Incongruent Patterns of Local and Global Genome Size Evolution in Cotton

Abstract

Genome sizes in plants vary over several orders of magnitude, reflecting a combination of differentially acting local and global forces such as biases in indel accumulation and transposable element proliferation or removal. To gain insight into the relative role of these and other forces, ∼105 kb of contiguous sequence surrounding the cellulose synthase gene CesA1 was compared for the two coresident genomes (A_T and D_T) of the allopolyploid cotton species, Gossypium hirsutum. These two genomes differ approximately twofold in size, having diverged from a common ancestor ∼5–10 million years ago (Mya) and been reunited in the same nucleus at the time of polyploid formation, ∼1–2 Mya. Gene content, order, and spacing are largely conserved between the two genomes, although a few transposable elements and a single cpDNA fragment distinguish the two homoeologs. Sequence conservation is high in both intergenic and genic regions, with 14 conserved genes detected in both genomes yielding a density of 1 gene every 7.5 kb. In contrast to the twofold overall difference in DNA content, no disparity in size was observed for this 105-kb region, and 555 indels were detected that distinguish the two homoeologous BACs, approximately equally distributed between A_T and D_T in number and aggregate size. The data demonstrate that genome size evolution at this phylogenetic scale is not primarily caused by mechanisms that operate uniformly across different genomic regions and components; instead, the twofold overall difference in DNA content must reflect locally operating forces between gene islands or in largely gene-free regions.

The lack of correlation between genome size and organism complexity, known as the “C-value paradox” (Thomas 1971) or “G-value/N-value paradox” (Claverie 2000; Bertran and Long 2002), has been recognized for more than half a century (Mirsky and Ris 1951). Genome size in eukaryotes varies >200,000-fold, from ∼2.8 Mb in Encephalitozoon cuniculi (Biderre et al. 1998) to >690,000 Mb in the diatom Navicola pelliculosa (Cavalier-Smith 1985; Li and Graur 1991). Even within various eukaryotic groups, there are remarkable differences in genome size. Protozoans display a 5800-fold genome size variation, vertebrates a 330-fold variation, and angiosperms display a 2300-fold variation in genome size (Cavalier-Smith 1985; Gregory 2001; Bennett and Leitch 2003). Significant genome size variation has also been observed among closely related species; for instance, the plant genus Crepsis displays a ninefold variation (Jones and Brown 1976), whereas another plant genus, Vicia, displays a sixfold variation in genome size (Chooi 1971). Despite this impressive variation in genome size, the amount of variation in the numbers of protein coding genes is only about 20-fold (Li 1997).

Although it is generally agreed that the majority of genome size variation can be accounted for by differences in the amount of noncoding DNA, the relative importance of mechanisms that generate genome size variation is not well-understood. In plants, the most prominent forces involved in genomic expansion are acknowledged to be polyploidy (Wendel 2000) and transposable element (TE) amplification (Bennetzen 2002), complemented by smaller-scale processes such as increases in pseudogene number (Zhang 2003), intron size (Deutsch and Long 1999; Vinogradov 1999), and incorporation of organellar genome fragments into the nucleus (Adams and Palmer 2003; Shahmuradov et al. 2003). Taken alone, these forces would cause an upward spiral toward bloated genomes (Bennetzen and Kellogg 1997). This one-way ticket to obesity is contraindicated by the phylogenetic distribution of plants with smaller genomes (Bennett and Leitch 1995, 1997; Leitch et al. 1998; Wendel et al. 2002b), as well as by the existence of many plants, such as Arabidopsis (Vision et al. 2000) and maize (Ilic et al. 2003), that clearly have eliminated massive amounts of DNA following polyploidization. Less well understood are evolutionary mechanisms that reduce genome size. Global mechanisms, such as small indel (<400 bp) mutational bias (Petrov 2002b) and species-specific differences in nonhomoeologous end joining (Kirik et al. 2000; Orel and Puchta 2003), have the potential to stochastically and differentially contract genomes. Sequence-specific mechanisms, such as LTR recombination (Shirasu et al. 2000; Vitte and Panaud 2003), ectopic recombination (Langley et al. 1988; Bennetzen 2000b; Petrov et al. 2003), and illegitimate recombination (Devos et al. 2002; Ma et al. 2004), have been shown to be capable of removing larger segments of DNA. Superimposed on these internal molecular genetic mechanisms are external factors and selective forces that may mold genome size; cell size limitations and cell division rate selection, for example, may constrain genome size (Gregory 2002).

Some mechanisms of genome size evolution, such as polyploidy and global deletional biases, are expected to affect all genomic constituents approximately equally, whereas others, such as proliferation of transposable elements, are likely to be more heterogeneous in their impacts on various genomic regions. To evaluate these alternatives, it may be informative to compare closely related species that differ dramatically in genome size. Here we demonstrate this approach using model species from the genus Gossypium. Despite its relatively young age (5–10 million years old; Cronn et al. 2002) and conserved complement of genes, DNA content varies more than threefold within the genus, from 980 to 3425 Mb per 1C nucleus (Wendel and Cronn 2003). Two diploid groups of species, designated A-genome and D-genome, diverged from a common ancestor about 5–10 million years ago (Mya) and acquired genomes that differ approximately twofold in size. Approximately 1–2 Mya, these two genomes became reunited in a common nucleus through allopolyploidization, leading to the evolution of the modern polyploid cotton species, including Gossypium hirsutum, the primary cotton of commerce. Backed by a well-studied phylogeny (Fig. 1), we have embarked on comparative BAC sequencing to illuminate the patterns and processes responsible for modern-day genome size differences. For our initial study, we compared 100 kb+ of homoeologous sequence surrounding a cellulose synthase gene (CesA1) from the two genomes (designated A_T and D_T) that comprise the allotetraploid, G. hirsutum, and that differ overall in genome size by a factor of 2 (1C = 980 Mb and 1860 Mb for D_T and A_T, respectively; Endrizzi et al. 1985). Remarkably, sequence conservation between the A_T and D_T genomes is shown to be high, even in intergenic regions. No evidence of mechanisms that underlie the twofold genome size difference is observed within this genomic region, where even the >550 small indels detected are evenly divided among the two genomes. The results show that genome size evolution operates regionally rather than globally at this phylogenetic scale, perhaps largely between gene islands.

Figure 1.

The evolutionary history of diploid and tetraploid Gossypium, as inferred by numerous chloroplast and nuclear data sets (Seelanan et al. 1997; Small et al. 1998; Cronn et al. 2002). Genome groups designate closely related species, as determined by interspecific meiotic pairing and chromosome size (Endrizzi et al. 1985). All diploid species have the same base chromosome number (n = 13); however, each genome group varies in genome size (1C content indicated in circles). Polyploid species are thought to have originated 1–2 Mya, following divergence of their diploid progenitors 5–10 Mya.

RESULTS

General Sequence Comparison of the Homologous BACs

The CesA1 BACs from the A_T and D_T genomes were shotgun-sequenced and assembled, giving a total of 2311 sequence reads and 4019 sequence reads, respectively. The overall gapped, aligned length of A_T with D_T is 123.8 kb. The ungapped aligned length of the A_T BAC is 103.9 kb, and the ungapped aligned length of the D_T BAC is 107.9 kb. Thus, for the CesA1 region in G. hirsutum, there is only a 4-kb difference in length between the A_T and D_T genomes. Both BACs are equal in GC content (33% GC). Database searches led to the inference of 14 genes in the CesA1 region, shared by both genomes. The total length of these genes was calculated to be 29.2 kb, or about one-third of the sequence. Excluding the 555 gapped positions (see below), which collectively exclude 36 kb and distinguish the two homoeologs, sequence identity over the aligned, ungapped positions was extraordinarily high (95%).

Analysis of Potential Genes

Fourteen genes were predicted along the colinear segment (Fig. 2), giving an average density of 1 gene per 7.5 kb of sequence. This is slightly less than the average Arabidopsis gene density of 1 gene per 4.5 kb of sequence (The Arabidopsis Genome Initiative 2000) and similar to the average gene density in rice (Rice Chromosome 10 Sequencing Consortium 2003). The CesA1 region appears to be part of a gene island, as the gene density is fairly high and the non-genic DNA content low. The predicted genes (Table 1) range in size from a partial 244-bp fragment of a putative ABC-transporter to 4.3 kb in a predicted gene that is similar to an expressed Arabidopsis protein (gi: 18396997). Silent and replacement site substitutions were calculated for each gene (Table 1). Synonymous substitution rates between homoeologous genes vary over a 10-fold range, from 0.008 to 0.084, with a weighted mean of 0.037; this value is identical to the weighted mean of 0.037 that was previously reported for a set of ∼40 homoeologous genes in polyploid Gossypium (Senchina et al. 2003).

Figure 2.

Pairwise alignment of CesA1 homoeologous BACs, A_T and D_T, to scale. A_T and D_T are genes shown as block diagrams: numbered boxes are predicted corresponding to the list presented in Table 1; rTE1, rTE2, and rTE3 represent the three largely intact retrotransposons identified (rTE1 encompasses two predicted copia elements); the POGO and MuDR-like TEs are indicated individually, as is the ycf2 fragment of plastidial origin. The bottom panel indicates a continuous window of sequence identity between the two BACs, scaled from 50% to 100%.

Table 1.

Gene Features Predicted Along Homoeologous A_T and D_T BACs Surrounding the CesA1 Gene in Allopolyploid Cotton

		Length (bp)b			Total length, exonsb			Total length, introns		Length (a.a.)		Divergencec
Gene	Putative functiona	DT	AT	Exons	DT	AT	Introns	DT	AT	DT	AT	Ks	Ka	Ksil
1	ABC transporterd	244	244	*	244	244	*	0	0	81	81	0.035	0.000	0.035
2	GTP-binding protein	1909	1925	2	657	657	1	1252	1268	218	218	0.054	0.006	0.047
3	WRKY TF	1024	1013	3	822	819	2	202	194	273	274	0.064	0.014	0.041
4	Arabidopsis hypothetical protein	994	992	3	471	471	2	523	521	156	156	0.020	0.017	0.061
5	No BLAST homoeology	519	519	1	519	519	0	0	0	172	172	0.084	0.026	0.084
6	G-protein B	2376	2376	9	1068	1068	8	1308	1308	355	355	0.043	0.012	0.029
7	CesA	4080	4083	12	2925	2925	11	1155	1158	974	974	0.041	0.004	0.033
8	LeuRR	1308	1308	1	1308	1308	0	0	0	435	435	0.020	0.012	0.020
9	PRR/Se-binding protein	2628	2628	3	2418	2418	2	210	210	805	805	0.019	0.013	0.018
10	Ribosomal protein L11	1273	1275	4	519	519	3	754	756	172	172	0.008	0.000	0.029
11	LeuRR transmembrane or kinase	3189	3197	11	1857	1857	10	1332	1340	618	618	0.042	0.006	0.031
12	Growth regulator	2631	2637	10	1797	1800	9	834	837	598	599	0.050	0.015	0.037
13	Permease	2633	2629	13	1575	1575	12	1058	1054	524	524	0.016	0.006	0.021
14	Arabidopsis expressed protein	4349	4325	8	1287	1278	7	3062	3047	425	422	0.037	0.014	0.029
										Weighted average			0.037	0.010	0.032

^aPutative function is assigned by BLAST homoeology to genes in GenBank. The locations of genes along the BAC contigs are represented in Figure 2.

^bTotal length including stop codon.

^cK_s, K_a, and K_sil denote rate of substitution across all sites, substitutions at nonsynonymous sites, and synonymous sites within codons plus all noncoding positions, respectively.

^dThis predicted gene is fragmented in the BAC; a start codon was identified, but no intronic sequences or stop codon was found. This gene would presumably be full length, were it not fragmented in the generation of the BAC library. An asterisk (*) signifies that the number of exons and introns is not known.

We searched a growing collection of cotton EST data sets for evidence of transcription of the predicted genes. To date, ∼150,000 ESTs have been generated from various tissues and organs of diploid and polyploid cotton (J.A. Udall, J. Hatfield, R.A. Rapp, Y. Wu, L. Dennis, A.B. Arapat, T. Wilkins, J. Guo, X. Chen, E. Taliercio et al., unpubl.). Searches of these data sets revealed evidence for expression of five of the 14 genes inferred to reside on the CesA1 BACs. This, in addition to the sequence divergence evidence and low levels of replacement substitutions (Table 1), lends support to the gene predictions.

Analysis of Potential Transposable Elements

Differential insertions of transposable elements (TEs) are recognized as a prominent force in genome size expansion. Thus, we examined the CesA1 BACs for evidence of transposable elements. A total of six largely intact TEs were detected in the two G. hirsutum homoeologs, two that are shared, one that is unique to A_T, and three that are unique to D_T. The two genomes also share a highly degraded EnSpm (class 2) remnant, identified by CENSOR (Jurka et al. 1996), as well as a potential and highly degraded retrotransposon, identified by BLAST homoeology to elements characterized in an ongoing study (J.S. Hawkins, H-R. Kim, R.A. Wing, J.D. Nason, and J.F. Wendel, unpubl.). The A_T genome has a series of potential, highly degraded retrotransposons of indeterminate number, again identified by BLAST homoeology. Additionally, nine potential miniature inverted-repeat TEs (MITEs) were predicted in the CesA1 region, seven shared between A_T and D_T, and two that are unique to A_T. Overall, transposable elements (including remnants) account for 28.5 kb of sequence in the region, 10.8 kb in A_T, and 17.7 kb in D_T.

The two shared intact transposable elements belong to different classes. One of the shared transposable elements has similarity to known POGO elements from Arabidopsis. The putative POGO is flanked by 15-bp terminal inverted repeats (TIR), which have 73% identity (5′-TIR vs. 3′-TIR) and which retain the typical TA dinucleotide target site duplication. Each Gossypium POGO element retains ∼90% identity over the TIR to several Arabidopsis POGO elements (Feschotte and Mouches 2000) and 35% identity over the entire element. Compared with each other, the A_T POGO (1940 bp) and the D_T POGO (2150 bp) have 83% sequence identity, including gaps, and 92% sequence identity when gaps are excluded.

The other shared intact transposable element is a retrotransposon of unidentified type. This element was identified through its BLASTX identity to known reverse transcriptase (RT) sequences (40% identity and 65% similar over 100 amino acids to numerous RT sequences from Arabidopsis). There is evidence that this RT sequence may have been derived from a degraded non-LTR retroelement, as a few BLASTX hits were to non-LTR RT sequences and no vestige of ancient LTRs was identified.

The A_T and D_T genomes also share what appears to be a 45-bp remnant of a highly degraded EnSpm transposon. This remnant was identified by CENSOR as having identity to the described EnSpm element ATENSPM5 from Arabidopsis (Jurka 2000). Sequence identity between A_T and D_T over the remnant is 100%, and the sequence identity between either Gossypium remnant and ATENSPM5 is 82%.

Finally, the A_T and D_T genomes also share a potential, highly degraded gypsy retrotransposon. The D_T element shows 164 bp identity to Gossypium gypsy elements, whereas 204 bp of identity was observed for the A_T element.

The A_T BAC sequence contains only one identified largely intact transposable element that is not shared with the D_T genome. This element is a predicted long terminal repeat (LTR) retrotransposon of unknown type. The element contains 612-bp LTRs, which retain 98% sequence identity with each other. The element is 3138 bp in length and contains homoeology to identified tomato (gi: 4235644) and Arabidopsis pol proteins of unspecified type.

The A_T BAC sequence contains two potential and extremely degraded retroelement clusters. The first retroelement cluster spans 7.5 kb of sequence, although only 1573 bp can be identified as belonging to degraded TEs. This cluster may have contained two to three gypsy elements, one shared with the D_T BAC sequence (204 bp; mentioned above), and contains moderate sequence identity (60%–70%) to previous reported A-genome-specific repetitive sequences (Zhao et al. 1998). The second degraded retroelement cluster contains a potential gypsy remnant and a potential copia remnant. These two remnants are situated on either side of a cpDNA insertion (see below), and were likely genomic neighbors before being separated by the cpDNA insertion.

The D_T BAC contains three transposable elements (one DNA element and two copia retrotransposons) that are not shared with the A_T genome. The DNA element has homoeology to several Oryza mutator (MuDR) elements, as well as some limited homoeology to the Arabidopsis Vandal12 DNA element (Jurka et al. 1996). The element appears to be degraded, as the protein alignment generated by BLASTX shows only 26% identity (44% similarity) over 536 amino acids.

The two copia insertions that are D_T-specific for this BAC are nested within the POGO insertion (see Fig. 3). The outer copia has 200-bp LTRs that are 97% identical. The element is 5.3 kb in length and has well-defined reverse transcriptase-, integrase-, and protease-coding domains. The inner copia has 561-bp LTRs that are 99.7% identical. This element also is 5.3 kb in length and has well-defined reverse transcriptase- and integrase-coding domains. The protease-coding domain for this element could not be identified. The inner copia inserted between the protease-coding domain and LTR of the outer copia, after the outer one had inserted. These copia insertions share no identity with each other; thus, they probably belong to different families. Retrotransposon insertions can be dated based on LTR divergence (SanMiguel et al. 1998), although these estimates provide only approximations, given the unknown absolute rate of mutation. Previous data on sequence divergence in Gossypium (Senchina et al. 2003) can be used to infer the relative insertion times of each copia. The percent divergence between LTRs of the outer copia (3%) is similar to that estimated for divergence of A- and D-genome diploids, suggesting transposition shortly after the divergence of these two species groups. Similarly, LTR divergence of the inner copia (0.3%) is slightly less than that estimated for comparisons between model diploid progenitors and their counterparts in the polyploid, suggesting insertion of the internal copia subsequent to polyploidization.

Figure 3.

Nested insertions of retroelements in the A_T BAC of Gossypium hirsutum. The outer copia is shown in gray and the inner copia in black. Four LTRs, corresponding to the two copia insertions, are shown as triangles. The three coding domains of the copias, reverse transcriptase (RT), integrase (INT), and protease (PRO), are designated by the labeled boxes within the LTRs. Surrounding the copia nest is a single POGO element that is shared by A_T and D_T, and which was split in two when the copias inserted.

Miniature inverted-repeat transposable elements (MITEs) are a common feature of gene-rich regions (Feschotte et al. 2002). Although they are categorized as class 2 elements, these nonautonomous TEs do not encode a transposase or transposase remnant; thus, the prediction and classification of potential MITEs is primarily achieved through terminal inverted repeat (TIR) and target site duplication (TSD) identification (Feschotte et al. 2002). Considering this, two approaches were used to predict MITEs in the A_T and D_T CesA BACs. The first approach, which attempted to predict MITEs from known families (Stowaway, Tourist, etc.) by searching for similarity to TIRs from known MITEs in Arabidopsis, Brassica, and the grasses, did not reveal any known MITEs in the CesA BACs. The second approach used a de novo search method (Tu 2001), which inspects the sequence for potential TIRs that also have a TSD. Although this method predicted many MITEs in both A_T and D_T, subsequent inspection revealed that a majority of the predicted TIRs and TSDs contained simple sequence repeats (SSRs). Predicted MITEs whose TIRs were comprised mostly of SSRs were considered probable artifacts. In total, 16 MITEs were predicted in the CesA1 region, seven shared and two unique to A_T, accounting for 2.5 kb and 2 kb in A_T and D_T, respectively.

Other Potential Mechanisms of Genome Size Evolution

In addition to TE insertions, the CesA1 BAC alignments were examined for other distinctions. Most prominent among these is a 900-bp fragment of the plastid gene ycf2, which was inserted into a noncoding region of the A_T genome (Fig. 2; 90% identity over 897 bp to ycf2 from Arabidopsis) and was flanked by 1.5 kb of A_T-specific sequence of undetermined identity. While accounting for a mere 0.86% of the total aligned length of the homoeologous BACs, the 900-bp ycf2 fragment accounts for 5.6% of the A_T-specific sequence.

Intron sizes for each gene were compared for all inferred genes on the homoeologous BACs to evaluate their potential contribution to the genome size variation. Intron sizes deviated by an average of 4.3 bp per gene, with a range of 0–16 bp. The total contribution of intron size differences to the size difference of the region was a mere gain of 3 bp in A_T. This result provides a striking contrast to reports of intron sizes contributing to genome size differences over much longer evolutionary timescales (Deutsch and Long 1999; Vinogradov 1999). The present study concurs with previous data on Gossypium intron size variation, which suggested that there exists little intron size variation among Gossypium species, irrespective of genome size (Wendel et al. 2002a).

Evidence for a bias in small indel number and length was also examined for the homoeologous sequences (Fig. 4; Table 2). The frequency of small indels was computed for any gapped position <400 bp in length. A total of 555 small gaps were scored in the two BACs, approximately equally distributed between A_T and D_T in number and aggregate size. Of the 269 indels in A_T and 286 indels in D_T, 264 and 279 were classified as small indels, respectively. Moreover, small indels account for 2777 bp of missing sequence in A_T and 2897 bp of missing sequence in D_T, a difference of only 120 bp. In addition to similarities in number and aggregate size, the frequency spectrum of small indels is similar in shape and position between the A_T and D_T BACs; that is, the number of indels of any length is similar between A_T and D_T (Fig. 4). Overall, small indels account for 14% and 18% of the total length in A_T and D_T, respectively, but fail to contribute significantly to the overall size difference in the aligned region.

Figure 4.

The spectrum of small indels inferred from sequence alignment of the A_T and D_T CesA1 BACs. For A_T (solid bars), “differences” are gapped positions relative to D_T, whereas for D_T (open bars), differences reflect gaps relative to A_T. These indels are not although the phylogenetically polarized, spectrum of indels is equivalent in the two genomes.

Table 2.

Spectrum of Small Indels^a in the Comparison Between A_T and D_T Homoeologous BACs of Gossypium hirsutum

	AT Genome		DT Genome
	No. of indels	bp	No. of indels	bp
1-10 bp	210	593	207	489
11-20 bp	25	369	34	506
21-30 bp	10	260	17	414
31-40 bp	8	274	5	239
41-50 bp	1	49	8	346
51-100 bp	5	353	5	351
101-200 bp	4	665	2	321
200-400 bp	1	214	1	231
Small indels	264	2777	279	2897
All indels	269	19,977	286	16,009

^aIndels are binned in multiples of 10 bp because of up to an indel length of 50 bp; the last three bins span 50 bp, 100 bp, and 200 bp, respectively, because of the infrequency of these larger indels in either genome. The last two rows tally totals for the number and amount of sequence accounted for by small indels (<400 bp) and all indels, respectively.

One hallmark of illegitimate recombination is the presence of direct flanking repeats 2–15 bp in size (Ma et al. 2004). We searched all indels discovered here for flanking repeats, restricting our attention to the 144-bp indels that were at least 10 bp in length (Ma et al. 2004). Of these, 55 (38%) showed flanking repeats of 2–15 bp (excluding possible mono- or dinucleotide and microsatellite expansion/contraction events). These flanking repeats were unequally distributed in number between the A_T and D_T genomes (19 vs. 36), but encompassed approximately the same amount of sequence (11,720 and 13,164, respectively). These data suggest that illegitimate recombination is a common mechanism of sequence evolution in Gossypium, and that it may play a role in genome size evolution. Additional analyses that include outgroups for phylogenetic polarization of indels will shed light on the extent and importance of this mechanism.

DISCUSSION

In recent years there has been a rapidly accumulating literature focused on comparative analyses of contiguous, homoeologous stretches of genomic sequence in plants. Stimulated by the seminal investigations of Bennetzen and colleagues on the maize, rice, and sorghum sh2/a1 and Adh regions (SanMiguel et al. 1996; Chen et al. 1997, 1998; Tikhonov et al. 1999) and the increasing accessibility of genomic tools, “microcollinearity” has been studied for numerous other genomic regions and taxa (Ku et al. 2000; Tarchini et al. 2000; Dubcovsky et al. 2001; Rossberg et al. 2001; Wicker et al. 2001; Fu and Dooner 2002; Ramakrishna et al. 2002a; Vandepoele et al. 2002; van Leeuwen et al. 2003; Chantret et al. 2004). Among the generalizations and insights that emerged from these analyses is the concept that gene order and content may be conserved over long periods of evolutionary time (Gale and Devos 1998; Bennetzen 2000a), that polyploidy may lead to a rapid decay in synteny and gene content preservation among homoeologs (Ilic et al. 2003; Kellis et al. 2004; Langham et al. 2004), and that intergenic regions may be subject to more dramatic and rapid evolutionary alterations. The latter in particular has led to the notion that much of the genome size evolution that takes place in plant genomes is caused by differential accumulation of retroelements in intergenic regions (Bennetzen 2000b), although it also is evident from the draft Oryza sativa genome sequence (Goff et al. 2002; Yu et al. 2002) that retroelements may be concentrated near centromeres and other largely heterochromatic regions. Superimposed on these ideas has been the concept that genome size itself may have biological significance and be visible to natural selection (Bennett 1985, 1987; Gregory and Hebert 1999), applying directional pressure on all genomic constituents simultaneously, perhaps through molding genome-specific mutational processes that determine the frequency and spectrum of deletions or insertions (Kirik et al. 2000; Petrov 2002b; Orel and Puchta 2003). Based on the foregoing, we anticipated that the twofold genome size difference that exists between A- and D-genome cotton species might reflect similar phenomena of either differential intergenic retroelement accumulation or perhaps a more globally operating bias in the prevalence and size of insertions and deletions. Neither of these expectations was realized, however, and in addition, a remarkable degree of conservation of the entire CesA1 region was observed, including the size and sequence of most intergenic regions.

Genome Evolution in the CesA1 Region of Polyploid Cotton

The most likely process responsible for the twofold genome size difference between the A_T and D_T genomes is differential accumulation or retention of transposable elements, particularly retroelements. In the region studied here, however, relatively few TEs were detected, and their differential presence does not correspond with the genome size difference; three of the four unique and intact TE insertions are found in the smaller (D_T) of the two genomes, accounting for 15.5 kb in D_T versus 5.8 kb in A_T. The presence of unique MITEs in A_T did little to counteract the disparity, only accounting for +500 bp in A_T. Thus, although transposable element amplification may have contributed to the twofold genome size difference, this phenomenon is not evidenced in this genomic region.

The CesA1 region was examined for evidence of ectopic recombination among retroelements. If ectopic recombination has played a role in shaping the CesA1 region, then footprints of the recombined elements should be apparent, such as solo LTRs resulting from recombination between LTRs of individual retroelements or between LTRs of distinct but linked elements (Vicient et al. 1999; Kalendar et al. 2000; Shirasu et al. 2000; Devos et al. 2002; Vitte and Panaud 2003). In the CesA1 region, however, all elements are either fully situated within a span of unique noncoding DNA or are identifiably full length. Thus, although ectopic recombination may play a role in shaping the genome and genome sizes in Gossypium, no evidence of that role was seen here.

Similarly, illegitimate recombination has recently been shown to have the ability to reduce genome size more than was previously anticipated (Ma et al. 2004). The current comparison does not distinguish insertions from deletions, and thus we are unable to accurately gauge the extent to which illegitimate recombination has shaped this region. However, as slightly more than a third of the indels >10 bp in size had flanking repeats, illegitimate recombination may prove to be an active force contributing to genome size evolution in Gossypium. Follow-up studies that distinguish insertions from deletions will further enable an evaluation of the importance of illegitimate recombination in cotton.

Analysis of the sequenced Arabidopsis and rice genomes showed that organelle–nuclear transfers (fragmented or full length) can be common in some genomes (rice) and relatively infrequent in others (Arabidopsis; Shahmuradov et al. 2003). In the present study, one chloroplast DNA (cpDNA) fragment was found in the A_T BAC, nestled among unique noncoding DNA. This fragment was only 900 bp in length, however, therefore it does not contribute significantly to genome size evolution in this region.

Despite evidence from broader phylogenetic surveys and some other systems that intron size may be correlated with genome size (Moriyama et al. 1998; Deutsch and Long 1999; Vinogradov 1999; McLysaght et al. 2000), this is not true for genes in the CesA1 region. The average intron size deviation was 4.3 bp per gene (60 bp total). Intron size deviation was not biased with respect to genome; the A_T BAC sequence contains a total of only 3 bp more intronic sequence than does its homoeolog. This result is not surprising for Gossypium; a previous study reported for 40 nuclear genes that there exists no significant size variation between Gossypium species groups with varying genome sizes (Senchina et al. 2003). Thus, although intron size expansion/contraction may play a role in shaping the size of other genomes, evidence from Gossypium indicates that it has not played a similar role at the phylogenetic scale encompassed by this genus.

One of the attractive proposals that attempts to account for genome size variation is that there exist biases in the frequency and size of insertions and deletions (Bensasson et al. 2001; Petrov 2002a,b). To evaluate this possibility, we tabulated the spectrum of small indels in the CesA1 region of the A_T and D_T genomes. The data reveal no evidence of an indel bias (Table 2; Fig. 4). For each indel bin, there were approximately the same number of indels accounting for a similar total of nucleotides. The maximum difference for any bin was 344 bp, which was counteracted through indels in other bins. Overall, the total difference in genome size attributable to small indels is a scant 120 bp (in D_T). These observations demonstrate the absence of a globally operating indel bias in Gossypium, despite evidence to the contrary in some other plants (Kirik et al. 2000; Orel and Puchta 2003).

Remarkable Conservation of Intergenic Space

Aside from several TE insertions and a single chloroplast insertion, most intergenic space between the A_T and D_T genomes is highly conserved. This contrasts with most other studies of microcollinearity, reflecting both the absence of major structural alterations in this genomic region (as discussed above) and perhaps the amount of time that has elapsed since the A and D genomes diverged from their last common ancestor. Yet reports from grasses suggest that ∼11 million years is sufficient to remove homoeology outside of genes (Ramakrishna et al. 2002b; SanMiguel et al. 2002), and, in some cases, only 0.5–1 million years is required (Wicker et al. 2003). Because the A_T and D_T genomes evolved in isolation on different continents for 5–10 million years prior to becoming reunited by polyploidy ∼1 Mya (Cronn et al. 2002; Senchina et al. 2003), one might not have been surprised by detecting a larger amount of intergenic divergence and lessened sequence identity. The remarkable conservation we observed indicates that the evolutionary forces and molecular mechanisms responsible for rapid intergenic divergence in other plant systems do not operate similarly in this region of Gossypium.

Conclusions

A large body of empirical evidence has demonstrated that myriad external forces and internal molecular genetic mechanisms are involved in the complex suite of phenomena that collectively mold genome size (Petrov 2001; Petrov and Wendel 2004). Recent technological advances in large insert libraries and high-throughput sequencing have made genomic comparisons accessible and feasible, thereby promising increasing application to nonmodel organisms. These comparisons will enable insights into the organization of genomes and their evolution, and are likely to be more informative when conducted within well-understood phylogenetic frameworks. The research described here represents a first step in this direction for Gossypium, which contains, in addition to the A and D genomes, other diploid groups (Fig. 1) whose genome sizes span an even greater range than the twofold size difference studied here. Extension of the present study to include more of this diversity, as well as to additional genomic regions, will enable us to more critically evaluate the suggestion of relative stasis in gene islands and conservation of intergenic sequence reported here. In this regard, the recent publication of a high-density genetic map for Gossypium (Rong et al. 2004) will facilitate targeted selection of genomic regions for analysis.

METHODS

BAC Library Screening and BAC Selection

A cotton (G. hirsutum L.) BAC library (Tomkins et al. 2001) was screened for clones containing a gene encoding cellulose synthase (CesA1), as previously reported (Tomkins et al. 2001). This gene was previously isolated and its sequence determined from A- and D-genome diploid cottons as well as from both genomes of polyploid cotton (Senchina et al. 2003), which facilitated identification of the genomic origin of each BAC. PCR and sequencing were used to verify the presence of CesA1 and to determine which homoeolog of the tetraploid (A_T or D_T) was represented by each BAC screened. The largest clone from the D_T genome (BAC clone 106I22) was sequenced to completion first. Following contig assembly, candidate A_T BACs for comparison were evaluated for maximal overlap with the sequenced D_T BAC, using a combination of PCR screening of inferred genes (3 and 11; see Fig. 2) as well as BAC-end sequencing. Because the G. hirsutum BAC library was created using partially digested (HindIII) genomic DNA, some BAC ends were conserved and shared among homoeologs. Thus, an A_T clone that shared a BAC end sequence and tested genes with 106I22 (A_T BAC clone 155C17) was verified as providing maximum overlap for the region. This clone was then sequenced as described below.

Shotgun Sequencing, Assembly, and Analysis

BAC DNA was sheared using a HydroShear (GeneMachines) DNA shearing device at speed code 12 with 25 cycles at room temperature. Fragmented DNA was end-repaired using the “End-it” DNA end repair kit (Epicentre), separated on an agarose gel, and size-selected for a range of 2–6 kb. This prepared insert DNA was randomly cloned into a pBluescript II KS+ vector (Strategene) and sequenced with the universal vector primers T7 and T3 to an average depth of 8× (∼1152 clones in A_T and 1920 clones in D_T). The resulting sequences were base-called using the program Phred (Ewing and Green 1998; Ewing et al. 1998), vector sequences were removed by CROSS_MATCH (Ewing and Green 1998; Ewing et al. 1998), and assembled by the program Phrap (Green 1999). Contigs were visualized and edited with CONSED (Gordon et al. 1998). Potential genes were predicted by three independent programs: FGENESH (http://www.softberry.com/), GeneMark.hmm (Lukashin and Borodovsky 1998), and GENSCAN+ (Burge and Karlin 1997). Predicted proteins were used as input for BLASTP searches against the nonredundant GenBank protein database. To further investigate potential genes in the assembled sequence, 500-bp segments of each assembled BAC were subjected to BLASTX queries against the nonredundant GenBank protein database and BLASTN queries against the cotton EST database.

Alignment of the homoeologous BACs to each other was accomplished using LAGAN (Brudno et al. 2003). The resulting alignment was checked manually for errors using BIOEDIT (http://www.mbio.ncsu.edu/BioEdit/bioedit.html).

Preliminary mining for repetitive elements was accomplished through RepeatMasker (http://ftp.genome.washington.edu/RM/RepeatMasker.html), CENSOR (Jurka et al. 1996), and BLAST homoeology to known elements in RepBase (version 8.5; Jurka 2000). MITEs were mined using the program FINDMITE (Tu 2001) and querying the results for repetitiveness in the genome (J.S. Hawkins, H-R. Kim, R.A. Wing, J.D. Nason, and J.F. Wendel, unpubl.), as well as by searching for conserved Arabidopsis TIR and TSD sequences. Each potential MITE was inspected manually to ensure that the predicted TIRs were not composed primarily of simple sequence repeats that would generate a false prediction. In addition, each BAC was queried against itself in 500-bp fragments to reveal potentially missed repetitive elements. Finally, each BAC was again queried in 500-bp fragments against whole-genomic shotgun sequences characterized by an ongoing study (J.S. Hawkins, H-R. Kim, R.A. Wing, J.D. Nason, and J.F. Wendel, unpubl.).