Sequences of the LytM genes in Supplementary Table 1 were retrieved by pBLAST searches on the Integrated Microbial Genomes and Microbiomes (IMG/M) database (Chen et al., 2017). Multiple alignments were achieved with MUSCLE (Edgar, 2004) and manually adjusted and visualized with Jalview (Waterhouse et al., 2009). Before performing phylogenetic reconstruction, the sequences were truncated to only the conserved LytM domain. The LytM domain from S. aureus (GenBank ID: MBH4889575.1) was used in alignments to determine the boundaries of the conserved domain.

Phylogenetic reconstruction was performed using MEGA11 (Kumar et al., 2018; Stecher et al., 2020) to estimate consensus phylogenies and carry out bootstrapping analysis. The LytM gene tree was inferred by using the Maximum Likelihood method and Le Gascuel 2008 model (Le and Gascuel, 2008). A discrete Gamma distribution was used to model evolutionary rate differences among sites [four categories (+G, parameter = 1.6678)]. The rate variation model allowed for some sites to be evolutionary invariable [(+l), 0.00% sites]. This analysis involved 178 amino acid sequences (Supplementary Table 1). The tree was visualized and formatted using iTol (Letunic and Bork, 2016). Presence and absence of LytM clade members was determined from the tree and tabulated in Supplementary Table 2.

Full length genes were screened for N-terminal secretory signal sequences using the SignalP-5.0 Server (Almagro Armenteros et al., 2019) and for transmembrane segments using TMHMM (Krogh et al., 2001). Additionally, TREND was used to quickly identify conserved protein domains (Gumerov and Zhulin, 2020) and DeepCoil was used to identify coiled-coil domains (Ludwiczak et al., 2019). The results of this analysis are summarized in Supplementary Table 1.

WebLogo 3 (Crooks et al., 2004) was used to plot the amino acid distribution at each position of the LysM domain. To create the alignments for logo generation, LytM sequences from the defined five clades (and the additional subgroup of active NlpD genes) in the tree were simultaneously aligned to the LytM domain from S. aureus (GenBank ID: MBH4889575.1) and any insertions causing gaps in the LytM_Sa sequence were removed for uniform comparison. This analysis included the 38 MepM, 36 EnvC, 31 LmdC, 27 NlpD, 6 NlpD (with conserved active site HXXXD, HXH), and 21 LpdB LytM domains from the genes listed in Supplementary Table 1.

Agrobacterium tumefaciens C58 and derived strains were grown in LB rich medium or ATGN (0.5% glucose) minimal medium (Morton and Fuqua, 2012) at 28°C with shaking. When appropriate, antibiotics were used at the following working concentrations: kanamycin 300 μg/ml and gentamycin 200 μg/ml. When indicated, IPTG was used as an inducer at a concentration of 1 mM. In some cases, cumate was used as an inducer at a concentration of 0.1 mM. E. coli DH5α and S17-1 were routinely cultivated on LB agar or liquid LB medium at 37°C with shaking. When needed antibiotics were used at the following concentrations: kanamycin 50 μg/ml and gentamycin 20 μg/ml. All strains used in this study are listed in Supplementary Table 3. Unless otherwise specified, the experiments were conducted in LB.

All strains and plasmids used are listed in Supplementary Table 3 and primers are listed in Supplementary Table 4. To construct expression vectors containing lmdC-sfgp, rgsM-sfgfp, dipM-sfgfp, and envC-sfgfp, the corresponding coding sequence was amplified from purified A. tumefaciens C58 genomic DNA without a stop codon. The amplicons and plasmids pSRKKM-Plac-sfgfp, pSRKKM-Cym-sfgfp, pSRKKM-T7-sfgfp, pSRKKM-PenvC-sfgfp, were digested overnight and ligated overnight at 4°C using NEB T4 DNA ligase. Ligations were transformed into E. coli DH5α and purified plasmids were sequenced to verify expected translational fusions. Atu1832 was amplified from genomic DNA and inserted into pSRKKM-Cym-sfgfp linearized with NdeI and BamHI using Gibson cloning (NEB) according to manufacturer’s protocols.

For complementation of ΔenvC, the EnvC native promoter was cloned into the pSRKKM plasmid. Complementation of ΔamiC and ΔamiD was achieved by constitutively expressing amiC and amiD under the T7 medium promoter. The T7 medium promoter was introduced into EcoRI and NdeI sites of pSRKKM using the oligo annealing/ligation method and the T7forwardEcoRINdeI and T7revEcoRINdeI primers. Lastly, dipM with a stop codon was ligated into pSRKKM-Ptac-sfgfp to allow conditional overexpression of dipM from the Ptac promoter in the presence of IPTG.

Vectors for gene deletions by allelic exchange were constructed using recommended methods for A. tumefaciens (Morton and Fuqua, 2012). Gene deletions were achieved by allelic exchange and vectors were constructed as previously described (Howell et al., 2019). Briefly, 500 bp fragments upstream and downstream of the gene of interest (envC, amiC, ampD) were amplified using primer pairs P1/P2 and P3/P4. Overlapping PCR was used to the amplicons generated by P1/P2 and P3/P4, using primer pair P1/P4. The amplicon was digested and ligated into pNTPS139. Ligations were transformed into E. coli DH5α and purified plasmids were sequenced to verify expected inserts. The deletion plasmids were introduced into A. tumefaciens by mating using an E. coli S17 conjugation strain to create kanamycin resistant, sucrose sensitive primary exconjugants. Primary exconjugants were grown overnight in media with no selection. Secondary recombinants were screened by patching for sucrose resistance and kanamycin sensitivity. Colony PCR with primers P5/P6 for the respective gene target was used to confirm deletion. PCR products from P5/P6 primer sets were sequenced to further confirm deletions.

For the insertional knock-out of Atu1832, nucleotides 1172–1771 of Atu1832 were amplified from genomic DNA and cloned into pMCS-2 (Thanbichler et al., 2007) linearized with EcoR1 using Gibson cloning. Integrants were isolated by antibiotic selection and checked for correct insertion of the plasmid using primers upstream of this sequence and inside the plasmid.

To construct the DipM and RgsM depletion strains, the target gene was amplified, digested and ligated into the pUC18-mini-Tn7T-GM-Ptac and pUC18-mini-Tn7T-GM-Plac, respectively. The mini-Tn7 vector, along with the pTNS3 helper plasmid, were introduced into C58ΔtetRA:a-attTn7 as described previously (Figueroa-Cuilan et al., 2016). Transformants were selected for gentamycin resistance and insertion of the target gene into the a-att site was verified by colony PCR using the tet forward and Tn7R109 primers. PCR products were sequenced to confirm insertion of the correct gene. Next, the target genes were deleted from the native locus as described above in the presence of 1 mM IPTG to drive expression of the target gene from the engineered site. Additional deletions were introduced into the DipM depletion strain following the protocol described above with IPTG present during all steps.

Exponentially growing cells (OD600 ∼ 0.6) were immobilized on 1.25% LB or ATGN agarose pads as described previously (Howell et al., 2017). Phase contrast, differential interference (DIC) and epifluorescence microscopy was performed with an inverted Nikon Eclipse TiE equipped with a QImaging Rolera Em-C2 1K EMCCD camera and Nikon Elements Imaging Software. For time-lapse imaging, cells were imaged every 5 or 10 min for the duration of the experiment. Time-lapse microscopy is conducted at room temperature (∼23°C).

To construct demographs of LmdC-GFP, RgsM-GFP, EnvC-GFP, or DipM-GFP localization (Figure 2), cells (∼100 to 500) were imaged for each strain. A GFP channel profile was taken along the medial axis for each imaged cell. These medial axis profiles were then aligned by cell length using MicrobeJ software (Ducret et al., 2016).

Exponentially growing cells were imaged using phase microscopy as described above. Bacterial cell length was detected by MicrobeJ software whereas principal component analysis (PCA) to identify shape variations was performed using CellTool (Pincus and Theriot, 2007). Phase contrast images were converted and edited to binary masks in FIJI (Schindelin et al., 2012) before using CellTool.

For cell viability spot assays, cultures were grown overnight and diluted to an OD600 = 0.05 and serially diluted in LB or ATGN agar plates. Four microliters of each dilution was spotted and plates were incubated at 28°C for 36 h (LB) or 48 h (ATGN) before imaging. When appropriate plates contained 300 μg/ml kanamycin and/or IPTG at 1 mM as indicated in figure legends. For the RgsM, DipM, or DipM-derived depletion strains, cells were grown in liquid media in the presence of 1 mM IPTG to an OD600 = 0.6. The cells were washed three times by centrifugation in media to remove the inducer and resuspended to an OD600 = 0.05 in media. Serial dilutions were performed, and 4 μl of each dilution was spotted onto and plates were incubated at 28°C for 36 or 48 h before imaging. Each condition was tested in duplicate during three independent experiments (n = 3).

For the SDS susceptibility assay, exponentially growing cells were diluted to an OD600 = 0.5. A total of 0.25 ml of each culture was aliquoted into 1.5 ml tubes and centrifuged at 4.5 × g for 5 min. Cells were resuspended in 0.25 ml of various concentrations (0.625, 1.25, 2.5, and 5%) of SDS solubilized in HEPES buffer at a pH 7.4, and incubated for 5 min. Untreated controls were resuspended in 0.25 ml of HEPES buffer only. Four microliter of each dilution was spotted onto LB and stored at 28°C for 24 h before imaging. When treating depletion strains with SDS, cells were pre-depleted as indicated, treated with SDS for 5 min and spotted on plates containing IPTG and no SDS. To determine the strain sensitivity to high salt, cell viability assays were conducted as described above on LB plates containing 2% NaCl.

Wildtype and ΔenvC strains were grown in LB or ATGN for ∼16 h to early stationary or late exponential, respectively, before being spun down and resuspended in fixative (2% paraformaldehyde, 2% glutaraldehyde in 100 mM sodium cacodylate buffer pH = 7.35). DipM depletion strains were grown as described and depleted for the indicated time before being collected, spun down, and resuspended in fixative. Unless otherwise stated, all reagents were purchased from Electron Microscopy Sciences and all specimen preparation was performed at the Electron Microscopy Core Facility, University of Missouri.

For whole cell TEM, fixed whole cells were rinsed by pelleting at 2,500 g and resuspending in water. Samples were placed on negatively charged carbon coated copper grids. A negative charge was applied to the grid using a PELCO easiGlow Glow Discharge Cleaning System. Cells were allowed to settle on prepared grids for 2 min before removing the water solution and replacing it with 1% aqueous uranyl acetate for 2 min before drying and imaging. Images were acquired with a JEOL JEM 1400 Transmission Electron Microscope (JEOL, Peabody, MA, United States) at 80 kV on a Gatan Ultrascan 1000 CCD (Gatan, Inc., Pleasanton, CA, United States).

For SEM, fixed whole cells were plated overnight on cell culture treated coverslips to ensure adhesion and rinsed with 100 mM sodium cacodylate buffer, pH 7.35 containing 130 mM sucrose. Secondary fixation was performed using 1% osmium tetroxide (Ted Pella, Inc., Redding, CA, United States) in cacodylate buffer using a PELCO BioWave (Ted Pella, Inc., Redding, CA, United States) operated at 100 W for 1 min. Specimens were next incubated at 4°C for 1 h, then rinsed with cacodylate buffer and further with distilled water. Using the PELCO BioWave, a graded dehydration series (per exchange, 100 W for 40 s) was performed using ethanol. Samples were dried using the Tousimis Autosamdri 815 (Tousimis, Rockville, MD, United States) and samples were sputter coated with 5 nm of platinum using the EMS 150T-ES Sputter Coater. Images were acquired with a FEI Quanta 600F scanning electron microscope (FEI, Hillsboro, OR, United States) at 2 kV, Spot 7, with the Everhart–Thornley secondary electron detector.

For thin sectioning, each sample was centrifuged at 2,500 × g and the resulting pellet was resuspended in HistoGel (Thermo Scientific, Kalamazoo, MI, United States). Next, fixed pellets were rinsed with 100 mM sodium cacodylate buffer, pH 7.35 (Sigma Aldrich, St. Louis, MO, United States), and 130 mM sucrose. Secondary fixation was performed using 1% osmium tetroxide (Ted Pella, Inc., Redding, CA, United States) in 2-ME buffer using a PELCO BioWave (Ted Pella, Inc., Redding, California) operated at 100 W for 1 min. Specimens were next incubated at 4°C for 1 h, then rinsed with cacodylate buffer and further with distilled water. En bloc staining was performed using 1% aqueous uranyl acetate and incubated at 4°C overnight, then rinsed with distilled water. Using the PELCO BioWave, a graded dehydration series (per exchange, 100 W for 40 s) was performed using ethanol, transitioned into acetone, and dehydrated specimens were then infiltrated with EPON resin (250 W for 3 min) and polymerized at 60°C overnight. Sections were cut to a thickness of 75 nm using an ultramicrotome (Ultracut UCT, Leica Microsystems, Germany) and a diamond knife (Diatome, Hatfield, PA, United States). Images were acquired with a JEOL JEM 1400 Transmission Electron Microscope (JEOL, Peabody, MA, United States) at 80 kV on a Gatan Ultrascan 1000 CCD (Gatan, Inc., Pleasanton, CA, United States).

For monitoring DipM expression in LB and ATGN, 20 ml cultures were grown to an OD600 = 0.6 in the presence of IPTG. A 2-ml aliquot of the culture was removed from each culture, and the pellets were stored to serve as time zero (DipM-replete). The remaining 18 ml of cultures were washed three times with LB or ATGN via centrifugation to remove the IPTG. The washed pellets were resuspended and normalized to an OD600 of 0.4 and kept in exponential phase throughout the experiment. Aliquots were collected after 2, 4, 8, 16, and 24 h to monitor the levels of DipM production during the depletion. The cell pellets were stored at -80°C until further analysis. The cell pellets were incubated with 100 μl of a master mix containing 1 ml of BugBuster protein extraction reagent (Novagen) and supplemented with 1 (EDTA-free protease inhibitor cocktail (Sigma), 10 μl of lysonase (Novagen), 2,500 U/ml DNase I (Thermo Scientific), and 1 mM dithiothreitol (DTT) (Thermo Scientific) for 30 min with shaking at room temperature to lyse the cell pellets. The whole-cell lysates were clarified by centrifugation at 17,136 × g for 15 min. The BCA protein concentration kit (Thermo Scientific Pierce) was used to measure the total protein concentration of each sample. A final concentration of 1 × Laemmli buffer (BioRad) was added to the cleared cell lysates. Samples were boiled at ∼100°C for 5 min prior to loading equivalent amounts in each well of an SDS-10% PAGE gel. The separated proteins were electroblotted onto polyvinylidene difluoride (PVDF) membranes (Bio-Rad) and blocked for 1 hour in 5% non-fat dry milk powder solubilized in 1% TBST (1 × Tris-buffered saline, 1% Tween 20). The blocked PVDF membranes were probed primary antibody (Caulobacter crescentus α-DipM diluted 1:2,000) for 1.5 h in 2.5% milk-TBST, followed by incubation with secondary antibody (horseradish peroxidase-conjugated goat anti-rabbit at a dilution of 1:10,000) for 1 h in 2.5% milk-TBST. The secondary antibody was detected using the ECL Plus-HRP substrate (Thermo Scientific Pierce).

For Figure 3G, DIC time-lapse microscopy of cells growing on agarose pads (as described above) was used to monitor cell cycle progression. Typical cell cycle maintenance indicates that an individual cell elongates, constricts at mid-cell, and divides with growth resuming at new poles generated by cell division. Cell growth inhibition indicates that the cell stops growing; cell division inhibition indicates that cells do not properly constrict or divide; and polarity defect indicates that growth appears to occur from an old pole. The % of cells displaying each cell cycle phenotype was calculated by counting the number of cells displaying the indicated phenotypes and dividing it by the total number of cells analyzed for each strain. For Figure 4E, DIC time-lapse microscopy was employed to track growth poles for 1–3 generations. In WT cells, elongation occurs at the new poles generated by cell division in both mother cells and daughter cells. Mother cells are defined as the cells which inherit the oldest (non-growth) pole when a cell divides whereas daughter cells inherit the former growth pole. Cells resuming growth at the newly formed cell poles were categorized as growing through the new pole. Cells resuming growth at the pole opposite to the new pole generated by cell division were categorized as resuming growth at the old pole.

The A. tumefaciens genome encodes four open reading frames that contain LytM (M23) enzymatic domains. Because of the wide diversity of genes containing LytM domains among Proteobacteria, it can be difficult to infer whether a given factor should have an ancestral or species-specific role. We took a phylogenetic approach to place the four LytM factors within a greater functional and evolutionary context: we collected all detectable LytM factors from representatives of all the proteobacterial classes, including some deep-branching proteobacterial classes, and constructed a maximum likelihood gene tree using only the LytM domains (Figure 1A). In this gene tree, five clades emerged that were distinguished by the conservation of N-terminal protein domains outside of the LytM domain in the full length genes (see also Supplementary Figures 1, 2). All of these clades have characterized members, and we named the clades using the most characterized member. In general, almost all surveyed bacterial representatives encoded orthologs of EnvC, LmdC, and MepM whereas Csd and NlpD orthologs were constrained to or have been lost in certain bacterial classes (Figure 1B). Divisome components NlpD and EnvC were further distinguished from other LytM factors and each other by dLytM domains with distinct inactivation patterns of the catalytic site (Figure 1A and Supplementary Figure 3). LytM endopeptidases belong to the large family of lysostaphin-type metalloenzymes that conserve a zinc chelating active site that catalyzes the cleavage of peptide bonds (Firczuk and Bochtler, 2007). This catalytic site is defined by the conserved signature sequence HXXXD, HXH (Bochtler et al., 2004; Firczuk et al., 2005; Grabowska et al., 2015). Crystal structures of LytM domains with substitutions at these positions do not contain the Zn²⁺ cofactor (Peters et al., 2013; An et al., 2016; Cook et al., 2020). The inactivation of the catalytic site has been well-documented among various studies of dLytM factors in the Proteobacteria and it has been shown in various members that these factors activate cognate amidases rather than directly hydrolyzing peptide bonds themselves (Uehara et al., 2009; Stohl et al., 2016; Yang et al., 2018; Gurnani Serrano et al., 2021).

The A. tumefaciens genome encodes four of the five major LytM factor classes detected in our approach. The genome encodes two enzymatically active LytM endopeptidases and two dLytM factors. It does not encode a member of the LpdB clade, which is particularly conserved in the Delta- and Epsilonproteobacteria and Spirochaetia (Figures 1A,B). This clade features no distinguishing domains by sequence analysis and we were unable to determine if any orthologs of members of this clade have been well-characterized. Currently this clade is named for the C. crescentus ortholog that exhibited no phenotype when deleted (Zielińska et al., 2017).

Agrobacterium tumefaciens encodes one LmdC ortholog (Atu1832). The LmdC clade features an N-terminal transmembrane (TM) segment, one to two coiled coil domains, and an active LytM domain (Figure 1A). The LmdC clade has representatives from most classes but is particularly conserved in the Alpha- and Deltaproteobacteria and deep-branching proteobacteria (Figure 1B). This clade has few characterized representatives but is essential in H. neptunium where it could contribute to growth and division of daughter cells that bud from the reproductive prostheca (Cserti et al., 2017). It has also been implicated in prostheca (stalk) biogenesis in C. crescentus (Billini et al., 2019). Our phylogenetic analysis indicated that in all alphaproteobacterial orders, with the exception of Rhizobiales, the coiled-coil domain is preceded by a DUF5930 domain (Figure 1A). Overall, our analysis supports the idea that this gene is broadly conserved with specialized roles in the Alphaproteobacteria that has undergone further derivation in the Rhizobiales. To visualize the localization of this ortholog in A. tumefaciens, we constructed an cumate-inducible plasmid expressing LmdC_Atu with monomeric superfolder GFP fused to its C-terminus. After inducing overnight, the cells showed low, diffuse fluorescence (Figure 2A). Attempts at fusing msfGFP at the native locus resulted in even lower levels of diffuse fluorescence (data not shown). An insertional knock-out of this gene exhibited no effect on cell viability or any morphological phenotype when grown in LB or ATGN (Figures 2E–G).

Agrobacterium tumefaciens encodes one MepM ortholog (Atu4178). This broadly conserved active LytM clade of DD-endopeptidases is generally distinguished by an N-terminal Csd3 inactivation domain (Figures 1A,B), which has been shown to obstruct the substrate binding site of the Helicobacter pylori MepM ortholog Csd3 in its crystal structure (An et al., 2015). MepM has been genetically associated with the elongasome in E. coli (Park et al., 2020) whereas the ortholog in the alphaproteobacterium Sinorhizobium meliloti (named “RgsM”) is a component of the polar elongasome (Krol et al., 2020). In agreement with S. meliloti and A. tumefaciens being close relatives and having polar elongation machinery in common, Atu4178 (RgsM_Atu) had similar properties as RgsM_Sm: RgsM_Atu-msfGFP was mostly diffuse with some observable foci at mid-cell and at one pole (Figure 2B). After several attempts to delete RgsM_Atu, we concluded it was essential, in agreement with its identification as an essential gene from high throughput transposon mutagenesis sequencing (Curtis and Brun, 2014). Thus, we constructed an RgsM depletion strain. In comparison to WT cells (Supplementary Video 1), depletion of RgsM inhibited polar elongation resulting in shortening of cells by 20 h (Figure 2E and Supplementary Video 2). Analysis of cell length indicated a significant decrease in cell length after depletion in comparison to WT or RgsM replete (+RgsM) cells (Figure 2F). Depletion of RgsM (−RgsM) results in a 5-log loss of viability confirming its essentiality (Figure 2G). The depletion of RgsM phenocopies the depletion of PBP1a, the major cell wall synthase driving polar elongation (Williams et al., 2021). These observations are consistent with characterization of RgsM_Sm (Krol et al., 2020) suggesting that RgsM orthologs may function as a space-making DD-endopeptidases involved in the polar growth of Rhizobiales.

Agrobacterium tumefaciens conserves both divisome associated dLytM clades, EnvC (Atu2775) and NlpD (Atu1700). The EnvC protein architecture is characterized by a cleavable signal sequence, two or more coiled-coil domains, and an inactivated LytM domain (Figure 1A). The wide conservation of this gene suggests a common role in the proteobacterial divisome, where it activates cognate amidases, such as AmiAB in E. coli (Uehara et al., 2010) and AmiB in V. cholerae (Möll et al., 2014). EnvC_Atu-msfGFP expressed from a plasmid under native promoter control localized to mid-cell during cell division, suggesting that this role is also conserved in Agrobacterium (Figure 2C). However, deleting envC_Atu resulted in shorter, hypercurved cells with a viability defect (Figures 2E–G and Supplementary Video 3). This phenotype differs from what has been reported for ΔenvC strains in most other proteobacteria, where the deletion phenotype has been associated with cell separation defects, chaining, and often little to no viability loss (Hara et al., 2002; Uehara et al., 2009; Cserti et al., 2017; Zielińska et al., 2017).

Members of the NlpD clade contain a lipidation signal sequence for outer membrane anchoring, one or more PG-binding LysM domains that have been shown to help localize various NlpD orthologs (Goley et al., 2010; Möll et al., 2010; Poggio et al., 2010; Tsang et al., 2017), and a dLytM domain with a different inactivation pattern (two of the four chelating residues are generally conserved) (Figure 1A). In contrast to EnvC, this clade, as defined by its dLytM domain, appears to be constrained to the Alpha-, Beta-, and Gammaproteobacteria, while a few deep-branching proteobacteria have genes with a similar genetic architecture but active LytM domains (Figures 1A,B). NlpD has been extensively studied in E. coli where it activates AmiC and plays a role in outer membrane invagination during division (Uehara et al., 2009, 2010; Tsang et al., 2017). In C. crescentus, the NlpD ortholog “DipM” has been linked to cell separation and outer membrane invagination albeit with different morphological phenotypes than those typified by E. coli (Collier, 2010; Goley et al., 2010; Möll et al., 2010; Poggio et al., 2010; Zielińska et al., 2017). Expression of dipM_Atu-msfGFP from a constitutive promoter on a pSRK plasmid indicated that localization was cell cycle dependent, localizing to mid-cell in pre-divisional and dividing cells and to the new pole in non-dividing cells (Figure 2D). This is a typical localization pattern for divisome factors in A. tumefaciens, such as FtsZ and FtsA (Howell et al., 2019), supporting DipM_Atu’s role as a divisome component. However, we determined that dipM was essential, unlike NlpD orthologs in other proteobacteria. After multiple attempts to delete dipM, we constructed a depletion strain for DipM. Depletion of DipM resulted in filamented single cells with mid-cell bulges and loss of viability (Figures 2E–G and Supplementary Video 4). Similar to our observations for ΔenvC, this phenotype was unexpectedly different from other proteobacteria, where NlpD is non-essential and the single mutant phenotype is either silent such as in E. coli (Uehara et al., 2009, 2010) and V. cholerae (Möll et al., 2014), or associated with cell separation defects such as in N. gonorrhoeae (Stohl et al., 2016), H. influenzae (Ercoli et al., 2015), or C. crescentus (Collier, 2010; Goley et al., 2010; Möll et al., 2010; Poggio et al., 2010; Zielińska et al., 2017). Given the uncharacteristic phenotypes of the divisome-associated LytM orthologs in A. tumefaciens, we focused on how these factors may function differently in this polar-growing Alphaproteobacterium.

Instead of a cell separation defect, which is expected for an ortholog of an amidase activator that functions in the final steps of division, ΔenvC cells exhibited hypercurvature that was readily detectable in both phase contrast (Figure 2E) and DIC (Figure 3A and Supplementary Video 3) in LB. Cells appeared curled and at times smaller than WT cells with no signs of cell separation defects. Similar, but less severe effects were seen in ATGN, a defined glucose-based medium (Supplementary Figures 4A,B). Although we were able to isolate ΔenvC mutants on ATGN, the ΔenvC strain exhibited very low viability in LB (Figures 2G, 3B) and suppressors arose fairly easily (data not shown), suggesting that EnvC plays a critical role in cell physiology. The morphological defects of ΔenvC were complemented with a plasmid containing envC under native promoter control (Figures 3A,B). Although the cells did not exhibit observable cell separation defects, we tested other properties that have been associated with dLytM and amidase deletion strains in other proteobacteria, such as heightened susceptibility to detergents and salt concentrations (Möll et al., 2014; Zielińska et al., 2017). ΔenvC was more sensitive to SDS and changes to salt concentrations (Figure 3C), suggesting that the defect may also include altered outer membrane stability and/or induce envelope stress. In addition, overproduction of EnvC resulted in a loss of viability (Figure 3D), suggesting it may activate cell wall hydrolases.

Given the data that EnvC localized to mid-cell during division and that ΔenvC exhibited altered outer membrane stability, we used TEM to investigate the subcellular effects of the mutant. Whole cell TEM of ΔenvC cells grown overnight indicated that the cells were hypercurved, bent, or kinked (Figure 3E). At low frequency in LB, some cells exhibited even more pleomorphic traits (Supplementary Figure 4C). The kinked or bent morphotypes were more frequently observed in ΔenvC cells growing in ATGN as cells are generally longer and generate a more uniform rod morphology in this medium (Supplementary Figure 4D). In these cells, rather than being gradually curved along the entire cell length, the long axis appeared to be disrupted at one point that skewed the direction/alignment of elongation. This morphological defect may indicate that during the cell cycle, polar growth has become askew or misdirected from its normal polar trajectory that creates reliably straight rods. Thin section TEM indicated that cells did not exhibit any separation or outer membrane defects (Figure 3F), even in the most severe cases (see also Supplementary Figures 4E,F). Although dividing cells were rarely captured in our preparations of whole cells or thin sections, the few cases we did observe did not display any abnormalities in envelope structure (Supplementary Figure 4F). These data confirmed that loss of EnvC did not result in any typical cell separation defects, such as chaining or incomplete constriction of the PG and outer membrane layers, which is in marked contrast to what has been shown for other proteobacteria.

To observe how the ΔenvC mutant grew and divided in real time, we performed time-lapse microscopy on cells growing on agar pads made with LB. The highly curved cell morphology made determination of division events difficult and hard to characterize (Supplementary Video 3). We were able to observe that the mutant displayed several defects in addition to hypercurvature that were generally associated with daughter cells (Figure 3G): a small percentage (about 10%) of cells elongated but did not appear to divide during the course of the experiment, indicating that division might be inhibited in a minority of cases. About 14% of the cells on the pad failed to elongate during the full course of the time-lapse experiment or after a division event, suggesting that despite division being possible, a portion of the population was unable to complete another cell cycle. Finally, about 25% of the cells exhibited a polarity defect: we were surprised to observe that after division, some daughter cells were seen to initiate elongation at the opposite pole than expected (Figure 3H). In WT cells, both the mother cell and daughter cell initiate growth at the pole produced by the division event, but in a subset of ΔenvC_Atu cells, the daughter cells instead grew from the other pole (Figure 3G). Together, these data suggest that, unlike other proteobacteria, loss of EnvC in A. tumefaciens has clear morphological defects that are not due to defects in cell separation but may instead be due to disruption of early polar growth patterning.

Disruption of polar growth patterning has been linked to loss of the AmiC and EnvC orthologs in S. meliloti (Krol et al., 2021), leading us to wonder if this relationship was conserved in A. tumefaciens. Indeed, ΔamiC cells exhibited a similar hypercurved phenotype, decreased viability on LB, and increased sensitivity to SDS and salt (Figures 4A–C and Supplementary Video 5). The ΔamiC phenotype was complemented with a plasmid carrying amiC under T7-medium constitutive promoter control (Figures 4A,B). Time-lapse microscopy indicated that ΔamiC cells displayed the same distinct growth behavior on agarose pads as ΔenvC_Atu (Figure 4D). The double mutant ΔenvCΔamiC exhibited a slightly more severe viability defect than ΔamiC but otherwise appeared to phenocopy the single mutants (Figures 3B, 4B,D and Supplementary Video 6). Together, this data suggested that, like in S. meliloti, EnvC and AmiC function in the same pathway.

We used CellTool (Pincus and Theriot, 2007) to run PCA on the mutant phenotypes to better compare them to WT and to each other (Supplementary Figure 5A). The largest deviation between the mutants and WT was attributed to cell curvature (shape mode 2, Figure 4E). Between mutants, the majority of the variance between ΔenvC and ΔamiC was explained by cell length (shape mode 1, Supplementary Figure 5B). Both ΔenvC and the double mutant were significantly shorter than WT and ΔamiC. Cell width (shape mode 3) contributed to a small amount of the morphological variance due to additive effects in the double mutant (Supplementary Figure 5C). While both single mutant strains were wider than WT, the double mutant strain was significantly wider than both. Overall, these differences support the idea that while EnvC and AmiC likely function in the same pathway (possibly by activation of AmiC by EnvC), EnvC appears to have additional roles in the cell and perhaps other regulatory partners.

All three mutants exhibited a similar polarity defect, in which the daughter cell that inherits the former growth pole initiates growth from either the new or old pole (Figures 4F,G). These observations are in agreement with data previously reported for S. meliloti (Krol et al., 2021). Like all alphaproteobacteria, Agrobacterium undergoes a dimorphic cell cycle with a non-motile mother cell that gives rise to a flagellated daughter cell. The cell cycle requires strict asymmetrical cellular patterning with unique poles. As a member of the polar-elongating Rhizobiales, during the cell cycle, the Agrobacterium mother cell grows polarly at the new pole, the machinery relocates to mid-cell to drive division, and the division event produces two new poles that are primed to be elongation poles (Figure 4G). In all three mutants, about half of the daughter cells do not elongate from the newly created pole after completing division, but rather from the opposite (old) pole.

Together, these strong converging phenotypes suggest that the EnvC ortholog may activate AmiC in A. tumefaciens, and, moreover, that A. tumefaciens can complete division without an amidase. AmpD (Atu2113) is the only other canonical amidase in the A. tumefaciens genome and has been previously shown to function in the PG recycling pathway (Figueroa-Cuilan et al., 2020). Moreover, ΔampD had no morphological defects and the ΔampDΔamiC double mutant morphologically phenocopied ΔamiC (Supplementary Figures 6A–C). We hypothesize that EnvC may activate AmiC; however, we find that the absence of AmiC does not bypass the toxicity induced by EnvC overexpression (Supplementary Figure 6D), consistent with an additional role for EnvC beyond the putative regulation of AmiC. Remarkably, AmiC no longer appears to be necessary for cell separation in Rhizobiales genera and instead it predominantly functions to help establish growth pole patterning in the daughter cell.

Like all NlpD orthologs studied so far, DipM_Atu-sfGFP localized to mid-cell during division (Figure 2A); unlike most NlpD orthologs studied so far, DipM_Atu was essential. Depletion of DipM led to highly pleomorphic defects and lysis in most cells after 16 h of growth in LB and a strong cell viability defect (Figures 5A,B and Supplementary Video 4). DIC imaging of DipM-depleted cells showed that cells exhibited different combinations of mid-cell bulging, branching, overall loss of width control, and lysis by 16 h. The onset of these morphological effects is delayed in ATGN due to increased time for sufficient depletion of DipM; nevertheless DipM-depleted cells showed severe loss of viability on ATGN media in spot assays. Western blot analysis confirmed that the depletion of DipM was more rapid in LB than ATGN medium (Supplementary Figure 7A), presumably due to the faster growth rate.

To compare DipM to known properties of NlpD orthologs in other proteobacteria, we determined its overexpression phenotype and sensitivity to detergent. We overexpressed dipM by driving dipM expression from a plasmid under the pTAC promoter in WT cells. Overproduction of DipM severely reduced cell viability (Figure 5C), in agreement with the idea that it may activate hydrolytic factors such as shown in Caulobacter (Goley et al., 2010; Möll et al., 2010). While both depletion and overproduction of DipM resulted in loss of viability and lysis, these strains manifested lysis differently in that the overexpression strain did not exhibit the extreme pleomorphic traits of the depletion strain (Supplementary Figures 6C,E). Both the depletion strain and the overexpression strains were sensitive to SDS (Figure 5D), suggesting that both hypo- and hyperactivity of DipM regulation networks impact outer membrane stability or permeability. All of these observations are in line with what has been demonstrated in other proteobacterial genera (Goley et al., 2010; Möll et al., 2010, 2014; Tsang et al., 2017).

Because the depletion of DipM resulted in pleomorphic defects, we tracked the onset of changes in cell morphology and division over time using multiple approaches. Phase-contrast images of cells undergoing DipM depletion in liquid culture indicated morphological changes accumulate over time (Figure 5E): early on, cells elongated and often formed mid-cell bulges, and at later time points most cells appeared to be inhibited in division and had undergone both mid-cell bulging and various kinds of branching. Overall, these observations suggested that cells experienced loss of regulation of PG synthesis at mid-cell at low DipM levels and that in the absence of DipM, division was completely inhibited and the initiation of elongation was no longer strictly at the new pole. Next, we used time-lapse DIC microscopy to observe cells undergoing DipM depletion on agarose pads. In these experiments, we pre-depleted DipM for 6 h in liquid to avoid cell crowding caused by division events prior to sufficient DipM depletion (Figure 5F and Supplementary Video 4). It is likely that tracking the depletion in time-lapse allowed us to see many events and certain morphological trajectories that were more vulnerable to lysis and lost in sample handling for liquid culture time points. We found that each cell had its own unique trajectory that made it hard to generalize this phenotype and have included one example in Figure 5F to outline some major themes (see also Supplementary Video 4, for other examples). Of note, the cell in Figure 5F never terminated polar growth before division events, leading to tip-splitting and bipolar growth (see daughter cell behavior at 10, 15 h), formed ectopic poles at mid-cell bulges (see mother cell behavior at 10, 15h), and underwent erratic and atypical division events. The different types of branching (tip splitting or ectopic pole formation) observed during DipM depletion have been observed previously: tip splitting with FtsZ depletions and branching from mid-cell with FtsA and FtsW depletions (Howell et al., 2019). Moreover, bulging at mid-cell occurs in the absence of FtsW or both PBP3a and PBP3b and has been interpreted as an indicator of inhibition or misregulation of sPG synthesis (Howell et al., 2019; Williams et al., 2021). The combination of these defects in the depletion of DipM suggests a central role for DipM in the redirection of PG synthesis to mid-cell and subsequent cell division.

We used various forms of electron microscopy to determine the effect of the DipM depletion at the subcellular level. TEM and SEM of whole cells indicated a similar pattern to that seen in phase-contrast and DIC images of cells undergoing DipM depletion (Figure 5G and see also Supplementary Figures 7B,C). At 8 h, we observed particularly long cells that exhibited multiple bulges, suggesting that if these bulges are mid-cell bulges, these cells might have scaffolded more than one division plane (mid-cell region) over the course of their cell cycles but failed to divide (Supplementary Figure 7B). SEM of cells depleted for 9 h indicated rounds of failed constriction as well as multiple mid-cell bulges (Figure 5G and Supplementary Figure 7D). The accumulation of phenomena associated with the division plane suggests that DipM may play an indirect role in early division, such as stabilizing the divisome, and that depleting DipM may indirectly destabilize nascent, constricting Z-rings.

It seemed possible that some of the branched cellular filaments created by depleting DipM could consist of chained cells, or that areas with constrictions might have completed cytokinesis of the inner membrane but were still linked by PG and the outer membrane. However, thin section TEM of cells at various time points indicated that even cells with massive morphological defects did not exhibit any abnormalities in their envelope layers along the cell length or at division sites (Figure 5H and see also Supplementary Figure 7C). Slices containing cells sectioned parallel to their long axis did not reveal any internal compartments separated by IM layers, verifying that cell chaining did not underlie the filamenting phenotype. These data indicate that the predominant effect of DipM depletion in A. tumefaciens was inhibition of division during early steps of constriction and not the final steps of septation, as shown for other proteobacteria.

Because the defects from depleting DipM could be associated with misregulation of septal hydrolases and because AmiC is directly regulated by the NlpD ortholog in E. coli (Uehara et al., 2010), we sought to determine any possible connection between these two genes in A. tumefaciens. We first wanted to determine if loss of AmiC would alleviate any of the manifold morphological defects of the DipM depletion. Depleting DipM in the ΔamiC background resulted in bloated, curved cells that appeared to exhibit a mixture of the two phenotypes with no improved outcomes in filamentation or viability (Figures 6A,C,E and Supplementary Video 7). In both time points of liquid culture and in time-lapse observations, it seemed possible that the mid-cell bulges were non-existent or reduced; however, the cumulative effects on cell width made this possible effect hard to determine. We also depleted DipM in the ΔenvC background to determine if there were any EnvC-specific effects on the DipM depletion phenotype. The outcome was similar to, if not worse than, depleting DipM in the ΔamiC background, with additive effects on loss of width control (Figures 6B,D,E and Supplementary Video 8). In both cases, the double mutants appeared to exhibit a combination of the predominant features of ΔamiC/ΔenvC and the DipM depletion.

The absence of EnvC or AmiC pathways exacerbated the DipM depletion phenotype suggesting that DipM functions in a pathway distinct from EnvC–AmiC or other EnvC pathways. We were therefore surprised to find that loss of AmiC completely alleviated the toxicity of dipM overexpression (Figure 6F and Supplementary Figure 6D). The absence of AmpD was not sufficient to relieve the toxicity of dipM overexpression (Supplementary Figures 6C,D). These results suggest that the toxic effects of dipM overexpression stem predominantly from crosstalk with the EnvC–AmiC pathway and is unrelated to its roles in regulating division or termination of elongation. Therefore, there is some overlap between the DipM and EnvC–AmiC pathways, but DipM’s other roles in division and regulation of termination of elongation obscure this relationship during DipM depletion.