Superimposition of one structure against another was performed using PyMOL v1.7.4. The atomic coordinates and structure information of the proteins of interest were recovered from the Protein Data Bank (PDB) (Berman et al., 2000a,b). The structures were aligned to minimise the RMSD (Root Mean Square Deviation) between the aligned atoms (C-alfas). For multiple sequence alignment, a special mode of T-Coffee (Notredame et al., 2000; Di Tommaso et al., 2011; Magis et al., 2014) was used to incorporate structural information, i.e. Expresso (Armougom et al., 2006), which is an extension of 3D-Coffee where structure based alignment is used as a template for realigning the original sequences, which results in a structure-based multiple sequence alignment (O'sullivan et al., 2004). The alignment was then visualised with ENDscript, which combines both primary sequence and secondary structure alignment (Gouet et al., 2003; Robert and Gouet, 2014).

The AC domains were also identified in autotransporters by submitting passenger sequences to Phyre v2.0 for automated modeling. The dataset included the 1523 well-defined and genuine autotransporters identified by the twin-HMM autotransporter procedure designed by Celik et al. (2012). In parallel, using the refined AC structures here defined from structural alignments of IcsA, EspP, Hbp, Pet, pertactin P69, Hap, and IgA1 (PDB files provided as Supplementary Materials), the presence of the AC in other autotransporters was searched using Phyre in reverse, i.e., BackPhyre (Kelley and Sternberg, 2009; Kelley et al., 2015). To achieve a high degree of reliability with respect to the predicted domain fold and modeling of the protein core at high accuracy (<4 Å RMSD from native, true structure), only structural matches with a high level of confidence (>90%) were considered.

The ACs identified by structural alignment with Phyre/BackPhyre were aligned with T-Coffee in the expresso mode, using PDB files restricted to the AC domains of IcsA (PDB: 3ML3; D₆₀₆–L₇₂₀), EspP (PDB: 3SZE; D₈₆₉–I₉₇₉), Hbp (PDB: 1WXR; N₉₄₈–L₁₀₅₆), Pet (PDB: 4OM9; N₈₆₅–I₉₇₄), P69 (PDB: 1DAB; D₄₄₄–L₅₅₆), Hap (PDB: 3SYJ; D₈₃₀–L₉₆₄), and IgA1 (PDB: 3H09; D₈₆₅–L₉₇₇) to seed the alignment. A BioNJ tree (Gascuel, 1997) based on observed divergences between pairs of sequences was obtained using SplitsTree (Kloepper and Huson, 2008). The most relevant clusters, i.e., monophyletic groups or clades, were identified and selected based on splits showing bootstrap values above 80% over 1,000 pseudo-replicates.

Functional motifs were searched using InterProScan v5.22 (Jones et al., 2014) and interrogating InterPro (IPR) v61.0 database (Finn et al., 2017), which includes CATH-Gene3D v4.1 (Sillitoe et al., 2015; Lam et al., 2016), CDD (Marchler-Bauer et al., 2017), MobiDB v2.3.2014.07 (Potenza et al., 2015), HAMAP (Pedruzzi et al., 2015), PANTHER v11.1 (Mi et al., 2017), Pfam v30.0 (Finn et al., 2016), PIRSF (Wu et al., 2004), PRINTS (Attwood et al., 2003), ProDom v2012.1 (Servant et al., 2002), PROSITE v2017.01 (Sigrist et al., 2013), SFLD (Akiva et al., 2014), SMART v7.0 (Schultz et al., 1998), SUPERFAMILY v1.75 (Wilson et al., 2009), TIGRFAMs v15.0 (Haft et al., 2003). Besides structure modeling using Phyre v2, β-helix folds were predicted with BetaWrap, using rung profile (3–7 rungs) (Bradley et al., 2001), as well as PfScan to identify the types of β-solenoid motif (Kajava and Steven, 2006).

While the autochaperone (AC) was experimentally identified in BrkA, AIDA-I, EspP, Hbp, Pet, and Ssp, neither the structure of the passenger of BrkA, AIDA-I nor Ssp has been experimentally resolved by crystallography, or by any other mean, contrary to EspP (PDB: 3SZE) (Khan et al., 2011), Hbp (PDB: 1WXR) (Otto et al., 2005), and Pet (PDB: 40M9) (Domingo Meza-Aguilar et al., 2014) passengers. Using the structure of the AC of IcsA as reference (AC^IcsA; PDB: 3ML3) (Kuhnel and Diezmann, 2011), identification of the AC in the C-terminal part of EspP, Hbp, and Pet was first attempted. While the resolution of the secondary structure of Pet was low in the C-terminal region of the passenger (PDB: 40M9) and the RMSD was slightly high for Hbp when superimposed to AC^IcsA (RMSD > 4 Å over 54 atoms), the 3D-structure of the C-terminal part of EspP (D₈₆₉–I₉₇₉) superimposed onto residues D₆₀₆–L₇₂₀ of IcsA (Figure 1) with RMSD of 2.51 Å over 43 atoms (Table 1). The structural region here identified as the AC^EspP completely agrees with previous experimental investigations by functional mutagenesis where the AC was identified within the 821–997 region of EspP (Dutta et al., 2003; Velarde and Nataro, 2004; Skillman et al., 2005; Brockmeyer et al., 2009). Similarly, AC^Pet (N₈₆₅–I₉₇₄) was here identified within the 819–992 region of Pet, functionally characterised as an autochaperone (Dutta et al., 2003) and within the 950–1,048 region for the AC^Hbp (N₉₄₈–L₁₀₅₆) (Soprova et al., 2010). Using the AC^IcsA structure (D₆₀₆–L₇₂₀), this domain was structurally aligned within the C-terminal region of all other crystallised autotransporter passengers, namely pertactin P69 (PDB: 1DAB), Hap (PDB: 3SYJ), IgA1 (PDB: 3H09), EstA (PDB: 3KVN), Ag43 (PDB: 4KH3), and VacA (PDB: 2QV3). Except for EstA, Ag43, and VacA, an AC could be identified in all other resolved AT passengers. Following structure alignment, the RMSD varied between 1.51 Å for AC^IcsA vs. AC^P69 and 3.40 Å for AC^IcsA vs. AC^IgA1 (Table 1). Once identified in each passenger, these AC domains were further structurally aligned one with another and appeared to superimpose with RMSD ranging from 0.44 to 2.97 Å for AC^Hbp vs. AC^EspP and for AC^P69 vs. AC^Pet respectively (Table 1). Of note, the AC^Hbp superimposed to the AC of these other ATs with RMSD systematically lower than 2 Å (Table 1). With a size ranging from 109 to 135 amino acid (a.a.) residues, the AC domains displayed a conserved structure forming coils of parallel and anti-parallel β-sheets and a couple of short α-helices (Figure 2A). For IgA1 and Pet, however, the resolution of the structure in this region was not high enough to provide information on the secondary structures. Sequence similarities of the identified AC were further confirmed following multiple sequence alignment incorporating structural information, where regions with β-strands or α-helices aligned one another in the different ACs (Figure 2B).

To determine whether similar AC structures are present in other autotransporters, the well-defined ACs here identified from solved tertiary structures of passengers, namely AC^IcsA (D₆₀₆–L₇₂₀), AC^EspP (D₈₆₉–I₉₇₉), AC^Hbp (N₉₄₈–L₁₀₅₆), AC^P69 (D₄₄₄–L₅₅₆), and AC^Hap (D₈₃₀–L₉₆₄), were used as queries to search for similar structures using BackPhyre. In parallel, the AC domains were also identified by submitting the sequences corresponding to the passenger domain to Phyre for automated modeling. For data mining, the dataset was constituted of the well-defined and genuine ATs identified by Celik et al. (2012). Out of these 1523 sequences of autotransporters, an AC could be identified in 708 of them with confidence levels higher than 90.0%, which corresponded to 429 non-redundant and distinct sequences (Supplementary Material Table 1S). These AC domains have an average size of about 110 a.a. residues (mediane = 112 and mode = 115 a.a.) and were essentially present toward the C-terminal region of the passengers, adjacent to the translocator region at an average distance of about 100 a.a. residues upstream (mediane = 84 and mode = 67 a.a.). With respect to the ATs for which an AC was experimental identified but no structural information was available, an AC could indeed be identified in the C-terminal region of the BrkA and AIDA-I passengers but not Ssp. Regarding ATs for which the passenger structure was resolved but an AC domain could not be identified in the first instance, an AC was finally identified in the C-terminal part of the Ag43 passenger (position 525-638), i.e., a region that has not been crystallised (Heras et al., 2014), but this was not the case either for EstA or for VacA. Following multiple sequence alignment incorporating structural information, phylogenetic analysis revealed that the AC domains distribute into deeply rooted branches, from which 20 main clusters emerged (named according to cyrillic alphabet with phonetics in brackets), namely A [a], Б [b], B [v], Г [g], Д [d], E [je], Ж [ʒ], З [z], И [i], Й [j], К [k], Л [l], M [m], H [n], O [o], П [p], P [r], C [s], T [t], and У [u] (Figure 3), where clusters A and Б form the largest groups hosting some autotransporter members of Iba (Inducible Bartonella autotransporter) (Eicher and Dehio, 2012) and AutA (Auto-transporter A) groups respectively (Ait-Tahar et al., 2000). Clusters Г, Д, З, И, Й, K, Л, O, C, and У harbor ACs from autotransporters that have not been characterised yet. The resolved AC domains from IcsA, EspP, Pet, or pertactin P69 were not found in clusters but in deeply rooted branches, whereas the ACs from Hbp, IgA1, and Hap were part of clusters M, P, and T respectively (Figure 3). Considering other characterised autotransporters, clusters B, E, Ж, H, and П harbor ACs from EhaC, Ag43, AIDA-I, EhaD, and YcgV autotransporter members (Vo et al., 2017), whereas the ACs from BrkA and Pet were found in deeply rooted branches.

Looking for a correlation with the presence of the AC, functional genomic analysis of the ATs was performed (Supplementary Material Table 1S). First, an ESPR (IPR024973) could be predicted in the SP region of 134 ATs; in 84% of these an AC was also identified. Based on the functional motifs identified in the passengers, the ATs could be further classified into 6 main and distinct functional categories, i.e. the (i) protease autotransporters (PATs), (ii) self-associating autotransporters (SAATs), (iii) phosphatase/hydrolase autotransporters (PHATs), (iv) lipase/esterase autotransporters (LEATs), (v) vacuolating autotransporters (VATs), and (vi) autotransporters of unknown function. The PATs, like Pet, Ssp, or EspP, could belong to different peptidase families, essentially the (i) serine peptidases S1 (IPR001314), S6 (IPR000710) or S8/S53 (IPR000209), (ii) cysteine peptidase C1 (IPR025660), or (iv) metallopeptidases M10 (IPR011049) or M28 (IPR007484). SAATs systematically feature an adhesion domain of ATs (IPR030930) like AIDA or Ag43. PHATs mainly belong to either the (i) phosphatidic phosphatase (IPR000326), (ii) tyrosine phosphatase (IPR029021), or (iii) glycoside hydrolase (IPR002772) family. LEATs, like EstA, possess GSDL lipase/esterase (IPR001087) and/or SGNH esterase (IPR013830) domains. Like VacA, the VATs systemically exhibit a vacuolating cytotoxin domain (IPR004311). The most striking observation was that no AC could be identified in any of the LEATs or VATs, whereas it was identified in the large majority of the SAATs (83%). Among the PATs and PHATs, an AC was present or absent from some functional subcategories, e.g. the M28 metalloproteases or the tyrosine phosphatases. Regarding the ATs for which no function could be inferred, homology to pectin lyase fold (IPR011050), pertactin P69 (IPR004899; IPR003991) and/or P22 tailspike protein (IPR012332) was predicted in the passengers. These regions correspond to β-helix topologies as encountered in most ATs (Jenkins et al., 1998), including the SAATs. Interestingly, all passengers belonging to the LEATs, as well as to the M28 metalloprotease or the tyrosine phosphatase ATs, exhibit α-helical folds for which no AC could be identified. Of note, the passenger of Ssp was predicted to essentially display only α-helices, like EstA, for which no AC structure could be identified either (Figure 4). Following structural modeling of the passengers, it appeared that the AC is systematically associated with passenger exhibiting β-helix folds, e.g. as observed in the passengers with a resolved structure, namely Ag43, EspP, Hap, Hbp, IcsA, IgA1, pertactin P69, and Pet, or in predicted β-helical passengers where an AC was experimentally identified, namely BrkA and AIDA-I. Despite predominantly exhibiting a single-stranded right-handed parallel β-helix in the passenger, no AC could be identified in VacA or any VATs, indicating that such as a fold is not always associated with an AC domain. Considering the oval (O), triangular (T) or L-shaped coil cross-sections, 16 different β-solenoids are currently recognised (Kajava and Steven, 2006). As expected, no O1 or O2 repeat could be identified in any of the AT passengers as these coils occur upon oligomerisation, as observed in trimeric ATs belonging to the T5cSS. Besides L1, L3, T4, T5, and T6-type β-solenoids previously reported as specific to AT passengers, some coils originally considered as specific to the T5bSS, i.e., the TPS (two-partner system), were also identified and encountered in association with an AC, namely the L2, T1, and T8 coils. Rather than being associated to some specific functions, the AC thus appears exclusively associated with β-helical passengers.

RR and MP are permanent employees of GSK, which provided support in the form of salaries. The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.