Cyclic nucleotide binding domains in the National Center for Biotechnology Information's non-redundant amino acid database (NR) and Global Ocean Sampling (GOS) 1920 data were identified using a combination of psi-blast profiles and motif models (see Materials and methods). This resulted in nearly 5,241 significant hits in NR and 2,455 hits in the GOS data. Most of the identified sequences were multi-domain proteins in that they contained other functional domains covalently linked to the CNB domain. Because these functional domains play an important role in CNB domain functions, they were used as markers for annotation and classification (see below).

The 7,696 CNB domain containing sequences can be classified into 30 distinct families (Figure 1) based on the sequence similarity within the CNB domain (see Materials and methods). These 30 families are predominantly eukaryotic or bacterial in origin (Table 1). The only significant hit in Archea was to a hypothetical protein (gi: 11498576) from Archaeoglobus fulgidus. CNB domains in eukaryotes can be broadly classified into five major categories: the kinase domain associated PKA and PKG families; the guaninine nucleotide exchange factor (Epac's); transmembrane domain containing HCN and Na channels; HCN type channels in protozoans; and CNB domains in metazoans and plants that are fused to functional domains such as PAS domains, PP2C like phosphatases and phospholipases ('Other_Eukaryotic' in Table 1). Several of these families/subfamilies are lineage-specific and contain domain combinations that have not been reported before. The PP2C like phosphatase, for instance, is a plant specific subfamily that contains a kinase domain carboxy-terminal of the CNB domain. The co-occurrence of kinases, phosphatase and CNB domains in the same operon is interesting because previous bioinformatics analysis had failed to provide any evidence for a cAMP or cGMP dependent regulation of kinase activity in plants 21.

CNB domains are also prevalent in prokaryotes and some of the major groups include: the CRP family members (Marr, Arsr, AsnC, ICLR, GNTR) that contain a DNA binding domain covalently linked to the CNB domain; and a distinct class of DNA binding domain containing proteins (NnR, ArcR, Fnr and FixK) that are activated by second messenger signals such as NO, oxygen and heme 10. In addition, our analysis reveals several novel families (CBS, HisK and AAA ATPases) in prokaryotes that lack the DNA binding domain, but conserve other functional domains (Table 1) such as histidine kinases (HisKs), cystathionine beta synthase (CBS) domains and AAA ATPases (AAA_Atpases in Table 1).

Until now, the primary function of CNB domains in prokaryotes was believed to be in the transcriptional regulation of genes. However, our analysis suggests that other cellular processes, such as ATP production, protein phosphorylation and NADH production, may also involve CNB domain functions (Table 1). Of particular interest is the CBS domain associated CNB domains. CBS domains are known to function as sensors of cellular energy levels in eukaryotes as they are activated by AMP and inhibited by ATP. They are also implicated in various hereditary diseases in humans 24. The function of CBS domains in prokaryotes, however, is poorly understood, although the crystal structure of a CBS domain from Thermotoga maritime has been determined as part of the structural genomics initiative 25. The occurrence of both a CBS domain and a CNB domain in the same open reading frame suggests that, in some bacteria, ATP levels may be regulated in a cAMP-dependent manner. Structurally characterizing the full-length protein (CBS + CNB domain) may shed light on this regulatory mechanism in prokaryotes.

Other novel domains in prokaryotes that are fused to CNB domains include the HisKs that are involved in bacterial two component signaling, and the AAA class of ATPases (AAA_Atpases in Table 1) that control a wide variety of cellular functions in both eukaryotes and prokaryotes 26.

While the functional domain linked to the CNB domain is unique to a given family or subfamily, the CNB domain is shared by the entire superfamily. A multiple alignment of nearly 7,000 CNB domain sequences (Figure 2) reveals key sequence motifs that are shared by the entire superfamily (Figure 2). These residues/motifs define the core of the CNB domain. Several of these core residues correspond to glycines (Gly159, Gly166, Gly178, Gly195, and Gly199) that are located in loops connecting the beta strands of the beta subdomain (Figure 3). Note that the residue numbers correspond to PKA-mouse numbering in Figure 2. The most conserved of these glycines is Gly178, which is located in the β3-β4 loop and adopts a main-chain conformation (phi = 85.0; psi = -176.5) that is disallowed for other amino acids in the Ramachandran map. The role of Gly178 is not obvious from crystal structure analysis; however, the remarkable conservation of this residue across diverse eukaryotic and prokaryotic phyla suggests an important role in CNB domain structure and function.

In addition to the conserved glycines, CNB domains also conserve a hydrophobic core in the alpha and beta subdomains. The hydrophobic core in the alpha subdomain is formed by residues Phe136, Ile147, Tyr229, and Ile224, while the core in the beta subdomain is formed by residues Ile175, Met180, Val213, Val162, Phe198 and Tyr173 (Figures 2 and 4a). Comparison of the cAMP-bound and the catalytic subunit-bound structures of the PKA regulatory subunit (R1alpha) reveals that while the hydrophobic core in the beta subdomain is relatively stable in the two functional states, the hydrophobic core in the alpha subdomain is malleable and undergoes a conformational change upon binding to the catalytic subunit (Figure 4b). In particular, Tyr229, which packs up against the PBC in the cAMP-bound structure moves away from the PBC upon binding to the catalytic subunit (Figure 4b). Likewise, Phe136, which typically points away from the PBC, moves closer toward the PBC upon binding to the catalytic subunit. These coordinated changes in the helical subdomain were recently proposed to function as a latch for gating cAMP 13 and also shield cAMP from solvent. The conservation of these core residues across diverse families suggests that the conformational changes in the alpha subdomain may be a fundamental feature of all CNB domain functions.

Having delineated the core residues/motifs of the CNB superfamily, we focused on motifs that contribute to the functional specificity of individual families. In particular, we focused on the PBC region (Figure 5a), which displays a strikingly different pattern of conservation in some families (Figure 5b). The canonical sequence motif in the PBC region is the FGE [L,I,V]AL [LIMV]X [PV]R²⁰⁹ [ANQV] motif, where X is any amino acid. A key residue within this motif is a conserved arginine (Arg209), which coordinates with the phosphate group of cAMP (Figure 5c). While mutation of this arginine to a lysine in PKA reduces the affinity for cAMP by nearly ten-fold 27, some eukaryotic families, such as PDZ_GEF (PDZ domain associated family closely related to Epac), naturally contain a methionine or histidine at the Arg209 position (Figure 5b). Although the functional implications of this variation in PDZ_GEF (Figure 5d) are currently unclear, it is likely that this may alter the affinity for cAMP or facilitate binding of a different small molecule ligand. Notably, in the crystal structure of PDZ_GEF, which was solved as part of the RIKEN structural genomics initiative, the region analogous to the PBC region in PKA adopts a strikingly different conformation (Figure 5d) and is not bound to any ligand.

The ability of the CNB domain to bind to diverse ligands raises an important question: what features distinguish the cAMP binding families (ones that conserve a canonical PBC motif) from those that bind to other ligands? In order to address this question we used the CHAIN (Contrast Hierarchical Alignment and Interaction Network analysis) program, which quantifies the differences between two functionally divergent groups of sequences using statistical methods 28. Using this program, we identified sequence features that distinguish the canonical PBC motif containing CNB domains from those that lack the canonical PBC motif. Analyzing these features in light of existing structural and biochemical data provides a model for allosteric regulation, which is likely conserved in all cAMP binding modules.

CNB domains in GOS and NR data were identified using a combination of psi-blast 29 and Gibbs motif sampling procedures 30. Psi-blast profiles and motif models were initially built using CNB domains of known structures. These models were then iteratively updated as distant members from NR and GOS data were identified. An e-value cutoff of 0.001 was used for psi-blast searches.

CNB domains identified from NR (5,241 sequences) were multiply aligned using the CHAIN analysis program 28. The aligned sequences were clustered into families and sub-families using the clustering option in the CHAIN program and the SECATOR program 31. Families were annotated by identifying the functional domains linked to the CNB domain. The taxonomic origin of the sequences was also taken into account in the annotation processes. For instance, PKG-like CNB domains from parasitic organisms were annotated as 'PKG_parasites'. Functional domains were identified using rpsblast, which was run against a collection of conserved domains in CDD, Smart and Pfam 32 with an e-value cutoff of 0.0001.

Because CNB domains in the GOS data displayed significant sequence similarity to known CNB domains, they were assigned to one of the 30 families by running them against 30 family specific blast profiles. The taxonomic assignment for the GOS sequences was likewise done based on their similarity to known NR sequences 19. Examination of the domain organization in individual families indicated that while the NR sequence contained both the CNB domain and functional domains, GOS sequences usually contained only the CNB domain. This presumably is due to the fragmentary nature of the GOS data. In any case, nearly all the CNB domain containing GOS sequences could be assigned to one of the 30 families based on the similarity within the CNB domain alone.

In order to visually examine the evolutionary relationship between the identified sequences, we first constructed a phylogentic tree of all the 7,696 CNB sequences. The resulting tree, however, was very complex and hard to interpret. Therefore, we decided to take an alternative approach where we depicted each family by a consensus sequence. The 30 consensus sequences, corresponding to each of the 30 families, were generated from multiple alignments of individual families. The neighbor joining algorithm as implemented in the Molecular Evolutionary Genetics and Analysis (MEGA) program 33 was used for tree construction and visualization. Bootstrap test was done using default settings in MEGA.

The evolutionary constraints imposed on CNB sequences were measured using the CHAIN program 28. In brief, the CHAIN program identifies co-conserved residues that distinguish two related sets of sequences (foreground and background) by measuring the degree to which aligned residue positions in the foreground set are shifted away from the corresponding position in the background set. Residue positions that are shifted the most (indicated by red histograms above the alignment) contribute to the functional divergence of the foreground set from the background set. In the current study, all the CNB sequences that contain the canonical PBC motif constitute the foreground set, while the ones that lack the canonical motif constitute the background set.

The sequence identifiers for the sequences used in alignments Figures 2, 5b and 6a are: 94370018|PDZ_GEF-mouse; 93138731|K-channel-plant; 9857982|FixK-bacteria; 6759981|Fnr-bacteria; 15675445|ArcR-bacteria; 17989331|NnR-bacteria; 68552962|CBS-bacteria; 15673985|Flp-bacteria; 56419292|ARSR-bacteria; 1942960|PKA-mouse; 37964177|PKG-seahare; 68076807|PKA-parasite; 76609590|Epac-cattle; 68402320|HCN-zebrafish; 89309052|channel_Tetrahymena; 87198326|Bact_Pyrredox; 22298372|channel_Bact; 76259471|HisK-bacteria; 106879720|AAA_Atpase-bacteria; 462748|NtcA-bacteria; 86610079|ICLR-bacteria; 71367866|GNTR-bacteria; 111225891|CRP-bacteria; 115352640|MARR-bacteria; 116183754|ASNC-bacteria; 1FT9|pdb|cooA-bacteria; 2D93|pdb|PDZ_GEF_human; 2H6B|pdb|CprK-human.