Evolution of multicellular longitudinally dividing oral cavity symbionts (Neisseriaceae)


 In spite of the staggering number of bacteria that live associated with animals, the growth mode of only a few symbionts has been studied so far. Here, we focused on multicellular longitudinally dividing (MuLDi) Neisseriaceae occurring in the oral cavity of mammals and belonging to the genera Alysiella, Simonsiella and Conchiformibius. Firstly, by applying comparative genomics coupled with ultrastructural analysis, we inferred that longitudinal division evolved from a rod-shaped ancestor of the Neisseriaceae family. Secondly, transmission electron microscopy on cells and sacculi showed that, within each A. filiformis, S. muelleri or C. steedae filament, neighbouring cells are attached by their lateral cell walls. Thirdly, by applying a palette of peptidoglycan metabolic precursors to track their growth, we showed that A. filiformis septates in a distal-to-proximal fashion. In S. muelleri and C. steedae, instead, septation proceeds synchronously from the host-attached poles to midcell. Strikingly, based on confocal-based 3D reconstructions, PG did not appear to be inserted concentrically from the cell periphery to its centre, but as a medial sheet guillotining each cell. Finally, comparative genomics revealed MuLDi-specific differences that set them apart from rod-shaped members of the Neisseriaceae. These MuLDi-specific genetic differences comprise the acquisition of the amidase-encoding gene amiC2, the loss of dgt, gloB, mraZ (an activator of the dcw cluster), rapZ, and amino acids changes in 7 proteins, including the actin homolog MreB and FtsA. Strikingly, introduction of amiC2 and allelic substitution of mreB in the rod-shaped Neisseria elongata resulted in cells with longer septa. In conclusion, we identified genetic events that may have allowed rod-shaped Neisseriaceae to evolve multicellularity and longitudinal division. The morphological plasticity of Neisseriaceae together with their genetic tractability, make them archetypal models for understanding the evolution of bacterial shape, as well as that of animal-bacterium symbioses.


Introduction
Allometry of animal-microbe associations suggests that 10 25 prokaryotes thrive on animals and 10 23 on humans (Kieft and Simmons, 2015;Whitman et al., 1998) and, yet, the morphology and growth mode of animal symbionts are underexplored (Bulgheresi, 2016). Although many may form bio lms (see for example Buskermolen et al., 2016;Kosten et al., 2015), intestinal segmented lamentous bacteria (SFB; Hampton and Rosario, 1965;Jonsson et al., 2020;Schnupf et al., 2015) and three genera of Neisseriaceae that occur in the oral cavity (e.g., species belonging to the genera Alysiella, Simonsiella and Conchiformibius; Hedlund and Kuhn, 2006;Hedlund and Tønjum, 2015;Kuhn et al., 1978;Xie and Yokota, 2005), are the only known animal symbionts that may be regarded as multicellular, i.e. they invariably form stable laments of more than two cells. SFB occur in the small intestine of several animals and play a primal role in pathogen resistance and gut homeostasis (Ericsson et al., 2014;Schnupf et al., 2017). In contrast to SFB, multicellular oral cavity Neisseriaceae are relatively understudied. They are closely related to the other » 30 species of Neisseriaceae occurring, for the majority, in the buccal cavity of warmblooded vertebrates, they are cultivable and some are genetically tractable (Nyongesa et al., n.d. submitted publication; Veyrier et al., 2015). Besides multicellular, Neisseriaceae may be rod-shaped (e.g., Neisseria elongata) or coccoid (e.g., the human pathogen Neisseria meningitidis and Neisseria gonorrhoeae). A. liformis cells are 2 µm-long and 0.6 µm-wide on average and form upright-standing palisades on the squamous epithelium of the mouth, so that each cell has a proximal pole attached to the host epithelium and a distal, free pole (Figures 1, 2B). Furthermore, within each lament, A. liformis cells appear as paired. Concerning S. muelleri and C. steedae (previously known as Simonsiella steedae; Kuhn et al., 1978), they are thinner, but can be up to 4 and 7 µm-long, respectively. Unlike A. liformis, both poles of S. muelleri and C. steedae are attached to the mouth (Kuhn et al., 1978;Pangborn et al., 1977). This confers S. muelleri and C. steedae cells a curved (or crescent-shaped) morphology and we will henceforth refer to their host-attached poles as proximal and to their midcell as their most-distal region (Figure 1 and Figure 2C-D). Although mbriae were detected on the proximal pole of A. liformis (G. E. ) and on the host-proximal side of S. muelleri (previously referred to as ventral or concave side; Pangborn et al., 1977), up to this study, C. steedae mbriae localization pattern was unknown.
Besides multicellularity, another peculiarity of Alysiella, Simonsiella and Conchiformibius is that they divide longitudinally (Gary E Kuhn et al., 1978;Murray et al., 1965 and this manuscript). This is extraordinary, given that, except for nematode (Leisch et al., 2016(Leisch et al., , 2012, insect (Ramond et al., 2016) and dolphin symbionts (Dudek et al., 2021), rod-shaped bacteria typically elongate and divide by transverse ssion, two processes coordinated by the elongasome and divisome, respectively. In model bacteria, each of these machineries is constituted by over a dozen proteins, with the actin homologue MreB and the tubulin homologue FtsZ, respectively, orchestrating cell elongation and division (Szwedziak and Löwe, 2013): deletion of ftsZ results in lamentation (Bi and Lutkenhaus, 1991), whereas inactivation of mreB turned rods into cocci (Höltje, 1998;Veyrier et al., 2015). Even more striking was the effect of speci c amino acid changes: in MreB, they resulted in irregularly sized, bent or branched Escherichia coli cells (Shi et al., 2018(Shi et al., , 2017 and, when affecting FtsZ, they led to misplaced septa in E. coli, Bacillus subtilis and Streptomyces spp. (Addinall and Lutkenhaus, 1996;Monahan et al., 2009;Sen et al., 2019). Curiously, single amino acid mutations in the FtsZ-binding protein Ssg resulted in longitudinally dividing Streptomyces (Xiao et al., 2021). Collectively, these ndings led to the hypothesis that longitudinal division might have evolved from differential regulation of subtly different MreB and/or FtsZ variants (den Blaauwen, 2018;Thanbichler, 2018).
Here, we sought to nd out whether a similar path led to the evolution of Alysiella, Simonsiella and Conchiformibius -henceforth, collectively referred to as multicellular longitudinally dividing (MuLDi) Neisseriaceae -by applying comparative genomics to back-track their molecular evolution. To this aim, we closed the genomes of 21 out of 42 Neisseriaceae species to obtain a robust phylogeny and correlated it with both ultrastructural analysis and peptidoglycan (PG) mass spectrometry. This approach, which previously identi ed genetic events underlying the rod-to-coccus transition in the Neisseriaceae (Veyrier et al., 2015), revealed that also MuLDi Neisseriaceae evolved from a rod-shaped ancestor.
Moreover, incubation of A. liformis, S. muelleri and C. steedae with a palette of uorescent D-amino acids (FDAAs) revealed that these MuLDi Neisseriaceae employ a unique PG insertion pattern to grow.
Indeed, nascent septa cross the cells medially as to guillotine them -from the proximal to the distal pole in A. liformis, or from both poles to midcell in S. muelleri and C. steedae.
Finally, recapitulation of MuLDi-speci c allelic changes in the rod-shaped N. elongata resulted in longer septa, suggesting that the transition from transverse to longitudinal division required, at least, the deletion of mraZ -here shown to activate the Neisseriaceae division cell wall (dcw) cluster -the acquisition amiC2 and MreB amino acid permutations.
The capacity of oral cavity Neisseriaceae to have evolved -more than once -into coccoid or MuLDi cells from a rod-shaped ancestor, together with their amenability to cultivability and genetic manipulation, makes them ideal models to understand the evolution of bacterial cell division, as well as that of animal-bacterium symbioses.

Results
Core genome-based phylogeny of Neisseriaceae suggests that MuLDi evolved from a rod-shaped ancestor The Neisseriales order comprise the Chromobacteriaceae family and the Neisseriaceae family. Recently three new families have been suggested (Aquaspirillaceae; Chitinibacteraceae and Leeiaceae) (Chen et al., 2021). The Neisseriaceae family now includes 12 genera (Alysiella; Bergeriella; Conchiformibius; Eikenella; Kingella; Morococcus; Neisseria; Simonsiella; Snodgrassella; Stenoxybacter; Uruburuella; Vitreoscilla). We selected species from each of these Neisseriaceae genera and used SMRT (PacBio) and Minion (Nanopore) technologies to obtain 21 additional closed genomes (Table S1). Genomes obtained in this study were combined with Neisseriaceae draft genomes (n=365) from NCBI database to calculate the Average Nucleotide Identity (ANI) in order to identify the main species corresponding to genomes with ANI>95% (Table S2). We obtained 69 species for the construction of a core genome-based phylogeny using closed genomes or, if not, the draft genome for each species (Figure 1). Phylogeny results were similar to a recently published study (Chen et al., 2021). Of note, although most genomes available in the NCBI database originated from coccoid Neisseria (lineage 1; dark blue in Figure 1), the detailed phylogenetic analysis of this lineage, which likely evolved from an ancestral rod (Veyrier et al., 2015) and which include the well-known pathogens N. meningitidis and N. gonorrhoeae, will be presented elsewhere (Veyrier lab, in preparation).
Using Scanning-Electron Microscopy (SEM), we imaged all the species that are available in public collections to morphologically classify them as rod, cocci or MuLDi. Of note, we used sublethal concentrations of Penicillin G to test the elongation capacity of species that could not be unambiguously classi ed as rods or cocci by SEM, as previously described (Veyrier et al., 2015). This allowed us to con rm that all Neisseriaceae are rod-shaped (bacilli), except for two closely related species (N. wadsworthii and N. canis; henceforth referred to as coccoid lineage 2, light blue branches in Figure 1), which did not lengthen upon Penicillin G treatment. Remarkably, we found that coccoid species belonging to lineage 2 harbour genes encoding for the elongasome, but lost yacF/zapD (Figure 4), a major genetic event which also allowed the emergence of coccoid lineage 1 (Veyrier et al., 2015). Moreover, most species from the Chromobacteriaceae family (which, as aforementioned, belongs to the Neisseriales order) are also described as typical rod-shaped cells (Adeolu and Gupta, 2013), which suggests that the shape of the ancestor of all Neisseriaceae was a rod.
Collectively, our phylogenetic analysis indicates that two lineages of cocci (coccoid lineages 1 and 2) evolved independently from a rod-shaped ancestor and that two lineages of MuLDi evolved from a rodshaped ancestor, Simonsiella/Alysiella and Conchiformibius, referred to as MuLDi lineage 1 and 2, respectively.
MuLDi Neisseriaceae are attached by their lateral cell walls and harbour a characteristic signature in their muropeptide composition Previous (Gary E Kuhn et al., 1978;Murray et al., 1965), as well as our microscopic analyses (see electron micrographs of cells shown in Figure 1, Figure 2, Figure S1b-d and Supplementary Movies 1-4) suggested that A. liformis, S. muelleri and C. steedae laments result from incomplete cell separation. Moreover, Nile red staining con rmed the presence of membranes between adjoining cells (Figure S1e and f). To understand whether adjoining MuLDi Neisseriaceae share additional cellular structures, which prevent them to separate from one another, we performed transmission electron microscopy (TEM) of sacculi extracted from A. liformis, S. muelleri and C. steedae, as well as from the transversally dividing rod-shaped N. elongata, for comparison ( Figure 2 and Figure  S1a). We observed that, the sacculi of the three MuLDi symbionts, even upon extraction, remained attached, laterally, to one another (Figure 2b-d, bottom panels). We concluded that in the Neisseriaceae A. liformis, S. muelleri and C. steedae multicellularity results from adjoining cells attached by a cell-wall anked by two inner membranes.
We previously showed that a modi cation in the PG composition of the Neisseriaceae (increased proportion of pentapeptides) accompanied their rod-to-coccus transition (Veyrier et al., 2015). To nd out whether the rod-to-MuLDi transition would also correlate with a change in total muropeptide composition, we applied mass spectrometry to analyse the PG of three MuLDi A. liformis, S. muelleri and C. steedae, as well as that of 14 rod-shaped Neisseriaceae ( Figure S2). The abundance of dimers (Di), trimers (Tri) and tetramers (Tetra) relative to the abundance of monomers and the estimated total crosslinked were generally higher in MuLDi ( Figure S2 c and d). We concluded that, when compared to rod-shaped Neisseriaceae, MuLDi Neisseriaceae PG was more crosslinked ( Figure S2).
Alysiella liformis nascent septa guillotine the cells from their distal to their proximal poles Fimbriae-like structures were detected by TEM on the regions of A. liformis attached to oral epithelial cells (Gary E Murray et al., 1965;Pangborn et al., 1977). To con rm the presence of mbriae at the proximal pole, we immunostained them with an anti-mbriae antibody and found its signal to be localized at the proximal pole, consistent with the seminal ultrastructural data. Moreover, we noticed that, when observed at the epi uorescence microscope, the proximal, mbriae-rich side of each lament was invariably the convex one ( Figure S3 a-d), which allowed us to establish A. liformis polarity in the absence of mbriae localization in all the following microscopic analyses.
After con rming A. liformis polarity, we proceed to determine its growth mode by tracking PG synthesis by consecutively applying the three uorescent D-amino acids (FDAAs) HADA (blue), BADA (green) and TADA (red), which are labelled D-Ala residues incorporated into the peptide side chains of new PG. When imaged by epi uorescence microscopy, A. liformis cells sequentially labelled with HADA 30 min, BADA 15 min and TADA 15 min showed strongest uorescent signal at their septation planes. The virtual time-lapse obtained by the triple FDAAs labelling revealed that A. liformis starts to septate at the distal pole and that PG synthesis is continued unidirectionally toward the proximal cell pole (Figure 2 f, i, k and S4 a-c; Movie S5), consistent with what observed by thin section TEM ( Figure 2b). To view the PG insertion pattern in 3D, we performed confocal microscopy ( Figure 3). Surprisingly, the septal signal appeared as a sheet when viewed from the side, and, contrarily to what observed in other transversally (Bisson-Filho et al., 2017;Yang et al., 2017) or longitudinally (Pende et al., 2018) dividing bacteria, we did not observe PG discs or arcs at any septation stage.
In conclusion, we showed that A. liformis septation is asynchronous (i.e., it proceeds from the distal to the proximal pole) and non-centripetal, i.e., the PG appeared not to be inserted concentrically, from the periphery to the centre of the cell, but as a sheet that guillotined each cell from its distal to its proximal pole.
Simonsiella muelleri and Conchiformibius steedae septation starts at both poles synchronously and proceeds from the poles to midcell Based on previous ultrustructural studies, S. muelleri mbriae are situated on the cell side facing the epithelial cells (Murray et al., 1965;Pangborn et al., 1977), here referred to as the proximal side. To test whether this was also the case for C. steedae cells, we immunostained them with an anti-mbriae antibody and con rmed that mbrial appendages covered the proximal (concave) side of each lament ( Figure S3c and d).
To understand how they grow, we then tracked the synthesis of PG by subsequently applying HADA (blue), BADA (green) and TADA (red) to S. muelleri and C. steedae. When imaged by epi uorescence microscopy, S. muelleri and C. steedae sequentially labelled with HADA 30 min, BADA 15 min and TADA 15 min and with HADA 1h, BADA 45 min and TADA 45 min, respectively, showed strongest uorescent signal at the septation plane. However, the virtual time-lapse obtained by the triple FDAAs labelling differed from that obtained for A. liformis. Namely, both S. muelleri and C. steedae, appeared to start septation at both poles synchronously and PG insertion continued, bidirectionally, until midcell was reached (Figure 2g, h, j and k and Figure S4d-f).
To view the PG insertion pattern in 3D, we performed confocal microscopy on FDAA-labelled C. steedae (Figure 3c-e; Movie S6). At septation onset (Figure 3d), FDAA signal appeared as two juxtaposed triangular sheets, each emerging from one cell pole (green signal in septum 1, Figure 3d; red signal indicates the two leading edges). With septation progression, the two leading edges merged at midcell (red oval signal in Figure 3d, septum 2) and nally appeared as a circular disk at the very last septation stage (red signal in septum 3, Figure 3d and S5b).
Summarizing, we propose that the two curved oral symbionts S. muelleri and C. steedae start septation at each pole independently, but synchronously, and septation ends when the two poleoriginated PG sheets meet and merge at midcell.
Multiple genetic events associated with the cell shape transition from rod-shaped to MuLDi Neisseriaceae By applying exhaustive comparative genomics, we previously discovered that mutations at speci c genetic loci mediated the rod-to-coccus transition of the ancestor of pathogenic Neisseria (Veyrier et al., 2015). We therefore hypothesised that mutations at speci c genetic loci, had mediated their evolution from an ancestral, transversally dividing rod-shaped Neisseriaceae (Figure 1). Our approach was to detected shared events by the two MuLDi Neisseriaceae lineages (the Simonsiella/Alysiella lineage 1 and the Conchiformibius lineage 2), To identify these genetic loci, we applied previously described pipelines (Guerra Maldonado et al., 2020;Veyrier et al., 2009Veyrier et al., , 2015 to determine the presence/absence of proteins in 37 species of Neisseriaceae, 32 rod-shaped and 5 MuLDi (all displayed in Figure 1). Of note, we excluded both lineages of coccoid Neisseriaceae from our analysis, as they underwent a different evolutionary path (Veyrier et al., 2015).
Firstly, we identi ed 7 genes that were present in MuLDi, but absent in rod-shaped Neisseriaceae ( Figure  4). These genes comprised a gene encoding for an AmiC-like amidase, henceforth referred to as AmiC2. Interestingly, the amiC2 gene is always associated with cdsA, a gene encoding for the phosphatidate cytidylyltransferase CdsA in all MuLDi species ( Figure S6). As amiC2 and cdsA are either anked by a transposase (in the MuLDi lineage 1) or by a restriction/modi cation system (in the MuLDi lineage 2), we hypothesize that amiC2 was acquired by horizontal gene transfer, possibly from a Fusobacterium-related bacterium (see AmiC and AmiC-like phylogeny in Figure S7). Intriguingly, Fusobacteria, as the Neisseriaceae, are common members of the oral, gastrointestinal and genital ora (Brennan and Garrett, 2018). As for the remaining 6 MuLDi-speci c genes, four are predicted to encode for hypothetical proteins and two for the hemolysin transporter ShlB.
Secondly, we found that only 4 genes were absent in MuLDi Neisseriaceae when compared to rod-shaped ones ( Figure 4). Surprisingly, most of the genes lost during the rod-to-MuLDi transition are implicated in PG synthesis and cell division (Figure 4a). These genes include: mraZ and rapZ. mraZ is the rst gene of the dcw cluster ( Figure S6) in most bacteria, where it encodes for a poorly characterized, but highly conserved transcriptional regulator MraZ (Eraso et al., 2014;Fisunov et al., 2016;Mengin-Lecreulx et al., 1998). On the other hand, rapZ encodes the small RNA adaptor protein RapZ, implicated in cell envelope precursor sensing and signalling (Khan et al., 2020).
Thirdly and lastly, we used our software called CapriB (Guerra Maldonado et al., 2020) to search for amino-acid changes in the 438 proteins strictly conserved among the 39 Neisseriaceae species (core proteome) ( Figure 4b). Strikingly, we detected amino acid permutations in only 7 out of the 438 proteins (1.6%). Namely, three permutations and two permutations were found in FtsA and MreB, respectively, two proteins which are both involved in bacterial morphogenesis (Busiek and Margolin, 2015). Moreover, we found two permutations in the e ux pump membrane transporter MtrD and one permutation per protein in the DNA-directed RNA polymerase subunit RpoZ, single stranded DNA-binding protein Ssb, twocomponent regulator MisR and the long-chain-fatty-acid-CoA ligase FadD.
Altogether, comparative genomics of rod-shaped versus MuLDi Neisseriaceae identi ed 18 genetic loci at which mutations might have mediated rod-to-MuLDi transition. Notably, these genetic loci include mreB, encoding for the actin homologue, and amiC2, encoding for a cell wall amidase.

Downregulation of dcw cluster genes in MuLDi Neisseriaceae
As several genes encoding for regulators were mutated in MuLDi, we employed RNAseq to determine differential gene expression patterns between MuLDi (n=5) and rod-shaped (n=5) Neisseriaceae cultured in the same condition (GCB agar Media, 6h, 37°C 5% CO 2 ). To compare gene expression between species, we standardized the annotation of the ve rod-shaped and the 5 MuLDi Neisseriaceae genomes by inferring gene orthology using BlastP. Using NetworkX python programming package (Hagberg et al., 2008), we reannotated clusters of homologous genes in each genome (for example, ftsZ gene will be called NEISS_1241 in all the genomes). By doing so, we were able to count the reads associated with each gene in each species and perform DESeq2 statistical analyses using the core transcriptome. Strikingly, our analysis ( Figure 4c) showed that the majority of the signi cantly differentially regulated genes are involved in cell envelope synthesis (as demonstrated by their clustering in the String analysis shown in Figure 4d). Namely, we observed that 12 genes were upregulated in MuLDi species, most notably minE, ftsX and ftsY involved in cell division. More importantly, the 19 down-regulated genes in MuLDi species include ftsA, murE and ftsI, which are part of the dcw cluster.
To conclude, comparative RNA-seq suggests that the loss of mraZ in MuLDi Neisseriaceae impacted the expression of dcw cluster including ftsI.

Downregulation of dcw cluster genes in N. elongata mraZ deletion mutants
To test whether deletion of mraZ in the rod-shaped Neisseriaceae N. elongata could cause downregulation of dcw cluster genes (similarly to what observed in MuLDi Neisseriaceae, that naturally lost mraZ), we compared the transcriptomes of wild-type N. elongate and a mraZ deletion mutant thereof. This revealed that ve genes located downstream of mraZ (mraW, ftsL, ftsI, murE and murF) were downregulated (Figure 5a and b). These results were validated by quantitative real-time PCR (Figure 5c). Moreover, overexpressing mraZ (by inserting it, ectopically, after the nrq locus in the N. elongata ΔmraZ mutant) restored wild-type expression. Although the N. elongata ΔmraZ mutant did not display strong morphological defects (Figure 5d), ΔmraZ pilEp-mraZ N. elongata, which overexpresses MraZ, tended to be smaller ( Figure 5 d and e).
Collectively, we showed that mraZ is an activator of transcription of the rst ve genes of the N. elongata dcw cluster, similar to what was proposed for Mycoplasma spp. (Eraso et al., 2014).

Recapitulation of MuLDi-speci c genetic changes in the rod-shaped Neisseriaceae Neisseria elongate resulted in longer septa
After deleting mraZ, we tested whether changes at other MuLDi-speci c loci could turn the rodshaped Neisseriaceae N. elongata in a MuLDi bacterium. However, as for the mraZ deletion, deletion of dgt, gloB, or rapZ either singularly or in combination did not change N. elongata morphology (data not shown). All the same, introduction of amiC2 (along with its neighbouring gene cdsA) in N. elongata did not result in signi cant shape or growth anomalies (Figure 6a), unless it was accompanied by the allelic exchange of N. elongata mreB with S. muelleri mreB, which resulted in longer cells (Figure 6a and Figure  S8).
In a nal attempt to turn the rod-shaped N. elongata into a MuLDi Neisseriaceae, we used an unmarked deletion-based technique developed by us (Nyongesa et al., n.d. submitted) to, concomitantly, delete dgt, gloB, mraZ and rapZ, substitute N. elongata mreB with S. muelleri mreB and introduce amiC2/cdsA. As shown in Figure 6 and Figure S8, N. elongata Ddgt, DgloB, DmraZ, DrapZ with mreB sm cells were longer and branched. More importantly, the substitution of mreB ne with mreB sm together with the introduction of amiC2/cdsA resulted in cells which had a longer septum and a shorter axis perpendicular to the septum ( Figure 6). Namely, the ratio between the two cell axes changed from 0.61 ± 0.25 (n=186), for the wildtype, to 0.95 ± 0.29 (n=174) for the mutant N. elongata.
All in all, even if our attempt to genetically manipulate the rod-shaped N. elongata into a MuLDi did not result into a complete transverse-to-longitudinal division switch (ratio between the two cell axes >1), the observed increase in septum length suggests that the genetic events identi ed by comparative genomics have participated in the rod-to-MuLDi transition in the Neisseriaceae.

Discussion
There is a huge discrepancy between the number of known prokaryotic species and how many of them have been characterized morphologically. This makes it hard to predict how the shape and the growth mode of bacteria evolved. In an attempt to ll this knowledge gap, we focused on MuLDi Neisseriaceae occurring in the oral cavity of warm-blooded vertebrates, including humans. Whole genome-based phylogenetic analysis, coupled with ultrastructural analysis, indicated that MuLDi Neisseriaceae evolved from a rod-shaped, Neisseriaceae. Although these rod-shaped septate transversally, our incubations with a set of uorescently labelled PG precursors showed that MuLDi Neisseriaceae septate longitudinally -in A. liformis in a distal-to-proximal fashion, in S. muelleri and C. steedae synchronously, from both poles to midcell (notably, the other two known species of the Alysiella and Conchiformibius genera, A. crassa and C. kuhniea, also septate longitudinally, the former unidirectionally and the latter bidirectionally; Figure  S4g). Irrespective of the uni-or bi-directionality of cell wall construction (that remains to be mechanistically deciphered by genomic comparison between Alysiella and Conchiformibius and Simonsiella), we further observed that in these bacteria, new PG was not inserted concentrically, but as a medial sheet guillotining each cell. Finally, full-scale comparative genomics revealed MuLDi-speci c differences that set them apart from rod-shaped members of the Neisseriaceae (e.g., amiC2 acquisition, mraZ loss and amino acid changes in the cytoskeletal proteins MreB and FtsA). Supporting the role of speci c genetic changes in the rod-to-MuLDi transition, introduction of amiC2 and allelic substitution of mreB in the rod-shaped Neisseriaceae N. elongata resulted in cells with longer septa.
Taken together, we presented two novel modes of septal growth and we identi ed genetic events that contributed to the evolution of bacterial multicellularity, longitudinal division and, possibly, hostpolarization in a group of mammalian symbionts.
Multiple phylogenetic studies have suggested that the wide palette of bacterial morphotypes we observe today evolved from rod-shaped bacteria, which makes us consider their shape as the ancestral one (Siefert and Fox, 1998;Young, 2006). Our genome-based phylogenetic reconstruction revealed that, even in the Neisseriaceae, MuLDi evolved from an ancestral rod-shaped bacterium. It remains uncertain whether these two MuLDi lineages have convergently evolved twice, or whether species belonging to the genus Kingella also evolved into MuLDi, but subsequently reverted to the rod-shaped morphology. We speculate that the MuLDi phenotype may have favoured colonisation of-or nutrient uptake from the buccal cavity which is characterized by rapidly shedding epithelial cells and salivary ow (Mark Welch et al., 2020). Indeed, multicellularity makes cooperation among cells possible in the form, for example, of division of labour and may, therefore, help bacteria to survive nutrient stress (see for example Claessen et al., 2014). Although, previous morphological studies suggested that the terminal cells of S. muelleri (Hedlund and Tønjum, 2015;Pangborn et al., 1977) and C. steedae (Kuhn et al., 1978) might phenotypically differ from the central ones and although we observed a thinner cell every approximately 14 cells in C. steedae (Figure S1, S5 and Movie S7), future studies are needed to clarify whether different cell types exist within each lament.
Multicellularity may arise via three distinct processes: aggregation of individual cells resembling the initial stages of bio lm formation (Monds and O'Toole, 2009), the formation of syncytial laments via crosswalls segmenting the mother cell, but not separating it into daughter cells (streptomycetes; Zhang et al., 2016) and incomplete cell ssion after cell division to produce chains of cells (referred to as clustered growth, e.g., lamentous cyanobacteria; Flores et al., 2019). TEM analysis of MuLDi sacculi revealed that these Neisseriaceae share one cell wall which makes them resemble to cyanobacteria. If MuLDi cells belonging to the same lament appear to be synchronized (Figure 2 and Figure S4), future studies are needed to nd out whether their cytoplasms are connected by septal junctions and/or hemidesmosomes (Flores et al., 2019).
Although longitudinal septation is clearly not a prerequisite of bacterial multicellularity (here de ned as clusters of at least 3 cells), these two phenotypic traits appeared to have evolved jointly in the Neisseriaceae. Longitudinal septation has also been shown in the nematode symbionts Candidatus T. oneisti and T. hypermnestrae (Leisch et al., 2016(Leisch et al., , 2012, as well as in the fruit y endosymbiont Spiroplasma poulsonii (Ramond et al., 2016). In these three unicellular symbionts, the tubulin homolog FtsZ localized at the septal plane and was therefore thought to mediate septal PG insertion. As for the actin homolog MreB, it was shown to form a medial ring-like structure in Ca. Thiosymbion throughout the cell cycle and to be required for septal FtsZ localization and PG insertion (Pende et al., 2018). Indeed, its pharmacological inactivation impaired both Ca. Thiosymbion growth and division (Pende et al., 2018). Although the localization pattern of MreB in MuLDi Neisseriaceae is currently unknown, its (1) presence in their genomes, (2) its transcriptional expression, (3) the identi cation of two MuLDi-speci c amino acid permutations (H185Q and T247A), and (4) the fact that introducing a MuLDi MreB in the rod-shaped N. elongata (with or without the concomitant insertion of the amiC2 gene) led to shape aberrations suggest that MreB is involved in PG insertion in MuLDi Neisseriaceae. Intriguingly, amino acid 185, located after the GVVYS motif, is substituted in MuLDi MreBs when compared to rod-shaped Neisseriaceae (H185Q), but also in longitudinally dividing Ca. Thiosymbion when compared to E. coli (S185N) (den Blaauwen, 2018). It should also be noted that the allelic substitution of MreB affected differently the morphology of N. elongata depending on the genetic background (i.e., presence or absence of other MuLDi-speci c genes). This, in addition to the pleiotropic effect of MreB reported in other studies (Shi et al., 2018), can make this protein accountable for accommodating multiple cell shape adaptations (e.g., rod-to-coccus, rod-to-MuLDi). If we still do not know whether MreB and/or FtsZ place the insertion of the PG synthesis machinery at the septum, based on our confocal-based 3D reconstructions, new septal PG is not inserted in successive, concentric rings or ellipses, as shown for model rods (Bisson-Filho et al., 2017;Yang et al., 2017) (Du and Lutkenhaus, 2019) and nematode symbionts (Pende et al., 2018), respectively.
In addition to MreB amino acid changes, MuLDi-speci c loss of mraZ led to the misregulation of the dcw cluster. mraZ has been described as a highly conserved transcriptional regulator of the dcw cluster, of which mraZ is the rst gene (Eraso et al., 2014;Fisunov et al., 2016). The dcw cluster is a group of genes involved in the synthesis of PG precursors and cell division (Ayala et al., 1994)  Throughout the Neisseriaceae, the dcw cluster consists of 14-16 tightly packed genes in the same orientation and mostly in the same order, with midA located before the cluster in reverse orientation ( Figure S6). The fact that, in the Neisseriaceae, the gene content and orientation of the dcw cluster mostly mirrored the phylogenetic placement of each species, suggests that the dcw cluster evolved vertically. Moreover, having a fragmented dcw cluster (as in the case of MuLDi Neisseriaceae and some Kingella species) does not seem to impact cell morphology, given that both rod-shaped and coccoid species may bear or not bear split dcw clusters. Of note, in spite of fragmentation, bacteria can retain some gene subclusters (e.g., "mraW-ftsL, ftsI, murE and murF", "ftsW, murG" and "murC, ddI, ftsQ, ftsA, ftsZ"), probably because the genes grouped in a given sub-cluster need to be co-transcribed (Mingorance et al., 2004). If several studies agree on the regulatory role of MraZ, its effect seems to vary depending on the organism.
In Neisseriaceae (this study) and in Mycoplasma spp. (Fisunov et al., 2016;Martínez-Torró et al., 2021) it is described as an activator of the dcw gene cluster whereas in E. coli, MraZ is described as a transcriptional repressor that controls its own expression and that of other dcw cluster genes (Eraso et al., 2014). Importantly, deletion or overexpression of MraZ led to cell lamentation in bacteria where MraZ, respectively, activates (e.g., Mycoplasma spp.) or represses (e.g., E. coli) the dcw cluster. If downregulation of the dcw cluster did not result in N. elongata lamentation, its upregulation, in MraZoverexpressing N. elongata, led to shorter cells. Altogether, our data suggest that, in the Neisseriaceae, MraZ is controlling cell elongation via the regulation of the dcw cluster and we speculate that it may have altered the balance between the divisome and the elongasome machineries (i.e., in mraZ-less MuLDi, the elongasome might contribute to PG synthesis in the septum).
Finally, comparative genomics highlighted the importance of the acquisition of the cdsA/amiC2 locus. Although its sole deletion does not result in morphological changes of N. elongata, when combined with the allelic substitution of mreB sm , we observed cells with longer septa. This suggests that the AmiC2 amidase may regulate MuLDi septation. Intriguingly, HPLC analyses of PG extracted from 17 rod-shaped bacteria and from three MuLDi Neisseriaceae (A. liformis, S. muelleri and C. steedae) showed that MuLDi PG is richer in M44 ( Figure S2), suggesting higher amidase activity in these Neisseriaceae. Concerning the amiC2-associated genetic locus cdsA, it encodes for a CDP-diacylglycerol synthase putatively implicated in phospholipid biosynthesis. Given that the presence of anionic phospholipids (cardiolipin and phosphatidylglycerol) has been shown to repel MreB (Kawazura et al., 2017), we can hypothesise that CdsA affects the composition of the membrane and, therefore, the localisation of MreB.
Despite all our efforts, we could not turn the rod-shaped N. elongata into a complete MuLDi Neisseriaceae even upon, concomitantly, replacing MreB, inserting amiC2/cdsA and deleting dgt, gloB, mraZ and rapZ. This could be due to the fact that we could not recreate all genetic events (such as replacing ftsA ne with ftsA sm due to its proximity to ftsZ) or due to the existence of other undetected events (such as species-speci c events that resulted in a convergent phenotype).
How could rod-shaped, transversally dividing bacteria evolve into longitudinally dividing ones? Permanent cell shape transitions may have resulted from modi cations (e.g., gene deletions, insertions and nucleotide polymorphisms) of genetic loci involved in morphogenesis (e.g., mreB, amiC2) and, additionally, in those involved in their transcriptional regulation (e.g., mraZ). Two evolutionary scenarios were proposed (den Blaauwen, 2018;Pende et al., 2018;Thanbichler, 2018): (1) an ancestral rod was compressed by its poles so that it got shorter and fatter, or (2) an ancestral rod rotated its septation axis by 90 degrees. Our results suggest that, in the course of evolution, the cell width of an ancestral rod increased (and its length decreased), perhaps following a misbalance between elongation and division. However, genetic tools are needed to gain insights on MuLDi Neisseriaceae evolution by, for example, visualizing the localization pattern of FtsZ and MreB or by attempting reversion into unicellular, possibly, transversally dividing bacteria such as N. elongata.
To date, most protein function studies have been conducted in either pathogenic or bacterial species that are easy to culture and manipulate in the laboratory such as E. coli and B. subtilis. In addition to these models, efforts to study other morphologies including commensal species are necessary to understand bacterial cell evolution, but also to increase the pool of protein targets (e.g., antibiotic targets) for industrial and biopharmaceutical applications. Throughout their evolution, Neisseriaceae succeeded in repeatedly, and seemingly effortlessly, evolve different cell shapes (e.g., coccoid, MuLDi). Moreover, they are the only known multicellular longitudinally dividing bacteria that may thrive in humans, but which are also cultivable and, likely, genetically tractable. We hence propose the use of Neisseriaceae as models to study how longitudinal division and multicellularity evolved as well as the molecular and cell biological mechanisms underlying the establishment of bacterium-animal symbioses.
Single colonies were subcultured into the respective liquid media with agitation at 120 rpm and grown to exponential phase (OD 600 0.1-0.6). For cloning experiments E. coli DH5α cells were cultured onto Luria-Bertani Media at 37 o C. When required, antibiotics were used as follows: kanamycin (50 μg/ml for E. coli; 100 μg/ml for N. elongata), erythromycin (300 μg/ml for E. coli; 3 μg/ml for N. elongata), chloramphenicol (25 μg/ml for E. coli; 5 μg/ml for N. elongata), and streptomycin (100 μg/ml for N. elongata). Transformation of N. elongata was done using linearized plasmid or PCR product by dropping approximately 500 ng of DNA on fresh cultures on GCB media supplemented with 10 mM MgCl 2 and incubated for 6-12 hours before subculturing on GCB media containing the appropriate antibiotics and Xgal if needed as described previously (Veyrier et al., 2015).

Time-lapse imaging of N. elongata
Strains were streaked from -70°C freezer stocks onto BSTSY agar plates and grown overnight at 37°C with 5% CO2. Single colonies were transferred to liquid culture and grown to exponential phase (OD 600 0.2). Cells were spotted onto pads made of 0.8% SeaKem LE Agarose (Lonza, Cat. No. 50000) in BSTSY and topped with a glass coverslip. Cells were transferred to an Okolab stage top chamber to control temperature (37°C) and gas (CO 2 5% and O 2 18%). Images were recorded with inverted Nikon Ti-2 microscopes using a Plan Apo 100X 1.40 NA oil Ph3 DM objective using Hamamatsu Orca FLASH 4 camera. Images were processed with NIS Elements software (Nikon). In all experiments, multiple x/y positions were imaged. Representative images were processed using the Fiji Software package.
Time-lapse imaging of A. liformis, S. muelleri and C. steedae Strains were streaked from -70°C freezer stocks onto PY (A. liformis), meat extract (S. muelleri) or BSTSY (C. steedae) agar plates grown overnight at 37°C with 5% CO2. Single colonies were transferred to liquid culture and grown to exponential phase (OD 600 0.2-0.5) at 37°C shaking at 180 rpm agitation. For all strains, 250 μL of diluted exponential phase cultures (OD 0.025) were loaded into the cell loading well of a prepared (shipping solution removed and washed three times with sterile appropriate media) B04A-03 micro uidic plate (Merck-Millipore). Time-lapse imaging was performed using CellASIC® ONIX Microfuidic System. The ONIX manifold was sealed to the B04A-03 plate. CellASIC® ONIX2 System was used as the micro uidics control software. First, a ow program was set up to prime ow channel and culture chamber by owing medium from inlet wells 1 to 5 at 34.5 kPa for 2 min. Second, cells were loaded onto the plate at 13.8 kPa for 15 s. Priming run was performed for 5 min with pressure set to 34.5 kPa. The medium ow was set at 12 kPa throughout the experiment for 12h with sterile appropriate media. Images were recorded with an inverted Nikon Ti-E microscope using a Plan Apo 60XA oil Ph3 DM objective using Hamamatsu Orca FLASH 4 camera. Images were processed with NIS Elements software (Nikon). In all experiments, multiple x/y positions were imaged. Representative images were processed using the Fiji Software package.

Electron microscopy
For transmission electron microscopy, half a loopful of 6-8 h old bacterial cultures were xed by direct resuspension in 500 μl of 2.5% glutaraldehyde in 0.1 M cacodylate buffer and incubated for at least 1h at 4°C. Cells were then pelleted through centrifugation at 5000 rpm for 3 min and washed 3 times in 500 μl 0.2 M cacodylate wash buffer solution (PH 7.2). 30-50 μl of wash solution containing bacterial cells was pipetted onto Formvar Carbon 200 mesh copper grids (Sigma-Aldrich) and negative staining done using 1% phosphotungstic acid (PTA) for 2 sec before imaging at the INRS-CAFSB platform using a Hitachi H-7100 electron microscope.
For scanning electron microscopy, fresh bacterial cells were cultured for 6h in liquid media containing poly-L-Lysine (Sigma) coated glass slides. Cells were xed using 2.5% glutaraldehyde in 0.1 M cacodylate buffer for 1h at 4°C then rinsed 3 times in 0.2 M cacodylate wash buffer solution (PH 7.2). Post xation was subsequently done using 1% osmium tetroxide (in 0.2 M cacodylate) before gradual dehydration through increasing ethanol concentrations ( 25%, 50%,75%, 95% and 100%). Carbon dioxide critical point drying (CPD) and gold sputtering were done on Leica EM CPD300 and Leica EM ACE600 instruments respectively. The imaging was done at the electron Imaging Facility (Faculty of dental medicine, Université de Montréal, Québec, Canada) using a Hitachi Regulas 8220 electron microscope.

Peptidoglycan extraction and analysis
Peptidoglycan extraction was performed as previously described (Veyrier et al., 2015). Bacterial cultures were harvested from solid agar plates using inoculation loops and emulsi ed in 10 ml of distilled water, the suspension mix was added drop by drop into 10 ml of 8% boiling sodium dodecyl sulfate (SDS) and boiled for and extra hour. After overnight storage at room temperature, the cells were washed six times using distilled water (pH 6.0) through ultracentrifugation at 39000 x g for 30 min. The nal pellet was lyophilized and resuspended in distilled water (concentration 6mg/ml or more) and stored at -20°C until further use. Analysis of the muropeptide composition was performed essentially as described previously (Alvarez et al., 2020). Samples were treated with Proteinase K (20 μg/mL, 1 h, 37°C). The reaction was heat-inactivated and sacculi were further washed by ultracentrifugation. Finally, samples were digested overnight with muramidase (100 μg/mL) at 37°C. Muramidase digestion was stopped by boiling and coagulated proteins were removed by centrifugation (15 min, 14,000 rpm). For sample reduction, the pH of the supernatants was adjusted to pH 8.5-9.0 with sodium borate buffer and sodium borohydride was added to a nal concentration of 10 mg/mL. After incubating for 30 min at room temperature, pH was adjusted to 3.5 with orthophosphoric acid. The soluble muropeptides were analysed by high-performance liquid chromatography (HPLC; Waters Corporation, USA) on a Kinetex C18 column (150 x 4.6 mm; 2.6 µm particle size, 100 Å) (Phenomenex, USA) and detected at 204 nm with UV detector (2489 UV/Visible, Waters Corporation, USA). Muropeptides were separated with organic buffers at 45°C using a linear gradient from buffer A (formic acid 0.1% (v/v) in water) to buffer B (formic acid 0.1% (v/v) in 40% acetonitrile) in a 18 minutes run with a 1 ml/min ow. Quanti cation of muropeptides was based on their relative abundances (relative area of the corresponding peak) normalized to their molar ratio. The molar percentage was calculated for each muropeptide. This relative molarity was also used to calculate the molar percentage of crosslinked muropeptides. Muropeptide identity was con rmed by MS analysis, using a Xevo G2-XS QTof system (Waters Corporation, USA).

FDAA incubations
For the sequential labelling of cells with HADA (7-hydroxycoumarin-3-carboxylic acid-D-alanine, blue), BADA (BODIPY FL-D-alanine; green) and TADA (TAMRA-D-alanine; red), exponential phase cells were pelleted, resuspended in medium containing the rst label and then grown at 37°C. Speci c media, incubation intervals and order of the labels for each Neisseriaceae species are listed in Table S2 and S3, respectively. After the rst interval cells were washed twice with fresh medium (37°C) and centrifuged between washes (7,000 g for 2 min at RT). After this, the cell pellets were resuspended in pre-warmed medium containing label two. For triple labelling, cells were washed twice and resuspended in medium containing the third label. Cells were then immediately treated with 70% ice-cold ethanol and incubated on ice for 1 hour. Ethanol-xed cells were collected via centrifugation (7,000 g for 2 minutes at RT), washed twice with 4°C 1 x Phosphate Buffered Saline (PBS, pH 7.4), resuspended in PBS, and stored on ice before imaging.

EDA-DA incubation and click-chemistry
To track symbiont cell wall growth followed by immunolabeling Alysiella liformis cells were grown over night on PY plates. Single colonies were incubated in 10 mM ethynyl-D-alanyl-D-alanine (EDA-DA, a Damino acid carrying a clickable ethynyl group) for 30 min, resuspended in pre-warmed PY medium, washed twice (7000 g for 2 minutes at RT) and treated with 70% ethanol like described before. After that, cells were rehydrated and washed in PBS containing 0.1% Tween 20 (PBT). Blocking was carried out for 30 min in PBS containing 0.1% Tween 20 (PBT) and 2% (wt/vol) bovine serum albumin (blocking solution) at room temperature. An Alexa488 uorophore was covalently bound to EDA-DA via copper catalysed click-chemistry by following the user manual protocol for the Click-iT reaction cocktail (Click-iT EdU Imaging Kit, Invitrogen). The cells were incubated with the Click-iT reaction cocktail for 30 min at RT in the dark. Unbound dye was removed by a 10-min wash in PBT and one wash in PBS. For immunostaining of clicked bacterial cells, cells were washed for 10 min in PBT and subsequently incubated with blocking solution for 30 min at room temperature in the dark. From here on, immunostaining was performed as described below.

Western Blots
Proteins from bacteria cells were separated by reduced sodium dodecyl sulfate (SDS)-polyacrylamide gel electrophoresis (PAGE) on NuPAGE 4%-12% Bis-Tris pre-cast MOPS gel (Invitrogen), respectively, and each blotted onto Hybond ECL nitrocellulose membranes (Amersham Biosciences). Membranes were blocked for 45 min in PBS containing 5% (wt/vol) nonfat milk (PBSM) at room temperature and incubated overnight at 4°C with a 1:1,000 dilution of sheep polyclonal anti-E. coli K88 mbrial protein AB/FaeG antibody (ab35292, Abcam) in PBSM. For the negative control, the primary antibody was omitted. After ve 6 min-long washes in PBSM and one nal wash in PBS containing 0.1% Tween20, the blot was incubated for 1 h at room temperature with a horseradish peroxidase-conjugated anti-sheep secondary antibody (1:10,000; Amersham Biosciences) in PBSM. Protein-antibody complexes were visualized using ECL Plus detection reagents (Amersham Biosciences).

Immunostaining
Exponential phase cells were xed overnight in 3% formaldehyde at 4°C. Cells were collected via centrifugation (7000 g for 2 minutes at RT), washed twice with PBS and resuspended in PBS containing 0.1 % Tween 20 (PBT). Blocking was carried out for 1 h in PBT containing 2% (wt/vol) bovine serum albumin (blocking solution) at room temperature. After that, cells were incubated with a 1:500 dilution of sheep polyclonal anti-E. coli K88 mbrial protein AB/FaeG antibody (ab35292, Abcam) overnight at 4°C in blocking solution. Upon incubation with primary antibody (or without in the case of the negative control) samples were washed three times in PBT and incubated with an Alexa555 conjugated anti-sheep antibody (Thermo Fisher Scienti c) at 1:500 dilution in blocking solution for 1 h at room temperature. Unbound secondary antibody was removed by two washing steps one in PBT and one in PBS. Cell pellets were resuspended in PBS containing 5 µg/mL Hoechst for 20 min and subsequently washed and resuspended with PBS. 1 µL of the bacterial solution was mixed with 0.5 µL of Vectashield mounting medium (Vector Labs) and mounted on an agarose slide.
Nile red membrane staining Exponential phase cells were xed overnight in 2 % formaldehyde at 4°C. Cells were collected via centrifugation (7000 g for 2 minutes at RT), washed twice with PBS and resuspended in PBS containing 10 µg/mL Nile Red (Stock is prepared with DMSO; ThermoFisher N1142) and 5 µg/mL Hoechst for 15 min in the dark at room temperature. Cells were washed and resuspended in PBS and subsequently 1 µL of the bacterial solution was mixed with 0.5 µL of Vectashield mounting medium (Vector Labs) and mounted on an agarose slide.

Fluorescence microscopy
For Figure 2 and Supplementary Figures S1 and S4 immunostained or FDAA-labelled bacteria were imaged using a Nikon Eclipse NI-U microscope equipped with a MFCool camera (Jenoptik) and images were acquired using the ProgRes Capture Pro 2.8.8 software (Jenoptik). For Figure 3 and Supplementary Figure S5, FDAA-labelled bacteria were visualized with a Leica TCS SP8 X confocal laser scanning microscope. Images were taken with a 63X Plan-Apochromat glycerine objective with a NA of 1.30 and a refraction index of 1.46 (glass slide, glycerine and antifade mounting medium). The Leica software LASX (3.7.2.22383) was used for image acquisition and post-processing if necessary.

FDAA uorescence quanti cation and statistical analysis
Microscopic images were processed using the public domain program ImageJ (Schneider et al., 2012) in combination with plugin Fil-Tracer (this study). Cell outlines were traced and morphometric measurements recorded. Fluorescent intensities were measured along the septal plane and plotted as fraction of the normalized cell length. Automatic cell recognition was double-checked manually. For representative images, the background subtraction function of ImageJ was used and brightness and contrast were adjusted for better visibility. Data analysis was performed using Excel 2021 (Microsoft Corporation, USA), plots were created with ggplot2 in R (http://www.R-project.org/). Septa length (Figure 6 and S8) of BADA and TADA labelled cells were analysed using the public domain software Fiji (Schindelin et al., 2012). Cell and septa lengths were measured manually. Notably, only cells that showed a BADA and TADA signal were considered for the septa length measurements. Two-tailed unpaired T tests were performed using GraphPad Prism version 9.3.0 for Mac (La Jolla California USA, www.graphpad.com). Figures were compiled using Adobe Photoshop and Illustrator 2021 (Adobe Systems, USA).

Genome sequencing and assembly
Genomic DNA for WGS of Neisseriaceae species and PCR ampli cation of DNA used for cloning purposes or sequence veri cations were extracted using Genomic Tip 20/G or 100/G kits (Qiagen) according to the manufacturer's instructions. The genome sequencing results are presented in Table S1.
Core-genome based phylogeny of Neisseriaceae All genomes were annotated with Prokka v1.14.5 (Seemann, 2014). Core-genome genes were obtained by MAFFT alignment through Roary v3.11.2 (Page et al., 2015), considering a minimum percentage identity of 55 for blastp, and occurrence in at least 90% of the isolates. Best evolutionary model for each partition was found by IQ-TREE version 1.6.3 (Kalyaanamoorthy et al., 2017) and maximum-likelihood phylogenetic analysis was also performed using IQ-TREE (Nguyen et al., 2015) using 10,000 ultrafast bootstrap replicates .

Genomic comparisons
For gene insertion and deletion, we have used the MycoHIT pipeline that was described before (Veyrier et al., 2009(Veyrier et al., , 2011. We used complete genomes of all the rod-shaped and MuLDi Neisseriaceae species represented in Figure 1. We excluded the second coccus lineage (Neisseria wadsworthii; Neisseria canis and N. sp. 83E034). We performed an alignment search with the standalone TBLASTN program (Gerts et al., 2006), using the 2105 predicted proteins from N. elongata ATCC29315 or the 2349 predicted proteins from S. muelleri as the query sequences to search for matches in the genomic DNA of other organisms. We obtained two matrices of around 80000 scores (2063 or 2105 protein sequences blasted against 38 genomes) providing two types of output: categorical (hit versus no hit) and quantitative (degree of similarity). To categorically assign that there was no hit, we employed the default E-value of e-10. Thus, if the statistical signi cance ascribed to a comparison is greater than this E value, we assigned a percentage of similarity of 0% to that comparison. To analyse quantitative results, we used MycoHIT (Veyrier et al., 2009) to assign absence of gene in all MuLDi with presence of the gene in all rodshape or vice versa. "Absence" was de ned as lower values and "presence" as higher values, than the 95 percentile of tested species.
Study of a possible association between amino acid changes and their in uence on cell shape was approached using CapriB (Guerra Maldonado JF, 2020). Brie y, two databases were generated considering Simonsiella muelleri ATCC 29453 (multicellular, accession number GCA_002951835.1) and Neisseria elongata subs. glycolytica ATCC 29315 (bacilli, accession number GCA_000818035.1) as references. The proteins encoded in each genome under study here were further compared against these two references by TBLASTN. Once the blast results were obtained, and the groups to be compared were de ned, i.e. multicellular versus bacilli, amino acid changes in proteins shared by both groups (identity threshold 60%) were investigated. In this regard, CapriB has several analysis options, but the study focused on those amino acids that remain identical in the members of one group but are modi ed in the other (I vs D option).

RNA sequencing and analysis
Total RNA was extracted from 6h cultures grown on GCB agar plates. The cells were harvested in RNA protect reagent (Qiagen). RNeasy Mini Kit (Qiagen) with RNase Free DNAse set (Qiagen) was used for RNA extraction according to the manufacturer's instructions.
The removal of ribosomal RNA for cDNA synthesis was done with NEBNext rRNA Depletion kit with 1 ug of total RNA in the puri cation using 1.8X Cytiva Sera-mag. For results presented in Figure 9, the rRNA depleted mRNA were processed using the Illumina® Stranded mRNA Prep protocol without modi cation by Génome Québec Innovation Centre (McGill University, Montréal, Canada). 100bp Pair-End Sequencing was performed with the NovaSeq 6000 system. Sequence reads were processed with FastQC (Version 0.73) to determine the quality before grooming by FastQ Groomer (Version 1.1.5). Paired FastQ reads were then aligned against Neisseria elongata subsp. glycolytica ATCC 29315 (accession number NZ_CP007726.1) genome using Bowtie2 (Version 2.4.2) and read counts were determined using htseq_count (Version 0.9.1) tool. Subsequently the gene expression of the transcripts was determined using DESeq2 (Version 2.22.40.6). Visualization of differentially expressed genes was done with Venn diagrams, drawn by a Web-based platform Venny 2.1 (https://bioinfogp.cnb.csic.es/tools/venny/).
For intra-genus transcriptomic comparison presented in Figure 8, rRNA depleted mRNA were treated using the RevertAid RT Reverse Transcription Kit (K1691; Thermo Scienti c™) with some adjustments. For rst strand cDNA synthesis, 1 ul of random primer (3 μg/μL ; 48190011; Invitrogen™) was added and the solution was incubated at 65°C. For the second strand cDNA synthesis, procedure was followed without RNA removal step and by purifying the double-stranded cDNA with 1.8X Cytiva Sera-mag. The cDNA were eluted in 24 μL of nuclease-free water. Libraries were prepared by PCR BARCODING (96) AMPLICONS (SQK-LSK109) and PCR BARCODING (SQK-PBK004) (Oxford Nanopore technologies), as described by the manufacturer. The base call was carried out using guppy_basecaller (version 5.0.11+2b6dbff) in sup mode, adapters were removed and ltered by quality Q> 8, they were separated by barcodes using guppy_barcoder (version 5.0.11+2b6dbff). In parallel, the ten indicated genomes were annotated with Prokka v1.14.5 (Seemann, 2014). Using each of the protein sequences (.faa) les, a standalone BLASTP (Camacho et al., 2009) was performed for each dyad possibility. Network connection was thereafter established with the python programming package NetworkX version 2.6.2 (Hagberg et al., 2008) with a cut-off of 60% of similarity. Basically, all proteins showing more than 60% similarities with one of the members (putative homologues) were clustered together. Each cluster of proteins was named (example NEISS_1) and this name was used to replace the original locus-tags in the .GFF le (previously generated by Prokka). This was done using an homemade python script and has generated a new le that we called .GTF. This le was used to map the reads to the corresponding genomes using minimpa2 (Li, 2018). The .GTF and .sam les were used to perform the reads counts using featureCounts of Subread package (Liao et al., 2014). The count les for each sample were joined into a table using a homemade script and these results were analysed using DESeq2 version 3.14 (Love et al., 2014).
Parameter used were Reads >1 in the 10 genomes (core-transcriptome: genes that were showing at least one read mapped in all genomes). We investigated the biological functions of the gene differentially expressed and the putative pathways that could link them through a STRING analysis (Szklarczyk et al., 2021).
Quantitative real-time PCR RNA samples were standardized to a nal concentration of 1 μg with addition of DNaseI Ampli cation grade (Invitrogen) for genomic DNA removal. Random primers (Invitrogen), and RevertAid H-Minus reverse transcriptase (Thermo Scienti c) were used for complementary DNA synthesis (cDNA) according to the manufacturer's instructions. Absence of contaminating gDNA was veri ed by conventional PCR of RNA samples in the absence of reverse transcriptase. Gene expression of dcw cluster was veri ed by quantitative real-time PCR (qRT-PCR) using Power SYBR Green PCR master mix (Applied Biosystems) using primers listed in supplementary Table S2. Differential gene expression was calculated using ΔΔCT method using the mean CT value of each target, normalization was done relative to gyrA gene. Standard T-test using (GraphPad Prism v9.0; GraphPad Software, CA) was used to ascertain statistical signi cance of gene expression between the strains, where P<0.05 was considered signi cant.

AmiC and AmiC2 phylogeny
Sequences of N-acetylmuramoyl-alanine amidase from S. muelleri ATCC 29453 (AmiC, accession number AUX62143.1) and C. kuhniae (AmiC2, accession number WP_027009548.1) were searched by blastp against all the Neisseriaceae genomes included in this study, as well as the complete bacterial repertoire found at NCBI. Amino acid sequences of the hits obtained by blastp were retrieved from the entire set of genomes using faSomeRecords (https://github.com/santiagosnchez/faSomeRecords/blob/master/faSomeRecords.pl). The resulting sequences were aligned with MAFFT v7 (Katoh et al., 2002), and maximum-likelihood phylogenetic analysis was performed using IQ-TREE using 1,000 ultrafast bootstrap replicates.
Genomic organization of the dcw cluster and cdsA loci in the Neisseriaceae Coordinates of the dcw cluster and of the cdsA loci were obtained by tblastn for each Neisseriaceae genome. Once the genomic location of each sequence was determined, the sequences were extracted using tools available in the EMBOSS package (Rice et al., 2000). The resulting sequences were annotated with Prokka, and the output gbk les were used to construct the synteny by employing EasyFig 2.2.2 (Sullivan et al., 2011).

Construction of mutant strains
Neisseria elongata mutant strains were done in N. elongata subsp. glycolytica (ATCC 29315) for single gene mutation and its streptomycin-resistant variant with a point mutations K88R rpsL* for unmarked and multiple gene editing. mraZ was deleted by replacing mraZ with an mCherry-encoding gene. The construct for mraZ deletion was obtained by fusing multiple PCR fragments using Phusion DNA polymerase according to the protocol (New England Biolabs) as follows: rstly, N. elongata gDNA was used to amplify approximately 500 bp of regions up and down stream of mraZ using, respectively, primer pairs 5'KoMraZF-R and 3'KoMraZF-R. The promoter "pdcwSm", located upstream the S. muelleri dcw cluster, was ampli ed from S. muelleri gDNA using primer pairs pdcwsmF/pdcwsmR. Primer pairs 5MraZKmF and KmpSimR were used to amplify the kanamycin resistance cassette from pGEM::Km plasmid DNA (Veyrier et al., 2015), while the Mcherry cassette was obtained by PCR ampli cation of pMcherry10 (Addgene) using primer pairs pdcwsmMcherry F and McherryNsilR. Subsequently, the 5'MraZ and Km cassette were fused using primer pairs 5KoMraZF and KmpSimR, while Mcherry and 3'MraZ were fused using primer pairs pdcwMcherryF and 3'KoMraZ R. Finally, 5'MraZ-KM, pdcwSm and Mcherry-3'MraZ fragments were fused using primer pairs 5KoMraZ F and 3KoMraZ R and the resulting DNA was used for transformation in N. elongata.
To overexpress mraZ, Neisseria meningitidis promoter, porB was ampli ed from N. meningitis gDNA using primer pairs (porBpF-porBpbluntR) while the mraZ gene was ampli ed from N. elongata gDNA using primer pairs (MraZSphIF-3MraZR). The porB promoter from N. meningitis and the mraZ gene from N. elongata were subsequently fused by PCR. This resulted in an approximately 1.6 kb-long porB:MraZ cassette that was digested using the restriction enzymes NheI and KpnI and then ligated with NheI-KpnI digested plasmid p5nrq3::Cm (Veyrier et al., 2015). The ligation mix was transformed in E. coli DH5α cells to obtain the porBMraZ::p5nrq3::Cm plasmid. The plasmid was subsequently linearized before transformation into the Neisseria elongata DmraZ strain.
For the single knockout of ΔmraZ, ΔrapZ, ΔgloB or Δdgt, we used a cassette developed in our laboratory named RPLK (Nyongesa et al., n.d. submitted) that contains the wild-type N. elongata rpsL gene, N. meningitidis promoter porBp, the blue-white screening selection marker lacZ and the kanamycin resistance marker that facilitated the generation of unmarked deletion in addition to multiple gene editing. We used synthesized DNAs (BioBasic) that contain approximately 500 bp each 5' and 3' regions surrounding the respective genes with a central BglII restriction site and cloned into pUC57 plasmid. The plasmids were linearized using BglII and ligated with RPLK cassette (Nyongesa et al., n.d. submitted).
Mutants were obtained by transforming either N. elongata wild-type (single KO) or an N. elongata streptomycin-resistant strain (indicated rpsL*) (multiple KO) with the linearized plasmid of the targeted gene that resulted in blue, kanamycin-resistant, streptomycin sensitive clones. Markerless deletion was achieved by introducing DNA of the 5'-3' homologous regions of the target gene thereby excising the RPLK cassette resulting in white, kanamycin sensitive and streptomycin resistant clones. Subsequent genes of interest were edited by repeating this procedure and veri cations of the correct excision was done by PCR.
For allelic switching of N. elongata mreB with that from S. muelleri, plasmid pMreBSimon-3'RD3Ne was obtained by amplifying S. muelleri mreB using primer pairs MreBsimonF -MreBsimonR, while the subsequent region of the locus (3'RD3Ne that comprise a piece of mreCD) was ampli ed from N. elongata using primer pairs 3'RD3NeF-3'RD3NeR. The two products were fused using primer pairs MreBsimonF -3'RD3NeR. This generated a cassette of mreB sm fused with mreCD ne that was then digested by restriction enzymes BamHI and SpeI before ligation with plasmid p5KORD1Ne::cm (Veyrier et al., 2015) digested with the same enzymes to obtain plasmid pMreBsimon3'RD3Ne::cm. The plasmid was linearized with ScaI before transformation in N. elongata strains. mreB sm positive and mreB ne negative clones were con rmed by PCR.
For the cdsA-amiC2 knock-in constructs, we used the plasmid pUCNe ::ampR that contains 5' and 3' Neisseria elongata homologous regions to the intergenic locus between two genes coding for hypothetical proteins at position 888015 (insertion site). We rst constructed the pUCNe::RPLK plasmid by ligating the RPLK cassette using BglII. Secondly, cdsA-AmiC2 PCR product was obtained using primer pairs cdsAmiC2F-amiC2R, was digested using BglI and ligated to pUCNe::ampR to produce the pUCcdsamiC2::ampR plasmid. The mutants were obtained with a two-step methods (Nyongesa et al., n.d. submitted). First, we transformed the plasmid pUCNe::RPLK into N. elongata rpsl* to obtain N. elongata RPLK (RPLK inserted at position 888015). In the second step, we have replaced the RPLK cassette with cdsA-amiC2 genes, by transforming the pUCcdsamiC2::ampR plasmid linearized using ScaI into N. elongata RPLK. cdsamic2 positive transformants were con rmed by PCR.
Data and code availability. The documentation for the ImageJ plugin Fil-Tracer can be accessed here:   Comparative genomics and transcriptomic of rod-shaped and MuLDi Neisseriaceae.
(a) Phylogenetic tree of Neisseriaceae species (left) and distribution, within the family, of the genes that were inserted (left part of the table) or deleted (middle part of the table). In addition, selected proteins known to be involved in cell growth and division are also presented (right part of the table). Inserted genes are indicated with S. muelleri locus_tag (such as RS00570 for BWP33_RS00570). All the other genes are indicated with N. elongata locus_tag (such as RS02740 for NELON_RS02740). The putative encoded protein associated with each gene are also speci ed (HP stands for hypothetical protein). The green and black squares of the table indicate genes that are present or absent, respectively. Individual genes were considered to be present when they had a sequence similarity ≥60% relative to N. elongata [an e-value cut-off of 1e-10 has also been applied in TBLASTN version 2.7.1 (Altschul et al., 1997)

. (b)
Weblogo of the amino acid sequences, of the 7 proteins displaying amino acid permutations rod-shaped or for MuLDi, detected with amino-acids permutations between rod-shaped and MuLDi Neisseriaceae. (c) Volcano-plot: p value have been plotted with fold change. Points coloured in red are genes with lower expression in MuLDi and green correspond to genes with higher expression. (d) STRING association analysis. ftsA, ftsI and murE from the dcw cluster are highlighted. In red are genes with lower expression in MuLDi and in green are genes with higher expression.

Figure 5
Downregulation of the dcw cluster in N. elongata ΔmraZ. (a) Volcano plot of RNAseq analysis of an N. elongata ΔmraZ and complemented. p value is plotted against fold change. Red points represent genes with higher expression in MraZ-overexpressing N. elongata (ΔmraZ; pilEp-mraZ), as compared to N. elongata ΔmraZ. (b) Venn diagram showing genes (mraZ, mraW, ftsL and ftsI) upregulated in N. elongata wild-type as compared to N. elongata ΔmraZ. (c) Transcript abundance of dcw cluster genes measured cell length of N. elongata expressing or not expressing MraZ. Scale bar is 2 µm. Figure 6