Two novel heteropolymer‐forming proteins maintain the multicellular shape of the cyanobacterium Anabaena sp. PCC 7120

Polymerizing and filament‐forming proteins are instrumental for numerous cellular processes such as cell division and growth. Their function in stabilization and localization of protein complexes and replicons is achieved by a filamentous structure. Known filamentous proteins assemble into homopolymers consisting of single subunits – for example, MreB and FtsZ in bacteria – or heteropolymers that are composed of two subunits, for example, keratin and α/β tubulin in eukaryotes. Here, we describe two novel coiled‐coil‐rich proteins (CCRPs) in the filament‐forming cyanobacterium Anabaena sp. PCC 7120 (hereafter Anabaena) that assemble into a heteropolymer and function in the maintenance of the Anabaena multicellular shape (termed trichome). The two CCRPs – Alr4504 and Alr4505 (named ZicK and ZacK) – are strictly interdependent for the assembly of protein filaments in vivo and polymerize nucleotide independently in vitro, similar to known intermediate filament (IF) proteins. A ΔzicKΔzacK double mutant is characterized by a zigzagged cell arrangement and hence a loss of the typical linear Anabaena trichome shape. ZicK and ZacK interact with themselves, with each other, with the elongasome protein MreB, the septal junction protein SepJ and the divisome associate septal protein SepI. Our results suggest that ZicK and ZacK function in cooperation with SepJ and MreB to stabilize the Anabaena trichome and are likely essential for the manifestation of the multicellular shape in Anabaena. Our study reveals the presence of filament‐forming IF‐like proteins whose function is achieved through the formation of heteropolymers in cyanobacteria.

Polymerizing and filament-forming proteins are instrumental for numerous cellular processes such as cell division and growth. Their function in stabilization and localization of protein complexes and replicons is achieved by a filamentous structure. Known filamentous proteins assemble into homopolymers consisting of single subunitsfor example, MreB and FtsZ in bacteriaor heteropolymers that are composed of two subunits, for example, keratin and α/β tubulin in eukaryotes. Here, we describe two novel coiled-coil-rich proteins (CCRPs) in the filament-forming cyanobacterium Anabaena sp. PCC 7120 (hereafter Anabaena) that assemble into a heteropolymer and function in the maintenance of the Anabaena multicellular shape (termed trichome). The two CCRPs -Alr4504 and Alr4505 (named ZicK and ZacK)are strictly interdependent for the assembly of protein filaments in vivo and polymerize nucleotide independently in vitro, similar to known intermediate filament (IF) proteins. A ΔzicKΔzacK double mutant is characterized by a zigzagged cell arrangement and hence a loss of the typical linear Anabaena trichome shape. ZicK and ZacK interact with themselves, with each other, with the elongasome protein MreB, the septal junction protein SepJ and the divisome associate septal protein SepI. Our results suggest that ZicK and ZacK function in cooperation with SepJ and MreB to stabilize the Anabaena trichome and are likely essential for the manifestation of the multicellular shape in Anabaena. Our study reveals the presence of filament-forming IF-like proteins whose function is achieved through the formation of heteropolymers in cyanobacteria.

Introduction
Cytoskeletal proteins that polymerize to form protein filaments are paramount in bacterial cell biology where they play a role in cell division, alignment of bacterial microcompartments (BMCs), chromosome and plasmid segregation, organization of cell polarity and the determination of cell shape [1]. For example, FtsZ [2], the prokaryotic homolog to the eukaryotic tubulin [3], is a main component of the divisome [4], a multiprotein complex that governs cell division in bacteria and self-assembles into a proteinaceous ring (called Z-ring) at the midcell position [5]. Another key bacterial cytoskeletal protein is MreB [6], which is a homolog of the eukaryotic actin [2,7] and a crucial component of the multiprotein complexes termed the elongasome. This complex modulates cell elongation in many rodshaped bacteria through regulating peptidoglycan (PG) biogenesis [8]. Both, FtsZ and MreB monomers assemble into filamentous strands (protofilaments), consisting of only one type of monomer, termed homopolymers [1]. The cell division in prokaryotes markedly contrasts the division of plastid organelles in photosynthetic eukaryotes that are of cyanobacterial origin [9]. Cell division in plastids is dependent on the cooperative function and heteropolymerization of two FtsZ homologs, FtsZ1 and FtsZ2 in the green lineages and FtsZA and FtsZB in the red lineage. However, each FtsZ homolog is also self-sufficient to form homopolymers (reviewed by [10]). In contrast, a likely horizontally transferred pair of tubulin homologs, BtubA and BtubB from Prothescobacter spp., exclusively assembles into heteropolymers in vitro [11], revealing similar properties than eukaryotic microtubules that are heteropolymers composed of α and β tubulin monomers [12]. Eukaryotic IF proteins, despite sharing substantially the same building blocks and a high degree of coiled-coil (CC) domains [13], which are considered excellent mediators of protein-protein interactions [14], only form heteropolymers with a subset of other IF proteins within their same assembly group but otherwise form obligate homopolymers [15].
Polymer-forming coiled-coil-rich proteins (CCRPs) have been shown to play a role also in multicellularity traits in myxobacteria and actinomycetes (reviewed by [1,16]). Similar to eukaryotic IFs [13], many bacterial CCRPs perform cytoskeletal functions through their ability to self-assemble into filaments in vivo and in vitro in a self-sufficient and cofactor independent manner [17][18][19][20]. The CCRP Crescentin determines the Caulobacter crescentus typical curved morphology by aligning to the inner cell curvature and exuding local mechanical constraints on the PG biosynthesis, likely through cooperation with MreB [17,21]. In analogy to eukaryotic IF proteins, Crescentin assembles into straight protein filaments with a width of 10 nm and displays a similar domain organization [17]. However, while Crescentin is often considered a prokaryotic homologue to eukaryotic IF proteins, its restricted distribution to only one identified organism questions real homologous relationships and rather suggests that it was acquired by horizontal gene transfer [22]. Multicellular actinobacteria, such as Streptomyces spp., grow by building new cell wall (i.e., PG) only at the cell poles, independent of MreB [23], a striking different cell growth than in most other bacteria [24]. This characteristic polar growth mode is organized by a cytoskeletal network of at least three CCRPs -DivIVA, Scy and FilPthat directly interact with each other to form the polarisome [18]. FilP and Scy independently self-assemble into filaments in vitro [18,25], thereby fulfilling major IF-like criteria [1]. In vivo, however, Scy does not form filaments and instead accumulates as foci at future branching points [18], while FilP localizes as gradient-like caudates at the hyphal tips [26], instead of forming distinct filaments as observed for Crescentin [17].
Among prokaryotes, cyanobacteria exhibit the largest morphological diversity, comprising unicellular species as well as complex cell-differentiating multicellular species [27]. For the model multicellular cyanobacterium Anabaena, it is imperative to form stable trichomes in order to cope with external influences such as shearing stress [28,29]. Under nitrogendeprived growth conditions, Anabaena develops specialized cell types for nitrogen fixation (heterocysts), which are evenly spaced among the Anabaena trichome and provide other vegetative cells with fixed nitrogen compounds like glutamine [30]. In Anabaena, proteinaceous cell-joining structures that allow intercellular transport (i.e., cell-cell communication) and function by gating are termed septal junctions (analogous to eukaryotic gap junctions; [31,32]). Septal junctions consist of several structural elements, an intracellular cap, a plug inside the cytoplasmic membrane formed by the septal junction protein FraD and a tube traversing the septum through nanopores in the peptidoglycan [31]. The correct positioning of the septal protein SepJ, which is involved in septum maturation and filament stability, among others, depends on the FtsZ-driven divisome component FtsQ [33], which links the early and late assembly components of the divisome [34]. FtsZ was shown to be essential for Anabaena viability and to assemble in a typical Z-ring structure at future septum sites in vegetative cells while being downregulated in heterocysts [35][36][37]. In contrast, MreB is dispensable for Anabaena viability but determines its typical cylindrical cell shape, since ΔmreB mutant cells display a pronounced rounded and swollen morphotype. This cell swelling, however, does not affect chromosome segregation, which was found to be governed, at least in part, by random segregation in Anabaena [38]. Maintenance of the cylindrical cell shape is furthermore dependent on a class B penicillin-binding protein (PBP) [39] and AmiC-type cell wall amidases in Anabaena [40,41], suggesting that loss of normal cell shape is commonly associated with defects in PG biogenesis.
In this work, we aimed to identify proteins that play a role in cyanobacterial morphology and multicellularity. Searching for IF-like CCRPs, we identified two novel CCRPs in Anabaena that are capable of assembly exclusively into a heteropolymer in vitro and in vivo and that have a putative role in the Anabaena linear trichome shape.

Results
CCRPs ZicK and ZacK from Anabaena are conserved in cyanobacteria A computational survey of the Anabaena genome for protein-coding genes containing a high coiled-coil content [42] revealed two CCRPs Alr4504 and Alr4505; here, we term the two CCRPs ZicK and ZacK, respectively (i.e., zig and zag in German). ZicK is predicted to contain five distinct coiled-coil (CC) domains, while ZacK has four CC domains ( Fig. 1A; File S1). Using PSORTb (v.3.0.2), both ZicK and ZacK are predicted to be cytoplasmic proteins, which is corroborated by the absence of detectable transmembrane domains (predicted using TMHMM v. 2.0). Since both proteins (and their homologs) are annotated as hypothetical proteins (File S2), we validated their transcription under standard (BG11) and diazotrophic (BG11 0 ) growth conditions (Fig. S1B,C). The genomic neighbourhood of zicK and zacK motivated us to test for a common transcriptional regulation of both genes (i.e., an operon structure); however, we did not identify a common transcript (using RT-PCR; Fig. S1A,B). Searching for known proteins sharing structural similarities to ZicK/ZacK using I-TASSER revealed structural similarities between ZicK and the eukaryotic cytolinker protein plectin, and of ZacK with the cell division protein EzrA, a predicted structural similarity that was previously associated with other bacterial CCRPs, including Crescentin, HmpF Syn and HmpF Syc [42]. Further annotation using the NCBI conserved domain search [43] showed that ZicK and ZacK contain 'structural maintenance of chromosomes' (SMC) domains (Fig. 1B), similarly to what we previously identified in other self-polymerizing cyanobacterial CCRPs [42]. A search for ZicK/ZacK homologs by sequence similarity revealed that they are absent in picocyanobacteria (i.e., Synechococcus/Prochlorococcus) and generally rare in unicellular cyanobacteria. Otherwise, about 50% of the examined 168 cyanobacterial genomes have homologs for the two genes ( Fig. 1C and File S2). Several heterocystous cyanobacteria lacking ZicK/ZacK homologs are characterized by a reduced genome (e.g., Nostoc azollae PCC 0708 and Richelia intracellularis). Additionally, Chlorogloeopsis spp. that forms multiseriate filaments is lacking the homologs and several strains of Fischerella spp., which form true branching filaments, harbour only a ZacK homolog. The protein sequence of ZicK and ZacK homologs is well conserved, with about 55% amino acids identity among the homologs. The number of CC domains, however, differ among ZicK and ZacK homologs: between 3-6 CC domains in ZicK homologs, and 3-10 domains in ZacK homologs (File S1). Notably, ZicK and ZacK are neighbours in 53 out 72 genomes; both proteins and their genomic neighbourhood are highly conserved among heterocystous cyanobacteria (Fig. 1C,D).

ZicK and ZacK are interdependent for polymerization in vitro
As a prerequisite for proteins to be considered as IFlike proteins, it is imperative for them to be able to self-interact in vivo and to polymerize into long protein filaments in vitro [1]. To investigate the in vitro polymerization properties of ZicK and ZacK, we employed an in vitro polymerization assay that we previously established to test CCRPs' polymerization properties [42]. As a positive control for our approach, we used Crescentin (Fig. 1A), which formed an extensive filamentous network in our in vitro assay (Fig. S2). As negative controls, we included empty vector-carrying Escherichia coli cells, GroEL1.2 from Chlorogloeopsis fritschii PCC 6912 (known to form oligomers [44]) and the highly soluble maltose-binding protein (MBP), all of which were tested negative for filament formation in vitro using our approach (Fig. S2). Purified ZicK protein formed into amorphous, nonfilamentous protein aggregates while for ZacK low levels of aggregated sheet-like structures could be observed ( Fig. 2A). However, the vast majority of ZacK protein precipitated into clumps of aggregates upon renaturation, which resembled the structures observed for GroEL1.2, suggesting that ZacK has only a partial capacity to self-polymerize or, more likely, is highly unstable on its own in vitro. Inspired by the close genomic localization of zicK and zacK, we next tested for a potential heteropolymerization of both proteins. This revealed that ZicK and ZacK co-assembled into a meshwork of protein heteropolymers upon corenaturation ( Fig. 2A, Fig. S3). While both, ZicK and ZacK renatured alone, formed aggregates in the dialysis tubes that were detectable with the naked eye (similar to our observations from GroEL1.2), corenatured ZicK/ZacK remained in solution, a common property of eukaryotic IFs [45]. Next, we tested whether the cofilamentation was dosage dependent and observed that distinct protein filaments could be detected in vitro only with equal amounts of ZicK and ZacK (Fig.  S3). To further test for an in vivo self-interaction, we analysed the self-binding capacity ZicK and ZacK using the bacterial adenylate cyclase two-hybrid (BACTH) assay and found that ZicK and ZacK interact with themselves and also with each other (Fig. 2B) confirming the heterologous binding capacity of both proteins. Consequently, ZicK and ZacK fulfil a major characteristic of IF and IF-like proteins as they are able to self-assemble into filament-like structures in vitro, although, unlike other bacterial IF-like CCRPs, this assembly exclusively occurs as a heteropolymer.

ZicK and ZacK are interdependent for polymerization in vivo
To examine the in vivo localization pattern of ZicK and ZacK, we initially expressed translational GFP  fusions of both proteins from the replicative pRL25C plasmid, which is commonly used in experimental work in Anabaena [38,46]. The expression of ZicK-GFP and ZacK-GFP from their respective native promoters (as predicted using BPROM [47]) revealed no discernible expression of ZacK-GFP, while ZicK-GFP accumulated within the cells as dot-like aggregates (Fig. S4A). We assume that the lack of ZacK-GFP expression from its predicted native promoter is based on the uncertainty of the precise promoter site prediction. Alternatively, and although not expressed as an operon, the expression of zacK could be affected by the expression of zicK by so far unknown mechanisms. Consequently, we proceeded to investigate the in vivo localization of both proteins from the copper-regulated petE promoter (P petE ), which is commonly used to study the subcellular protein localization in Anabaena (e.g., FtsZ, MreB and SepI; [36,38,48,49]). The expression of ZicK-GFP and ZacK-GFP from P petE in Anabaena independently did not reveal distinct or filamentous structures but resulted in the formation of inclusion body-like aggregates within the cells (Fig. 3 A), similar to those observed for P zicK ::zicK-gfp (Fig.  S4A). We could not detect any structures when we expressed ZicK and ZacK N-terminally fused to GFP, suggesting that the N terminus is key for proper protein folding. Consequently, we next proceeded to coexpress the two proteins with different fluorophores from P petE : ZicK C-terminally fused to eCFP (ZicK-eCFP) and ZacK C-terminally fused to GFP (ZacK-GFP). Heterologous co-expression revealed the formation of a ZicK/ZacK heteropolymer filament-like structure that usually localized along the longitudinal cell axis and in rare occasions also perpendicular to the cell axis (Fig. 3B). The prominent formation of the ZicK/ZacK heteropolymer was furthermore evident as To further explore whether the ZicK/ZacK heteropolymer assembly is restricted to Anabaena, we proceeded to analyse ZicK and ZacK in vivo in an unrelated heterologous system using the E. coli split GFP assay [50]. Clearly, discernible filamentous-like structures (reminiscent of FilP-GFP [20]) could be detected upon coexpression of ZicK and ZacK C-terminally fused to the split GFP products (ZicK-NGFP and ZacK-CGFP; Fig. S5). This is in agreement with the lack of discernible structures upon expression of N-terminally GFP-fused ZicK and ZacK in Anabaena. Nonetheless, some indications for heteropolymerization were also present upon co-expression of NGFP-ZicK with ZacK-CGFP, which is in agreement with the BACTH results that indicated that both, N and C-terminal fusions of ZicK and ZacK, are potentially able to interact with each other (Fig. 2B). The different Deletion of zicK and zacK leads to defects in trichome and cell shape and Anabaena viability In contrast to the obtained ΔzicKΔzacK double mutant, single ΔzicK or ΔzacK mutant strains could not be generated, suggesting that the presence (or absence) of ZicK or ZacK alone is lethal for Anabaena. Further investigating the ΔzicKΔzacK mutant phenotype revealed an altered trichome and cell shape phenotype and a reduced trichome viability (Fig. 4). Unlike the linear trichome growth pattern of the WT, the ΔzicKΔzacK mutant strain mostly grew as zigzagged trichomes (Fig. 4A,B), a phenotype that could be rescued by heterologous expression of P zicK :: zicK-zacK from pRL25C but not from P zicK ::zicK-ecfp+zacK-gfp (Figs S4C (Fig. 4C,D), reminiscent of the ΔmreB mutant strain [38]. The round and swollen cell phenotypes of the ΔzicKΔzacK mutant strains are indicative of an impairment in cell wall integrity and/ or defects in PG biogenesis as well as an elevated sensitivity to turgor pressure [51,52]. Consequently, we tested for an elevated sensitivity of the ΔzicKΔzacK mutant to cell wall damaging enzymes. This showed that the ΔzicKΔzacK mutant is slightly more sensitive to Proteinase K treatment but was unaffected by lysozyme treatment and still retained the ability to grow diazotrophically (i.e., on BG11 0 plates) (Fig. 4E) and to form heterocysts (Fig. S6B). More importantly, however, we found that the ΔzicKΔzacK mutant lost the ability to grow in liquid culture (with and without agitation; Fig. 4F), hinting for an elevated sensitivity to fluid shear stress or turgor pressure. This defect could be complemented by heterologous expression of P zicK ::zicK-zacK from pRL25C (Fig. S6C). We note, however, that we could not fully complement the mutant phenotype using our in transcomplementation system as indicated by the decreased growth rate and seemingly smaller cell volumes (Fig. S6A,C), likely due to the variation in copy number of pRL25C [53].

ZicK and ZacK interact with proteins involved in cell shape and trichome integrity
Considering the impact of the deletion of zicK and zacK on cell and trichome shape and the assumed septal docking of the ZicK/ZacK heteropolymer, we next wanted to investigate whether both, ZicK and ZacK, physically interact with other proteins known to function in cell shape determination and cell-cell communication. Using the BACTH assay, we found that both, ZicK and ZacK, interacted with the divisome-associated septal protein SepI [49], the septal protein SepJ [54], the cell shape-determining protein MreB as well an elongasome associated protein (ZipM; covered in more detail in subsequent study) and the Anabaena homolog to HmpF (here named HmpF Ana ), whose homologs were shown to be involved in motility in Nostoc punctiforme ATCC 29133 [55] and Synechocystis sp. PCC 6803 [42] (Fig. 5A). No interactions were found with FtsZ, FraC and FraD (Fig. S7). We attempted to further confirm our interaction results with affinity co-elution experiments but found that Ni-NTA-bound ZicK and ZacK purified from E. coli readily precipitated upon transfer from denaturing to native buffer conditions, precluding further co-elution studies. Additionally, we observed that nondenaturing conditions failed to purify overexpressed CCRPs from E. coli, confirming their inherent insoluble nature, a property known to eukaryotic IFs [56]. Instead, we surveyed for further interaction partners by anti-GFP co-immunoprecipitation experiments of Anabaena cells expressing ZicK-GFP and analysed coprecipitated proteins by LC-MS/MS analytics (all 27 identified possible interactors are listed in File S3). This analysis confirmed that ZicK and ZacK interact with each other in vivo and further strengthened the observed association of ZicK with MreB (Fig. 5B). Furthermore, ZicK coprecipitated ParA (Fig. 5B), a Walker A-type ATPase, involved in chromosome and plasmid partitioning [57].
Deletion of zicK and zacK affects the localization of MreB and the chromosomes As our BACTH analysis identified SepI and SepJ as interaction partners, and both proteins are involved in intercellular transport and cell-cell communication in Anabaena [49,58], we proceeded to test whether the ΔzicKΔzacK mutant is also affected in solute diffusion using fluorescence recovery after photobleaching (FRAP) experiments of calcein-stained ΔzicKΔzacK mutant. However, this analysis did not reveal any defect in cell-cell communication in the ΔzicKΔzacK mutant ( Fig. S8A-C), and hence, ZicK and ZacK do not affect septal junction functionality. Additionally, electron microscopy of ultrathin sections of the ΔzicKΔzacK mutant did not show any discernible differences in the ultrastructure of the cells compared to cells of Anabaena WT (Fig. S8D). In accordance with a lack of interaction between FtsZ and ZicK/ZacK, FtsZ placement was unaffected in the ΔzicKΔzacK mutant as shown using anti-FtsZ immunofluorescence (Fig. S9A). Following the lead of ZicK/ZacK interaction partners, we next analysed the localization of MreB in the Anabaena WT and the ΔzicKΔzacK mutant using a functional P petE ::gfp-mreB fusion [38]. In Anabaena WT, we predominantly observed GFP-MreB filaments throughout the cells without any directional preferences (96% or 202 of 211 counted cells) and rarely cell pole-localized foci (4% or 9 of 211 counted cells) (Fig. 6A). Even though GFP-MreB filaments were present in 40% (81 of 199 counted cells) of the ΔzicKΔzacK mutant cells expressing GFP-MreB (Fig. 6A inlay), we only detected those filaments in nonrounded cells that seemingly had a WT-like phenotype (Fig. 6A). In contrast, in rounded/swollen cells of zigzagged trichomes, GFP-MreB was mostly (60% or 118 of 199 counted cells) restricted to the cell poles (Fig. 6A). WT-like nonrounded cells accounted for about 24% of the ΔzicKΔzacK mutant cells expressing GFP-MreB (245 of 1040 cells counted), whereas rounded/swollen cells of zigzagged trichomes accounted for 76% of counted ΔzicKΔzacK mutant cells (795 of 1040 counted cells; note that not all cells counted for cell shape characteristics also contained GFP-MreB signals, thus explaining the discrepancies between the different counted cell numbers. This is likely due to the copy number variation of the pRL25C plasmid). To further investigate the potential  [59]). The staining pattern between the WT and the ΔzicKΔzacK mutant was indistinguishable but the fluorescence intensity levels were slightly decreased in the ΔzicKΔzacK mutant (Fig. S9B,C). Nonetheless, this is likely accounted for by the reduced growth rate of the ΔzicKΔzacK mutant ( Fig. 4E and general observation on growth plates).
Considering the interaction of ZicK/ZacK with ParA, we further tested for a function of ZicK and ZacK in DNA placement and compared the DNA distribution in the WT and the ΔzicKΔzacK mutant as measured by distribution of 4 0 ,6-diamidino-2-phenylindol (DAPI) staining intensity (Fig. 6B,C). For that, we calculated the width of the DAPI focal area as the range of DAPI staining around the maximum intensity focus (AE10 grey intensity in arbitrary units). This revealed that the staining focal area size was significantly different among the WT and the ΔzicKΔzacK mutant. The DAPI signal observed in the ΔzicKΔzacK mutant appears more condensed, and indeed, the ΔzicKΔzacK mutant focal DAPI area was smaller than the WT (Fig. 6C). Unlike the WT, DAPI signals in the ΔzicKΔzacK mutant were also observed between two neighbouring cells (Fig. 6B). Overall, our results may suggest the involvement of ZicK/ZacK in DNA distribution and segregation in dividing cells.

Discussion
Here, we provide evidence for the general capacity of two Anabaena CCRPs, which we termed ZicK and ZacK, to form polymers in vitro and in vivo. While the previously described prokaryotic filament-forming CCRPs formed homopolymers [17,19,20,60], ZicK and ZacK exclusively assembled into a heteropolymer in vitro and in vivo, thus revealing a new property of bacterial CCRPs. The inherent heteropolymerization tendency of ZicK and ZacK was confirmed in a heterologous and evolutionary distant E. coli system, which was previously used to investigate other known CCRPs such as Scc from Leptospira biflexa [61] or Crescentin [62]. Although heteropolymerization has previously been described for prokaryotic cytoskeletal proteins, none of those polymerization pairs both belonged to the group of CCRPs. BacA and BacB, members of the widely conserved class of bactofilins, both independently polymerize into filaments in vitro, colocalize in vivo in C. crescentus and interact directly with each other as indicated by co- Only translational fusion combinations that resulted in a significant interaction between two analysed proteins are shown. All other combinations were negative for interaction. Note that we used full-length proteins for all our BACTH analysis, including SepJ, whose precise subcellular localization remains to be identified [49,97] but could be different in E. coli and Anabaena [49]. BacA and BacB has not been studied in vitro. Another interesting pair of potential copolymerizing cytoskeletal proteins that both independently assemble into homopolymers but also co-align in vivo and affect each other's properties are Crescentin and the CtpS enzyme from C. crescentus [62]. However, again, the co-assembly in vitro is not reported in the literature. Despite the numerous independently confirmed heteropolymerization properties of ZicK and ZacK, we note that the results from our in vivo experiments are based on artificial and heterologous expression of the two CCRPs and thus in itself could lead to artefacts. We hypothesize that the absence of a ZicK/ZacK heteropolymer in strains expressing ZicK-GFP or ZacK-GFP alone (with the WT zicK and zacK alleles still present) may be due to a dosage-dependent effect, where the presence of unequal concentration of ZicK or ZacK in the cell leads to protein aggregates. Our observation of ZicK-GFP or ZacK-GFP aggregates upon expression alone in the ΔzicKΔzacK mutant strain supports the dosage effect hypothesis. Also, in our in vitro polymerization assay, ZicK and ZacK only formed clear and distinct filament-like structures when both proteins are present in equal concentrations. Nonetheless, co-expression of ZicK-eCFP and ZacK-GFP could not complement the ΔzicKΔzacK mutant. Attempts to express both proteins fused to a fluorophore from the native promoter remained unsuccessful, possibly a result of the close genomic proximity. Furthermore, the genomic neighbourhood of zicK and zacK suggests that the ZicK/ZacK heteropolymer formation could be relying on cotranslational assembly (e.g., as observed for LuxA/LuxB [64]). Cotranslational assembly of natively present ZicK and ZacK would lead to an efficient binding of the two subunits such that the additional expression of one unit only in excess (i.e., ZicK-GFP or ZacK-GFP alone) would lead to the formation of aggregates. As such, it remains to be elucidated to what extent the ZicK/ZacK heteropolymer actually exists in Anabaena WT. Although we could not identify any protein filaments in our ultrathin sections from Anabaena WT, other studies have previously described filamentous strings and even longitudinal cell-spanning polymers in multicellular Anabaena and Nostoc strains [65,66]. Despite compelling evidence for the existence of a cyanobacterial Z-ring structure during cell division [48,67], no Z-ring ultrastructures have yet been identified, and consequently, the absence of longitudinal ZicK/ZacK filaments in ultrathin sections does not rule out that they exist but could also indicate that they could not be visualized yet.

A B
Our results indicate that ZicK and ZacK are associated with the elongasome (through their interaction with MreB) and proteins in the septal cell wall (through the interaction with SepJ and SepI) and might affect cellular DNA placement (Fig. 7). A function of ZicK/ZacK in chromosome segregation would be in concert with the identified interaction of ZicK with ParA; this, however, remains to be elucidated as it could also be an indirect consequence of the swollen/rounded cell shape in the ΔzicKΔzacK mutant, similar to the indirect effect of mreB deletion of carboxysome alignment in Synechococcus elongatus PCC 7942 [68]. Nonetheless, so far, no chromosome partitioning system has yet been identified in multicellular cyanobacteria. In Anabaena, deletion of mreB did not affect chromosome segregation but induced a swollen cell phenotype [38], similar to the ΔzicKΔzacK mutant. Consequently, MreB and ZicK/ZacK likely share functional properties but are not exclusively involved in the same cellular processes. Swollen cell morphotypes were also described for Anabaena or Synechocystis mutants lacking penicillin-binding proteins (PBPs), which are enzymes that are directly involved in cell wall biogenesis through the modification of the PG layer [39]. This presumed link of ZicK/ ZacK to the actin-like MreB cytoskeleton and the PG biogenesis apparatus is also indicated by the altered localization of GFP-MreB and the decreased PG staining intensity in the ΔzicKΔzacK mutant strain. Consequently, ZicK and ZacK might indirectly act to positively regulate PG biogenesis, although we cannot exclude that the reduced staining intensity in the ΔzicKΔzacK mutant simply reflects the slower growth rate of this strain. An interaction or involvement of prokaryotic filament-forming CCRPs with MreB and Fig. 7. Interaction network of known divisome, elongasome and septal proteins in Anabaena. A model for a partial divisome, elongasome and septal junction network in Anabaena as deduced from BACTH and co-IP analyses. Black arrows indicate interactions that have been previously described by [33,49,67]. Red arrows indicate interactions identified in the current analysis. PG synthesis was previously observed in other bacteria. Examples are the gliding motility in Myxococcus xanthus, where a multiprotein complex, including the filament-forming CCRP AglZ and MreB, was found to coordinate type A-motility [69]. Similarly, the curved morphotype of C. crescentus is induced by Crescentin, which functionally associates with MreB and likely modulates PG biogenesis by exuding local mechanical forces to the cell membrane [16,21]. Other aspects like a decreased cell envelope permeability of the ΔzicKΔzacK mutant are also conceivable, although we did not detect any cell wall defects in the ΔzicKΔzacK mutant. MreB and the elongasome are the main determinants of the PG exoskeleton, which provides the cell with structural integrity and resistance to turgor pressure [70]. The lack of liquid growth of the ΔzicKΔzacK mutant would also argue for a defect in the resistance to turgor pressure.
Together with the cell shape-determining protein MreB, ZicK and ZacK could possibly contribute to normal cell shape and relay trichome shape-stabilizing properties to neighbouring cells in the trichome by means of their association with the filament stabilizing protein SepJ (Fig. 7). As such, they are important for maintaining the linear Anabaena trichome phenotype. ZicK and ZacK polymers might constitute stabilizing platforms or scaffolds for other proteinaceous structures, similarly to the stabilizing function of the eukaryotic cytoskeleton for cell-cell contacts (i.e., desmosomes). Furthermore, ZicK shares in silico predicted structural similarities with the spectrin repeats of plectin, a well-described eukaryotic cytolinker protein. Plectin links the three eukaryotic cytoskeletal systems (actin filaments, microtubules and IFs), thereby contributing to the resistance to deformation of vertebrate cells [12]. They stabilize desmosomes and are hence directly involved in cell-cell contact integrity [71]. An analogous cytolinker function of ZicK could explain why ZacK alone did not form properly folded protein filaments on its own and suggests that ZacK requires ZicK as the linking protein for polymerization. As plectin not only stabilizes but also dynamically disassembles IF protein filaments (i.e., vimentin) in a concentration-dependent manner [72], this would further support a dosage-dependent effect of ZicK and ZacK for heteropolymerization.
The ubiquity of ZicK and ZacK in heterocystous cyanobacteria that form linear trichomes (Fig. 1C, File  S2) indicates that the presence of ZicK and ZacK contributes to the maintenance of a linear trichome shape. The ΔzicKΔzacK mutant had a zigzagged phenotype and was unable to grow in liquid culture. We hypothesize that the loss of linear trichome shape led to an increase in accessible surface for the acting mechanical forces in liquid [73], including fluid shear stress [74], ultimately resulting in forces that cannot be endured by the abnormal mutant trichomes. Nonetheless, the absence of ZicK and/or ZacK in several heterocystous cyanobacteria species, including the fern-symbiont N. azollae and ramified cyanobacteria, may suggest that these species are less sensitive to mechanical stress (i.e., due to their interaction with the host tissues or complex filament formation), and hints for the presence of additional, yet unknown, mechanisms for the determination of filament shape.
The key hallmarks of permanent bacterial multicellularity are morphological differentiation and a welldefined and reproducible shape, termed patterned multicellularity [75]. In sporulating actinomycetes, this is exemplified by the highly reproducible cell division, cell proliferation and cell differentiation pattern [76]. In filamentous cyanobacteria, the reproducible linear trichomes are considered a major contributor to the cyanobacterial patterned multicellularity [75], manifesting a selective advantage to biotic and abiotic environmental factors [77,78]. Our results indicate that ZicK and ZacK serve as regulators of the typical linear Anabaena trichome and as such as regulators of Anabaena patterned multicellularity. The evolution of patterned multicellularity is considered an important step towards a sustainable division of labour and the development of cell differentiation [75]. Our study provides initial evidence for a role of two heteropolymer-forming CCRPs in the evolution and maintenance of cyanobacterial multicellular forms.

Bacterial strains and growth conditions
Anabaena sp. PCC 7120 was obtained from the Pasteur Culture Collection (PCC) of cyanobacteria (France). Cells were grown photoautotrophically in BG11 or without combined nitrogen (BG11 0 ) at constant light with a light intensity of 30 µmolÁm −2 Ás −1 in liquid culture or on agar plates (1.5% w/v agar). When appropriate, 5 µgÁmL −1 spectinomycin (Sp), 5 µgÁmL −1 streptomycin (Sm) or 30 µgÁmL −1 neomycin (Nm) was added to strains carrying respective plasmids or chromosomal insertions. In some cases, basal copper-regulated petE-driven expression of gene candidates in Anabaena cells was lethal or growth inhibiting, therefore these strains were grown in BG11 without copper and protein expression was later induced by the addition of CuSO 4 at indicated concentrations to the culture. E. coli strains DH5α, DH5αMCR, XL1-blue and HB101 were used for cloning and conjugation by triparental mating. BTH101 was used for BACTH system, and BL21 (DE3) was used for expression of His 6 -tagged proteins in E. coli. All E. coli strains were grown in LB medium containing the appropriate antibiotics at standard concentrations. Tables S1-S4 list all used bacterial strains, plasmids and oligonucleotides.

Prediction of coiled-coil-rich proteins
Genome sequence of Anabaena (GCA_000009705.1) was analysed by the COILS algorithm [79] as previously described [20]. The algorithm was run with a window width of 21, and the cut-off for amino acids in coiled-coil conformation was set to ≥ 80 amino acid residues. The resulting set of protein candidates was further manually examined with online available bioinformatic tools, including NCBI conserved domain search [80], NCBI BLAST [81], TMHMM [82] and I-TASSER [83]. Protein candidates exhibiting BLAST hits involved in cytoskeletal processes or similar domain architectures as known IF and IF-like proteins like Crescentin, FilP, vimentin, desmin or keratin were selected, and enzymatic proteins as well as proteins predicted to be involved in other cellular processes were excluded.

Distribution of homologs in cyanobacteria
Homologs to the Anabaena proteins were extracted from precalculated cyanobacterial protein families [42]. Conserved syntenic blocks (i.e., gene order) were identified using CSBFinder-S [84].

Transformation
Anabaena was transformed by triparental mating as previously described [85].

Plasmid construction
Ectopic expression of Anabaena protein candidates was achieved from a self-replicating plasmid (pRL25C [86]) under the control of the copper-inducible petE promoter (P petE ) or the native promoter (predicted by BPROM [47]) of the respective gene. All constructs were verified by Sanger sequencing (Eurofins Genomics, Ebersberg, Germany). Plasmids were created using standard restriction enzymebased techniques or Gibson assembly. Information about precise plasmid construction strategies is available from the authors upon request.

Anabaena mutant strain construction
The ΔzicKΔzacK mutant strain was generated using the pRL278-based double homologous recombination system employing the conditionally lethal sacB gene [87]. For this, 1500 bp upstream and downstream of zicK-zacK was generated by PCR from Anabaena gDNA. Upstream region of zicK was amplified using primers #97/#98, and downstream region of zacK was amplified using primers #99/#100. Calcein was excited at 488 nm and fluorescence emission monitored in the range from 500 to 530 nm with a maximally opened pinhole (600 µm). FRAP experiments were carried out by an automated routine as previously described [58]. After recording an initial image, selected cells were bleached by increasing the laser intensity by a factor of 5 for two subsequent scans and the fluorescence recovery followed in 0.5-s intervals for 30 s was recorded using the Leica LAS X software. Exchange coefficients (E) were then calculated as previously described [58,90].

BODIPY ™ FL vancomycin (Van-FL) staining
Van-FL staining of strains grown on BG11 plates was essentially performed as described previously [91,92]. Briefly, cells were resuspended in BG11 medium, washed once in BG11 by centrifugation (6500 g, 4 min, RT) and incubated with 5 µgÁmL −1 Van-FL (dissolved in methanol; Thermo Fischer Scientific). Cells were incubated in the dark for 1 h at 30°C, washed three times with BG11 and immobilized on an agarose pad. Van-FL fluorescence signals were then visualized using epifluorescence microscopy with an excitation time of 130 ms. Arithmetic mean fluorescence intensities were recorded from the septa between two cells with a measured area of 3.52 µm 2 using the histogram option of the ZEN BLUE 2.3 software (Carl Zeiss).

Alcian blue staining
Anabaena WT and ΔzicKΔzicK cells were grown on BG11 0 plates, resuspended in BG11 0 liquid medium and stained with 0.05% (w/v) alcian blue (final concentration). Polysaccharide staining of cells immobilized on an agarose pad was observed with an Axiocam ERc 5s colour camera (Carl Zeiss).

Data analysis
Cell volume and roundness were determined using the imaging software IMAGEJ [93], and a perfect circle is defined to have a roundness of 1. Cell volume was calculated based on the assumption of an elliptic cell shape of Anabaena cells using the Major Axis and Minor Axis values given by IMAGEJ and the formula for the volume of an ellipsoid Angles between neighbouring cells were determined from chlorophyll autofluorescence micrographs from individual Anabaena trichomes in IMAGEJ. For this, the trichomes were separated from the background using the 'set autothreshold' command. The adjustable Watershed plug-in was used to separate individual cells in one trichome (tolerance = 1). After selecting cells with the analyse particle command (size = 200-2000 pixel; circularity = 0.20-1.00), the centre of each cell was determined as XY coordinates using the centroid measurement setting. The angles between three of

Bacterial two-hybrid and beta-galactosidase assays
Chemically competent E. coli BTH101 cells were cotransformed with 5 ng of plasmids carrying the respective T18 and T25 translational fusion constructs, plated onto LB plates supplemented with 200 µgÁmL −1 X-gal, 0.5 mM IPTG, Amp, Km and grown at 30°C for 24-36 h. Interactions were quantified by beta-galactosidase assays from three independent colonies. For this aim, cultures were grown for 2 days at 20°C in LB Amp, Km, 0.5 mM IPTG and beta-galactosidase activity was recorded as described in the manufacturer's instructions (Euromedex, Souffelweyersheim, France; BACTH System Kit Bacterial Adenylate Cyclase Two-Hybrid System Kit) in a 96-well plate format as previously described [94].

GFP-fragment reassembly assay
Chemically competent E. coli BL21 (DE3) were cotransformed with indicated plasmid combinations, plated on LB Amp, Km and grown over night at 37°C. Liquid overnight cultures of single colonies of the respective plasmid-bearing E. coli strains were then diluted 1 : 40 in the same medium the following day. Cells were grown for 2 h at 37°C, briefly acclimated to 20°C for 10 min, and protein expression was induced with 0.05 mM IPTG and 0.2% (w/v) Larabinose. Pictures of induced cultures grown at 20°C were taken 48 h after induction.

Mass spectrometry
Mass spectrometry of coprecipitated proteins was performed as previously described [49].

Immunofluorescence
Immunolocalization of FtsZ in Anabaena WT and ΔzicKΔzacK mutant was essentially performed as previously described [33]. For this, strains were scraped off from growth plates (BG11 and BG11 0 plates), resuspended in a small volume of distilled water and air-dried on Polysine ® adhesion slides (Menzel) at RT followed by fixation and permeabilization with 70% ethanol for 30 min at −20°C. Cells were allowed to air dry for 30 min at RT and then washed two times with PBST [PBS supplemented with 0.1% (v/v) Tween-20] for 2 min. Unspecific binding sites were blocked for 30 min at RT with blocking buffer (1× Roti ® -ImmunoBlock in PBST; Carl Roth, Karlsruhe, Germany), and afterwards, rabbit anti-FtsZ (Agrisera; raised against Anabaena FtsZ; 1 : 150 diluted) antibody in blocking buffer was added to the cells and incubated for 1.5 h at RT in a self-made humidity chamber followed by five washing steps with PBST. 7.5 µgÁmL −1 (final concentration) Alexa Fluor 488-conjugated goat anti-rabbit IgG (H+L) secondary antibody (Thermo Fischer Scientific) in blocking buffer was added to the cells and incubated for 1 h at RT in the dark in a self-made humidity chamber. Subsequently, cells were washed five times with PBST, air-dried and mounted with ProLong ™ Diamond Antifade Mountant (Thermo Fischer Scientific) overnight at 4°C. Immunolocalization of FtsZ was then analysed by epifluorescence microscopy.

Spot assays
For spot assays, Anabaena WT and ΔzicKΔzacK mutant strain was grown on BG11 growth plates, resuspended in BG11 liquid medium and adjusted to an OD 750 of 0.4. Cells were then spotted in triplicates of 5 µL onto the respective

Protein purification and in vitro filamentation assays
For protein purification, E. coli BL21 (DE3) cells carrying His-tagged protein candidates were grown in overnight cultures at 37°C and 250 r.p.m. The next day, overnight cultures were diluted 1 : 40 in the same medium and grown at 37°C until they reached an OD 600 of 0.5-0.6. Protein expression was induced with 0.5 mM IPTG for 3-4 h at 37°C and 250 r.p.m. Afterwards, cell suspensions of 50 mL aliquots were harvested by centrifugation, washed once in PBS and stored at −80°C until further use. For in vitro filamentation assays, cell pellets were resuspended in urea lysis buffer (ULB: 50 mM NaH 2 PO 4 , 300 mM NaCl, 25 mM imidazole, 6 M urea; pH 8.0) and lysed in a Precellys ® 24 homogenizer (3× 6500 r.p.m. for 30 s) using the 2 mL microorganism lysis kit (VK01; Bertin) or self-packed Precellys tubes with 0.1-mm glass beads. The resulting cell debris was pelleted by centrifugation at 21 000 g (10 min, 4°C), and the supernatant was incubated with 1 mL His-Pur ™ Ni-NTA resin (Thermo Fischer Scientific) for 1 h at 4°C in an overhead rotator. The resin was washed five times with 4× resin-bed volumes ULB and eluted in urea elution buffer (UEB: ULB supplemented with 225 mM imidazole). Total protein concentration was measured using the Qubit ® 3.0 Fluorometer (Thermo Fischer Scientific). Filament formation of purified proteins was induced by overnight dialysis against polymerization buffer (PLB: 50 mM PIPES, 100 mM KCl, pH 7.0; HLB: 25 mM HEPES, 150 mM NaCl, pH 7.4; or 25 mM HEPES pH 7.5) at 20°C and 180 r.p.m. with three bath changes using a Slide-A-Lyzer ™ MINI Dialysis Device (10K MWCO, 0.5 mL or 2 mL; Thermo Fischer Scientific). Purified proteins were stained with an excess of NHS-fluorescein (dissolved in DMSO; Thermo Fischer Scientific), and in vitro filamentation was analysed by epifluorescence microscopy. The NHS-fluorescein dye was previously successfully used to visualize in vitro FtsZ and CCRP protein filaments [42,95]. And we note that the His 6 -tag did not impact the in vitro polymerization properties of the CCRP FilP [25], confirming the applicability of our approach.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. zicK and zacK are expressed at standard growth conditions. Fig. S2. In vitro polymerization assay controls. Fig. S3. Co-polymerization of ZicK and ZacK is dosage-dependent. Fig. S4. Ectopic expression of ZicK and ZacK. Fig. S5. Host-independent heteropolymerization of ZicK and ZacK upon heterologous expression in E. coli. Fig. S6. Complementation of the zigzagged growth phenotype of the ΔzicKΔzacK mutant and its ability to grow under diazotrophic conditions.