Figure 1 shows this for the three frames in a coding sense strand (CSF1, CSF2, CSF3), and for the three corresponding frames in a coding antisense strand (CAF1, CAF2, CAF3). It is seen that the ORF designated CSF1 is the only frame in the sense strand which corresponds to RNY. Similarly, CAF1 is the only frame in the antisense strand which corresponds to RNY. Thus, an antisense NSF should usually be in the same frame as that of the corresponding ORF (Alff-Steinberger 1987). It can also be seen that CSF2 (NYR) corresponds to CAF3, not CAF2. Similarly, CSF3 (YRN) corresponds to CAF2, not CAF3. This was recently confirmed by Yomo and Urabe (1994) in a study of long codon sets generated by combining E. coli coding regions into one long sequence. Thus their results are consistent with the proposal of a prototypic RNY code, as well as with the existence of genes in the antisense strand (as they suggest).

Long NSFs in High (G+C) DNA

    If there are no stop codons (TAG, TAA, TGA) in the NSF, then there can be no complementary codons (CTA, TTA, TCA) in the corresponding ORF. CTA and TTA code for leucine. TCA codes for serine. These two amino acids are found frequently in proteins. Thus, except for an unlikely ORF with no leucine or serine, whether there will be a long NSF depends on which synonymous codons for leucine and serine are used in the corresponding ORF. Leucine and serine are two of the three amino acids for which there are six synonymous codons. An important factor determining codon choice is the species-specific pressure on a genome to adopt a particular (G+C)% base composition ("GC/AT pressure"; reviewed by Filipski 1990). AT-rich codons are used infrequently in species with a high (G+C)% (i.e. RNY is usually GNC).

    In Figure 2a the combined average frequencies of the usages of CTA, TTA and TCA as codons in sets of genes from various species are plotted against the corresponding average percentages of G+C in coding regions calculated from the frequency and GC content of all the codons of each gene set. The latter should provide a measure of GC/AT pressure. The data are taken from the species codon usage profiles of Wada et al. (1990). The right ordinate shows the average length of NSF expected at particular values of the usage of CTA, TTA and TCA. Thus 1% usage of the codons implies an average NSF of 100 codons. An extreme example is Rhodobacter capsulatum (RCA) which uses CTA, TTA and TCA at a frequency of only 0.05%, implying an average NSF of 2000 codons in the 21 genes examined by Wada et al. (1990).

     As the (G+C)% increases, preferred codons become increasingly GC-rich, and the combined usage of the AT-rich codons CTA, TTA and TCA decreases. It was found empirically that there is an approximately rectilinear relationship between the logarithm of the sum of the percentages of CTA, TTA and TCA, and the average (G+C)% of the corresponding coding regions. However, at high G+C percentages, values for the frequency of CTA+TTA+TCA for two species, Azotobacter vinelandii (AVI) and Rhodobacter capsulatus, are below the lower limit of the 95% prediction interval (estimated using Minitab statistics software; Ryan and Joiner 1994). TCA participates equally with CTA and TTA in decreasing in frequency disproportionately at high (G+C)% in these species. A species with more GC-rich codons, Streptomyces (STM), remains close to the regression line.

     The divergence is also apparent in the case of GC-rich plasmid genes (shown as triangles), which encode bacterial nylon-degrading enzymes (Yomo, Urabe and Okada 1992). Except for the gene NA26AHDH, the values for the sum of the percentages of CTA+TTA+TCA as codons in these genes are zero (corresponding to a completely open average NSF). The value for NA26AHDH is close to the regression line and corresponds to an average NSF of 400 codons. However, the difference between NA26AHDH and the other plasmid genes is due to only one codon (TCA is present once in NA26AHDH). In a series of only four genes this difference is not statistically significant, so does not support the suggestion that there is some special mechanism protecting the plasmid NSFs from accumulating mutations that generate stop codons (Yomo et al 1992;1994). Such a special mechanism might be expected to affect only CTA, TTA and TCA. However, in the plasmids there are zero frequencies not only of these codons, but also of seven of the eight codons of general formula W₃ (e.g. TTT, ATA, etc.), and of about half the 24 codons of general formula W₂S (e.g. AGA, TGT, etc.). It seems likely that, as suggested by Ikehara and Okazawa (1993), the plasmid genes are merely responding to GC/AT pressure, and not to some "unknown force".

An "Unknown Force" in Rhodobacter and Azotobacter?

    Assuming that the observed regression line (Fig. 2a) is a close representation of the relationship between GC/AT pressure and the frequencies of CTA+TTA+TCA, then species whose frequencies of these codons fall close to the regression line would appear to be responding simply to GC/AT pressure. The species whose frequencies of the codons diverge significantly below the line at high GC/AT pressures would lend some credence to the "unknown force" postulate.

    The average NSF for all the 37 genes of Azotobacter vinelandii, and for all the 21 genes of Rhodobacter capsulatus (that were studied by Wada et al. 1990), is greater than predicted by the regression line. The "unknown force" would seem to have acted on all the genes of these species. It would seem unlikely that all 58 genes would have sufficient flexibility in the coding of the sense strand product to permit the evolution of a useful antisense strand product. Thus, either pressures to form each antisense protein have not been causing these genes to loose CTA, TTA and TCA, or the only way pressure to form antisense protein can work is by committing numerous "innocent bystander" genes to the same strategy; that is, the evolution of one particular antisense product was so advantageous that all the genes in the organism were required to respond to some "unknown force" in order to acquire the potential to generate an antisense product. Alternatively, the organisms could be responding to some intrinsic species specific "unknown force" the blind consequences of which would be NSFs longer than predicted from (G+C)%.

Role for TpA Pressure in Rhodobacter?

    In Rhodobacter capsulatus two codons of general form NTA (CTA and TTA) have zero frequency, and TCA has a frequency of 0.05%. The latter value is the lowest of all the species tabulated by Wada et al. (1990), except for Azotobacter vinelandii (0.01%). The low values are not specific to these three codons. Figure 2b, shows plots of the combined frequencies of two other codons of general form NTA (GTA, encoding valine; ATA, encoding isoleucine). As in Figure 2a, most of the data-points are close to the regression line. The point corresponding to Rhodobacter capsulatus diverges below the regression line. Indeed in this species the usage of GTA + ATA is zero, which is beyond the 95% prediction interval (Ryan and Joiner 1994). Thus, it is possible that the divergence may be ascribed, in part, to "TpA pressure", which is a "known" but not well understood pressure (Nussinov 1984; Alff-Steinberger, 1987; Barrai et al. 1991). TpA pressure affects all codons of general forms NTA and TAN (Forsdyke, 1995d).

     The point corresponding to Azotobacter vinelandii is closer to the regression line and is within the 95% prediction interval. In this species the average usages of GTA and ATA are 0.27% and 0.06% respectively. With respect to these codons this organism may simply be responding to GC/AT pressure. Thus TpA pressure does not appear to explain the low levels of CTA and TTA in Azotobacter vinelandii. Here the "unknown force" postulate may be valid.

Similar Codon Frequencies in Sense and Antisense Strands

     Sense and antisense strands have similar codon frequencies in various organisms (Alff-Steinberger 1984, 1987). Furthermore, trinucleotides with the potential to encode hydrophilic amino acids (e.g. GAA; glutamate), are present at approximately the same frequencies as their complements, which often have the potential to encode hydrophobic amino acids (e.g. UUC; phenylalanine). However, studies of the frequencies of dinucleotides (Nussinov 1984), trinucleotides (Yomo and Ohno 1989; Ohno and Yomo 1991), and higher oligonucleotides (Pradhu 1993), show this to be a general characteristic of DNA sequences, both coding and in non-coding.

    One explanation for this, proposed by Nussinov (1982), was that DNA might contain numerous inverted repeats. These would confer on DNA the ability to form stem-loop, cruciform, structures. Indeed, evidence obtained by comparing the optimum folded structures of "windows" in natural sequences with the optimum folded structures of the same windows in which the order of bases have been randomized, suggests that all DNA sequences have been under an evolutionary selection pressure to maximize the ability to form local stem-loop structures ("fold pressure"). This affects both protein-coding and non-coding regions (introns and intergenic regions; Forsdyke 1995b,c).