Evolution of Relative Reading Frame Bias in Unidirectional
Prokaryotic Gene Overlaps
Peter J. A. Cock*,1,2 and David E. Whitworth3
1
MOAC Doctoral Training Centre, University of Warwick, Coventry, United Kingdom
Plant Pathology, Scottish Crop Research Institute, Invergowrie, Dundee, United Kingdom
3
Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, Ceredigion, United Kingdom
*Corresponding author: E-mail: [email protected].
Associate editor: James McInerney
2
Abstract
Pairs of unidirectional (same strand) genes can overlap in one of two phases (relative reading frames). There is a striking bias
in the relative abundance of prokaryotic gene overlaps in the two possible phases. A simple model is presented based on
unidirectional gene overlaps evolving from nonoverlapping gene pairs, through the adoption of alternative start codons by
the downstream genes. Potential alternative start codons within upstream gene sequences were found to occur at greater
frequencies in one phase, corresponding to the most prevalent phase of gene overlaps. We therefore suggest that the phase
bias of overlapping genes is primarily a consequence of the N-terminal extension of downstream genes through adoption of
new start codons.
Key words: annotation, overlaps, start codon, phase, reading frame.
observed frequencies at which alternative stop codons are
found in the reverse complement sequence of nonoverlapping convergent genes. Herein, we extend this idea to
unidirectional overlaps evolving from nonoverlapping unidirectional genes. First, C-terminal extension of the upstream gene is considered by looking at the frequencies of
alternative stop codons within the downstream gene. The
mechanism in mind is that the existing stop codon of
a nonoverlapping upstream gene is lost through a point
mutation or indel, and the gene therefore is extended to
the next stop codon, which may cause an overlap with
the downstream gene. Second, N-terminal extension of the
downstream gene by the adoption of a new start codon
is considered by looking at the frequencies of alternative
start codons within the upstream gene. Here, the creation
of overlaps from unidirectional neighboring genes is by the
adoption of a new start codon for a downstream gene (e.g.,
due to an indel or an accumulation of point mutations, perhaps in association with the loss of the original start codon).
We demonstrate that C-terminal extension of an upstream gene does not give the observed phase bias, but that
N-terminal extension of a downstream gene does. Coupled
with an exponential fitness cost to the overlap length, this
model reproduces the general character of the observed distribution of overlaps.
In total, 3, 153, 393 gene pairs were considered: 460, 065
divergent, 460, 751 convergent, and 2, 232, 577 unidirectional; a split of 14.6%, 14.6%, and 70.8%, respectively
(3 significant figures [sf]). As noted by Lillo and Krakauer
(2007), the number of divergent and convergent pairs are
expected to be almost equal. Within each orientation, the
proportion of overlapping gene pairs varies considerably.
Only 3.1% of divergent genes are annotated as overlapping
compared with 13% of convergent pairs and 21% (2 sf) of
© The Author 2009. Published by Oxford University Press on behalf of the Society for Molecular Biology and Evolution. All rights reserved. For permissions, please
e-mail: [email protected]
Mol. Biol. Evol. 27(4):753–756. 2010 doi:10.1093/molbev/msp302 Advance Access publication December 14, 2009
753
Letter
Overlapping genes are ubiquitous in viruses, and common in prokaryotes, but are also found in eukaryotes
(Normark et al. 1983; Rogozin et al. 2002; Makalowska et al.
2005). In overlapping coding sequences, the DNA simultaneously encodes portions of two separate polypeptides.
Consequently, mutation of the overlapping region has implications for both proteins, and it is therefore nontrivial
to characterize the selective pressures at work on such sequences. Neighboring genes on opposite strands of the DNA
can be convergent (tail-to-tail, →←) or divergent (headto-head, ←→). Same strand genes (head-to-tail, →→) are
unidirectional, with an upstream and a downstream gene. In
general, phases +0, +1, and +2 can be defined by the gene
separation modulo 3 (Kingsford et al. 2007) or equivalently
by overlaps of 3i , 3i − 1, and 3i − 2 (for an integer i ). For
unidirectional genes, this phase is also the relative reading
frame (Cock and Whitworth 2007).
Unidirectional (same strand) overlaps are the most common overlap orientation in prokaryotes (Fukuda et al. 2003).
Only overlaps in phases +1 and +2 are considered because
in-phase overlaps can be regarded as a single gene with alternative initiation sites. Most unidirectional gene overlaps are
of 1 or 4 bp (phase +2), but for longer overlaps, phase +1
is much more common than phase +2 (Eyre-Walker 1996;
Borodovsky et al. 1999; Johnson and Chisholm 2004; Cock
and Whitworth 2007; Lillo and Krakauer 2007).
We previously highlighted this interesting phase bias in
prokaryotic unidirectional overlaps (Cock and Whitworth
2007) and attempted to explain it with a mutual constraint
argument (akin to that of Rogozin et al. 2002 for convergent
overlaps). Although this mechanism could explain the phase
bias, it did not make testable predictions.
Kingsford et al. (2007) provided a simple model to
explain the phase bias in convergent overlaps based on the
Cock and Whitworth · doi:10.1093/molbev/msp302
FIG. 1. Observed unidirectional overlaps, showing the clear +1 phase
bias for overlaps of 7 bp or more. This figure includes 28 overlaps of
length 5, which all use the alternative start codon ATT.
unidirectional pairs. These ratios are in good agreement
with published results (Fukuda et al. 2003; Cock and
Whitworth 2007; Kingsford et al. 2007). For the rest of this
paper, we focus on unidirectional overlaps. Figure 1 shows
the observed distribution, with the clear phase +1 bias in
overlaps of 7 bp or more. This bias persists even for genomes
grouped by GC% (data not shown).
Figure 2 shows overlaps generated from the last out-offrame stop codon within each nonoverlapped downstream
gene, that is, consideration of C-terminal extension of an upstream gene by adoption of a new stop codon. This shows
only a small phase bias, but it is opposite to that observed in
the annotated genomes.
Figure 3 shows overlaps generated from the first out-offrame start codons within each nonoverlapped upstream
gene, that is, consideration of N-terminal extension of a
downstream gene by adoption of a new start codon. This
does show the same phase bias observed in the annotated
genomes (fig. 1), although the distribution of short overlaps
is very different. In particular, a large number of potential
overlaps of length 5 bp have been identified, all of which
use the alternative start codon ATT. However, only a handful of such overlaps were found in the original survey, and
this analysis is overly simplistic as the different start codons
are not necessarily equally likely. Indeed, some of the start
codons defined in the relevant NCBI translation tables (11
and 4) are only used in a small subset of the prokaryotes.
It would be possible to extend the above analysis with
an acceptance weighting based on observed start codon frequencies. Instead, figure 4 shows the more straightforward
approach of only considering the three most common start
codons (ATG, GTG, or TTG). Again there is a strong phase
754
MBE
FIG. 2. Hypothetical unidirectional overlaps generated by any potential alternative stop codon within the downstream gene of nonoverlapping unidirectional gene pairs. Note the slight +2 phase bias for
longer overlaps, counter to the observed distribution (fig. 1).
+1 bias in the longer overlaps, and now the ratios of overlaps 1 and 4 are much closer to those observed (fig. 1).
One marked difference between the observed distribution (fig. 1) and figures 3 and 4 are the very different decay rates. Kingsford et al. (2007) observed a similar
phenomenon in their analysis of convergent overlaps and
FIG. 3. Hypothetical unidirectional overlaps generated by any valid
start codon within the upstream gene of nonoverlapping unidirectional gene pairs. Any valid start codon in the relevant codon table
was accepted. Note the high number of overlaps of length 5, which all
use the start codon ATT.
Relative Reading Frame Bias in Unidirectional Prokaryotic Gene Overlaps · doi:10.1093/molbev/msp302
MBE
FIG. 4. Hypothetical unidirectional overlaps generated by any common start codon (ATG, GTG, or TTG) within the upstream gene of
nonoverlapping unidirectional gene pairs. Note that there are no overlaps of length 5.
FIG. 5. Hypothetical unidirectional overlaps generated by any common start codon (ATG, GTG, or TTG) within the upstream gene
of nonoverlapping unidirectional gene pairs, with exponential decay
scaling. Note that there are no overlaps of length 5.
resolved this with a two-stage model, whereby there is a
phase bias at overlap creation due to codon frequencies, but
with overlap length subject to an exponential fitness, which
can be determined empirically to match the observed data.
Figure 5 show the data normalized with an exponential decay of 0.0931 (least squares difference fitting for overlaps up
to 200 bp). There is good agreement with figure 1, but the
phase bias is slightly less pronounced.
Two mechanisms for neighboring unidirectional genes
to become overlapped were considered. C-terminal extension of the upstream gene (fig. 2) does not explain the
observed pattern in unidirectional gene overlaps (fig. 1).
However, considering N-terminal extension of the downstream gene by looking for alternative start codons does
predict the observed phase bias (figs. 3 and 4), largely explained as due to the relative frequencies of alternative start
codons in the two reading frames. Together with an exponential fitness criteria on the overlap length, this predicts
a distribution close to that observed (fig. 5). This proposed
exponential fitness cost could be due to the metabolic burden of making a longer protein, the increased likelihood of
problems with protein misfolding/aggregation, or a combination of these or other effects. This model is compatible
with mutual sequence constraint arguments as in Cock and
Whitworth (2007) but provides a much clearer explanation.
Sabath et al. (2008) describe a related analysis of unidirectional overlaps, which also found that the phase bias in
longer unidirectional overlaps could be explained in terms
of the relative abundance of alternative start codons within
an upstream gene and rejected the complementary explanation of the adoption of alternative stop codons within a
downstream gene. Rather than observing alternative start
and stop codons directly from real gene sequences, their frequencies were inferred from di-codon frequencies taken as
the product of observed codon frequencies (assuming dicodon frequencies are independent). This cannot capture
any differences in codon bias within genes, for example, between the 5 and 3 terminal regions of a gene. Also, a much
smaller data set was used, drawing on annotated overlaps
from only 167 genomes. Nevertheless, their work is supportive of the results given here.
Translational coupling provides a biological reason for
short unidirectional gene overlaps and thus, the very large
number of overlaps of 1 or 4 bp. This may also apply to 5
bp overlaps, which could be tested in vitro, suggesting that
the high number of unidirectional overlaps generated using the “rare” alternative start codon ATT (fig. 3) may be
of biological relevance, with the handful of cases annotated
being just the tip of the iceberg. Eyre-Walker (1996) noted
skewed ratios of alternative stop codons in short overlaps of
1 or 4 bp, so atypical start codon usage in this context is not
unreasonable.
Although the genetic code itself appears to induce these
phase biases in longer overlaps, without searching for Shine–
Dalgarno translation initiation sites or direct experimental evidence, it is not clear how many of these annotated
long unidirectional overlaps are biologically relevant. Although translational coupling provides a biological reason
for short unidirectional gene overlaps, it may not apply to
the longer overlaps reported. A recent analysis by Pallejá
et al. (2008) concluded all unidirectional overlaps over 60
bp in their data set of 338 prokaryotic genomes were misannotations, but did identify some “real overlaps.” However,
the phase bias is still found when only genes with annotated
755
MBE
Cock and Whitworth · doi:10.1093/molbev/msp302
functionality are considered (Cock and Whitworth 2007;
Lillo and Krakauer 2007; Sabath et al. 2008), and 98.1 % (3
sf) of annotated unidirectional overlaps are 60 bp or less.
The phase patterns in overlapping genes have no immediately apparent evolutionary function, but rather are inherently linked to the genetic code itself, which has evolved
under various pressures. Itzkovitz and Alon (2007) explored
a range of hypothetical genetic codes and concluded that
those observed in nature are near optimal for encoding additional information within a protein sequence. This work
did not specifically mention nucleotide sequences simultaneously encoding two proteins, but rather arbitrary (short)
sequences representing possible DNA-binding regions or
other motifs. Furthermore, the presence of alternative outof-frame stop codons within a gene (hidden stop codons)
has been looked at from the point of view of robustness
to translational frameshift errors (Seligmann and Pollock
2004). The standard genetic code was found to terminate
erroneous reads sooner than hypothetical genetic codes,
which is advantageous as less resources are wasted constructing and degrading nonfunctional proteins. It seems
reasonable that functions like double coding and hidden
stop codons may have shaped the genetic code and thus indirectly contributed to the overlap phase patterns observed.
Methods
Separation/overlap frequencies were tabulated for divergent, convergent, and unidirectional adjacent annotated
genes in 1, 800 GenBank files for the 962 bacterial or archaeal
species available from the NCBI as of 7 September 2009. For
simplicity, any genes with nonexact locations, ambiguous
sequences, internal stop codons, invalid start or stop codons
(as verified using the declared genetic code), or special cases
with noncontinuous coding sequences (e.g., from ribosomal
slippages) were excluded, as were cases where one gene was
entirely within another (fully overlapped). Separated gene
pairs with any ambiguous sequence between them were also
excluded.
For each nonoverlapping unidirectional gene pair, the
downstream gene was searched for the first out-of-frame
stop codon, whose location determined the length of a hypothetical unidirectional gene overlap. Additionally, the upstream gene was searched for the last out-of-frame start
codon, the location of which also determined a hypothetical unidirectional gene overlap. If, due to the presence of a
nearby in-frame stop codon, this hypothetical gene would
be encoded within the upstream gene, it was rejected. Two
variants of the start codon analysis were performed, first
looking for any valid potential start codon in the genetic
code declared for that organism and second looking only for
the most typically used start codons (ATG, GTG, or TTG).
756
The analysis was written in Python using Biopython
(Cock et al. 2009). Figures were drawn with R (R Development Core Team 2007).
Acknowledgment
This work was supported by the Engineering and Physical
Sciences Research Council via the MOAC Doctoral Training
Centre (studentship to P.J.A.C.).
References
Borodovsky M, Hayes WS, Lukashin AV. 1999. Statistical predictions
of coding regions in prokaryotic genomes by using inhomogeneous Markov models. In: Charlebois R, editor. Organisation of the
prokaryotic genome. Washington, DC: ASM Press. p. 11–33.
Cock PJA, Antao T, Chang JT, et al. (11 co-authors). 2009. Biopython:
freely available Python tools for computational molecular biology
and bioinformatics. Bioinformatics 25(11):1422–1423.
Cock PJA, Whitworth DE. 2007. Evolution of gene overlaps: relative
reading frame bias in prokaryotic two-component system genes.
J Mol Evol. 64(4):457–462.
Eyre-Walker A. 1996. The close proximity of Escherichia coli genes: consequences for stop codon and synonymous codon use. J Mol Evol.
42(2):73–78.
Fukuda Y, Nakayama Y, Tomita M. 2003. On dynamics of overlapping
genes in bacterial genomes. Gene 323:181–187.
Itzkovitz S, Alon U. 2007. The genetic code is nearly optimal for allowing additional information within protein-coding sequences.
Genome Res. 17:405–412.
Johnson ZI, Chisholm SW. 2004. Properties of overlapping genes are
conserved across microbial genomes. Genome Res. 14:2268–2272.
Kingsford C, Delcher AL, Salzberg SL. 2007. A unified model explaining
the offsets of overlapping and near-overlapping prokaryotic genes.
Mol Biol Evol. 24(9):2091–2098.
Lillo F, Krakauer D. 2007. A statistical analysis of the three-fold evolution of genomic compression through frame overlaps in prokaryotes. Biol Direct 2:22.
Makalowska I, Lin C-F, Makalowski W. 2005. Overlapping genes in vertebrate genomes. Comput Biol Chem. 29(1):1–12.
Normark S, Bergstrom S, Edlund T, Grundstrom T, Jaurin B,
Lindberg FP, Olsson O. (1983). Overlapping genes. Annu Rev Genet.
17:499–525.
Pallejá A, Harrington ED, Bork P. 2008. Large gene overlaps in prokaryotic genomes: result of functional constraints or mispredictions?
BMC Genomics 9:335.
R Development Core Team. 2007. R: a language and environment for
statistical computing [Internet]. Vienna (Austria): R Foundation
for Statistical Computing. ISBN 3-900051-07-0.
Rogozin IB, Spiridonov AN, Sorokin AV, Wolf YI, Jordan IK, Tatusov RL,
Koonin EV. 2002. Purifying and directional selection in overlapping
prokaryotic genes. Trends Genet. 18(5):228–232.
Sabath N, Graur D, Landan G. 2008. Same-strand overlapping genes
in bacteria: compositional determinants of phase bias. Biol Direct
3(1):36.
Seligmann H, Pollock DD. 2004. The ambush hypothesis: hidden stop
codons prevent off-frame gene reading. DNA Cell Biol. 23(10):
701–705.