Early biophysical characterization showed that the N- and C-termini of monomeric flagellin are mainly disordered and critical for polymerization, although they become highly structured in the filament [58,59]. Due to the tendency of full-length flagellin to polymerize, crystallization of S. Typhimurium flagellin FliC was achieved only after the removal of its terminal regions by limited proteolysis [60]. The overall shape of FliC obtained by combining X-ray crystallography and cryo-EM approaches revealed a structure in the form of a Greek letter Γ with four distinct domains (Figure 1a) [61,62]. Starting from the N-terminus, the polypeptide chain makes a single helix that forms one half of an α-helical domain called D0, continues with the partial formation of domains D1 and D2, and forms domain D3 prior to completing the folds of the previous three domains. Thus, domains D0-D2 have N- and C-terminal moieties.

The N- and C-terminal helices of D0 form a coiled coil in the filament. The organization of D1 with three α-helices and a β-hairpin resembles a four-helix bundle with an extensive hydrophobic core. D2 and D3 consist primarily of β-strands organized in a specific fold called β-foliums in which the tips of the β-hairpins are bent or twisted [61,62].

While all flagellin structures known so far exhibit a high degree of structural similarity in the D0 and D1 region, the structures of the variable regions differ extensively (Figure 1b,c). The structure of the P. aeruginosa type A FliC lacking the D0 domain showed that the D1 domain is highly similar to its counterpart in S. Typhimurium [63]. Conversely, the D2 domain adopts a different fold with two β-sheets and one α-helix between them forming a less flexible cup-like structure positioned parallel to the D1 domain, instead of pointing away from it as in Salmonella (Figure 1c). In the case of Campylobacter, the variable region of the flagellin FlaA is larger and forms three domains (D2, D3 and D4) (Figure 1c) [64]. The D2 and D3 domains of the Campylobacter FlaA are structural homologs of each other and of the Pseudomonas FliC D2. D4 is inserted between C-terminal moieties of D2 and D1, with most of the FlaA glycans located on this domain. It is the most exposed outer domain resembling a shield for both D2 and D3 and, contrary to them, D4 does not have structural homologs.

The presence of posttranslational modifications in flagellins was suspected early on, but their extent and importance became apparent only recently. Investigating discrepancies between theoretical and observed molecular weights showed that Pseudomonas type A flagellin was glycosylated [55]. A genomic island between flgL and fliC genes containing 14 open reading frames (ORFs) is responsible for glycosylation and it specifically modifies type A flagellin, while having no effect on type B [65]. These glycan chains in the PAK strain are O-linked to residues T189 and S260 through rhamnose and vary in length (up to 11 monosaccharides) and composition [66].

The type B flagellin characteristic of the PAO1 strain also contains O-linked glycans at serine residues 191 and 195, although in this case it is a much simpler modification with a single monosaccharide [67]. The genomic island responsible for type B glycosylation is smaller than its type A counterpart and contains four ORFs. Pseudomonas type A and B mutants in which glycosylation is absent or incomplete do not affect motility in general, although the lack of glycosylation significantly decreased virulence of both PAK and PAO1 strains in a burned-mouse model of infection [68]. Apart from O-linked glycans, N-linked glycosylation was also reported in P. aeruginosa strain PA14, with three sites on the D0 and D1 domains proposed as the most likely candidate positions [69].

Campylobacter flagellin is glycosylated at 19 serine and threonine residues with O-linked pseudaminic acid or related derivatives [70,71,72]. Glycosylation in C. jejuni is responsible for the 6 kDa difference between predicted and experimentally obtained molecular weight. Approximately 50 genes in a locus that is adjacent to the flaA and flaB genes encode the Campylobacter glycosylation machinery [73]. Glycosylation of flagellin in Campylobacter is necessary for filament assembly as mutants that produce unglycosylated flagellin do not have filaments and flagellin accumulates intracellularly [74].

H. pylori flagellins are also glycosylated with pseudaminic acid with most glycosylation sites in the central core region of the protein [75]. As in Campylobacter, glycosylation plays an important role during assembly of the filament; abolishing glycosylation leads to loss of the filament and motility, even though the levels of flagellin mRNA are unaffected [75].

The flagellin of Listeria monocytogenes has O-linked N-acetylglucosamines at up to six sites located in the central surface exposed region of the protein [66]. The flagellin of P. syringae is also glycosylated and these modifications are important for its virulence and swarming [76,77].

Another modification that is present in flagellins is methylation. In S. Typhimurium, the presence of methylated lysines was reported as early as 1959 [78]. Lysine methylation is performed by the enzyme FliB. Abolishing methylation does not affect filament assembly or motility [79]. However, it was shown that it promotes adhesion and host cell invasion [80]. N-methylation was also detected in Shewanella oneidensis [81].

One of the questions that draws much attention is the structural basis for the polymorphic switching of the filament. Comparison of the L- and R-structures obtained by cryo-EM revealed that the D0 domains align well with some local changes in conformation, while the rest of the molecule is twisted by 5° in the counterclockwise direction [90]. Previous X-ray diffraction study on L- and R-types from Salmonella showed that there were no changes in helical parameters in the inner region of the core made of D0 domains [91]. However, the two types differ in the outer regions and the distance between two subunits in the protofilament is 0.8 Å shorter in the R-type compared to the L-type. Maki-Yonekura et al. [90] found that the packing of the hydrophobic side chains at the lower portion of D1 is different between L- and R-type and that the intersubunit interactions along the 5-start helix are important in stabilizing the conformation. Based on molecular dynamics simulations of the polymorphic supercoiling mechanism, Kitao et al. [92] identified three types of interaction between subunits: permanent (the same pair of residues in different states), sliding (the same types of interaction with variable partners) and switch interactions that were responsible for locking the protofilament interface in R- or L-handed state. The switch interactions were found only along the 5-start helices. However, this was not supported by the analysis of the corresponding interactions in L- and R-filaments of B. subtilis [86].

Originally designated as a HAP2, FliD is localized at the tip of the filament where several copies of it assemble into a capping structure. The presence of the cap prevents the leakage of flagellin that is continuously exported through the flagellar type III secretion system into the filament channel and facilitates its insertion at the growing end of the filament. Early cryo-EM studies of the Salmonella cap showed that it is a stool-like pentameric structure with the flat star-shaped head region and five legs positioned almost at the right angle. The head region is 145 Å in dimeter and the height of the cap is 125 Å.

While they vary in size, FliDs from different bacteria have conserved N- and C-terminal regions predicted to be coiled-coils and are largely unstructured in the monomeric state [93]. The first high-resolution structure of the P. aeruginosa FliD showed that this 50 kDa protein has three distinct domains: a highly flexible D1 domain that corresponds to the leg domains seen in the cap structure, and compact D2 and D3 domains that comprise the head [94]. Much as in flagellin, a continuous sequence in the middle of the polypeptide folds into D3 while domains D1 and D2 consist of N- and C-terminal portions of the chain relative to D3. Domains D2 and D3 are each made of two antiparallel β-sheets, while D1 could not be resolved with confidence apart from one helix. The structure of the E. coli FliD offered more insight into D1, showing that it is a helical bundle of at least four helices with a β-hairpin similar to the flagellin D1 domain, although approximately 50 terminal residues from both sides were not present in the construct (Figure 3a,c) [95]. These structures, together with the ones from Salmonella and Serratia marcescens [95,96], indicated that the structural organization of the capping proteins among different bacteria is preserved. However, not all FliDs are of the same size. For example, the FliD of H. pylori is 74 kDa and the crystal structure revealed that apart from domains D1–3 present in the previously studied proteins, it has two additional domains, D4 and D5, both of which reside in the head region with similar overall folds to D2 and D3 (Figure 3b) [97]. These domains may be related to the specific role FliD plays as a virulence factor during H. pylori infection; stronger antigenicity of D4 and D5 further supports this [97].

Once the hook and the junction are formed, FliD forms the filament cap, promoting the polymerization of FliC. The oligomeric state of the filament cap is still an open question. Initial characterization of the recombinant FliD of Salmonella showed the existence of decamers in solution, but the combination of cross-linking experiments, analytical ultracentrifugation and electron microscopy revealed that a pentamer is the functional unit, decamers being the product of a non-physiological cap dimerization. This was generally also accepted as the case for other bacteria, and the recent cryo-EM structure of the pentameric C. jejuni cap supports it [98]. However, both Pseudomonas and E. coli FliDs crystallized as hexamers and Serratia FliD as a tetramer (Figure 3d) [94,95,96]. Moreover, tetramers and pentamers of Pseudomonas FliD and tetramers of Salmonella FliD were observed in vitro [94,98]. While these stoichiometries could arise due to non-physiological conditions, in the case of P. aeruginosa hexamers were also detected in vivo [94].

The mechanism of cap-filament interaction is yet to be explained. During filament elongation, new flagellin molecules arrive at the tip continuously. This would require high mobility of the cap and continuous conformational changes of the D1 domains, most probably independently from one another, in order to accommodate arriving FliC subunits, all while staying firmly attached to the filament. One way in which FliD could interact with FliC is through its highly flexible terminal regions, present in all proteins in the outer part of the flagellum and are shown to mediate protein–protein interactions.

The genetic organization of the flagellar components is highly complex and relatively well conserved across bacteria with the same type of flagellar arrangement (i.e., peritrhichous, monotrichous/polar, etc.) but quite different between them [1,99,100]. The bacterial flagellar regulon commonly consists of dozens of genes grouped in several operons, encoding all structural proteins of the flagellum, the chemosensory apparatus and the regulators that control gene expression. For example, there are about 50 genes involved in the synthesis of the polar P. aeruginosa flagella, distributed in 17 putative operons comprising 41 flagellar genes (26 encoding structural proteins, eight encoding regulators and six involved in the export apparatus), linked to nine genes involved in chemotaxis and clustered in three regions of the chromosome constituting the fla regulon [101,102]. On the other hand, in peritrichous bacteria such as Salmonella spp. and Escherichia coli, the flagellar regulon contains 70 genes distributed in at least 25 operons [79,99]. The transcription of these genes is hierarchical and controlled by different promoter classes that are temporally regulated (Figure 4, Table 1). In peritrichous bacteria, flagellar genes are organized into three classes (I-III), starting with the transcription of the class I master operon as a response to the environmental stimuli that leads to the synthesis of the transcriptional regulator FlhDC. This regulator, in turn, activates the transcription of the class II genes responsible for the assembly of the hook-basal body (HBB) complex. Among class II genes are two crucial regulators of the class III genes: transcription factor FliA (σ²⁸) responsible for the transcription of class III genes, and FlgM, which acts as a FliA inhibitor. Once the HBB is assembled, FlgM is exported out of the cell via the HBB central channel, releasing FliA to interact with RNA polymerase and guide the transcription of the class III genes, including flagellin. This three-tiered organization ensures that cells produce flagella in response to the environmental signals and that energetically costly synthesis of the filament proceeds only when enough HBBs have been formed. An exception is the Actinoplanes missouriensis zoospore in which there is no hierarchical coordination and 33 flagellar genes are transcribed simultaneously during the sporangium formation [103]. The filament cap gene fliD belongs to the group of class II genes, whereas flagellin genes belong to class III. Although fliD is expressed as a class II gene, it is not assembled into the growing flagellar structure until the hook and the junction proteins are expressed from class III (including flagellin genes) and incorporated into the flagellar organelle [104]. Hence, this cascade serves to control the timing of gene expression to coincide with the assembly of the flagellar apparatus and filament.

As mentioned above, S. Typhimurium exhibits phase variation, or alternate expression of two flagellin genes, fliC and fljB. This process is regulated by a control region upstream of fljB that is a subject of inversion, which orients the region forward and generates a promoter. This allows co-transcription of fljB and fljA, a transcriptional repressor that inhibits the expression of FliC [105].

Recently, Rao et al. found that flagellar gene expression in Salmonella is bimodal, meaning that in a population of genetically identical cells under the same conditions only a fraction of them is motile [106,107]. This bimodality is present at both the class II and class III levels and is governed by separate mechanisms. While a double negative feedback loop involving two flagellar regulatory proteins, RflP and FliZ, controls the expression of the class II genes in response to nutrient availability, class III gene expression is tuned by the secretion of FlgM and there is a minimum number of HBBs necessary for the cell to pass this checkpoint.

In bacteria with a polar flagellum, flagellar genes are transcribed in a four-tiered hierarchy, with genes encoding components of the HBB split between class II and class III (Figure 4) [101]. Gene regulation in Pseudomonas is more complex and involves the activation of the two-component system FleS-FleR and another sigma factor RpoN (σ⁵⁴) to activate class III genes. In addition, unlike in Salmonella, FliA (σ²⁸) is constitutively expressed in Pseudomonas independently of other flagellar genes, which makes it one of the class I genes [101].

In bacteria with multiple flagellins, such as Helicobacter and Campylobacter, expression is under the control of different sigma factors with FlaA under control of σ²⁸, while FlaB is under control of σ⁵⁴ [108,109]. Although the major flagellin is usually under the control of σ²⁸, there are exceptions such as V. cholera and S. oneidensis in which the major flagellin is dependent on σ⁵⁴, while commensal gut bacteria of Eubacterium and Roseburia species are under the control of σ²⁸ and σ⁴³ [110,111]. In the phylogenetically related Butyrivibrio fibrisolvens, transcription of one fliC gene is driven from two different promoters, yielding two transcripts with alternative transcription start-sites [111].

In some Gram-positive bacteria, changes in the environmental conditions such as nutrient limitation induce variations in the levels of the intracellular messengers guanosine tetra/pentaphosphate, (p)ppGpp, guanosine nucleoside triphosphate (GTP) and branched chain amino-acid pools. These variations are detected by a conserved GTP-sensing protein CodY and a global regulator CsrA that modulate flagellin expression. Cell motility can be repressed under elevated intracellular cyclic-di-GMP levels, by impeding transcription of some of the flagellum genes [112,113,114,115,116]. Indeed, the expression of fliD and fliC is repressed in phosphodiesterase 3 (PDE3) knockout mutants triggered by elevated c-di-GMP accumulation [117].

Because the presence of flagellin can be deleterious to the bacterium on account of inducing host immunity, flagellated bacteria downregulate (or turn off) flagellin expression during the host invasion to avoid host immune responses. Therefore, normal microbiota within the healthy adult mammalian gut has been shown to have overall relatively low levels of flagellin expression, while TLR5−/− mice exhibited a diversity of gut microbiome members with overexpressed flagellar genes [118,119].

Both commensal gut strains of motile E. coli and pathogenic S. Typhimurium strains strongly down-regulate their genes coding for flagellar machinery and lose their motility once inside the host [120,121]. Under environmental temperature conditions (22–30 °C), the expression of flagellar genes in the human pathogens Listeria monocytogenes and Legionela pneumophila is normal, but significantly reduced when raised to 37 °C, revealing temperature-dependent transcription mainly controlled by the protein GmaR, which acts as a protein expression thermostat [122,123]. These types of system provide a pathogen with the ability to turn off immune-stimulating antigens before they trigger adverse host defense mechanisms once inside their target host.

Regulation of flagellin expression also occurs on the post-transcriptional level by proteins that bind to the untranslated leader region of the flagellin mRNA, affecting transcript stability and/or ribosome access [124,125,126]. For instance, the RNA-binding protein CsrA of B. burgdorferi specifically mediates synthesis of the major flagellin by inhibiting translation initiation of its transcript. In some cases, post-transcriptional regulators repress the accumulation of flagellin when cells are defective for flagellar hook assembly [127,128].

An assembly checkpoint that prevents flagellin translation and assembly prior to hook completion was discovered in B. subtilis. This is governed by a homeostatic mechanism in which the flagellin protein itself is a critical regulator. In a partner switching mechanism, the flagellar assembly factor FliW binds flagellin, while a global regulator CsrA binds flagellin mRNA, repressing its translation. After completion of the hook, flagellin is secreted, and the released FliW binds CsrA, thus derepressing flagellin translation [129]. An interesting case is A. missouriensis, in which all flagellar genes are transcribed simultaneously during sporangium formation and there is no checkpoint mechanism in the process of flagellar gene transcription to optimize the efficiency of the flagellar assembly [103]. The post-transcriptional level of regulation between protein production and assembly could play an important role, although this remains unknown.

Accumulation and premature oligomerization of flagellar axial proteins in the cytosol could be wasteful and detrimental to the cell. For this reason, bacteria encode flagellar-specific chaperones dedicated to facilitating the assembly and export of the flagellum components mainly by protecting and/or preventing their cognate flagellar protein substrates from aggregation or avoiding premature undesired interactions of flagellar proteins in the cytoplasm prior to interaction with the export gate [130]. Due to the small diameter of the flagellar export central channel (20–30 Å), the proteins destined for incorporation into the growing flagellum must be exported in a partially or completely unfolded state, implying that premature folding and oligomerization in the cytosol must be prevented to keep them in a secretion-competent conformation for flagellum assembly [131]. Hence, the external flagellar components require T3SS-specific chaperones to facilitate their efficient export [99].