Structural insights into the Escherichia coli lysine decarboxylases and molecular determinants of interaction with the AAA+ ATPase RavA

The inducible lysine decarboxylase LdcI is an important enterobacterial acid stress response enzyme whereas LdcC is its close paralogue thought to play mainly a metabolic role. A unique macromolecular cage formed by two decamers of the Escherichia coli LdcI and five hexamers of the AAA+ ATPase RavA was shown to counteract acid stress under starvation. Previously, we proposed a pseudoatomic model of the LdcI-RavA cage based on its cryo-electron microscopy map and crystal structures of an inactive LdcI decamer and a RavA monomer. We now present cryo-electron microscopy 3D reconstructions of the E. coli LdcI and LdcC, and an improved map of the LdcI bound to the LARA domain of RavA, at pH optimal for their enzymatic activity. Comparison with each other and with available structures uncovers differences between LdcI and LdcC explaining why only the acid stress response enzyme is capable of binding RavA. We identify interdomain movements associated with the pH-dependent enzyme activation and with the RavA binding. Multiple sequence alignment coupled to a phylogenetic analysis reveals that certain enterobacteria exert evolutionary pressure on the lysine decarboxylase towards the cage-like assembly with RavA, implying that this complex may have an important function under particular stress conditions.


Results and Discussion
CryoEM 3D reconstructions of LdcC, LdcI a and LdcI-LARA. In the frame of this work, we produced two novel subnanometer resolution cryoEM reconstructions of the E. coli lysine decarboxylases at pH optimal for their enzymatic activity -a 5.5 Å resolution cryoEM map of the LdcC (pH 7.5) for which no 3D structural information has been previously available (Figs 1A,B and S1), and a 6.1 Å resolution cryoEM map of the LdcI a , (pH 6.2) (Figs 1C,D and S2). In addition, we improved our earlier cryoEM map of the LdcI-LARA complex from 7.5 Å to 6.2 Å resolution (Figs 1E,F and S3). Based on these reconstructions, reliable pseudoatomic models of the three assemblies were obtained by flexible fitting of either the crystal structure of LdcI i or a derived structural homology model of LdcC (Table S1). Significant differences between these pseudoatomic models can be interpreted as movements between specific biological states of the proteins as described below.
The wing domains as a stable anchor at the center of the double-ring. As a first step of a comparative analysis, we superimposed the three cryoEM reconstructions (LdcI a , LdcI-LARA and LdcC) and the crystal structure of the LdcI i decamer ( Fig. 2 and Movie S1). This superposition reveals that the densities lining the central hole of the toroid are roughly at the same location, while the rest of the structure exhibits noticeable changes. Specifically, at the center of the double-ring the wing domains of the subunits provide the conserved basis for the assembly with the lowest root mean square deviation (RMSD) (between 1.4 and 2 Å for the Cα atoms only), whereas the peripheral CTDs containing the RavA binding interface manifest the highest RMSD (up to 4.2 Å) (Table S2). In addition, the wing domains of all structures are very similar, with the RMSD after optimal rigid Scientific RepoRts | 6:24601 | DOI: 10.1038/srep24601 body alignment (RMSD min ) less than 1.1 Å. Thus, taking the limited resolution of the cryoEM maps into account, we consider that the wing domains of all the four structures are essentially identical and that in the present study the RMSD of less than 2 Å can serve as a baseline below which differences may be assumed as insignificant. This preservation of the central part of the double-ring assembly may help the enzymes to maintain their decameric state upon activation and incorporation into the LdcI-RavA cage.
The core domain and the active site rearrangements upon pH-dependent enzyme activation and LARA binding. Both visual inspection (Fig. 2) and RMSD calculations (Table S2) show that globally the three structures at active pH (LdcI a , LdcI-LARA and LdcC) are more similar to each other than to the structure determined at high pH conditions (LdcI i ). The decameric enzyme is built of five dimers associating into a 5-fold symmetrical double-ring 17 (two monomers making a dimer are delineated in Fig. 1). As common for the α family of the PLP-dependent decarboxylases 17,25 , dimerization is required for the enzymatic activity because the active site is buried in the dimer interface (Fig. 3A,B). This interface is formed essentially by the core domains with some contribution of the CTDs. The core domain is built by the PLP-binding subdomain (PLP-SD, residues 184-417) flanked by two smaller subdomains rich in partly disordered loops -the linker region (residues 130-183) and the subdomain 4 (residues 418-563). Zooming in the variations in the PLP-SD shows that most of the structural changes concern displacements in the active site ( Fig. 3C-F). The most conspicuous differences between the PLP-SDs can be linked to the pH-dependent activation of the enzymes. The resolution of the cryoEM maps does not allow modeling the position of the PLP moiety and calls for caution in detailed mechanistic interpretations in terms of individual amino acids. Therefore we restrict our analysis to secondary structure elements. In particular, transition from LdcI i to LdcI-LARA involves ~3.5 Å and ~4.5 Å shifts away from the 5-fold axis in LdcI a (C) and LdcI-LARA (E) decamers with one protomer in light grey. In the rest of the protomers, the wing, core and C-terminal domains are colored from light to dark in shades of green for LdcC (A), pink for LdcI a (C) and blue for LdcI in LdcI-LARA (E). In (E), the LARA domain density is shown in dark grey. Two monomers making a dimer are delineated. Scale bar 50 Å. (B,D,F) One protomer from the cryoEM map of the LdcC (B), LdcI a (D) and LdcI-LARA (F) in light grey with the pseudoatomic model represented as cartoons and colored as the densities in (A,C,E). Each domain is indicated for clarity. Scale bar 50 Å. See also Figs S1 and S3. the active site α -helices spanning residues 218-232 and 246-254 respectively ( Fig. 3C-E). Between these two extremes, the PLP-SDs of LdcI a and LdcC are similar both in the context of the decamer (Fig. 3F) and in terms of RMSD min = 0.9 Å, which probably reflects the fact that, at the optimal pH, these lysine decarboxylases have a similar enzymatic activity 26 . In addition, our earlier biochemical observation that the enzymatic activity of LdcI a is unaffected by RavA binding 19 is consistent with the relatively small changes undergone by the active site upon transition from LdcI a to LdcI-LARA. Worthy of note, our previous comparison of the crystal structure of LdcI i with that of the inducible arginine decarboxylase AdiA 17 revealed high conservation of the PLP-coordinating residues and identified a patch of negatively charged residues lining the active site channel as a potential binding site for the target amino acid substrate 17  Rearrangements of the ppGpp binding pocket upon pH-dependent enzyme activation and LARA binding. An inhibitor of the LdcI and LdcC activity, the stringent response alarmone ppGpp, is known to bind at the interface between neighboring monomers within each ring (Fig. S4). The ppGpp binding pocket is made up by residues from all domains and is located approximately 30 Å away from the PLP moiety 17 . Whereas the crystal structure of the ppGpp-LdcI i was solved to 2 Å resolution, only a 4.1 Å resolution structure of the ppGpp-free LdcI i could be obtained 17 . At this resolution, the apo-LdcI i and ppGpp-LdcI i structures (both solved at pH 8.5) appeared indistinguishable except for the presence of ppGpp 17 (Fig. S11 in ref. 17). Thus, we speculated that inhibition of LdcI by ppGpp would be accompanied by a transduction of subtle structural changes at the level of individual amino acid side chains between the ppGpp binding pocket and the active site of the enzyme 17 . All our current cryoEM reconstructions of the lysine decarboxylases were obtained in the absence of ppGpp in order to be closer to the active state of the enzymes under study. While differences in the ppGpp binding site could indeed be visualized (Fig. S4), the level of resolution warns against speculations about their significance. The fact that interaction with RavA reduces the ppGpp affinity for LdcI 21 despite the long distance of ~30 Å between the LARA domain binding site and the closest ppGpp binding pocket (Fig. S5) seems to favor an allosteric regulation mechanism. Interestingly, although a number of ppGpp binding residues are strictly conserved between LdcI and AdiA that also forms decamers at low pH optimal for its arginine decarboxylase activity, no ppGpp regulation of AdiA could be demonstrated 26 . Swinging and stretching of the CTDs upon pH-dependent LdcI activation and LARA binding.
Inspection of the superimposed decameric structures (Figs 2 and S6) suggests a depiction of the wing domains as an anchor around which the peripheral CTDs swing. This swinging movement seems to be mediated by the core domains and is accompanied by a stretching of the whole LdcI subunits attracted by the RavA magnets. Indeed, all CTDs have very similar structures (RMSD min < 1 Å). Yet the superposition of the decamers lays bare a progressive movement of the CTD as a whole upon enzyme activation by pH and the binding of LARA. The LdcI i monomer is the most compact, whereas LdcI a and especially LdcI-LARA gradually extend their CTDs towards The C-terminal β-sheet of a lysine decarboxylase as a major determinant of the interaction with RavA. In our previous contribution, based on the fit of the LdcI i and the LARA crystal structures into the LdcI-LARA cryoEM density, we predicted that the LdcI-RavA interaction should involve the C-terminal two-stranded β-sheet of the LdcI 23 . Our present cryoEM maps and pseudoatomic models provide first structure-based insights into the differences between the inducible and the constitutive lysine decarboxylases. However, at the level of this structural element the two proteins are actually surpisingly similar. Therefore, we wanted to check the influence of the primary sequence of the two proteins in this region on their ability to interact with RavA. To this end, we swapped the relevant β -sheets of the two proteins and produced their chimeras, namely LdcIC (i.e. LdcI with the C-terminal β -sheet of LdcC) and LdcCI (i.e. LdcC with the C-terminal β -sheet of LdcI) (Fig. 5A-C). Both constructs could be purified and could form decamers visually indistinguishable from the wild-type proteins. As expected, binding of LdcI to RavA was completely abolished by this procedure and no LdcIC-RavA complex could be detected. On the contrary, introduction of the C-terminal β -sheet of LdcI into LdcC led to an assembly of the LdcCI-RavA complex. On the negative stain EM grid, the chimeric cages appeared less rigid than the native LdcI-RavA, which probably means that the environment of the β -sheet contributes to the efficiency of the interaction and the stability of the entire architecture ( Fig. 5D-F). The C-terminal β-sheet of a lysine decarboxylase is a highly conserved signature allowing to distinguish between LdcI and LdcC. Alignment of the primary sequences of the E. coli LdcI and LdcC shows that some amino acid residues of the C-terminal β -sheet are the same in the two proteins, whereas others are notably different in chemical nature. Importantly, most of the amino acid differences between the two enzymes are located in this very region. Thus, to advance beyond our experimental confirmation of the C-terminal β -sheet as a major determinant of the capacity of a particular lysine decarboxylase to form a cage with RavA, we set out to investigate whether certain residues in this β -sheet are conserved in lysine decarboxylases of different enterobacteria that have the ravA-viaA operon in their genome. We inspected the genetic environment of lysine decarboxylases from 22 enterobacterial species referenced in the NCBI database, corrected the gene annotation where necessary (Tables S3 and S4), and performed multiple sequence alignment coupled to a phylogenetic analysis (see Methods). This procedure yielded several unexpected and exciting results. First of all, consensus sequence for the entire lysine decarboxylase family was derived. Second, the phylogenetic analysis clearly split the lysine decarboxylases into two groups (Fig. 6A). All lysine decarboxylases predicted to be "LdcI-like" or biodegradable based on their genetic environment, as for example their organization in an operon with a gene encoding the CadB antiporter (see Methods), were found in one group, whereas all enzymes predicted as "LdcC-like" or biosynthetic partitioned into another group. Thus, consensus sequences could also be determined for each of the two groups (Figs 6B,C and S7). Inspection of these consensus sequences revealed important differences between the groups regarding charge, size and hydrophobicity of several residues precisely at the level of the C-terminal β -sheet that is responsible for the interaction with RavA (Fig. 6B-D). For example, in our previous study 23 , site-directed mutations identified Y697 as critically required for the RavA binding. Our current analysis shows that Y697 is strictly conserved in the "LdcI-like" group whereas the "LdcC-like" enzymes always have a lysine in this position; it also uncovers several other residues potentially essential for the interaction with RavA which can now be addressed by site-directed mutagenesis. The third and most remarkable finding was that exactly the same separation into "LdcI-like" and "LdcC"-like groups can be obtained based on a comparison of the C-terminal β-sheets only, without taking the rest of the primary sequence into account. Therefore the C-terminal β-sheet emerges as being a highly conserved signature sequence, sufficient to unambiguously discriminate between the "LdcI-like" and "LdcC-like" enterobacterial lysine decarboxylases independently of any other information (Figs 6 and S7). Our structures show that this motif is not involved in the enzymatic activity or the oligomeric state of the proteins. Thus, enterobacteria identified here (Fig. 6, Table S4) appear to exert evolutionary pressure on the biodegradative lysine decarboxylase towards the RavA binding. One of the elucidated roles of the LdcI-RavA cage is to maintain LdcI activity under conditions of enterobacterial starvation by preventing LdcI inhibition by the stringent response alarmone ppGpp 21 . Furthermore, the recently documented interaction of both LdcI 27 and RavA 22 with specific subunits of the respiratory complex I, together with the unanticipated link between RavA and maturation of numerous iron-sulfur proteins, tend to suggest an additional intriguing function for this 3.5 MDa assembly. The conformational rearrangements of LdcI upon enzyme activation and RavA binding revealed in this work, and our amazing finding that the molecular determinant of the LdcI-RavA interaction is the one that straightforwardly determines if a particular enterobacterial lysine decarboxylase belongs to "LdcI-like" or "LdcC-like" proteins, should give a new impetus to functional studies of the unique LdcI-RavA cage. Besides, the structures and the pseudoatomic models of the active ppGpp-free states of both the biodegradative and the biosynthetic E. coli lysine decarboxylases offer an additional tool for analysis of their role in UPEC infectivity. Together with the apo-LdcI and ppGpp-LdcI i crystal structures, our cryoEM reconstructions provide a structural framework for future studies of structure-function relationships of lysine decarboxylases from other enterobacteria and even of their homologues outside Enterobacteriaceae. For example, the lysine decarboxylase of Eikenella corrodens is thought to play a major role in the periodontal disease and its inhibitors were shown to retard gingivitis development [28][29][30] . Finally, cadaverine being an important platform chemical for the production of industrial polymers such as nylon, structural information is valuable for optimisation of bacterial lysine decarboxylases used for its production in biotechnology [31][32][33] .
Chimeric LdcIC and LdcCI were constructed, expressed and purified as follows. The chimeras were designed by exchange, between LdcI and LdcC, of residues from 631 to 640 and from 697 to the C-terminus, corresponding to the regions around the two strands of the C-terminal β -sheet (Fig. 5B,C), while leaving the rest of the sequence unaltered. The synthetic ldcIC and ldcCI genes (2148 bp and 2154 bp respectively), provided within a pUC57 vector (GenScript) were subcloned into pET-TEV vector based on pET-28a (Invitrogen) containing an N-terminal TEV-cleavable 6x-His-Tag. Proteins were expressed in Rosetta 2 (DE3) cells (Novagen) in LB medium supplemented with kanamycin and chloramphenicol at 37 °C, upon induction with 0.5 mM IPTG at 18 °C. Cells were harvested by centrifugation, the pellet resuspended in a 50 mM Tris-HCl, 150 mM NaCl, pH 8 buffer supplemented with Complete EDTA free (Roche) and 0.1 mM PMSF (Sigma), and disrupted by sonication at 4 °C. After centrifugation at 75000 g at 4 °C for 20 min, the supernatant was loaded on a Ni-NTA column. The eluted protein-containing fractions were pooled and the His-Tag removed by incubation with the TEV protease at 1/100 ratio and an extensive dialysis in a 50 mM Tris-HCl, 150 mM NaCl, 1 mM DTT, 5 mM EDTA, pH 8 buffer. After a second dialysis in a 50 mM Tris-HCl, 150 mM NaCl, pH 8 buffer supplemented with 10 mM imidazole, the sample was loaded on a Ni-NTA column in the same buffer, which allowed to separate the TEV protease and LdcCI/LdcIC. Finally, the pure proteins were obtained by size exclusion chromatography on a Superdex-S200 column in buffer A.
LdcI a -cryoEM data collection and 3D reconstruction. LdcI was prepared at 2 mg/ml in a buffer containing 25 mM MES, 100 mM NaCl, 0.2 mM PLP, 1 mM DTT, pH 6.2. 3 μl of sample were applied to glow-discharged quantifoil grids 300 mesh 2/1 (Quantifoil Micro Tools GmbH, Germany), excess solution was blotted during 2.5 s with a Vitrobot (FEI) and the grid frozen in liquid ethane 34 . Data collection was performed on a FEI Polara microscope operated at 300 kV under low dose conditions. Micrographs were recorded on Kodak SO-163 film at 59,000 magnification, with defocus ranging from 0.6 to 4.9 μm. Films were digitized on a Zeiss scanner (Photoscan) at a step size of 7 μm giving a pixel size of 1.186 Å. The contrast transfer function (CTF) for each micrograph was determined with CTFFIND3 35 .
Initially ~2500 particles of 256 × 256 pixels were extracted manually, binned 4 times and subjected to one round of multivariate statistical analysis and classification using IMAGIC 36 . Representative class averages corresponding to projections in different orientations were used as input for an ab-initio 3D reconstruction by RICOserver (rico.ibs.fr/) 37 . The resulting 3D reconstruction was refined using RELION 38 , which yielded an 18 Å resolution map. Using projections of this model as a template, particles of size 256 × 256 pixels were semi-automatically selected from all the micrographs using the Fast Projection Matching (FPM) algorithm 39 . The resulting dataset of ~46000 particles was processed in RELION with the previous map as an initial model and with a full CTF correction after the first peak. The final map comprised 44207 particles with a resolution of 7.4 Å as per the gold-standard FSC = 0.143 criterion 40 . It was sharpened with EMBfactor 41 using calculated B-factor of − 350 Å 2 and masked with a soft mask to obtain a final map with a resolution of 6.1 Å (Fig. S3, Table S1).

LdcC -cryoEM data collection and 3D reconstruction.
LdcC was prepared at 2 mg/ml in a buffer containing 25 mM HEPES, 100 mM NaCl, 0.2 mM PLP, 1 mM DTT, pH 7.2. Grids were prepared and sample imaged as LdcI a . Data were processed essentially as LdcI a described above. Briefly, an initial ~20 Å resolution model was generated by angular reconstitution after manual picking of circa 3000 particles from the first micrographs, filtered to 60 Å resolution, refined with RELION and used for a semi-automatic selection with FPM. The dataset was processed in RELION with a full CTF correction to yield a final reconstruction comprising 61000 particles. The map was sharpened with Relion postprocessing, using a soft mask and automated B-factor estimation (− 690 Å 2 ), yielding a map at 5.5 Å resolution (Fig. S1, Table S1).
LdcI-LARA -3D reconstruction. In our first study 23 , the dataset was processed in SPIDER and the CTF correction involved a simple phase-flipping. Therefore, for consistency with the present work, we revisited the dataset and processed it in RELION with a full CTF correction after the first peak. It was sharpened with EMBfactor 41 using calculated B-factor of − 350 Å 2 and masked with a soft mask to obtain a final map with a resolution of 6.2 Å (Fig. S2). This reconstruction is of a slightly better quality in terms of the continuity of the internal density. Therefore we used this improved map for fitting of the atomic model and further analysis (Fig. S2, Table S1).
Scientific RepoRts | 6:24601 | DOI: 10.1038/srep24601 Additional image processing. As a crosscheck, each data set was also refined either from other initial models: a "featureless donut" with approximate dimensions of the decamer, and low pass-filtered reconstructions from the two other data sets (i.e. the LdcC reconstruction was used as a model for the LdcI a and LdcI-LARA data sets, etc). All refinements converged to the same solutions independently of the starting model. Local resolution of all maps was determined with ResMap 42 .
LdcCI and LdcIC chimeras -negative stain EM and 2D image analysis. 0.4 mg/ml of RavA (in a 20 mM Tris-HCl, 500 mM NaCl, 10 mM MgCl 2 , 1 mM DTT, 5% glycerol, pH 6.8 buffer) was mixed with 0.3 mg/ml of either LdcI, LdcC, LdcCI or LdcIC in the presence of 2 mM ADP and 10 mM MgCl 2 in a buffer containing 20 mM Hepes and 150 mM NaCl at pH 7.4. After 10 minutes incubation at room temperature, 3 μl of mixture were applied to the clear side of the carbon on a carbon-mica interface and negatively stained with 2% uranyl acetate. Images were recorded with a JEOL 1200 EX II microscope at 100 kV at a nominal magnification of 15000 on a CCD camera yielding a pixel size of 4.667 Å. No complexes between RavA and LdcC or LdcIC could be observed, whereas the LdcCI-RavA preparation manifested cage-like particles similar to the previously published LdcI-RavA 19 , but also unbound RavA and LdcCI, which implies that the affinity of RavA to the LdcCI chimera is lower than its affinity to the native LdcI. 1260 particles of 96 × 96 pixels were extracted interactively from several micrographs. 2D centering, multivariate statistical analysis and classification were performed using IMAGIC 36 . Class-averages similar to the cage-like LdcI-RavA complex were used as references for multi-reference alignment followed by multivariate statistical analysis and classification.
Fitting of atomic models into cryoEM maps. A homology model of LdcC was obtained using the atomic coordinates of the LdcI monomer (3N75) as the template in SWISS-MODEL server 43 . The RMSD between the template and the resulting model was 0.26 Å. The LdcC model was then fitted as a rigid body into the LdcC cryoEM map using the fit-in-map module of UCSF Chimera 44 . This rigid fit indicated movements of several parts of the protein. Therefore, the density corresponding to one LdcC monomer was extracted and flexible fitting was performed using IMODFIT 45 at 8 Å resolution. This monomeric model was then docked into the decameric cryoEM map with URO 46 and its graphical version VEDA 47 that use symmetry information for fitting in Fourier space. The Cα RMSD min between the initial model of the LdcC monomer and the final IMODFIT LdcC model is 1.2 Å. In the case of LdcI a , the density corresponding to an individual monomer was extracted and the fit performed similarly to the one described above, with the final model of the decameric particle obtained with URO and VEDA. The Cα RMSD min between the LdcI i monomer and the final IMODFIT model is 1.4 Å. For LdcI-LARA, the density accounting for one LdcI monomer bound to a LARA domain was extracted and further separated into individual densities corresponding to LdcI and to LARA. The fit of LdcI was performed as for LdcC and LdcI a , while the crystal structure of LARA was docked into the monomeric LdcI-LARA map as a rigid body using SITUS. The resulting pseudoatomic models were used to create the final model of the LdcI-LARA decamer with URO and VEDA. The Cα RMSD min between the LdcI i monomer and the final IMODFIT model is 1.4 Å. A brief summary of relevant parameters is provided in Table S1.

Sequence analysis.
Out of a non-exhaustive list of 50 species of Enterobacteriaceae (Table S3), 22 were found to contain genes annotated as ldcI or ldcC and containing the ravA-viaA operon (Table S4). An analysis using MUSCLE 48 with default parameters showed that these genes share more than 65% identity. To verify annotation of these genes, we compared their genetic environment with that of E. coli ldcI and ldcC. Indeed, in E. coli the ldcI gene is in operon with the cadB gene encoding a lysine-cadaverine antiporter, whereas the ldcC gene is present between the accA gene (encoding an acetyl-CoA carboxylase alpha subunit carboxyltransferase) and the yaeR gene (coding for an unknown protein belonging to the family of Glyoxalase/Dioxygenase/Bleomycin resistance proteins). Compared with this genetic environment, the annotation of several ldcI and ldcC genes in enterobacteria was found to be inconsistent (Table S4). For example, several strains contain genes annotated as ldcC in the genetic background of ldcI and vice versa, as in the case of Salmonella enterica and Trabulsiella guamensi. Furthermore, the gene with an "ldcC-like" environment was found to be annotated as cadA in particular strains of Citrobacter freundii, Cronobacter sakazakii, Enterobacter cloacae subsp. Cloaca, Erwinia amylovora, Pantoea agglomerans, Rahnella aquatilis, Shigella dysenteriae, and Yersinia enterocolitica subsp. enterocolitica, whereas in Hafnia alvei, Kluyvera ascorbata, and Serratia marcescens subsp. marcescens, the gene with an "ldcI-like" environment was found to be annotated as ldcC. In addition, as far as the genetic environment is concerned, Plesiomonas appears to have two ldc genes with the organization of the E. coli ldcI (operon cadA-cadB). Consequently, we corrected for gene annotation consistent with the genetic environment and made multiple sequence alignments using version 8.0.1 of the CLC Genomics Workbench software. A phylogenetic tree was generated based on Maximum Likelihood and following the Neighbor-Joining method with the WAG protein substitution model 49 . The reliability of the generated phylogenetic tree was assessed by bootstrap analysis. The presented unrooted phylogenetic tree shows the nodes that are reliable over 95% (Fig. 6A). Remarkably, the multiple sequence alignment and the resulting phylogenetic tree clearly grouped together all sequences annotated as ldcI on the one side, and all sequences annotated as ldcC on the other side. Thus, we conclude that all modifications in gene annotations that we introduced for the sake of consistency with the genetic environment are perfectly corroborated by the multiple sequence alignment and the phylogenetic analysis. Since the regulation of the lysine decarboxylase gene (i.e. inducible or constitutive) cannot be assessed by this analysis, we call the resulting groups "ldcI-like" and "ldcC-like" as referred to the E. coli enzymes, and make a parallel between the membership in a given group and the ability of the protein to form a cage complex with RavA.