In eukaryotes, meiotic exchange of genetic information, or recombination, between homologous chromosomes is a critical step in generating genetic diversity required for adaptation. Recombination is also a crucial tool in plant improvement efforts. Local genome architecture is sculpted by the recombination process, and genome architecture, in turn, drives recombination. This interplay helps to create variability in genomic space, defining relatively stable and plastic genomic regions. This fluctuation in genomic stability is critical for balancing adaptation and stability on the phenotypic level.

Recombination has direct implications for GC patterns and vice versa. GC content refers to the percentage of guanine and cytosine bases in a DNA sequence, as opposed to adenine and thymidine bases. There have been many studies substantiating the positive correlation between recombination and GC content (Ikemura and Wada, 1991; Eyre-Walker, 1993; Fullerton et al., 2001; Galtier et al., 2001; Marais et al., 2001; Duret and Arndt, 2008; Haudry et al., 2008; Escobar et al., 2010; Muyle et al., 2011). Crossovers have been found to be correlated with high GC content in rat, mouse, human, zebrafish, bee, and maize at a broad scale (Jensen-Seaman et al., 2004; Beye et al., 2006; Gore et al., 2009; Backstrom et al., 2010; Giraut et al., 2011), while other studies detected strong correlation only at a fine scale (∼5 kb for yeast, ∼15–128 kb for human) and rather weak correlation at a broad scale (∼30 kb for yeast, ∼1 Mb for human; Gerton et al., 2000; Myers et al., 2006; Marsolier-Kergoat and Yeramian, 2009).

In plants, correlation of recombination and GC content has been demonstrated in multiple species (Haudry et al., 2008; Escobar et al., 2010; Muyle et al., 2011). A study examining three different grasses (rice, maize, and Brachypodium) revealed significant correlation of the local recombination rate with high GC content, especially in the wobble codon position (third position in the codon; Serres-Giardi et al., 2012). This was in contrast to prior studies performed with lower levels of resolution, which found only weak correlation in maize and none in rice or Brachypodium (Gore et al., 2009; Tian et al., 2009; Huo et al., 2011). However, negative correlation between crossovers and high GC content was reported for Arabidopsis (Drouaud et al., 2006) in spite of the fact that a crossover motif has been identified that has high GC content every third nucleotide (Wijnker et al., 2013).

While most studies agreed on a positive correlation between recombination and high GC content, the cause for this has been disputed (Duret and Arndt, 2008; Marsolier-Kergoat and Yeramian, 2009; Giraut et al., 2011). Possible reasons suggested for the high GC/crossover correlation include selection on codon usage, mutational bias, or GC-biased gene conversion (Eyre-Walker and Hurst, 2001; Duret and Galtier, 2009). The latter is seen as the most likely cause for GC enrichment (Figure 1), and has been suggested for diverse organisms such as yeast, mammals, and birds (Webster et al., 2006; Duret and Arndt, 2008; Mancera et al., 2008; Nabholz et al., 2011). A study by Birdsell (2002) presented compelling evidence demonstrating a highly significant positive correlation between GC in the wobble position and recombination within 6,143 ORFs analyzed in the yeast (Saccharomyces cerevisiae) genome. This study also showed a significant correlation between recombination and the mean GC content of the first and second codon and recombination, but not to the same extent as with the GC content in the third, or wobble, position (Birdsell, 2002).

In biased gene conversion, repair tracts at recombination sites that are not necessarily resolved into crossovers favor changes to GC over AT (Eyre-Walker, 1993; Birdsell, 2002). This can occur simply by using the mismatch repair machinery (Marais, 2003) or by repair of double strand breaks (DSBs) in GC-poor alleles with GC rich ones, which occurs in human but not in yeast (Duret and Arndt, 2008; Marsolier-Kergoat and Yeramian, 2009; Marsolier-Kergoat, 2011). In yeast, GC-biased gene conversion was found specifically at crossovers, not at DSBs that did not involve gene conversion and did not result in a crossover, and was shown to be due to mismatch repair rather than base excision repair or DSB repair (Lesecque et al., 2013). Even within-species variation in GC content is partly related to recombination, being correlated in mammals, birds, and yeast (Birdsell, 2002; Duret and Arndt, 2008; Nabholz et al., 2011). Notably, it has been proposed for angiosperms, that recombination and GC-biased gene conversion drives gene GC content patterns, even forming a 5′–3′ gradient along many genes (Glémin et al., 2014). However, GC-biased gene conversion might be attenuated in inbreeding species such as selfing grasses leading to no apparent correlation (Giraut et al., 2011).

Triplets of the four major nucleotide bases (guanine, cytosine, adenine, and thymine/uracil) can be combined in 64 (4³) different ways but encode for only 20 amino acids and the stop codon. Thus, there is redundancy in the genetic code, resulting in triplets (codons) that differ but are synonymous concerning their matching amino acid. GC content, mutations, and selection pressure are critical for the evolution of heterogeneity in codon usage (Sharp et al., 1988; Sharp and Matassi, 1994; Karlin and Mrazek, 1996; Sueoka and Kawanishi, 2000). By now, it has been well documented that synonymous codon usage varies significantly among genomes and, in addition, among different genes within any given genome (Grantham et al., 1980; Sharp et al., 1988; Wang and Hickey, 2007). One implication of higher GC content from silent codon positions can be a general increased expression due to higher stability of mRNA, as demonstrated in mammals (Kudla et al., 2006). Intra- and inter-genomic deviations in codon usage can be attributed to many factors. In the case of prokaryotes and unicellular organisms, it is thought to be a result of natural selection during which protein production is optimized (Gouy and Gautier, 1982; Sharp and Li, 1986; Shields et al., 1988; Sharp et al., 2005). Likewise, in some higher eukaryotes, there is also evidence that codon bias may occur due to selection for translational efficiency (Shields et al., 1988; Stenico et al., 1994). Particularly, there is a positive correlation between the complementarity of the codons in highly expressed genes and the anticodons of the most abundant tRNAs (Ikemura, 1981; Moriyama and Powell, 1997; Kanaya et al., 1999).

Numerous studies have been conducted on codon usage in plant species including the model plants Arabidopsis (Arabidopsis thaliana), rice (Oryza sativa), and a moss (Physcomitrella patens), as well as in poplar, citrus and true grass (Poaceae/Gramineae) species (Chiapello et al., 1998; Liu et al., 2003; Guo et al., 2007; Ingvarsson, 2008; Xu et al., 2013). In addition to the reported factors involved in codon usage bias, higher average GC content in monocot compared to dicot genomes plays an important role in dictating codon usage differences between the two groups of plants (Fennoy and Bailey-Serres, 1993; Carels and Bernardi, 2000). These compositional variations are illustrated by an abundance of genes with high GC levels in the third codon position in monocots (Matassi et al., 1989). The third position in the codon, also referred to as the wobble position, is less biased for the amino acid than the other two bases and accounts for most of the degeneracy of the genetic code. The wobble position functions as a marker for GC richness, and the frequency of GC nucleotides at the third position is defined as GC₃ (Chiapello et al., 1998; Tatarinova et al., 2010; Elhaik et al., 2014).

In some organisms, two classes of genes can be distinguished by their high or low GC₃ content (Carels and Bernardi, 2000; Tatarinova et al., 2010). This trait is thought to be ancestral to monocots, with some lineages losing the bimodal GC₃ distribution (Clément et al., 2014). These two GC₃ classes in monocots experience divergent evolutionary pressure and contain different functional categories of genes. High GC₃ genes within monocots were mainly categorized into functions involving electron transport or energy pathways, response to biotic and abiotic stressors and signal transduction (Carels and Bernardi, 2000; Tatarinova et al., 2010). High GC₃ genes can be turned on quickly and experience accelerated evolution (Tatarinova et al., 2010).

Kawabe and Miyashita (2003) went beyond GC₃ content and described GC content in the first, second, and third codon positions for seven plant species, including monocots and dicots, showing that GC₁ and GC₂ in addition to GC₃ were significantly higher in monocots than dicots. The differences in GC content were the largest in the third codon position, followed by the first and then the second position. Studies by Cruveiller et al. (2000, 2004) report a linear correlation between genic GC₂ and GC₃ levels, which is found in species as distant as human and bacteria.

As part of a larger project, we have explored the genomic landscape of meiosis, specifically looking at early prophase I in isolated plant meiocytes, the cells which undergo meiosis and recombination. Previous publications have focused on the functional aspects of gene expression (Chen et al., 2010; Dukowic-Schulze et al., 2014a,b) and the landscape of DSB sites and its distribution (He et al., in review). This article focuses primarily on the implications of recombination on genome evolution, as measured by GC patterns, and vice versa.

Recombination occurs during meiosis, specifically during prophase I when DSBs are formed and homologous chromosomes pair and recombine (Padmore et al., 1991; Sheehan and Pawlowski, 2012; Dukowic-Schulze et al., 2014b). While DSBs initiate recombination and are a necessary prerequisite, most DSBs will not be resolved into crossovers. Two Arabidopsis motifs associated with crossovers have been identified, including one that showed high GC content every three nucleotides (Wijnker et al., 2013). In maize, a previously identified variable motif underlying genic DSB hotspots is GC-rich and also shows high GC periodicity every three nucleotides, reminiscent of GC periodicity within the codon (He et al., in review).

In this study, we used maize as a model to better under stand the relationship between genome architecture and recombination. We examine the interplay between genome evolution (including divergent evolutionary trajectories within a single genome), GC patterns, and recombination initiation in maize. Specifically, we address whether the GC-rich, three nucleotide-periodic motif underlying DSB hotspots in maize correlates with GC₃ or other codon-driven GC patterns. In addition, we address how meiotic genes fit into the DSB and GC landscapes. Concurrently, we extend present knowledge of GC₁, GC₂, and GC₃, collectively termed GC_x, in Zea mays, by examining the relationships among them. In short, we aim to learn whether DSB-associated motifs with high GC content, particularly at every third nucleotide, could be the driving force behind bimodal GC patterns that split the maize genome into labile and stable evolutionary trajectories.

We explored the correlations of genes with instances of a motif found at meiotic DSB hotspots in maize with GC patterns as well as meiotic expression levels. GC-rich motifs with three-nucleotide periodicity underlie hotspots of DSBs that are precursors to recombination between homologous chromosomes. We looked at the genomic landscape of these motifs, including motifs in regions outside of the DSB hotspots that were identified in our dataset. While most DSB hotspots occurred near the transcription start or stop sites, GC_x content is, by definition, within the codons. Nevertheless, genes tend to be either high or low GC_x as a whole though there is a slight increase in GC content in the 5′–3′ direction (Tatarinova et al., 2010) and GC_x determination included bases between the transcription start and stop sites.

These DSB hotspot motifs tend to occur in genes that frequently have high GC₁, GC₂, or GC₃ content. High GC₁, GC₂, and GC₃ content genes are negatively correlated with each other, indicating that the extremely high GC content (≥80%) that occurs in genes with high GC_x almost exclusively derive from only one frame.

To incorporate meiotic gene expression, we sequenced RNA from meiocytes (the cells where meiosis takes place) that were just beginning the meiotic process. Specifically, the cells were in the early stages of prophase I, which itself consists of five sequential stages (leptotene, zygotene, pachytene, diplotene, and diakinesis) and is the platform of DSBs and crossovers. Early prophase I includes the formation of DSBs (leptotene), resection and single-end invasion (zygotene), and resolution of some of the DSBs into crossovers (pachytene) and therefore is highly relevant to the generation of novel genetic combinations. Notably, genes that are up-regulated in early meiosis tend to have low GC_x and are biased against containing DSB hotspot motifs. Our data thus indicates that genes up-regulated in early meiosis might be protected against DSBs. Our observation that genes that are up-regulated in meiocytes are significantly biased against genes with DSB hotspot motifs makes sense given the environment in early prophase I. Meiotic genes induced during early prophase I, by necessity, need to be genes that are not simultaneously undergoing DSBs, though we emphasize that the presence of a motif does not necessarily indicate the presence of a recombination hotspot. As a key point, our analysis suggests that DSB formation could be targeted to or avoided in specific functional classes, regulated by the periodic GC-rich motif underlying DSBs.

In general, the low GC_x gene classes in our dataset were underrepresented for the DSB hotspot motif (Table 4; Figure 7) and showed bias toward housekeeping functions. In contrast, we found genes with high GC_x that are enriched for more specialized functions, including sexual reproduction, though genes did not appear to be specific to meiotic recombination. Possibly, these latter genes are needed during other stages of development or meiosis, and not around the time when DSBs are generated, but benefit from recombination-driven evolution.

GC bimodality refers to the occurrence of two classes of genes, distinguished by their GC content. They have been reported in monocot but not in dicot plants (Clément et al., 2014). To date, there have been several reports explaining the differences in relative distribution of genes based on GC₃ between monocots and dicots. Additionally, pronounced differences were observed in GC₃ between close relatives in plant families. For example, a comparative study between A. thaliana, Raphanus sativus, Brassica rapa, and Brassica napus revealed that the GC₃ values of R. sativus, B. rapa, and B. napus are on an average 5% higher than that of A. thaliana orthologs (Villagomez and Kuleck, 2009; Tatarinova et al., 2010). Further, many plants, including grasses are prone to genome duplications, which provides redundancy that can relax selection pressure for individual genes though genome-wide some stabilization must occur after a polyploidy event. Nevertheless, the flexibility provided by redundancy that occurs after polyploidization may enable the evolution of genomic regions that have varied recombination rates and GC contents.

Degeneracy in the genetic code in the third codon position gave rise to the wobble hypothesis (Crick, 1966) which has been invoked to explain the bimodal GC₃ distribution of genes in maize and other monocots (Tatarinova et al., 2010). In the wobble position, tRNA modifications allow pairing of tRNA with multiple codons, allowing multiple codons to code for a single amino acid (reviewed in Agris et al., 2007). This degeneracy of the genetic code gives freedom, mainly in the third codon, for GC shifts to take place.

While the wobble hypothesis is often only applied to the third codon position, there is degeneracy in the genetic code in the first and second positions of the codon as well, though much less than is seen in the third position (Supplementary Table 1). Given the much larger amount of wobble in the third codon position compared to the first and second, however, it is surprising that the size of the peak with high GC₁ content genes is nearly two-thirds that of the corresponding high GC₃ peak in our dataset of DSB hotspot motif containing genes (Figure 3; Table 1). Further, the amount of wobble in the first and second positions is quite similar (Supplementary Table 1), yet the second position has only 17% the number of genes with high GC₂ content compared to GC₁ (Figure 3; Table 1). If GC bimodality patterns in maize were enabled simply by the freedom that codon wobble gives for GC shifts, than nearly all of the high GC_x genes would be in the high GC₃ category. The fact that nearly half of the high GC_x genes fall into the GC₁ and GC₂ categories clearly indicates that the wobble hypothesis, on its own, is insufficient for explaining the GC_x bimodality across all three frames.

The ability in many parts of a protein to interchange amino acids with amino acids of similar biochemical properties (Henikoff and Henikoff, 1992) might help explain the discrepancy between the numbers of genes in each high GC_x class and the codon position’s corresponding wobble. Substitution of amino acids of similar biochemical properties may allow a genome to shift GC content at all the codon positions. More work needs to be done to test this hypothesis and discover its effect on the varying sizes of the high GC_x classes across all genes but is not in the scope of this work.

The importance of GC_x bimodality is underscored by its maintenance through mutational pressure and selective restraint though the mechanism is yet unknown (Liu et al., 2012). The propensity for recombination to increase GC content (Figure 1) may be a contributing mechanism of positive reinforcement for maintaining genes in high GC_x content and DSB hotspot pools (Galtier et al., 2001). Further, the typically high evolutionary rates associated with high GC_x genes (Tatarinova et al., 2010) could be due to their association with the GC-rich DSB motifs, leading to increased rates of recombination. Genetic recombination is crucial to an organism’s ability to survive, adapt, and evolve. Genetic recombination occurs through crossing over during meiosis and, furthermore, gene conversion occurs with a frequent GC bias in the case of non-crossovers.

DSBs were assayed using ChIP-seq targeting the RAD51 gene (He et al., 2013), which binds to DSBs during zygotene, resulting in the discovery of a maize DSB hotspot motif (He et al., in review). Given the three nucleotide-based GC periodicity found in the DSB hotspot motifs and the documented GC₃ bias found in maize, it was natural to look for correlations between the two. High correlation was found between genes containing DSB hotspot motifs and GC₃ levels, both of the motif itself and the containing gene. Interestingly, similar correlations were found for GC₁ and GC₂ levels. Indeed, 70% of all genic DSB hotspots fall into high GC_x genes, a remarkable and significant proportion given that only one-quarter of genes overall fall into the high GC_x category. Even more intriguingly, genes with high GC₂ motifs make the biggest contribution, although high GC₂ genes are far less prevalent than GC₁ or GC₃ genes (Figures 5C and 6B). Further work needs to be done in order to determine whether there is a direct or secondary relationship between copies of the motif, double-strand breaks, and, by extension, recombination. Nevertheless, this data suggests intriguing hypotheses that imply a role for recombination in maintaining bimodal GC_x distributions that cannot fully be explained by the wobble hypothesis.

Sexual reproduction coupled with meiotic recombination is an important evolutionary strategy for generating genetic diversity crucial to adaptation and evolution of species. As the precursors to crossovers and non-crossovers, DSBs and their locations are critical to understanding recombination processes and biases. Here we showed that DSB hotspot motifs have a strong tendency to occur in genes with high GC_x content. These genes have extremely high GC content in one frame, have high evolutionary rates, and are biased toward SOS-functional classes, implicating support for adaptation in these genes. Further, genes up-regulated in meiocytes have a strong bias against high GC_x classes and DSB motifs, indicating the importance of these genes and their acquisition of protective strategies against DSBs. Intriguingly, wobble bases and degeneracy in the genetic code are not sufficient for explaining the high GC_x classes and their sizes across all three frames. The presence of the three nucleotide-based periodic GC-rich motifs underlying double strand hotspot motifs may provide a first glimpse at additional selection pressure driving the generation of functionally biased high GC_x classes in maize and related organisms. More work needs to be done to determine whether the motif provides recognition for the DSB machinery, thus providing a possible mechanism promoting high GC_x content.

AS, SD-S, and JM analyzed the data and wrote the manuscript. MK, KE, OO, NG, TR, and MG helped with data analysis. MW, YH, QS, and WP performed experiments to identify DSB hotspot motifs and provided data. WP and CC edited the manuscript. WP, CC, SK, JP and JM are investigators on the NSF IOS-1025881 grant of which this manuscript is a part.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.