Abstract

    Models for meiotic recombination based on Crick's "unpairing postulate" require symmetrical extrusion of stem-loop structures from homologous DNA duplexes. The potential for such extrusion is abundant in many species and, for a given single-strand segment, can be quantitated as the "folding of natural sequence" (FONS) energy value. This, in turn, can be decomposed into base order-dependent and base composition-dependent components. The FONS values of top and bottom strands in most C. elegans segments are close, as are the corresponding base order-dependent and base composition-dependent components; any discrepancies are in the base composition-dependent component. This suggests that the strands would extrude with similar kinetics. 

    However, interspersed among these segments and at the ends of chromosomes (telomeres) are segments containing short tandem repeats (microsatellites) which, by virtue of their high variability, have been postulated to inhibit the pairing of homologous chromosomes and hence drive speciation. In these segments there are usually wide discrepancies between the FONS values of top and bottom strands, mainly attributable to differences in base order-dependent components. Analyses of artificial microsatellites of different unit sizes and base compositions show that this asymmetrical distribution of folding potential is greatest for microsatellites when the units are short and violate Chargaff's second parity rule. 

    It is proposed that when there is folding asymmetry, recombination proceeds by special, strand-biased, somatic mechanisms analogous to those operating with Chi sequences in E. coli. If meiotic recombination in the germ-line requires extrusion symmetry, then a general inhibitory influence of microsatellite-containing segments could mask the antirecombinational influence of their variability. Thus, microsatellites may not have driven speciation.

Repetitive sequences form much of the genomes of organisms considered higher on the evolutionary scale and were postulated to play a role at the level of individual organisms (Britten and Davidson, 1971). However, the concept of "selfish DNA" led to the view that repetitive sequences might make "no specific contribution to the phenotype" (Orgel and Crick, 1980). Conceding this, it was argued that, since repetitive sequences tend to vary (a) more than unique sequences and (b) more between species than within a species, they "could act as a blunt instrument driving speciation by reproductive isolation alone, without reference to adaptation" (Robertson, 1981). Thus, repetitive sequences may be more important at the species, than at the individual, level. Sequence differences would impede the meiotic pairing of homologous chromosomes so impeding recombination and making interspecies hybrids infertile. Unable to continue the line through their hybrid, parents would be reproductively isolated from each other, and so would be deemed members of distinct species with the potential to pursue independent reproductive paths (Flavell, 1982). Consistent with this, it has been proposed that DNA sequence divergence ("chromosomal hypothesis"), rather than differences in gene function ("genic hypothesis"), underlies the majority of speciation events (Forsdyke, 2007a). We consider here whether divergence might be a general genomic property or might be driven by special non-genic sequences - namely, a class of highly variable repetitive sequence containing short tandem repeats (STRs) often referred to as "microsatellites" or "minisatellites" (Ellegren, 2004).

Aggregations between molecules may be broadly classified as like-with-like and like-with-unlike. Like can aggregate with like by virtue of shared regularities in structure and/or resonance frequencies, as when pure crystals appear within a mixture of solutes in solution, or virus coat proteins aggregate (Lauffer, 1975; Muller, 1941). Like can aggregate with unlike by virtue of complementarity of structures and compatible resonance frequencies. Ostensibly, the pairing of homologous chromosomes in meiosis appears as an example of like-with-like aggregation. Yet, at the molecular level, Crowther (1922) drew an analogy with a sword pairing with its scabbard (i.e. complementarity of structure). Decades later the discovery of the double-helical structure of DNA revealed complementarity (A pairing with T, and G pairing with C) as a fundamental feature of this major chromosomal component (Watson and Crick, 1953). 

In his "unpairing postulate" Crick (1971) proposed that to initiate meiosis the strands in a DNA duplex would unpair thus allowing cross-pairing of strands without strand breakage (paranemic pairing). In other words, the "sword" of one duplex would pair with the matching "scabbard" of the other, and vice versa. If scabbards did not match (absence of homology) the pairing would fail, hence avoiding commitment to strand-breakage and launching a potential speciation event. Seeking broad generalizations, Crick considered the unpaired DNA strands as single-stranded bubbles. He did not explore the possibility that, by virtue of their palindrome content, the unpaired strands might rapidly adopt higher ordered stem-loop structures. In other words, the "swords" in single stranded sequences might find local "scabbards" within their own strands, thus perhaps militating against the cross-pairing between strands that he had postulated.

Yet sequencing studies were already suggesting (since verified; e.g. (Forsdyke, 1995a) ) that the potential to assume local intra-strand stem-loop structures is abundant in nucleic acids of many species. Such structures usually contain sequences displaying a folding symmetry that can be described as "palindromic" since, at the duplex level, they read the same forward and backward. A duplex containing 5'AAAAAAAATTTTTTTT3' is considered palindromic, whereas a duplex containing 5'AAAAAAAACCCCCCCC3' is not (see below). A duplex with two complementary limbs separated by an intermediate region (e.g. 5'AAAAAAAAGGGGTTTTTTTT3') would form a stem and a loop (in this case G-rich in the top strand) and would be considered to display partial palindromicity. In 1974 Crick wrote:

While intra-strand pairing between complementary sequences that had adopted stem-loop configurations seemed to contradict Crick's model for inter-strand meiotic pairing, several more specific recombination models invoked intra-strand pairing, thus providing an adaptive explanation for the abundance of palindromes (Doyle, 1978; Sobell, 1972; Wagner and Radman, 1975). Consistent with the latter models, Tomizawa demonstrated for RNA that inter-strand pairing could follow exploratory "kissing" interactions between the loops in complementary single-strand molecules. Furthermore, Kleckner and Weiner (1993) suggested that the Tomizawa mechanism might apply to the pairing between stem loops extruded from the DNAs in homologous chromosomes (Zickler, 2006). Small differences in GC% (i.e. no violation of Chargaff's second parity rule; see section 3) would suffice to disrupt pairing and impede recombination (Forsdyke, 2006; 2007a). There is a well-known association of recombination with palindromes (Leach, 1994), and a role for stem-loop intermediates has been suggested from studies of recombination in HIV and in its coreceptor gene (Zhang et al., 2005a; 2005b). Recent studies have shown that, despite the presence of heterologous duplexes, homologous DNA duplexes can aggregate in simple salt solution in the absence of proteins. However, whether stem-loop intermediates are involved is undetermined (Baldwin et al., 2008).

A prediction of the stem-loop models is that complementary kissing interactions would be favored if the "sword" strands and "scabbard" strands of a DNA duplex were extruded with similar kinetics (Forsdyke, 2006; 2007a). As an indicator of this possibility, the "top" ("W") strand of a duplex might show the same propensity to fold as its complementary "bottom" ("C") strand. There would then be symmetry between the folding potential of top and bottom strands. We here report studies with the nematode worm C. elegans (supported by our unpublished studies with other organisms) that show this is generally true. However, the symmetry is often lost in microsatellite-containing regions, raising the possibility that, by virtue of the asymmetry, microsatellites might exert a general antirecombinational effect. Yet, microsatellites are thought to promote, not impede, recombination (Ellegren, 2004). We explore here the basis of this apparent paradox. 

For the nuclear genomes of most biological species Chargaff's first parity rule for duplex DNA (A = T, G = C) also applies, to a close approximation, to single stranded DNA (Chargaff's second parity rule; Forsdyke and Mortimer, 2000; Phillips et al., 1987; Prabhu, 1993; Rogerson, 1989). Thus, the AT-rich DNA sequence 5'AAAAAAAATTTTTTTT3' obeys Chargaff's second parity rule and has the potential to fold into a stem-loop (hairpin-like) configuration (Gierer, 1966). In an antiparallel duplex containing this sequence in its "top" (W) strand (as listed in GenBank), the corresponding "bottom" (C) strand, 3'TTTTTTTTAAAAAAAA5', has the same base composition and the same potential to fold into a stem-loop configuration. Thus, there is symmetry in the folding energies of top and bottom strands. Furthermore, the two strands would display the same buoyant density when subjected to centrifugation in alkaline CsCl gradients. In contrast, the AC-rich sequence 5'AAAAAAAACCCCCCCC3' (top strand) does not obey the second parity rule, has little potential to fold, and has a different base composition from the corresponding GT-rich bottom strand, 3'TTTTTTTTTGGGGGGGG5'. However, since G can form a weak Watson-Crick base-pair with T (Allawi and SantaLucia, 1998), the latter sequence retains some potential to fold. Thus, there is asymmetry in the folding energies of top and bottom strands. By virtue of their composition differences, the two stands display different buoyant densities in alkaline CsCl gradients. When banded in such gradients one strand can be regarded as a "satellite" of the other. Indeed, a significant class of repetitive sequences was discovered by virtue of this satellite property (Flamm et al., 1969). Base composition and order both influence folding symmetry. Thus, the degrees of symmetry/asymmetry would be modified if the orders of the bases were less simple than shown above (e.g. if each sequence were shuffled).

One strand of many microsatellite sequences being GT-rich in violation of Chargaff's second parity rule, the corresponding duplexes would be expected to display folding asymmetry and hence impede meiotic pairing (see Section 2). Yet, paradoxically, GT-richness appears to favor recombination. In prokaryotes recombination may be associated with "GT-rich islands" (approx. 1 kb) that contain strand-specific "crossover hotspot instigator" sequences - Chi sequences (GCTGGTGG; E. coli) or Chi-like sequences (GNTGGTGG; H. influenzae; Bell et al., 1998; Lao and Forsdyke, 2000; Tracy et al., 1997). For their function Chi sequences require a GT-rich base composition, but the order of bases is also important. Specific GT-rich sequences (GGGGCTGGG) embedded in GT-rich regions are also important in the strand-biased recombination events that allow immunoglobulin class switching in higher eukaryotes. Here folding asymmetry would facilitate access of the mutagenic deoxycytosine deaminase to the AC-rich strand (Huang et al., 2007). The role of GT-richness in germ-line recombination is less clear (Majewski and Ott, 2000; Wahls, 1998). Genetic studies that relate high linkage to low recombination between polymorphic loci indicate that in recombination hot spots "there exist multiple fuzzy DNA sequence determinants … based on the nature of the allele present" (Nishant and Rao, 2006). Our approach is less specific in that we study DNA, as DNA, irrespective of the underlying function.

    For a given duplex DNA segment, the potential for a strand to be extruded to adopt a stem-loop structure can be quantitated as the "folding of natural sequence" (FONS) value of that structure. This, in turn, can be decomposed into a base order-dependent component (quantitated as the "folding of randomized sequence difference;" FORS-D) and a base composition-dependent component (quantitated as the "folding of randomized sequence mean;" FORS-M). These terms refer to methods of calculation from the energy values of computer-folded sequences (Mathews and Turner, 2006; Zuker, 2000). The average value of several independently shuffled versions of a sequence (base order having thus been eliminated) provides the base composition-dependent component. This value is subtracted from the corresponding value for the total folding energy (FONS) to obtain the base order-dependent component. Whatever the relative contributions of base order and composition, when the FONS values of top and bottom strands are close to zero, classical duplex forms would be expected to predominate in a DNA segment. High negative values would be expected to stabilize extruded stem-loops, a process that would be favoured by negative supercoiling (Forsdyke, 1995b; 1998; Wang et al., 1990). Misconceptions concerning the shuffling of sequences at the dinucleotide level, rather than at the level of single bases (mononucleotides), are discussed elsewhere (Forsdyke, 2007b; Xu et al., 2007).

Folding energy values were determined for 200 nt segments ("windows") from the genome of C. elegans that were moved in steps of 50 nt along a sequence, using our program "Random_fold_scan," with base order being randomized at the mononucleotide level by shuffling (Xu et al., 2007; Zhang et al., 2008). A representative result is shown in Figure 1.

Total folding energy values fluctuated widely (Fig. 1a), and these fluctuations reflected changes in the base order-dependent component of the folding energy (Fig. 1b) more than the base composition-dependent component (Fig. 1c). The folding energies of top and bottom strands followed each other closely. However, there were small differences. Sometimes the top strand had the greatest folding potential (more negative FONS value), and sometimes the bottom strand had the greatest folding potential. The similarity between the two strands was also reflected in the values for FORS-D (Fig. 1b) and FORS-M (Fig. 1c). As described previously for various biological species, whereas FONS and FORS-M values were always negative, some FORS-D values were positive. Sometimes base order and base composition work together to generate the total folding energy (FONS) value and sometimes they work in opposition (Forsdyke, 1998; Xue and Forsdyke, 2003) .   

To investigate other chromosomes, 600 windows (200 nt) were randomly selected from each chromosome (software for this selection is available on request). Table 1 shows that no significant overall differences were observed between the mean FORS-D values of top and bottom strands of each chromosome. Whereas the autosomes (chromosomes I to V) had mean values around -4.32 kcal/mol, the X chromosome had a significantly lower mean value (-2.35 kcal/mol; P<0.0001). Thus, on average, the X chromosome of C. elegans displayed lower base order-dependent stem-loop potential than autosomes. For each chromosome the average of the absolute differences between individual paired FORS-D values of top and bottom strands was significantly different from zero (Table 1; P<0.0001). The average X chromosome absolute FORS-D difference (1.74 kcal/mol) resembled that of chromosome I (1.73 kcal/mol), but differed significantly from the other autosomes (average difference 4.23 kcal/mol; P<0.0001).    
    In the case of the base composition-dependent component (FORS-M), although there were small, but significant, absolute differences, there were no significant overall differences between the FORS-M values of top and bottom strands (Table 1). However, in this case the X chromosome resembled chromosome I both in its mean folding value and in the average of the absolute differences. These differences between individual chromosomes were not further investigated.

The symmetry of folding energies between the two strands was lost in telomeric regions which contain STRs (Cangiano and La Volpe, 1993). Figure 2 compares the two telomeric regions of chromosome I with a region randomly selected from the middle of the chromosome. In the telomeric regions the strands containing GT-rich repeats (bottom strand in Fig. 2a and top strand in Fig. 2c) had greater total folding energy (FONS) than the AC-rich complementary strands. This was due to differences in base order (Figs. 2d, f) more than in base composition (Figs. 2g, i). In contrast, relative symmetry was displayed by a short 144 nt segment in a telomeric region that was devoid of GT-rich STRs (Figs. 2a, d, g), and by the central non-telomeric region (Figs. 2b, e, h). Similar results were obtained with the telomeres of the other five C. elegans chromosomes (data not shown).

The symmetry of folding energies between the two strands was also lost in many internal regions containing STRs (with repeat unit sizes starting at 2 nt). A sample of these regions was obtained from chromosomes I and II using Tandem Repeats Finder (Benson, 1999). STR-containing segments were required to have individual repeats that (i) were short (less than 13nt), (ii) closely matched the other repeats in the segment (>90%), and (iii) had a collective length of at least 200nt. These were selected for fold analysis together with their flanking sequences. The results showed that some, but not all, of these internal STR sequences displayed the asymmetry in FONS (Figs. 3 a-f, h).  A region on chromosome II (bases 6183938-6186538), which contained the 7nt-STR unit ACGCTAT that had a 100% match with its companion repeats and only slightly violated Chargaff's second parity rule, did not display the asymmetry (Fig. 3g). Sometimes the top strand displayed the greatest folding propensity (more negative FONS value; Figs. 3 a, d, h), and sometimes the bottom strand (Figs. 3 b, c, e, f). Again, the asymmetries reflected differences in base order (FORS-D; Figs. 3i-p) more than in base composition (FORS-M; Figs. 3q-x).

Rather than continue with natural sequences, to identify general sequence characteristics that might confer large differences in folding values between top and bottom strands, we generated an artificial set of top strand STR-containing sequences that varied in repeat unit length (5 - 50 nt), base composition, and base order. To avoid duplication in or between different groups, these STR sequences had to meet the following criteria: (i) Ability to fold using RNAstructure 4.2 with local data files for DNA (Mathews and Turner, 2006). (ii) Absence of internal tandem repeats. A repeat unit such as 5'ATGAATGA3' would not qualify since it contains two shorter repeat units. (iii) No cyclic permutability. For all possible tandem repeat sequences with the same base order but different start nucleotides (e.g. 5' GGACTAAT3', 5' GACTAATG3', 5' ACTAATGG3' and 5' CTAATGGA3'), only one can be selected. (iv) No reverse complementarity. If two sequences are reverse complementary (e.g. 5' GGACTAAT 3' and 5'ATTAGTCC 3'), only one can be selected. A program (written with ActivePerl-5.8.8.820) that can generate these sequences when given the size of repeat units and their duplication number, is available on request. Previous studies having shown the utility of 200nt windows sizes, here we used the program to generate STRs of this size. A 200 nt sequence window might contain 20 copies of a 10 nt repeat unit, or 4 copies of a 50 nt repeat unit. For a given size of repeat unit a total of 600 STRs were generated randomly. For 5nt and 6nt repeat units only 58 and 220 sequences, respectively, satisfied the above criteria. STRs with smaller repeat units were not generated.

600 top strand FORS-D values for such computer-generated STRs were hierarchically ordered and plotted with the corresponding bottom strand values (Fig. 4). In some cases the bottom strand values were more negative than top strand values. In other cases the bottom strand values were less negative than top strand values. Thus, while the top strand values gave smooth lines (because of the hierarchical ordering), bottom strand values gave irregular lines. These irregular fluctuations were greater in the case of STRs with shorter subunits (Fig. 4a), and were least in the controls (a set of natural sequences; Fig. 4c). Studies with other subunit sizes indicated progressively greater fluctuations as subunit size decreased (data not shown). The asymmetries were much less evident in the case of FORS-M values (Figs. 4d-f).

The signs of the fluctuations were eliminated by taking absolute differences between top and bottom strand FORS-D values, which were again hierarchically ordered (Fig. 5). When compared with corresponding absolute FORS-M differences, that were relatively constant, absolute FORS-D differences of STR sequences exhibit larger values and great variation (Figs. 5a, b).

For 600 computer-generated 10 nt repeat unit STRs, the influence of top strand base composition (e.g. [G + C]% written as GC%) on the difference between top and bottom strand FORS-D and FORS-M values was investigated by first order linear regression. Differences in GC% (i.e. Chargaff's second parity rule was not violated) affected the differences between the strands neither of the STR-containing sequences (Figs. 6a, 7a), nor of the control series of 600 natural 200 nt sequences (Figs. 6d, 7d). Thus, there were no differences in FONS values (the sum of FORS-D and FORS-M values). Stem-loop extrusion from duplexes would be symmetrical both at low and high GC% values.

However, differences in AG% (that would reciprocally correspond to bottom strand CT%), and especially differences in GT% (that would reciprocally correspond to bottom strand AC%), were correlated with large differences between both the FORS-D values and FORS-M values in the case of the computer-generated STR-containing sequences (Figs. 6b,c; Figs. 7b,c). Thus, this group of sequences included sequences that, by virtue both of their base composition and order, would have differed widely in the FONS values of their complementary strands. Stem-loop extrusion from such duplexes would have been increasingly asymmetrical as their A+G or G+T content increased.

In the case of the natural sequences, differences in AG% and GT% either did not correlate with differences in strand FORS-D values (Fig. 6e), or correlated very weakly (Fig. 6f). On the other hand, differences in AG%, and especially differences in GT%, correlated well with differences in strand FORS-M values (Figs. 7e, f). Thus, total folding energy values (FONS) would differ between top and bottom strands as the A+G or G+T content of the top strand of a segment increased, but this would mainly reflect differences in base composition. The particular computer-generated sequences which, by virtue of their base orders, accounted for the differences between FORS-D values (Figs. 6b,c) would either not have been present, or would have been present at very low frequencies, in the natural sequences (Figs. 6e,f). Thus, for natural sequences it is usually differences in base composition alone (AG% or GT%), that correlate with asymmetrical folding of top and bottom strands. Differences in GC% do not violate the second parity rule so folding is symmetrical.

For most sequence segments violations of Chargaff's second parity rule are minimal, and the strands of a duplex have the potential to adopt single-strand stem-loop configurations in a symmetrical manner (Fig. 1; Table 1). In this circumstance, recombination could be in accord with models for germ-line meiotic strand pairing based on Crick's unpairing postulate. The major disruptor of the pairing between two independent duplexes would be differences in GC% between the two duplexes, the individual strands of each duplex being extruded symmetrically by virtue of their common GC% (Forsdyke, 2006; 2007a). Where symmetry failed it would be due to strand differences in certain bases (i.e. AG% and/or GT%), irrespective of their order (Figs. 6, 7). Thus, distinct sequences would not be involved. However, the symmetry would fail in telomeric regions (Fig. 2) and in other regions containing STRs (Fig. 3), due to both the composition and the order of bases (Figs. 6, 7). Distinct sequences would be involved. The failure would be greatest when the STRs were small and violations of Chargaff's second parity rule were maximal (Figs. 4, 5; Table 2).

In view of the highly specialized nature of telomeres (Verdun and Karlseder, 2007) and the fact that asymmetry between top and bottom strands can be associated with specialized forms of recombination (Huang et al., 2007), it seems likely that telomeric and interspersed regions containing STRs (Figs. 2, 3) are, in some way, specialized for functions involving somatic recombination, and contain distinct sequences adapted for this. In such regions recombination would be highly regulated, requiring specialized proteins analogous to some of those mediating recombination in the GT-rich islands containing Chi sequences in E. coli.

For example, the telomeric DNA of eukaryotes such as C. elegans has a single-stranded GT-rich 3' extension that would only weakly bond with itself. As such, under the influence of a variety of specialized proteins (Raices et al., 2008) it might readily invade a neighboring duplex with its identical strand being displaced as a "D-loop." The resulting "T-loop" might decrease telomere erosion. Alternatively, the single stranded form might more readily engage in a recombination-dependent form of telomere regeneration known as ALT ("alternative lengthening of telomeres"; Verdun and Karlseder, 2007). It should be noted that, depending on salt concentration, G-rich regions have the potential (i) to aggregate (Forsdyke, 1984) with the formation of G-quartets (Henderson et al., 1987; Williamson et al., 1989) , and (ii) to form Z-DNA (Haniford and Pulleyblank, 1983). The folding programs we have employed (Mathews and Turner, 2006; Zuker, 2000) do not take such possibilities into account.  

Apart from a somatic role, their association with meiotic recombination hotspots suggests a role for STRs in the germ line. That the association is indicative of a role in the termination, rather than initiation, of meiotic recombination has been suggested by Wahls (1998); some characteristic of satellites is held to arrest the branch migration that may follow formation of Holliday junctions. Thus, recombination could initiate in a flanking non-satellite region and terminate within the satellite due to some inhibitory influence. Perhaps the enzymes involved would sense the potential extrusion asymmetry. 

In summary, extrusions of higher ordered structures from DNA duplexes vary between the extremes of close symmetry and highly asymmetry. We have argued that a germ-line event (the initiation of divergence into distinct species) is likely to be influenced by meiotic recombination when there is symmetry of strand extrusion from DNA duplexes (i.e. Chargaff's second parity rule applies). If this is true then, by virtue of their extrusion asymmetry, regions adapted for special forms of somatic recombination might be less favorably adapted for the strand pairing that initiates meiotic recombination (for an alternative view see Rockmill and Roeder, 1998). High variability (associated with microsatellite content) might then fail to register in terms of the increased pairing incompatability between homologous chromosomes that is expected to precede speciation. In this circumstance, and without excluding roles for other classes of repeat sequence, a role for microsatellites as drivers of speciation is in doubt.