Exon structure conservation despite low sequence similarity: a relic of dramatic events in evolution?

Abstract

The evolutionary significance of introns remains a mystery. The current availability of several complete eukaryotic genomes permits new studies to probe the possible function of these peculiar genomic features. Here we investigate the degree to which gene structure (intron position, phase and length) is conserved between homologous protein domains. We find that for certain extracellular-signalling and nuclear domains, gene structures are similar even when pro tein sequence similarity is low or not significant and sequences can only be aligned with a knowledge of protein tertiary structure. In contrast, other domains, including most intracellular signalling modules, show little gene structure conservation. Intriguingly, many domains with conserved gene structures, such as cytokines, are involved in similar biological processes, such as the immune response. This suggests that gene structure conservation may be a record of key events in evolution, such as the origin of the vertebrate immune system or the duplication of nuclear receptors in nematodes. The results suggest ways to detect new and potentially very remote homologues, and to construct phylogenies for proteins with limited sequence similarity.

Introduction

A major feature of eukaryotic genomes is that the DNA coding for proteins is frequently interrupted by introns. Splicing machinery in cells removes introns from mRNA (thus joining together exons) before it is translated into protein. Why organisms have evolved such a seemingly convoluted system for transmitting information between the nucleus and the rest of the cell has been the subject of much debate during the past three decades. Most theories, including the ‘introns-early’ (Gilbert, 1986) and ‘introns-late’ (Orgel and Crick, 1980; Cavalier-Smith, 1985; Sharp, 1985; Stoltzfus et al., 1994) theories, argue for a role of exon shuffling in the evolution of proteins. For example, studies of intron positions within modular extracellular proteins have suggested that exon shuffling played a key role in the evolution of complex multi-domain proteins first used for cell–cell communication within multicellular animals (Patthy, 1994, 1999). However, until recently the relative paucity of genomic sequence information meant that most studies were limited to a handful of protein families.

The current abundance of genomic sequence data (The C.elegans sequencing consortium, 1998; Adams et al., 2000; The Arabidopsis Initiative, 2000; Lander et al., 2001) permits large-scale studies on the relationship between introns and their associated proteins. Here we assess the degree to which exon/intron (gene) structure, defined by the position, phase and length of introns, is conserved within diverse protein domains. Like previous studies on single proteins (e.g. Craik et al., 1981, 1983; Rokas et al., 1999; Venkatesh et al., 1999), we mapped genomic DNA sequence onto protein alignments. We took protein sequence alignment data from the SMART database (Schultz et al., 2000), which contains a diverse set of carefully aligned protein domains from all species and with diverse functions and cellular locations. We compared each protein sequence with genomic sequences and merged gene structure data with the aligned protein sequences. We devised a simple measure of gene structure similarity and sought groups of proteins with conserved gene structures at low sequence similarities. We discuss several examples where gene structures are conserved at low sequence similarity, and possible reasons for this strange phenomenon. We also describe how conserved gene structures could aid phylogenetic reconstruction and the detection of new members of diverse protein superfamilies.

Results

Figure 1 shows the distribution of gene structure similarities for all homologous pairs of protein domains with sequence identities of <40%. Although the majority of pairs show little similarity in gene structure, there are clearly peaks to the right of the figure indicating high gene structure similarities, particularly for nuclear and extracellular signalling domains (broad classifications from the SMART database; Schultz et al., 2000).

Fig. 1. Distribution of gene structure similarities for pairs of proteins with <40% sequence identity. Protein pairs are grouped according to the broad classifications for cellular location given in the SMART database: Extracellular signalling, Intracellular signalling, Nuclear, Others. Contributions to the frequencies were weighted according to the number of pairs from each family. This was to avoid large families (e.g. MATH domains with 4031 pairs) obscuring the plot.

Table I shows that several domains contain large clusters (sets of sequences from a larger alignment) with gene structure similarity despite low sequence identity, and for two (marked with +), gene structures are still similar even though protein sequences can only be accurately aligned with a knowledge of protein tertiary structure (see Materials and methods). These clusters come frequently from extracellular signalling and nuclear domains, though only rarely from intracellular signalling or other domains. Although the location and phase of introns is conserved for the clusters shown in Table I, intron length is highly variable (from <10 to >1000 nucleotides), even at comparatively high sequence similarities (results not shown).

Table I. Protein domains with similar gene structures.

Domain	Np	Ng	Taxg	N2+	Nc	%Imin	%Iave	Taxc	Introns
Extracellular signalling (162)	10752	2396		964
Meprin and traf homology domains	116	98	V F M	93	77	6	26	At Ce	2.	•
					7	11	18	At Ce	21	•
Chemokines	64	30	Vert	29	29	13	31	Vert	2	•
Four-helical cytokines +	110	38	Vert	36	18	4	43	Vert	000	•
Trypsin-like proteases	223	54	F M	43	13	26	42	mammals	20.0	•
					10	27	37	Vert	2.10	•
Cystatins	56	19	V M	18	12	7	27	Vert	00	•
Lysozymes	97	21	M	14	12	30	53	Vert	112	•
β-trefoil cytokines +	67	17	M	12	11	7	33	M	10	•
Albumins	84	14	Vert	12	10	16	28	primates	202	•
Ly-6 like proteins	30	15	M	13	9	14	22	mammals	1	•
Granulins	18	9	V M	7	7	32	45	Mm	0	•
Hemopexin repeats	124	27	M	16	7	4	20	Ce Pl Hs	1	•
Phospholipases A2	142	11	Vert	7	7	30	48	Vert	21	•
Nuclear (77)	6143	1866		859
Homeobox	276	135	V F M	64	37	7	30	V M	.0	•
					7	19	29	V F M	00
Nuclear receptors (ligand binding)	311	128	M	120	28	8	22	M	..1
					11	9	29	Ce	02.	•
					10	13	27	Ce	021	•
					10	9	29	Ce	.21	•
					8	9	21	Ce	0.1
Krueppel associated box	58	23	Hs Mm	22	22	12	41	Hs Mm	1	•
Nuclear receptors (Zn finger domain)	215	126	M	115	20	27	38	Ce	....1.
					16	28	39	Ce	...2..
					14	34	46	Ce	...2.2	•
					13	26	39	Ce	0...1.	•
					9	28	38	Ce	.2....
					8	28	42	Ce	.2..1.	•
					7	42	54	M	..0...
					6	40	45	Ce	.....2
					6	32	38	Ce	0.....
Leucine-rich region	20	12	Hs	12	12	47	58	Hs	0	•
B-box Zn finger	61	23	V M	9	8	15	32	At Ce	0	•
IPT domains	103	32	F M	21	6	13	19	Vert	20	•
Intracellular signalling (138)	10773	2808		977
C1 domains	58	27	V F M	16	9	16	27	F M	1	•

The table is divided into three sections corresponding to extracellular, nuclear and signalling classes as defined by the SMART database. Domain names are taken from SMART; N_p gives the number of protein sequences for each domain; N_g, the number of genomic DNA entries found; N₂₊, the number of domains containing two or more exons (or parts thereof); and N_c, the number (out of N₂₊) having pairwise S_G ≥0.8. %I_min gives the lowest pairwise protein % sequence identity observed for the cluster; %I_ave gives the average. Taxonomic data were extracted from DDBJ/EMBL/GenBank and the descriptions given for all genomic DNA sequences (N_g, Tax_g) and for those in the clusters (N_c, Tax_c). When more than three species were represented in the genomic DNA or the clusters, the taxonomic groups were abbreviated as: V, Viriplantae; F, Fungi; M, Metazoan; Vert, vertebrates or mammals; otherwise species are given: At, Arabidopsis thaliana; Ce, Caenorhabditis elegans; Mm, Mus musculus; Pl, Paracentrotus lividus (urchin); Hs, Homo sapiens. ‘Introns’ gives a short description of the conserved gene structure by giving the phases of the equivalent introns. Where more than one group is present in a single SMART domain, the intron patterns are aligned to show their relationships to one another (a ‘.’ denotes a gap). Groups are separated according to the four broad classes in SMART, with the number in parentheses giving the total number of domains in the class, and N_p, N_g and N₂₊ in these rows giving totals for the class. No groups with similar gene structures were found for the ‘Others’ class (N_p = 3168, N_g = 560, N₂₊ = 159). Domain names are taken from SMART: Meprin and traf homology, MATH; chemokines, SCY; trypsin-like proteases, Tryp_SPc; cystatins, CY; lysozymes, LYZ1; albumins, ALBUMIN; Ly-6 like, LU; granulins, GRAN; hemopexin repeats, HX; phospholipases A2, PA2; homeobox, HOX; nuclear receptors ligand binding, HOLI; Krueppel associated box, KRAB; nuclear receptors Zn finger, ZnF_C4; leucine-rich region, LRR; B-box Zn finger, BBOX; IPT domains, IPT; C1 domains, C1. Those domains marked with a ‘+’ symbol are those comprising multiple SMART alignments linked via structure based sequence alignment: four-helical cytokines, CSF2 IFabd IL10 IL2 IL4_13 IL6 LIF_OSM; β-trefoil cytokines, FGF IL1.

It is clear from the table that most clusters from extracellular signalling domains come from vertebrates, and that the domains themselves are often involved in the immune system. Four of these domains are effector or cytokine molecules: the chemokines, four-helical cytokines (the cluster includes interleukins 2, 4, 6, 10, 13 and colony stimulating factor), β-trefoil cytokines (fibroblast growth factors and interleukin 1), and granulins (GRAN). Other domains involved in the immune system include the lysozymes, one cluster of 13 trypsin-like proteases (including granzyme B, mast-cell proteases, cathepsin G and leukocyte elastases) and Ly-6-like proteins. Within these clusters there is only one sequence from the β-trefoil cytokines (an FGF homologue from Drosophila melanogaster) from outside the vertebrates. Vertebrate-specific clusters are also found in the cystatins and albumins, which can be loosely grouped into blood constituents, and in the phospholipases A2, which include several snake venom components.

Figure 2 shows an alignment of four-helical cytokines with the location, length and phase of introns, and a representative tertiary structure coloured according to the location of the core exons. For this diverse superfamily sequence identities are as low as 4% and homology can only be inferred from a similar protein structure and function. Nevertheless, many family members share a common gene structure comprising four core exons, with all introns having phase 0. Despite great sequence diversity this superfamily is very specific to the vertebrates; with the exception of viral homologues, no non-vertebrate members have been found (Lander et al., 2001), nor has any non-vertebrate protein to date been found to adopt the unusual up-up-down-down topology (Lo Conte et al., 2000).

Fig. 2. Gene structure conservation within the four helical cytokines. (A) Alscript (Barton, 1993) figure showing a member of the four-helical cytokine superfamily (SMART domains: IL2, IL6, CSF2, IL4_13, LIM_OSF, IL10, IFabcd merged via the SCOP ‘four-helical cytokine’ superfamily). Aligned sequences for individual SMART entries were taken from the SMART database. Representative three-dimensional structures were aligned using the STAMP package (Russell and Barton, 1992), and this was used to merge the different SMART alignments in this superfamily. Residues are coloured according to conservation of hydrophobic properties: yellow background, conserved hydrophobic character; red characters, conserved hydrophobic. Structurally similar regions, defined by STAMP (Russell and Barton, 1992) are boxed. The location of introns within each sequence is shown within braces; the two numbers denote the phase and length of the intron. A ‘×’ denotes regions where unconserved sequence is not shown for clarity. The location of the four helices common to all proteins in this superfamily is shown below the alignment, coloured blue, yellow and red to agree with the order shown in (B). (B) Molscript (Kraulis, 1991) figure showing the structure of granulocyte macrophage colony stimulating factor (PDB code 2gmf). α-helices are shown as ribbons, β-strands as arrows and other structures as coil. Segments are coloured blue, green, yellow and red to show the order of core exons in the superfamily.

Gene structures for meprin and traf homology (MATH) domains from Arabidopsis thaliana and Caenorhabditis elegans are also surprisingly similar. A total of 87 out of 98 proteins with known gene structures contain a phase 2 intron at the same position, and for 86 the codon interrupted is TGG (tryptophan). Conserved or nearly conserved tryptophans like this are also found in the first (phase 1) and last (phase 2) introns of the lysozymes, and a thymidine is conserved in the second codon position within the phase 2 intron in the chemokines, which leads to conservation of hydrophobic character in the amino acid.

Several nuclear domains also show conservation of gene structure despite little sequence similarity. For example, several large clusters with similar gene structures are found in homeodomains and nuclear receptors. Here there is no obvious common theme uniting the functions of the domains, although the many C.elegans-specific clusters in the nuclear receptors are possibly associated with the diversification of this protein family within the nematodes (Sluder et al., 1999).

Table I also contains instances where domains duplicated in tandem show preservation of gene structure. Clusters from albumins, hemopexin repeats (Altruda et al., 1988) and granulins show a similarity in gene structure within tandem repeats despite limited sequence similarity. This parallels the situation seen previously for enzymes such as phosphoglycerate kinase (PGK) (Michelson, 1985).

The majority of domains containing introns show little conservation of gene structure. Figure 3 shows an alignment and introns for the GAP superfamily, a diverse group of intracellular signalling molecules. Modifications in the number, position and phase of introns are common, even with phase differences at the same approximate alignment positions. The lack of gene structure conservation within these and other intracellular signalling domains broadly supports suggestions that the spread of these modules may have preceded the rise of exon shuffling (Patthy, 1999).

Fig. 3. Gene structure differences within the GAP superfamily. Alscript (Barton, 1993) figure showing GAP domains (SMART domains RasGAP and RhoGAP linked by SCOP ‘GTPase activation domain, GAP’ superfamily). Details are as for Figure 2.

Discussion

Why can gene structure be conserved in certain domains despite little or no protein sequence similarity? For clusters specific to vertebrates or to C.elegans the most plausible explanation is that the proteins have diversified quickly via point mutations, and within too short a time to permit changes to the gene structure (e.g. such changes are rare in vertebrates; Guigo et al., unpublished results; Venkatesh et al., 1999). Rapid evolution has clearly lead to situations where the position of introns and the protein three-dimensional structure are apparently more conserved than the protein sequence. There could well be other examples where conservation of gene structure occurs within domains under evolutionary pressure, but there is currently not enough genomic DNA sequence from a sufficient diversity of organisms to see them.

There could also be functional reasons for the conservation of intron position. For example, conservation of phase 1 or 2 introns (where an amino acid is split across an intron) could be due to pressure to conserve the ‘split’ amino acid and/or the integrity of the splice site (Fichant, 1992). This could lead to a situation where an intron position becomes inexorably linked to protein structure and/or function (e.g. MATH domains). Another possible role for intron position conservation may be in alternative splicing, reflecting sites on protein structures that are useful for modulating function. An investigation of this possibility awaits large sets of carefully annotated splice variants.

The fact that intron positions (particularly phase 1 introns) sometimes correspond to domain boundaries in modular extracellular proteins led to the suggestion that exonic recombination played a key role in the ‘big bang’ of metazoan radiation (Patthy, 1994, 1999). During this period many new multi-domain, modular proteins evolved to function in the cell–cell communication processes first seen in the multicellular animals. The observation that phase and position of introns is conserved within certain diverse domain families similarly suggests that exon shuffling may have played key roles during other dramatic events in evolutionary history, such as the evolution of molecules involved in the immune system, or the diversification of nuclear receptors in C.elegans.

An intriguing possibility is that fixation of intron positions permitted the construction of chimeric domains through unequal crossing-over events between members of the same family. Shuffling equivalent parts of different proteins sharing the same fold could be a mechanism for rapid evolution. This phenomenon would suggest that one might obtain different phylogenies for the exons comprising the domains. We could not identify a clear example where this was the case, although the sequences are often too short, or have diverged beyond the point where phylogenetic reconstruction gives reliable trees. Other evidence for unequal crossing over comes from clustering of gene families on chromosomes. This is indeed the case for several of the cytokines (e.g. Kelleher et al., 1991) and a preliminary investigation with human proteins found that several of the clusters in Table I are within close proximity on the same chromosome. For example, two clusters of trypsin-like proteases on human chromosomes 14 and 19 correlate with the different gene structures given in Table I.

Genetic features beyond DNA and protein sequences, such as intron position and phase, can be excellent sources of phylogenetic information (e.g. Patthy, 1999; Rokas et al., 1999; Venkatesh et al., 1999; Rokas and Holland, 2000; Telford et al., 2000; Krem and Di Cera, 2001). Several of the groups shown in Table I contain similar but not identical gene structures. For example, the nuclear receptor ligand binding domains in Table I all appear to contain gene structures that are variations on a similar three intron (phase 0, 2 and 1) theme, with different clusters being formed possibly by different intron loss events. Differences in gene structure might thus provide clues as to the phylogeny of distantly related proteins.

Intron position could also be used for the detection and accurate alignment of new members of diverse protein families. Where gene structure conservation is seen in the absence of protein sequence identity it might be possible to detect additional, and very remote members of families simply by searching for genes with a similar set of intron positions and phases (Brown et al., 1995). Similarly, in regions of alignment ambiguity (e.g. due to low local sequence similarity) it may be fruitful to consider whether or not intron positions are equivalent.

Our results argue strongly neither for the ‘introns-early’ (Gilbert, 1986; Dorit et al., 1990; de Souza et al., 1998) nor the ‘introns-late’ (Orgel and Crick, 1980; Cavalier-Smith, 1985; Sharp, 1985; Stoltzfus et al., 1994) hypotheses. However, they do suggest that the conservation of intron location and phase within diverse protein domains might be explained by comparatively recent evolutionary events, rather than by an ancient correspondence between exons and local regions of structure or function.

We have found clear instances of gene structure conservation despite low protein sequence similarity. We anticipate that more examples will be uncovered as genomic sequences are determined for more organisms, giving further clues to events in evolution.

Materials and methods

We took 419 protein domain sequence alignments from SMART (Schultz et al., 2000) and information on superfamilies defined by analysis of protein structure from SCOP (Lo Conte et al., 2000), which groups proteins into homologous superfamilies based on similarities beyond sequence (e.g. structural features, common active sites, etc.). A total of 58 SCOP superfamilies were found to contain two or more SMART domains, and these were merged by structure alignment (Russell and Barton, 1992).

We extracted a database of eukaryotic genomic DNA sequences from DDBJ/EMBL/GenBank release 120 (rod, mam, pri, pln, inv, vrt). Each domain sequence was compared against this database to identify putative identical matches using WU-tblastn (W = 5 hitdist = 11 –filter xnu + seg T = 1000 –matrix identity –nogap S = 40 topcomboN = 2). Each pair of protein–genomic DNA sequences was aligned using genewise (Birney, unpublished) to refine the exon–intron boundaries and to obtain phase information. The SMART database contained a total of 27 168 protein sequences (including Eukarya, Archaea and Bacteria), of which 7701 gene sequences could be matched to a genomic DNA sequence. Ignoring the genes that lacked introns left 2987 gene sequences. Of the 419 protein domain alignments, 250 contained at least two gene sequences with one or more introns, and 151 contained six or more. We merged gene structures with their associated proteins and incorporated the gene structures into the alignment. For each pair of proteins in every alignment the associated sets of intron locations were matched by finding the optimal set of exon matches, fusions and insertions that could transform one gene structure into the other.

We constructed a simple gene structure similarity measure, which assesses the similarity between a pair of aligned proteins based solely on the location and phase of introns:

a_i and b_i refer to the ith equivalent intron positions in the two proteins (one on each side of an exon), N_max is the largest number of introns found in either protein, N_equiv is the number of introns that have been assigned to be equivalent, d_i is the difference in position of the introns within the two proteins (in amino acids), and ϕ(a_i,b_i) is 1 if the phases are the same and 0 otherwise. γ and δ are constants (0.2, 30) in a sigmoid function (i.e. the first term above), and were chosen such that the function was insensitive to small changes in intron positions (i.e. ± 10 residues). We weighted the terms for position and phase equally, although modifications to weights do not change the results significantly. We found intron size varied considerably even in closely related proteins, and thus did not develop a scoring term for differences between intron lengths. S_G ranges from 0 (no similarity) to 1 (identical), and we found that requiring values to be ≥0.8 gives pairs or groups of proteins with gene structures that we consider to be similar. This allows for short insertions/deletions or errors/ambiguities in the alignments.

For each domain we calculated all pairwise S_G values and used these to cluster sequences by complete linkage. Only clusters with six or more sequences were considered further. We performed a simple χ²-test with 1 degree of freedom for each domain and each gene structure cluster, to test the hypothesis that the fraction of domains with that gene structure (considering only intron phases) was significantly more than the background frequency of the gene structure in the entire data set. Table I shows domains containing at least one significant cluster (p <0.001), which are indicated by black dots in the rightmost column.

All data used in this study are available from the authors.