An ultimate validation of the genomics-based metabolic reconstructions was attained by experimental testing of growth phenotypes. Four UCC isolates including the Halomonas sp. HL-48 and HL-93 strains, Roseibaca calidilacus HL-91 and Marinobacter sp. HL-58 were tested for their ability to grow on a panel of various carbon sources as a sole carbon and energy source. Cells were grown in Hot Lake Heterotroph (HLH) medium (Cole et al., 2014), containing 10 mM TES pH 8.0, 400 mM MgSO₄, 80 mM Na₂SO₄, 20 mM KCl, 1 mM NaHCO₃, 5 mM NH₄Cl, and supplemented with 5 mM of a specific carbohydrate as a sole carbon source. For initial validation of genomic prediction of carbohydrate utilization we used bioMerieux^TM Api^TM50 CH carbohydrate fermentation strips. The Api^TM50 CH strip contains 49 wells with different carbohydrates in each well and one negative control well (no carbon source). The HLH media for starter cultures was supplemented with 5 mM glycerol (for HL-91) or 5 mM sucrose (for HL-48, HL-93, and HL-58). The mid-log phase grown cells were further washed three times with HLH medium without any carbon source to eliminate carry over of carbon substrates from starter media prior to inoculation and washed cells were used as inoculums in Api50 CH strip. Utilization of carbon sources were indicated by the growth on the well. For each strain, two independent repetitions were performed. The incubation time for Api50 CH strip measurements was 3 days (for HL-48, HL-93, and HL-58) and 7 days (for HL-91). Growth phenotype of Api50 CH strip results were further validated by growth measurements using selected carbohydrates. An optical density (OD₆₀₀) was measured to monitor cell growth during 60–100 h using a plate reader instrument Norden Lab Professional-Bioscreen. 250 μL culture volumes in the 100 well Bioscreen plate, and each growth experiments were performed in 10 replicates.

In both studied Halomonas species we found a new gene locus potentially involved in the utilization of L-arabinose and L-arabinonate (Figure 2). It encodes proteins that are orthologous to (i) the ABC-type arabinose uptake transporter system AraFHG and (ii) the AraA, AraC, and AraE enzymes from the oxidative arabinose degradation pathway in Azospirillum brasiliense (Watanabe et al., 2006a,b). Based on genome context and distant homology analysis we have identified candidates for the missing 2-keto-3-deoxy-L-arabonate dehydratase (AraD) and a second isozyme of arabinose-1-dehydrogenase (AraY). A member of aldose 1-epimerase family encoded in the ara gene cluster was previously assigned the functional role L-arabinose mutarotase (AraM), which interconverts alpha and beta anomers of L-arabinose. The ara gene locus in Halomonas encodes two novel TFs from the LysR and GntR families (named AraR and AraR2, respectively). Reconstruction of their cognate regulons using these and other Halomonas genomes has revealed two different DNA motifs (Supplementary Table S2). AraR presumably controls the divergently transcribed araMFGHCY and araR genes, whereas AraR2 is predicted to co-regulate the araD-araT-araR2-araA operon, the araE gene and several other genes encoding a novel transporter from the tripartite ATP-independent periplasmic (TRAP) transporter family and a hypothetical lactonase, which was assigned the missing arabinolactonase function (named AraB). Nearly all known TRAP-family transporters have specificities to organic acids (Vetting et al., 2015), suggesting the novel AraR2-regulated transporter is specific to L-arabinonate, an intermediate of the L-arabinose catabolism. In summary, the ara gene locus represents interconnection of two regulatory systems controlling a shared catabolic pathway for utilization of L-arabinose and L-arabinonate.

The oxidative pathway for utilization of L-fucose, where L-fuconate is an intermediate, was shown in Xanthomonas campestris (Yew et al., 2006). We observed loci containing genes from this pathway in both Halomonas species (HL-48 and HL-93), A. marincola HL-49 and Rhodobacteriaceae bin08 (Figure 3). HL-48 possesses genes encoding the last two steps of this pathway, COG0179 and COG1028. However, the absence of genes for the first two steps of the fucose pathway suggests that HL-48 can only utilize the L-fuconate intermediate. In contrast, HL-93 and bin08 have the complete L-fucose utilization pathway, whereas in HL-49 L-fuconate dehydratase FucD is missing and L-fucose dehydrogenase is substituted with a novel non-orthologous dehydrogenase (named FucO^II). The four UCC genomes have different transporters encoded in the fuc loci. HL-49 has a gene encoding an ortholog of fucose permease FucP from X. campestris. HL-93 and bin08 have two non-orthologous ABC systems that we predict to be involved in uptake of L-fucose.

The L-fucose/L-fuconate utilization gene loci in both Halomonas species contain the lgoD gene encoding L-galactonate-5-dehydrogenase, which is the signature gene of L-galactonate utilization (Kuivanen and Richard, 2014), the uxaA and uxaB (or uxaF) genes involved in the downstream steps of L-galactonate catabolism, as well as a novel TRAP-family transporter operon (Figure 3). A similar L-galactonate catabolic gene locus in Chromohalobacter salexigens (Csal_1738-1731) encodes the same set of catabolic enzymes and a non-orthologous TRAP transporter, which was previously characterized to have a dual specificity toward L-fuconate and L-galactonate (Vetting et al., 2015). Based on these observations, we propose that the novel TRAP system encoded within the L-fucose/L-fuconate/L-galactonate gene loci in Halomonas species is involved in the utilization of both L-fuconate and L-galactonate and thus was named Lgo/Lfo. This example introduces an interesting case of chromosomal co-localization (and, likely, co-regulation) of genes that are involved in the shared carbohydrate utilization pathways.

D-Galacturonate and D-glucuronate are hexuronic acids that are commonly found in pectins, proteoglycans and glucuronans. D-Galactarate and D-glucarate are the ring opened hexose diacids (or aldaric acids) that serve as a growth substrate to many microorganisms. Both Halomonas species (HL-48 and HL-93) contain a gene cluster encoding enzymes involved in the galactarate/glucarate utilization, as well as a novel predicted transporter from the tripartite tricarboxylate transporter (TTT) family (named TctABC) and a novel GntR-family TF (termed GguR). The TctABC transporter is predicted to be involved in galactarate/glucarate uptake. The reconstructed GguR regulon in both Halomonas genomes includes the galactarate/glucarate utilization operon, whereas HL-93 has an additional GguR-regulated operon, which encodes the glucoronate/galacturonate utilization enzymes Udh, Gli, and Gci, as well as an ortholog of the known hexuronate transporter UxuPQM. We concluded that HL-93 (but not HL-48) has an additional capability to utilize galacturonate and glucoronate using the pathway partially shared with the galactarate/glucarate pathway (Figure 4).

The R. calidilacus HL-91 and Rhodobacteriaceae bin08 and bin18 genomes encode a different variant of the glucuronate utilization pathway, which starts from the UxaC isomerase and continues through 2-dehydro-3-deoxygluconate (KDG) and its phosphorylated derivative, KDG-6P. The corresponding glucoronate utilization gene loci include a novel ABC-family transporter operon, which was predicted to be co-regulated with the glucuronate utilization genes by a novel GntR-family regulator (UxuR). However, the corresponding transporter and regulator are missing in HL-91, thus the mechanism of glucuronate uptake in yet unknown in this organism. We speculate that it can take up glucuronides that are hydrolyzed in the cytoplasm by the LfaA glucosidase, thus providing glucoronate to feed the catabolic pathway.

Algoriphagus marincola HL-49 has two separate loci with galacturonate and glucuronate utilization genes. Both of these gene loci are controlled by a novel LacI-family TF (UxuR2) and encode a novel or TRAP-family transporter (UxuPQM^II) and an uncharacterized glycosyl hydrolase from the GH109 family. UxuPQM^II is distantly related to the previously characterized UxuPQM transporters in various Proteobacteria (Vetting et al., 2015), however, it belongs to a distinct orthologous group of TRAP transporters that are mostly present in the Bacteroidetes phylum. We propose that UxuPQM^II also has the dual specificity for both hexuronic acids it is co-regulated with both galacturonate and glucuronate utilization genes in HL-49.

In both studied Halomonas genomes, we identified a conserved locus encoding proteins homologous to catabolic enzymes, a TRAP-family transporter and a GntR-family regulator that were previously characterized as a part of the L-gulonate utilization pathway in C. salexigens (Wichelecki et al., 2014). Thus, we predict that HL-48 and HL-93 are able to utilize L-gulonate.

In Escherichia coli and other Enterobacteria, D-galactose is utilized via the Leloir pathway, which involves galactokinase GalK, galactose-1-phosphate uridylyltransferase GalT and UDP-glucose 4-epimerase GalE (Holden et al., 2003). The Porphyrobacter sp. HL-46 and A. marincola HL-49 genomes contain the Leloir pathway genes, suggesting there are able to utilize D-galactose (Figure 5). The galactose gene locus in HL-46 contains a predicted galactose permease from the SSS family, which is orthologous to the GalP^II transporter previously identified in the galactose catabolic gene loci in Shewanella spp. (Rodionov et al., 2010). Additionally, the gal locus in HL-46 contain two genes encoding cytoplasmic galactosidases, RafA and BgaL, suggesting galactose-containing oligosaccharides may serve as additional inputs to the galactose catabolic pathway. In contrast, the galactose utilization pathway in HL-49 is incomplete with both the GalT uridylyltransferase and a galactose-specific transporter missing.

A different variant of D-galactose catabolic pathway, which is known as the DeLey-Doudoroff pathway, was identified in three Rhodobacteriaceae isolates, namely bin08, bin09 and bin18 (Figure 5). In this pathway, D-galactose is first oxidized to D-galactonate, which is then converted to pyruvate and GAP through the subsequent action of a dehydratase, a kinase and an aldolase (Wong and Yao, 1994). In addition to the DeLey-Doudoroff pathway enzymes, the galactose utilization gene loci in these three Rhodobacteriaceae genomes include genes encoding a cytoplasmic α-galactosidase, an unknown TF from the IclR family and a novel ABC-type transporter. This novel ABC transport system belongs to the Carbohydrate Uptake Transporter-1 (CUT1) family that mostly known to transporter di- and oligo-saccharides, according to the TCDB database (Saier et al., 2014). Thus we propose that this novel transport system is involved in uptake of α-galactosides and that the Rhodobacteriaceae spp. are able to utilize as α-galactosides rather than D-galactose.

In Salinivirga fredricksonii HL-109, the DeLey-Doudoroff pathway locus is missing the galactose dehydrogenase and lactonase that are required for conversion of D-galactose to D-galactonate. The dgoK gene in HL-109 is clustered with genes encoding a novel IclR-family TF (termed GalR) and a novel TTT-family transporter. Known transporters from the TTT family are specific to tricarboxylate and sugar acids (Saier et al., 2014). We predicted that a TTT-family transporter from the incomplete galactose catabolic gene locus is involved in D-galactonate uptake. We also propose that the GalR TF encoded in the same locus senses D-galactonate as an effector. Overall, HL-109 is the only UCC member that is able to utilize this sugar acid.

The DeLey-Doudoroff pathway genes for D-galactonate catabolism are present in Halomonas HL-48 and HL-93, as well as in several other reference Halomonas genomes. The corresponding dgo gene loci contain a hypothetical sugar lactone lactonase from (COG3386, named DgoL), which can serve as a non-orthologous gene displacement for D-galactono-1,4-lactone lactonase GalA. The reconstructed DgoR regulon in the Halomonas genomes includes additional candidate co-regulated gene encoding a SSS-family transporter with predicted galactose specificity (GalP^II). Although orthologs of known D-galactose dehydrogenase (GalD) are missing in Halomonas spp., we tentatively assigned them the galactose utilization capability, which is supported by growth phenotype testing (see below). Further similarity searches revealed one possible candidate for the missing GalD reaction in Halomonas spp. – a D-xylose dehydrogenase (XylD) from the xylose utilization gene cluster (it has 47% identity with the characterized GalD enzyme from Rhizobium meliloti). Thus, we tentatively propose that XylD in Halomonas spp. has specificity to multiple substrates including D-galactose and D-xylose.

The reconstructed peripheral pathways in heterotrophic UCC members contain 171 novel genes distinguishing them from those previously described in model species. These include 22 genes encoding novel enzymes with assigned function, 107 genes encoding components of novel sugar transporters and 42 novel sugar-specific transcriptional regulators. Most common are numerous cases of non-orthologous gene displacement, when a novel functional role is encoded by a gene that is not orthologous to any of the previously known genes of the same function. Several predicted non-orthologous enzymes involved in utilization of arabinose (AraB, AraD), fucose (FucO^II), galactose (DgoD^II, DgoL), and L-galactonate (UxaF) are described in details above. Other proposed cases of non-orthologous enzymes include a novel N-acetylglucosamine kinase (COG1070) in bin09, the putative sorbitol dehydrogenase SorD^II in bin07, and the predicted fructokinase MtlZ involved in utilization of arabinitol, sorbitol, and mannitol in both Halomonas spp.

Carbohydrate uptake transporters constitute the largest group of newly functionally assigned genes in UCC genomes. Most of these genes encode components of 27 multicomponent transport systems from the ABC, TRAP, and TTT families (Supplementary Table S2). Among 18 novel ABC systems most are predicted to transport hexoses (N-acetylglucosamine, N-acetylgalactosamine, fucose), oligosaccharides (α-/β-galactosides, α-glucosides, fructooligosaccharides), as well as glucoronate, sorbitol, erythritol, and mannoheptulose. All five newly predicted TRAP systems and three TTT-family transporters are specific to sugar acids (hexuronates, arabinonate, galactonate, fuconate) and hexose diacids (glucarate, galactarate). We also identified 12 novel single-component sugar permeases located in the inner membrane (AraT, BglT, BglT^II, GalP^II, FruT, MalP, COG2211), three TonB-dependent outer membrane transporters (with predicted specificities to sucrose and β-glucosides) and two novel outer membrane porins (possibly involved in uptake of glucose and glycerol).

Transcriptional regulation is another highly variable aspect of the sugar utilization pathways in UCC genomes. Indeed, 42 of the 88 TFs tentatively associated this UCC sugar utilization pathways are non-orthologous to their counterparts previously characterized in other bacteria, as captured in the RegPrecise database (Novichkov et al., 2013). We identified candidate TFBSs and reconstructed regulons for 60 TFs including 30 novel regulators (Supplementary Table S2). The majority of genes from sugar catabolic pathways were identified as candidate members of respective sugar-specific TF regulons in UCC genomes.

In Rhodobacteriaceae bin08, we identified a new gene locus encoding an ABC-family transporter, an aldolase from the tagatose-1,6-bisphosphate aldolase family (COG3684) and two kinases (COG1940 from the ROK family and COG0529 from the adenylylsulfate kinase family). The substrate-binding component of this ABC transport system has 83% similarity to Avi_5339 from Agrobacterium vitis, which was previously found to bind mannoheptulose (Steven Almo and John Gerlt, unpublished observation). Mannoheptulose is a heptose, which is structurally similar to D-tagatose. Based on these observations and known substrate specificities for other kinases from the COG1940 and COG0529 families, we propose the following hypothetical pathway for mannoheptulose utilization. The COG1940 kinase first phosphorylates the substrate to produce mannoheptulose-7-phosphate, then the COG0529 kinase further produces mannoheptulose-1,7-biphosphate, which is subject to the COG3684 aldolase reaction producing glycerone phosphate and erythrose-4-phosphate.

Three UCC members are predicted to utilize xylose though the classical isomerase pathway (XylB, XylA), whereas both Halomonas species have a different pathway for xylose utilization including xylose dehydrogenase XylD. In Caulobacter crescentus, xylose is converted to α-ketoglutarate by xylose dehydrogenase, xylonolactonase, and xylonate dehydratase (Stephens et al., 2007). However, we have not identified candidate genes for xylonolactonase and xylonate dehydratase in Halomonas spp. In contrast, the xylose utilization operon in Halomonas spp. encodes two hypothetical enzymes, a sugar phosphate isomerase/epimerase (COG1082) and a Gfo/Idh/MocA-family oxidoreductase (COG0673), however, their exact biochemical functions require further experimental characterization.

Another yet unknown pathway involved in rhamnose utilization was identified in Rhodobacteriaceae bin08. Its putative rhamnose catabolic locus contains genes encoding the RhaFGHJ transporter and the RhaM mutarotase, however, bin08 lacks other candidate genes required for utilization of rhamnose (Rodionova et al., 2013b). However, function of other six genes in this locus encoding putative enzymes is unclear. Comparative genomics shows that orthologous loci are present in other Rhodobacteriaceae genomes and in some cases they include genes encoding rhamnose dehydrogenase and rhamnonate dehydratase, however bin08 lacks orthologs of these genes, suggesting the existence of a novel yet uncharacterized rhamnose catabolic pathway in the Rhodobacteriaceae species.