On 2017 Jun 23, Joshua L Cherry commented:

NO EVIDENCE THAT SELECTION DRIVES SWITCHING

This article claims that “the great majority of codon set switches proceed by two consecutive nucleotide substitutions…and are driven by selection”. The data in fact support a predominance of simultaneous switches that are not driven by selection. Even if we assume that sequential switches predominate, the implication that selection increases their rate by a factor of ~50 is unjustified. Moreover, selection against the non-serine intermediate is expected to decrease the rate of sequential switches, not to drive them. The authors’ argument to the contrary is analogous to arguing that a mountain range between two locations speeds the journey between them because it accelerates the downhill portion of the trip.

Inappropriate standard of comparsion

The usual way to establish that an evolutionary process is driven by selection is show that it is faster than some process that is largely unaffected by selection. The authors instead made comparisons to “expectations” derived from nonsynonymous substitution rates, which are greatly decreased by selection. This comparison cannot establish that switching is driven by selection, and would greatly overestimate any such effect.

A more appropriate comparison would be to expectations derived from synonymous rates. These are several-fold higher than nonsynonymous rates, and “expectations” involve products of two rates. Thus, the claimed acceleration by selection mostly or entirely disappears with a proper standard of comparison.

Unjustified rejection of simultaneous switching

The authors reject a significant role for simultaneous double mutation in switching because the rate of switching is higher than the rate of analogous double changes in noncoding sequences by a factor of 5-10. This argument would be valid if non-coding regions were evolving nearly neutrally, but this is far from the case: rates of non-coding transversions (Fig. 3) are comparable to the rates of some nonsynonymous transversions (Fig 2).

I have determined that the rates of the relevant synonymous transversions are higher than the corresponding single-base non-coding changes by a factor of >5. This presumably reflects purifying selection in non-coding regions. The effect of selection on simultaneous tandem changes is expected to be larger. Thus, the excess of serine switches over non-coding tandem changes can easily be explained by selection in non-coding regions. Put differently, we can estimate a lower limit on the rate of simultaneous serine switches, and it corresponds to the majority of the observed switches.

Slow- vs. fast-evolving genes

The article claims to have shown that the rate of switching is “higher in conserved genes than in nonconserved genes in full agreement with the selection hypothesis”. The results (Fig. 6) in fact demonstrate just the opposite: the rate of switching in “nonconserved” genes (0.0032 or 0.0022) is about three times higher than that in “conserved” genes (0.0010 or 0.0008).

The authors considered the ratio of the switching rate to a sum of products of nonsynonymous rates. This ratio is higher in “conserved” genes only because the nonsynonymous rates are lower in “conserved” genes. This is true almost by the definition of “conserved” (low dN/dS), and has nothing to do with serine codon switches.

Theoretical expectation

Under the simple selection scheme considered by the authors, selection will actually decrease the rate of sequential switching. After fixation of a deleterious Ser->Thr or Ser->Cys mutant, selection will indeed increase the fixation probability of a mutant that restores Ser. However, selection always decreases the fixation probability of the initial deleterious mutant by a larger factor. As illustrated here, the product of the two relative fixation probabilities, and hence the relative probability of a switch during a short interval, is always less than one (selection slows switching) for nonzero s, and it decreases monotonically and approaches zero as the strength of selection (|Nes|) increases.

The above implicitly assumes weak mutation, but the same conclusion holds outside of this regime (Kimura, 1985).

Conclusion

Neither the data nor the authors’ model supports the claim that serine codon switching is driven by selection or has an especially “high frequency”. In fact, both data and theory point to the opposite conclusion.

References

Kimura, M (1985) The role of compensatory neutral mutations in molecular evolution. Journal of Genetics 64(1):7-19.

On 2016 Nov 03, Donald Forsdyke commented:

CONSIDERATION OF NUCLEIC ACID LEVEL SELECTION?

The authors set out "to investigate the evolutionary factors that affect serine codon set switches" (i.e. between TCN and AGY). Their "findings imply unexpectedly high levels of selection" (1). Indeed, the data strongly support the conclusion that codon mutations "are driven by selection." It is conjectured that the codon mutation "switch would involve as an intermediate either threonine ACN or cysteine TGY, amino acid residues with properties substantially different from those of serine, so that such changes are unlikely to be tolerated at critical functional or structural sites of a protein."

However, it does not follow that the unsuitability of the interim amino acids drove the rapid tandem substitutions. Choice of "coincident codons" has long been seen as influenced by pressures acting at the nucleic acid level (2-4). These pressures evolve in parallel with, and sometimes dominate, protein pressures. One example is purine-loading pressure (3). If this cannot be satisfied by changes at third codon positions, then sometimes the organism must accept a less favorable amino acid. With serine codons, a change from TCN to AGY (i.e. first and second codon positions) can increase purine-loading pressure without compromising the amino acid that is encoded see Ref. 3.

1.Rogozin IB, Belinky F, Pavlenko V, Shabilina SA, Kristensen DM, Koonin EV (2016) Evolutionary switches between two serine codon sets are driven by selection. Proc Natl Acad Sci USA www.pnas.org/cgi/doi/10.1073/pnas.1615832113 Rogozin IB, 2016

2.Bains W. (1987) Codon distribution in vertebrate genes may be used to predict gene length. J Mol Biol 197:379-388. Bains W, 1987

3.Mortimer JR, Forsdyke DR (2003) Comparison of responses by bacteriophage and bacteria to pressures on the base composition of open reading frames. Appl Bioinf 2: 47-62. Mortimer JR, 2003

4.Forsdyke DR (2016) Evolutionary Bioinformatics, 3rd edition (Springer, New York).

On 2016 Nov 03, Donald Forsdyke commented:

CONSIDERATION OF NUCLEIC ACID LEVEL SELECTION?

The authors set out "to investigate the evolutionary factors that affect serine codon set switches" (i.e. between TCN and AGY). Their "findings imply unexpectedly high levels of selection" (1). Indeed, the data strongly support the conclusion that codon mutations "are driven by selection." It is conjectured that the codon mutation "switch would involve as an intermediate either threonine ACN or cysteine TGY, amino acid residues with properties substantially different from those of serine, so that such changes are unlikely to be tolerated at critical functional or structural sites of a protein."

However, it does not follow that the unsuitability of the interim amino acids drove the rapid tandem substitutions. Choice of "coincident codons" has long been seen as influenced by pressures acting at the nucleic acid level (2-4). These pressures evolve in parallel with, and sometimes dominate, protein pressures. One example is purine-loading pressure (3). If this cannot be satisfied by changes at third codon positions, then sometimes the organism must accept a less favorable amino acid. With serine codons, a change from TCN to AGY (i.e. first and second codon positions) can increase purine-loading pressure without compromising the amino acid that is encoded see Ref. 3.

1.Rogozin IB, Belinky F, Pavlenko V, Shabilina SA, Kristensen DM, Koonin EV (2016) Evolutionary switches between two serine codon sets are driven by selection. Proc Natl Acad Sci USA www.pnas.org/cgi/doi/10.1073/pnas.1615832113 Rogozin IB, 2016

2.Bains W. (1987) Codon distribution in vertebrate genes may be used to predict gene length. J Mol Biol 197:379-388. Bains W, 1987

3.Mortimer JR, Forsdyke DR (2003) Comparison of responses by bacteriophage and bacteria to pressures on the base composition of open reading frames. Appl Bioinf 2: 47-62. Mortimer JR, 2003

4.Forsdyke DR (2016) Evolutionary Bioinformatics, 3rd edition (Springer, New York).

On 2017 Jun 23, Joshua L Cherry commented:

NO EVIDENCE THAT SELECTION DRIVES SWITCHING

This article claims that “the great majority of codon set switches proceed by two consecutive nucleotide substitutions…and are driven by selection”. The data in fact support a predominance of simultaneous switches that are not driven by selection. Even if we assume that sequential switches predominate, the implication that selection increases their rate by a factor of ~50 is unjustified. Moreover, selection against the non-serine intermediate is expected to decrease the rate of sequential switches, not to drive them. The authors’ argument to the contrary is analogous to arguing that a mountain range between two locations speeds the journey between them because it accelerates the downhill portion of the trip.

Inappropriate standard of comparsion

The usual way to establish that an evolutionary process is driven by selection is show that it is faster than some process that is largely unaffected by selection. The authors instead made comparisons to “expectations” derived from nonsynonymous substitution rates, which are greatly decreased by selection. This comparison cannot establish that switching is driven by selection, and would greatly overestimate any such effect.

A more appropriate comparison would be to expectations derived from synonymous rates. These are several-fold higher than nonsynonymous rates, and “expectations” involve products of two rates. Thus, the claimed acceleration by selection mostly or entirely disappears with a proper standard of comparison.

Unjustified rejection of simultaneous switching

The authors reject a significant role for simultaneous double mutation in switching because the rate of switching is higher than the rate of analogous double changes in noncoding sequences by a factor of 5-10. This argument would be valid if non-coding regions were evolving nearly neutrally, but this is far from the case: rates of non-coding transversions (Fig. 3) are comparable to the rates of some nonsynonymous transversions (Fig 2).

I have determined that the rates of the relevant synonymous transversions are higher than the corresponding single-base non-coding changes by a factor of >5. This presumably reflects purifying selection in non-coding regions. The effect of selection on simultaneous tandem changes is expected to be larger. Thus, the excess of serine switches over non-coding tandem changes can easily be explained by selection in non-coding regions. Put differently, we can estimate a lower limit on the rate of simultaneous serine switches, and it corresponds to the majority of the observed switches.

Slow- vs. fast-evolving genes

The article claims to have shown that the rate of switching is “higher in conserved genes than in nonconserved genes in full agreement with the selection hypothesis”. The results (Fig. 6) in fact demonstrate just the opposite: the rate of switching in “nonconserved” genes (0.0032 or 0.0022) is about three times higher than that in “conserved” genes (0.0010 or 0.0008).

The authors considered the ratio of the switching rate to a sum of products of nonsynonymous rates. This ratio is higher in “conserved” genes only because the nonsynonymous rates are lower in “conserved” genes. This is true almost by the definition of “conserved” (low dN/dS), and has nothing to do with serine codon switches.

Theoretical expectation

Under the simple selection scheme considered by the authors, selection will actually decrease the rate of sequential switching. After fixation of a deleterious Ser->Thr or Ser->Cys mutant, selection will indeed increase the fixation probability of a mutant that restores Ser. However, selection always decreases the fixation probability of the initial deleterious mutant by a larger factor. As illustrated here, the product of the two relative fixation probabilities, and hence the relative probability of a switch during a short interval, is always less than one (selection slows switching) for nonzero s, and it decreases monotonically and approaches zero as the strength of selection (|Nes|) increases.

The above implicitly assumes weak mutation, but the same conclusion holds outside of this regime (Kimura, 1985).

Conclusion

Neither the data nor the authors’ model supports the claim that serine codon switching is driven by selection or has an especially “high frequency”. In fact, both data and theory point to the opposite conclusion.

References

Kimura, M (1985) The role of compensatory neutral mutations in molecular evolution. Journal of Genetics 64(1):7-19.