Introduction

Without mutational or selective pressures on the composition of the two strands of DNA in genomes, the base frequencies within each strand will be at an equilibrium of A = T and G = C (Lobry 1995; Sueoka 1995; Lobry and Lobry 1999). Deviation from these rules (usually called parity rules type 2, i.e., PR2 [see Frank and Lobry 1999]) means that two DNA strands experience different mutational and/or selective pressures, which leads to asymmetric nucleotide substitution (Kowalczuk et al. 2001). There are two ways of studying strand asymmetry: detection of the deviations of nucleotide composition from PR2, and phylogenetic analysis of base substitution pattern (Frank and Lobry 1999).

The compositional asymmetries between leading and lagging strand have been observed in many eubacterial, organelle and viral genomes (Lobry 1996a; Blattner et al. 1997; Fraser et al. 1997; Kunst et al. 1997; Grigoriev 1998; McLean et al. 1998; Mrazek and Karlin 1998; Grigoriev 1999; Rocha et al. 1999; Tillier and Collins 2000; Rocha and Danchin 2001; Shioiri and Takahata 2001). The leading strands are usually richer in G than in C and in T than in A. Similarly, there is much evidence that certain substitutions occur with different frequencies on the leading and the lagging strands of bacterial and organelle genomes (Fijalkowska et al. 1998; Reyes et al. 1998; Rocha and Danchin 2001). The asymmetries in prokaryotic genomes are so strong that they can be used to search for the origin and terminus of DNA replication (Grigoriev 1998; Picardeau et al. 2000; Zawilak et al. 2001).

In eukaryotic genomes, however, such studies are limited and yield inconsistent results. Most analyses of eukaryotic or archaeal genomes did not show strand compositional asymmetry on large scales (Grigoriev 1998; Mrazek and Karlin 1998; Gierlik et al. 2000). However, the examination of three contigs from human chromosomes gave some evidence of strand compositional asymmetries in human genome (Shioiri and Takahata 2001). In addition, local asymmetries have been found in the telomere regions of Saccharomyces cerevisiae chromosomes (Grigoriev 1998; Gierlik et al. 2000). Besides the deviations of nucleotide composition from the PR2, studies on the nucleotide substitution patterns in eukaryotic genomes did not give consistent results. Early studies found small local asymmetries in the beta-globin region of six primate genomes (Wu and Maeda 1987). But in later studies, neither the examination of longer sequences (Bulmer 1991) nor the analysis based on the exact position of replication origin within the beta-globin region (Francino and Ochman 2000) showed the existence of strand asymmetry.

One difficulty in the study of DNA strand asymmetry in eukaryotic genomes is that most replication origins are unknown (Francino and Ochman 2000). It is virtually impossible to detect the deviation from PR2 or to analyze the substitution pattern phylogenetically without the knowledge of the origin positions of DNA replication. Some methods, e.g., cumulative skew diagrams (Grigoriev 1998) and DNA walk (Lobry 1996b), can be used to show the strand asymmetries of prokaryotic genomes without referring to the replication origin positions. Unfortunately, for reasons discussed below, such methods did not lead to conclusive results in eukaryotes.

In DNA synthesis, the template of the lagging strand (i.e., the leading strand in the previous DNA synthesis process) stays in the single-stranded state for a longer period than its counterpart. Similarly, the coding strand (equivalent to the RNA produced) of a gene is exposed for a longer time as the single strand in transcription than the noncoding strand (serving as the template in transcription). The rate of deamination of cytosine and 5-methylcytosine residues in single-stranded DNA is much greater than that in double helix (Frederico et al. 1990; Beletskii and Bhagwat 2001). Deamination of cytosine and 5-methylcytosine generates uracil and thymine. Replication of these deaminated products will produce C-to-T transition. So the lagging strand and the coding strand are expected to experience a higher pressure of C-to-T mutation. The differences in the processes and enzymes involved in the synthesis of leading strand and lagging strand may lead to some differences in the accuracy of replication (Kunkel 1992; Izuta et al. 1995; Gawel et al. 2002). The transcription-coupled repair induced by RNA polymerases stalled at lesions on the template strand decreases the G-to-A substitution in the noncoding strand (Francino and Ochman 2001).

The method of DNA walk is simply to assign a direction of a virtual walker to each nucleotide. The trace indicates relative cumulative abundance of A over T and G over C in the analyzed sequence (Lobry 1996b). Cumulative skew is a sum of the AT skew or of the GC skew in adjacent windows from arbitrary start to a given point in a sequence. The diagram also indicates the relative cumulative abundance of the nucleotides (Grigoriev 1998). In both cases, the distinct turning point usually coincides with the replication origin of chromosome. If replication-associated mutation pressure is the most important cause of the strand asymmetry (Tillier and Collins 2000), the line of cumulative skew or of DNA walk is expected to be smooth, and the turning point is distinct. However, in the cases where the effect of transcription on the asymmetry is comparable to that of replication, the issue becomes much more complicated. In human papillomavirus, where all the genes are transcribed from one strand while the DNA replication is bidirectional, only half of the cumulative skew diagram shows clear trend (Grigoriev 1999).

The Okazaki fragments in nuclear DNA replication are much shorter than those in prokaryotic DNA replication. The replication-associated mutation pressure resulting from single-strand cytosine deamination is expected to be much weaker in eukaryotes than in prokaryotes. In addition, the replicons of eukaryotes are much shorter than that of prokaryotes (Lewin 1997). If a replicon contains only several genes randomly distributed on the two strands, and if the effect of transcription is comparable to that of replication, a general trend or a turning point across the replication origin in the cumulative skew diagram or DNA walk seems unlikely. Therefore, the aim of this paper is to propose a new method with which we can determine if the strand asymmetries exist in eukaryotic genomes.

Eukaryotic chromosomal DNA replication initiates at multiple origins, with an average replicon size ranging from 10 to 300 kb (Brewer and Fangman 1993; Lewin 1997; Hyrien et al. 2003). If two genes, like P and Q in Fig. 1, are closely located on a chromosome (say, the gene pair length is only 2 kb), it is very likely that they are located at the same side of a bidirectional replicon. Furthermore, if they are coded from two different strands, it is expected that one is coded from the leading strand while the other from the lagging strands of the same replicon. To examine the nucleotide compositional skews of such a pair of genes is to study the asymmetries of leading and lagging strands of the same replicon. In this way, the effects of replication and transcription on strand asymmetry can be separated.

Figure 1
figure 1

Gene P and gene Q are closely located on different strands. The arrowed lines between the two strands show the directions of transcription and the mRNA transcribed. Gap: sequence between the two start points (or two endpoints) of the gene pair; downstream (or upstream) elements like promoters of nearby genes may be included. Gene pairlength: the total length of P, Q, and their gap. a Tail-to-tail gene pair. b Head-to-head gene pair.

Full size image

Recent advances in genome sequencing make it possible to study the strand compositional asymmetries of eukaryotic nuclear DNA at chromosomal or genomic scale. In this paper, we make use of all the gene pairs defined in Fig. 1 from the genomes of several model eukaryotic organisms, including Saccharomyces cerevisiae, Schizosaccharomyces pombe, Plasmodium falciparum, Encephalitozoon cuniculi, Arabidopsis thaliana, Caenorhabditis elegans, Drosophila melanogaster, Anopheles gambiae, Mus musculus, and Homo sapiens. In order to eliminate the selective effects on strand compositional asymmetries (Frank and Lobry 1999), we examine the introns of the gene pairs. As the fourfold degenerate codons are thought to experience weak selection (Graur and Li 2000), we also study the skews of the fourfold degenerate sites of codons in the gene pairs. Strand compositional asymmetries are found in most genomes we analyze, in which both replication and transcription play a role.

Materials and Methods

The complete genomes of S. cerevisiae (Goffeau et al. 1996), S. pombe (Wood et al. 2002), E. cuniculi (Katinka et al. 2001), A. thaliana (The Arabidopsis Genome Initiative 2000), C. elegans (The C. elegans Sequencing Consortium 1998), D. melanogaster (Adams et al. 2000), and A. Gambiae (Holt et al. 2002), and all the available contigs of H. sapiens (International Human Genome Sequencing Consortium 2001) and M. musculus (Mouse Genome Sequencing Consortium 2002) were downloaded from NCBI’s genome database (ftp://ncbi.nlm.nh.gov/genomes/). The sequences of E. cuniculi and H. sapiens were downloaded in October 2002; other sequences were downloaded in September 2002. The complete genome of P. falciparum (Gardner et al. 2002) was downloaded from PlasmoDB (http://plasmodb.org ) in October 2002.

In previous studies, the window sizes strongly affect the results (Lobry 1996a; Grigoriev 1998; Shioiri and Takahata 2001). Similarly, the gene pair length (Fig. 1) is also critical in this study. If the gene pair length is too large, the two genes may be located on different side of a replicon or even on different replicons, and the strand asymmetries, if any, become undetectable. On the other hand, if the gene pair length is too small, stochastic noise may be so strong that the pattern is difficult to be detected with sufficient statistical power. To maximize the likelihood that both partners of any gene pair are located at the same side of a bidirectional replicon, the gene pair length must be much smaller than the length of a replicon. However, shorter gene pair length means less data to analyze. This represents a difficult compromise that we nonetheless have to strike. On the basis of many trials, we finally set the gene pair length to be 1.5 to 9 kb for human and mouse and 1.5 to 3 kb for other species according to the estimated sizes of eukaryotic replicons (Brewer and Fangman 1993; Lewin 1997; Hyrien et al. 2003). For those genes involved in two or more gene pairs, only the shortest gene pair was retained to ensure the independence of samples.

If the gap of a gene pair (Fig. 1) is too short, upstream (or downstream) regulatory elements (e.g., promoter) of a gene are likely to locate at the coding sequence of the other. To ensure the neutrality of the intron sequences and the third positions of the fourfold degenerate codons, those gene pairs with a gap less than 400 bp were discarded. The first 6 nucleotides at 5′ end and the last 30 nucleotides at the 3′ end encompassing the splicing sites in the introns, which may be subject to the action of some selective forces, were removed. Those genes with total intron length less than 40 nucleotides after the removal of splicing sites were excluded to minimize the stochastic noise. The fourfold degenerate codons are defined as GCX (Ala), CCX (Pro), TCX (Ser), ACX (Thr), CGX (Arg), GGX (Gly), CTX (Leu), and GTX (Val). Thus, for Leu, Ser, and Arg, we used four of their six codons. Those genes with less than 40 fourfold degenerate codons were eliminated to increase the statistical confidence.

The genomes of S. cerevisiae, E. cuniculi, and P. falciparum, with an insufficient number of suitable gene pairs (i.e., less than 20, from the viewpoint of the chi-square test), were excluded from further analysis of introns. Similarly, the genome of P. falciparum was not included in further analysis of the fourfold degenerate codons. The numbers of gene pairs analyzed in this study are given in Figs. 2 and 3.

Figure 2
figure 2

Scatter diagrams for the strand asymmetries found from introns. The X axis represents the skews of the introns of P genes, and the Y axis represents the skews of the introns of Q genes. Each point in the diagrams corresponds to a gene pair as shown in Fig. 1.

Full size image
Figure 3
figure 3

Scatter diagrams for the strand asymmetries found at the fourfold degenerate sites of codons. The X axis represents the skews of the introns of P genes, and the Y axis represents the skews of the introns of Q genes. Each point in the diagrams corresponds to a gene pair as shown in Fig. 1.

Full size image

Analogous to the studies in bacteria, the strand compositional asymmetries were estimated by the GC skew and AT skew of the leading strand and the lagging strand of replication, and by the differences of GC skew and AT skew between the leading strand and the lagging strand (Lobry 1996a). For a given gene, we calculated the GC skew and the AT skew with all the intron sequences of the gene as a unit, and those skews with all the fourfold degenerate sites of codons. The GC skew was defined as

and the AT skew was defined as

For lack of knowledge on the exact position of replication origin, we are not sure whether P or Q is coded from the leading strand. So, it is impossible to rearrange the gene pairs of certain species into leading strand group and lagging strand group. The classical method (Lobry 1996a) to examine whether GCS, ATS, and their differences between leading strand and lagging strand are significantly deviated from zero is not applicable to the present study. We used scatter diagrams to visualize the strand compositional asymmetries (Fig. 2 for those of introns and Fig. 3 for those of fourfold degenerate sites of codons). The skews of those coded from one strand were defined as variable P, and the skews of those coded from the other strand were defined as variable Q. Variable P was plotted along the horizontal axis, while variable Q was plotted along the vertical axis.

Results

Symmetry or Asymmetry

The scatter diagrams for the skews of the introns from the genomes of S. pombe, A. thaliana, C. elegans, D. melanogaster, A. gambiae, M. musculus, and H. sapiens are shown in Fig. 2. And those for the skews of the fourfold degenerate sites of codons from the genomes of S. cerevisiae, S. pombe, E. cuniculi, A. thaliana, C. elegans, D. melanogaster, A. gambiae, M. musculus, and H. sapiens are shown in Fig. 3.

Ideally, both ATS and GCS should be zero if there is no strand compositional asymmetry. However, random variations (e.g., errors in sequence data gathering and manipulation, small sample size) in ATS or GCS will result in their being randomly distributed among the four quadrants of the scatter diagram (Fig. 4a). Significant deviations from the random distribution suggest the existence of strand asymmetries. The results of χ2 test for Figs. 2 and 3 are presented in Tables 1 and 2, respectively.

Table 1 Results of χ2 tests for the intron skews and the relative importance of replication (r), transcription-associated mutation bias, and codon bias (q1 and q3) as the causes of the asymmetries in seven eukaryotic genomes studied
Full size table
Table 2 Results of χ2 tests for the skews of the fourfold degenerate sites of codons and the relative importance of replication (r), transcription-associated mutation bias, and codon bias (q1 and q3) as the causes of the asymmetries in nine eukaryotic genomes studied
Full size table
Figure 4
figure 4

Schematic diagrams. a Random distribution—strand symmetry. b Nonrandom distribution—strand asymmetry resulting from DNA replication. c Uneven distribution—strand asymmetry resulting from transcription-associated mutation bias and/or codon bias. d Uneven distribution—strand asymmetry resulting from codon bias.

Full size image

In M. musculus and H. sapiens, the ATS and GCS of both introns and the fourfold degenerate sites of codons are unevenly distributed in four quadrants, suggesting that there exist strand compositional asymmetries in the genome of human and mouse. In S. pombe and A. thaliana, the strands are significantly asymmetric except the GCS of the introns. In A. gambiae, significant nonrandom distributions are found for the skews of the fourfold degenerate sites of codons, but for introns they are only marginally significant (Table 1). In other species, however, there are large differences between ATS and GCS (Figs. 2 and 3, Tables 1 and 2). The D. melanogaster genome does not exhibit asymmetry in the light of ATS, while its strand asymmetry is significant or marginally significant in the light of GCS. The results of C. elegans are inconsistent (Figs. 2 and 3, Tables 1 and 2). For S. cerevisiae, the asymmetry exists only for ATS. In contrast, the E. cuniculi genome is significantly asymmetric only for the GCS.

Possible Causes of the Asymmetries

As shown in Fig. 4b–d, there are different causes of strand asymmetry. If the structural asymmetry of DNA replication is the main cause of compositional asymmetries, the skews of gene P (Fig. 1) should be similar to the opposites of the skews of gene Q (i.e., ATS P ≈ −ATS Q , GCS P ≈ −GCS Q ) due to the complementary nature of DNA double strands. Thus, most AT skews should fall into quadrant 2 and quadrant 4 in Fig. 4b; so do GC skews. Alternatively, if transcription-associated mechanisms are the major cause of the compositional asymmetries, the skews of gene P should be similar to those of gene Q (i.e., ATS p ≈ ATS Q , GCS p ≈ GCS Q ). Therefore, AT or GC skews should fall into quadrant 1 and quadrant 3 in Fig. 4c and d. If C-to-T or A-to-G mutation of the coding strand is the main cause, both ATS and GCS should be negative (Fig. 4c). Finally, if T-to-C or G-to-A mutation is the main cause, most skews should be in quadrant 1 in Fig. 4d. In addition to mutational pressure, selective codon bias may also contribute to the strand asymmetries (Frank and Lobry 1999). If A or C is preferred at the fourfold degenerate sites of codons, both AT and GC skews should be positive (Fig. 4d). If T or G is preferred, then both AT and GC skews should be negative (Fig. 4c).

Either codon bias or transcription-associated mutation bias could result in the distribution pattern of skews as shown in Fig. 4c and d. To show the codon usage bias, we calculated the genome-wide percentages of A, T, C, and G at the fourfold degenerate sites of codons from all studied genomes (Table 3). The percentages of the points in quadrant 1 and quadrant 3 in Figs. 2 and 3 are denoted as q1 and q3, respectively (Tables l and 2). To access the importance of replication as a cause of the strand compositional asymmetries, the proportion of the gene pairs that falls in quadrants 2 and 4 (Fig. 2) is introduced, and denoted r. The larger r the more important replication is to the asymmetry. The results are presented in Tables 1 and 2.

Table 3 The genome-wide usage of different nucleotides at the fourfold degenerate sites of codons in nine eukaryotic genomes studied
Full size table

In S. cerevisiae, r is 44.9%, q3 is 38.4%, and there is a small difference between the genome-wide usages of A (30.1%) and T (33.9%) at the fourfold degenerate sites of codons (Tables 2 and 3). Thus, all three factors (replication, transcription-associated mutation bias, and codon bias) may play an important role in causing the strand asymmetry.

In S. pombe, we focus on the AT skews only, since the intronic GC skew does not show a statistically significant departure from randomness. In this case, DNA replication contributes little to the strand asymmetry (Tables 1 and 2). The genome-wide A and T usages at the fourfold degenerate sites of codons are highly skewed (A = 27.3%, T = 42.4%; see Table 3). At first sight, codon usage bias may be regarded as the main cause of the strand asymmetry. But the values of q3 for the codons (70.3%) and for the introns (72%) are almost the same (Tables 1 and 2). Meanwhile, the proportions of A (30.5%) and T (40.3%) in all the introns of the examined gene pairs are comparable to that of genome-wide fourfold degenerate codon usage (A = 27.3% and T = 42.4%; see Table 3). Thus, transcription-associated mutation bias is more likely to be the major cause of the DNA strand asymmetry. Codon usage bias may result from either selection or mutation (Graur and Li 2000), while intron base composition bias can only result from mutation. The similarity in the magnitude of codon bias and intron base composition bias indicates that the genome-wide codon bias observed in S. pombe may result from transcription-associated mutation. Similar results were observed in prokaryotic genomes too (McLean et al. 1998). Since GC skews do not show a pattern similar with AT skews, transcription-associated C-to-T mutation bias cannot account for the observed results in S. pombe. As a consequence, it must be asymmetric A → T or A → C → T substitution that essentially causes the DNA strand asymmetry in S. pombe.

In A. thaliana, the distribution patterns of AT and GC skews are similar to those of S. pombe, therefore, we are led to the same conclusion, that the transcription-associated asymmetric A → T or A → C → T substitution may be the genuine cause of the asymmetry.

In E. cuniculi, the replication-associated mutation bias is the main cause of the DNA strand asymmetry observed from GC skew (Table 2).

In C. elegans and A. gambiae, no interesting conclusions could be drawn due to the poor statistical results.

In D. melanogaster, only the GCS of the fourfold degenerate sites of codons show statistically significant departure from randomness, which can be attributed to the selective codon usage bias.

In M. musculus and H. sapiens, we obtain similar results. Replication may contribute, to some extent, to the strand asymmetries (Figs. 2 and 3, Tables 1 and 2). For the GC skews of both M. musculus and H. sapiens, most skews of introns are in quadrant 3 (Fig. 2, Table 1), while most skews of the fourfold degenerate sites of codons are in quadrant 1 (Fig. 3, Table 2). The genome-wide usage of G and C at the fourfold degenerate sites of codons is highly skewed (Table 3). As the introns and the exons of the same gene are expected to experience similar mutation pressures, the differences in GC skews could be attributed only to selection on the synonymous codons. From the AT skews and the intronic GC skews, we can conclude that transcription-associated mutation biases (like C to T) may also contribute to the strand asymmetries observed.

Positive correlation between AT skews and GC skews has not been found in all the studied genomes. Instead, we find a significantly negative correlation in the fourfold degenerate sites of codons in S. cerevisiae, S. pombe, C. elegans, D. melanogaster, A. gambiae, M. musculus, and H. sapiens and in the introns of H. sapiens (data not shown). Therefore, we may make an assertion that there is some tendency to change the ratio of purine (or pyrimidine) at the fourfold degenerate sites of codons regardless of A or G (or between C or T).

Discussion

Previous studies on eukaryotic genomes did not show the strand compositional asymmetry on chromosomal scale (Grigoriev 1998; Mrazek and Karlin 1998; Gierlik et al. 2000), although distinct local asymmetry was observed in the telomere regions of yeast chromosomes (Grigoriev 1998; Gierlik et al. 2000). Current methods (cumulative skew diagram and DNA walk) show the asymmetry by the general trend and a major turning point across the replication origin. As we have shown, transcription-associated mutation bias and/or codon bias may have comparable effect on the asymmetry. Random distribution of genes on two strands is expected to produce local zigzags in the figures of cumulative skew diagram and DNA walk. Because eukaryotic nuclear replicons are very short (containing only a few genes), it is unlikely to find a distinct turning point across the replication origin from the zigzag noises. By means of a novel method proposed in this study, we demonstrate the existence of strand compositional asymmetries on large scale in most eukaryotic genomes studied.

In the studies of bacterial strand asymmetries, the results of AT skew and GC skew are not always identical, e.g., the GC skew tends to be much stronger than AT skew (Lobry 1996a; Blattner et al. 1997; Grigoriev 1998, 1999; Rocha and Danchin 2001). Because the AT skews are still significant in most cases, the difference between AT skew and GC skew deserves attention. In our study, the AT skew and the GC skew are not positively correlated but, rather, negatively correlated in some cases. Because base transitions (C ↔ T and A ↔ G) affect AT skew and GC skew equally, the strand bias cannot result only from cytosine deamination in single strand or differential repair in transcription (Grigoriev 1998; Frank and Lobry 1999; Francino and Ochman 2001; Rocha and Danchin 2001). Asymmetric C → G or C → A → G substitution was suggested to be a potential source of the difference between AT skew and GC skew in bacterial genomes (Rocha and Danchin 2001). Similarly, asymmetric A → T or A → C → T substitution may account for the difference between AT skew and GC skew in the genomes studied in this paper.

In prokaryotic genomes, replication-associated mutation pressure is the main source of the strand compositional asymmetries (Tillier and Collins 2000). Our study reveals that transcription-associated mutation bias and selective codon usage bias may cause the strand compositional asymmetries in eukaryotes. Similar evidence had been accumulated in bacterial and viral genomes. Asymmetric substitution pattern had been discovered in the transcribed E. coli sequences (Francino and Ochman 2001). In the strand compositional asymmetries of some viral genomes, the effect of transcription is also comparable to that of replication (Grigoriev 1999). Because of the segregation of germ cells and somatic cells in metazoans (Buss 1987), transcription-associated mutations in somatic cells cannot be transmitted to next generation. So the effect of transcription-associated mutation pressure on strand asymmetric substitution in metazoans may be underestimated by our method. One way to test this idea is to study the expression pattern of the genes that fall into different quadrants (Figs. 2 and 3). Unfortunately, only a few genes are well studied. The spatial-temporal expression patterns of most genes are still not clear now.

The cumulative skew diagrams of three contigs for human chromosomes gave evidence on strand compositional asymmetries in the human genome (Shioiri and Takahata 2001). Although the cumulative skew diagram is a powerful method to predict the origin position of DNA replication in mitochondrial or prokaryotic genomes (Grigoriev 1998), there is no convincing evidence for replication origin position in diagrams by Shioiri and Takahata. In addition, the window size they used for two human contigs is 100 kb, which is too large for most windows to correspond to sequences at one side of bidirectional replicons (Brewer and Fangman 1993; Lewin 1997; Hyrien et al. 2003). It seems reasonable to assume that the asymmetries that Shioiri and Takahata found result mainly from transcription.

If transcription is the main cause of strand compositional asymmetries in the human genome, it is then understandable why Francino and Ochman (2000) failed to detect the asymmetry of certain replicons by the phylogenetic method. The influence of replication may be canceled out by that of other processes occurring in the region, such as transcription and recombination. In addition, the asymmetric mutation bias may not be strong enough to overcome random genetic drift.

If both intron and the synonymous codons are selectively neutral, they are expected to experience similar mutation pressures. In this study, the skews of introns and fourfold degenerate sites of codons are generally similar except for the GC skews of D. melanogaster, M. musculus, and H. sapiens (Figs. 2 and 3). These exceptions can be attributed to the selective choices of the synonymous codons. Actually we found excessive cytosine at the fourfold degenerate sites of codons (Table 3). Thus, selective codon biases, along with replication- and transcription-associated mutations, may have played a role in the strand asymmetries of the nine genomes studied.

In order to enlarge samples, the upper limits of gene pair length in our study are set to be less than 7.5% of the estimated replicon size (Lewin 1997; Hyrien et al. 2003). Inaccurate data on replicon size and too long gene pair length may result in underestimation of replication effects on the strand compositional asymmetries.

Because of the segregation of germ and soma in metazoans, transcription-associated mutations in somatic cells have no chance to enter descendents, while in germ cells, both replication- and transcription-associated mutations can be inherited. Studies on the genes expressed only in somatic cells will give a clear picture of whether replication is involved in the asymmetry. Similarly, if we know precisely which genes are expressed in germ cells, we can separate the effects of transcription-associated mutation from that of replication-associated mutation on the compositional asymmetry. The difference between the compositional skews of somatic cell-expressed genes and germ cell-expressed genes will give a clear manifestation of the effects of transcription-associated mutation on the strand asymmetries. By examining the nucleotide substitution of the transcribed and untranscribed sequences of one orthologous region in nine mammalian genomes, Green and coauthors (2003) claimed that they had found transcription-associated mutational asymmetry. However, they did not mention any evidence about whether the genes they studied are expressed in germ cells.

Distinct local asymmetry was observed in the telomere regions of S. cerevisiae chromosomes in previous studies (Grigoriev 1998; Gierlik et al. 2000). It was suggested that relative excess of autonomously replicating sequences and random choice of these sequences in each replication cycle obscure the effects of replication-associated mutation pressure in other chromosomal regions. Fortunately, the origin sites of yeast chromosomal DNA replication have become increasingly clear (Raghuraman et al. 2001; Wyrick et al. 2001), and detailed analysis of the strand compositional skews and the differences between the leading strands and the lagging strands in S. cerevisiae may be a reality in the near future.