supplementary Figure S6), confirming that the HCHO sensitivity of the pepP-ΔyhbO strain is due solely to the PepP defect.

Conclusions
The core parts of our hypothesis connecting the Xaa-Pro aminopeptidase PepP with HCHO resistance are (i) that thioproline-containing peptides (at least those with thioproline as second residue) are toxic, and (ii) that PepP can cleave Xaa-thioproline peptides ( Figure 5A). Part (ii) is validated by the evidence presented above. Part (i) is a strong inference -and will likely remain so because it is so challenging to test. With thioproline at position two, there are 20 2 (400) possible tripeptides, 20 3 (8,000) possible tetrapeptides, and so forth up to 20 9 (512 billion) possible decapeptides. Assuming, as for other peptides [46,47], that toxicity is restricted to a tiny fraction of the total sequence space, it would be necessary to screen an intractable number of peptides of various sizes and sequences in order to identify the few that are toxic. Moreover, these peptides would have to be supplied exogenously but might not be taken up because bacterial peptide transporters have various degrees of substrate specificity [48], or might be hydrolyzed before uptake by peptidases released by dead cells [49].
Although toxicity testing of thioproline-containing peptides is infeasible, it is interesting to ask what the basis of their toxicity could be. We suggest the following general explanation, based on the findings that certain non-natural oligopeptides inhibit E. coli growth [50] and that thioproline-containing polypeptides can differ greatly in physicochemical properties from their natural proline-containing counterparts [34]. Perhaps certain thioproline-containing peptides simply differ enough from their harmless natural analogs to be toxic via the same undefined mechanisms as other alien oligopeptides [50].
In sum, our genetic and biochemical findings on PepP provide strong evidence that the formation of thioproline-containing peptides is a substantial driver of HCHO toxicity. From an engineering standpoint, this evidence makes PepP a potential damage-control component [17] for use in designing and building pathways in which HCHO is an intermediate (Figure 1). Although the damage-control pathway involving PepP ( Figure 5A) ultimately converts HCHO to formate, the flux via this pathway is constrained by the rate of proteolysis and is hence almost certainly too low to constitute a major drain on the HCHO pool. From a physiological standpoint, the evidence that PepP deals with a toxic consequence of HCHO formation -which happens in normal metabolism [1-3] -raises a question: Is cleaving thioproline peptides the primary role of PepP? If so, the well-characterized activity of PepP against proline peptides would be of secondary importance. That the ΔpepP strain has no growth defect on minimal medium ( Figure 3A

Table 1 Kinetic parameters of recombinant E. coli PepP
A coupled spectrophotometric assay was used. Values are means ± s.e. (n = 3). The Km value for Ala-Pro-Ala is close to the reported value (0.77 mM) [40] and the kcat value is consistent with that reported previously (85 s -1 ) [40], allowing for differences in assay pH and temperature. All values for Ala-thioproline-Ala fall in the 95% confidence intervals for Ala-Pro-Ala values, i.e. are statistically the same.
(A) PepP (Xaa-Pro aminopeptidase) homologs from four bacterial phyla cluster on the chromosome with genes encoding enzymes of one-carbon metabolism: 5-formyl-THF cyclo-ligase (fau) and the T, H, and P subunits of the glycine cleavage complex (gcvT, gcvH, and gcvP, respectively). Gray genes encode proteins whose function is either unrelated to one-carbon metabolism or unknown.  (A) E. coli BW25113 wild type (WT) and ΔpepP strains harboring the pBAD24 vector alone (V) or containing pepP (P) were cultured on M9 minimal medium containing 0.4% glucose and 0.02% arabinose, minus or plus 400 μM HCHO. Overnight liquid cultures of each strain were ten-fold serially diluted and 3.5-μl aliquots were spotted on the plates. Images were captured after incubation at 37°C for 1.5 d.  E. coli BW25113 wild type (WT) and ΔpepP strains were grown in M9 minimal medium containing 0.4% glucose until OD 600 reached 0.4. Control (-HCHO) cells were then harvested; HCHO-treated (+ HCHO) cells were cultured for another 2 h after adding HCHO (final concentration 1 mM) and then harvested. Extracted proteins were hydrolyzed with S. griseus protease; the hydrolysate was analyzed by LC-MS for N 6 -formyl lysine. Data are means ± s.e. (n = 6). Significance was determined by Student's t-test. *P <0.05; ns, non-significant.