Structural genomics reveals EVE as a new ASCH/PUA-related domain

Summary

We report on several proteins recently solved by structural genomics consortia, in particular by the Northeast Structural Genomics consortium (NESG). The proteins considered in this study differ substantially in their sequences but they share a similar structural core, characterized by a pseudobarrel five-stranded beta sheet. This core corresponds to the PUA domain-like architecture in the SCOP database. By connecting sequence information with structural knowledge, we characterize a new subgroup of these proteins that we propose to be distinctly different from previously described PUA domain-like domains such as PUA proper or ASCH. We refer to these newly defined domains as EVE. Although EVE may have retained the ability of PUA domains to bind RNA, the available experimental and computational data suggests that both the details of its molecular function and its cellular function differ from those of other PUA domain-like domains. This study of EVE and its relatives illustrates how the combination of structure and genomics creates new insights by connecting a cornucopia of structures that map to the same evolutionary potential. Primary sequence information alone would have not been sufficient to reveal these evolutionary links.

Introduction

Structural Genomics has high impact on sequence space

Structural Genomics (SG) is a worldwide effort that has many objectives. One goal is the increase of the structural coverage, or the percentage of proteins for which we have direct experimental or indirect computational information about their three-dimensional (3D) structure. One way of realizing this goal is to identify families for which we currently do not have 3D coverage and that are large, so that each experimental structure yields reliable models for many proteins. Experimentally determining the structure for at least one representative in this sequence-structure family yields high impact ¹. Over the last few years, SG consortia have deposited thousands of structures ^2,3 into the Protein Data Bank (PDB) ⁴. If the aim is to increase structural coverage as much as possible given certain resources, then SG undoubtedly has become the most efficient way of reaching this aim ⁵. The systematic identification of large families with no previous structural characterization leads to a new scenario. While, traditionally, the details of experimental structures have confirmed hypotheses about protein function, now the 3D structures themselves generate functional hypotheses. In other words, the approach shifts from hypothesis-driven to hypothesis-generating science.

Large families are biologically important

One assumption of target selection in structural genomics is that the larger and more diverse a protein family, the more likely its proteins perform essential functions ⁶; this assumption has been proven in hundreds of cases ⁷. Solving the structure for one family-representative may unravel evolutionary relations with a different family, and may thereby provide new insights into the functional traits of the latter as well as into the function of the target protein. Indeed, SG is likely to impact biology more by unraveling relations that are undetectable from sequence analysis alone, than by discovering new exotic folds ⁸. In its targeting of large un-annotated families, SG is clearly orthogonal to traditional structural biology, which focuses more on providing a better understanding of what is partially known than on exploring what is largely unknown.

Structural genomics data reveals a new PUA domain-like domain

Here, we investigated several proteins recently solved by the Northeast Structural Genomics consortium (NESG) ⁹ and other SG consortia. Many of these proteins have levels of sequence similarity that pairwise sequence comparisons would consider as below random. Still, they can be defined by a common structural framework/architecture, namely the PUA domain-like fold in SCOP ¹⁰. These proteins constitute a family with (1) still limited experimental exploration, (2) a great number of sequences, and (3) with a large number of diverse structures. This combination is exactly the target that structural genomics has successfully been addressing over the last eight years. The SCOP fold is named after the pseudouridine synthase and archaeosine transglycosylase (PUA) domain, a highly conserved RNA binding motif ¹¹. PUA is present, among other proteins, in RNA modification enzymes and ribonucleoproteins; in human, it has been related to disease, including cancer ¹². SCOP ¹⁰ classifies PUA domain-like domains (PUA-like domains hereafter) by their conserved 5-stranded core pseudo-barrel beta sheet (strands 1 to 5, Fig. 1A). Many family members additionally feature an alpha helix (helix A) between strands 1 and 2.

Fig. 1. PUA-like domain topologies.

In blue and orange we denote strands and helices respectively. Insertions (INS0, INS1, INS2) characteristic of specific domains are marked in red and additional strands (i.e. strands within insertions, namely 6 and 2′) in lavender. The four panels represent: (A) the basic elements (core) of the PUA-like domains topology as reported in SCOP ¹⁰, (B) the ASCH domain architecture ²¹, (C) the EVE domain architecture, (D) the architecture of the PUA domains described in the text.

We revisited previous studies in light of the new structural information available and provided an in depth-analysis of PUA-like domains. Although we consider information from sequence including alignments and genomic profiles, our study is primarily based on a comparative structural analysis of the domains. We use structural alignments, cavity identification, surface residue conservation and surface electrostatic potential calculations. We show that seven of the proteins with a PUA-like architecture recently solved by SG consortia exhibit features clearly distinct from other PUA domain-like domains. These structural differences predict that they also carry different functional traits. We refer to these newly identified domains as EVE (following the PDB identifier of one of the available structures: 2eve ¹³). Overall, this study illustrates how structural knowledge can be essential not only for evolutionary linking proteins without recognizable sequence similarity but also for highlighting differences between structurally related domains that may reflect more subtle divergences in function and history. Sequence and structural data for the proteins that we analyze are available at http://luna.bioc.columbia.edu/honiglab/PUAlike (see also Methods).

Results

PUA domain-like structures from NESG suggest EVE as a new structural family

Structural comparisons using Skan ^14,15 (Methods) revealed that at least ten of the proteins for which structures have been experimentally determined by the NESG consortium ⁹ included domains with a PUA-like architecture ^16,11 (Table 1) (note that, in this paper, we will refer to individual proteins by preferentially using their PDB identifier). Interestingly, querying Pfam ¹⁷ with our NESG protein sequences produced no match with the PF01472 PUA proper family. Instead, three proteins were assigned to the PF04266 ASCH (Activating Signal Cointegrator 1 Homology) family, four to the PF01878-DUF55 family of unknown function, one (1zbo ¹⁸) to the PF02190 ATP-dependent protease La family, one (1nxz ¹⁹) to the PF04452 RNA methyltransferase family, and one (1t5y ²⁰) could not be assigned to any Pfam-A family. Aspects of function could be inferred by homology for only two out of the ten proteins (ASCH domains have been connected to RNA metabolism ²¹, but more detailed characterizations are yet missing).

Table 1.

Proteins analyzed in this paper ^*

PDBid [NESGid]	Our 3D classification	SCOP family	Pfam family	Experimental method	References
1J2B	PUA	PUA domain	PF01472	X-ray	28
2HVY	PUA	PUA domain	PF01472	X-ray	38
1R3E	PUA	PUA domain	PF01472	X-ray	39
1T5Y (95-170)# [HR2118]	PUA	PUA domain	PB010048 (PFAM-B)	X-ray	20
1NXZ (2-73)# [IR73]	Other (RNA methyltransferase)	YggJ N-terminal domain-like	PF04452	X-ray	19
1G8F(2-168)	Other (ATP sulfurylase N-terminal domain)	ATP sulfurylase N-terminal domain	PF01747	X-ray	63
1XNE# [PfR14]	ASCH (family 321)	ProFAR isomerase associated	PF04266 (DUF)	NMR	23
1S04# [PfR13]	ASCH (family 321)	ProFAR isomerase associated	PF04266 (DUF)	NMR	21,22
1TE7# [ET99]	ASCH (family 621)	yqfB-like	PF04266 (DUF)	NMR	24
1T62	ASCH (family 421)	EF3133	PF04266 (DUF)	X-ray	21,47
2DP9/1WK2	ASCH (family 121)	TTHA0113	PF04266 (DUF)	X-ray	21,46
1ZCE# [AtR33]	EVE	Atu2648/PH1033-like	PF01878 (DUF)	X-ray	25
2EVE# [PsR62]	EVE	Atu2648/PH1033-like	PF01878 (DUF)	X-ray	13
2AR1	EVE	Atu2648/PH1033-like	PF01878 (DUF)	X-ray	29
2G2X# [PpR72]	EVE	Atu2648/PH1033-like	PF01878 (DUF)	X-ray	27
2GBS# [RpR3]	EVE	Atu2648/PH1033-like	PF01878 (DUF)	NMR	26
2HD9/1WMM	EVE	Atu2648/PH1033-like	PF01878 (DUF)	X-ray	31,32
2P5D	EVE	not in SCOP	PF01878 (DUF)	X-ray	30
2YUD	EVE-like	not in SCOP	PF04146	NMR	35
2YU6	EVE-like	not in SCOP	PF04146	NMR	36
1ZBO# [BoR27]	Other (ATP-dependent protease La)	LON domain like	PF02190	X-ray	18

^*PDBid: PDB identifier ⁴; number signs mark structures determined by NESG (with NESG ID in squared parentheses); Our 3D classification: structural classification of the target based on the analysis performed by the authors (see text for more details); family numbers for ASCH domains follow Iyer et al. ²¹; SCOP family: all SCOP ¹⁰ families listed belong to the PUA domain-like superfamily and fold in SCOP; Pfam family: families annotated by Pfam ¹⁷ as belonging to the ASCH/PUA clan in bold; Experimental method: method used for structure determination; References: PDB reference or publications reporting on the experimental structures when available (note: the link to reference ²¹ does not indicate the structural paper but simply that the structures were discussed in ²¹).

Most PUA-like structures from NESG can be grouped into two topological themes of variation of the 5-stranded pseudo-barrel beta sheet core architecture of SCOP. The first theme (Fig. 1B) is a long insertion (INS1) of about 40 residues, featuring a number of short helices, between strands 4 and 5. Proteins carrying this insertion (1s04 ^21,22, 1xne ²³, and 1te7 ²⁴; Table 1) belong to the ASCH family ²¹ (details below). The second theme (Fig. 1C) is a shorter insertion (INS1: 10–20 residues) between strands 4 and 5 coupled with a long (about 40 residues) second insertion (INS2) at the C-terminus (1zce ²⁵, 2eve ¹³, 2gbs ²⁶ and 2g2x ²⁷; Table 1). INS2 contains several helices of variable lengths and an additional strand (strand 6 in Fig. 1C) that pairs in anti-parallel fashion with strand 1 of the core beta sheet. Following the PDB identifier of one of the representatives from this group (2eve ¹³, Table 1), we dubbed this latter topological theme EVE. The addition of a strand that pairs with strand 1 is not unique to EVE: PUA domains (e.g. 1j2b ²⁸, Table 1) also have an extra strand pairing to strand 1 (strand 2′, Fig. 1D). However, topologically this strand is very different from strand 6 in EVE. In fact, since PUA domains lack INS2, strand 2′ is located between strands 2 and 3. Also, although strands 6 (Fig 1C) and 2′ (Fig 1D) seem to pair with strand 1 in a similar way, they have a very different effect on the solvent accessible surface of their respective domains (below).

Among the remaining NESG structures, 1t5y ²⁰ is similar to PUA, while 1nxz ¹⁹ resembles the SCOP-core architecture (Fig. 1A). Protein 1nxz is classified by Pfam as an RNA methyltransferase, due to an additional C-terminal domain carrying this annotation. Finally, 1zbo ¹⁸ features a long (~100 residues) C-terminal insertion unrelated to INS2 and is classified by Pfam as part of the ATP-dependent protease La domain family (Table 1).

EVE domains share structural features and signature conserved residues

Four of the NESG structures are characterized by the topology of 2eve that we named EVE (Fig. 1C: 2eve, 2gbs, 2g2x, 1zce). Structural comparisons with Skan ^14,15 revealed three more proteins with similar topology in the PDB (solved by other SG consortia), namely 2ar1 ²⁹, 2p5d ³⁰ and 2hd9/1wmm (1wmm ³¹ and 2hd9 ³² are two crystal structures of the same protein; Table 1). Structural alignments highlighted a strong sequence/structure relation between these seven single-domain proteins (Table 2 and Fig. 2); a subset of these relations has been analyzed before ²⁹. 2hd9 ³² and 2p5d ³⁰ are relatively sequence distant from the others (<25% pairwise sequence identity in the structural alignment); all align at Evalues <10⁻³⁹ to the same Pfam family (PF01878-DUF55) (data obtained via InterProScan ³³).

Table 2.

Conservation of structure and sequence among EVE domains ^*

RMSD/AL (%ID)	1ZCE	2EVE	2AR1	2G2X	2GBS	2HD9/ 1WMM	2P5D
1ZCE	0.0/146
2EVE	1.5/132 (40%)	0.0/149
2AR1	1.4/135 (45%)	0.8/147 (50%)	0.0/157
2G2X	1.4/132 (42%)	0.6/147 (78%)	1.0/147 (50%)	0.0/149
2GBS	1.3/139 (53%)	1.6/132 (40%)	1.9/136 (40%)	1.5/132 (38%)	0.0/145
2HD9/ 1WMM	2.7/117 (22%)	2.8/121 (20%)	2.6/123 (19%)	2.8/120 (20%)	2.9/115 (22%)	0.0/143
2P5D	3.0/116 (21%)	3.1/122 (19%)	2.6/123 (23%)	3.1/119 (18%)	2.9/115 (22%)	1.4/138 (59%)	0.0/145

^*First row and first column: PDB identifiers of EVE domains. Other rows and columns: the first number represents the RMSD in (in Angstroem) between EVE domain pairs as obtained by means of the 3D alignment program Skan ^14,15; the second number is the length of the alignment; the third number in parentheses is the percent sequence identity between the two domains as derived from the structural alignment (note: the diagonals show the length of the domains that varies from 143 to 157 residues). For instance, 1zce and 2eve have 1.5 Angstroem RMSD over 132 structurally aligned residues and 40% sequence identity. Cell colors reflect RMSD values, ranging from low=similar (dark) to high=dissimilar (light), and according to the following intervals (in Angstroem): 0.0–1.0 (dark), 1.1–1.5, 1.6–2.0, more than 2.0 (light).

Fig. 2. Structure-based sequence alignment of 2eve ¹³, 2ar1 ²⁹, 2g2x ²⁷, 1zce ²⁵, 2gbs ²⁶, 2hd9/1wmm ^32-31 and 2p5d ³⁰ (EVE domains).

The sequence alignments are obtained automatically from pairwise structural alignments between 2eve and the other proteins (Table 2). Each protein is represented by two lines, one giving the sequence, the other the secondary structure assignment (according to Skan ^14,15). Numbers at each end indicate the first and last residue found in the structure, following the PDB ⁴ sequence numbering for each protein. Below the alignment we highlight the topology of the EVE domains (with strands 1–6, helix A and the two insertions INS1 and INS2). Boxes mark residues that are generally not only conserved in the structural alignment but also in sequence-based alignments from PSI-BLAST ⁵¹ and CLUSTALW ⁵² or Pfam ¹⁷. Conserved residues and numbers reported above the boxes are as in the 2eve PDB file.

Sequence conservation analysis (Methods) identified several conserved aromatic residues at the N-terminus of EVE; these included Y3, W4 and W26 (Fig. 2; sequence and numbering as in the 2eve ¹³ PDB file). Note that the sequence-based Pfam seed alignment (PF01878-DUF55) of the EVE domain N-terminus is incorrect (Supplementary Online Material). While W4 is mostly conserved, Y3 is often substituted by other aromatic residues such as phenylalanine and histidine. This appears to reflect the biophysical features of Y3 as well as its characteristic interactions in EVE: in all known structures, Y3 seems to form a π-π bond with W142, another well-conserved aromatic residue located in the C-terminal insertion. W142 variants include phenylalanine or tyrosine. Highly conserved W4, on the other hand, interacts with highly conserved D45 at the beginning of strand 3. Hence, Y3 and W4, interacting with sequence-distant residues, seem to be important for stabilizing the domain structure. In contrast, the strong conservation of W26 may have functional implications. W26 extends its bulky side chain into the domain’s main surface cavity, which is lined by helix A, strands 1, 2, and 6, and has an overall surface area of about 200 Ų (Fig. 3A). In three of the six available crystal structures this cavity is partially occupied by a ligand. Note that the structure of 2gbs is from NMR data, i.e. data from which ligand identification is difficult. The observed ligands include a molecule of 3[N-morpholino]propane sulfonic acid (MPO) in 2eve ¹³ (Fig. 3A and 3B) and a molecule of glycerol (GOL) in both 2ar1 ²⁹ and 2hd9 ³². In all three instances, W26 appears to play a major role in shaping the binding pocket. Neither MPO nor GOL, however, are functional ligands. Their presence in the structures is related to their use as buffer or cryoprotectant ²⁹ substances. Note that MPO should not be confused with morpholino oligonucleotides, which are nucleotide analogs that have become an important knockdown tool in developmental biology ³⁴. Indeed, no nucleobase is present in the MPO molecule bound to 2eve.

Fig. 3. Ligand binding and surface conservation in the main surface cavity of EVE domains.

(A) 2eve ¹³ structure in cartoon representation (green) with co-crystallized ligands 3[N-morpholino]propane sulfonic acid (MPO) and tris-hydroxymethyl-methyl-ammonium (TRIS) in ball and stick representation (CPK colors). In red (licorice representation) we show conserved residues (Y3, W4, F13, W26, V29, D45, Y50, W142). (B) 2eve molecular surface colored by residue conservation score (as calculated by ConSurf ⁵³), conserved residues are in purple, variable residues are in cyan, the orientation of 2eve is as in (A). MPO and TRIS are in ball and stick representation (CPK colors).

Among the residues lining the cavity, V29 and F13 are also conserved (Fig. 3A); however, F13 is less conserved than W26 and V29 (in particular, it is mutated into non-aromatics in 2hd9 ³² and 2g2x ²⁷). Overall, the region around the cavity appears to be well conserved in sequence (Fig. 3B). An analysis of the electrostatic potential of the cavity by GRASP2 ¹⁴ noted the shift from a slightly negatively charged surface potential in 2eve and 2ar1 to a slightly positive potential in 2hd9 (data not shown).

Another interesting pattern of conservation is constituted by the previously mentioned D45, along with G44 and Y50, which are all highly conserved in the EVE family (Fig. 2). While G44 and D45 are also conserved in other PUA-like domains (see below), the conservation of Y50 seems to be unique to EVE domains. Y50 is at the center of a crevice lined by strands 1, 3, 4, and INS2, which can be seen as a continuation of the cavity containing F13, W26 and V29 (Fig. 3A and 3B). In 2eve, this additional cavity hosts a small non-functional ligand, a molecule of tris-hydroxymethyl-methyl-ammonium that interacts with Y50 and with a series of positively charged residues (K7, K115, and R131; mostly non-conserved).

YTH domains: EVE-like structure of a putative splicing factor in human

Structural alignments ^14,15 revealed two additional EVE-like structures in the PDB, namely the human proteins 2yud ³⁵ and 2yu6 ³⁶ (Table 1) belonging to the putative splicing factor YT521-B homology (YTH) family (PF04146) ³⁷. Both 2yud and 2yu6 feature the characteristic C-terminal insertion INS2 including strand 6, while they differ substantially in sequence from EVE domains with maximally 19% pairwise sequence identity (between 2yud and 2hd9; Table S1, Supplementary Online Material). Structural alignments to EVE (Fig. S1, Supplementary Online Material) and multiple sequence alignments of the family ³⁷ showed that YTH domains conserve the N-terminal pair of aromatic residues Y3W4 (YF in 2yu6 and FF in 2yud) and the cavity-lining W26 typical of EVE. Y50 is instead mutated into a highly conserved serine. The main surface cavity of EVE is also present in YTH and it is of comparable size. On the other hand, EVE-conserved residues V29, G44, D45 and W142 (Fig. S1, Supplementary Online Material) are not found in YTH and YTH domains feature several well-conserved residues not present in EVE (including several aromatic residues ³⁷). Also, while EVE domains are well represented in prokaryotes, YTH domains are only found in eukaryotes. Differences notwithstanding, YTH domains appear to be as the most similar to EVE among all other PUA-like domains currently in the PDB.

EVE differ from ASCH and PUA domains in important ways

Iyer et al. ²¹ identified ten families (including ASC-1) within what they defined as the ASCH superfamily (Fig. 1B). Experimental structures are now known for five ASCH domains. These five single-domain proteins vary substantially in sequence; they map to four of the ten families described by Iyer et al. ²¹ (Table 1). In Pfam, ASCH domains have recently been reclassified into a single family (ASCH-PF04266) of unknown function. We have already described the main differences between the EVE and ASCH structural themes. Sequence-wise, ASCH domains have an N-terminal sequence signature ²¹ that is conserved across all ten families and includes a glycine and a lysine (G19-x-K21; numbers as in the 1xne ²³ PDB file). In most ASCH domains, two conserved, adjacent residues (E24 and R26) contribute to the longer G-x-K-x-x-[ETS]-x-R motif. This motif is in the “loop” region connecting helix A and strand 2. Notably, EVE domains do not conserve any of these residues (Fig. 2 and Fig. 4). Similarly, most of the EVE signatures discussed previously are not conserved in ASCH domains, specifically: residues Y3 and W4 at the N-terminus, W26 and V29 in the main surface cavity, Y50 in the second EVE cavity. Finally, residue W142 in INS2 is also not conserved in ASCH, as INS2 is completely absent from these domains. The largest surface cavity in ASCH (>300 Å ²) roughly corresponds to the main cavity in EVE, although the latter is ~30% smaller due to the extra strand 6. Residue conservation within the cavity is markedly different in ASCH and EVE. In ASCH, most of the conserved residues protruding into the cavity are polar, acidic or basic, in particular: E24, R26 and in at least some cases K21 (e.g. in 1xne). E24 and R26 structurally align to hydrophobic conserved residues W26 and V29 in EVE (Fig. 4), giving a completely different ‘flavor’ to the ASCH cavity. An exception is constituted by aromatic residue Y12, often mutated into a phenylalanine or a tryptophan (but a leucine in ASCH family 4). It corresponds to F13 in EVE. Note that this conserved position in ASCH domains was not identified by Iyer et al ²¹ (due to their use of sequence-based alignments) and that one of the tRNA binding residues of PUA domain 1j2b, F519, is found in a similar structural position (see below). Although the hypothesis awaits experimental confirmation, the largest surface cavity of ASCH and EVE, with its size, conservation and capability of accommodating small ligands (Fig. 3A and 3B), is likely to play an important functional role. The significant differences observed between the conserved residues lining the cavity in ASCH and EVE, along with the additional presence of a nearby cavity surrounding highly conserved Y50 in EVE, seem hence to point to important differences in the molecular function of the two domains.

Fig. 4. Structure-based sequence alignment between ASCH and EVE domains.

Notations are as in Fig. 2. The alignment is obtained from pairwise structural alignments between 1s04 ^21,22 and the other proteins, except for 2g2x ²⁷, whose sequence was manually added taking its pairwise structural alignment to 2ar1 ²⁹ as a template (since Skan failed to identify 2g2x as a structural homolog of 1s04). For simplicity, alignments are reported only up to INS1, i.e. the insertion between strand 4 and 5. ASCH sequences are on top and EVE at the bottom. Numbers at each end of a sequence indicate the first and last residues following the PDB ⁴ sequence numbering for each protein. As in Fig. 3, boxes highlight conserved residues generally indicating not only conservation in the structural alignment but also in sequence-based alignments obtained using PSI-BLAST ⁵¹ and CLUSTALW ⁵² or Pfam ¹⁷. Type and numbers for these residues are as in the PDB sequence of 2eve ¹³. Note that in this alignment W26 is slightly misaligned in the EVE domains (compare to Fig. 2) and so is the ASCH motif ²¹ GxKxxE/T/SxR in the ASCH domains.

Most of the conserved residues characterizing EVE domains and analyzed above (i.e. Y3, W4, W26, V29, Y50 and W142) are not conserved in the PUA domains. Although some PUA domains feature a phenylalanine at the end of strand 2 (F527 in 1j2b, see Fig. 2 in ¹¹), structural alignments show that it does not correspond to W26 in EVE (Fig. S2, Supplementary Online Material). The PUA cavity corresponding to the main EVE (and ASCH) cavity is very small (surface area of < 100 Å ²). This is due to the presence of strand 2′ (Fig. 2D), which contrary to the shorter strand 6 of EVE fully extends into the core of the cavity (Fig. S2, Supplementary Online Material). Indeed, a simple structural superposition by Skan ^14,15 indicated that there is no space to dock the ligands bound to EVE (specifically, MPO in 2eve ¹³ and GOL in 2ar1 ²⁹ and 2hd9 ³²) into the PUA domain structures of 1j2b ²⁸, 2hvy ³⁸ and 1r3e ³⁹. These differences may lead to important functional consequences as we discuss in the next paragraph.

EVE structure compatible with RNA binding but RNA binding, if any, likely to be different than in PUA

PUA domains are known to interact with RNA as part of multi-domain RNA modifying proteins. Although no EVE domain appears to be fused to any known RNA modifying protein, EVE domains could in principle act in concert with other proteins to bind and modify RNA. To further investigate this hypothesis, we compared the regions involved in RNA binding in PUA with the corresponding regions in EVE.

Four PUA-RNA complexes are available, including archaeosine tRNA-guanine transglycosylase ArcTGT (PDB identifier 1j2b ²⁸) and pseudouridine synthases Cbf5 and TruB (PDB identifiers 2hvy ³⁸, 1r3e ³⁹ and 1k8w ⁴⁰). Although RNA binding in known PUA-RNA complexes presents important differences, some common elements have been identified ¹¹. In particular, three structural elements seem to always participate in RNA recognition: helix A, strand 2 and strand 5 (note, naming and numbering used in Perez-Arellano et al. ¹¹ differ from ours). While strand 5 and the loopy region connecting helix A and strand 2 bind to the double stranded RNA stems, a cleft between helix A and strand 2 provides recognition for single stranded RNA overhangs. As detailed above, in EVE and ASCH, this shallow cleft is part of a much larger cavity. Double stranded RNA binding residues include polar and charged residues found in the loop connecting helix A and strand 2 and on strand 5. Single stranded RNA overhangs are instead anchored to the PUA domain differently in the ArcTGT-RNA and in the Cbf5-RNA complex: via a cluster of phenylalanines in ArcTGT (F519, F527, F530) and via a C-terminal tryptophan (W337) and hydrogen bonds with the protein backbone in Cbf5. Note that no single stranded RNA binds the PUA domain in TruB structures 1r3e ³⁹ and 1k8w ⁴⁰.

Structural alignments of ArcTGT PUA with EVE revealed that EVE domains expose several positively charged residues at the end of helix A and on strand 4 and 5 that may be able to mediate recognition of double stranded RNAs (Fig. S3–S5, Supplementary Online Material). To better investigate the putative binding of single stranded RNA overhangs to EVE domains, we performed the following simple experiment. We first superimposed the structures of 2eve ¹³ and of the PUA domain of ArcTGT 1j2b ²⁸ (residues 507–582) (using GRASP2 ¹⁴). Then, we removed 1j2b while leaving its bound tRNA and 2eve in place (Fig. 5A and 5B). The resulting virtual complex provides us with two important indications. First, only one of the three phenylalanines that bind RNA in the PUA domain is conserved in EVE (Fig. 5A and S2, Supplementary Online Material) (the only conserved residue in EVE is F13, sometimes mutated into non-aromatic hydrophobic residues; in ASCH it corresponds to conserved residue Y12). Second, although the 2eve region where the single stranded RNA overhang is found in the virtual complex is part of the previously characterized large EVE cavity, it does not correspond to the highly conserved pocket where MPO (and GOL) is found (Fig. 5A and 5B). Neither it corresponds to the small cavity into which conserved residue Y50 protrudes (Fig. 5A). So, hypothesizing that indeed EVE binds RNA, it is likely that its large cavity either hosts a cofactor (in place of MPO and GOL) or allows single stranded RNAs to bind more deeply into the domain (with W26 possibly playing a major role in binding). Another possibility is that single stranded RNAs anchor themselves to EVE by binding on the opposite side of the cavity with respect to what we see in PUA domains. In conclusion, our analysis points to important differences between EVE and PUA in a functionally relevant region of the PUA domain.

Fig. 5. Virtual complex of 2eve and tRNA from 1j2b.

The figure was obtained according to the following procedure: first, 2eve ¹⁹ was superimposed on the PUA domain of ArcTGT 1j2b complexed with tRNA ²⁸ (residues 507–582) (using GRASP2 ¹⁴); then, the 1j2b PUA domain structure was removed while its bound tRNA and 2eve and its ligand (MPO) was kept in place. (A) 2eve structure is cartoon representation (blue), tRNA (CPK colors), MPO (green) are in ball and stick representation. Relevant residues (see text) are in licorice representation (blue). (B) 2eve molecular surface colored by residue conservation score (as calculated by ConSurf ⁵³), conserved residues are in purple, variable residues are in cyan, the orientation of 2eve is as in (A). tRNA (CPK colors), MPO (green) are in ball and stick representation.

Another indirect evidence supporting an RNA binding role for EVE comes from the EVE-like eukaryotic YTH domains. Stoilov et al. ³⁷ suggested that YTH (2yud ³⁵ and 2yu6 ³⁶) could act as mRNA-binding domains, taking part in alternative splicing. In multi-domain proteins, YTH can be found fused to zinc fingers, RRM motifs, and HA2 domains, all of which are thought or known to be involved in nucleic acid binding. Although a role in alternative splicing for EVE domains can be excluded given their extensive presence in prokaryotes (see below), it is tempting to postulate that they could still perform a function similar to YTH at the molecular level.

Genomics and phylogeny of EVE domains

PUA domains almost always occur in multi-domain proteins, often in fusion with known RNA modifying enzymes. ASCH domains, although mostly found as single-domain proteins, were originally identified at the C-terminus of ASC-1 proteins that in vertebrates are known to physically interact with proteins involved in RNA processing ²¹. Different from PUA and ASCH, EVE domains seem to exist almost exclusively as single-domain proteins. Exceptions are proteins from Staphylococcus aureus and Staphylococcus saprophyticus, in which EVE is flanked by N- and C-terminal regions that have not yet been functionally characterized. In Eukarya, most proteins feature an insertion N-terminal to EVE that may or may not constitute a separate domain. In fact, the only examples of annotated domains associated to EVE are found in Aspergillus fumigatus, Aspergillus oryzae and Aspergillus nidulans. In these genomes, proteins that contain the EVE domain carry three copies of the AT-hook motif at the N-terminus. This domain is typically found in proteins that bind preferentially to the minor groove of AT-rich regions in double-stranded DNA ^41,42. Finally, several eukaryotic EVE domains (including one in Homo sapiens protein Q9HC20_HUMAN) are annotated as thymocyte proteins (e.g. in UniProt ⁴³), where thymocytes are the T-cell precursors that mature in the thymus.

EVE, ASCH and PUA domains occur in all three super-kingdoms of life (Eukaryota, Bacteria, and Archaea). Although the analysis of EVE sequences included in the PF01878-DUF55 Pfam family reveals a predominant presence of EVE domains in Bacteria, EVE are further present in both the Crenarchaeota and Euryarchaeota phyla of Archaea and in Eukaryota, including representatives in Euglenozoa, Metazoa, Fungi and Viridiplantae. In Eukaryota, however, EVE domains are relatively rare. This may suggest that the acquisition of the EVE domain by eukaryotic organisms has occurred via lateral gene transfers from Bacteria. The phyletic distribution of YTH (PF04146), exclusively found in Eukaryota, is very similar to that one of EVE (PF01878-DUF55). This observation, along with the discussed similarities in sequence and structure, may indicate that YTH originated from EVE domains following gene duplication in Eukaryota. ASCH domains (PF04266) also have a phyletic pattern similar to EVE and ASCH although, contrary to the other two, they seem to be practically absent in Fungi.

Discussion

EVE: a novel PUA-like domain

We characterized a new PUA-like domain, that we named EVE (Fig. 1C). Structure comparison revealed problems in alignments based on sequence alone (Supplementary Online Material), and showed that EVE domains are related but importantly distinct from ASCH and PUA domains (Table 3). The evidence includes the following observations: (i) differences in topology (Fig. 1B, 1C and 1D), in particular the characteristic EVE C-terminal insertion (INS2); (ii) residue conservation, with Y3, W4, W26, V29, Y50 and W142 that appear to be conserved only in EVE and not in other PUA-like domains; (iii) surface cavity properties, including cavities’ size and residue conservation within them; in EVE the main surface cavity is lined by three conserved hydrophobic residues, two of which (W26 and V29) structurally align to conserved E24 and R26 in ASCH. In PUA domains the same cavity is partially occluded by the presence of an additional strand (strand 2′, Fig. 1D).

Table 3.

Summary of differences between EVE, ASCH and PUA domains *

	EVE	ASCH	PUA (1j2b, 2vhy, 1re3)
Topology	Add* strand 6 paired to strand 1; add INS1+INS2 (Fig. 1C)	Add INS1 (Fig 1B)	Add strand 2′ paired to strand 1; add INS1+INS0 (Fig 1D)
Conserved residues	Y3, W4, F13, W26, V29, G44, D45, W142 (numbers as in 2eve)	G19-x-K21-x-x-E24-x-R26 (numbers as in 1xne) Y12 (aligned to F13 of EVE), G38 (aligned to G44), D39 (aligned to D45)	Depend on PUA domain
Cavity size	Lined by helix A, strand 1,2 and 6. Surface area ~ 200 Å2	Lined by helix A, strand 1 and strand 2. Surface area ~ 300 Å2	Lined by helix A and strand 2. Surface area < 100 Å 2
Residues lining the cavity	Mainly hydrophobic (F13, W26, V29)	Mainly charged and polar residues (Y12, E24 and R26)	Presence of a positive patch although not completely conserved throughout the family
Sequence identity (alignment length)	EVE/ASCH: 18%, (77) EVE/PUA: 9%, (66)	ASCH/EVE: 18%, (77) ASCH/PUA:10%, (68)	PUA/EVE:9%, (66) PUA/ASCH:10%, (68)
RMSD (alignment length)	EVE/ASCH: 3.4 Å, (84) EVE/PUA: 3.3 Å, (66)	ASCH/EVE:3.4 Å, (84) ASCH/PUA:3.7 Å, (68)	PUA/EVE:3.3 Å, (66) PUA/ASCH:3.7 Å, (68)

^*‘Add’ refers to structural changes with respect to the SCOP-fold core topology (Fig. 1A). Sequence identities in row 7 are the highest among any pair of domains belonging to the two structural themes indicated. For example, EVE/ASCH 18% sequence identity means that 18% is the maximum sequence identity (according to Skan structure-based sequence alignments) between any two EVE and ASCH domains. Similarly, 3.4 EVE/ASCH RMSD is the minimum RMSD between any two EVE and ASCH domains. Numbers in parentheses represent alignment lengths.

While PUA and ASCH domains are sometimes found as part of multi-domain proteins in which one domain is an RNA modifying enzyme, we were not able to identify any such case for EVE. Analysis of PUA-RNA complexes revealed that EVE sequence and structural features are compatible with RNA binding. The comparison, however, also points to important differences between the two domains. Conserved EVE cavities, in particular, much reduced or absent in PUA, are likely to be relevant for function. This suggests that RNA binding in EVE, if any, may be very different from the one observed in PUA. Another hint for a possible RNA binding role of EVE comes from eukaryotic YTH domains, which have basically the same topology as EVE (albeit some interestingly different conservation patterns). In fact, YTH domains have been suggested to bind mRNA. Overall, the exact functional role of EVE domains remains unclear and awaits experimental investigation.

EVE and other PUA-like domains are evolutionary related

Although our evidence suggested important functional differences between EVE and other PUA-like domains, the strong conservation of a structural core between all PUA-like domains suggests an evolutionary connection. The GD sequence motif found in several families (including EVE, ASCH and PUA but not YTH) may be a surviving legacy of this common ancestry. The glycine (G at the inception of strand 3, Fig. 2) may provide the protein backbone with the flexibility needed for accommodating the insertions found immediately before this strand (e.g. strand 2′, Fig. 1D). Thereby, it may help preventing the disruption of the core structure. The aspartate (D) often interacts with residues on strand 1, likely increasing the stability of the pairing between strands 1 and 3 and hence the stability of the whole domain.

Considering that EVE, PUA and ASCH domains all span the three super-kingdoms, it is possible that PUA-like domains all originated from an early common ancestor. Later they may have diverged functionally, giving rise to several functional families.

Suggested alterations to Pfam

Pfam groups families that are likely to have arisen from a single evolutionary origin ⁴⁴ into clans. Several PUA-like domains are included into the PUA (CL0178) clan, in particular: PUA (PF01472), ASCH (PF04266), PF02594-DUF167 and PF03657-UPF0113 (Table 1). Our results seem to clearly indicate that EVE (PF01878-DUF55) and YTH (PF04146) domains should be added to this clan. On the other hand, structure analysis of family PF02594 (dubbed Uncharacterized ACR, YggU family and including PDB proteins 1jrm, 1n91 and 1yh5) indicates that the PUA-like core architecture (Fig. 1A) is not conserved in these domains. This suggests that PF02594 may not have the same evolutionary origin as the other families in the PUA clan, and that it should therefore be removed from the clan. Other Pfam families that could possibly be included in the clan are PF09157 (Pseudouridine synthase II TruB, C-terminal, see 1k8w ⁴⁰), PF09142 (tRNA Pseudouridine synthase II, C terminal, see 1sgv ⁴⁵) and PF02190 (ATP-dependent protease La (LON) domain, see 1zbo ¹⁸).

More PUA-like targets for structural genomics

What are the most interesting PUA-like domains still lacking a structural representative in the PDB? Among EVE domains, it would be interesting to target proteins in the PF01878-DUF55 Pfam family seed alignment such as YOUG_BACLI and Y3206_RHILO. These proteins feature, at the N-terminus, two conserved histidines and in some instances one cysteine. Given their position along the sequence, these residues could protrude into the main surface cavity of the domain, thus potentially affecting electrostatic potential and putative ligand binding properties within the cavity. Among ASCH domains, the ones that stand out are those found in the ASC-1 family. Functionally, they are part of transcriptional coactivator proteins while, structurally, they feature an about 30 residue-long insertion between strand 3 and strand 4, not found in any of the available structures that we analyzed.

Conclusions

Structural genomics consortia continuously deposit high-resolution structural data for a great number of proteins into the public domain. Most of these proteins are outstandingly important as revealed by their evolutionary conservation. Surprisingly often we have no functional annotations from sequence alone for many of these important discoveries. Adding so many new structures is an important achievement per se, irrespective of what is known today about the function of these proteins. Sequences from genome sequencing projects are an obvious analogy. The simultaneous availability of both a wealth of sequence and structural data for evolutionary related proteins allows for detailed investigations of functional similarities and differences. Sequence analysis is in this a first step, but often it requires structures to generate hypotheses that will then drive further experimental studies.

Here, we applied comparative sequence-structure analysis to characterize a novel domain that we called EVE. Taking advantage of several structures within the PUA domain-like SCOP-fold solved by different Structural Genomics consortia, we showed that differences in topology, residue conservation and surface cavities clearly separate EVE from other PUA-like domains (e.g. PUA and ASCH). While the molecular function of EVE is most likely RNA-binding, we predict that EVE will have functional roles not shared by other PUA-like domains. Further functional characterization of EVE domains will have to await systematic experimental studies.

Methods

General approach to sequence and structural analysis of PUA-like domains

On our main web page http://luna.bioc.columbia.edu/honiglab/PUAlike, we show the PDB identifiers of PUA-like domains separated into 5 groups: EVE, EVE-like, ASCH, PUA, and others. For EVE (1zce ²⁵, 2eve ¹³, 2gbs ²⁶, 2g2x ²⁷, 2ar1 ²⁹, 2p5d ³⁰, 2hd9 ³²/1wmm ³¹), EVE-like (2yud ³⁵, 2yu6 ³⁶) and ASCH (1s04 ^21,22, 1xne ²³, 1te7 ²⁴, 1wk2 ⁴⁶, 1t62 ⁴⁷) we included all such domains that we could find in the PDB. For the other two groups, we only included a few representative domains (i.e., those that are mentioned throughout the paper; for PUA: 1j2b ²⁸, 2hvy ³⁸ and 1r3e ³⁹) (Table 1). PDB identifiers for all these proteins are linked to the pages containing the sequence and structure analysis results. Our database is accessible to any JavaScript capable browser. For protein structure visualization Java has to be enabled. Protein structures are visualized using the AstexViewer 2.0 ⁴⁸, which allows for interactive manipulation.

We submitted the sequence and structure of each of the proteins found on our main web page to several servers. Structure comparison methods (Skan ^14,15 and Dali ⁴⁹) identified proteins with similar structures. The protein surface was scanned for cavities using SCREEN ⁵⁰. In parallel, the target sequence was PSI-blasted ⁵¹ against UniProt ⁴³ to identify related sequences and submitted to InterProScan ³³ to provide links to databases such as Pfam ^17,44. Proteins with related sequences were then aligned using ClustalW ⁵²; the resulting multiple sequence alignment was submitted to ConSurf ⁵³ assigning a conservation score to each residue in the target. All these data were then used for the manual analysis of function presented in this paper. All details on how the servers and programs were run and on how data were collected and processed follow.

Sequence analysis

Sequence conservation was assessed by three different means: (1) structure-based sequence alignments, i.e. from 3D superimpositions, for all available 3D structures (Skan ^14,15; Dali ⁴⁹; details below), (2) ClustalW ⁵⁴ alignments with two versions of generating the input: (i) sequences retrieved by PSI-BLAST ⁵¹ using as query the sequence each protein of known structure in our dataset, and (ii) sequences from the Pfam-A family associated with each protein of known structure in our dataset, and (3) the original Pfam family alignments ¹⁷. For the domains within the ASCH group, we also considered the alignments of ten distinct ASCH families reported by Iyer et al. ²¹ (in that paper, these correspond to families one to nine plus ASC-1).

For each protein in our dataset (Table 1), we ran three iterations of PSI-BLAST ⁵¹ against UniProt ⁴³ (release 9.0) and using the E-value cutoffs 10⁻¹, 10⁻³, and 10⁻¹⁰ for the final iteration. Redundant sequences were removed by filtering with CD-HIT ⁵⁵ at 80% (E-values 10⁻¹ and 10⁻³) and at 90% (E-value 10⁻¹⁰) pairwise sequence identity. The remaining sequences were aligned by ClustalW 1.83 ⁵² using default parameters (gap open: 10, gap extension: 0.2, matrix: Gonnet ⁵⁶). In addition to the PSI-BLAST based sequence alignment that we just described, we produced two alternative multiple sequence alignments: one by aligning the target to the Pfam ^17,44 seeds, the other by aligning the target to the full Pfam family if available (i.e. the seed list expanded by using HMMer ⁵⁷), using the profile alignment method implemented in CLUSTALW 1.83 ⁵².

We assigned each protein to a particular Pfam ^17,44 family, by submitting its sequence to InterPro ⁵⁸ using InterProScan ³³. Interpro runs the protein sequence against every Pfam family HMM model (using HMMer ⁵⁷), returning a Pfam family assignment for that protein if a match is found. All ASCH and all EVE domains here analyzed (Results) were assigned with high reliability to a Pfam family (E-values <10⁻¹⁰). We used Pfam to obtain several different types of information: 1) to extract sequences later aligned with CLUSTALW ⁵²; 2) to extract the original Pfam alignments, which might differ from the automatic CLUSTALW alignments that use the same sequences since they are manually adjusted; 3) to analyze the distribution of the members of a Pfam family in Eukarya, Bacteria, and Archaea; and, finally, 4) to study gene fusion events between members of a given Pfam family and other domains. All this information is directly available from the Pfam website.

Structure analysis

We used the two structural alignment programs Skan ^14,15 and Dali ⁴⁹ to identify proteins with structures similar to those of the targets. All hits (proposed structural relations) were ranked by a variety of scores: 1) the internal scores of the 3D alignment method (PSD and Z-score for Skan and Dali, respectively); 2) the RMSD; 3) the alignment length; 4) the percentage of pairwise sequence identity; and (5) the structure alignment score (SAS) ⁵⁹, which combines alignment length and RMSD. We used VISTAL ⁶⁰ for pairwise structure superimposition according to Skan alignments and to visualize aligned secondary structure elements. To identify residues that may play a key role in protein function, each multiple sequence alignment was input to ConSurf ⁵³. ConSurf’s scores were then mapped onto the targets for visual inspection. Solvent accessible cavities on the protein surface were identified by SCREEN ⁵⁰. Cavities were visualized by displaying their molecular surface and with the option of mapping the ConSurf scores ⁵³ onto them.

The electrostatic surface potentials were calculated using Delphi ⁶¹ as implemented in Grasp2 ¹⁴ (with internal and external dielectric constants set to 1.0 and 80.0, respectively; constant salt concentration at 0.2 M and the probe radius equal to 1.4 Å). We used charmm22 parameters ⁶² for atomic charges and atomic radii.

Genomic analysis

The genomic analysis of the PUA-like domains was performed using all sequences found in the corresponding Pfam families as a reference (e.g. family PF01878-DUF55 for EVE). Hence, hypotheses about the evolutionary history of the domains are based on the assumption that the sequences found in Pfam are representative of the domains’ actual genomic distribution.

Abbreviations used

ASCH (domain)

Activating Signal Cointegrator 1 Homology

DUF

Domains of Unknown Function

EVE (domain)

a newly characterized PUA-like domain

GOL

glycerol

MPO

3[N-morpholino]propane sulfonic acid

NESG

Northeast Structural Genomics

PDB

Proteins Data Bank

PUA (domain)

Pseudouridine synthase and Archaeosine transglycosylase

PUA-like domains

PUA domain-like domains

RMSD

Root Mean Square Displacement

SAS

Structure Alignment Score

SG

Structural Genomics

YTH (domain)

YT521-B Homology