Resubmission to Molecular Biology and Evolution as a Research Article.

Transcription

1 Selection on synonymous sites for increased accessibility around mirna binding sites in plants Wanjun Gu 1,*, Xiaofei Wang 1, Chuanying Zhai 1, Xueying Xie 1 and Tong Zhou 2,3,* Resubmission to Molecular Biology and Evolution as a Research Article. 1Key Laboratory of Child Development and Learning Science of Ministry of Education of China, Southeast University, Nanjing, Jiangsu , China 2Institute for Personalized Respiratory Medicine, The University of Illinois at Chicago, Chicago, IL 60612, USA 3Section of Pulmonary, Critical Care, Sleep & Allergy, Department of Medicine, The University of Illinois at Chicago, Chicago, IL 60612, USA *Corresponding authors: Wanjun Gu, Tel: , wanjungu@gmail.com; and Tong Zhou, Tel: , tongzhou@uic.edu Keywords: synonymous selection, microrna function, site accessibility Running Head: Synonymous selection for mirna binding

2 Abstract Synonymous codons are widely selected for various biological mechanisms in both prokaryotes and eukaryotes. Recent evidences have suggested microrna (mirna) function may affect synonymous codon choices near mirna target sites. To better understand this, we perform genome- wide analysis on synonymous codon usage around microrna (mirna) target sites in four plant genomes. We observe a general trend of increased site accessibility around mirna target sites in plants. GC- poor codons are preferred in the flank region of mirna target sites. Within genome analyses show significant variation among mirna targets in species. GC content of the target gene can partly explain the variation of site accessibility among mirna targets. MiRNA targets in GC- rich genes show stronger selection signals than those in GC- poor genes. Gene s codon usage bias and the conservation level of mirna and its target also have some effects on site accessibility. But, the expression level of mirna or its target and the mechanism of mirna activity do not contribute to site accessibility differences among mirna targets. We suggest synonymous codons near mirna targets are selected for efficient mirna binding and proper mirna function. Our results present a new dimension of natural selection on synonymous codons near mirna target sites in plants, which will have important implications of coding sequence evolution.

3 Introduction Mutations at synonymous codon sites are normally assumed to be evolutionarily neutral because these mutations don t change its encoded protein sequences (Yang and Nielsen, 2000). Under this assumption, functional sites in protein coding sequences are identified by comparing non- synonymous substitution rate with synonymous substitution rate (Yang and Bielawski, 2000). However, increasing evidences have suggested synonymous codons are under many selection pressures in both prokaryotes and eukaryotes (Chamary, et al., 2006; Duret, 2002). A well- known selection constraint in bacteria, yeast, worm and plant is translational accuracy and efficiency (Akashi and Eyre- Walker, 1998; Chamary and Hurst, 2005; Cohen and Mayfield, 1997; Duret, 2002; Ikemura, 1985; Morton and Wright, 2007; Zhou, et al., 2009). In mammals, synonymous codons are selected for stable DNA secondary structures (Vinogradov, 2003), proper nucleosome positioning (Warnecke, et al., 2008), optimized mrna stability (Chamary and Hurst, 2005; Stoletzki, 2008), efficient mrna splicing (Parmley, et al., 2006; Warnecke and Hurst, 2007), proper protein co- translational folding (Komar, et al., 1999; Thanaraj and Argos, 1996) and efficient translation initiation (Gu, et al., 2010; Kudla, et al., 2009; Tuller, et al., 2010). The inclusion of synonymous codon selection has made more accurate inferences in evolutionary analysis of protein coding sequences (Nielsen, et al., 2007; Yang and Nielsen, 2008; Zhou, et al., 2010). Recently, many studies have suggested mirna function may affect synonymous codon choices as well (Brest, et al., 2011; Hurst, 2006; Hurst, 2011; Tay, et al., 2008). MiRNAs are small RNAs as important post- transcriptional

4 regulators in animals, plants and viruses (Axtell, et al., 2011). MiRNAs can bind complementarily to their target mrna sequences, resulting in translational repression, RNA degradation or RNA cleavage (Bartel, 2009). Tay et al. (Tay, et al., 2008) experimentally validated that mirna interactions with Nanog, Oct4, and Sox2 coding regions could modulate embryonic stem cell differentiation in mouse. Notably, they observed some silent mutations in mirna target sites could abolish mirna activity and delay the induced phenotype (Tay, et al., 2008). Brest et al. (Brest, et al., 2011) concluded that human Crohn s disease was associated with a synonymous substitution located in mir- 196 target region in IRGM. Another computational analysis revealed a low synonymous substitution rate at several mirna target sites in human protein coding sequences (Hurst, 2006). All these observations suggested synonymous codons in mirna target sites (especially nucleotides pairing with mirna seed region) were selected since their base paring interactions with mirna sequences are crucial in mirna function (Bartel, 2009). Other than nucleotides directly targeted by mirnas, sequences near mirna target sites are also important for mirna function (Long, et al., 2007). In the process of mirna activity, binding of gene transcripts with RNA- induced silencing complexes (RISCs) is an important step (Brodersen and Voinnet, 2009; Fabian, et al., 2010). Several structural and sequence features around mirna target sites, especially mirna target site accessibility (Long, et al., 2007), were found to be related to mirna binding (Gu, et al., 2009; Hausser, et al., 2009; Lin and Ganem, 2011; Sun, et al., 2010). Hence, synonymous codons around mirna

5 target sites are probably selected to facilitate mirna binding by making mirna target region more accessible. Here, we tried to investigate this possible selection on synonymous codons near mirna target sites at the genome scale. Although many mirna target sites were exemplified in protein coding sequences in animals (Forman and Coller, 2010; Forman, et al., 2008; Rigoutsos, 2009; Schnall- Levin, et al., 2011; Schnall- Levin, et al., 2010; Tay, et al., 2008), it is still a big challenge in identifying all mirna target regions in protein coding sequences of animal genomes (Miranda, et al., 2006). Unlike animals, most mirna targets are located in protein coding regions in plants (Jones- Rhoades and Bartel, 2004). Since plant mirnas are nearly perfect complementary to their target sites (Dai, et al., 2011), computational prediction of mirna targets is much more efficient in plants. Hence, we performed our analysis on four plant genomes to test our hypothesis. We used computational methods to calculate site accessibility of each mirna target region. To estimate the significance of synonymous codon selection for site accessibility, we permutated mrna sequences near mirna target sites and assessed the deviation of site accessibility in wild type mirna target region from random expectation. We tested whether there were selection signals on synonymous sites around mirna target sites for increased site accessibility in plants. We also addressed whether there were variations of site accessibility among mirna targets in a genome. If there were variations among mirna targets within species, we further looked into several factors that may affect site accessibility. Materials and Methods

6 Genome data We performed our analysis on four plants, including Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). We downloaded protein coding sequences (CDS) of all four genomes from Ensembl ( release 9, April 2011) (Kersey, et al., 2010) using BioMart (Haider, et al., 2009). For genes with multiple transcripts, we chose the longest transcript. MiRNA data We downloaded known mirnas in these species from mirbase ( release 16, September 2010) (Kozomara and Griffiths- Jones, 2011). We then used pstarget server ( (Dai and Zhao, 2011) to identify all putative mirna targets in the CDSs. All mirna target regions in CDSs are available in Supplementary Data 1. MiRNAs are dynamically gained and lost along plant evolution (Axtell and Bowman, 2008). Some mirnas are well conserved in all land plants, while some are specific to a single species or lineage. To investigate the effects of mirna conservation on site accessibility, we classified mirnas in Arabidopsis thaliana into three groups - Arabidopsis thaliana unique mirnas, Arabidopsis thaliana and Arabidopsis lyrata specific mirnas, conserved mirnas. We obtained the conservation category of each mirna from Fahlgren et al. (Fahlgren, et al., 2010). Similarly, mirna target regions experience dynamic gain or loss in evolution (Axtell and Bowman, 2008). We classified mirna target regions in

7 Arabidopsis thaliana into two groups - Arabidopsis thaliana specific targets, conserved targets in Arabidopsis thaliana and Arabidopsis lyrata. We also parsed the conservation category of each mirna target region from Fahlgren et al. (Fahlgren, et al., 2010). Expression data To evaluate if gene expression or mirna expression have effects on site accessibility of mirna target region, we downloaded Massively Parallel Signature Sequencing (MPSS) data for mrnas and mirnas in Arabidopsis thaliana from plant MPSS database ( (Meyers, et al., 2004). We used two different measures in estimating the expression level from MPSS data. We summed the number of sequenced short tags in each mrna or mirna over all tissues ( Exp sum ) as one measure of expression level. Other than Exp sum, we counted the number of tissues with sequenced tags in each mrna or mirna ( Exp tissue ) as another measure. To eliminate the false positives of low expression levels caused by sequencing errors, we only considered mrnas or mirnas with 4 or more tags in a tissue as valid MPSS data (Wright, et al., 2004). Site accessibility Site accessibility represents the difficulty in opening mirna target region for its binding with RISCs (Axtell, et al., 2011). In our analysis, a mirna target region contained 21 nucleotides directly targeted by mirna, 17 flank upstream and 10 flank downstream nucleotides. We chose this window of 48 nucleotides as mirna target region since site accessibility of this region has the best correlation

8 to measured expression of mirna target transcript (Kertesz, et al., 2007). We used ΔG open to quantify site accessibility of each mirna target. ΔG open is the difference between the free energy of the ensemble of all secondary structures of the target region and the free energy of all target region structures in which the target sites are unpaired region (Kertesz, et al., 2007). Other than ΔG open, we calculated the free energy of local secondary structure of the target region ( ΔG local ) to measure RNA stability near mirna target regions. We used RNAddG4 program in PITA package (Kertesz, et al., 2007) to calculate ΔG and ΔG open local with default parameters. In RNAddG4, RNAfold (Hofacker, 2003) was used to calculate the free energy of RNA secondary structures (Kertesz, et al., 2007). As suggested in previous studies (Kertesz, et al., 2007; Lu and Mathews, 2008; Richter, et al., 2010; Ying, et al., 2011; Zhao, et al., 2005), we used a segment of mrna sequences, rather than the full- length mrnas, when calculating ΔG open and ΔG local. The input mrna segment included 48 nucleotides in mirna target region, and additional 140 flank upstream and downstream nucleotides. The use of this mrna segment was based on the facts that the probability of base pairing when nucleotides are separated by more than 140 nucleotides was low (data not shown) and it could substantially reduce computational complexity. mrna randomization and Z-score calculation If selection acts on synonymous codons near mirna target sites to facilitate mirna binding, site accessibility of this region in real mrnas should be statistically different from that of randomized sequences. Therefore, separately for each gene, we randomly shuffled synonymous codons among sites with

9 identical amino acid, which keeps the same peptide sequence, global codon usage pattern and GC composition. Since mirna target sites are important for mirna recognition and function, codons targeted by mirnas are not shuffled in the randomization. We generated 1,000 such resampled sequences for each gene. For real mrna sequence and each permutated sequence, we calculated ΔG open, ΔG local and GC content in sliding windows of 48 nucleotides with a step of 16 nucleotides around mirna target region. To determine the deviation of the real sequence from the randomized ones, we calculated the Z- score of ΔG open (Z ΔGopen ) for each target by: Z ΔGopen = n i =1 (ΔG open ) N (ΔG open ) P ((ΔG open ) Pi (ΔG open ) P ) 2 n 1 (1) Here, (ΔG open ) N is the site accessibility for the naturally occurring target site under consideration. (ΔG open ) Pi is the site accessibility for the target site in permuted sequence, and (ΔG open ) P is the mean of (ΔG open ) Pi over all permuted sequences. The variable represents the total number of permuted sequences. Here, we have at 1,000. Similarly, we also evaluated the difference of free energy of local mrna structure and GC content around mirna targets between natural mrna sequences and permutated sequences. We calculated Z- score of free energy of local mrna secondary structure (Z ΔGlocal ) and Z- score of local GC content (Z GC ) as formulas below.

10 Z ΔGlocal = n i =1 (ΔG local ) N (ΔG local ) P ((ΔG local ) Pi (ΔG local ) P ) 2 n 1 (2) Z GC = n i =1 GC N GC P (GC Pi GC P ) 2 n 1 (3) (ΔG local ) N, and (ΔG local ) P, GC N, GC Pi and GC P are analogous to (ΔG open ) N, (ΔG open ) Pi and (ΔG open ) P, but refer to the free energy of local RNA secondary structure or GC content rather than the free energy required to open target region. Results Synonymous codons are selected for increased site accessibility in plants We first calculated ΔG open in sliding windows of 48 nucleotides in length, moving upstream and downstream from the real mirna target region in steps of 16 nucleotides (for a total of 13 windows). To compare the ΔG open with randomized pattern, we shuffled mirna target gene for 1,000 times while keeping gene s codon usage bias, amino acid sequence and mirna target sites. We then calculated ΔG open in sliding windows in 1,000 permutated genes. Next, we calculated Z- score of site accessibility, Z ΔGopen, by comparing ΔG open of the real mirna targets to the distribution of permutated mirna targets. Finally, we calculated the mean of Z ΔGopen across all mirna targets in a genome. Z ΔGopen measures the deviation of the site accessibility in real mrna from randomized

11 sequences. A negative Z ΔGopen means that less energy is required in opening mirna target region in the real mrna. We observed a general trend of Z ΔGopen changes in sliding windows along the mrna sequence for all species (Figure 1 and Supplementary Data 2). Z ΔGopen of mirna target region was significantly less than zero in Oryza sativa (t- test, P = ), Populus trichocarpa (t- test, P = 0.008) and Zea mays (t- test, P = 0.006). But, the decrease of ΔG open in real mirna target from that in permutated mirna targets was only marginally significant in Arabidopsis thaliana (t- test, P = 0.096). When sliding windows moved downstream or upstream from mirna target region, Z ΔGopen increased and many of them had positive values (Figure 1). To investigate the effect of window size on site accessibility, we repeated our analysis using sliding windows with different number of nucleotides in mirna target region. We tested different numbers of flank upstream and downstream nucleotides with 21 nucleotides bound to mirnas. Results from these analyses were comparable with those we observed above in all four species (Table S1). Similarly, we performed the sliding window analysis of ΔG local in all four genomes (Figure S1 and Supplementary Data 3). ΔG local is the free energy of local secondary structure of mirna target region. A positive Z ΔGlocal means selection for loose RNA secondary structure near mirna target sites. We observed significantly reduced mrna stability near mirna targets in Oryza sativa (t- test, P = ) and Zea mays (t- test, P = 0.003). When sliding windows moved upstream or downstream from mirna target region, Z ΔGlocal decreased and some of them had negative values (Figure S1). But, mean ΔG local value of all mirna

12 targets was positive in Arabidopsis thaliana (t- test, P = 0.46) and Populus trichocarpa (t- test, P = 0.07) (Figure S1). GC-poor codons are preferred near mirna target sites in plants We observed increased site accessibility near mirna target sites in four genomes. But, how synonymous codons were selected to have increased site accessibility in mirna target region? We calculated Z GC in mirna target region for all four species. Z GC is the deviation of local GC content in real mrna sequences from that of random expectation. Negative Z GC means GC- poor codons are preferred in mirna target region. We found strong correlations between Z ΔGopen and Z GC in all four species (Figure 2). Therefore, GC- poor codons were likely to be selected in mirna target region for increased site accessibility to facilitate mirna binding. Within genome variation of Z ΔGopen in mirna target region We used the mean value of Z ΔGopen across all mirna targets in a genome to detect selection signals at the genome level. Substantial variation of Z ΔGopen value was observed among mirna targets within species (Figure 1). Therefore, we next investigated several potential factors that may affect site accessibility near mirna target sites. We first considered gene GC content. We found significant negative correlations between gene GC content and Z ΔGopen near mirna target sites in Arabidopsis thaliana, Oryza sativa and Zea mays (Figure 3). We further extracted mirna targets in genes with the highest 5% and the lowest 5% GC content. We

13 compared mean Z ΔGopen of these two groups of mirna targets. mirna target sites in genes with higher GC content tended to have lower Z ΔGopen in all species (Figure S2). Z ΔGopen of target sites in genes with the lowest 5% GC content did not show significant deviations from zero (t- test, P > 0.05 for all species; Figure S2). Notably, even though we only found marginally significant signal of selection near mirna target sites when all mirna target regions were considered in Arabidopsis thaliana (Figure 1), mirna targets in GC- rich genes did show significant deviations of reduced site accessibility (t- test: P = 0.002; Figure S2). Next, we considered gene codon usage bias. We used ENC (Effective Number of Codons) to measure gene s codon usage bias (Wright, 1990). The higher a gene s codon usage bias, the lower is the gene s ENC. We compared Z ΔGopen of target sites in genes with the top 5% ENC to genes with the bottom 5% ENC in all species. We observed significant differences between Z ΔGopen mean of these two groups of mirna targets in Populus trichocarpa (t- test, P = 0.028) and Zea mays (t- test, P = ), but not in Arabidopsis thaliana (t- test, P = 0.054) and Oryza sativa (t- test, P = 0.08) (Figure S3). In Populus trichocarpa, Oryza sativa and Zea mays, mirna targets in genes with stronger codon usage bias (lower ENC values) showed smaller Z ΔGopen values (Figure S3). But, the difference was inverted in Arabidopsis thaliana (Figure S3). We further considered the conservation level of mirna and its target. First, we compared Z ΔGopen mean of target sites with different mirna conservation level in Arabidopsis thaliana (Figure 4). Target sites with mirnas conserved in Arabidopsis lineage had smaller Z ΔGopen than those with mirnas unique to

14 Arabidopsis thaliana (t- test: P = 0.04). But, comparison of site accessibility of target sites with Arabidopsis lineage- specific mirnas and those with deep conserved plant mirnas did not show significant difference (t- test: P = 0.25). Second, we compared Z ΔGopen between targets with different conservation level. We also observed some difference (Figure 4). Targets conserved in Arabidopsis lineage also had smaller Z ΔGopen than those unique to Arabidopsis thaliana (t- test, P = 0.016). We finally tested whether expression level of mirna or its target mrna could account for Z ΔGopen variation within species. There was no obvious Z ΔGopen difference between mirna targets with the highest 5% expression level and the lowest 5% expression level of mirna (t- test, P = 0.52 for Exp sum and P = 0.69 for Exp tissue ) or its target genes (t- test, P = 0.14 for Exp sum and P = 0.10 for Exp tissue ) in Arabidopsis thaliana (Figure S4 and Figure S5). Other than expression of mirna or its target, we also considered the mechanisms that mirna used to regulate the expression of target mrna. We compared Z ΔGopen of mirna targets with different mirna activity mechanisms in all four species. We did not observe any Z ΔGopen difference (t- test, ath: P = 0.08, osa: P = 0.14, ptc: P = 0.37, zma: P = 0.67) between targets regulated by RNA degradation and those regulated by translational repression in all four genomes (Figure S6). Discussion The degeneracy of the genetic code allows protein coding sequences to carry abundant additional information other than amino acid sequences (Bollenbach,

15 et al., 2007; Itzkovitz, et al., 2010). Synonymous codons have been related to DNA structure (Vinogradov, 2003), nucleosome positioning (Warnecke, et al., 2008), RNA stability (Chamary and Hurst, 2005; Stoletzki, 2008), RNA splicing (Parmley, et al., 2006; Warnecke and Hurst, 2007), translation accuracy and efficiency (Chamary, et al., 2006; Duret, 2002) and translation initiation (Gu, et al., 2010; Kudla, et al., 2009; Tuller, et al., 2010). Recently, synonymous mutations in mirna target sites have been observed to be able to change mirna activities (Tay, et al., 2008) and cause disease (Brest, et al., 2011). We have surveyed mirna binding information in protein coding sequences using four plant genomes in this study. We have found synonymous codons are generally selected near mirna target sites in four genomes (Figure 1). Site accessibility in mirna target region tends to be higher than randomly expected given target gene s amino acid sequence, codon usage bias and mirna target sites. Our results are comparable with those reported by some experimental studies (Gu, et al., 2009; Hausser, et al., 2009; Kertesz, et al., 2007). They have observed decreased site accessibility near mirna target region can substantially abolish mirna activity (Gu, et al., 2009; Hausser, et al., 2009; Kertesz, et al., 2007). We have found the signal of negative Z ΔGopen is uniquely located near mirna target region along the mrna sequence (Figure 1). This suggests the increased site accessibility near mirna target region is related to mirna function. When sliding window moves upstream or downstream from mirna target region, Z ΔGopen increases gradually and shifts to a positive value (Figure 1). The positive Z ΔGopen value in other sliding windows means mrna segments elsewhere in the gene tend to be locally structured. This is consistent with previous observations

16 that synonymous codons are selected for stable mrna secondary structure in many organisms (Chamary and Hurst, 2005; Gu, et al., 2010; Seffens and Digby, 1999). We have performed our analysis using a window of 48 nucleotides (21 nucleotides targeted by mirna, 17 flank upstream and 10 flank downstream nucleotides) as the mirna target region. This is comparable with the region proposed in Kertesz et al. (Kertesz, et al., 2007) and Hausser et al. (Hausser, et al., 2009). Kertesz et al. have experimentally altered nucleotides near mirna target sites and measured mirna activity for different mirnas in human, mouse, fly and worm (Kertesz, et al., 2007). They have found a maximum correlation between site accessibility and mirna activity when the region has 17 flank upstream and 13 flank downstream nucleotides. Hausser et al. have shown site accessibility of a target region with 12 flank upstream and 12 flank downstream nucleotides can explain most of the activity variations among mirna targets in human (Hausser, et al., 2009). Our results have observed site accessibility of a similar region in protein coding sequences is evolutionary selected in plant genomes, which suggest its important biological role in mirna action. In addition to site accessibility, RNA secondary structure near mirna target sites is also important in mirna activity (Long, et al., 2007). We have found reduced RNA stability near mirna target sites in some species (Figure S1). GC- poor codons are locally preferred around mirna target sites for loose secondary structure (Figure 2). Site accessibility and RNA secondary structure are related to each other, since site accessibility is based on RNA secondary structure in calculation. In our results, we have observed significant correlations between

17 Z ΔGopen and Z ΔG local in all genomes (Figure S7). But, selection signals in mirna target region for reduced site accessibility (Figure 1) are more obvious than that for reduced local RNA stability (Figure S1). This is comparable with the observations in two previous studies (Hausser, et al., 2009; Kertesz, et al., 2007) that site accessibility is more related to mirna activity than RNA secondary structure. We have observed substantial variation in the extent to which increased site accessibility is selected among mirna targets within species. We have found several factors, such as gene s GC content, conservation level of mirna and its target, and gene s codon usage bias, can explain the variation among mirna targets. First, the increase of site accessibility is greater for mirna targets in genes with higher GC content. For genes with lower GC content, site accessibility near mirna targets can be high enough for mirna binding. Therefore, there is no extra need to selectively use GC- poor codons near mirna target sites to facilitate mirna binding in genes with lower GC content. Second, we have found mirna target regions with conserved mirnas or mirna targets have significant increases of site accessibility. However, recently gained mirnas or mirna targets in Arabidopsis thaliana don t show any selection signal in mirna target region. This is consistent with the observation of reduced purifying selection with young mirna genes and its targets in Arabidopsis thaliana (Axtell and Bowman, 2008; Cuperus, et al., 2011; Fahlgren, et al., 2010). Third, we have found gene s codon usage bias has some effects on site accessibility near mirna target sites. In principle, genes with higher codon usage bias (lower ENC ) should have lower Z ΔGopen. In our results, we have observed Z ΔGopen is significantly

18 lower in genes with the highest 5% codon usage bias in Oryza sativa, Populus trichocarpa and Zea mays (Figure S3). Other than gene s GC content, codon usage bias and the conservation level of mirna and its target, we have considered some other factors, including mirna expression, mirna target expression and mechanism of mirna activity. We haven t fount any effect on site accessibility near mirna target sites for these factors. In fact, while mrna expression level is the input quantity of mirnas for mirna activity, the target gene expression level is the final output of mirna activity. Mechanisms of mirna action are the underlying biological pathways by which mrnas are repressed by mirnas. Even though these factors are relevant to the whole process of mirna- mediated gene regulation, they may have little role in the process of mirna binding. Since site accessibility is an important feature in mirna binding, it is reasonable to observe no effects on Z ΔGopen variations when these factors are considered. Although site accessibility is generally selected in mirna target region in several plant genomes, the deviation of Z ΔGopen from zero is marginally significant in Arabidopsis thaliana when all mirna targets are considered together (Figure 1). However, it is notable some subsets of mirna targets in Arabidopsis thaliana do show significantly decreased site accessibility in mirna target region. For example, mirna target regions in genes with the highest 5% GC content (Figure S2), those targeted by Arabidopsis lineage specific mirnas (Figure 4), and those are conserved in Arabidopsis lineage (Figure 4), have Z ΔGopen significantly less than zero in Arabidopsis thaliana. This suggests site accessibility in mirna target region is selected in Arabidopsis thaliana, although the selection signal is weak. The reason why we have observed site accessibility differences among species is

19 largely unknown. Previous studies have suggested synonymous codons are primarily selected for efficient and accurate mrna translation in Arabidopsis thaliana (Morton and Wright, 2007). Compared to translation efficiency and accuracy, site accessibility is local selection constraint acting on synonymous codons in mirna target region. Therefore, selection signal of increased site accessibility near mirna target sites is probably screwed up by strong signals induced by translational selection in Arabidopsis thaliana. We have presented a new dimension of selection constraint on synonymous codons for mirna function in plants. Given increasing evidences of mirna targets in animal genomes (Forman and Coller, 2010; Forman, et al., 2008; Rigoutsos, 2009; Schnall- Levin, et al., 2011; Schnall- Levin, et al., 2010; Tay, et al., 2008), this selection constraint may exist in animal genomes as well. It is worthy of taking a systematic survey of site accessibility in mirna target region when more mirna targets in protein coding sequences are identified in animal genomes. Supplementary Material Supplementary Data 1. mirna targets in protein coding sequences of four plants: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Supplementary Data 2. Z ΔGopen of each sliding window for each mirna target in four plant genomes: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma).

20 Supplementary Data 3. Z ΔGlocal of each sliding window for each mirna target in four plant genomes: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Table S1. The mean Z ΔGopen of mirna target regions with various number of flank upstream and downstream nucleotides in four plants: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Figure S1. Mean and standard error of Z ΔGlocal of each sliding window in four plant genomes: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Each point represents a single mirna target in the genome. Figure S2. Comparisonon mean and standard error of Z ΔGopen of mirna target sites between genes with the highest 5% and lowest 5% GC content in four plant genomes: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Figure S3. Comparison on mean and standard error of Z ΔGopen of mirna targets between genes with top 5% and bottom 5% codon usage bias in four plant genomes: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Figure S4. Comparison on mean and standard error of Z ΔGopen of mirna targets between mirnas with top 5% and bottom 5% mirna expression level in Arabidopsis thaliana. We considered both expressed tag counts ( Exp sum ) and number of expressed tissues ( Exp tissue ) as mirna expression level. Figure S5. Comparison on mean and standard error of Z ΔGopen of mirna targets between target genes with top 5% and bottom 5% expression level in Arabidopsis thaliana. We considered both expressed tag counts ( Exp sum ) and number of expressed tissues ( Exp tissue ) as expression level of mirna target gene.

21 Figure S6. Comparison on mean and standard error of Z ΔGopen of mirna targets between mirnas with different activity mechanisms in four plant genomes: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Figure S7. The correlation between Z ΔGopen and Z ΔG local of each mirna target in all four species: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Each point represents a mirna target in the genome. Solid lines show lowess smoothed data. Acknowledgement We thank two anonymous reviewers for helpful comments. This work was supported by grants from the National Basic Research Program of China [2012CB to WG and XX], the National High Technology Research and Development Program of China (863 Project)[2012AA to WG], National Natural Science Foundation of China [ and to WG, and to XX], and a SRF for ROCS, SEM, China. Literatures Cited Akashi, H. and Eyre- Walker, A. (1998) Translational selection and molecular evolution, Curr Opin Genet Dev, 8, Axtell, M.J. and Bowman, J.L. (2008) Evolution of plant micrornas and their targets, Trends Plant Sci, 13, Axtell, M.J., Westholm, J.O. and Lai, E.C. (2011) Vive la difference: biogenesis and evolution of micrornas in plants and animals, Genome Biol, 12, 221. Bartel, D.P. (2009) MicroRNAs: target recognition and regulatory functions, Cell, 136, Bollenbach, T., Vetsigian, K. and Kishony, R. (2007) Evolution and multilevel optimization of the genetic code, Genome Res, 17, Brest, P., et al. (2011) A synonymous variant in IRGM alters a binding site for mir- 196 and causes deregulation of IRGM- dependent xenophagy in Crohn's disease, Nat Genet, 43, Brodersen, P. and Voinnet, O. (2009) Revisiting the principles of microrna target recognition and mode of action, Nat Rev Mol Cell Biol, 10, Chamary, J.V. and Hurst, L.D. (2005) Biased codon usage near intron- exon junctions: selection on splicing enhancers, splice- site recognition or something else?, Trends Genet, 21,

22 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. Chamary, J.V., Parmley, J.L. and Hurst, L.D. (2006) Hearing silence: non- neutral evolution at synonymous sites in mammals, Nat Rev Genet, 7, Cohen, A. and Mayfield, S.P. (1997) Translational regulation of gene expression in plants, Curr Opin Biotechnol, 8, Cuperus, J.T., Fahlgren, N. and Carrington, J.C. (2011) Evolution and functional diversification of MIRNA genes, Plant Cell, 23, Dai, X. and Zhao, P.X. (2011) psrnatarget: a plant small RNA target analysis server, Nucleic Acids Res, 39, W Dai, X., Zhuang, Z. and Zhao, P.X. (2011) Computational analysis of mirna targets in plants: current status and challenges, Brief Bioinform, 12, Duret, L. (2002) Evolution of synonymous codon usage in metazoans, Curr Opin Genet Dev, 12, Fabian, M.R., Sonenberg, N. and Filipowicz, W. (2010) Regulation of mrna translation and stability by micrornas, Annu Rev Biochem, 79, Fahlgren, N., et al. (2010) MicroRNA gene evolution in Arabidopsis lyrata and Arabidopsis thaliana, Plant Cell, 22, Forman, J.J. and Coller, H.A. (2010) The code within the code: micrornas target coding regions, Cell Cycle, 9, Forman, J.J., Legesse- Miller, A. and Coller, H.A. (2008) A search for conserved sequences in coding regions reveals that the let- 7 microrna targets Dicer within its coding sequence, Proc Natl Acad Sci U S A, 105, Gu, S., et al. (2009) Biological basis for restriction of microrna targets to the 3' untranslated region in mammalian mrnas, Nat Struct Mol Biol, 16, Gu, W., Zhou, T. and Wilke, C.O. (2010) A universal trend of reduced mrna stability near the translation- initiation site in prokaryotes and eukaryotes, PLoS Comput Biol, 6, e Haider, S., et al. (2009) BioMart Central Portal- - unified access to biological data, Nucleic Acids Res, 37, W Hausser, J., et al. (2009) Relative contribution of sequence and structure features to the mrna binding of Argonaute/EIF2C- mirna complexes and the degradation of mirna targets, Genome Res, 19, Hofacker, I.L. (2003) Vienna RNA secondary structure server, Nucleic Acids Res, 31, Hurst, L.D. (2006) Preliminary assessment of the impact of microrna- mediated regulation on coding sequence evolution in mammals, J Mol Evol, 63, Hurst, L.D. (2011) Molecular genetics: The sound of silence, Nature, 471, Ikemura, T. (1985) Codon usage and trna content in unicellular and multicellular organisms, Mol Biol Evol, 2, Itzkovitz, S., Hodis, E. and Segal, E. (2010) Overlapping codes within protein- coding sequences, Genome Res, 20, Jones- Rhoades, M.W. and Bartel, D.P. (2004) Computational identification of plant micrornas and their targets, including a stress- induced mirna, Mol Cell, 14, Kersey, P.J., et al. (2010) Ensembl Genomes: extending Ensembl across the taxonomic space, Nucleic Acids Res, 38, D

23 Kertesz, M., et al. (2007) The role of site accessibility in microrna target recognition, Nat Genet, 39, Komar, A.A., Lesnik, T. and Reiss, C. (1999) Synonymous codon substitutions affect ribosome traffic and protein folding during in vitro translation, FEBS Lett, 462, Kozomara, A. and Griffiths- Jones, S. (2011) mirbase: integrating microrna annotation and deep- sequencing data, Nucleic Acids Res, 39, D Kudla, G., et al. (2009) Coding- sequence determinants of gene expression in Escherichia coli, Science, 324, Lin, H.R. and Ganem, D. (2011) Viral microrna target allows insight into the role of translation in governing microrna target accessibility, Proc Natl Acad Sci U S A, 108, Long, D., et al. (2007) Potent effect of target structure on microrna function, Nat Struct Mol Biol, 14, Lu, Z.J. and Mathews, D.H. (2008) Efficient sirna selection using hybridization thermodynamics, Nucleic Acids Res, 36, Meyers, B.C., et al. (2004) Arabidopsis MPSS. An online resource for quantitative expression analysis, Plant Physiol, 135, Miranda, K.C., et al. (2006) A pattern- based method for the identification of MicroRNA binding sites and their corresponding heteroduplexes, Cell, 126, Morton, B.R. and Wright, S.I. (2007) Selective constraints on codon usage of nuclear genes from Arabidopsis thaliana, Mol Biol Evol, 24, Nielsen, R., et al. (2007) Maximum likelihood estimation of ancestral codon usage bias parameters in Drosophila, Mol Biol Evol, 24, Parmley, J.L., Chamary, J.V. and Hurst, L.D. (2006) Evidence for purifying selection against synonymous mutations in mammalian exonic splicing enhancers, Mol Biol Evol, 23, Richter, A.S., et al. (2010) Seed- based INTARNA prediction combined with GFP- reporter system identifies mrna targets of the small RNA Yfr1, Bioinformatics, 26, 1-5. Rigoutsos, I. (2009) New tricks for animal micrornas: targeting of amino acid coding regions at conserved and nonconserved sites, Cancer Res, 69, Schnall- Levin, M., et al. (2011) Unusually effective microrna targeting within repeat- rich coding regions of mammalian mrnas, Genome Res. Schnall- Levin, M., et al. (2010) Conserved microrna targeting in Drosophila is as widespread in coding regions as in 3'UTRs, Proc Natl Acad Sci U S A, 107, Seffens, W. and Digby, D. (1999) mrnas have greater negative folding free energies than shuffled or codon choice randomized sequences, Nucleic Acids Res, 27, Stoletzki, N. (2008) Conflicting selection pressures on synonymous codon use in yeast suggest selection on mrna secondary structures, BMC Evol Biol, 8, 224. Sun, G., Li, H. and Rossi, J.J. (2010) Sequence context outside the target region influences the effectiveness of mir- 223 target sites in the RhoB 3'UTR, Nucleic Acids Res, 38, Tay, Y., et al. (2008) MicroRNAs to Nanog, Oct4 and Sox2 coding regions modulate embryonic stem cell differentiation, Nature, 455,

24 Thanaraj, T.A. and Argos, P. (1996) Ribosome- mediated translational pause and protein domain organization, Protein Sci, 5, Tuller, T., et al. (2010) Translation efficiency is determined by both codon bias and folding energy, Proc Natl Acad Sci U S A, 107, Vinogradov, A.E. (2003) DNA helix: the importance of being GC- rich, Nucleic Acids Res, 31, Warnecke, T., Batada, N.N. and Hurst, L.D. (2008) The impact of the nucleosome code on protein- coding sequence evolution in yeast, PLoS Genet, 4, e Warnecke, T. and Hurst, L.D. (2007) Evidence for a trade- off between translational efficiency and splicing regulation in determining synonymous codon usage in Drosophila melanogaster, Mol Biol Evol, 24, Wright, F. (1990) The 'effective number of codons' used in a gene, Gene, 87, Wright, S.I., et al. (2004) Effects of gene expression on molecular evolution in Arabidopsis thaliana and Arabidopsis lyrata, Mol Biol Evol, 21, Yang, Z. and Bielawski, J.P. (2000) Statistical methods for detecting molecular adaptation, Trends Ecol Evol, 15, Yang, Z. and Nielsen, R. (2000) Estimating synonymous and nonsynonymous substitution rates under realistic evolutionary models, Mol Biol Evol, 17, Yang, Z. and Nielsen, R. (2008) Mutation- selection models of codon substitution and their use to estimate selective strengths on codon usage, Mol Biol Evol, 25, Ying, X., et al. (2011) starpicker: a method for efficient prediction of bacterial srna targets based on a two- step model for hybridization, PLoS One, 6, e Zhao, Y., Samal, E. and Srivastava, D. (2005) Serum response factor regulates a muscle- specific microrna that targets Hand2 during cardiogenesis, Nature, 436, Zhou, T., Gu, W. and Wilke, C.O. (2010) Detecting positive and purifying selection at synonymous sites in yeast and worm, Mol Biol Evol, 27, Zhou, T., Weems, M. and Wilke, C.O. (2009) Translationally optimal codons associate with structurally sensitive sites in proteins, Mol Biol Evol, 26,

25 Figure Legends Figure 1. Mean and standard error of Z ΔGopen of each sliding window in four plant genomes: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Figure 2. The correlation between Z ΔGopen and Z GC of each mirna target in all four species: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Each point represents a mirna target in the genome. Solid lines show lowess smoothed data. Figure 3. The correlation between Z ΔGopen of each mirna target and GC content of its target gene in all four species: Arabidopsis thaliana (ath), Oryza sativa (osa), Populus trichocarpa (ptr) and Zea mays (zma). Each point represents a mirna target in the genome. Solid lines show lowess smoothed data. Figure 4. Mean and standard error of Z ΔGopen of mirna targets with different conservation levels of mirna or its target in Arabidopsis thaliana. The conservation of mirna has three levels: Arabidopsis thaliana specific (ath- specific), Arabidopsis lineage specific (ath- aly specific) and deep conserved (universal). Instead, the conservation of mirna targets had two levels: Arabidopsis thaliana specific (ath- specific) and Arabidopsis lineage specific (ath- aly specific).

26 Mean Z Gopen Mean Z Gopen Mean Z Gopen Mean Z Gopen ath osa ptc zma Offset Figure 1. Mean and standard error of of each sliding window in four plant genomes: Arabidopsis thaliana (ath), Oryza sa2va (osa), Populus trichocarpa (ptr) and Zea mays (zma).

27 ath osa ptc zma Z G open R = 0.2 P < Z G open R = 0.38 P = Z G open R = 0.34 P < Z G open R = 0.34 P < Z GC Z GC Z GC Z GC Figure 2. The correla;on between and of each mirna target in all four species: Arabidopsis thaliana (ath), Oryza sa2va (osa), Populus trichocarpa (ptr) and Zea mays (zma). Each point represents a mirna target in the genome. Solid lines show lowess smoothed data.

28 Z Gopen ath GC content R = 0.07 P = Z Gopen osa GC content R = 0.15 P = Z Gopen ptc GC content R = 0.02 P = Z Gopen zma GC content R = 0.21 P < Figure 3. The correla;on between of each mirna target and GC content of its target gene in all four species: Arabidopsis thaliana (ath), Oryza sa2va (osa), Populus trichocarpa (ptr) and Zea mays (zma). Each point represents a mirna target in the genome. Solid lines show lowess smoothed data.

29 Mean Z G open mirna ath specific ath aly specific universal Target Figure 4. Mean and standard error of of mirna targets with different conserva;on levels of mirna or its target in Arabidopsis thaliana. The conserva;on of mirna has three levels: Arabidopsis thaliana specific (ath- specific), Arabidopsis lineage specific (ath- aly specific) and deep conserved (universal). Instead, the conserva;on of mirna targets had two levels: Arabidopsis thaliana specific (ath- specific) and Arabidopsis lineage specific (ath- aly specific).