Cross-over hot-spot instigator (Chi)sequences (5'-GCTGGTGG-3') are orientation-dependent, strand-specific sequences implicated in RecA-mediated DNA recombination. In E. coli and H. influenzae Chi and Chi-like sequences preferentially locate to approx. 1 kb recombination "islands" in the mRNA-synonymous strands of open reading frames (ORFs).

    Since mRNA-synonymous strands follow Szybalski's transcription direction rule in being G-rich, and the average ORF is about 1 kb, then, on this basis alone, Chi sequences are seen to reside in 1 kb G-rich "islands". However, RecA preferentially binds GT-rich sequences, suggesting that genomic context might potentiate Chi action. Consist with this, we report for E. coli that 1 kb sequence windows with Chi near their centres are a distinct subset of total 1 kb windows, the mRNA-synonymous strands being preferentially enriched in both G and T.

    Chi function might be particularly important for bacteria which survive high temperature and radiation. These often exist in habitats where recombination with E. coli DNA would be unlikely, so canonical Chi sequences might not confer a selective disadvantage in this respect. In general, Chi sequences are not more frequent in thermophilic bacteria and D. radiodurans, than in E. coli and other mesophilic bacteria. Only two of five thermophilic bacteria examined showed preferential location of Chi sequences to mRNA-synonymous strands. In the thermophile M. jannaschii, windows containing the canonical Chi sequence do not form a distinct subset.

     We suggest that in thermophilic bacteria and D. radiodurans the Chi function may be achieved by sequences which differ from the canonical Chi sequence, or the number of these sequences is sufficient, or the Chi function is unnecessary.

Cross-over hot-spot instigator (Chi) sequences (5'-GCTGGTGG-3') are abundant, strand-specific, sequences, which locally increase recombination in Escherichia coli. Noting regularities in surrounding sequences, Tracy et al. (1997b) suggested that canonical 8-base Chi sequences in E. coli usually exist within approximately one kb GT-rich "recombination islands". A correlation of Chi orientation with transcription direction (Burland et al., 1993) suggested that the islands might also be "transcription islands", implying a relationship between transcription and recombination (Bell et al., 1998; Bell and Forsdyke, 1999a,b). Since in E. coli open reading frames (ORFs) form the main component of transcripts, the islands could also be considered "translational islands" (Biaudet et al., 1998; Colbert et al., 1998).

     Mutations creating or inactivating Chi activity occur only in Chi, not in surrounding DNA (Smith, 1997), and insertion of the Chi octomer into DNA enhances recombination in all cases reported (Dixon and Kowalczykowski, 1991; Dabert et al. 1992; Kuzminov et al., 1994). These observations imply that the octomer alone is sufficient for activity. However, recombination hotspots also occur in eukaryotes, where they may be influenced by chromosomal context (Ponticelli and Smith, 1992). Noting that RecA, a key enzyme influenced by Chi, binds preferentially to GT-rich DNA, it was suggested that the genomic context of Chi might potentiate its effectiveness (Tracy et al., 1997a,b). If this were so, then it is possible that one kb sequence "windows" with Chi sequences at their centres might form a distinct subset of the corresponding total set of one kb windows.

      This characteristic, together with a preference for mRNA-synonymous regions of DNA, could also indicate whether Chi or Chi-like sequences in other genomes are likely to be functional. Absence of the appropriate Chi function "password" targets an exogenous DNA for exonucleolytic degradation by the host cell RecBCD homolog (Kuzminov et al., 1994), thus militating against interspecies recombination, while permitting intraspecies recombination. Disruption of such interspecies barriers facilitates recombination (Rayssiguier et al., 1989). Just as a general (C+G)% similarity between the DNAs of members of a species may be necessary for the initiation of homologous recombination (Forsdyke, 1996), so exogenous DNA must have sufficient Chi sequence similarity with the DNA of host bacteria. However, the selection pressure promoting differentiation of sequences mediating the Chi function would be unlikely to be sustained between species occupying different ecological niches. This might explain why E. coli and H. influenzae have similar Chi sequences (Sourice et al., 1998).

    To examine these issues, we here present a study of

• (1) the frequency of Chi sequences in bacterial genomes,

• (2) their orientation with respect to transcription, and

• (3) their correspondence with a distinct subset of sequence windows of characteristic base composition.

Since Chi-dependent recombination could be involved in the repair of DNA strand breaks, the study included bacteria which both occupy ecological niches where recombination with E. coli would be unlikely (hence no selection pressure against the canonical Chi sequence), and survive under conditions where DNA strand breakage is increased (high temperature and radiation).

2.1. Genomic sequences

Sequence information refers to the "top" strand as designated in the GenBank record. Unedited segments of the Deinococcus radiodurans genome were obtained in 1997 from the Institute for Genome Research, Rockville, MD. Five of the longest segments, which had been designated gdr_5 (77466 nt), gdr_19 (68159 nt), grd_23 (50515 nt), gdr_31 (60954 nt), and gdr_34 (61120 nt), were presumed to originate from disparate parts of the genome, and hence were considered likely to be representative of the genome. Putative ORFs were determined using the Glimmer program.

2.2. Location of Chi sequences

    Chi sequences were located using the Findpatterns program of the Genetics Computing Group, Madison, WI. Since it is established experimentally that H. influenzae uses both the canonical and degenerate "Chi-like" sequences (Sourice et al., 1998), we sought such sequences (GNTGGTGG) and their top strand complement (CCACCANC) in this organism. In all other cases, we sought the canonical Chi sequence in the top strand, with Chis in the bottom strand being detected as the complement (CCACCAGC) in the top strand. Chi sequences were classified according to their location in leftward or rightward-transcribed ORFs. For E. coli and H. influenzae locations in non-protein-encoding genes (e.g. ribosomal RNA) were also sought in the examined segments, but were found only in H. influenzae.

    For each organism, the frequency of Chi sequences (Chi sequences/kb) expected if base order were random was calculated from the base composition of the combined genomic segments under study; this was the product of each base's statistical chance of occurrence in each Chi octomer, multiplied by the number of possible independent octomers in 1 kb (993). Thus, if P(A)=P(T)=0.2 and P(C)=P(G)=0.3, then the expected frequency for GCTGGTGG and CCACCAGC in a sequence is [[(0.3)⁶(0.2)²]+[(0.3)⁶(0.2)²]] x 993 = 0.058.

2.3. Chargaff difference analysis

     Chargaff's first parity rule for duplex DNA (%A = %T; %C = %G) applies, to a close approximation, to single stranded DNA (Chargaff's second parity rule). Deviations from parity are referred to as "Chargaff differences", which often depend on the "purine loading" of mRNA synonymous strands (providing an explanation for Szybalski's transcription direction rule; Bell and Forsdyke, 1999b). The base-composition of successive 1 kb windows, moved in steps of 0.1 kb, was assessed as described by Dang et al. (1998). Chargaff differences were calculated as (A-T)/W and (C-G)/S and expressed as percentages. Here, A, T, C and G refer to the frequency of the corresponding base in a window. The direction of subtraction (A-T or T-A) is determined alphabetically. W is the sum of the W base frequencies (A+T) and S is the sum of the S base frequencies (C+G). A window of 1 kb was chosen as the size which optimizes the difference between the Chargaff differences of a natural and the corresponding shuffled sequence (Bell and Forsdyke, 1999a). A 1 kb window selected by virtue of containing a Chi octomer near its centre would receive a contribution to the Chargaff difference from the octomer by virtue of the excess of Gs over Cs (4) and of Ts over As (1). Thus, for an organism of 50% (C+G), the G excess would contribute [4/500] x 100 = 0.8% to the Chargaff difference percentage value. Differences of this order would only marginally have affected results due to compositional differences in the entire 1 kb "island" surrounding a Chi sequence.

Due to base composition alone, the frequency of the G-rich Chi sequence would be expected to increase as genome (C+G)% increases. However, whatever the (C+G)%, in most bacteria examined the observed frequency of the canonical Chi sequence is greater than expected (Fig. 1; exceptions A. aeolicus, B. subtilis, T. pallidum). The points corresponding to observed and expected Chi frequencies fit simple exponential curves, with the prominent exception of E. coli. However, the observed/expected frequency ratio for E. coli (6.51) is of the same order as that of most other bacteria, and is slightly less than that of the thermophile M. jannaschii (8.20). For this organism with four other thermophiles the mean ratio is 3.20 se 1.28; the mean ratio of the eleven mesophilic bacteria studied is 3.32 se 0.52.

Higher than expected frequency of canonical Chi sequences in many bacterial genomes. Points for observed frequency (red circles) and for the frequency expected if base order were random (green circles) were reiteratively fitted to corresponding least-squares curves to arrive at the formulae:

Observed frequency = -1.049 + 0.2917e^{0.0681 x (C+G)%} (r² = 0.749; P<0.0001);

Expected frequency = -2.118 + 0.6867e^{0.04 x (C+G)%} (r² = 0.997; P<0.0001).

Organisms corresponding to each point are identified, with ratios of observed to expected frequencies in parentheses. segments examined are as in Table 1. The standard error of estimation for the curve fitted to the observed frequencies was 4.17, which gives a standardized residual for the E. coli outlier of >3.0 (indicating that its outlier status is statistically highly significant). For H. influenzae only data for canonical Chi sequences occurrence are shown.

ORF DNA    

A. aeolicus  

M. jannaschii

D.radiodurans

Quadrant analysis of base compositions (expressed as Chargaff differences) in 1 kb windows from the top strand of three segments from the E. coli genome, either combined (a, b), and independent (c-h).

     The centres of windows overlap either leftward-transcribed ORFs (a, c, e, g), or rightward-transcribed ORFs (b, d, f, h). Each quadrant corresponds to windows enriched for two particular bases (indicated at the corners). Each point corresponds to a window.

     Listed in each figure are the slopes (Sl) of the least square regression lines (shown as a solid diagonal line), the adjusted squares of the correlation coefficients (r²), and the probabilities (P) that slopes are not significantly different from zero. Only (c) and (f) had significant slopes (P = 0.023 and 0.002, respectively), when the regression analysis used every tenth window to avoid overlapping windows.

• (a) The 22 windows corresponding to top-strand Chi sequences, whose centres overlap leftward-transcribed ORFs.
  
• (b) The 52 windows corresponding to top-strand Chi sequences, whose centres overlap rightward-transcribed ORFs.
  
• (c) The 61 windows corresponding to bottom-strand Chi sequences, whose centres overlap leftward-transcribed ORFs.
  
• (d) The 17 windows corresponding to bottom-strand Chi sequences, whose centres overlap rightward-transcribed ORFs. Details are as in Fig. 2.

P^c

N^b

.

.

.

.

.

.

.

^a Chargaff differences (%) are presented together with the standard error of the mean (se).
   All data refer to the base composition of the top strand.

• (a) All 7494 windows overlapping leftward-transcribed ORFs.
  

• (b) All 8202 windows overlapping rightward-transcribed ORFs.
  

• (c) The 21 windows corresponding to bottom-strand Chi sequences, with those overlapping leftward-transcribed ORFs (14) shown as large open circles, and those overlapping rightward-transcribed ORFs shown as small open circles.
  

• (d) The 27 windows corresponding to top-strand Chi sequences, with those overlapping rightward-transcribed ORFs (18) shown as large open circles, and those overlapping leftward-transcribed ORFs shown as small open circles.

Regression lines are for points corresponding to leftward-transcribed ORFs (a, c), and to rightward-transcribed ORFs (b, d). Y₀ is the value of the intercept on the Y axis. Other details are as in previous figures. In (a) and (b) slopes remained significantly different from zero when every tenth point was employed for the regression analysis to avoid overlapping windows.

4.1. Mechanism of homologous recombination

    A new view, arising from studies of homologous recombination in eukaryotes, is that recombination follows an initial homology search, which may involve exploratory "kissing" interactions (Eguchi et al., 1991) between the tips of single-strand DNA loops extruded from intact DNA duplexes (Kleckner and Weiner, 1993; Forsdyke 1996). Only when such homology has been "tested" through formation of paranemic joints would there be a commitment to recombination (strand-breakage). The initial homology search might involve special regions of DNA, recognized as "hot-spots" or "islands of preferred DNA pairing" (Tracy et al., 1997b). Consistent with this, single-strand loops are gaining recognition as important intermediates in the initiation phase of recombination in bacteria (Anderson and Kowalczykowski, 1998). Bacterial RecA can connect two intact DNA duplexes, which should facilitate homology recognition (Leger et al., 1998; Zaitsev and Kowalczykowski, 1999), and a human RecA homolog has been shown to form homologous joints in the absence of net strand exchange (Gupta et al., 1999).

    In homologous recombination, an "island of preferred pairing" might suffice to guide certain enzymes of recombination to a region where subsequent strand breakage would occur. However, enzymes such as RecA are also involved in the repair of double-strand breaks. In this case, the initial break might occur at some distance from an "island", so that enzymes recognizing the break (e.g. the RecBCD complex) would have to translocate along the DNA towards an "island", where RecA and other enzymes would then be locally recruited.

    The "recombination island" proposal emerged at a time when Chargaff difference analysis was revealing genomes as consisting of distinct gene-sized compositional domains characterized by base compositions often in accord with Szybalski's transcription direction rule. However, in one organism (Drosophila melanogaster) which possesses a genome sectored into regions where genes are compactly arranged prokaryote-style and regions where genes are interrupted by introns mammalian-style, the compositional domains were found in both sectors (Dang et al., 1998). Indeed, introns and intergenic DNA of mammals show compositional domains (patterns of Chargaff differences) similar to those in protein-coding regions. This suggested that these domains might be related to some function other than the genic function, perhaps recombination (Bell and Forsdyke, 1999a,b).

4.2. Evolution of recombination

    Both the bacterial RecA protein and its eukaryotic homologs prefer GT-rich sequences (Tracy et al., 1997a). Furthermore, GT-rich loci are recombinogenic in various organisms (e.g. Jeffreys et al., 1985). Thus, Tracy et al. (1997b) proposed that Chi sequences initially evolved in GT-rich domains which already possessed a low level of intrinsic hot-spot activity. In support of this, they found sequence regularities extending 400 bases on either side of Chi sequences, suggesting a fundamental GT-rich "recombination island" size of about 1 kb, which is about the size both of the average ORF, and of sequence windows at which deviations from Chargaff's second parity rule are maximum when comparing natural with the corresponding shuffled sequences (Bell and Forsdyke, 1999a).

     Accordingly, it was proposed that early genomes evolved as a mosaic of regions of distinct base composition to facilitate, through loop-loop kissing interactions the interstrand and intrastand "accounting" necessary for recombination repair (Bell and Forsdyke, 1999b). From studies of "Chargaff differences" it was suggested that Chi was critically orientated with respect to transcription direction (Bell et al., 1998), and that the "recombination islands" of Tracy et al. (1997b) had become "recombination/transcription" islands in regions where transcriptional activity had evolved.

     The adaptive value of recombination was likely to have become evident in the early "RNA world" prior to the development of protein-synthesis (Bell and Forsdyke, 1999b). Thus, early genomes may have contained recombinational domains of varying base compositions, and the transcriptional function may have later imposed upon this recombinational mosaic, prior to the evolution of protein-encoding potential. While we do not know at what stage Chi functions evolved, it seems unlikely that the small Chi octomer was itself sufficient for early recombinational events. More likely, Chi-encoding potential and protein-encoding potential had to adapt independently to the preexisting compositional bias of recombination domains (as well as to the species-specific (C+G)% bias; Forsdyke, 1996).

4.3. Codon usage "explains"?

    In this light we should reexamine the work of two groups who recently criticised the "recombinational island" suggestion of Tracy et al. (1997b), on the grounds that the frequency and distribution of Chi can be "accounted for" (Colbert et al., 1998), or "explained" (Biaudet et al., 1998) by a preexisting codon usage bias. The "islands of preferred DNA pairing" are held to be merely "oceans of ORFs" (Colbert et al., 1998). However, on the basis of an initial study of a 136 kb segment of the E. coli genome, Burland et al. (1993) argued against an influence of codon usage. We would further argue that codon usage no more explains Chi frequency and distribution than that the disappearance of storks from the rooftops in Holland in the war years explained the concomitant decrease in the human birth rate. Just as these events were both likely to be due to the disturbance of warfare, so a more fundamental cause, the compositional bias of preexisting recombinational domains is likely to have created a genomic environment to which both Chi-encoding potential and protein-encoding potential had to adapt.

    Colbert et al. (1998) further argue that "Except for the sequence of Chi and its orientation-dependence, no deterministic features of Chi sites are apparent". However, in addition to strand orientation (top or bottom strand), Chi sequences selectively appear in leftward-transcribed ORFs in the bottom strand and in rightward-transcribed ORFs in the top strand (Burland et al., 1993; Bell et al., 1998). This provides a novel deterministic feature of Chi sites, which is the basis of the present analysis. While we agree that in bacteria "recombination islands are equivalent to ORFs" (Colbert et al., 1998), Chargaff difference analysis using 1 kb window sizes provides evidence, at least for E. coli, that Chi windows may correspond to a distinct subset of ORFs (Table 2). To this extent, the "island" proposal is supported. We also note that four Chi-like sequences (GTTGGTGG) in the H. influenzae segments which we examined, occur in 16S RNA genes, where there is no question of an involvement of codons. Three of these four Chi-like sequences occur in the top strand where, as expected, transcription is to the right, and the other occurs in the bottom strand where, as expected, transcription is to the left.

4.4. Chi sequences in other bacteria

      In this initial study we have limited ourselves to the canonical Chi sequence, except in the case of H. influenzae Chi-like sequences, which are likely to be functional (Sourice et al., 1998). sequences with Chi-like function, but with little similarity to the canonical Chi sequence, have been identified in some bacteria (Smith, 1997; Chedin et al., 1998). In these cases, perhaps because their ecological niches overlap with that of E. coli, there might be a chance of recombination with E. coli DNA (not-self), so there would have been a selection pressure militating against a similarity of sequences with Chi function (so assisting the preservation of species integrity; Forsdyke 1996). Thus, it is pertinent to ask whether canonical Chi sequences are more frequent in bacteria other than E. coli, particularly if they are likely to occupy different niches. If so, one can examine whether these Chi sequences have deterministic features of Chi other than sequence similarity. Indeed, in many such cases, Chi sequences do locate preferentially to mRNA synonymous strands, as in E. coli (Table 1).

     The fact that Chi abundance in E. coli appears as a distinct outlier when compared with that of other bacteria (Fig. 1), suggests that the canonical Chi is particular to E. coli. However, when expressed as the observed/expected ratio, Chi frequency in E. coli is of the same order as in most other organisms, and less than that of the AT-rich thermophile M. jannaschii. Some bacteria (A. aeolicus, B. subtilis, T. pallidum) had no more canonical Chi sequences than expected from base composition, and in at least one of these (B. subtilis) the Chi function could be mediated by an independent sequence (Chedin et al. 1998).

     Since recombination repair may be important for correction of DNA damage, and Chi promotes recombination, it was considered possible that the Chi function might be generally increased in thermophilic and radiation-resistant bacteria. These generally occupy ecological niches different from E .coli, so that recombination with the E. coli would be unlikely. Thus, they might have even more canonical Chi sequences than E. coli. However, when base composition was taken into account, D. radiodurans did not have more canonical Chi sequences than E .coli, although many preferentially located to the mRNA synonymous stands (Fig. 1; Table 1). Furthermore, only two of the five thermophilic bacteria examined showed preferential location of Chi sequences to mRNA-synonymous strands. Thus, in thermophilic bacteria and D. radiodurans, the Chi function may be achieved by sequences which differ from the canonical Chi sequence, or the number of these sequences is sufficient, or the Chi function is unnecessary.