Many eukaryotic genes are split into exons and introns, the latter being removed post-transcriptionally so that only exon sequences appear in cytoplasmic RNAs. Since introns appear in both protein-encoding RNAs and non-protein-coding RNAs, they interrupt genetic information per se, not just protein-encoding information. 

    A DNA sequence has the potential to carry more than one type of genetic information, but different types may conflict. Thus, it has been proposed that introns arose because sequences were unable to contain concomitantly complete information for the encoding both of stem-loops and of cytoplasmic products (protein and/or RNA). Stem-loop potential is held to be selectively advantageous since it promotes the recombination-dependent correction of genetic errors. Stem-loop potential, the best local measure of which is base order-dependent stem-loop potential, tends to be less in exons than in introns. This is particularly evident in genes evolving rapidly under positive Darwinian selection, where the protein-encoding function is dominant. 

    Evidence is now presented that the rare regions where genes overlap also impose excessive encoding demands so that the concomitant coding of base order-dependent stem-loop potential is decreased. Our results are consistent with the hypothesis that sequences with high stem-loop potential arose in the early "RNA world." Ancestors of modern genes would have entered this world when sequences (exons) encoding cytoplasmic products, were interspersed with sequences (introns) encoding selectively advantageous stem-loops. Purine-loading pressure would also have favoured intron formation.

Key words: Base order; Information conflict; Introns; Overlapping genes; Purine-loading; Recombination; Secondary structure; Stem-loop potential

The sequence of a biological nucleic acid is the outcome of numerous, potentially conflicting, evolutionary pressures. These include the pressures to encode RNAs and proteins, to adopt a particular GC%, to purine-load mRNA-synonymous strands, and to form stem-loops (Forsdyke and Mortimer, 2000; Forsdyke, 2001a, b). One consequence of this seems to be that in many species the encoding of an RNA or protein cannot be achieved by one contiguous sequence of DNA bases (e.g. one open reading frame; ORF). Instead, the RNA or protein is encoded in segments ("exons"), which are interrupted by segments ("introns") which do not usually encode cytoplasmic products. The pressure to form stem-loops, a pressure whose adaptive value may relate to meiotic synapsis, appears of particular importance in this respect.

    In 1922 Muller suggested that the pairing of genes as parts of chromosomes undergoing meiotic synapsis, might provide clues on gene structure and replication. He noted that: 

     In 1954 he set his students an essay "How does the Watson-Crick model account for synapsis?" (Carlson, 1981). Crick took up the challenge in 1971 with his "unpairing postulate" by which the two strands of the classical DNA duplex would unpair to allow a homology search. The unpaired single strands would form stem-loop structures, and this would facilitate synapsis and recombination (Doyle 1978). Maternal and paternal chromosome homologs would mutually explore each other and test for "self" DNA complementarity, using a loop-loop "kissing" mechanism (Eguchi et al., 1991). If sufficient complementarity were found, then crossing over and recombination would occur. The main adaptive value of this would be to provide for the correction of errors in the individual homologs (Winge, 1917; Forsdyke, 1981, 1996b).

    A potential conflict between protein-encoding capacity and nucleic acid secondary structure was recognized when the first sequences became available (Salser, 1970). However, this was considered the likely result of pressures operating at the mRNA level (Ball, 1973), a view held even by some recent commentators (Seffens and Digby, 1999). 

     A role of genomic forces was suggested by the finding that stem-loop potential was genome-wide, affecting both genic and non-genic DNA (Forsdyke, 1995a-c; Heximer et al., 1996). It was proposed that the potential would conflict with a sequence's protein-encoding function to such an extent that open reading frames (ORFs) would have to be interrupted by introns in order to accommodate stem-loops. If so, the conflict would be most evident when the sequence was under strong selective pressure with respect to protein function. This was found to hold for genes evolving rapidly under positive Darwinian selection (Forsdyke, 1995b, 1996a). It should also hold for overlapping genes. Protein-encoding segments of genes which overlap protein-encoding segments of other genes should be more constrained than segments which do not overlap, and so should more powerfully exclude stem-loop potential. We here explore this hypothesis.

    A region required simultaneously to encode two proteins either in different reading frames in one strand (unidirectional transcription), or in complementary strands (bi-directional transcription), should be less able to accommodate stem-loop potential.

    A member of the tumor necrosis factor receptor family, the human B cell maturation protein encoded by the BCMA gene, has an antisense transcript encoding a putative 115 amino acid protein. Figure 1 shows that the overall folding potential (FONS) is greater (i.e. high negative values) in introns and flanking regions, than in exons. The potential is contributed partly by base composition (FORS-M), and partly by base order (FORS-D). As in previous work (Forsdyke 1995a,b), the difference between introns and exons is far from dramatic. There appears to be much "noise," and the case is best made statistically, not visually. The average base order-dependent stem-loop potential of the two introns (average FORS-D of -2.94 sd 0.56), is significantly greater (P<0.004) than that of the three exons (average FORS-D of -0.24 sd 0.56). However, this does not establish that overlapping coding functions have influenced the difference. For this, it is necessary to examine sequences in GenBank that contain both overlapping and non-overlapping exons in proximity.

    The human HLA class III region contains a gene encoding the extracellular matrix protein tenascin-X, the antisense strand of which contains the gene encoding adrenal steroid 21-hydroxylase (P450-C21). Figure 2 shows that, where exon 45 of the former overlaps exon 10 of the latter, FORS-D values (measuring base order-dependent stem-loop potential) are positive. 

    The figure also illustrates some of the difficulties in this type of analysis. In the region of exons 40 to 45 there is, as expected, an alternation between low and high base order-dependent stem-loop potentials, with low potentials corresponding with exons, and high potentials corresponding with introns. However, this is not so apparent in the case of exons 34 to 39. Indeed the largest exon (34) has high base order-dependent stem-loop potential, with a region of low potential (positive FORS-D values) in its 3' flank. This indicates that the segment of the protein corresponding to exon 34 can be encoded without impairing stem-loop potential, whereas in the 3' flank there might be a conserved region, perhaps playing a regulatory role, whose sequence cannot accommodate stem-loop potential (Forsdyke, 1995a).

GENE OR GENOME SEGMENT

NON-OVERLAP SEGMENTS

FORS-D
DIFFER-
ENCE^c

-1.38sd0.64

-3.08sd0.26

1.7

0.004

1.28sd0.32

F, G, H

-2.28sd0.49

3.57

-1.73sd0.35

-3.29sd0.94

1.55

0.077

1.24sd0.85

-0.81sd0.56

2.04

0.066

0.68sd0.53

2.23

0.01

-2.88sd0.47

-0.84sd0.74

-2.04

0.02

0.63sd0.68

1.23

0.145

3.25sd1.00

0.76sd0.47

2.50

0.042

^bT4 phage ORF 30.3 overlaps ORFs 30.4, 30.3' and 30.2. ORF Y overlaps ORF 30.2. G4 phage ORFs A and D 
overlap ORFs B, K. and E. Achyla klebsiana glutamate dehydrogenase exon 10 overlaps heat shock protein 70 
(U02504). Drosophila melanogaster fructose 1,6-bisphosphate aldolase exon 4A overlaps exon 4B. Rattus 
norvegicus gonadotrophic-releasing hormone exons 2 and 3 overlap exons 0, 1 and 4 of the SH-4 protein gene. 
Exon 2 of serine proteinase inhibitor-2  overlaps exon 1 of a mitochondrial cytochrome P-450 C27-steroid 
hydroxylase (U17375). Exon 10 of Sus scrofa cytochrome P-450 C21-steroid hydroxylase overlaps exon 39 of 
the putative swine tenascin-X gene. Exon 45 of the human tenascin-X gene overlaps exon 10 of the adrenal 
C21-steroid hydroxylase gene (u24488).

^dProbability that not significantly different from zero (unpaired t-test).

    To demonstrate that exon 45 was under significantly greater constraint than most neighboring exons, the average FORS-D values of windows whose centers overlapped the exon (3.25 kcal/mol) was compared with the average FORS-D values of windows whose centers overlapped exons 34 to 44 (0.76 kcal/mol). The difference was 2.5 kcal/mol (P = 0.042). This approach was applied to seven other genes from a variety of organisms. In all cases except one, differences were positive (penultimate column of Table 1). The average difference was 1.60 sd 0.58 (P <0.03).

    The exception is shown in Figure 3. In this case, the overlap is between a large exon (exon 2) of the gene encoding rat serine proteinase inhibitor-2, and exon 1 of a steroid 27-hydroxylase. It is possible that exon 2 is large (i.e. has not been split in the course of evolution) because it has sufficient flexibility in choice of amino acids and codons to accommodate both to stem-loop potential and to the coding needs of the antisense strand.

    In the 1950s it was found that much RNA freshly synthesized in the nucleus was locally degraded and did not reach the cytoplasm. In the 1960s eukaryotic rRNAs were found to be synthesized as long precursor RNAs ("heterogenous nuclear RNAs") which were subsequently processed by the intranuclear removal and degradation of apparently functionless internal "spacer" sequences (Harris, 1994). When prokaryotic rRNAs were found to be more compactly organized, it was appropriate to ask whether the first rRNAs to evolve had the spacer sequences, which subsequently decreased in prokaryotes, or whether the spacer sequences were later acquired in eukaryotes.

   A similar processing was later found to apply to eukaryotic mRNAs. After transcription internal sequences ("introns") were removed, and what remained in the processed mRNA constituted the "exons". Since the phenomenon had first been noted in the case of rRNA genes, which were not translated into a protein product, it was not surprising that introns were found in other RNAs with no protein product (Pfeifer and Tilghman, 1994), as well as in non-coding parts of mRNAs which had a protein product. Thus, introns interrupted genetic information per se, not just protein-encoding information, and it was difficult to associate exons with domains of protein structure or function (Weber and Kabsch, 1994; Stoltzfus et al., 1994). The same questions remained. Were introns "early" or "late"? What function(s), if any, did introns have?

    If bacteria could exist without introns, then perhaps introns had no function. Alternatively, whatever function introns had, either was not necessary in bacteria, or might be achieved in other ways. Since members of many bacterial species appeared to be under intense pressure to streamline their genomes to facilitate rapid replication, if it were possible they would have dispensed with any preexisting introns and/or would have been reluctant to acquire them. 

    On the other hand, if introns had a function and/or did not present too great a selective burden, eukaryotes would have tended to retain preexisting introns, or could have acquired them. Knowing the function of introns seemed critical for sorting out these issues.

4.2. Error-correction

    Some principles to guide investigation of a possible error-checking role for introns were presented (Forsdyke, 1981). Although the mechanism may be different to that originally proposed (Liebovitch et al., 1996), the present work is part of a growing body of evidence that introns play such a role (Forsdyke 1995a-d; 1996a, b; 1998; Heximer et al., 1996). It appears possible that the order of bases in nucleic acids have been under evolutionary pressure to develop the potential to form stem-loop structures which would facilitate "in-series" or "in-parallel" error-correction by recombination (Forsdyke, 1996b; Bell and Forsdyke, 1999; Forsdyke, 2001a,b).

    Whereas base composition tends to be a characteristic of entire genomes or large genome sectors, base order tends to be a local characteristic, like the ability to encode a protein. Despite the background "noise," which makes statistical analysis essential, base order-dependent stem-loop potential can be a sensitive indicator of functional conflict acting at the local level between a potential error-correcting function and a protein-encoding function. 

    One example of such conflict is that of genes evolving rapidly under positive Darwinian selection. Here there is a negative correlation (reciprocal relationship) between base order-dependent stem-loop potential and sequence variability (Forsdyke, 1995b, d; 1996a). Sequences varying rapidly in response to powerful environmental selective forces ("arms races") appear unable simultaneously to encode base sequences favoring the elaboration of a higher order structure of a type which might mediate meiotic chromosomal interactions. The latter function must either be relegated to those less rapidly evolving protein-encoding sequences (or parts of such sequences) where there is some flexibility in codon or amino acid choice (the main option in compact genomes), or to non-protein-encoding regions (introns and non-genic DNA).

    On the other hand, in the case of slowly evolving sequences, there may be a direct relationship between base order-dependent stem-loop potential and sequence variability. The demand of faithful reproduction of a protein, with negative Darwinian selection of individuals with mutations affecting the functionally most important parts of the protein, leaves the co-encoding of stem-loops to regions encoding functionally less important parts of the protein (such as the protein surface; Alvarez-Valin et al., 2000), and to non-protein-encoding regions (introns, non-genic DNA). High stem-loop potential can then correlate positively with high substitution rates (variability) when homologous sequences from different species are compared (Heximer et al., 1996). In the case of compact genomes with no intron option, the need to form stem-loops could sometimes have over-ridden the demands of protein structure and function, so that less-than-perfect proteins resulted (Ball, 1973). In this case the negative selection of individuals because of impaired stem-loop forming ability, would have been greater than the negative selection resulting from impaired protein function.

4.4. Demonstration of conflict

    High base order-dependent stem-loop potential (assessed here as high negative FORS-D values), appears as the "default" state of genomes, which probably arose in the early RNA world (Forsdyke, 1995a, c; Heximer et al., 1996). We have sought here "echoes" of the evolutionary "big bang" that occurred when protein-encoding capacity arose to compete with stem-loop potential for genome space. Despite the "noise" associated with DNA's evolutionary role as a "channel" for multiple forms of information, if the need to encode a protein were to conflict with the need to encode stem-loops, then it would be predicted that stem-loop potential would detectably decrease (i.e. FORS-D values would tend to become positive) in modern protein-encoding regions (e.g. exons). Both the degeneracy of the genetic code allowing changes in codons without changing the corresponding amino acid, and exchanges between amino acids of similar function, might mask the conflict. If so, an exon could be large. If not so, a potentially large exon would be split into smaller exons by the interspersion of introns with sequences of high stem-loop potential.

    In initial studies, a decrease in base order-dependent stem-loop potential (FORS-D) in exons was demonstrated, but sometimes required removing the 3' terminal exon (often the largest) from the analysis (Forsdyke, 1995a). It was found that the relationship was best shown by genes which were evolving rapidly under positive Darwinian selection. These included the snake venom phospholipases (Forsdyke, 1995b), retroviral genomes (Forsdyke, 1995d), MHC proteins (Forsdyke, 1996a), and troponin C (Forsdyke, 1996b). The extreme pressure to adapt protein function appeared to countermand accommodation of base order-dependent stem-loop potential.

    In the present work we examined the effect of the extreme demand made on the function when genes overlap. Although not visually obvious, from statistical analyses we found that base order-dependent stem-loop potential was usually countermanded (Figs. 1, 2; Table 1), with one unexplained exception (Fig. 3). Collected from a wide variety of species, we believe that the genes studied are likely to be generally representative of overlapping genes. Thus, consistent with the conflict hypothesis under investigation, our results support the generalization that regions of gene overlap support stem-loop potential less than regions of non-overlap.

4.5. Gene size and role of junk DNA

     Regions where the constraints of classical Darwinian negative and positive selection, or of the encoding of overlapping genes, would not permit accommodation of DNA stem-loop potential, would be expected to be rich in introns. The more severe the constraints, the more introns there would be, and the longer would be the length of DNA occupied by the gene (Naora and Deacon, 1982). The adaptive value of recombination involving stem-loops could have been manifest in the early RNA world, so that protein-encoding capacity would have intruded on genomes already adapted for stem-loop formation. In this respect, introns would have been "early", not "late." With each speciation event there should be changes in stem-loop potential (Forsdyke, 1996b, 1998), with the possibility of further changes in intron number and distribution. However, although this provides a possible explanation for the origin of introns, it does not appear to explain why introns are sometimes of extremely large size. A full understanding of factors affecting intron size may require a better understanding of the functions of intronic and non-genic DNA other than the stem-loop function (Forsdyke and Mortimer, 2000).

4.6. Purine-loading should promote intron formation

     Another problem relates to the existence of introns in genes with no protein product. Instead the products are RNAs, which have specific functions dependent on their secondary structures (usually selected for at the cytoplasmic level). In broad features we find that computer-derived secondary structures for an RNA molecule (using pairing energy tables for RNA bases), are similar to the structures derived for the corresponding DNA (using pairing energy tables for DNA bases; D. R. Forsdyke, unpublished work). If such stem-loop structures were sufficient for function at the DNA level, then there would be no need for introns in genes for non-protein-encoding RNAs. Thus we surmise that stem-loop structures which suffice for function at the RNA level, do not suffice for function at the DNA level. Since patterns of RNA stem-loops are influenced by the purine-loading of loops (the selective force for which probably operates at the mRNA level; Forsdyke and Mortimer, 2000), then purine-loading pressure (which would constrain DNA-specific stem-loop patterns in exons) should support stem-loop pressure in provoking the splitting of large exons.

     We thank J. Gerlach for assistance with computer configuration. Academic Press and Elsevier Science gave permissions for the inclusion of full-text versions of relevant preceding papers at our internet site.

Alvarez-Valin, F., Tort, J. F., Bernardi, G. 2000. Nonrandom spatial distribution of synonymous substitutions in the GP63 gene from Leishmania. Genetics 155,1683-1692.

Ball, L. A. 1973. Mutual influence of the secondary structure and information content of a messenger RNA. J. Theor. Biol. 41, 243-247.

Bell, S. J., Forsdyke, D. R.1999. Deviations from Chargaff's second parity rule correlate with direction of transcription. J. Theor. Biol. 197, 63-76.

Bronson, E. C., Anderson, J. N. 1994. Nucleotide composition as a driving force in the evolution of retroviruses. J. Mol. Evol. 38, 506-532.

Carlson, E. A. 1981. Genes, Radiation and Society. The Life and Work of H. J. Muller. Cornell University Press, Ithaca, pp. 390-392.

Crick, F. 1971. General model for the chromosomes of higher organisms. Nature 234, 25-27.

Doyle, G. G. 1978. A general theory of chromosome pairing based on the palindromic model of Sobell with modifications and amplification. J. Theor. Biol. 70,171-184.

Eguchi, Y., Itoh, T., Tomizawa, J. 1991. Antisense RNA. Annu. Rev. Biochem. 60, 631-652.

Forsdyke, D. R. 1981. Are Introns In-series Error-detecting Sequences? J. Theor. Biol. 93, 861-866.

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. 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. 1995c. 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. 1995d. Reciprocal relationship between stem-loop potential and substitution density in retroviral quasispecies under positive Darwinian selection. J. Mol. Evol. 41, 1022-1037.

Forsdyke, D. R. 1996a. Stem-loop potential in MHC genes: a new way of evaluating positive Darwinian selection. Immunogenetics 43, 182-189.

Forsdyke, D. R. 1996b. Different biological species "broadcast" their DNAs at different (C+G)% "wavelengths". J. Theor. Biol. 178, 405-417.

Forsdyke, D. R. 1998. An alternative way of thinking about stem-loops in DNA. A case study of the human G0S2 gene. J. Theor. Biol. 192, 489-504.

Forsdyke, D. R. 2001a. Functional constraint and molecular evolution. Encyclopedia of Life Sciences, Macmillan Reference Ltd., London.

Forsdyke, D. R. 2001b. The Origin of Species, Revisited. McGill-Queen's University Press, Montreal.

Forsdyke, D. R., Mortimer, J. R. 2000. Chargaff's legacy. Gene 261, 127-137.

Harris, H. 1994. An RNA heresy in the fifties. Trends. Biochem. Sci. 19, 303-305.

Heximer, S. P., Cristillo, A. D., Russell, L., Forsdyke, D. R. 1996. Sequence analysis and expression in cultured lymphocytes of the human FOSB gene (G0S3). DNA Cell Biol. 15, 1025-1038.

Le, S. H. Maizel, J. V. (1989). A method for assessing the statistical significance of RNA folding. J. Theor. Biol. 138, 495-510.

Liebovitch, L. S., Tao, Y., Todorov, A. T., Levine, L. 1996. Is there an error-correcting code in the base sequence of DNA? Biophys. J. 71, 1539-1544.

Muller, H. J. 1922. Variation due to change in the individual gene. Am. Nat. 56, 32-50.

Naora, H., Deacon, N. J. 1982. Relationship between the total size of exons and introns in protein-coding genes of higher eukaryotes. Proc. Natl. Acad. Sci. USA 79, 6196-6200.

Pfeifer, K., Tilghman, S. M. 1994. Allele-specific gene expression in mammals: the curious case of imprinted RNAs. Genes Devel. 8, 1867-1874.

Salser, W. 1970. Discussion. Cold Spring Harb. Symp. Quant Biol. 35, 19.

Santalucia, J., Allawi, H. T., Seneviratne, P. A. 1996. Improved nearest neighbour parameters for predicting DNA duplex stability. Biochemistry 35, 3555-3562.

Seffens, W., Digby, D. 1999. mRNAs have greater negative folding free energies than shuffled or codon choice randomized sequences. Nucleic Acids Res. 27, 1578-1584.

Stoltzfus, A., Spencer, D. F., Zuker, M., Logsdon, J. M., Doolittle, W. F. 1994. Testing the exon theory of genes: the evidence from protein structure. Science 265, 202-207.

Weber, K., Kabsch, W. 1994. Intron positions in actin genes seem unrelated to the secondary structure of protein. EMBO. J. 13, 1280-1286.

Winge, Ö. 1917. The chromosomes, their number and general importance. Compte. Rend. Trav. Lab. Carlsberg  13, 131-275.