Early midcell localization of Escherichia coli PBP4 supports the function of peptidoglycan amidases

Insertion of new material into the Escherichia coli peptidoglycan (PG) sacculus between the cytoplasmic membrane and the outer membrane requires a well-organized balance between synthetic and hydrolytic activities to maintain cell shape and avoid lysis. Since most bacteria carry multiple enzymes carrying the same type of PG hydrolytic activity, we know little about the specific function of given enzymes. Here we show that the DD-carboxy/endopeptidase PBP4 localizes in a PBP1A/LpoA and FtsEX dependent fashion at midcell during septal PG synthesis. Midcell localization of PBP4 requires its non-catalytic domain 3 of unknown function, but not the activity of PBP4 or FtsE. Microscale thermophoresis with isolated proteins shows that PBP4 interacts with NlpI and the FtsEX-interacting protein EnvC, an activator of amidases AmiA and AmiB, which are needed to generate denuded glycan strands to recruit the initiator of septal PG synthesis, FtsN. The domain 3 of PBP4 is needed for the interaction with NlpI and EnvC, but not PBP1A or LpoA. In vivo crosslinking experiments confirm the interaction of PBP4 with PBP1A and LpoA. We propose that the interaction of PBP4 with EnvC, whilst not absolutely necessary for mid-cell recruitment of either protein, coordinates the activities of PBP4 and the amidases, which affects the formation of denuded glycan strands that attract FtsN. Consistent with this model, we found that the divisome assembly at midcell was premature in cells lacking PBP4, illustrating how the complexity of interactions affect the timing of cell division initiation.


Introduction
The peptidoglycan (PG) layer is sandwiched between the cytoplasmic membrane (CM) and the outer membrane (OM) of the Gram-negative bacterium Escherichia coli forming a covalently closed network of glycan strands that are interconnected by short peptides [1]. The PG layer maintains the shape of the bacterium by stabilizing the cell against its high internal osmotic pressure. To proliferate, the rod-shaped bacterium grows in length and then divides by binary fission into two equally sized daughter cells, with nascent PG inserted into the lateral cell wall and septum.
Binary fission in E. coli is initiated by the assembly of the Z-ring at midcell, which consists of a network of dynamic FtsZ filaments that are anchored to the cytoplasmic membrane by FtsA and ZipA [14,15]. The Z-ring also recruits the PG synthases PBP1A and PBP1B to midcell [2,16]. A number of Z-associated proteins play a role in the organization and dynamics of the ring [17][18][19][20][21][22]. Of these, the integral membrane protein FtsX interacts with FtsA [23] and its cytoplasmic ATPase partner protein FtsE interacts with FtsZ [24]. FtsE and FtsX localize early to midcell, together with the Z-ring [25]. The recruitment of the other cell division proteins occurs with a time delay of about 20% of the cell division cycle, depending on the growth condition [26][27][28]. FtsK, FtsBLQ, FtsW, PBP3 and FtsN localize in an interdependent fashion. The FtsBLQ complex inhibits the PG synthase activities of PBP3, FtsW and PBP1B until it is outcompeted by the accumulation of FtsN that relieves this inhibition and initiates septum synthesis [29][30][31][32].
The maintenance of the PG sacculus is a dynamic process, which is achieved by the balanced activities of PG synthases and hydrolases that ensures a safe extension of the PG layer without defects that would cause cell lysis [33]. Beta-lactam antibiotics disturb this balance by inactivating the transpeptidase (TPase) activity of the penicillin binding proteins (PBP)s causing lysis of growing cells. We currently know more than twenty PG hydrolases that can cleave domain of unknown function (domain 2), which is inserted into the transpeptidase domain 1, and a third domain (domain 3), which is inserted into domain 2. Domain 3 is positioned above the active site (Serine 62, in domain 1), and might be involved in substrate binding or regulation [58,59] (Fig 1A). Site directed mutagenesis suggested that residues from domain 1 (R361) and 2 (D155) are important for the endopeptidase activity [60]. Deletion of dacB, which encodes PBP4, does not have morphological consequences but causes a decrease in the percentage of PG crosslinks and of PG-attached lipoprotein [61] and an increased sensitivity to bile salts [62]. A dacB deletion aggravates the morphological defects of a dacA (PBP5) deletion mutants [40], and the chaining phenotype of the ΔamiC strain [45]. PBP4 has also been reported to be involved in exopolysaccharide synthesis regulation [63]. Overproduction of PBP4 is toxic [64] presumably because the enhanced cleavage of peptide cross-links weakens the PG mesh and eventually causes lysis. There is litttle knowledge about the spatio-temporal regulation of PBP4 activity and its coordination with cell cycle events. Here we determined the cellular localization of PBP4 during the cell cycle of E. coli. The timing of localization at midcell and interactions of PBP4 support a role in remodeling of PG synthesized by the divisome.

PBP4 localizes to the lateral wall and at midcell
To determine the localization of PBP4 in E. coli cells as a function of the bacterial cell cycle, we generated an antiserum against PBP4 and removed unspecific antibodies by adsorption to cells of a dacB deletion mutant that lacks PBP4. The purified antibodies were specific for PBP4 (S1 Fig) and used to immunolabel the wild-type strain MC4100 grown in rich (TY) medium at 37˚C. PBP4 localized strongly at midcell and to a lesser extent in the lateral wall in wild-type cells (Figs 1 and 2).

PBP4 localizes close to the peptidoglycan layer and the outer membrane
Although it was reported that PBP4 has a stretch of weakly amphipathic amino acids at its Cterminus [57], it was not known whether PBP4 freely diffuses in the periplasm or associates with the membrane or other proteins. It has been previously observed that chemically fixing freely diffusing periplasmic proteins can shock them into the cell poles [10,65]. The fact that we did not observe any polar localization of PBP4 after fixation (Figs 1B, 1C and S1), suggests that it is either associated with the membrane or other proteins. Next, we investigated whether PBP4 was accessible to anti-PBP4 in cells with an intact PG layer. For this, wild-type cells were grown in TY at 37˚C and immunolabeled without the typically used Triton X-100-mediated permeabilization of the OM or without digesting the PG mesh with lysozyme. PBP4 was not accessible without permeabilization of the OM but was fully accessible in cells with an intact PG (S1D Fig). In contrast, cytoplasmic membrane bound proteins like FtsN and PBP3 were not accessible without degradation of the PG layer by lysozyme, whereas the outer membrane lipoprotein LpoB is accessible with Triton only (S1D Fig). This suggests that PBP4 resides close to the OM and the PG layer.

PBP4 localizes as early divisome protein and its concentration in the envelope is not constant
Since PBP4 localizes at midcell it could be part of or associated with the divisome. The division machinery is assembled in two successive steps [26,66]. First, FtsZ and its membrane bound partners ZipA and FtsA in conjunction with other Z-ring associated proteins form the protoring that localizes at midcell [67]. Second, after approximately 20% of the cell division cycle the The extra fluorescence at midcell compared to the rest of the cell (FCPlus) is plotted as function of the normalized cell division cycle age. The dots (grey for MC4100 and light pink for the BW25113 ΔdacB strain) are the values measured for the individual cells and the markers with bars (black for the wild-type MC4100, red for the BW25113 ΔdacB) are 5% age bins with 95% confidence. (D) Demographs of the localization of FtsZ, PBP4 and FtsN in the cells sorted according to cell length. The white line indicates the length of the cells. (E) Comparison of the timing of the localization at midcell of FtsZ (black), PBP4 (green) and FtsN (blue) during the cell division cycle age. The FCPlus values are min-max normalized to enable timescale comparison, despite differences in molecule number and antibody affinities. (F) BW25113 wild-type cells were grown to "steady state" in minimal glucose medium at 28˚C and immunolabeled with antibodies specific for FtsZ, PBP4 and FtsN. The FCPlus of three independent biological experiments in 5% age bins was determined. The binned FCPlus values were min-max normalized. The average min-max normalized FCPlus values of the three experiments were subsequently plotted as function of the cell division age cycle.
https://doi.org/10.1371/journal.pgen.1010222.g002 remaining proteins assemble subsequently at the division site. To obtain more details on PBP4 localization in relation to other cell division proteins, MC4100 cells were grown to steady state in minimal glucose medium at 28˚C. The cells were immunolabeled with antibodies against PBP4, FtsZ (early localizing), or FtsN (late localizing). Interestingly, the concentration of PBP4 was not constant during the cell division cycle but increased simultaneously with its enhanced localization at midcell (Fig 2B and 2C). Using the ribosome profiling data from [68] on the average number of 133 PBP4 molecules per cell in a culture grown in MOPS minimal medium, we calculated from the total fluorescence per average cell in the culture and the amount of fluorescence at midcell (FCPlus) that about 100 PBP4 molecules are present at midcell during the process of septal PG synthesis. The strong localization bias of PBP4 at the midcell (Fig 2), suggests a dedicated function in the cell division process. To determine whether PBP4 belongs to the early or late localizing proteins of the divisome, we analyzed the timing of its midcell localization. To this end, we plotted the extra fluorescence (FCPlus) present at midcell in comparison to the amount of fluorescence in the rest of the cell (Fig 2D, 2E, and 2F). PBP4 localization coincided with that of FtsZ, indicating that PBP4 localizes relatively early in the cell cycle to midcell.

Midcell localization of PBP4 depends on the FtsEX complex
To investigate if PBP4 localization is dependent on the presence of the Z-ring or the later localizing division associated proteins we localized PBP4 in various division defective strains. First, the localization of PBP4 was assessed in a strain that expressed Tre1 that ADP-ribosylates 6 proteins including FtsZ at R174, which renders it unable to participate in protofilament formation [69,70]. The inactive E415Q variant of Tre1 was used as negative control. PBP4 was unable to localize in cells with insufficient FtsZ protofilaments to form division Z-rings ( Fig  3A), whereas it localized as in WT cells in the presence of the TreE415Q variant, proving that a Z-ring is needed for PBP4 localization. Next, strains harboring temperature sensitive alleles of FtsE, FtsQ and PBP3 (FtsI) were grown in minimal glucose medium at 28˚C to an OD 450 of 0.2, diluted 1:4 in prewarmed medium of 28˚C (permissive temperature) or 42˚C (non-permissive temperature) and grown for two mass doublings before PBP4 immunolabelling. PBP4 localized at stalled division sites in filamentous cells with thermo-labile FtsQ or PBP3 indicating that these proteins are not needed to recruit PBP4 ( Fig 3B). However, PBP4 localized very diffuse around potential division sites in the filamentous ftsE(ts) cells ( Fig 3B). This suggests that a functional FtsE is important for PBP4 localization. The FtsEX/EnvC complex assists in the assembly of the divisome [71,72] and is essential to activate the amidase function of AmiA and AmiB to i) generate denuded glycan strands for the recruitment of the septal PG synthesis activator FtsN and ii) cleave PG for cell separation during constriction [46]. To investigate a possible link between PBP4 and the amidase activation complex we further looked for the localization requirement of PBP4 in a ΔenvC strain, as EnvC bridges FtsEX with AmiA and AmiB, and is required for their activation [43,45]. EnvC was not necessary for PBP4 localization (Fig 4A), although PBP4 failed to localize in a ΔftsEX strain ( Fig 4A) and a newly constructed ΔftsE single mutant strain. The ftsE and ftsX genes reside in an operon and the expression of ftsX requires the ftsE gene [73]. We therefore constructed a ΔftsE strain with a weakened pTrc99A promoter placed upstream of ftsX to ensure sufficient expression of FtsX. The resulting strain, XL36 grew with a mild filamentous phenotype in rich medium that could be complemented by the expression of FtsE(wt) (S2 Fig). PBP4 did not localize at septal positions in the ΔftsE strain XL36 (Fig 4A) but localized normally when FtsE was expressed to complement this strain (Fig 4B). This suggests that the periplasmic part of FtsX enhances the septal localization of PBP4 and that this requires FtsE.
To determine whether the activity of FtsE was needed for the localization of PBP4, we expressed three inactive FtsE variants that had been shown to localize to midcell [73]. The K41Q and D162A versions of FtsE cannot bind ATP and FtsE(E163A) binds ATP but cannot hydrolyze it. PBP4 was able to localize at the division site in cells that expressed any of these inactive variants of FtsE as the only copy (Fig 4B), demonstrating that the activity of FtsE is not required for the localization of PBP4. PBP4 localization is dependent on the presence of the proto-ring. A. MC4100 was transform with a plasmid that expressed either Tre1 that abolished the ability of FtsZ to polymerize by ADP-ribosylation of residue R174 or with the inactive variant Tre1(E415Q). Cells were grown in Gb1 medium at 28˚C and expression of the inhibitor was induced for 2 mass doublings (MD) with 0.15% arabinose. Subsequently, the cells were fixed and immunolabeled with antibodies against FtsZ or PBP4. From left to right: the phase contrast (PC), corresponding fluorescence image of the PBP4 labeling and the merged images are shown. In the demographs the cells are sorted according to cell length (contours in white). The number of cells analyzed were 3694 and 3756 for Tre1E415Q and 1275 and 1621 for Tre1 for anti FtsZ and anti PBP4, respectively. B. Isogenic strains of MC4100 producing different temperature sensitive (ts) versions of cell division proteins were grown to steady state in Gb1 medium at 28˚C and split in two parts. These were 1:4 diluted in prewarned medium and grown for two MD at either 28˚C or 42˚C. The cells were fixed, labeled with antibodies specific for PBP4 (anti4). From left to right: the phase contrast (PC), corresponding fluorescence image of the PBP4 labeling and the merged images at the permissive temperature, demograph of PBP4 fluorescence localization where cells are sorted according to their cell length (contours in white). This is followed by the same series from the non-permissive temperature samples. The number of cells analyzed were 2809 and 1576 for LMC515 FtsE(ts), 3833 and 1203 for LMC531 FtsQ (ts) and 3876 and 926 for LMC510 PBP3(ts), for the cells grown at 28˚C and 42˚C, respectively. The scale bar equals 5 μm. Within one antibody staining the brightness and contrast of the samples is identical and therefore directly comparable. https://doi.org/10.1371/journal.pgen.1010222.g003

Amidases affect the localization of PBP4
Interestingly, PBP4 localized strongly at division sites in a strain lacking the three amidases or lacking amiA and B but more diffuse in strains that lacked amiC or nlpD (S3 and S4 Figs). The absence of amiAB or amiABC also resulted in a 2-or 3-times enhanced expression of PBP4 compared to the parental wild-type strain. Furthermore, PBP4 localized as in wild-type in a strain lacking essential components of the Tol-Pal system, which is involved in the coordination of OM constriction and PG invagination during cell division (S4 Fig).

PBP1A/LpoA contribute to midcell localization of PBP4
PBP4 was recently reported to interact with PBP1A, LpoA and the EPase adaptor protein NlpI [55]. To investigate whether PBP1A and/or LpoA proteins were needed for midcell localization of PBP4, strains deleted for mrcA (PBP1A), lpoA, mrcB (PBP1B) or lpoB were grown in rich medium at 37˚C and immunolabeled with anti-PBP4. The amount of PBP4 that localized at midcell was identical for the parental wild-type strain BW25113, ΔmrcB and ΔlpoB, but reduced in the ΔmrcA and ΔlpoA strains (even after correction for the smaller diameter of these cells, [74], whereas the PBP4 concentration was similar for all strains (S5 Fig). This indicates that PBP1A/LpoA could assist in the localization of PBP4 at midcell.

NlpI seems not to be involved in midcell localization of PBP4
As NlpI was reported to interact with PBP4 [55], we next tested if the interaction with the OM anchored NlpI affected the localization of PBP4. In the ΔnlpI cells the amount of PBP4 at midcell per μm average cell circumference was reduced by 42 ± 6.7% (n = 3) of the wild-type and the protein localized later at midcell than in the wild-type cells (S6 Fig). However, the PBP4 concentration in the ΔnlpI strain was also lower than in the wild-type cells (64.8 ± 17.6%, n = 3). This could indicate that the cells down-regulate PBP4 expression or that PBP4 becomes unstable in the ΔnlpI strain. Labeling of PBP4 in a ΔnlpI strain did not require lysozyme (S6 Fig), suggesting that NlpI is not needed to maintain PBP4 localized outside the PG layer, which could be facilitated by additional interactions of PBP4.
In conclusion, the midcell localization of PBP4 depends on the presence of the Z-ring and FtsEX, its distribution in the lateral wall and at midcell is affected by PBP1A, LpoA, and the amidases. Its concentration during the cells cycle is not constant and mutants have either less (ΔnlpI) or more ΔamiABC) PBP4 per cell than wild-type cells.

PBP4 localizes at inactive divisomes
To further dissect whether the localization of PBP4 at the division site is dependent on the availability of its substrate or the presence of other proteins at the division site, cell division was inhibited through the specific inactivation of PBP3 by aztreonam [75]. Aztreonam treatment for 1-3 mass doublings stalls the division machinery at midcell resulting in filamentous growth. In the longer filaments new division machineries are formed and localize at future division sites [3]. In these filaments PBP4 still localized albeit somewhat weaker than in untreated cells at all possible division sites (S7 Fig), confirming that the activity of PBP3 was not required for the recruitment of PBP4. While PBP3 is inhibited, preseptal PG synthesis [76] still occurs at the potential division sites in these filamenting cells [16,36,77]. The localization of PBP4 mimics the localization of PBP1A in cells treated with aztreonam, which was previously shown to be recruited to preseptal PG synthesis sites by early cell division proteins [2,16].

Activity of PBP4 is not needed for midcell localization
Substrate binding can be a key determinant in the localization of proteins [36,78,79]. To test whether PBP4 substrate binding is needed to localize at division sites we expressed the active site mutants PBP4(S62G) and (S62A) [59] that fail to bind β-lactams and are not able to hydrolyze the PG stem peptide (S8 and S9 Figs). A plasmid encoding the mutant protein was constitutively expressed in a ΔdacB background by basal activity of a weakened pTrc99A promoter. Interestingly, the wild-type and active site mutant when expressed from plasmid were both able to localize at midcell, although the S62G variant localized more diffusely (Fig 5). This indicates that although PBP4(S62G) cannot hydrolyze its substrate or bind β-lactams covalently (S8C and S9 Figs), it is either still able to interact with peptidoglycan or it localizes through protein interactions.
Since PBP4(S62G) could still interact with its PG substrate, we mutagenized the gene to express PBP4 versions in which residues D155 or R361 involved in the binding of the DAP residue in the stem peptide [60] were replaced by alanine. The genes encoding PBP4(D155A) or PBP4(R361A) were expressed again from same plasmid (pTrc99A) in a ΔdacB background, and both PBP4 versions localized at midcell (Fig 5). An immunoblot analysis showed that all mutants were expressed to similar levels and that PBP4(R361A) and PBP4(D155A) were able to bind the fluorescent β-lactam Bocilin-FL (S8 Fig). Therefore, all three active site protein variants are potentially able to interact with substrate as an inactive protein. The truncated protein interacted with PG sacculi, albeit somewhat weaker than the native protein and perhaps non-specific, (S10C Fig). However, although the inactive PBP4ΔD3 was capable of binding substrate, it did not localize at midcell (Fig 5). Hence, we conclude that the midcell localization of PBP4 is driven by protein-protein interactions and/or by substrate interaction but is independent from its own activity.
Although the PBP4 variants were not induced, the leakiness of the plasmids was sufficient to produce PBP4 in 3 x excess to the wild-type strain, which clearly affected their morphological parameters (Fig 5C and 5D). The timing of the PBP4 localization of these mutants was also quite variable ( Fig 5A). To verify whether the timing of the rest of the divisome had adapted itself to the timing of PBP4, cells grown to steady state in minimal medium at 28˚C, were fixed, divided into three aliquots, and labeled with antibodies against FtsZ, PBP4 or FtsN. The timing of the arrival of these proteins at midcell was determined by analyzing fluorescence signals of several thousands of cells [80]. The initiation of localization at midcell was defined as the timepoint during the cell division cycle age where more fluorescence is detected at mid cell than in the rest of the cell (t 0 ) and the time it would take to reach half of the maximum value of the fluorescence at midcell (t 1/2 ) was determined by dividing the maximum FCPlus by two and noting the corresponding timepoint (see for a graphical explanation S11A or 6A Figs). We decided against using the maximum because this time point was more difficult to determine as the maximum FCPlus often reached a plateau before decreasing at the end of the cell cycle. Indeed, the variation in the timing of the PBP4 variants localization reflected differences in the timing of the assembly of the divisome (S11 Fig).

PBP4 affects the timing of the divisome assembly
Because the PBP4 variants affected the morphology and the timing of the divisome assembly (Figs 5 and S11), we hypothesized a role for PBP4 in the divisome assembly. To verify this, we measured the timing of the divisome assembly in the absence of PBP4 in the ΔdacB strain and replicates, respectively) and its parental BW25113 (7 and 6 replicates, respectively) and the ΔdacB strain (6 and 6 replicates, respectively) and its parental LMC500 (13 and 12 replicates, respectively). Using an unpaired T-test: P 0.0001 = ��� , P 0.001 = �� .
https://doi.org/10.1371/journal.pgen.1010222.g006 in its parental BW25113. A difference was observed between the two strains, which was not significant due to the large variation in the localization timing of FtsZ and FtsN in both strains ( Fig 6B). We inferred that BW25113 is not able to reach a reproducible steady state, possibly due to a known frameshift in the rph gene, which leads to pyrimidine starvation and hence irregular DNA replication [81]. The uracil and thymine that we add as standard to the minimal medium for this strain is evidently unable to fully compensate the deficiency. To solve this possible issue, the dacB gene was deleted from the wild-type strain MC4100 and the cells were grown to steady state in minimal medium in the absence of an antibiotic. The results reproduced the difference between the ΔdacB and wild-type strains but now with a significance of P = 0.0001 ( Fig 6B). No significant difference between the ΔdacB and its parental in other parameters that could account for a change in cell cycle such as mass doubling time, cell length, diameter or the percentage of constricting cells was detected (S1 Table). However, FtsZ started to localize (t 0 ) 8% of the cell division cycle time earlier in the ΔdacB strain and reached 50% of its maximum intensity (t 1/2 ) 9% earlier compared to the parental strain. FtsN localized 8.9% (t 0 ) and 6.5% (t 1/2 ) earlier (see for an explanation of t 0 and t 1/2 Fig 6A). These results show that the cells initiated divisome assembly significantly earlier in the absence of PBP4.

The absence of PBP4 exacerbates filamentation of ΔenvC
The early localization, the delay in divisome assembly, and the dependence on FtsEX for midcell localization in combination with the enhanced expression of PBP4 in a ΔamiAB strain, suggested a connection between the activity of PBP4 and the start of cell division. To test the hypothesis that PBP4 could help AmiAB to provide denuded glycan strands that would attract FtsN, the morphology of a series of deletion combinations was analyzed (S2 Table). Interestingly, ΔenvC cell chains became twice as long when dacB was also deleted (Fig 7). In contrast, the morphology of mutants lacking single amidase genes or a double amiAB deletion mutant was not affected by the additional deletion of dacB (S2 Table). From these results we conclude that EnvC and PBP4 are both needed for the correct function of AmiA and/or AmiB.

PBP4 domain 3 is needed for interaction with NlpI but not PBP1A/LpoA
We next used microscale thermophoresis (MST) in which one protein is labelled with a fluorophore and titrated with an unlabeled partner protein to analyze the interactions between PBP4 and its known and putative interaction partners. Purified proteins were assayed for direct protein-protein interactions. We confirmed our previous result showing that wild-type PBP4 interacted with PBP1A, LpoA and NlpI [55], and we could now cross-link PBP1A to PBP4 in cells, followed by co-immunoprecipitation ( Fig 8C). Interestingly, PBP4 lacking domain 3 (PBP4ΔD3) did not interact with NlpI (Fig 8).
We found that PBP4 and PBP4ΔD3 interacted with full length LpoA with apparent K D values of 315 ± 38 nM and 153 ± 31 nM, respectively (Figs 8A and S12) suggesting that domains 1 or 2 of PBP4 are sufficient for the interaction with LpoA. MST assays of PBP4 or PBP4ΔD3 with LpoA's C-terminal (residues N257-S678; LpoA C ) or N-terminal (LpoA residues G28-T256; LpoA N ) domains also yielded positive binding curves (Figs 8A and S12) [82][83][84]. LpoA C and the full length LpoA version had comparable affinities for both PBP4 versions, suggesting that the C-terminal domain of LpoA is sufficient for interaction with PBP4 ( Fig 8A). Interestingly, the removal of domain 3 from PBP4 resulted in an increased affinity of LpoA N for PBP4 (app. K D decreased by~50 fold) (Figs 8A and S12). Hence, the interaction between the N-terminal domain of LpoA and PBP4 possibly occurs via a different mechanism and may be enhanced by a conformational rearrangement of domain 3 of PBP4. In conclusion PBP4 robustly interacts with PBP1A/LpoA independent of its domain 3.

LpoA has a small effect on PBP4 activity
We found previously that NlpI does not affect the activity of PBP4 [55] but it was possible that PBP1A/LpoA have an effect. We first verified that NlpI or PBP4 did not alter the activity of PBP1A in the presence or absence of LpoA using an in vitro glycosyltransferase assay (S13

Amidase activator EnvC interacts with PBP4 and NlpI
PBP4 localizes at midcell in cells lacking EnvC but not in cells lacking membrane anchored FtsEX. To begin investigating interactions of PBP4 with members of the amidase activation pathway, we first tested for the interactions with EnvC, a periplasmic protein with a coiled-coil and a LytM domain that activates AmiA and B upon interaction with FtsEX [85]. MST experiments showed that full-length EnvC interacted with full-length PBP4 (K D = 307±45 nM), but substantially weaker with PBP4ΔD3 (K D > 5900±2823 nM) (S14 Fig). The LytM domain of EnvC did not interact with PBP4 suggesting that the interaction between both proteins is mediated by domain 3 of PBP4 and the coiled-coil domain of EnvC. Hence domain 3 of PBP4 is needed for midcell localization ( Fig  6) and is also used to bind to EnvC (Fig 8) and NlpI (S15 Fig). Yet the absence of none of these proteins alone is enough to abrogate the PBP4 midcell localization (Figs 3, 4 and S6).
During these studies, we also discovered an interaction between EnvC with NlpI, via the LytM domain of EnvC (S15 Fig). The different binding domains in EnvC for PBP4 and NlpI suggested that a trimeric complex could be formed. Indeed, a pulldown assay with oligo-histidine-tagged NlpI and untagged PBP4 and EnvC supported the formation of a trimeric complex. In summary, we show that PBP4 interacts with a component of the amidase-activation pathway, EnvC and with NlpI in a manner that can imply the formation of dynamic/different tripartite complexes that requires further investigation. Discussion E. coli contains a large number of seemingly redundant PG hydrolases, which possibly fine tune peptidoglycan biogenesis and remodelling in response to environmental parameters [34]. Apart from their enzymatic activities, little is known about the function of most PG hydrolases. Here we report that the DD-carboxy/endopeptidase PBP4 localizes at midcell during septal PG synthesis in E. coli and is important for the timing of the assembly of the division machinery.

PBP4 localizes at midcell during cell division in an FtsEX dependent manner
PBP4 localizes predominantly at midcell during cell division in an FtsEX dependent manner. Inactive PBP4 variants were able to localize at midcell indicating that substrate hydrolysis is not a requirement for localization. However, a version of PBP4 lacking the non-catalytic domain 3 (PBP4ΔD3), which is required for activity but not β-lactam binding, did not localize to midcell. Domain 3 is also required for the interaction with EnvC and NlpI, but not with PBP1A/LpoA. From these results we infer that the midcell localization of PBP4 requires multiple protein interactions, but not substrate hdyrolysis.

PBP4 might be initially recruited by preseptal PG synthesis
PBP4 interacts with NlpI, PBP1A and LpoA and these proteins can form a ternary complex ( [55] and (Fig 8). PBP4ΔD3 interacted with PBP1A and LpoA but not with NlpI or EnvC, suggesting that NlpI and EnvC bind to domain 3. The interaction with PBP1A and/or LpoA may contribute to the recruitment of PBP4 at midcell, as the amount of PBP4 at midcell was slightly reduced in ΔmrcA and ΔlpoA strains. PBP1A and its partner LpoA are involved in preseptal PG synthesis [16,80]. Although somewhat weaker, PBP4 still localizes at putative division sites in cell filaments generated with aztreonam that have lost septal synthesis but not preseptal synthesis activity. Possibly, some PBP4 enriches at preseptal PG synthesis sites together with PBP1A/LpoA, which localize through the interaction of PBP1A with ZipA [16], and preseptal PG synthesis may provide initially the substrate for PBP4. The interaction between NlpI and PBP4 could function to sequester PBP4 in the lateral wall to prevent its interaction with potential substrate (Fig 9).

Why does the loss of PBP4 cause premature divisome assembly?
The absence of PBP4 interferes with the timing of divisome assembly. In most bacteria studied thus far, the Z-ring and associated early division proteins localize first to prepare the future cell division site, perhaps by generating the border between the side wall and the future septum [16,76,86]. After a time-delay, the proteins that synthesize the bulk of the septal PG arrive and septation starts [26][27][28]. The absence of PBP4 causes a premature initiation of the assembly of the divisome for yet unknown reasons. The absence of PBP4 might enable NlpI to bind other EPases [54,55], coordinating their activity for remodelling the PG layer. However, the robust localization of PBP4 at midcell suggests a more direct role in modulating the timing of division initiation. We hypothesize that during cell elongation sufficient PBP4 molecules are retained by NlpI in the lateral wall above the PG layer to avoid promiscuous PBP4 activity (Fig 9), consistent with the reduced amount of PBP4 in cells lacking NlpI (S6 Fig and [55]). PBP4 might be initially recruited to midcell through interactions with PBP1A and LpoA, and the FtsEX-EnvC complex (Fig 9). The interaction of PBP4 with EnvC (S15 Fig) might ensure a concerted endopeptidase/amidase activity as EnvC activates AmiA and AmiB, which generate denuded glycan chains that attract FtsN, the trigger of septal PG synthesis. This is also suggested by the observation that a double ΔdacB ΔamiC mutant forms longer cell chains than a single ΔamiC strain [45] and by the observation that a ΔenvC ΔdacB forms longer filaments than a ΔenvC strain (Fig 7), whereas the deletion of dacB in an amiA and amiB double deletion strain has no exacerbating effect. As deleting dacB does not prevent the assembly of the divisome, it is plausible that other endopeptidases either contribute and/or partially take over the function of PBP4. Once septal peptidoglycan synthesis starts by the mature divisome, PBP4 remains at midcell as shown by the immunolocalization experiments (Fig 2). This could suggest that continued PBP4 activity is needed in cooperation with the amidases and other endopeptidase to create sufficient binding substrate for FtsN.

Strain construction
Strain XL35, the FtsE partial deletion strain, was constructed as follows: Primers priXL221 and priXL222 were used to amplify the upstream homologous sequences (UHS) from the ftsE start codon, and primers priXL223 and priXL224 were used to amplify the downstream homologous sequence after the FtsE 151 th amino acid (the potential promoter of ftsX), from the MC4100 genome. Primers priXL51 and priXL54 were used to amplify the kanamycin resistance cassette from plasmid pKD4 [87]. The amplified overlap PCR product (UHS-FRT-KAN-FRT-DHS) with primers priXL221 and priXL224 was used to construct the recombinant strain XL34 in which the first 150 amino acids encoding sequence of ftsE was deleted [87]. Subsequently, the kanamycin selection cassette was removed with plasmid pCP20 [87] yielding XL35.
Strain XL36, the ftsE clean deletion strain, was constructed similarly as XL35. Primers priXL221 and priXL222 were used to amplify the upstream homologous sequences from the ftsE start codon, and primers priXL225 and priXL226 were used to amplify the downstream homologous sequences after the ftsE stop codon, from the MC4100 genome. In order to restore the FtsX expression, a pTrc99Adown promoter [88] was amplified from plasmid pSAV057 [89] with primers priXL77 and priXL227 and a chloramphenicol selection cassette was amplified from plasmid pKD3 with primers priXL52 and priXL54. The amplified overlap PCR product (UHS-FRT-CAM-FRT-DHS) with primers priXL221 and priXL226 was used to construct the recombinant strain XL33 in which the entire ftsE coding sequence was deleted and FtsX was expressed from the chromosomally encoded pTrc99Adown promoter. Subsequently, the chloramphenicol selection cassette was removed with plasmid pCP20 yielding XL36 (Table 1). Bacterial strains construction for ΔdacB and amidases/regulators: E. coli BW25113 was the parental strain used for these experiments. E. coli amiA::kan, amiB::kan, amiC::kan, envC::kan, nlpD::kan, dacB::kan mutants were obtained from the KEIO library [90] and the mutations transduced into a clean parental strain. The strain dacB::cam was obtained from a previous study [55]. Double mutants were created by P1 transduction as described previously [91]. The chromosomal modification was confirmed by PCR when a Cam R or Kan R cassette was transduced.

Plasmid construction
The ftsE gene was amplified from the chromosome of LMC500 [92] using the forward primer priXL187 and the reverse primer priXL188, and sequenced to characterize the mutation. The wild-type ftsE gene was cloned into pSAV057, p15A origin and Cam resistant [89] to produce the plasmid pXL135. FtsE was expressed under the control of the pTcr99A down promoter. QuickChange site directed mutagenesis (Agilent technologies, Santa Clara, CA) and Gibson assembly [93] approaches were applied afterwards to construct the inactive ftsE mutant plasmids pXL137, pXL138 and pXL139, using the primers listed in Table 2.
To construct the PBP4 expression plasmids, the wild-type dacB gene was firstly amplified from the E. coli MG1655 genome using primers priXL244 and priXL245, and subsequently cloned into plasmid pSAV057 and pSAV057-dsba ss -mCherry-PBP5 (pNM010 [38]) with EcoRI and HindIII, to generate the PBP4 plasmids without and with mCherry fusion. Similarly, QuickChange site directed mutagenesis and Gibson assembly approaches were applied afterwards to construct the inactive dacB mutant plasmids from these two wild-type plasmids, using the primers listed in Table 2. Tre1 and Tre1E415Q were expressed from the arabinose inducible promoter of pBAD18 as described [69].

Immunolabeling
After reaching steady state for minimal glucose grown cells or the desired OD for rich medium grown cells, the cells were fixed for 15 min by addition of a mixture of formaldehyde (f. c. 2.8%) and glutaraldehyde (f. c. 0.04%) to the cultures in the shaking water bath and immunolabeled as described [94] with Rabbit polyclonal antibodies against PBP4, NlpI [55] preabsorbed against ΔdacB or ΔnlpI, strains, respectively, or against FtsZ or FtsN [26]. As secondary antibody, donkey anti-rabbit conjugated to Cy3 or to Alexa488 (Jackson Immunochemistry, USA) diluted 1:300 in blocking buffer (0.5% (wt/vol) blocking reagents (Boehringer, Mannheim, Germany) in PBS) was used, and the samples were incubated for 30 minutes at 37˚C. For immunolocalization, cells were immobilized on 1% agarose in PBS slabs coated object glasses as described [70] and photographed with an Orca Flash 4.0 (Hamamatsu, Japan) CCD camera mounted on an Olympus BX-60 (Japan) fluorescence microscope through a 100x/N.A. 1.35 oil objective. Images were taken using the program ImageJ with MicroManager (https://www. micro-manager.org). SIM images were obtained with a Nikon Ti Eclipse microscope (Japan) and captured using a Hamamatsu Orca-Flash 4.0 LT camera. Phase contrast images were acquired with a Plan APO 100x/1.45 Ph3 oil objective. SIM images were obtained with a SR APO TIRF 100x/1.49 oil objective, using 3D-SIM illumination with a 561nm laser, and were reconstructed with Nikon-SIM software using the values 0.20-0.25-0.20 for the parameters Illumination Modulation Contrast (IMC), High Resolution Noise suppression (HNS) and Out of focus Blur Suppression (OBS), respectively.

Image analysis
Phase contrast and fluorescence images were combined into hyperstacks using ImageJ (http:// imagej.nih.gov/ij/) and these were linked to the project file of Coli-Inspector running in combination with the plugin ObjectJ (https://sils.fnwi.uva.nl/bcb/objectj/). The images were scaled to 15.28 pixel per μm. The fluorescence background has been subtracted using the modal values from the fluorescence images before analysis. Slight misalignment of fluorescence with priXW6-R AATCTGAAGCCCCGGCCATGTGCCGGGGTTTCTTTTTGAatgggaattagccatggtcc LMC500ΔdacB Env-F GGAATTCcatatg GATGAGCGTGACCAACTC pET28a-His-EnvC respect to the cell contours as found in phase contrast was corrected using Fast-Fourier techniques as described [80]. Data analysis was performed as described [80]. In brief, midcell was defined as the central part of the cell comprising 0.8 μm of the axis. From either cell part, midcell and remaining cell, the volume, the integrated fluorescence, and thus the concentration of fluorophores was calculated. The difference of the two concentrations is multiplied with the volume of midcell. It yields FCPlus (surplus of fluorescence) at midcell. For age calculation, all cell lengths are sorted in ascending order. Then the equation is used, where rank is a cell's index in the sorted array, nCells is the total number of cells, and age is the cell's age expressed in the range 0.. 1. To analyze the FCPlus or the total concentration of fluorescence as a function of cell division cycle age, the data points of the individual cells were binned in 5% age classes and plotted with the 95% confidence interval indicated as error bars. To determine the timing of the divisome assembly, the moment that more fluorescent material was present at midcell (FCPlus) than in the rest of the cell was taken as timepoint zero of arrival (t 0 ) and then the maximum FCPLus value (usually a small plateau) was divided by half and the corresponding timepoint was noted as the half maximum fluorescent intensity time value (t 1/2 ).

ΔdacB and amidases/regulators fluorescence imaging
To image membrane and DNA compartmentalisation of the cells, FM1-43FX dye (Invitrogen) and DAPI (Stratech Scientific) were used, respectively. Overnight cultures were adjusted to initial OD 600 = 0.01 in LB. The cells were incubated at 37˚C until OD 600 =~0.2. A volume of 500 μl of the culture was stained with 5 μg ml -1 of FM1-43FX at room temperature for 10 min. The cells were adjusted to 33 mM sodium phosphate pH 7.4 and fixed by addition of 2.4% formaldehyde and 0.04% glutaraldehyde. Fixed cells were inoculated on agarose pads, which were prepared with 1.5% agarose in PBS and set in Gene Frames (Thermo Scientific

Proteins
Proteins were purified from overexpression strains according to published procedures: NlpI, PBP4, PBP4(A62) and PBP4ΔD3 [55]. LpoA, LpoA N and LpoA C [82]; PBP1A [4]. EnvC was purified from BL21(DE3) pET28a-His-EnvC. The overexpression plasmid pET28a-His-EnvC was constructed by amplifying a soluble EnvC version without the N-terminal, 34 amino acid long, signal peptide into the pET28a (+) vector using the restriction enzymes NdeI and Hin-dIII. To purify recombinant EnvC, cells which were grown in 3 L LB with 50 μg/ml kanamycin to OD578 0.5-0.6, at 37˚C. Protein overproduction was induced with 1 mM IPTG for 3 h at 30˚C before cells were harvested by centrifugation and lysed by sonication before ultracentrifugation at 14,000 × g, 1 h, 4˚C. The cell lysate was centrifuged and the supernatant was applied to a 5 ml HisTrap HP column (GE healthcare). The column was washed with 4 column volumes of 25 mM Tris/HCl, 300 mM NaCl, 20 mM imidazole, pH 7.5 at a flow rate of 2 ml/min. The oligohistidine-tagged EnvC was eluted using 25 mM Tris/HCl, 300 mM NaCl, 400 mM imidazole, 10% glycerol, pH 7.5. Protein purity and yield was analysed by SDS-PAGE and fractions of interest were pooled. The oligohistidine tag was removed by incubating the samples with 1 unit/ml of restriction grade thrombin (Novagen), overnight at 4˚C, during dialyses against 25 mM Tris/HCl, 200 mM NaCl, pH 8.0. The sample was concentrated to~5 ml using Vivaspin concentrator spin columns (Sartorius) at 4,500 × g at 4˚C. The sample was then applied to a HiLoad 16/600 Superdex 200 column (GE healthcare) at 1 ml/min pre-equilibrated with 25 mM HEPES/NaOH, 300 mM NaCl, 10%, glycerol, pH 7.5. Fractions were analysed by SDS-PAGE and those with the highest protein purity and yield were pooled and stored at -80˚C.

Ni 2+ -NTA pulldown assays
Nickel-Nitrilotriacetic acid (Ni-NTA) beads (100μl) were pre-equilibrated with dH 2 O and binding buffer (10 mM HEPES/NaOH, 10 mM MgCl 2 , 150 mM NaCl, 0.05% Triton X-100, pH 7.5) by centrifugation at 4000 × g, 4 min at 4˚C, and then incubated with proteins of interest. Equimolar amounts (1-2 μM) of His 6 -tagged and untagged proteins were incubated alone (control) and in combination, for 10 min at 4˚C in a 200 μl reaction volume, containing 10 mM binding buffer. An aliquot of this mixture is taken as an 'applied' sample. The protein samples were then added to pre-equilibrated Ni-NTA beads and incubated overnight on a spinning plate at 4˚C with 1.3 ml of binding buffer. Beads were centrifuged at 4000 × g, 4 min, 4˚C and washed a further 4-6 times with 1 ml of washing buffer (binding buffer with 30 mM Imidazole) before re-suspending in 250 μl washing buffer and transferring beads to Proteus spin columns (Generon), centrifuging as described above. Bound proteins were eluted by the addition of 50 μl of SDS-PAGE loading buffer and boiling at 100˚C for 5 min. Spin columns were centrifuged a final time at 1500 × g, 5 min, RT, to collect 'eluted' protein before separation by SDS-PAGE. Eluted samples were run alongside applied samples for comparison.

Microscale thermophoresis
Microscale thermophoresis (MST) is an immobilisation-free method which allows the detection of biomolecular interactions in solution. This technique is based on the specific directed movement of a protein along a heat gradient (thermophoresis), an observation first reported by Carl Ludwig in 1856 [96]. The thermophoresis of a protein changes upon ligand binding due to one or more changes to size, charge and/or hydration shell. In MST, a localized heat gradient is initiated by an IR-laser (wavelength 1470 nm). A protein of interest is fluorescently labelled and the change in its thermophoretic mobility, in the presence of an unlabelled ligand, is measured and expressed as a change in the normalised fluorescence (FNorm).
MST measurements were carried out as described in [97], using standard or premium capillaries on a Monolith NT.115 MST machine (Nanotemper). Binding curves and Kinetic parameters were plotted and estimated using manufacturer provided software (NT Analysis 1.5.41 and MO. Affinity Analysis (x64)). Capillary scans were carried out prior to all measurements to check for consistent fluorescence counts, confirming that any subsequent change in fluorescence was due to ligand binding and not due to inaccurate pipetting or adsorption and dilution effects.

SDS-denaturation (SD) test
In instances where capillary scans showed a ligand concentration dependant change in raw fluorescence, we investigated whether this was a property of the binding interaction by carrying out an SD-test.
Samples (10 μL) with the highest and lowest concentration of unlabelled ligand were centrifuged 10 000 × g, 5 min, RT and mixed 1:1 volume ratio with SD-test buffer (40 mM DTT, 4% SDS). Mixtures were boiled at 100˚C for 10 min, to abolish ligand binding, before being spun down and subjected to another capillary scan. If the fluorescence between samples that contained the highest and lowest concentration of ligand, after SDS treatment, were now back to ± 10% of each other, the initial observations were a property of ligand binding, and a binding curve was plotted from the raw fluorescence data. If the ligand concentration dependent change in fluorescence was still observed, then this indicated that the fluorescently labelled protein was aggregated, and assay and buffer conditions were optimised accordingly.
Samples were ultracentrifuged (140,000 × g, 4˚C, 1 h) to remove debris, before removing 2 × 1.2 ml of each supernatant to be subsequently diluted with 0.6 ml of CL buffer 4 (75 mM Tris/HCl, 10 mM MgCl 2 , 1 M NaCl, pH 7.5). One sample was incubated with an optimised concentration of specific antibody with the other used as a negative control. Both samples were incubated at 4˚C with agitation for 5 h. For the isolation of antibodies, and thus crosslinked interaction partners, 100 μl of protein G-coupled agarose bead resin (Roche) were washed (2 × CL buffer 4, 2 × CL wash buffer [2:1 CL buffer 3 and CL buffer 4]) and added to each sample, and incubated o/n at 4˚C with agitation.
Samples were centrifuged and the supernatant retained before washing the beads 10 × 1 ml with CL wash buffer. After the final wash, beads were resuspended in 250 μl CL wash buffer and transferred to 2 ml spin dry columns and centrifuged to isolate the beads. These were then resuspended in 50 μl of fresh SDS-loading buffer and boiled to elute bound proteins, and reverse cross-linkage, and were collected by centrifugation (10,000 × g, RT, 5 min). Supernatant and elution samples were resolved by SDS-PAGE and transferred to a nitrocellulose membrane by Western blotting to detect for specific interaction partners using purified antibodies. The secondary antibody used here is Trueblot Anti-Rabbit IgG-HRP specific for native antibodies.

Peptidoglycan digestion assays
To test for hydrolase activity on PG, 10 μl of sacculi isolated from strains of interest (usually E. coli strains MC1061, BW25113 or D456) were incubated with 1-10 μM of respective enzymes, 37˚C for between 1 h and overnight, as indicated for respective proteins. The standard reaction conditions were 10 mM HEPES/NaOH, 10 mM MgCl 2 , 150 mM NaCl, 0.05% Triton X-100, pH 7.5, in 100 μl reaction volume. Following incubation, samples were boiled at 100˚C, 10 min, to terminate reactions before digesting remaining PG overnight at 37˚C, with 1 μM cellosyl. The samples were centrifuged at 10,000 × g for 5 min, RT, to obtain digested muropeptide products in the supernatant.
To test for activity against soluble muropeptides; first, 100 μl of intact sacculi were incubated overnight at 37˚C with 1 μM cellosyl and cellosyl digestion buffer (20 mM NaPO 4 , pH 4.8). Next, samples were boiled at 100˚C, 10 min, before centrifugation at 10,000 × g for 5 min, RT, to obtain the soluble muropeptides in the supernatant [99]. Ten μl of the supernatant was then used as muropeptide substrate for incubating with enzymes of interest, for the appropriate incubation time, at 37˚C. Reactions were then terminated by boiling at 100˚C for 10 min.
After digestion of respective substrates, products were reduced with NaBH 4 , adjusted to pH 4-5 and separated for analysis by reversed-phase HPLC as described below in HPLC analysis.

Reduction of muropeptides with sodium borohydride
PG digestion samples were transferred to 2 ml vials following centrifugation. Muropeptides were reduced in a 1:1 volume ratio of sodium borohydride buffer (0.5 M sodium borate, pH 9.0) and a small spatula of sodium borohydride pellets, centrifuging at 3000 × g, 30 min, RT as described in [99]. The samples were adjusted to pH 4-5 using HPLC grade phosphoric acid before separation and analysis by reversed-phase HPLC.

Reversed-phase HPLC analysis of muropeptides
Following the protocol of [99], reduced muropeptides were separated for analysis on reversedphase HPLC systems with a Prontosil 120-3-C18-AQ 3 μm reversed-phase HPLC column (Bischoff). A linear gradient of solvent A (50 mM sodium phosphate, pH 4.31 supplemented with 0.2% NaN 3 ) to 100% solvent B (75 mM sodium phosphate, pH 4.95, 15% methanol) over 90 or 180 min was used to separate muropeptides at 52˚C. Unlabelled muropeptides were detected at UV absorbance 205 nm. In assays where 14 C-radiolabelled muropeptides were used, detection was achieved by flowing scintillation cocktail along with standard buffers to give a radioisotope scintillation count (radioactivity CPM). Muropeptide profiles were recorded and analysed using Laura V4.2.11.129 (LabLogic Systems Ltd.).

Spectrophotometric D-alanine release assay (DD-carboxypeptidase assay)
This protocol was adapted from [60]. The carboxypeptidase (CPase) activity of PG hydrolases results in the release of the terminal D-Ala residue from the pentapeptide stem of PG precursors. Using UDP-MurNAc pentapeptide as a substrate, and in this case PBP4, it was possible to spectrophotometrically measure the release of D-Ala.
All constituents of the reaction were added and mixed directly in a quartz cuvette (Hellma, 10 mm light path, 15 mm centre), before the addition, and brief mixing by pipette, of purified UDP-MurNAc pentapeptide (BACWAN, Warwick University) to begin the reaction. The released D-Ala residues from the CPase activity of PBP4 are oxidatively deaminated by the action of D-amino acid oxidase to produce pyruvate and hydrogen peroxide (H 2 O 2 ). The released H 2 O 2 is reduced to H 2 O by HRP using Amplex Red as an electron donor. Oxidised Amplex Red produces resorufin, which has an intense pink colour and the production of which was measured spectrophotometrically using a Cary 100 Bio UV-visible spectrophotometer (wavelength 555 nm). The change in absorption over 10 min was measured and analysed using the complementing software.

Circular dichroism
Proteins were dialysed overnight against 10 mM NaPO 4 , pH 7.5 and concentrated/diluted to 0.4 mg/ml. CD measurements were taken using a Jasco J-810 spectropolarimeter (Jasco, Tokyo, Japan) using a wavelength range of 180-250 nm. The average of 10 runs was taken for each protein with a buffer control subtraction. For a direct comparison, correcting for the differing amino acid sequences, the collected data were converted to molecular CD and plotted against wavelength (nm). The resulting CD spectra are compared in S4 Fig and show that PBP4 lacking domain 3 is folded, consisting of both α-helices (~190, 208 and 222 nm) and βsheets (~210 nm).

Analytical ultracentrifugation
Purified PBP4 and PBP4ΔD3 were dialyzed over night against 25 mM HEPES/NaOH, 150 mM NaCl, pH 7.5, in preparation for AUC experiments. AUC sedimentation velocity (SV) experiments were carried out in a Beckman Coulter (Palo Alto, CA, USA) ProteomeLab XL-I analytical ultracentrifuge using absorbance at 280 nm and interference detection. The AUC runs were executed at the rotation speed of 45,000 rpm and the temperature of 20˚C using an 8-hole AnTi50 rotor and double-sector aluminium-Epon centerpieces. The sample volume was 400 μl and the sample concentrations ranged between approximately 0.25 and 1.4 mg/ml. The partial specific volumes (� v) of the proteins were calculated from their amino acid sequence, using the program SEDNTERP [100]. Sedimentation velocity boundaries were analyzed using the size-distribution c(s) model implemented in the program SEDFIT http://www. analyticalultracentrifugation.com [101]. The experimental values of the sedimentation coefficient were converted to the standard conditions (s20,w), which is the value of sedimentation coefficient in water at 20˚C. The size-distribution peaks were integrated to obtain the weightaveraged values for sedimentation coefficient and molecular mass.
The atomic coordinates from the published high-resolution structure of [58] (pdb accession code 2EX2) were used to calculate the sedimentation coefficient values for the monomeric and dimeric forms of PBP4 /PBP4ΔD3 using the program SoMo [102]. PBP4 /PBP4ΔD3 crystallographic dimer was built using program PyMol (https://pymol.org/2/).

PG binding assay
To assay for interactions of respective proteins with purified PG, we used a PG binding protocol as described in [103]. Briefly,~100 μg purified PG was pelleted by centrifugation (10,000 × g, 10 min, 4˚C) and resuspended in binding buffer (10 mM Tris/Maleate, 10 mM MgCl 2 , 50 mM NaCl, pH 6.8). PG was incubated with 10 μg of desired protein before incubating on ice for 30 min and centrifuging at 10,000 × g, 10 min, 4˚C. An aliquot of the supernatant (S) was taken for SDS-PAGE. The pelleted PG was washed with 200 μl binding buffer by centrifugation as before and the wash supernatant was taken as sample W for SDS-PAGE. The pelleted PG was then resuspended in 100 μl of 2% SDS and incubated for 1 h on a stirring plate, 4˚C. Mixtures were centrifuged a final time at 10,000 × g, 10 min, 4˚C, with the supernatant, now containing any initially bound protein, taken as the pellet sample (P). A negative control sample with no PG was assayed in parallel to determine that binding of protein was specific. Samples, along with corresponding controls, were analysed by SDS-PAGE and visualised by Coomassie staining. Presence of protein band in pellet sample (P) indicated binding to PG, whilst presence in supernatant sample (S) indicated no binding.

Bocillin binding assay
A fluorescent-bocillin binding assay was used to determine whether a PBP was purified with a correctly folded active site. Bocillin FL is a commercially available dye-β-lactam conjugate, synthesised from penicillin V and BODIPY (Molecular Probes, In., Eugene, Oreg.). Ten μg of purified PBP was incubated for 10 min at 37˚C, with 20 ng of Bocillin FL in a 50 μl volume with 10 mM HEPES/NaOH, 10 mM MgCl 2 , pH 7.5. A negative control sample was pre-incubated with 1 mM ampicillin for 30 min, 37˚C, to block the PBP active site, prior to incubation with Bocillin FL. After incubation, samples were boiled at 100˚C for 10 min with 40 μl of SDS-PAGE loading buffer and resolved by SDS-PAGE. Fluorescent signal was observed using Typhoon Fluorescence-imager (Excitation; 488 nm, emission; 520 nm, BP20 PMT 400-600 V), followed by visualisation of gels by Coomassie staining. Proteins incubated with ampicillin should have no detectable binding of Bocillin FL. For whole cells, a culture was grown in 5 mL TY and induced with 10 μM IPTG for 3 mass doublings to an OD 600 of 0.5 and the cells were pelleted and washed once with PBS (8 g NaCl, 1.8 g Na 2 HPO 4 , 0.24 g KH 2 PO 4 per liter, adjusted to pH 7.4) at 4˚C and resuspended in 100 μl PBS with 5μg/ml Bocilin FL and incubated for 20 min at RT. To remove the excess Bocillin FL, the samples were pelleted and resuspended in 100 μl PBS with cOmplete Mini Protease Inhibitor Cocktail (Sigma) and 2 U DNAse I (NEB). Then sonicated on ice for 10 sec and boiled 15' at 99˚C after adding 25 μL of SDS-PAGE 5x loading buffer and resolved by SDS-PAGE. Fluorescent signal was observed using a home-made imager for Midori green, followed by visualisation of gels by Coomassie staining.

In vitro PG synthesis assay with radiolabelled lipid II
The assay was published by Bertsche et al., [104]. Lipid II (1.2 nmol, 11,000 dpm) was vacuum dried and dissolved in 5 μl of methanol. The reaction was performed in a total volume of 50 μl in 25 mM Tris/HCl, 10 mM MgCl 2 , 100 mM NaCl, 0.05% Triton X-100, pH 7.5 (buffer B), with 0.5 μM PBP1A, 1 μM LpoA, 1 μM PBP4(S62A), 2 μM NlpI. The reaction was performed for 1 h at 30˚C. The pH of the samples was adjusted to 4.8 prior to boiling for 5 min followed by a digestion with 50 g/ml cellosyl for 1 h at 37˚C and boiling for 5 min. Cell division cycle age timing of FtsZ and FtsN for the ΔdacB strain expressing PBP4 variants from plasmid without induction and its parental BW25113. Because BW25113 cannot be grown to steady state, the difference in divisome assembly of the PBP4 variant should be interpretated as not identical to that of the wild type in the case of S63G and D155A, ignoring the absolute numbers. (TIF) S13 Fig. PBP4 or NlpI do not affect the activity of PBP1A or PBP1A/LpoA. Representative HPLC chromatograms of samples from an in vitro PG synthesis assay of PBP1A and radiolabelled Lipid II, in the presence of proteins indicated. The main PG products upon digestion with the muramidase cellosyl are the disaccharide pentapeptide (Penta) and the bis-disaccharide tetrapentapeptide (TetraPenta). The % peptides present in cross-links is quantified on the right side and presented as mean ± variation of two independent experiments. (TIF) S14 Fig. PBP4 or NlpI do not affect the activity of PBP1A whereas LpoA reduces the activity of PBP4. (A) Summary of the results from an in vitro PG synthesis assay with PBP1A (in the presence or absence of LpoA) and radio-labelled lipid II substrate. The presence of NlpI, catalytically inactive PBP4(S62A) or both together, does not affect the cross-linkage of the PG product of a PBP1A or PBP1A/LpoA reaction (left side), and NlpI, PBP4(S62A) or both together, do not induce enhanced carboxypeptidase activity of PBP1A, which would be seen as a higher content of tetrapeptides (right side). Values are mean ± variation of two independent repeats. Example chromatograms are shown in S12B Fig. (B) The presence of LpoA modestly reduces the activity of PBP4 in a DD-carboxypeptidase assay using the substrate UDP-MurNAc pentapeptide (C) LpoA also reduces slightly the activity of PBP4 against PG sacculi from BW25113, as seen by the reduced digestion of TetraTetra upon a 30 min incubation period.