A periodic pattern of mRNA secondary structure
created by the genetic code
Svetlana A. Shabalina*, Aleksey Y. Ogurtsov and Nikolay A. Spiridonov1
National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bethesda,
MD 20894, USA and1Division of Therapeutic Proteins, Center for Drug Evaluation and Research, US Food and Drug
Administration, Bethesda, MD 20892, USA
Received March 2, 2006; Revised March 21, 2006; Accepted April 5, 2006
Single-stranded mRNA molecules form secondary
structures through complementary self-interactions.
Several hypotheses have been proposed on the rela-
tionship between the nucleotide sequence, encoded
We performed the first transcriptome-wide in silico
analysis of the human and mouse mRNA foldings
and found a pronounced periodic pattern of nucleo-
tide involvement in mRNA secondary structure.
We show that this pattern is created by the structure
of the genetic code, and the dinucleotide relative
abundances are important for the maintenance of
mRNA secondary structure. Although synonymous
codon usage contributes to this pattern, it is intrinsic
to the structure of the genetic code and manifests
itself even in the absence of synonymous codon
usage bias at the 4-fold degenerate sites. While all
codon sites are important for the maintenance of
mRNA secondary structure, degeneracy of the code
allows regulation of stabilityand periodicityof mRNA
secondary structure. We demonstrate that the third
degenerate codon sites contribute most strongly to
mRNA stability. These results convincingly support
the hypothesis that redundancies in the genetic code
tein structure and RNA structure. Our data show that
selection may be operating on synonymous codons
ary structure, which is likely to be important for tran-
script stability and translation. We also demonstrate
region (50-UTR), CDS and 30-UTR] preferentially fold
onto themselves, while the start codon and stop
codon regions are characterized by relaxed second-
ary structures, which may facilitate initiation and
termination of translation.
In 1972 White et al. (1) suggested that the redundancy in the
genetic code permits extensive variation of the nucleotide
sequence and allows the satisfaction of requirements for
both protein structure and RNA structure. Ball (2) proposed
three alternative hypotheses on the relationship between
amino acid sequences and mRNA secondary structure.
First, the choice of codons and their sequence in the message
could be independent of the resulting secondary structure of
the mRNA. Second, optimization of mRNA secondary struc-
ture may occur only within the limits of encoded amino acid
sequence. Third, selection pressure for specific RNA second-
ary structure could affect the choice of nucleotide at both
synonymous and non-synonymous positions. Fitch (3) exa-
mined these hypotheses and found evidence of the use of
the degeneracy of the genetic code to optimize base pairing
in mRNA molecules. He discussed the third hypothesis as
being biologically plausible. However, he did not find evid-
ence for (or against) the notion that the needs of RNA structure
and function must compete with the needs of protein structure
Since then, the idea that the redundancy of the genetic code
allows preservation of mRNA folding has been supported by
several lines of evidence. Periodical patterns complementary
to the proof-reading site in the ribosome and presumably
involved in the translation frame monitoring mechanism
have been found in many transcripts (4). It was shown that
synonymous substitutions affect mRNA translation in differ-
ent organisms (5–7). Strong mRNA secondary structures
formed due to gene-specific codon usage have been implicated
*To whom correspondence should be addressed. Tel: +1 301 594 5693; Fax: +1 301 480 2290; Email: firstname.lastname@example.org
The authors wish it to be known that, in their opinion, the first two authors should be regarded as joint First Authors
? The Author 2006. Published by Oxford University Press. All rights reserved.
The online version of this article has been published under an open access model. Users are entitled to use, reproduce, disseminate, or display the open access
version of this article for non-commercial purposes provided that: the original authorship is properly and fully attributed; the Journal and Oxford University Press
only in part or as a derivative work this must be clearly indicated. For commercial re-use, please contact email@example.com
Nucleic Acids Research, 2006, Vol. 34, No. 8
in discontinuous translation and pauses in synthesis of insect
silk fibroin, chicken collagen and other proteins (8,9). These
and similar works [for a review see (10)] gave rise to the
expectations that secondary structures can interfere with trans-
lation and therefore should be avoided in mRNA coding
regions. Contrarily to this opinion, significant biases in
favor of local RNA structures have been found in several
bacterial species and the yeast (11). Although evolutionarily
conserved local secondary structures were described in eukar-
yotic and mammalian mRNAs and pre-mRNAs (12), no con-
clusive evidence has been found to confirm or disprove the
hypothesis that selection for RNA structure can lead to non-
optimal amino acid usage. Seffens and Digby (13) reported
thatnativemRNAshavealower calculatedfolding freeenergy
than random sequences. Correlations between mRNA and pro-
tein secondary structures have been noted (14). Following Jia
et al. (14), Chamary and Hurst (15) suggested that base com-
position at the third synonymous site is driven by the nucleo-
tide usage of amino acids, and the requirement for elevated
C at 4-fold degenerate sites is related to usage of encoded
amino acids in alpha helices and beta sheets.
Non-random use of synonymous triplets coding for same
amino acids has been observed in genomes from different life
forms. The biological basis of unequal codon choice is not
completely clear. Positive correlation between synonymous
codon usage bias and gene expression level was established
in bacteria (5,16), yeast (5), nematode (17) and insect (18). In
many cases, preferred codons correspond to the most abundant
iso-accepting tRNAs, which was explained by evolutionary
ever, evidence supporting translational selection of codon
choice is arguable (19–21). No correspondence between the
usage of a codon in human protein coding sequences and the
abundance of iso-accepting tRNA has been found in several
studies (22–26). Recently, Lavner and Kotlar (27) reported a
weak positive correlation between expression level and fre-
quency of optimal codons for human genes.
Important observations suggesting the functional role for
degeneratesites inthe maintenanceofmRNAsecondary struc-
ture were derived from mutational studies. Analysis of syn-
onymous nucleotide polymorphism in enteric bacteria and
compensatory nucleotide substitutions in Drosophila sugg-
ested selective constraint on mRNA secondary structures
(28,29). Conservation of secondary structure features was
demonstrated for retroviral mRNA. It was shown that folding
in RNA stem regions disrupted by silent mutations on one
strand of retroviral RNA is restored by compensatory muta-
tions on the other strand (30). Mutations in GC-rich secondary
structures in complex 50-untranslated regions (50-UTRs) that
provide skaffold forinteractions with trans-acting proteins can
have implications in disease and tumorogenesis (31). It was
shownthat the location of synonymous mutations in the mouse
lineages is non-random with respect to mRNA stability (15),
and substitutions at the third synonymous positions affect
mRNA decay rates (32) and translation (5–7). Moreover, syn-
onymous mutations affecting mRNA structure and decay rate
can be highly deleterious and have implications in disease in
Here we report results of the transcriptome-scale in silico
analysis of the human and mouse mRNAs. We describe
general structural properties of mammalian protein coding
transcripts and demonstrate that the structure of the genetic
code creates specific periodic pattern of nucleotide base pair-
ing in mRNA coding regions. We show that degenerate codon
sites are important for maintaining a more ordered and stable
mRNA secondary structure in the protein coding regions. We
also demonstrate that mRNA functional domains (50-UTR,
CDS, 30-UTR) preferentially self-fold, and regions involved
in translation initiation and termination are characterized with
reduced levels of secondary structures.
MATERIALS AND METHODS
Non-redundant datasets of the human (19317 sequences) and
mouse (20892 sequences) mRNAs were compiled from the
RefSeq database for the human and mouse genomes (ftp://ftp.
ncbi.nlm.nih.gov/genomes). Only annotated mRNA sequ-
ences with complete CDSs (400 nt or longer) possessing
the 50- and 30-UTRs (50 nt or longer) were analyzed in this
study. A dataset of 6919 orthologous human–mouse mRNA
pairs with annotated 50-UTRs and 30-UTRs and with properly
aligned start and stop codons used in this study was described
previously (35). Vista computational tool was used for align-
ment visualization (36,37). Symmetrical best hits between
proteins from the respective genomes were identified using
the BLAST program (http://www.ncbi.nlm.nih.gov/BLAST/).
Nucleotide sequence alignments for identified orthologous
pairs of human–mouse 50-UTRs and 30-UTRs were produced
with the OWEN program (38). For the CDS, the alignment of
the nucleotide sequences was guided by the amino acid
sequence alignments. The positions of 50- and 30-UTRs
were taken from the feature tables of the GenBank entries.
The degree of conservation at each nucleotide position was
calculated as the number of matches over the number of pair-
wise alignments. The start codon and the stop codon provided
natural reference points for this analysis, the position number
was always determined as a distance from one of these codons.
Relative abundance of human mRNAs was estimated from the
numbers of the corresponding expressed sequence tags (ESTs)
from normal tissues in GenBank.
To investigate the role of the genetic code and codon usage
in mRNA secondary structure formation, we compared the
folding of native mRNAs with foldings computed for
sequences randomized with different methods. For each
mRNA sequence, we constructed several randomized sequ-
ences using procedures similar to those described by Seffens
and Digby (13). The first randomization procedure shuffled
nucleotides at the third 4-fold synonymous codon sites, retain-
ing nucleotide composition and amino acid sequence of the
native mRNAs. The second randomization procedure pre-
served the amino acid sequence, but eliminated codon
usage bias at the third codon positions by randomly choosing
synonymous codons from the genetic code table with equal
probability. The resulting shuffled sequences were ?80%
identical to the corresponding native mRNA sequences at
the nucleotide level (data not shown). The third randomization
procedure preserved the amino acid sequence and created
‘codon flat’ coding sequence by randomly choosing all
synonymous codons from the genetic code table. The forth
procedure randomly shuffled all nucleotides in CDSs of
the native mRNAs, preserving only the original nucleotide
Nucleic Acids Research, 2006, Vol. 34, No. 82429
content. Nucleotides in the 50-UTRs and 30-UTRs were
shuffled in all randomization procedures retaining nucleo-
tide composition. Additionally, we employed two dinucleotide
randomization procedures described by Katz and Burge (11).
The first dinucleotide randomization preserved dinucleo-
tide frequencies, codon frequencies, codon usage and
nucleotide composition of native mRNAs. The second dinuc-
leotide randomization randomly shuffled all dinucleotides,
retaining nucleotide composition of native mRNAs.
Native mRNAs and randomly generated sequences were
computationally ‘folded’ and the predicted minimum free sec-
ondarystructure energy wascalculated,usingourimplementa-
tion of the dynamic programming algorithm described by
Zuker (39) that employs nearest neighbor parameters for
evaluation of free energy. Energy minimization was per-
formed by dynamic programming method that finds the sec-
ondary structure with the minimum free energy with sums
contributing from stacking, loop length and the like using a
new algorithm for evaluation of internal loops (40). Our
program ‘folds’ sequences up to the 28000 nucleotide long.
The sequence fold variant with the lowest secondary structure
energy was used in our analysis. P-values for randomizations
were determined by paired t-tests. To test the program per-
formance, a part of the sequence dataset was folded with the
mfold v.3.2 server (http://www.bioinfo.rpi.edu/applications/
mfold/old/rna/) and RNAalifold program from Vienna server
comparative sequence information. All programs produced
similar results. As an additional control, we folded the com-
plete set of human tRNAs from an RNA database (http://
lowelab.ucsc.edu/GtRNAdb/). Distributions of tRNA pre-
ferred base pairing are presented in Supplementary Figure 2.
RESULTS AND DISCUSSION
Sequence conservation and mRNA stability
To study the relationship between the genetic code and mRNA
secondary structure, we evaluated sequence conservation, free
Gibbs energy of secondary structure formation and nucleotide
involvement in secondary structure elements. A total of 19317
human and 20892 mouse mRNA sequences were folded
in silico. Profiles of nucleotide base pairing and secondary
structure stability in the 50-UTR, CDS and 30-UTR in the
human mRNAs, and profiles of sequence conservation in
the human and mouse mRNAs are shown on Figure 1. A
well-defined periodic pattern of conservation is observed in
the coding region. Nucleotides in the first two codon sites are
the most conserved, while those in the third positions are the
least conserved. In the CDSs, we also found a pronounced
periodic pattern of nucleotide base pairing and mRNA sec-
ondary structure stability. Frequencies of paired nucleotides at
the first, second and third codon sites differed significantly
(0.62, 0.608 and 0.631, respectively, P < 10?5). Notably, base
pairing at the third GC-rich codon sites and their contribution
to mRNA secondary structure stability are significantly higher
than contributions of the first G-rich sites or the second AU-
rich sites. The same periodic pattern of nucleotide base pairing
was observed in the mouse mRNAs (Supplementary Figure 1).
Periodic patterns of nucleotide base pairing and secondary
structure stability in the CDSs follow the triplet pattern of
nucleotide conservation created by the genetic code. Correla-
tion coefficients between sequence conservation and nucle-
otide base pairing at different codon sites were as follows:
?0.373 for site 1 (P < 0.005), ?0.516 for site 2 (P < 10?5) and
0.733 for site 3 (P < 10?11). Correlation coefficients between
sequence conservation and free energy of base pairing at
different codon sites were as follows: 0.438 for site 1
(P < 0.0003), 0.662 for site 2 (P < 10?8) and ?0.836 for
site 3 (P < 10?17). Importantly, a significant negative correla-
tion was observed between sequence conservation and the free
energy of base pairing at the third codon sites, as opposite to
the first and the second codon sites (Figure 1), indicating that
the third sites make the greatest contribution to mRNA sec-
ondary structure stability.
The periodic pattern of nucleotide base pairing becomes
even more apparent when involvement of individual nucle-
otides in the secondary structure formation is considered
(Figure 2). This pattern emerges at nt ?9 upstream from
the start codon and terminates at the stop codon. Nucleotide
G preferentially participates in the secondary structures at
Figure 1. Profilesofnucleotideinvolvementinsecondarystructures,freeenergyofsecondarystructureformationandsequenceconservationaroundthestartcodon
(A) and the stop codon (B) in human mRNAs. Positionsfrom ?30 to ?1 correspond to 50-UTRs and positionsfrom 1 to 60 correspondto CDSs (A). Positions from
Red, base paired nucleotides in 19317 human mRNAs. Green, free Gibbs energy of base pairing in 19 317 human mRNAs.
2430 Nucleic Acids Research, 2006, Vol. 34, No. 8
codon site 1, nucleotides U and A at codon site 2, G and C at
codon site 3. These preferences are in agreement with the
relative nucleotide abundance (Supplementary Table 1) and
codon position biases in the human genes (41).
Pattern of nucleotide base pairing created by the
Messenger RNA self-folding allows three types of nucleotide
base pairing in the CDS in relation to codon sites (Figure 3),
which we denote as base pairing phase 1 (sites  paired
with sites ), phase 2 (sites  paired with sites ),
and phase 3 (sites  paired with sites ). As seen from
Figure 3, phases 1, 2 and 3 are supported by nucleotide base
pairing at the first, second and third codon sites, correspond-
ingly. A pronounced bias for preferential base pairing in
phase 3 is observed in the CDSs of human mRNAs (Figure 4
and Table 1). Obvious reasons for preferential realization of
pairing phase 3 are the elevated GC content and the near equi-
valent frequencies of base pairing nucleotides (A ? U, C ? G)
at the third codon site, relative to other sites (Supplementary
Table 1). Another reason is that frequencies of trinucleotides
Figure 2. Profilesofnucleotidebasepairingaroundthestartcodon(A,CandE)andthestopcodon(B,DandF)for19317humanmRNAs(AandB),forsequences
with randomly chosen synonymous codons (C and D), and sequences with randomly shuffled nucleotides and the same nucleotide composition as native mRNAs
(E and F). Blue, guanine; red, cytosine; green, adenosine; orange, uridine.
Nucleic Acids Research, 2006, Vol. 34, No. 82431
at codon sites [2,3,1] are close to frequencies of the correspond-
ing complementary trinucleotides at the same sites, which results
in a similarly high frequencies of paired nucleotides (Supple-
mentary Table 4). Furthermore, as seen from Table 1, phase
3 allows a more dense nucleotide pairing, as compared with
in pairing phase 3, relative to other pairing phases. This type of
also be preferentially realized in transcripts from other taxa. For
example, Carlini et al. (42) found that base pairing between
synonymous sites 3–3 and 3–1 are most frequent in the droso-
philid alcohol dehydrogenase genes.
To investigate the role of the genetic code and codon usage
in the formation of mRNA secondary structure, we compared
the folding computed for native mRNAs with the folding
computed for sequences randomized with different methods.
First, we eliminated codon usage bias by randomly choosing
synonymous codons from the genetic code table. This
significantly decreased base pairing for C and G at the third
codon sites (Figure 2 and Table 1), pattern of base pairing
relative to codon sites (Figure 4 and Table 1) and markedly
reduced free Gibbs energy of secondary structure formation at
all codon sites (Table 2). Shuffling of nucleotides at the third
synonymous codon sites which retained nucleotide composi-
tion and amino acid sequence of the native mRNAs affected
nucleotide base pairing to some extend due to changes in
codon context (Table 1). Random shuffling of all nucleotides
in the CDS while preserving the original nucleotide content of
the native mRNAs completely eliminated all the secondary
structure patterns and differences in Gibbs free energy
between all codon sites (Figures 2 and 4 and Tables 1 and 2).
Since it was argued that dinucleotide content is important
when assessing the predicted free energy of RNA secondary
structure (43), we additionally employed dinucleotide ran-
domization procedures suggested by Katz and Burge (11).
Results of these randomization experiments are presented in
Table 1. Frequencies of base paired nucleotides at different codon sites in the human mRNAs and randomized sequences
Sequences Codon site 1
Codon site 2
Codon site 3
Base paired G
Base paired C
Base paired U
Base paired A
Nucleotides paired with codon site 1
Nucleotides paired with codon site 2
Nucleotides paired with codon site 3
synonymous codons; Random CC·4, sequences with randomly chosen 4-fold degenerate synonymous codons; Random DCS, dicodone shuffling that preserved
dinucleotide frequencies, encoded amino acid sequence and codon usage of native mRNAs; Random DNS, random shuffling of all dinucleotides that retained
nucleotide composition of native mRNAs. Start and stop codons were excluded from this analysis.
2432 Nucleic Acids Research, 2006, Vol. 34, No. 8
Table 1. Dinucleotide shuffling that preserved dinucleotide
and codon frequencies, nucleotide composition, and codon
usage of native mRNAs had little effect on RNA base pairing
pattern. Random shuffling of all dinucleotides while pre-
serving the original nucleotide content of the native
mRNAs completely eliminated all the secondary structure
patterns, similar to random nucleotide shuffling.
We also evaluated thermodynamic stability of native human
mRNAs and randomized sequences. As Table 3 shows, all
shuffling procedures, with the exception of dicodon shuffling,
led to statistically significant increase in calculated free energy
of mRNA secondary structure formation. Analysis of base
pairing frequencies and differences in base pairing levels
between codon sites (Table 1 and Supplementary Table 5)
demonstrates that all randomization procedures, with the
exception of dicodon shuffle, changed base pairing pattern
of native human mRNAs. Thus, both synonymous codon
usage at degenerate sites and the dinucleotide relative abund-
ances at codon positions [1,2], [2,3] and [3,1] [defined as
genome specific codon signature by Karlin and Mrazek
(41)] are important for maintaining secondary structures in
human transcripts. Specifically, base pairing at the first and
the second codon positions is determined by the structure of
the genetic code and dinucleotide frequencies, while base
pairing at the third codon positions is largely determined by
the usage of 4-fold degenerate codons. Taken together, results
of our shuffling experiments indicate that periodic pattern of
secondary structure in mRNA coding regions is largely
determined by the structure of the genetic code, with contri-
bution from synonymous codon usage bias at degenerate
codon sites. Our data suggest that synonymous sites are
under selection for a more ordered and more stable mRNA
A characteristic feature of mRNA folding in the CDS is
periodic alternation of AT and GC base pairing, which may be
important for reduction of local strong secondary structures in
the protein coding regions. This alternation is largely due to
relatively high frequencies of AT at the second codon sites and
GC at the third codon sites in the human genome (41). Our
results on relative dinucleotide abundancy and codon position
bias for the human mRNAs (Supplementary Tables 2 and 3)
are consistent with data reported for the human genes (22,41).
The observed high frequency of dinucleotide GA at sites [1,2]
reflects high proportion of glutamate and aspartate in human
proteins. The relative abundance of AG and AC at codon sites
[2,3] may be explained with high usage of specific codons for
glutamine (CAG), lysine (AAG), histidine (CAC), glutamate
(GAG) and aspartate (GAC) in the human protein coding
genes. Two dinucleotides, UA and CG, were underrepresented
at all codon sites. The deficiency of CG has been explained by
the high mutability of this dinucleotide in the nuclear DNA
(44). The deficiency of UA is considered adaptive. This dinuc-
leotide is most successible for RNase activity, and under-
representation of UA in mRNAs may reflect a requirement
for transcript stability (45).
mRNA secondary structure and transcript abundance
To study the relationship between mRNA secondary structure
frequencies and dinucleotide frequencies for different subsets
of the human mRNA dataset. The average DG of dinucleotide
interaction was significantly lower for abundant messages
(Table 2). We found no major difference in the pattern and
frequencies of base paired nucleotides between the groups of
abundant and rare transcripts, and between the groups of long
and short transcripts (data not shown). At the same time, we
observed notable differences in the dinucleotide frequencies
and codon frequencies between abundant and rare mRNAs
(Supplementary Tables 2 and 3). Frequencies of codons for
histidine, proline, cysteine and tryptophan are significantly
enhanced, and codon frequencies for lysine, asparagine,
aspartate and glutamate are significantly reduced in abundant
transcripts, relatively to rare transcripts. However, this appear
to reflect differences in the amino acid content between the
groups of proteins encoded by abundant and rare transcripts,
and have little effect on the pattern and frequencies of nuc-
leotide base pairing.
Folding and conservation in mRNA functional domains
We studied distribution of secondary structures in different
domains of human mRNAs (i.e. within 50-UTR, CDS or
30-UTR). Overall, 50-UTRs are enriched with the secondary
structures, as compared with 30-UTRs, which can be explained
by higher GC content of 50-UTRs. Frequencies of paired nuc-
leotides in the 50-UTRs and 30-UTRs were 0.64 and 0.60,
correspondingly. In the 50-UTRs, the most pronounced con-
servation (over 75% identity) is seen in the nine positions
immediately upstream of the start codon (Figure 1A). The
nucleotide conservation level in the 30 nt region upstream
of the start codon is comparable with that of the synonymous
third positions in the CDSs. Further upstream into the 50-UTR,
Figure 3. The three phases of nucleotide base pairing in the mRNA CDS.
The numbers denote codon sites.
Nucleic Acids Research, 2006, Vol. 34, No. 82433
Figure 4. Profiles of base pairing for nucleotides around the start codon (A, C and E) and the stop codon (B, D and F) with different codon sites for 19 317 human
composition as native mRNAs (E and F). Nucleotides paired with codon sites 1, 2 and 3 are shown in blue, red and green, respectively.
Table 2. Free Gibbs energy (DG) for dinucleotide interaction in secondary structure formation at different codon sites in 19317 human mRNAs, randomized
sequences and in subsets of abundant and rare transcripts
Sequences Codon sites [1,2]
P-value Codon sites [2,3]
P-value Codon sites [3,1]
corresponding EST sequences in GenBank. Selection limits were ?15 and ?108 EST sequences for rare and abundant mRNAs, correspondingly.
2434 Nucleic Acids Research, 2006, Vol. 34, No. 8
conservation level decreases to a plateau of ?50% identity
(data not shown), which is close to the neutral conservation
level in human–mouse alignments (46,47). In the 30-UTR, the
level of conservation is the lowest within the 30 nt region
immediately downstream from the stop codon and steadily
rises further in the 30-UTR (Figure 1B). The frequency of
paired nucleotides within this 30 nt region is highest, and
steadily declines in the downstream 30-UTR region (data
not shown). Similar results were obtained for the mouse
mRNAs (Supplementary Figure 1). These data are in agree-
ment with the idea that the 30 nt GC-rich region downstream
of the stop codon could be involved in the post-termination
scanning and dissociation of the ribosome from the mRNA
(35). The requirement for the 30-UTR to bind miRNAs and
proteins which are responsible for transcript stability often
results in local instability around the binding site, which pro-
vides another reason to suggest that thermodynamic stability
of mRNA molecule is optimized, and not minimized (48–50).
We evaluated levels of secondary structure formation
Figure 5 shows, levels of base pairing within same domains
are the highest, while levels of base pairing between non-
adjacent domains (50-UTR–30-UTR) are the lowest. Moreover,
levels of base pairing with a neighboring domain rapidly
decrease, as the distance to the neighboring domain increases.
As seen from the figure, these levels drop 2-fold within 60 nt
regions around the start codon (50-UTR–CDS border) and
the stop codon (CDS–30-UTR border). These observations
underscore major role of local secondary structures in
mRNA folding. Our results indicate that secondary structures
are predominantly formed within the same functional domain,
and the three functional domains preferentially fold onto
Secondary structure is avoided at translation
initiation and termination sites
The 50-UTR–CDS and CDS–30-UTR boundaries are charac-
terized with conserved secondary structures. Sequence con-
servation and nucleotide base pairing profiles around the start
codon show a degree of symmetry around the AUG (Figures 1,
2 and 4). Apparently, the first 9 nt upstream of the start codon,
and the synonymous positions of codons directly following the
AUG are subject to a stronger purifying selection than those in
the rest of the CDS and the 50-UTR. Thus, there seems to be a
distinct functional signal around the start codon that extends
into the 50-UTR and CDS beyond the Kozak sequence. mRNA
secondary structures tend to be less stable at the first two
positions of the start codon and at the position ?1 upstream
of the start codon, and more stable at the last position of the
start codon and the first position of the second codon. The first
and second nucleotides of the start codon and three upstream
nucleotides tend to be unpaired, while the third nucleotide of
the start codon and several downstream nucleotides preferen-
tially base pair with the 50-UTR. Similarly, stop codons are
less frequently base paired than the rest of mRNA sequence,
and ?10 first nucleotides of the 30-UTR preferentially pair
with the CDS (Figures 1 and 5).
These data indicate that start codons and stop codons of a
large proportion of the human and mouse mRNAs reside in
local loop structures. This is consistent with the observation
that the start codon of the CAV1 and WNT2 transcripts are
contained in evolutionarily conserved loop regions of
hairpin-like secondary structure elements (12). Such second-
ary structures around the start and stop codon may be common
in mammals and may be important for efficient initiation and
termination of translation. Functional importance of the
nucleotide context flanking the AUG was demonstrated in
several studies that traced hereditary diseases to point
mutations around the start codon [for a review see (10)].
Table 3. Stability of the human mRNA secondary structures and randomized
?1032.0 ± 4.39
?952.67 ± 3.98
?986.38 ± 4.19
?993.01 ± 4.24
?976.42 ± 4.39
?1025.6 ± 4.37
?984.54 ± 4.39
with the 50-UTRs; red, nucleotides paired with the CDSs; green, nucleotides paired with the 30-UTRs. Data for 19317 human mRNAs.
Nucleic Acids Research, 2006, Vol. 34, No. 82435
Each of these mutations was shown to cause a decrease in
In this work, we computationally folded the human and mouse
mRNA sequences on the transcriptome scale and found a
pronounced periodic pattern of mRNA secondary structure
created by the structure of the genetic code. The dinucleotide
relative abundances at codon positions [1,2] and [2,3] that
induce periodic alteration of GC and AU base pairings, and
[3,1] that define the usage of 4-fold degenerate synonymous
codons are important for the maintenance of this pattern.
Although the third GC-rich codon positions support mRNA
folding in pairing phase 3 and contribute significantly to the
transcript thermodynamic stability, periodic pattern of mRNA
folding is also well pronounced in the absence of synonymous
codon usage bias. Our results convincingly support the hypo-
thesis that the structure of the genetic code contains provisions
for the optimal secondary structure of mRNA (1,2). While all
codon sites are important for the maintenance of mRNA sec-
ondary structure, degeneracy of the code allows regulation of
mRNA secondary structure stability and periodicity.
Our results support the idea that selection is operating on
synonymous codon sites to maintain structural features of
mRNA. The distribution of base paired nucleotides in pairing
phase 3 is mostly determined by the relative abundance of C
and G nucleotides at the third codon sites and the avoidance of
CpG and UpA context (Supplementary Figure 2 and Table 4).
This is in agreement with published data on weak selection in
favor of G and C at third 4-fold degenerate sites (51,52), on an
elevated rate of evolution of synonymous sites in Drosophila
and in hominids where selection favors GC pairs (53), and
with observation that the location of synonymous mutations in
the mouse lineages is non-random with respect to mRNA
stability (15). It is known that synonymous sites are occupied
by C and G more often than intron sites, especially in potential
CpG sites despite their enhanced mutability (54). A plausible
cause is synergistic epistasis due to the involvement of syn-
onymous sites in maintaining the structure of mRNA (51,55).
Periodicity in mRNA secondary structure facilitates forma-
tion of intramolecular helices and a more compact transcript
folding which may enhance resistance of the genetic message
to degradation and modification. We found that the average
DG of dinucleotide interaction was lower for abundant mes-
sages, as compared with rare messages, although this differ-
ence was notdramatic.Thus,selection seems tooperate notfor
the most stable, but for optimally stable and ordered mRNA
secondary structure. At the same time, periodic alteration of
GC and AU base pairings prevents the formation of strong
local secondary structures that may stall ribosome transloca-
tion and impede translation. Another possibility, suggested by
Lagunez-Otero and Trifonov (4), is a potential role of local
periodicities in mRNA structure in the control of reading
frame during translation. In this work, we analyzed mRNA
secondary structureswiththelowest predicted freeenergy,and
did not take into account dynamic behavior of mRNA
molecules. In the cell, the observed periodic properties may
be even more important for highly dynamic mRNA secondary
structure. Our results demonstrate that the genetic code allows
preservation of both protein and RNA structure, and under-
score the importance of the structure of the genetic code and
degenerate codon sites for the maintenance of mRNA folding.
Supplementary Data are available at NAR Online.
We thank Eugene Koonin, John Atkins, Luda Diatchenko and
Alexey Spiridonov for critical reading of the manuscript;
Michael Roytberg for helpful discussions. This research was
supported by the Intramural Research Program of the NIH,
NLM. Funding to pay the Open Access publication charges
for this article was provided by National Center for
Biotechnology Information, National Library of Medicine,
National Institutes of Health.
Conflict of interest statement. None declared.
1. White,H.B.,III, Laux,B.E. and Dennis,D. (1972) Messenger RNA
structure: compatibility of hairpin loops with protein sequence. Science,
2. Ball,L.A. (1973) Secondary structure and coding potential of the coat
protein gene of bacteriophage MS2. Nature New Biol., 242, 44–45.
3. Fitch,W.M. (1974) The large extent of putative secondary nucleic acid
structure in random nucleotide sequences or amino acid derived
messenger-RNA. J. Mol. Evol., 3, 279–291.
4. Lagunez-Otero,J. and Trifonov,E.N. (1992) mRNA periodical
infrastructure complementary to the proof-reading site in the ribosome.
J. Biomol. Struct. Dyn., 10, 455–464.
5. Ikemura,T. (1985) Codon usage and tRNA content in unicellular and
multicellular organisms. Mol. Biol. Evol., 2, 13–34.
6. Sorensen,M.A., Kurland,C.G. and Pedersen,S. (1989) Codon usage
determines translation rate in Escherichia coli. J. Mol. Biol., 207,
7. Sharp,P.M., Averof,M., Lloyd,A.T., Matassi,G. and Peden,J.F. (1995)
DNA sequence evolution: the sounds of silence. Philos. Trans. R Soc.
Lond. B Biol. Sci., 349, 241–247.
8. Mita,K., Ichimura,S., Zama,M. and James,T.C. (1988) Specific codon
usage pattern and its implications on the secondary structure of silk
fibroin mRNA. J. Mol. Biol., 203, 917–925.
9. Zama,M. (1999) Correlation between mRNA structure of the coding
region and translational pauses. Nucleic Acids Symp. Ser., 42, 81–82.
10. Kozak,M. (2002) Pushing the limits of the scanning mechanism for
initiation of translation. Gene, 299, 1–34.
11. Katz,L. and Burge,C.B. (2003) Widespread selection for local RNA
secondary structure in coding regions of bacterial genes. Genome Res.,
12. Meyer,I.M. and Miklos,I. (2005) Statistical evidence for conserved,
local secondary structure in the coding regions of eukaryotic mRNAs
and pre-mRNAs. Nucleic Acids Res., 33, 6338–6348.
13. Seffens,W. and Digby,D. (1999) mRNAs have greater negative folding
free energies than shuffled or codon choice randomized sequences.
Nucleic Acids Res., 27, 1578–1584.
14. Jia,M., Luo,L. and Liu,C. (2004) Statistical correlation between protein
secondary structure and messenger RNA stem–loop structure.
Biopolymers, 73, 16–26.
15. Chamary,J.V. and Hurst,L.D. (2005) Evidence for selection on
synonymous mutations affecting stability of mRNA secondary structure
in mammals. Genome Biol., 6, R75.
16. Bulmer,M. (1987) Coevolution of codon usage and transfer RNA
abundance. Nature, 325, 728–730.
17. Duret,L. (2000) tRNA gene number and codon usage in the C.elegans
genome are co-adapted for optimal translation of highly expressed
genes. Trends Genet., 16, 287–289.
2436 Nucleic Acids Research, 2006, Vol. 34, No. 8
18. Akashi,H. (1994) Synonymous codon usage in Drosophila Download full-text
melanogaster: natural selection and translational accuracy. Genetics,
19. dos Reis,M., Savva,R. and Wernisch,L. (2004) Solving the riddle of
codon usage preferences: a test for translational selection. Nucleic Acids
Res., 34, 5036–5044.
20. Chamary,J.V. and Hurst,L.D. (2004) Similar rates but different modes of
sequence evolution in introns and at exonic silent sites in rodents:
evidence for selectively driven codon usage. Mol. Biol. Evol., 21,
21. Chamary,J.V., Parmley,J.L. and Hurst,L.D. (2006) Hearing silence:
non-neutral evolution at synonymous sites in mammals. Nature Rev.
Genet., 7, 98–108.
22. Comeron,J.M. (2004) Selective and mutational patterns associated with
gene expression in humans: influences on synonymous composition and
intron presence. Genetics, 167, 1293–1304.
23. Urrutia,A.O. and Hurst,L.D. (2001) Codon usage bias covaries
with expression breadth and the rate of synonymous evolution
in humans, but this is not evidence for selection. Genetics, 159,
24. Urrutia,A.O. and Hurst,L.D. (2003) The signature of selection mediated
by expression on human genes. Genome Res., 13, 2260–2264.
25. Duret,L. (2002) Evolution of synonymous codon usage in metazoans.
Curr. Opin. Genet. Dev., 12, 640–649.
26. Kanaya,S., Yamada,Y., Kinouchi,M., Kudo,Y. and Ikemura,T. (2001)
Codon usage and tRNA genes in eukaryotes: correlation of codon
usage diversity with translation efficiency and with CG-dinucleotide
usage as assessed by multivariate analysis. J. Mol. Evol., 53,
27. Lavner,Y. and Kotlar,D. (2005) Codon bias as a factor in regulating
expression via translation rate in the human genome. Gene, 345,
28. Hartl,D.L., Moriyama,E.N. and Sawyer,S.A. (1994) Selection intensity
for codon bias. Genetics, 138, 227–234.
29. Innan,H. and Stephan,W. (2001) Selection intensity against deleterious
mutations in RNA secondary structures and rate of compensatory
nucleotide substitutions. Genetics, 159, 389–399.
30. Konecny,J., Schoniger,M., Hofacker,I., Weitze,M.D. and Hofacker,G.L.
(2000) Concurrent neutral evolution of mRNA secondary structures and
encoded proteins. J. Mol. Evol., 50, 238–242.
31. Pickering,B.M. and Willis,A.E. (2005) The implications of structured
32. Duan,J. and Antezana,M.A. (2003) Mammalian mutation pressure,
synonymous codon choice, and mRNA degradation. J. Mol. Evol., 57,
33. Duan,J., Wainwright,M.S., Comeron,J.M., Saitou,N., Sanders,A.R.,
Gelernter,J. and Gejman,P.V. (2003) Synonymous mutations in the
human dopamine receptor D2 (DRD2) affect mRNA stability and
synthesis of the receptor. Hum. Mol. Genet., 12, 205–216.
34. Capon,F., Allen,M.H., Ameen,M., Burden,A.D., Tillman,D.,
Barker,J.N. and Trembath,R.C. (2004) A synonymous SNP of the
corneodesmosin gene leads to increased mRNA stability and
demonstrates association with psoriasis across diverse ethnic groups.
Hum. Mol. Genet., 13, 2361–2368.
35. Shabalina,S.A., Ogurtsov,A.Y., Rogozin,I.B., Koonin,E.V. and
Lipman,D.J. (2004) Comparative analysis of orthologous eukaryotic
mRNAs: potential hidden functional signals. Nucleic Acids Res., 32,
36. Brudno,M., Poliakov,A., Salamov,A., Cooper,G.M., Sidow,A.,
Rubin,E.M., Solovyev,V., Batzoglou,S. and Dubchak,I. (2004)
Automatedwhole-genome multiplealignmentofrat, mouse, and human.
Genome Res., 14, 685–692.
37. Frazer,K.A.,Pachter,L.,Poliakov,A.,Rubin,E.M.and Dubchak,I. (2004)
VISTA: computational tools for comparative genomics. Nucleic Acids
Res., 32, W273–W279.
38. Ogurtsov,A.Y., Roytberg,M.A., Shabalina,S.A. and Kondrashov,A.S.
(2002) OWEN: aligning long collinear regions of genomes.
Bioinformatics., 18, 1703–1704.
39. Zuker,M. (2003) Mfold web server for nucleic acid folding and
hybridization prediction. Nucleic Acids Res., 31, 3406–3415.
40. Ogurtsov,A.Y., Shabalina,S.A., Kondrashov,A.S. and Roytberg,M.A.
(2006) Analysis of internal loops within RNA secondary structure in
almost quadratic time. Bioinformatics, in press.
41. Karlin,S. and Mrazek,J. (1996) What drives codon choices in human
genes? J. Mol. Biol., 262, 459–472.
42. Carlini,D.B., Chen,Y. and Stephan,W. (2001) The relationship
between third-codon position nucleotide content, codon bias, mRNA
secondary structure and gene expression in the drosophilid alcohol
dehydrogenase genes Adh and Adhr. Genetics, 159,
43. Workman,C. and Krogh,A. (1999) No evidence that mRNAs have lower
folding free energies than random sequences with the same dinucleotide
distribution. Nucleic Acids Res., 27, 4816–4822.
44. Hanawalt,P.C. (1994) Transcription-coupled repair and human disease.
Science, 266, 1957–1958.
45. Beutler,E., Gelbart,T., Han,J.H., Koziol,J.A. and Beutler,B. (1989)
Evolution of the genome and the genetic code: selection at the
dinucleotide level by methylation and polyribonucleotide cleavage.
Proc. Natl Acad. Sci. USA, 86, 192–196.
46. Ogurtsov,A.Y., Sunyaev,S. and Kondrashov,A.S. (2004) Indel-based
evolutionary distance and mouse–human divergence. Genome Res., 14,
47. Shabalina,S.A., Ogurtsov,A.Y., Kondrashov,V.A. and Kondrashov,A.S.
(2001) Selective constraint in intergenic regions of human and mouse
genomes. Trends Genet., 17, 373–376.
48. Kryukov,G.V., Castellano,S., Novoselov,S.V., Lobanov,A.V.,
Zehtab,O., Guigo,R. and Gladyshev,V.N. (2003) Characterization of
mammalian selenoproteomes. Science, 300, 1439–1443.
49. Shabalina,S.A., Ogurtsov,A.Y., Lipman,D.J. and Kondrashov,A.S.
(2003) Patterns in interspecies similarity correlate with nucleotide
composition in mammalian 30UTRs. Nucleic Acids Res., 31,
50. Shabalina,S.A. and Spiridonov,N.A. (2004) The mammalian
transcriptome and the function of non-coding DNA sequences. Genome
Biol., 5, 105.
51. Akashi,H. (1995) Inferring weak selection from patterns of
polymorphism and divergence at ‘silent’ sites in Drosophila DNA.
Genetics, 139, 1067–1076.
52. Eyre-Walker,A. (1999) Evidence of selection on silent site base
composition in mammals: potential implications for the evolution of
isochores and junk DNA. Genetics, 152, 675–683.
53. McVean,G.A. and Vieira,J. (2001) Inferring parameters of mutation,
selection and demography from patternsof synonymous site evolution in
Drosophila. Genetics, 157, 245–257.
54. Kondrashov,F.A., Ogurtsov,A.Y. and Kondrashov,A.S. (2005)
Selection in favor of nucleotides G and C diversifies evolution rates and
levels of polymorphism at mammalian synonymous sites. J. Theor.
Biol., in press.
55. Chen,Y., Carlini,D.B., Baines,J.F., Parsch,J., Braverman,J.M., Tanda,S.
and Stephan,W. (1999) RNA secondary structure and compensatory
evolution. Genes Genet. Syst., 74, 271–286.
Nucleic Acids Research, 2006, Vol. 34, No. 8 2437