Genes homologous to those encoding CsoSCA are quite widespread beyond carboxysome loci, and are present in genomes from autotrophic (e.g., Sulfuritortus caldifontis, Nitrospia marina) and heterotrophic (e.g., Cand. Accumulibacter phosphatis; Chrysiogenes arsenatis) bacteria. Given their widespread distribution, it is surprising that they have yet to be studied (Table 1; Figure 2; referred to as csoSCA2 to distinguish them from those present in carboxysome loci). The amino acid sequences predicted from csoSCA2 genes share many features with CsoSCA proteins; they include both an active and defunct domain (Sawaya et al., 2006), and the active domain includes all of the residues necessary for catalytic activity as carbonic anhydrase (Table 2; Figure 3).

There are two variants of CsoSCA2. The first variant closely resembles carboxysomal CsoSCA (found in Nitrobacter vulgaris, Nitrobacter winogradskyi, Nitrosomonas nitrosa, and Nitrosomonas sp. 51), clustering with carboxysomal CsoSCA in phylogenetic analyses, but missing the N-terminal 40 residues (Figure 2; Figure 3). The second variant is further truncated at the N-terminus, is missing short stretches of sequence throughout, and does not cluster with carboxysomal CsoSCA sequences (Figure 2; Figure 3). The more substantially truncated version has an N-terminal domain of only 40 amino acids (instead of 144 for the CsoSCA from Halothiobacillus neapolitanus) that does not align with the CsoSCA equivalent on a sequence level but is also predicted to form two short alpha helices in an AlphaFold2 model (Jumper et al., 2021; Figure 3). Further truncations include shorter loops connecting secondary structure elements (Figure 3). Some of those extra elements are involved in dimer contacts (Sawaya et al., 2006), so it is possible that this homolog has lost the ability to form dimers, which would be unusual for a β-carbonic anhydrase (Cannon et al., 2010); however, this would need to be verified experimentally. The N-terminus of CsoSCA from Htb. neapolitanus facilitates interaction between CsoSCA and RubisCO (Blikstad et al., 2021). Presumably, since CsoSCA2 proteins do not interact with RubisCO, this N-terminal region is not necessary for CsoSCA2 to function outside of carboxysomes. Altogether, this form of CsoSCA2 seems to be a more compact version, possibly due to the fact that it is not necessary to encapsulate this protein in a carboxysome. This is particularly interesting for genomes that include both csoSCA and csoSCA2 genes (e.g., members of Nitrobacter, Nitrosomonas, Ectothiorhodospira, and Halorhodospira). Presumably, the CsoSCA2 proteins cannot assemble within the carboxysomes present in these organisms.

The presence of CsoSCA2 sequences in numerous phyla, and the more restricted distribution of CsoSCA, suggest that CsoSCA may have originated from CsoSCA2. However, in some cases, the reverse appears to be the case. Nitrobacter vulgaris, Nb. winogradskyi (Nb-102, 106, and 255), Ns. nitrosa, and Nitrosomonas sp. 51 have genes encoding both a CsoSCA (encoded in a typical carboyxsome locus), and a CsoSCA2 encoded elsewhere. The two copies cluster together within the larger clade of carboxysomal CsoSCA sequences, despite having the truncated N-termini seen in other CsoSCA2 sequences (Figure 2; Figure 3). Sequence similarities between CsoSCA and CsoSCA2 proteins in these organisms suggest that these CsoSCA2 sequences duplicated and diverged from CsoSCA.

One wonders why these deeply divergent β-carbonic anhydrases are so widespread, and what the features of these proteins might be that make them particularly useful to their host organisms. Though CsoSCA2 proteins lack the residues needed to associate with RubisCO, they may have residues that facilitate the formation of other types of enzyme complexes. Alternatively, CsoSCA2 proteins may not require aggregation with other proteins for activity. Indeed, even carboxysomal CsoSCA is active when expressed in the absence of other carboxysomal proteins (Heinhorst et al., 2006), which suggests that “free” CsoSCA2 could also be active in the cytoplasm of its host organisms.

In genomes from some members of Alpha- and Betaproteobacteria, csoSCA homologs are present near genes encoding RubisCO (Figure 1; Figure 4). In Alphaproteobacteria, three strains of Nb. winogradskyi share this arrangement of genes (though the average nucleotide identities of strains Nb-102 and Nb-106 versus Nb-255 are 94.6%, suggesting they may be a different species; Richter and Rosselló-Móra, 2009). Among Betaproteobacteria, two species of Sulfuricella have cbbL and cbbS genes near csoSCA homologs (Figure 4).

For both the Alphaproteobacteria and Betaproteobacteria, if these genomically juxtaposed cbbL, cbbS, and csoSCA genes are the fragments of a single degraded carboxysome locus, one would anticipate that phylogenetic analysis would place them among genes encoding their carboxysomal cognates from taxonomically affiliated organisms. For the Nitrobacter spp. and Sulfuricella spp., the cbbLS genes cluster with noncarboxysomal RubisCO genes (Figure 4). The situation is more complicated for the csoSCA homologs (Figure 2). For the Nitrobacter spp., the csoSCA homologs appear to be recent duplicates of those present in the typical carboxysome loci in their genomes. For the Sulfuricella spp., the csoSCA homologs fall within the csoSCA2 clade and are unlikely to have arisen from carboxysome loci. Given the noncarboxysomal origin of the cbbLS genes in both classes, and csoSCA gene in the Sulfuricella spp., these are not fragments of a single degraded carboxysome locus.

Despite the likelihood that they do not share evolutionary history cohabitating carboxysomes, it is still possible that these two enzymes might function together in the cytoplasm to facilitate CO₂ fixation in their host organisms, all of which are capable of autotrophic growth (Winogradsky, 1892; Kojima and Fukui, 2010). Coregulation is possible for both, but seems more likely for the members of Nitrobacter, since their genes are <2 kb apart (Figure 1). The juxtaposition of noncarboxysomal RubisCO genes to those encoding typical β-carbonic anhydrase has been noted for two members of Hydrogenovibrio (Yoshizawa et al., 2004; Scott et al., 2006), and is apparent in genome data from in many other members of Hydrogenovibrio and Thiomicrorhabdus,¹ suggesting such juxtaposition may be selected for in some organisms. While the expression of cytoplasmic carbonic anhydrase results in a high CO₂-requiring phenotype in organisms with CCMs (Price and Badger, 1989), there is evidence for carbonic anhydrase activity in the chloroplasts of certain algae and plants (reviewed in Moroney et al., 2001). If these enzymes do function together in Nitrobacter and Sulfuricella, perhaps the carbonic anhydrase facilitates RubisCO-mediated CO₂ fixation by maintaining intracellular HCO₃⁻ and CO₂ near chemical equilibrium, preventing RubisCO from diminishing the concentration of intracellular CO₂ under conditions where CCMs are not induced (e.g., moderate environmental CO₂ concentrations; Yoshizawa et al., 2004).

Many members of family Thiobacillaceae (Boden et al., 2017; Boden, 2019) have csoS1–S4 genes in a separate genomic locus from cbbL and cbbS genes (Figure 1), as has previously been described for Thiobacillus denitrificans (Cannon et al., 2003; Beller et al., 2006a). Of the eight genome sequences from cultivated members of this family, all of which grow autotrophically (Boden et al., 2017; Boden, 2019), five include a homolog to csoS2 (Pfam012288; Annwoodia aquaesulis, Sulfuritortus calidifontis DSM103923 and J1A, Thiobacillus denitrificans ATCC25259, and Thiobacillus thioparus). In all cases, these csoS2 genes do not have cbbL and cbbS genes immediately upstream. Instead, cbbL and cbbS are located 2.6–21 kb away from csoS1-4 (Figure 1). The other three genome sequences lack csoS2 homologs; since these three sequences are incomplete (38–98 scaffolds), it is not possible to know whether csoS2 is truly absent from these organisms. Nine genomes (15–407 scaffolds) inferred to belong to members of Thiobacillaceae have been gathered from metagenomes, and five of these include csoS2 homologs. Three of these genes are present at the ends of scaffolds, making it impossible to determine whether cbbL and cbbS genes are nearby. For the two remaining (Thiobacillus spp. Bin4_E1B and BP01), cbbL and cbbS are encoded separately from csoS1–4. Based on these observations, it seems likely that having csoS1–4 genes apart from cbbL and cbbS genes may be a trait shared by all members of this family.

There are two mechanisms that could have resulted in the separate cbbLS and csoS1-4 loci found in members of Thiobacillaceae. In the first scenario, a typical ancestral carboxysome locus containing all of these genes was severed by genome re-arrangement. In the second scenario, cbbLS and csoS1-4 did not share an ancestral locus. Instead, carboxysomal cbbLS genes could have been lost from the genome entirely, and the cbbLS genes currently located 2–20 kb away are noncarboxysomal in origin. An additional possibility is that either cbbLS or csoS1-4 were acquired via horizontal gene transfer.

To provide evidence for different mechanisms for formation of cbbLS and csoS1-4 loci, phylogenetic analyses were conducted using concatenated alignments of cbbL and cbbS genes (cbbLS), and csoS2, S3, S4a, and S4b (csoS2-4). Genes encoding CsoS1A-C were omitted from these analyses, due to difficulties distinguishing the three types of csoS1A-C genes. The results of these analyses raise the possibility that cbbLS and csoS1-4 loci did not originate from a single ancestral typical carboxysome locus in these organisms (Figure 5). In Sf. calidifontis (here collapsed to strain J1A, since sequences for the two strains are identical), cbbLS genes fall in a small well-supported clade with Tb. denitrificans, and with other members of Thiobacillaceae in larger clades, though these larger clades are not as well-supported (Figure 5). The csoS2-4 genes from S. calidifontis fall among completely different taxa than its cbbLS genes, suggesting independent origins for its two loci. For all four isolates from Thiobacillaceae, cbbLS genes do not fall among those from typical carboxysome loci, though it should be noted that carboxysomal and non-carboxysomal cbbLS genes are not distinguished by two distinct, well-supported clades (Figure 5). Together, these observations suggest independent origins for cbbLS and csoS2-4 loci in Thiobacillaceae, but low bootstrap values for these phylogenetic analyses compromise the confidence of this assertion. If the cbbL and cbbS genes in members of Thiobacillaceae did not originate from a typical carboxysome locus, it would be very interesting to verify that they were capable of being packed into carboxysomes, as thus far it seems that noncarboxysomal RubisCO from other organisms cannot be packed into carboxysomes (Menon et al., 2008).

Currently, evidence for the presence of carboxysomes in members of Thiobacillaceae is limited to Tb. thioparus. Transmission electron micrographs have only been published for Tb. thioparus and Tb. denitrificans; polyhedral inclusions are apparent in Tb. thioparus cells, but not in Tb. denitrificans (Shively et al., 1970). Given the synteny of the carboxysome loci among Thiobacillus sp. Bin4 E1B, Thiobacillus sp. BP01, and Tb. thioparus, as well as the placement of their cbbLS and csoS2-4 genes together in clades (Figure 4), it seems likely that all three of these organisms are capable of synthesizing carboxysomes. For Tb. denitrificans, the absence of carboxysomes despite the presence of csoS1-4 is puzzling, and cannot be attributed to strain-level differences, since both ultrastructure and genome sequence were obtained from the same strain (ATCC25259). Perhaps their synthesis could be induced under growth conditions different from those used to cultivate the cells for ultrastructural study.

If carboxysomes are indeed synthesized by these organisms, one possible advantage of having separate loci would be independent regulation of cbbLS and csoS1-4 loci. Transcriptome analysis of Tb. denitrificans is consistent with this possibility. Based on hybridization with microarrays, transcripts of the Tb. denitrificans cbbL and cbbS genes are particularly abundant under aerobic conditions, but no such changes are apparent for csoS1-4 (Beller et al., 2006b). Other organisms have two sets of cbbLS, with one in a typical carboxysome locus and the other located elsewhere on the genome. These organisms synthesize noncarboxysomal RubisCO when CO₂ concentrations are moderate, and selectively synthesize carboxysomal RubisCO and shell proteins when CO₂ concentrations are very low (Yoshizawa et al., 2004). Perhaps some members of Thiobacillaceae upregulate the cbbLS locus when CO₂ concentrations are low to moderate, and reserve upregulation of the csoS1-4 locus for low CO₂ conditions, or other circumstances where carboxysomes facilitate growth.

Carboxysome loci lacking csoSCA genes arose multiple times; they are present in some autotrophic organisms from Beta- and Gammaproteobacteria. A carboxysome locus lacking csoSCA is also present in Pseudonocardia sp. N23, a member Actinobacteria (Figure 1); though it has not been determined whether this organism could grow autotrophically, other members of its genus can (e.g., Pseudonocardia autotrophica; Takamiya and Tubaki, 1956). None of these organisms have csoSCA homologs elsewhere in their genomes (aside from Thiomicrorhabdus sediminis, which has a copy in its “typical” carboxysome operon). In Betaproteobacteria, they are present in Nitrosospira muliformis and also Nitrosospira spp. Nsp5 and Nsp6, which may be strains of Nsp. multiformis, based on average nucleotide identities >99% (Richter and Rosselló-Móra, 2009). Within Gammaproteobacteria, they appear to have arisen independently three times. All members of Thiomicrospira have carboxysome loci lacking csoSCA. Within Thiomicrorhabdus, such loci seem to have arisen twice. In Thiomicrorhabdus sediminis, two carboxysome loci are present; one is typical, while the second appears to be a recent duplicate of the typical locus. The amino acid sequences predicted from both cbbL and cbbS genes are 100% identical. Both copies of CsoS2 are 100% identical at amino termini; however, at residue 330, they diverge, and this continues to the carboxy termini. CsoS1 sequences also are identical at the amino termini and have small differences at their carboxy termini. The carboxysome locus from Thiomicrorhabdus aquaedulcis does not fall within a clade with those from other members of its genus (Figure 6), suggesting that it may have been acquired via horizontal gene transfer.

The carboxysome locus from Pseudonocardia sp. N23 includes cbbL and cbbS genes distinct from those present in other members of phylum Actinobacteria (Figure 4). Other members of this phylum carry cbbL and cbbS, and Pseudonocardia sp. N23 does include a copy that falls within a clade of these sequences (Figure 4). However, the cbbL and cbbS genes present in the carboxysome locus, as well as csoS2, csoS4A, and csoS4B, fall among genes from carboxysome loci from members of Alpha- and Betaproteobacteria (Figures 4, 6), suggesting this locus was acquired via horizontal gene transfer.

It is apparent that these carboxysome loci originated from typical carboxysome loci, given that they cluster with others that contain csoSCA genes (Figure 6). Indeed, in members of Thiomicrospira, a gene is present between csoS2 and csoS4A (Figure 1), which is likely to be a degraded form of csoSCA. In Thiomicrospiras pelophila, thyasirae, and microaerophila, these genes do match Pfam08936 (csoSCA), but e-values range from 0.006 to 4.5e−05, and none of the residues necessary for carbonic anhydrase activity are present. However, the amino termini of the proteins predicted from these genes align well with those from CsoSCA proteins. Given that the amino termini of CsoSCA proteins may facilitate interactions among carboxysome proteins (Blikstad et al., 2021), perhaps these degraded genes may still encode proteins that facilitate packing of RubisCO molecules into carboxysomes.

There is evidence that these carboxysome loci are transcribed and translated. Carboxysome locus genes are transcribed in Tms. pelophila (Scott et al., 2019), carboxysomes are visible in transmission electron micrographs of members of Thiomicrospira (Sorokin et al., 2001, 2002a,b; Scott et al., 2019), and have been purified from Tms. thyasirae (Lanaras et al., 1991). Electron dark inclusions are abundant in Nsp. muliformis, but staining patterns suggest these consist of glycogen (Watson et al., 1971). Ultrastructural studies of Pseudonocardia sp. N23, as well as Tmr. aquadulcis and sediminis, have not been published.

Given the presence of carboxysomes in at least some of these taxa, the conservation of residues necessary for the function of the CbbL, CbbS, CsoS2, and CsoS4A and B proteins, and the convergent evolution of this sort of carboxysome locus in multiple lineages of microorganisms, they are likely to be functional in their host organisms (however, see the comments on CsoS1 sequences from Pseudonocardia sp. N23 in Section “Do the Genes From Atypical Carboxysome Loci Encode Functional Proteins” above). The current understanding of carboxysome function requires the presence of carbonic anhydrase activity within these microcompartments in order for them to facilitate CO₂ fixation by RubisCO (see above). One possibility is that these modified carboxysomes have shells that are permeable to CO₂, allowing this gas to enter from the cytoplasm. CsoS4 proteins are necessary for carboxysome shell impermeability to CO₂; the absence of CsoS4 to seal the vertices of their shells renders the microcompartments CO₂-permeable (Cai et al., 2009). Their critical function perhaps accounts for their strong sequence conservation, hence redundancy, which is unusual for bacterial microcompartments that have multiple pentamer-forming paralogs (Melnicki et al., 2021). Interestingly, the carboxysome locus from Tmr. aquaedulcis lacks genes encoding CsoS4A, and the Tmr. sediminis locus lacking csoSCA lacks both csoS4A and csoS4B. Perhaps carboxysomes from these organisms operate without CsoS4 proteins, and are permeable to CO₂. Given that carboxysome shells are assumed to require only 12 pentamers, and their pores are small (~4 Å in diameter), they are assumed to not play a significant role in metabolite conductance. However, a recent study of the protein stoichiometry of β-carboxysomes showed varying occupation of the vertices by the CcmL, the lone pentamer-forming gene product in beta carboxysome loci (Sun et al., 2019). The occupancy was correlated with environmental conditions, suggesting that pentamer association with shells is dynamic and perhaps serves as one way to alter permeability. Because Tmr. sediminis has two carboxysome loci (one typical, one lacking csoSCA, csoS4A, and csoS4B), determining the conditions under which it expresses typical, vs. atypical, carboxysomes could provide useful information about how its atypical carboxysomes might function, including whether pentamers and carbonic anhydrase are provided by the other locus. If these carboxysomes are permeable to CO₂, cytoplasmic CO₂ concentrations would need to be elevated in order to enhance RubisCO activity, running the risk of high rates of CO₂ loss from the cells via diffusion, unless this loss is counterbalanced by living in a high CO₂ habitat. These organisms have been cultivated in growth media supplemented with HCO₃⁻ (10–30 mM; Kojima and Fukui, 2019) or CO₂ (20% headspace; Liu et al., 2021). For Tmr. sediminis, the lack of csoS4 genes suggests that this organism may not be capable of growth under low CO₂ conditions, and it would be interesting to determine whether this is the case.

The other organisms lacking CsoSCA have loci including genes encoding CsoS1 and CsoS4; perhaps their shells are permeable to CO₂ based on modifications to these two types of shell proteins. However, such differences are not detected when shell proteins from typical carboxysomes are compared to those from carboxysomes lacking carbonic anhydrase. For CsoS1ABC proteins, the sequence FVGGGY, corresponding to residues 40–45 from Htb. neapolitanus, comprises the narrowest part of the pore and the residues surrounding it (Tsai et al., 2007). In all of the CsoS1ABC sequences from atypical carboxysome loci lacking csoSCA, these residues are conserved, suggesting the pores have characteristics similar to those in typical carboxysomes. To determine whether there are other residues that vary systematically for these atypical carboxysome loci, and to detect changes in the sequence that are more likely due to the presence/absence of CsoSCA rather than evolutionary distance, CsoS1ABC sequences within Piscirickettsiaceae were compared, since genomes from this family include both typical (all 10 species of Hydrogenovibrio, 8/10 species from Thiomicrorhabdus) and atypical (2/10 species from Thiomicrorhabdus, all six members of Thiomicrospira) loci. Among all of these organisms, CsoS1ABC sequences are highly conserved throughout the sequences. Likewise, CsoS1D sequences from these organisms have small differences throughout, and mapping those differences on a homology model does not reveal significant patches of variability. CsoS4A and B sequences are also very similar across all three genera and there are no distinguishable differences between them. If these shell proteins actually are permeable to CO₂, the mechanism mediating this change is not apparent from their sequences.

Another mechanism for preserving the activities of these carboxysomes would be their recruitment of a carbonic anhydrase encoded elsewhere on the genome, as may be the case for some β-carboxysomes from Cyanobacteria. β-carboxysomes carry homologs to γ-carbonic anhydrase (Dearaujo et al., 2014). In some cases, these homologs are enzymatically active as carbonic anhydrases, while in others, these homologs have apparently lost enzymatic activity (Cot et al., 2008), although the active site residues are intact. In these cases, the carboxysomes also carry a functional β-carbonic anhydrase (deeply divergent to CsoSCA; So et al., 2002; Cot et al., 2008; Rae et al., 2013), and the gene encoding this β-carbonic anhydrase is not present in or near the operon encoding the essential components of the carboxysome (Rae et al., 2013). Evaluation of these possibilities awaits purification of carboxysomes from organisms with carboxysome loci lacking csoSCA genes, to test the permeabilities of their shells and the potential presence of carbonic anhydrase activity within them.