Programs of the Genetics Computer Group Inc. (Gribskov and Devereux 1991) were made available online through the services of the Molecular Biology Data Service of the National Research Council, Ottawa, which include access to a Silicon Graphics Challenge XL computer. Sequences were randomized using the program SHUFFLE. The outputs from the latter could be used directly by the program FOLD (Zuker 1989), which finds a secondary structure of minimum free energy using the energy values for base stacking and loop destabilization assigned by Turner et al. (1988). This program was designed for the study of secondary structure in non-supercoiled, single-stranded RNA molecules. It is assumed here that it is applicable to DNA. This assumption is not entirely valid. For example, the 2' hydroxyl group in RNA can contribute of the order of I kcal/mol to helix stability (Turner and Bevilacqua 1993). Since the present study is mainly concerned with differences in fold energies rather than absolute values, differences between RNA and DNA are considered of minor importance. A unix script program, SHUFFOLD, was written to determine the minimum free energy of folding of natural and shuffled versions of successive overlapping 200nt windows from nucleic acid sequences. Another program, STATS, was written to subject the output from SHUFFOLD to statistical analysis in the Minitab system (Ryan and Joiner 1994).

    For each 200-nt window, FOLD first determines the minimum free energy value for folding of the natural sequence (FONS value). This is a function of both base composition and base order and measures the "total stem-loop potential" of a region. Then 10 random sequences are generated from the same window, and each randomized sequence is submitted to FOLD. The mean minimum free energy value for the 10 sequences (FORS-M value) provides a measure of the contribution of base composition alone to the stem-loop potential ("base composition-determined stem-loop potential"). Since the FONS value is usually more negative than the FORS-M value, the difference between the two values (FORS-M less FONS) is usually positive.

     This difference (the "FORS-D" value) provides a measure of the contribution of base order alone to the stem-loop potential. Thus, a positive FORS-D value defines and quantitates the "base order-determined stem-loop potential." This closely corresponds to the "segment score" of Le and Maizel (1989), which is used to assess "statistically significant" stem-loop potential. A negative FORS-D value in a region may mean that base order has been adapted to serve some other potential.

FORS-D Values of Intron-Containing Genes

      The human cytokine-encoding gene G0S19-1 is one of a series of potential lymphocyte G0/G1 switch regulatory genes ("G0S genes") which have been sequenced in my laboratory (Blum et al. 1990; [Now more widely known as MIP1-alpha or CCL3, whose receptor is a coreceptor for HIV-1]). The gene has three exons distributed over a region of 1,886 nt, and the cDNA derived by splicing is 778 nt in length [See diagram at top of page]. As noted above (fig. 1), FONS and FORS-M profiles tend to resemble each other (fig. 3A, lower panel). FORS-D values are generally positive in the gene; however, distinct local decreases occur in certain regions (fig. 3A, upper panel). When the introns are removed to generate the cDNA, decreased FORS-D values tend to be concentrated together, and general positivity is less evident (fig. 3B, upper panel). An extremely low FORS-D value in the immediate 5'flank of the gene corresponds to a region rich in potential regulatory motifs, which is highly conserved between G0S19-1 and its murine homolog (Russell and Forsdyke 1993).

FORS-D Values Correlate Negatively with Exon Overlap

   To obtain an objective measure of the tendency of small exons to associate with low FORS-D values, each of the 200-nt windows for which FORS-D values were determined was scored for its percentage of exon overlap. Figure 5A shows a linear regression plot for the three exons of the G0S19-1 gene. Consecutive windows extended from the first 5'window overlapping the first exon to the last 3'window overlapping the last exon. Although there was much scatter of data points, the slope of the least-squares regression line was significantly greater than zero (P = 0.048 that the slope is not greater than zero). Figure 5B and C show similar data for the oncogenes cfos and p53, except that the long last exons were omitted. Both slopes were significantly greater than zero (P = 0.018, c-fos, P = 0.002, p53). The corresponding P values when the last exons were included were 0.752 and 0.048. (A justification for omitting last exons is given later.)

Discussion

The Competing Needs of the Early Replicators

    The winner in the competition between replicator molecules in early evolution would have been the replicator which could most efficiently balance the needs for:

• (1) rapid replication,

• (2) accurate replication, and

• (3) stability.

    Intimately linked with the latter two needs would be the need to carry out repair of malreplicated or damaged molecules. In an early "RNA world," the ability to shuffle damaged segments between molecules so as to create segment combinations which could continue rapid and accurate replication could have been advantageous. Thus, it can be envisaged that a fourth need,

• (4) the need to recombine,

would have been another property which successful replicators would have optimized at an early stage.

     If Tomizawa's stem-loop "kissing" model is applicable to the early RNA world, then it can be further envisaged that replicators which modified their sequences to increase the probability of stem-loop formation (and hence of recombination), would have had a survival advantage. To reap the benefits of efficient recombination a replicator would have had to exchange segments with its own kind of replicator, not with other kinds. Thus need 4 implies a fifth need,

• (5) the need to distinguish "self" from "not-self."

Finally, when the competition to build better "survival machines" began, a sixth need arose. This was the need to dedicate parts of the genome to the

• (6) encoding of specific functions.

This discussion deals with potential conflicts between needs 4 and 6, and needs 4 and 5, and suggests how these conflicts might have been resolved.

Evidence for a Stem-Loop Model

    It can be imagined that as part of the attempt to improve replicator stability (need 3), RNA molecules would have been superseded by DNA molecules, but these would have retained the potential to form stem loops. This leads to the prediction explored here that, where not constrained by other needs, modern DNA molecules would have a special potential to form stemloop structures. This would be particularly evident during meiosis (Kleckner and Weiner 1993) and might require an exceptional degree of supercoiling for DNA to depart from its classical duplex structure (Murchie et al. 1992). Consistent with this, inhibitors of topoisomerases, which relax supercoiled DNA, greatly enhance recombination (Wang et al. 1990). Furthermore, endonucleases involved in recombination, which make single-strand nicks in double-stranded DNA, require a supercoiled substrate (Sung et al. 1993). Although a relatively crude measure, there is evidence from studies with anticruciform antibodies for the presence of stem-loop structures, which can vary with the physiological state of the cell (Ward et al. 1990). Further evidence for a role of stem loops in recombination is presented elsewhere (Romanova et al. 1986; Reed et al. 1994).

    Positive FORS-D values indicate the existence of an evolutionary pressure on DNA base order, rather than on base composition, to generate the potential to form stem-loop structures. Thus, in the absence of other constraints, base changes which enhance the formation of stem loops would have been accepted. I have reported here that dispersed positive FORS-D values are found in long human genome segments considered to consist mainly of nontranscribed intergenic DNA (fig. 1). This is found with DNA from various species, including some with much less intergenic DNA than humans (Forsdyke 1995b). The most obvious genome-wide selective force driving evolution of the potential to form stem loops is recombination.

Conflict Hypothesis and the Evolution of Introns

   A potential conflict between the genome-wide FORS-D pressure and pressures on DNA to encode specific functions is apparent (fig. 2). A sequence required to encode a protein might not at the same time be able locally to optimize its folding propensity. The conflict can be met in three ways.

• First, because of the redundancy of the genetic code, particular synonymous codons can be preferred.

• Second, amino acids with similar functions (e.g., serine and threonine) can be interchanged to widen the range of codon choices.

• Third, the sequences encoding a protein can be diffused over a wide region by permitting encoding to occur only in discrete segments.

If the first two options are not sufficient, then only the third option is left. Thus introns might correspond to parts of a gene where the constraints on the first two options are most severe. The more severe the constraints, the more introns there would be, and the longer would be the length of DNA occupied by the gene.

     This conflict hypothesis might explain how introns initially arose. However, it does not appear to address the problem of how introns came to vary so dramatically in length (Hawkins 1988). Variation in intron length would tend to prevent recombination between genes with homologous exon sequences by impairing the precise register required for successful "kissing" between the loops of stem-loop structures.

    In terms of FORS-D values, the hypothesis states that when exons first arose, the exon sequences themselves would have had low FORS-D values, except where, by chance, the need to encode a peptide happened not to be in conflict with the need to form stem loops. The low FORS-D exons would have been surrounded by high FORS-D introns and flanking sequences. Over evolutionary time the demands of regulation of gene expression would have been imposed upon this primitive arrangement, so that low FORS-D values might be present in regions of DNA where regulatory proteins bind (except, perhaps, where the regulatory proteins recognize DNA palindromes). Also, over evolutionary time a background "noise" might have been imposed by mobile genetic elements and recombinations accompanying exon shuffling (Gilbert and Glynias 1993).

    Taking into account the proposed long evolutionary period since introns were established in genes, the analyses of FORS-D values in various intron-containing genes (figs. 3-4; table 1) do provide some support for the idea of a conflict between genome-wide and local pressures. Further evidence has been provided by recent studies of snake venom phospholipase A2 genes, retroviral genomes, and major histocompatibility complex peptide-binding genes (Forsdyke 1995a, and unpublished data [see below]). All of these are under greater selection pressure (positive selection) than the genes studied here (with the possible exception of genes encoding troponin C; Ohta 1994).

     Some terminal long open-reading frames were excluded from the linear regression analyses shown in figure 5 and table 1. In terms of the hypothesis presented here, long open-reading frames would only exist in certain genes because those genes had been able to resolve the conflict in the regions where we now see the long open-reading frames; FORS-D pressure would have been accommodated by the choice of appropriate synonymous codons or of amino acids with similar functions. (These would have been the main choices open to organisms without introns.) Thus, FORS-D analyses of genes with long open-reading frames are not inconsistent with the conflict hypothesis (fig. 4). By omitting unduly long exons from regression analyses, one hopes to better assess genomic regions which are less able to adapt to FORS-D pressure by using synonymous codons or functionally similar amino acids.

     The conflict hypothesis would predict that cDNA sequences would tend to have lower average FORS-D values than the genes from which they derive. However, it does not follow that low FORS-D exons, when united, will always result in a low FORS-D cDNA. Two independently arising exons may each have a low potential to form stem loops but a high potential to form stem loops collectively. As a simple example, consider two exons of sequence CACACACA and TGTGTGTG. Both have low FORS-D values, but they would unite to generate a cDNA with a very stable stem loop (high FORS-D value). Only some genes show clear evidence for higher average FORS-D values than the corresponding cDNAs (G0SI 9-1, fig. 3; table 2).

Why Are FORS-D Values So Sensitive?

     Evolutionary pressures (mutation, selection, drift) determine how many bases there are in a sequence (DNA length), the proportions of different bases (base composition), and the order of the bases (sequence). The experimental shuffling of the order of bases in a sequence of given length and base composition disrupts information present in the natural sequence and generates a reference sequence of the same length and base composition against which the natural sequence can be compared. Thus one can distinguish evolutionary pressures affecting base order (primary sequence) from those affecting base composition.

    FORS-M and FONS profiles closely follow each other (figs. I B, 3B-4B), showing that base composition is the major determinant of the stability of stem loops. It thus might seem somewhat surprising that differences of only a few kilocalories (FORS-D values) provide such a sensitive indicator. That fluctuations in FORS-D values are not random is shown by their consistent average positivity (fig. 1) and the association of negative FORS-D values with functionally important regions of DNA (fig. 3).

    Base composition is a genome "strategy" (Grantham et al. 1980) and thus is less likely to be relevant to local gene function than primary sequence. FORS-D values provide a sensitive indicator because they measure the base order-determined stem-loop potential, not the base composition-determined stem-loop potential. A close examination of folding energy profiles reveals that high FORS-D values may sometimes be associated with quite low negative FONS values (a function of base order and base composition) and even lower negative FORSM values (a function of base composition). Thus, high negative FONS values (implying a very stable secondary structure) may sometimes have little local functional relevance.

     The recombination model of Wagner and Radman (1975) proposes that, after initial contact is made by loops at the tips of the stems, there are endonucleolytic cleavages of single strands of opposing stems. Single strand exchanges between the stems then follows. This process would presumably abort if precise complementarity were not present in the stems (Radman and Wagner 1993). Furthermore, Tomizawa (1993) has concluded that the major role of the stem in a stem-loop structure is the proper positioning of the loop. This allows the unpaired bases in the loop to pair, in register, with those of an appropriately positioned loop projecting from a complementary nucleic acid. This "kissing" is rate limiting in recombination. Thus, a departure, even by a few kilocalories, in complementarity between bases in the stem might prevent recombination. An evolutionary force adapting base order to favor stem-loop formation could be a very powerful one.

 Introns Early or Late?

    In that the ability to recombine is held to have evolved before the ability to encode proteins, this paper supports the "introns early" model. The positions of exon-intron junctions are held to have been determined by the need to form stem loops and are not necessarily related to protein domains (Gilbert and Glynias 1993). Stoltzfus et al. (1994, p. 202) concluded that "no significant correspondence between exons and units of protein structure was detected."

GC/AT Fine-Tuning and Speciation

    Different genomes, or genome compartments, have characteristic (G+C)/(A+T) ratios (Wyatt 1952; Bernardi 1989; Filipski 1990). In vertebrates this "GC/AT pressure" (fig. 2) is evident in intergenic regions, and introns, and in both coding and noncoding parts of exons. The "genome hypothesis" (Grantham et al. 1980) has led to the proposal that there is some intrinsic genome-wide pressure favoring the adoption of a particular (G+C)/(A+T) ratio in a particular species. The view that this primarily reflects mutational biases (Filipski 1990) has been challenged (Bernardi 1989). I argue elsewhere (unpublished data), that species-specific settings of the (G+C)/(A+T) ratio have arisen as part of a fine-tuning process which prevents recombination between species. Just as different radio transmitters broadcast their messages at different wavelengths to avoid interference, so different genomes "broadcast" their sequences at different (G+C)/(A+T) "wavelengths," to avoid undesirable recombination events.

     The present work suggests a mechanism by which this could occur. Base composition, rather than base order, is the major factor determining the folding energy of a nucleic acid segment (figs. 1B, 3B-4B). Thus small changes in base composition could greatly affect the looping pattern which a sequence could present for homology search. It would be more difficult to extrude loops from GC-rich DNA than from AT-rich DNA.

     If avoidance of recombination between different species (including a host species and its pathogen species) is evolutionarily advantageous (Bernstein and Bernstein 1991), then there should have been a selection pressure such that each species in a common environment with other species would have fine-tuned its (C+G)/(A+T) ratio to a level distinct from those of the other species. Thus, biologically similar species may have dissimilar ratios, whereas biologically dissimilar species (unlikely to interact sexually and with dissimilar pathogens) may have similar ratios (Wyatt 1952). The fine-tuning of base ratios could have been a key component of the postzygotic isolation process leading to speciation (unpublished data; [see Evolution Web Page]).

Acknowledgments

I thank D. Back for assistance in computer configuration and M. Go, C. Tittiger, V. K. Walker, and G. R. Wyatt for helpful comments on the manuscript. The work was supported by a grant from the Medical Research Council of Canada.

LITERATURE CITED

BERNARDI, G. 1989. The isochore organization of the human genome. Annu. Rev. Genet. 23:637-661.

BERNSTEIN, C., and H. BERNSTEIN. 199 1. Aging, sex and DNA repair. Academic Press, San Diego.

BLUM, S., R. E. FORSDYKE, and D. R. FORSDYKE. 1990. Three human homologs of a murine gene encoding an inhibitor of stem cell proliferation. DNA Cell. Biol. 9:589-602.

CRICK, F. 197 1. General model for the chromosomes of higher organisms. Nature 234:25-27.

FILIPSKI, J. 1990. Evolution of DNA sequence: contributions of mutational bias and selection to the origin of chromosomal compartments. Adv. Mutagenesis Res. 2:1-54.

FORSDYKE, D. R. 1995a. Conservation of stem-loop potential in introns of snake venom phospholipase A2 genes: an application of FORS-D analysis. Mol. Biol. Evol. 12:1157-1165.

FORSDYKE, D. R. 1995b. Relative roles of primary sequence and (G+C)% in determining the hierarchy of frequencies of complementary trinucleotide pairs in DNAs of different species. J. Mol. Evol. 41:573-581.

GILBERT, W., and M. GLYNIAS. 1993. On the ancient nature of introns. Gene 135:137-144.

GRANTHAM, R., G. GAUTIER, M. GOUY, R. MERCIER, and A. PARE. 1980. Codon catalog usage and the genome hypothesis. Nucleic Acids Res. 8: r49-r62.

GRIBSKOV, M., and J. DEVEREUX. 1991. Sequence analysis primer. Stockton, New York.

HAWKINS, J. D. 1988. A survey of intron and exon lengths.Nucleic Acids Res. 16:9893-9905.

HAWLEY, R. S., and T. ARBEL. 1993. Yeast genetics and the fall of the classical view of meiosis. Cell 72:301-303.

KLECKNER, N., and B. M. WEINER. 1993. Potential advantages of unstable interactions for pairing of chromosomes in meiotic, somatic and premeiotic cells. Cold Spring Harbor Symp. Quant. Biol. 58:553-565.

LE, S.-Y., and J. V. MAIZEL. 1989. A method for assessing the statistical significance of RNA folding. J. Theor. Biol. 138: 495-510.

MURCHIE, A. 1. H., R. B0WATER, F. ABOUL-ELA, and D. M. J. LILLEY. 1992. Helix opening transitions in supercoiled DNA. Biochem. Biophys. Acta 1131:1-15.

OHTA, T. 1994. Further examples of gene duplication revealed through DNA sequence comparisons. Genetics 138:1331-1337.

ORITA, M., H. IWAHANA, H. KANAZAWA, K. HAYASHI, and T. SEKIYA. 1989. Detection of polymorphisms of human DNA by gel electrophoresis as single strand conformation polymorphisms. Proc. Natl. Acad. Sci. USA 86:2766-2770.

PRABHU, V. V. 1993. Symmetry observations in long nucleotide sequences. Nucleic Acids Res. 21:2797-2800.

RADMAN, M., and R. WAGNER. 1993. Mismatch recognition in chromosomal interactions and specification. Chromosoma 102:369-373.

REED, K. M., L. W. BEUKEBOOM, D. G. EICKBUSH, and J. H. WERREN. 1994. Junction between repetitive DNAs on the PSR chromosome of Nasonia vitripennis.- association of palindromes with recombination. J. Mol. Evol. 38:352-362.

ROMANOVA, L. 1., V. M. BLINOV, E. A. TOLSKAYA, E. G. VIKTOROVA, M. S. KOLESNIKOVA, E. A. GUSEVA, and V. 1. AGOL. 1986. The primary structure of crossover regions of intertypic poliovirus recombinants: a model of recombination between RNA genomes. Virology 155:202-213.

RUSSELL, L., and D. R. FORSDYKE. 1993. The third human homolog of a murine gene encoding an inhibitor of stem cell proliferation is truncated and linked to a CpG island-containing upstream sequence. DNA Cell. Biol. 12:157-175.

RYAN, B. F., and B. L. JOINER. 1994. Minitab handbook. 3d ed. Wadsworth, Belmont, Calif

SOBELL, H. M. 1972. Molecular mechanism for genetic recombination. Proc. Natl. Acad. Sci. USA 69:2483-2487.

STOLTZFUS, A., D. F. SPENCER, M. ZUKER, J. M. LOGSDON, and W. F. DOOLITTLE. 1994. Testing the exon theory of genes: the evidence from protein structure. Science 265: 202-207.

SUNG, P., P. REYNOLDS, L. PRAKASH, and S. PRAKASH. 1993. Purification and characterization of the Saccharomyces cerevisiae RADI/RADIO endonuclease. J. Biol. Chem. 268: 26391-26399.

TOMIZAWA, J. 1984. Control of ColE I plasmid replication: the process of binding of RNA I to the primer transcript. Cell 38:861-870.

TOMIZAWA, J. 1993. Evolution of functional structures of RNA. Pp. 419-445 in R. F. GESTELAND and J. F. ATKINS, eds. The RNA world. Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y.

TURNER, D. H., and P. C. BEVILACQUA. 1993. Thermodynamic considerations for evolution by RNA. Pp. 447-464 in R. F. GESTELAND and J. F. ATKINS. eds. The RNA world. Cold Spring Harbor, Cold Spring Harbor, N.Y.

TURNER, D. H., N. SUGIMOTO, and S. M. FREIER. 1988. RNA structure prediction. Annu. Rev. Biophys. Chem. 17:167192.

WAGNER, R. E., and M. RADMAN. 1975. A mechanism for initiation of genetic recombination. Proc. Natl. Acad. Sci. USA 72:3619-3622.

WANG, J. C., P. R. CARON, and R. A. Kim. 1990. The role of DNA topoisomerase in recombination and genome stability: a double-edged sword? Cell 62:403-406.

WARD, G. K., R. McKENZIE, M. ZANNIS-HADJOPOULOS, and G. B. PRICE. 1990. The dynamic distribution and quantification of DNA cruciforms in eukaryotic nuclei. Exp. Cell Res. 188:235-246.

WYATT, G. R. 1952. The nucleic acids of some insect viruses. J. Gen. Physiol. 36:201-205.

ZUKER, M. 1989. Computer prediction of RNA secondary structure. Meth. Enzym. 180: 262-289.

Shen, L. X., Basilion, J. P. & Stanton, V. P. (1999) Single nucleotide polymorphisms can cause different structural folds of mRNA. Proc. Natl. Acad. Sci. USA 96, 7871-6.

Dong, F., Allawi, H. T., Anderson, T., Neri, B. P. & Lyamichev, V. I. (2001) Secondary structure prediction and structure-specific sequence analysis of single-stranded DNA. Nucleic Acids Res. 29, 3248-57.

Abbreviations:

• FONS = folding of natural sequence
  
• FORS-M = folding of randomized sequence mean
  
• FORS-D = folding of randomized sequence difference
  
• indel = insertion and/or deletion
  
• PLA₂ = phospholipase A₂.

      FORS-D analysis ("folding of randomized sequence difference analysis") implies that the sequences of introns of venom phospholipase A₂ genes have been under pressure to accept mutations which conserve the potential to form single-strand stem-loops, whereas exon sequences have evaded this pressure. These results are consistent with the hypothesis that stem-loop-based recombination evolved in the early "RNA world", and the potential to generate stem-loops became widely dispersed throughout genomes. The subsequent conflict between stem-loop potential and protein-encoding potential was resolved by:

• (i) use of synonymous codons,
  
  
• (ii) acceptance of conservative amino acid changes to widen the range of codon choice, and
  
  
• (iii) encoding proteins in dispersed segments interrupted by regions of high stem-loop potential (introns).

In genes under strong Darwinian selection, the pressure on genes to diversify in order to prevent unwanted recombination events is largely accommodated by exons. This leaves introns, which normally predominate in this role, free to accommodate to pressure for stem-loop formation.

Methods-

FORS-D analysis. This evaluates base order-determined stem-loop potential, which closely corresponds to the "statistically significant" stem-loop potential of Le and Maizel (1989). The theory and application of the method were described previously (Forsdyke, 1995a).

Determination of substitution density. Sequences of different PLA₂ genes were aligned with the program GENALIGN using the "regions" method, a matching weight of 1.0, and a deletion weight (gap penalty) of -0.5 (Martinez, 1988). The program was accessed through the Bionet on-line computing service (IntelliGenetics Inc., Mountain View, California). Base substitutions and "indels" (insertions or deletions) were counted in successive 200 nt windows, each of which overlapped the preceding window by 150 nt. Substitutions dominated the alignments and changes in deletion weight over a range from -0.25 to -1.0 caused few interchanges between substitutions and indels. Thus, the general patterns of substitution and indel densities were independent of deletion weight. No corrections were made for the possibility of multiple substitutions.

Results-

High substitution density in exons. Many species have multiple PLA₂ genes, which usually consist of four exons (Davidson and Dennis, 1990). In snakes a subset of these genes encodes proteins which contribute to the toxicity of venom. Intraspecies comparisons of base substitutions in snake venom phospholipase genes show high substitution densities in exons, and low substitution densities in introns. This applies to the habu snake (Nakashima et al, 1993), and to the rattlesnake (John et al. 1994).

Figure 1a demonstrates this, both within species, and between species. Substitutions are scored for successive 200 nt windows which overlap by 150 nt. Relative to the rattlesnake PLA₂ acidic subunit gene (GenBank designation U01026), the rattlesnake PLA₂ basic subunit gene (GenBank designation U01027), and two habu snake PLA₂ genes (GenBank designations TFLPA2P1 and TFLPA2P2), show many substitutions in all but the first exon, most of which does not code for protein. In the fourth exon, substitution density is greatest in the protein-encoding region. Because some of the 200 nt windows, centred in introns, overlap exons, sharp demarkations between the low substitution density introns and high substitution density exons, are not apparent. However, the relatively low intron substitution density is not uniform in the second intron; here the density increases in the 3' half of the intron. Essentially similar density patterns are obtained when different venom PLA₂ genes are taken as a basis for comparison (e.g. TFLPA2P2 compared with U01026, U01027 and TFLPA2P1; data not shown). Thus, although the particular bases substituted tend to be characteristic of a particular pair of genes, the overall substitution patterns are unlikely to be random.

     Figure 1b shows similar data for base insertions and deletions ("indels"). Indels are in both coding and non-coding regions. Major peaks occur at the 3' end of the second intron in the cases of the U01026/U1027 and U1026/TFLPA2P2 comparisons. In the former case the peak is associated with a 38 nt CG-rich sequence in U01026, which is also present in the habu snake genes, but not in U01027. In the latter case the peak is associated with a microsatellite repeat (CA)_n in TFLPA2P2, which is not so evident in the other snake venom PLA₂ genes. The indel density patterns shown in figure 1b are essentially similar to those obtained when different venom PLA₂ genes are taken as a basis for comparison (data not shown). Thus, although the actual bases inserted or deleted tend to be characteristic of a particular pair of genes, the overall patterns are unlikely to be random.

Reciprocal relationship between substitution density and FORS-D. The 200 nt windows used to characterize substitution and indel densities (fig. 1), were subjected to the program FOLD (Zuker, 1989) to determine the energetically most favourable folded structures.

In figures 2b (rattlesnake gene U01026) and 3b (rattlesnake gene U01027), the minimum free energies associated with each structure (FONS values) are shown together with the mean values for ten randomized versions of the same sequence windows (FORS-M values). In keeping with previous observations on a variety of genes (Forsdyke, 1995a), FONS values tend to be more negative than FORS-M values (indicating greater stability of the secondary structure of the natural sequence). This is particularly apparent in the 5' non-coding part of exon 1, in the first part of intron 2, and in intron 3. The differences (FORS-M less FONS) are expressed in figures 2a and 3a as the FORS-D ("folding of randomized sequence difference") values. Positive FORS-D values reflect a contribution of base order (primary sequence), rather than of base composition, to the stability of the optimum structure. The previously noted tendency for protein-encoding parts of exons to have zero or negative FORS-D values, and for introns to have positive FORS-D values (Forsdyke, 1995a), is generally apparent. In the case of U01026 the first intron has low values, but this is not seen in the case of U01027. In the case of U01027 the coding part of the fourth exon corresponds to a relative decline in FORS-D value, but remains positive. It should be noted that high FORS-D values, reflecting local adaptation of base order for the generation of secondary structure, are not necessarily correlated with high FONS values (e.g. high FORS-D values occur in the 3' half of the 2nd intron and in the 3' non-coding part of the fourth exon).

    Figures 2c,d and 3c,d show FONS, FORS-M and FORS-D values for U01026 and U01027 (rattlesnake) cDNAs. Fine vertical lines indicate sites of intron removal. FORS-D values are low in the regions corresponding to the middle two exons, but increase in the 5' and 3' exons, which contain extensive non-coding regions.

   Figures 2a and 3a also show substitution and indel densities taken from figure 1. There is an approximately reciprocal relationship between FORS-D values and substitution density. Where FORS-D values are low (coding parts of exons), substitution densities are high. Where FORS-D values are high (introns; 5' and 3' non-coding regions), substitution densities are low. However, the 3' end of the 2nd intron, where indel density is highest, is characterized by relatively high values both for FORS-D and substitution density. Similar results were obtained when the other venom PLA₂ genes under study were compared in this manner.

    More objective evidence for a relationship between secondary structure and substitution density was provided by plotting individual window values for FONS, FORS-M and FORS-D, against the corresponding substitution densities. For each pair of genes three linear regression plots were obtained.

LITERATURE CITED

BERNSTEIN, C., and H. BERNSTEIN. 1991. Aging, Sex and DNA Repair. Academic Press, San Diego.

BLUM, S., R. E. FORSDYKE, and D. R. FORSDYKE. 1990. Three human homologs of a murine gene encoding an inhibitor of stem cell proliferation. DNA Cell Biol. 9:589-602.

BRANNAN, C. I., E. C. DEES, D. S. INGRAM, and S. M. TILGHMAN. 1990. The product of the H19 gene may function as an RNA. Mol. Cell. Biol. 10:28-36.

BROCKDORFF, N., A. ASHWORTH, G. F. KAY, V. M. McCABE, D. P. NORRIS, P. J. COOPER, S. SWIFT, and S. RASTAN. 1992. The product of the mouse Xist gene is a 15 kb inactive X-specific transcript containing no conserved ORF and located in the nucleus. Cell 71:515-526.

DAVIDSON, F. F., and E. A. DENNIS. 1990. Evolutionary relationships and implications for the regulation of phospholipase A₂ from snake venom to human secreted forms. J. Mol. Evol. 31:228-238.

DURET, L., F. DORKELD, and C. GAUTIER. 1993. Strong conservation of non- coding sequences during vertebrate evolution: potential involvement in post- transcriptional regulation of gene expression. Nucleic Acids Res. 21:2315-2322.

FORSDYKE, D. R. 1991. Early evolution of MHC polymorphism. J. Theor. Biol. 150:451-456.

FORSDYKE, D. R. 1994. Relationship of X chromosome dosage compensation to intracellular self/not-self discrimination: a resolution of Muller's paradox? J. Theor. Biol. 167:7-12.

FORSDYKE, D. R. 1995a. A stem-loop "kissing" model for the initiation of recombination and the origin of introns. Mol. Biol. Evol. 12:949-958.

FORSDYKE, D. R. 1995b. Relative roles of primary sequence and (G + C)% in determining the hierarchy of frequencies of complementary trinucleotide pairs in DNAs of different species. J. Mol. Evol. 41:573-581.

FORSDYKE, D. R. 1995c. Different biological species "broadcast" their DNAs at different (G+C)% "wavelengths". (submitted) -- J. Theor. Biol. (1996a) 178:405-417.

FORSDYKE, D. R. 1995d. Reciprocal relationship between stem-loop potential and substitution density in retroviral quasispecies under positive Darwinian selection.(submitted) -- J. Mol. Evol.(1995) 41:1022-1037.

HARDING, K. A., and K. R. G. WELCH. 1980. Venomous Snakes of the World. Pergamon Press, Oxford.

HAWKINS, J. D. 1988. A survey of intron and exon lengths. Nucleic Acids Res. 16:9893-9905.

HILL, R. E., and N. D. HASTIE. 1987. Accelerated evolution in the reactive centre regions of serine protease inhibitors. Nature 326:96-99.

HUGHES, A. L., and M. NEI. 1989. Nucleotide substitution at major histocompatability complex class II loci: evidence for overdominant selection. Proc. Natl. Acad. Sci. USA. 86:958-962.

JOHN, T. R., L. A. SMITH, and I. I. KAISER. 1994. Genomic sequences encoding the acidic and basic subunits of Mojave toxin: unusually high sequence identity of non-coding regions. Gene 139:229-234.

JOYCE, G. F., and L. E. ORGEL. 1993. Prospects for understanding the origin of the RNA world. Pp 1-25 in R. F. GESTELAND, and J. F. ATKINS, eds. The RNA World. Cold Spring Harbour Laboratory Press, New York.

KLECKNER, N., R. PADMORE, and D. K. BISHOP. 1991. Meiotic chromosome metabolism: one view. Cold Spring Harbour Symp. Quant. Biol. 56:729-743.

KLECKNER, N. and B. M. WEINER. 1993. Potential advantages of unstable interactions for pairing of chromosomes in meiotic, somatic and premeiotic cells. Cold Spring Harbour Symp. Quant. Biol. 58:553-565.

LAMBEAU. G., P. ANCIAN, J. BARHANIN, and M. LAZDUNSKI. 1994. Cloning and expression of a membrane receptor for secretory phospholipase A₂. J. Biol. Chem. 269:1575-1578.

LE, S-Y., J-H. CHEN, M. BRAUN, M. GONDA, and J. V. MAIZEL.1988. Stability of RNA stem-loop structure and distribution of non-random structure in the human immunodeficiency virus (HIV-1). Nucleic Acids Res. 16:5153-5168.

LE, S-Y., J-H. CHEN, D. CHATTERJEE, and J. V. MAIZEL. 1989. Sequence divergence and open regions of RNA secondary structures in the envelope regions of the 17 human immunodeficiency virus isolates. Nucleic Acids Res. 17:3275-3288.

LE, S-Y. and J. V. MAIZEL. 1989. A method for assessing the statistical significance of RNA folding. J. Theor. Biol. 138:495-510.

MARTINEZ, H. M. 1988. A flexible multiple sequence alignment program. Nucleic Acids Res. 16:1683-1691.

NAKASHIMA, K-I., T. OGAWA, N. ODA, M. HATTORI, Y. SAKAKI, H. KIHARA, and M. OHNO. 1993. Accelerated evolution of Trimeresurus flavoviridis venom gland phospholipase A₂ isozymes. Proc. Natl. Acad. Sci. USA 90:5964-5968.

OGAWA, T., N. ODA, K-I NAKASHIMA, H. SASAKI, M. HATTORI, Y. SAKAKI, H. KIHARA, and M. OHNO. 1992. Unusually high conservation of untranslated sequences in cDNAs for Trimeresurus flavoviridis phospholipase A₂ isoenzymes. Proc. Natl. Acad. Sci. USA 89:8557-8561.

OHTA, T. 1994. Further examples of evolution by gene duplication revealed through DNA sequence comparisons. Genetics 138:1331-1337.

RADMAN, M., and R. WAGNER. 1993. Mismatch recognition in chromosomal interactions and speciation. Chromosoma 102:369-373.

RAYSSIGUIER, C., D. THALER, and M. RADMAN. 1989. The barrier to recombination between Echerichia coli and Salmonella typhimurium is disrupted in mismatch-repair mutants. Nature 342:396-401.

ROMANOVA, L. I., V. M. BLINOV, E. A. TOLSKAYA, E. G. VIKTOROVA, M. S. KOLESNIKOVA, E. A. GUSEVA, and V. I. AGOL. 1986. The primary structure of crossover regions of intertypic poliovirus recombinants: a model of recombination between RNA genomes. Virology 155:202-213.

SCPAER, E. G., and J. I. MULLINS. 1993. Rates of amino acid change in the envelope protein correlates with pathogenicity of primate lentiviruses. J Mol Evol 37:57-65

STOLTZFUS, A., D. F. SPENCER, M. ZUKER, J. M. LOGSDON, and W. F. DOOLOTTLE. 1994. Testing the exon theory of genes: the evidence from protein structure. Science 265:202-207.

TAKEDA, A., R. G. E. PALFREE, and D. R. FORSDYKE. 1982. Role of serum in inhibition of cultured lymphocytes by lysophosphatidylcholine. Biochem. Biophys. Acta 710:87-98.

TOMIZAWA, J. 1993. Evolution of functional structures of RNA. Pp. 419-445 in R. F. GESTELAND, and J. F. ATKINS, eds. The RNA World.Cold Spring Harbour Laboratory Press, New York.

TURNER, D. H., N. SUGIMOTO, and S. M. FREIER. 1988. RNA structure prediction. Annu. Rev. Biophys. Chem. 17:167-192.

ZUKER, M. 1989. Computer prediction of RNA secondary structure. Meth. Enzym. 180:262-289.

ZUKER, M., and P. STEIGLER. 1981. Optimal computer folding of large RNA sequences using thermodynamics and auxiliary information. Nucleic Acids Res. 9:133-148.