The Origin of Mitochondria-Specific Outer Membrane β-Barrels from an Ancestral Bacterial Fragment

Abstract

Outer membrane β-barrels (OMBBs) are toroidal arrays of antiparallel β-strands that span the outer membrane of Gram-negative bacteria and eukaryotic organelles. Although homologous, most families of bacterial OMBBs evolved through the independent amplification of an ancestral ββ-hairpin. In mitochondria, one family (SAM50) has a clear bacterial ancestry; the origin of the other family, consisting of 19-stranded OMBBs found only in mitochondria (MOMBBs), is substantially unclear. In a large-scale comparison of mitochondrial and bacterial OMBBs, we find evidence that the common ancestor of all MOMBBs emerged by the amplification of a double ββ-hairpin of bacterial origin, probably at the time of the Last Eukaryotic Common Ancestor. Thus, MOMBBs are indeed descended from bacterial OMBBs, but their fold formed independently in the proto-mitochondria, possibly in response to the need for a general-purpose polypeptide importer. This occurred by a process of amplification, despite the final fold having a prime number of strands.

Fig. 1.

—Three-dimensional structure and biological function of mitochondrial outer membrane β-barrels (MOMBBs). Six outer membrane β-barrel subfamilies have been described so far in mitochondria: The 16-stranded SAM50/TOB55, which belongs to the OMP85 family of bacterial OMBBs and is involved in the biogenesis and membrane insertion of OMBBs (Kozjak et al. 2003), and the five members of the 19-stranded OMBB family unique to mitochondria (MOMBBs) TOM40, VDAC, MDM10, ATOM, and TAC40 (Pusnik et al. 2011; Bay et al. 2012; Zarsky et al. 2012; Flinner et al. 2013; Schnarwiler et al. 2014). For those whose three-dimensional structure is known (VDAC and TOM40), the experimental structure is shown. For those whose structure was not yet experimentally determined, homology models are shown only for illustrative purposes. For that, the best templates for the reference sequences were identified with HHPred (Zimmermann et al. 2018) over the PDB70 (as of May 2018; supplementary table 1, Supplementary Material online) and the models built with SWISS-MODEL (Biasini et al. 2014) after target-template alignment with PROMALS3D (Pei and Grishin 2014). Long loops are shown in dashed cartoon lines for clarity and named Lx, where x refers to the ranking of the loop in the predicted structure.

Fig. 2.

—Classification and HMM-comparison of bacterial and mitochondrial OMBBs (MOMBBs). (a) Cluster map of a total of 5277 sequences collected for each MOMBB family and bacterial FapF-like and PgaA-like OMBBs. Clustering was performed with CLANS in 2D until equilibrium at a BLASTp P-value of 1.0. Connections represent similarities up to a P-value of 10⁻³ (darker means more similar). Black points represent sequences that do not connect to any cluster at P-values <10⁻⁴. The number of sequences in each cluster is shown within brackets. Clusters composed solely by hypothetical and nonannotated sequences but with significant homology to a known protein family, as detected with HHPred and PSI-BLAST searches, are referred by the name of the homologous family followed by Lx, where x represents the number of the cluster. The taxonomic distribution of the collected sequences is illustrated in supplementary figures 1 and 2, Supplementary Material online. A total of five eukaryotic and eight bacterial clusters were obtained. VDAC, TOM40, and MDM10 form a highly connected supercluster, which connects only marginally with bacterial OMBBs. TAC40 connects to VDAC, but ATOM does not connect to any cluster at a P-value <10⁻³. (b) Sequence homology matrix of OMBB clusters as measured by the hhalign probability of the alignment of their HMM-profiles. Those corresponding to the “Fatty-acid transporters” cluster were not included due to the high level of fragmentation of the sequences composing it. Bacterial and eukaryotic OMBBs define two different regions and all MOMBBs find only marginal matches to bacterial OMBBs, especially BcsC and PgaA, suggesting that all MOMBBs are monophyletic and share only local sequence similarity to OMBBs. (c) Strand composition predicted with Quick2D and repeat units identified with HHrepID for the HMM-profile consensus sequence. All MOMBBs are predicted to have a 19-stranded topology; additionally, VDAC and TOM40 show a repetitive sequence. No bacterial OMBB shows the same topology and repetition pattern. ββ: ββ-hairpin; ββββ: double ββ-hairpin; x: none or not clear.

Fig. 3.

—The repetitive nature of VDAC, TOM40, and BcsC/PgaA OMBBs. (a) Self-comparison dot plot of the PgaA, TOM40, and VDAC HMM-consensus sequence generated by HHrepID. The presence of diagonal lines indicates a repetitive sequence. Repeat families were identified at a P-value threshold of 10⁻¹. For VDAC and TOM40, the full consensus sequence included the N-terminal helix, colored grey in (b). Eight sequence repeats were identified in PgaA, whereas only five were found in VDAC and TOM40. (b) Three-dimensional mapping of the identified repeats on their reference three-dimensional structures (PgaA: 4y25_A; TOM40: 5o8o_A; VDAC: 4c69_X). The sequence repeats in PgaA correspond to single ββ-hairpins, while those of VDAC and TOM40 correspond to two ββ-hairpins. (c) Sequence homology matrix, measured as the hhalign probability of the alignment of the HMM-profiles built for VDAC, TOM40, and BcsC (as mapped over PgaA) double ββ-hairpins. The repeats in VDAC and TOM40 find significant matches only with the last C-terminal double ββ-hairpin of BcsC, with the best match found between this and the fourth repeat of VDAC. (d) Structural similarity matrix, measured as the TMscore from structural alignments with TMalign, of VDAC, TOM40, and PgaA double ββ-hairpins. A TMscore below 0.3 indicates random structural similarity while values above 0.5 suggests that both structures assume the same fold. A TMscore of 1.0 denotes a perfect match between the two structures. The predominantly blue matrix suggests that, despite their low sequence similarity, all double ββ-hairpins are structurally conserved and thus the high level of similarity between the fourth repeat of VDAC and the C-terminal strands of BcsC is not the result of structural constraints.

Fig. 4.

—Sequence and structural conservation of the repeat units from VDAC and TOM40. (a) Structure-based sequence alignment of the repeats identified in the HMM-consensus sequence of VDAC and TOM40 with HHrepID. Repeats were mapped onto their reference structure (TOM40: 5o8o_A; VDAC: 4c69_X), and (b) the structural superposition of the corresponding double ββ-hairpins was carried out with TMalign and manually adjusted using UCSF Chimera (Pettersen et al. 2004) without considering the N-terminal helix and the loop regions. In (a), the position and boundaries of the helices and strands, as of their reference structure, are shaded red and yellow, respectively; asterisks mark strand positions facing the outside of the barrel, with those in bold depicting hydrophobic, aromatic or small residues. Each sequence repeat is composed of two ββ-hairpins, with the exception of the first repeat, where the first strand appears to have been changed to an α-helix. All these double ββ-hairpins have a closely matching structure. In (b), a dashed line represents the transmembrane axis of the reference barrels, highlighting the strand tilt with respect to the membrane.

Fig. 5.

—Sequence homology matrix of double ββ-hairpins from MOMBBs, measured as the hhalign probability of the alignment of their HMM-profiles built with hhmake. For clarity, only the top match of a given query double ββ-hairpin in each of the MOMBB subfamilies is shown. Almost invariably, significant matches for repeat n of one MOMBB has its best match in repeat n′ of another MOMBB, suggesting that all 19-stranded MOMBBs diverged from a fully amplified ancestor and were not amplified individually.