The phylogeny of Acetobacteraceae: photosynthetic traits and deranged respiratory enzymes

The objective of this paper was to produce a comprehensive analysis of the Acetobacteraceae family encompassing both the taxonomic breath and the functional diversity in the energy metabolism of the family. To accomplish this objective, we initially relied on the DSMZ repository [cf. reference (4)], which listed 61 child genera of Acetobacteraceae (https://lpsn.dsmz.de/family/acetobacteraceae, first accessed on 26 May 2022). We then examined the GDTB database (28) (https://gtdb.ecogenomic.org/, first accessed on 5 September 2022), which included a total of 77 genera. Several of these genera did not correspond to those present in the DSMZ repository because they were based upon metagenome-assembled genomes (MAGs) or non-validated taxa (28), for which little genomic information was retrievable from publicly available NCBI resources. Useful genomic information was available for a set of taxa that incompletely overlapped those present in either the GTDB database or the DMSZ repository. We generally used strong taxon density, namely, wide sets of taxa, to evaluate the phylogenetic space of acidophilic taxa to which Acidocella belongs. Except for members of Acidocella and sister genera, we endeavored to limit taxonomic redundancy in terms of either protein similarity or taxonomic relatedness in the phylogenetic analysis of marker proteins. Along this effort, we analyzed many Antarctic MAGs (21) and novel MAGs contributed by this study. All the taxa studied and their genomic characteristics are listed in Extended Datasheet 1 in the repository https://osf.io/y6gxt/, which is associated with this manuscript.

We report here the novel genome of a strain derived from the original glycerol storage in which we maintained previously reported Acidocella MX-AZ02 (17); we re-isolated the strain from glycerol and selected one colony, which turned out to be a different strain than the original and was thus named MX-AZ03. Its DNA was extracted with the DNA Isolation Kit for Cells and Tissue (Roche), and genomic sequences were obtained using the PacBio RSII single-molecule real-time protocol RS_HGAP_Assembly.3 V2.2.0 from the SMRT portal version 2.3.0.140893 from Pacific Biosciences (Menlo Park, CA, USA). We used the following default parameters: minimum subread length = 500; minimum polymerase read quality = 0.80; and minimum polymerase read length = 100. The program “Flye assembler” version 2.8.3-b1695 (29) was then used to obtain a closed replicon. The genomic characteristics of MX-AZ03 obtained with CheckM (30) are presented in Table S1 and in Extended Datasheet 1 of the repository https://osf.io/y6gxt/.

Environmental samples for metagenomic sequencing were obtained from the water column at two sampling points along Rio Agrio Inferior (Neuquén province, Patagonia, Argentina), designated RAI-PG (−37,82884 S; −70,96715 W; 1,516 m.a.s.l.; average temperature 16.7ºC; average pH 3.2) and RAI-PT (−37,81318 S; −70,85157 W; 1,348 m.a.s.l.; average temperature 15.8ºC; average pH 4.4). Cells were separated from larger debris and sediment particles by serial filtration with 0.45- and 0.22-µm filters and preserved at −80°C. Prior to DNA extraction, the frozen filters were sheared, and the resulting pieces were placed in Eppendorf tubes, to which 1 mL of Tris-EDTA buffer was added. After vortexing, the suspension was processed for DNA extraction using phenol-chloroform-isoamyl alcohol. The DNA obtained was purified using the Genomic DNA Clean & Concentrator Kit (Zymo), quantified by fluorescence using the Quant-iT PicoGreen dsDNA Kit (Thermo Fisher), and qualitatively verified by spectrophotometry. Nearly 20 million paired-end (PE) reads were generated through Illumina sequencing systems PE 150 pb on the Hiseq platform through the CD Genomics sequencing service (https://www.cd-genomics.com, NY, USA). A sequence quality check was performed using fastqc v0.11.5 (Babraham Bioinformatics), while adapter removal and trimming were carried out with fastp v0.23.1. Shotgun reads with a >Q35 quality score were retained and de novo assembled using SPAdes v3.15.2 built in the pipeline SqueezeMeta v1.5.1 (31) with the parameters: -m sequential -t 90 -a spades -assembly_options ---meta --only-assembler y -t 90¨. Binning was performed using MaxBin2 v2.2.7 (32), Metabat2 (33), and CONCOCT v1.1.0 (25). Genomic characteristics were evaluated using CheckM (30), while preliminary taxonomic assignments were undertaken with GTDB-TK v2.1.0 (28). Additional methodological details are presented in the Extended Methods document posted in the repository https://osf.io/y6gxt/.

The analysis for gene annotation and function was initially performed on metagenomic data using conventional methods (19), exploiting multiple databases: GenBank, eggNOG KEGG, and Pfam (June 2022 versions). Details for the functional annotation and specific systems used are listed in Extended Datasheet 2 posted in the repository https://osf.io/y6gxt/, which is associated with the manuscript. Subsequently, genes encoding bioenergetic proteins were analyzed by iterative PSI-BLAST searches to evaluate the completeness of the conserved signatures defining function and the genomic clusters characteristic of complete operons, such as those of COX, essentially as described previously (34, 35).

Protein sequences retrieved from BLAST searches and various MAGs used in this study were first aligned using ClustalW embedded into MEGA programs and then manually refined using standard procedures that have been reported earlier (34, 35). N- and C-terminal regions were rarely trimmed to preserve information; indeed, bioenergetic proteins such as NuoD and the Rieske iron-sulfur protein (ISP) subunit of the bc ₁ complex have very conserved C-terminal regions. Phylogenetic trees were initially reconstructed using the neighbor joining (NJ) approach using the MEGA5 program (usually with the JTT model and 500 bootstraps) to orient the phylogenetic analysis. Subsequently, robust phylogenies were obtained in most cases with maximum likelihood (ML) inference (19, 34 – 36) using the program IQ-Tree as described recently (34) and methodological details that are presented in the figure legend. In some cases, Bayesian inference was additionally used (35). Following earlier works (37), we utilized both concatenated alignments of ribosomal proteins and single markers with strong phylogenetic signals, including the 16S rRNA gene (Fig. S1A) and the beta subunit of the F1 part of rotor-stator ATP synthase, called AtpD here after its gene atpD (as in Fig. 2A). We chose this protein marker because of its ubiquity and strong conservation across Acetobacteraceae (over 50%, hence very similar to the degree of conservation of 16S rRNA sequences). Moreover, the hydrophilic character of AtpD attenuates possible artifacts deriving from the substantially different GC content among diverse groups of Acetobacteraceae (4), since such artifacts are more pronounced for hydrophobic residues. Indeed, phylogenetic trees obtained with AtpD show stronger and better resolved internal nodes than those obtained with the very hydrophobic NuoL subunit of Complex I, despite the longer sequence of the latter (34). Trees similar to those of AtpD were reconstructed with the NuoD subunit of Complex I, which is hardly affected by artifacts derived from compositional and other biases (34). NuoD is significantly less conserved than AtpD across Acetobacteraceae, and its increased sequence variation provides additional characters to resolve the placement of “difficult” taxa such as Granulibacter, which has divergent proteins forming long branches. In principle, concatenated alignments of ribosomal proteins may provide better phylogenetic patterns than alignments of single marker proteins. However, we found that the taxonomic frequency of individual ribosomal proteins is not homogeneous across Acetobacteraceae (see Extended Datasheet 2 posted in the repository https://osf.io/y6gxt/ for details). We assembled a set of 15 ribosomal proteins that were present in the majority of these MAGs and in representative genomes of the rest of the family (using a threshold of three missing sequences for each protein) (Extended Datasheet 2 in the repository https://osf.io/y6gxt/ associated with this paper) to generate phylogenetic trees such as that shown in Fig. 1A. We noted that the branching pattern of such phylogenetic trees showed instability in their central region, despite the apparent strong support for various nodes. This was especially relevant for defining the sister clade to the acetous clade (Fig. 1B). Conversely, the tips of the branches consistently contained the same taxa in trees reconstructed with either concatenated ribosomal proteins or single marker proteins, thereby providing robust information for the taxonomic classification of the MAGs reported here. The same principle was applied for re-classifying previously reported MAGs that were either considered Rhodospirillales bacteria or misclassified, as in the case of Acidocella sp. C78, which firmly clusters with Acidiphilium taxa (Fig. 1A and 2A). Additional information was used to define the major clades of Antarctic MAGs (21): phylogenetic analyses with additional bioenergetic proteins such as cytochrome b (Fig. S1C), consideration of the length of marker proteins, and presence of photosynthetic traits (Table 2). We generally used proteins or genes from Skermanella to provide a valid root to our trees, contrary to previous questionable choices (1, 4, 5); Skermanella is a photosynthetic genus of the family Azospirillaceae, from which Acetobacteraceae branched off (3, 37).

In the course of this study, we realized that some MAGs classified as Rhodospirillales and Proteobacteria could also be part of the Acetobacteraceae family since their proteins clustered within the clade of orthologs from Acidocella or other genera of acidophilic Acetobacteraceae. Because such MAGs clearly required re-classification, we needed to integrate genomic information available from different resources (listed in Extended Datasheet 1 posted in the online repository https://osf.io/y6gxt/ associated with the manuscript) with our own phylogenetic results to comprehensively analyze Acetobacteraceae taxa. This approach of integrated analysis does not conform to standard bioinformatic studies that usually start with a set of taxa retrieved from a single resource. If we had applied standardized selection criteria, we would not have obtained a truly comprehensive analysis of Acetobacteraceae, as presented in Table S1. This table focuses on the genus Acidocella, to which we contribute additional taxa here. Moreover, our integrated approach enabled flexibility for functional genomic analysis; we progressively expanded or adjusted the taxonomic sampling of Acetobacteracae and outgroup taxa depending upon the presence of trait-defining proteins.

To frame the evolutionary steps that originated the diverse bioenergetic traits of Acetobacteraceae (Fig. 2B and C), we undertook a comprehensive analysis of the phylogeny of the entire family using concatenated core proteins and different single-copy markers, comparing the resulting trees with those recently reported in the literature. The family Acetobacteraceae, recently proposed to form the single family of the order Acetobacterales (28), is dominated by the acetous group (2) in both the overall number of taxa and the proportion of validated genera [ca. 50% of all genera in reference (4)]. Intriguingly, all acetous taxa encompass a compact phylogenetic space that begins with either endophytic Endobacter (41) or pathogenic Granulibacter and ends with late-branching symbionts of bees and ants, Bombella and Formicincola [Fig. 1A and 2A; Fig. S1A and D; cf. references (1, 2, 4, 5, 36, 41)]. This compact phylogenetic pattern implies that the fundamental contours of the acetous clade can be defined by a few representatives of the major subclades with acetous character, for example, Acetobacter and Gluconobacter, as shown in Fig. 1A. Granulibacter appears to have an odd position in phylogenetic trees obtained with the single marker AtpD (Fig. 2A), chiefly due to the highly divergent nature of its homolog protein.

The second largest group of Acetobacteraceae revolves around Roseomonas, for which there is an increasing number of taxa and closely related genera (1) that seldom have an acidophilic character, or mildly so (Table 1). Roseomonas and related taxa regularly form a compact clade that branches immediately after the clade of Elioraea (Fig. 1A, 2A, and 3A), which we confirmed to be the basal genus of Acetobacteraceae in phylogenetic trees reconstructed with concatenated ribosomal proteins [Fig. 1A, cf. references (3 – 5, 34)] and single marker proteins (Fig. 2A and 3A). Our results do not lend support to the proposal of a separate family for Elioraea (42). In expanded trees reconstructed with 16S rRNA sequences, some members of the Roseomonas clade, such as Roseococcus, form separate branches [Fig. S1A, cf. reference (41)]. This situation is most likely derived from the limited resolution of the very conserved sequences of rRNA. Indeed, phylogenetic trees reconstructed with single marker proteins (Fig. 2A and 3A, as well as Fig. S1C) or concatenated ribosomal proteins (Fig. 1A; Fig. S1D) invariably show a compact Roseomonas clade.

The central part of phylogenetic trees of Acetobacteraceae includes non-acetous acidophilic taxa of the family. Some of such taxa often cluster in a sister clade to the acetous taxa, while Acidocella and Acidiphilium form a basal branch that is closer to the Roseomonas clade than to the acetous clade [Fig. 1A, cf. references (1 – 5, 36)]. However, this common branching pattern depends upon the phylogenetic marker and taxonomic sampling used to reconstruct the trees. Even minimal differences in the taxonomic sampling of concatenated ribosomal proteins seem to alter the relative position of the acidophilic clade, as summarized in Fig. 1B. This was not the case for phylogenetic trees reconstructed with equivalent or larger taxonomic samplings of the AtpD or NuoD protein (Fig. 2A and 3A) and also with 16S rRNA sequences [Fig. S1A, cf. reference (41)]. These trees consistently showed the acetous clade in a sister position to the clade of acidophilic taxa, including Acidocella, Acidiphilium, and also Acidisoma, while Rhodovastum and other predominantly photosynthetic taxa generally formed a basal clade (Fig. 2A and 3A). Trees reconstructed with cytochrome b sequences clearly confirmed the sisterhood of the acidophilic and acetous clades (Fig. S1C), which would be consistent with the shared bioenergetic trait of the non-functional bc ₁ complex in these sister clades (Fig. 2C). Presumably, bc ₁ complex genes are retained in Acetobacteraceae predominantly using ubiquinol oxidases (9, 12 – 14) because the proteins coded by such genes fulfill a structural role in stabilizing Complex I (44), which is fundamental for their physiology (13 – 15). We subsequently used the phylogenetic tree of the NuoD subunit of Complex I as a framework to evaluate the evolutionary steps determining the respiratory chain changes in various Acetobacteraceae (Fig. 3A).

To sum up, two firm considerations emerged from our comprehensive analysis of the Acetobacteraceae phylogeny. The first is that the Roseomonas clade likely constitutes a separate subdivision of Acetobacteraceae, which we propose to designate the family Roseomonadaceae. The second is that Acidisoma taxa cluster with Acidocella and Acidiphillium taxa, together forming a newly defined clade of acidophilic members of Acetobacteraceae. We will preliminarily call this clade “acidophilic” to distinguish them from the generally less acidophilic but predominantly photosynthetic taxa such as Rhodovastum.

We have already mentioned that diverse members of Acetobacteraceae possess photosynthetic traits, for example, Rhodovastum. However, such traits have been scantly evaluated in the literature. For instance, a recent survey of Roseomonas taxa (1) did not consider photosynthetic traits, which we estimate to be present in about 40% of current taxa forming the Roseomonas clade (cf. Table 1). Here, we evaluated in granular detail the distribution of photosynthetic traits among Acetobacteraceae. Photosynthesis has been previously associated with Acidiphilium (45), Rhodopila, Rhodovastum (22, 46), and Acidisphaera (40, 47 – 49). However, our analysis indicates that only one-half of the genomes currently listed under Acidisphaera (six in GTDB: https://gtdb.ecogenomic.org/searches?s=al&q=g_Acidisphaera, accessed on 22 May 2023) possess photosynthetic traits. Indeed, the genus Acidisphaera is polyphyletic: non-photosynthetic Acidisphaera sp. S103 clusters with Rhodopila (Fig. 1A and 2A), while photosynthetic Acidisphaera sp. L21 (Table S2) clusters with various Antarctic MAGs (Fig. 1A, 2A, and 3A). The latter finding is consistent with the separate classification of this taxon under the LMUY01 genus in GTDB taxonomy (28) and illustrates the widespread distribution of photosynthetic traits among Acetobacteraceae MAGs from Antarctic environments (Table 2). Originally, the majority of Antarctic MAGs were reported to cluster with Rhodospirillales sp. 70–18 (21), a non-photosynthetic genus previously considered to be part of Acetobacterales (28, 50). Our analysis indicated that only a couple of Antarctic MAGs actually cluster with Rhodospirillales sp. 70–18 (Fig. 3A, cf. Table 2), while about one-half of the other MAGs cluster with established photosynthetic Acetobacteraceae such as Rhodopila (clades a, b, and LMUY01, Table 2). Conversely, two non-photosynthetic Antarctic MAGs (clade d, Table 2) cluster within the group of Lichenicoccus (36), which includes the recently reported (51) Acetobacteraceae KSS8 and KSS12 (Fig. 1A and 2A). Antarctic MAGs did not cluster within the acidophilic clade, which we expanded by adding novel taxa with photosynthetic traits: Proteobacteria bacterium isolate G5_19, re-classified in the Acidisoma genus, three Acidocella from acidic environments of Patagonia possessing photosynthetic traits, and various Rhodospirillales MAGs that we re-classified as members of either the Acidiphilium genus or the Acidocella genus (Table S1). Overall, our increased taxonomic sampling indicates that photosynthetic traits pervade the phylogenetic space encompassing Rhodovastum to Lichenicoccus, despite multiple instances of gene loss in closely related taxa. Moreover, the earliest branches of the Acetobacteraceae phylogeny, Elioraea and Roseomonadaceae, have a pervasive presence of photosynthetic traits too (Table 1; Fig. 3B). Therefore, we surmise that the common ancestor of Acetobacteraceae probably possessed photosynthetic traits. This is a novel concept in the phylogeny of Acetobacteraceae [cf. references (2) and (4, 36)], which deserved further detailed analysis.

We undertook the phylogenetic analysis of multiple proteins defining photosynthetic traits, starting with the largest subunit of the photosynthetic reaction center (RC), PufM (Fig. 4A and B; Fig. S3A). BLAST searches against the whole nr database have shown that PufM proteins of most Acetobacteraceae cluster in a clade that appears to be early branching (Fig. 4A). However, the PufM proteins of Rhodovastum and Rhodopila cluster in a separate branch containing various purple bacteria of the alphaproteobactera class intermixed with gammaprotobacteria such as Halorhodospira (Fig. 4A). All these bacteria have the physiology of anaerobic photosynthesis; namely, they use phototrophic autotrophy when light is available and oxygen is scarce, but revert to heterotrophic aerobic metabolism in the dark (40). Among Acetobacteraceae, only Rhodovastum (22) and Rhodopila (47, 52) have the same physiology, while Eliorea thermophila has been reported to grow phototrophically using substrates that typically sustain autotrophic growth with anaerobic photosynthesis (42, 53). This versatile phototrophic phenotype has been linked to the presence of Form I ribulose 1,5-bisphosphate carboxylase-oxygenase (Rubisco), an enzyme usually associated with photo-autotrophy (40, 43, 49) that is present only in E. thermophila among Elioraea species (53). Indeed, Rubisco has been used as a proxy to differentiate photo-autotrophic from photo-heterotrophic (40) or aerobic anoxygenic phototrophs (AAPs) (26, 43). AAP cannot grow phototrophically under anaerobic conditions because they lack Rubisco (43).

Here, we used the absence of Rubisco to define bona fide AAPs as in previous works (40, 43) but realized that this criterion would exclude bacteria previously recognized as AAPs, even if their genomes possess both subunits of Form I Rubisco, for example, Acidisphaera (43) (Fig. 4C). Remarkably, all possible combinations of photosynthetic traits with Rubisco are present in related acidophilic Acetobacteraceae (Fig. 4C; Fig. S4). This is consistent with the emerging picture of Rhodospirillales AAPs that have Rubisco and the whole Calvin cycle for carbon fixation (54). There also are taxa without photosynthetic traits that contain a catalytically active Form I Rubisco, as in the case of Acidiphilium iwatense (Fig. 4C; Fig. S4). Rubisco and its associated Calvin cycle can sustain other forms of autotrophic physiology, for example, methylotrophy (55). This explains the presence of two closely related Rubisco isoforms in Acidimonas methanolica (Fig. S4), a member of the acetous clade that uniquely has methylotrophic physiology (55). By analogy, it is possible that Acidiphilium iwatense and other Acetobacteraceae retain Rubisco to enable growth with chemolithotrophic pathways, in particular iron oxidation (ferrotrophy), which is widespread among acidophilic Acetobacteraceae (Table S2). Notably, Acidiphilium iwatense is one of the few acidophilic taxa that does not have a complete operon for the most common type of COX in alphaproteobacteria (20) (Fig. S4). This situation reflects a peculiar correlation between photosynthetic traits and COX (Fig. 3), which will be discussed next.

Early in our study, we realized that the distribution of photosynthetic traits among Acetobacteraceae essentially matches that of cytochrome oxidase (Fig. 3 and 4; Fig. S2 to S5), specifically of A1-type COX operon subtype b (20). Genomes that do not contain this operon have sometimes A2-type COX operon subtype a-I (Fig. S2B; Table S2), which is normally associated with nitrogen metabolism (20). Intriguingly, the distribution of COX operon subtype b stops at the phylogenetic junction of the acetous clade (Fig. 3C; Fig. S2B and C). However, there are acetous taxa that retain genes for the metabolism of photosynthetic carotenoids (Fig. S4), which is generally associated with photo-autotrophy (40). Indeed, the various genes responsible for the biosynthesis of carotenoids form part of the photosynthetic gene cluster, including the RC subunits and various enzymes for the biosynthesis of porphyrins and chlorophyll (26). The overall panorama of trait-defining proteins associated with photosynthesis in Acetobacteraceae (Fig. S4) suggests a pattern of differential loss with retention of vestigial characters, as in the case of Rubisco (Fig. 4C). This pattern is coupled to the loss of COX in several phylogenetically separate genera (Fig. 3; Fig. S4), a situation that differs from that present in Rhodobacterales (26). In the latter lineage, the loss of photosynthesis is rarely associated with that of COX but is generally linked to that of other proteins of the photosynthetic gene cluster, including those involved in carotenoid biosynthesis (26). The simplest explanation for such differences is that the distribution of photosynthetic and COX genes does not derive from the same evolutionary pattern in Acetobacteraceae and Rhodobacterales.

Lateral gene transfer (LGT) has been shown to be the dominant factor in the distribution of photosynthetic traits among Rhodobacterales (26, 40). Indeed, the photosynthetic gene cluster is located in plasmids of several Rhodobacterales (26) but not in Acetobacteraceae (45). The most evident LGT cases of photosynthetic traits in Acetobacteraceae regard the Rubisco present in Elioraea thermophila and two environmental MAGs, which belong to Form IA typical of gamma- and betaproteobacteria (27). In the case of PufM, there are likely LGT instances from Acetobacteraceae to other taxa, either somehow related, such as Skermanella, or clearly unrelated, such as deltaproteobacteria MAGs (Fig. S3A). This situation and the early branching position of the clade including most Acetobacteraceae PufM (Fig. 4A) suggest that photosynthesis entered early in the ancestral lineage of the family, perhaps concomitantly with the adaptation to stable levels of oxygen in the environment. The latter possibility emerged from combining our observations with established evidence regarding the evolution of photosynthesis in Proteobacteria (26, 40, 43, 49). The first piece of evidence is that anaerobic photo-autotrophy is ancestral to aerobic phototrophy (40); namely, photo-autotroph Rhodovastum is early divergent vs AAP such as Acidiphilium. Phylogenetic trees of the whole family sustain this (Fig. 2A and 3A, as well as Fig. S1A; Fig. S3B and C). The second piece of evidence is that RCs containing the tetraheme cytochrome c subunit, PufC, are more ancient than those having the PufX subunit instead (26, 40, 56). Connected to the latter, anaerobic photo-autototrophs have PufC proteins displaying a conserved Cys toward the N-terminus, which is used for post-translational attachment of a lipid anchor. Conversely, AAP taxa generally have PufC proteins without such a residue and remain attached to the membrane via an N-terminal transmembrane helix (56). This situation is present in Rhodopila (56) and the great majority of Acetobacteraceae, with the exception of a few members of the Roseomonas clade (Fig. S3B). Finally, the presence of Form I Rubisco in taxa that have been phenotypically characterized as typical AAPs such as Acidisphaera (43, 47, 49, 52) suggests that Acetobacteraceae evolved around the time of the transition from anaerobic to aerobic phototrophy, a transition that left extant taxa with a hybrid physiology (Fig. 4B). Clearly, this must have occurred as an adaptation to increasing and stable levels of oxygen in the environment (43).

COX evolution equally reflects bacterial adaptation to increasing levels of oxygen in the environment (18 – 20). Moreover, COX activity is required to maintain RC components that are sufficiently oxidized during photosynthetic electron transport in AAPs (43, 57, 58). This requirement may rationalize the correlation in the distribution of COX and photosynthetic traits (Fig. 3B and C; Table S4). The molecular signatures determining the different redox properties of RC components in AAPs remain elusive (43). However, the recently reported structure of the Rhodopila RC (56) may provide new clues. In particular, the C-terminal part of PufM extends at the periplasmic side of the RC complex, covering the PufC part that binds heme 3 (Fig. S5A). This cytochrome c heme directly transfers electrons to the photo-oxidized special pair of bacteriochlorophylls (56). The C-terminal part of PufM is missing in photo-autothrops such as R. sphaeroides and R. rubrum, the RC of which lacks the PufC subunit (Fig. S5B). Conversely, the PufM sequences of photosynthetic Acetobacteraceae and many other AAPs have a C-terminus with a length equivalent to that of Rhodopila. These sequences generally maintain the chemical character of PufC-interacting residues and may thus influence the electron transport properties between PufC and the RC (Fig. S5).

In sum, traces of a photosynthetic past are present in genomes of extant non-photosynthetic Acetobacteraceae, reflecting a pattern of vertical inheritance that is peppered with differential loss and transition states in overall physiology.

If the ancestor of Acetobacteraceae was originally photosynthetic, it follows that the loss of the RC function of ubiquinone (Q, Fig. 2B) reduction would be compensated by other Q reductases to maintain a redox balance in the respiratory chain. To verify such a possibility, we investigated the presence of a variety of Q reductases in the genomes of Acidocella and other acidophilic taxa (Fig. 5A). The various dehydrogenases that are constitutively present in basically all genomes are electron transfer flavoprotein dehydrogenase (ETF-Q) and NADH-quinone reductase (Complex I, Fig. 5A), in agreement with previous surveys (14, 59, 60). The SoxBY markers, representing alternative Q reduction pathways associated with sulfur oxidation (45), displayed instead a patchy distribution, not dissimilar to that of type 1 hydrogenase and carbon monoxide dehydrogenase (Codh) (Fig. 5A). Indeed, aerobic photosynthesis helps a variety of bacteria to survive in deserts and other extreme, oligotrophic environments (54, 61). Ultimately, the distribution pattern and the number of pyrroloquinoline (PQQ)-dependent Q reductases emerged as one of the most widespread traits compensating for the loss of RCs (Fig. 5A). PQQ-dependent dehydrogenases are the dominant metabolic enzymes of acetous taxa (2, 4, 14, 15, 59, 60); except for Rhodopila (52), these enzymes have a limited presence in other Acetobacteraceae (Fig. 5A), despite the widespread distribution of the operons for the biosynthesis of the PQQ cofactor (62). Intriguingly, Acidocella taxa that do not have photosynthetic traits possess two or more PQQ-dependent Q reductases (Fig. 5A), sustaining the possibility that these enzymes may compensate for the loss of Q reduction by RC.

During our search for additional Q reductases, we discovered a putative redox enzyme that is widespread among Acidocella taxa. We named this enzyme nSor, standing for novel Sulfoxide-Q oxidoreductase (Fig. 5B). It contains two different proteins coded by adjacent genes resembling the yedYZ of Escherichia coli, recently re-named MsrPQ for it constitutes a bacterial methionine sulfoxide reductase (66, 67). This enzyme belongs to the group of sulfite oxidases (68) and helps in quenching oxidative damage to proteins by reducing methionine sulfoxides (66, 67). Although MsrPQ normally oxidizes ubiquinol to reduce methionine sulfoxides in the periplasm (68), nSor likely functions as a Q reductase because its membrane di-heme cytochrome b is structurally similar (Fig. 5B) to the Q-reducing cytochrome b subunit of formate dehydrogenase (69). This hypothesis requires further experimental tests.

We next assessed the phyletic patterns and phylogeny of bioenergetic traits for the oxidation of ubiquinol. Soluble cytochrome c ₂ is reduced by the bc ₁ complex, which is the major partner of RCs in re-oxidizing ubiquinol [Fig. 2B, cf. references (26, 40, 49)]. However, Acetobacteraceae genomes contain other enzymes for ubiquinol oxidation that utilize oxygen as an electron acceptor, notably the bo ₃ ubiquinol oxidase (13, 14, 45, 59, 60) (Table S2). The most recent study on the origin of bo ₃ ubiquinol oxidase among Acetobacteraceae concluded that the bo ₃ oxidase of Acidiphilium was acquired from Acidithiobacillus spp. sharing the same extremely acidic habitats (45). We agree with this conclusion since our phylogenetic trees of the catalytic subunits CyoB (Fig. 6) and CyoA (Fig. S6) invariably show a close vicinity of Acidiphilium bo ₃ oxidase with homologs from Acidithiobacillus and Acidiferrobacter spp.

The clade of Acidiphilium CyoB, named clade 1 here, also includes proteins from Acidocella, Acidisoma, Lichenicoccus, and Lichenicola (Fig. 6; Fig. S7). Remarkably, the genomes of Lichenicoccus and Lichenicola contain two additional CyoB proteins belonging to separate operons segregating in different clades, labeled 2 and 3 here (Fig. 6). Previously, it was reported that multiple operons for the bo ₃ ubiquinol oxidase were specifically present in the Asaia genus among Acetobacteraceae (59). However, we found that some Acidocella, including our newly reported MX-AZ03 (Fig. S7), one Acidisoma, Acidisphaera rubrifaciens, and four acetous genera also have a second operon clustering in clade 2 with CyoB-2 from Asaia (59) (Fig. 6; Fig. S7). Clade 2 additionally includes the bo ₃ oxidase from some Antarctic MAGs and is separate from clade 3, which encompasses the second bo ₃ oxidase of acetous taxa, as well as the third bo ₃ operon of Lichenicoccus and Lichenicola (Fig. 6B; Figs. S6 and S7). Clade 3 is part of a very large cluster of bo ₃ oxidases from diverse proteobacteria (Fig. S6A), consistent with earlier studies (9, 13, 59). Distance trees obtained from BLAST searches of the nr database—accessed on 27 October 2022—are basically consistent with this picture but clearly show the early branching position of clade 1 proteins from acidophilic taxa and the Lichenicoccus group clustering together with those of acidithiobacilli and Acidiferrobacter spp. (Fig. S6A). This finding suggests that Acetobacteraceae initially acquired their bo ₃ oxidases from either Acidithiobacillus or Acidiferrobacter species and subsequently transmitted them to other lineages. The process was associated with operon duplication in the genome of the ancestor of some Acidocella strains, leading to the separation in different bo ₃ oxidases, one of which clusters in clade 2 (Fig. 6; Fig. S7). Instead, the second isoform clusters within a very large clade comprising the bo ₃ oxidases of other proteobacteria (Fig. S6A). Our results additionally suggest a late acquisition of bo ₃ oxidase by early branching Acetobacteraceae such as Rhodopila (Fig. 6). The question that then emerges is this: what has driven the diversification of bo ₃ oxidase from the original operon acquired from iron oxidizers?

To find the answer to the above question, we explored phyletic and molecular aspects of the bc ₁ complex (37, 70). Our results indicated a functional degeneration of the bc ₁ complex in various Acetobacteraceae, as shown in Fig. 7 (see also Fig. S1B and S8). It has been reported previously that the genome of Acetobacter spp. contains the gene for a deranged COX1-like protein lacking all the conserved residues binding the metal cofactors (13). However, this gene has been retained as part of a cluster containing the gene for CtaB, the enzyme that produces cytochrome o for the bo ₃ ubiquinol oxidase (13), probably fulfilling a structural role in stabilizing CtaB. Therefore, Acetobacteraceae genomes maintain genes for respiratory proteins that are deranged, a situation that is particularly relevant for the bc ₁ complex, which may be required for the stability of Complex I (44). This enzyme is of questionable function in acetous taxa (13, 15, 60) and even more in Acidiphilium (45), which lacks the gene for ISP (Fig. S8) and also the two conserved histidines that bind the cytochrome b _H [Fig. S1B, cf. reference (70)]. We found a similar situation in Acidocella genomes, which contain an ISP subunit that is even more deranged than cytochrome b (Fig. 7; Fig. S8). Acidocella ISP lacks three or four ligands of the Fe₂S₂ cluster but maintains the couple of cysteine residues that hold the cluster together by a disulfide bond (71) (Fig. S8C).

Conversely, the long ISP typical of acetous taxa and gammaproteobacteria (71) was found also in Rhodopila, Antarctic MAGs of clade c, and the Lichenicoccus group (Fig. S9). This long version of ISP generally shows the substitution of Tyr165 with Phe (Fig. S8C), which prevents the H-bond that Tyr165 normally entertains with a cluster ligand (71), thereby significantly reducing the activity of the bc ₁ complex (73, 74). It is therefore likely that the bc ₁ complex encoded by the split petA and petBC operons in Lichenicoccus, Lichenicola, Acetobacter, and related acetous taxa (13 – 15, 36, 72) has reduced catalytic activity. Intriguingly, the genome of the single Acidisoma taxon that retains the petABC operon shows intermediate features, consistent with the intermediate position of this taxon in the phylogenetic trees of the ISP protein (Fig. S9) and cytochrome b (Fig. 7C; Fig. S1B), as well as in those of cytochrome c ₁ (Fig. S10) and Cox15 (Fig. S11). In all such trees, the sisterhood of the acidophilic and acetous clades is strongly supported, as shown in Fig. 3A; Fig. S1C.

We undertook phylogenetic analyses of Cox15 (35) because its gene is present in the genome of Acetobacter spp. (13) and is necessary for A2-type COX subtype a-I present in some acetous taxa (Fig. S2; Table S2). Many other acetous taxa possess a cytochrome c peroxidase that may undertake the re-oxidation of reduced cytochrome c, as proposed in Acetobacter aceti (14) (Fig. S2A). Such a possibility is probably not as efficient as COX; therefore, the respiratory chain of Acetobacter, Gluconobacter, and related acetous taxa may be physiologically deranged at the level of cytochrome c (13). This indicates that both acetous and acidophilic taxa share a functionally deranged respiratory chain, as discussed earlier. Ubiquinol oxidation might be compensated by the cytochrome bd ubiquinol oxidases that are widespread among Acetobacteraceae (9, 13, 14) (Fig. S4; Table S2). However, the puzzling question remains: why do most Acidocella and Acidiphilium taxa possess an apparently functional cytochrome oxidase while having a deranged bc ₁ complex? Besides the mentioned structural role in stabilizing Complex I (44), a possible answer resides in iron oxidation physiology (Fig. S4).

Acidiphilium species have been known for a long time to reduce ferric iron under micro-aerobic conditions (75), while Acidocella aromatica can grow using the same pathway under anaerobic conditions (76). In acidithiobacilli, the pathway of ferric iron reduction consists of isoforms of the outer membrane Cyc2 and periplasmic Cyc1 c-type cytochromes, which are directly involved in the opposite reaction, the oxidation of ferrous iron (19, 75, 77). Cyc2 proteins have been shown to resemble a beta-barrel porin with the cytochrome c heme exposed to the periplasmic space; they are more widely distributed than previously thought (78). We have confirmed the presence of Cyc2 homologs also in Acidocella MX-AZ03 and one of the Acidocella MAGs reported here, as well as in early branching Acetobacteraceae such as Acidibrevibacterium (Fig. 8A) and some Antarctic MAGs (Fig. S4). Notably, the gene for Cyc2 is often adjacent to another encoding either a mono- or di-heme c-type cytochrome (Fig. S4), which likely represents a functional homolog of the Cyc1 protein involved in iron oxidation (75, 77, 78). In acidithiobacilli, HiPIP is another component of the electron wiring connecting the oxidation of extracellular ferrous iron with c-cytochromes (19, 75, 78), although its role remains unresolved. Conversely, HiPIP proteins assist cyclic electron flow in photosynthetic purple bacteria (79), as well as in photoferrotrophy (80). Our analysis indicates that HiPIP must be crucial for sustaining aerobic photosynthesis in Acidisphaera sp. L21, which lacks genes for cytochrome c ₂ and its homologs (Fig. S4). HiPIP can also function as the oxidizing substrate for the bc ₁ complex (79), but it is present in acidophilic taxa that have deranged forms of the complex (Fig. S4).

Notably, the bc ₁ complex can transfer electrons from ubiquinol to cytochrome c ₁ even when cytochrome b _H is blocked by inhibitors such as antimycin (83). The absence of cytochrome b _H ligands in the PetB protein of acidophilic Acetobacteraceae (Fig. S1B) would produce a linear electron transfer equivalent to that carried out by the antimycin-inhibited bc ₁ complex, with electrons flowing to cytochrome c ₁, which has an intact structure in acidophilic Acetobacteraceae (Fig. S10). This situation would resemble, in principle, that described for a mutant of chloroplast cytochrome b ₆ which lacks a ligand of cytochrome b _H but partially sustains photosynthetic electron flow (84). However, in acidophilic taxa, the mutations of cytochrome b _H ligands are always associated with those of the ISP (Fig. S1 and S9), thereby preventing “back door” pathways for ubiquinol oxidation (84). Much less speculative is the possibility that the Cyc1 and Cyc2 proteins present in acidophilic and photosynthetic Acetobacteraceae (Fig. S4) constitute an alternative pathway to reduce c-cytochromes in the periplasm, which in turn are re-oxidized by COX as in acidithiobacilli (19) (Fig. 8B). Therefore, the oxidation of ferrous iron, or perhaps other reduced metals, provides the physiological reason why a functional COX is maintained in acidophilic Acetobacteraceae that have a deranged bc ₁ complex (Fig. S4).

This work provides a comprehensive and updated analysis of the phylogeny of Acetobacteraceae, a large group of alphaproteobacteria that is in need of taxonomic revision. Our data support the rank elevation to the order Acetobacterales proposed previously (28) and indicate two new subdivisions to be added to the order: Roseomonadaceae, comprising the Roseomonas clade (Fig. 1 and 2; Table 1), and Acidocellaceae, encompassing the acidophilic taxa of Acidisoma, Acidocella, and Acidiphilium. The first conclusion emerging from our work is that the latter family is a sister to the acetous clade, which may be re-classified as the family Acetobacteraceae sensu stricto. This family could include not only the acetous taxa that originally defined the family (2 – 4) but also the non-acetous genera Entomobacter, Commensalibacter, and Asaia (59). The Lichenicoccus group that we have uncovered here (Fig. 1A, 2A, and 3A) forms a subclade that is consistently clustering with the acetous clade and, therefore, could be part of the family Acetobacteraceae sensu stricto. This taxonomic proposal would strengthen the concept that all Acetobacterales originated from a photosynthetic ancestor, a novel major conclusion of our work.

Phylogenomic analyses have shown that several Acetobacterales maintain a “historical genomic record” in the form of progressively deranged proteins (especially of the bc ₁ complex) and gradual loss of photosynthetic and bioenergetic traits (Table S2 and Fig. S4). Indeed, the differential loss of photosynthetic traits constitutes a trail linking together various clades of Acetobacterales, explaining the increased presence of PQQ-dependent Q reductases as potential compensation for the loss of the Q reductase function of RC. Pathways for oxidoreduction of external metals constitute essential traits in early branching and acidophilic Acetobacterales, providing a functional bypass for the deranged bc ₁ complex that such taxa often possess (Fig. 8B). Altogether, our results complete the evolutionary panorama of Acetobacterales showing a progressive transition from versatile photoferrotrophy to the incomplete oxidation of organic substrates defining acetous ecophysiology.