Beyond RuBisCO: convergent molecular evolution of multiple chloroplast genes in C₄ plants

Introduction

Convergent evolution represents the independent acquisition of similar phenotypic traits in phylogenetically distant organisms. Understanding the genomic changes underlying the recurrent emergence of phenotypes is a major goal of molecular evolution. The rapidly increasing taxonomic breadth of genomic resources combined with the development of rigorous frameworks to comparatively investigate molecular changes has accelerated the pace of discovery in this area. For instance, substitutions in coding regions of conserved genes have been implicated in phenotypic changes responsible for adaptation of marine mammals to an aquatic lifestyle (Foote et al., 2015; Zhou, Seim & Gladyshev, 2015). Other examples of convergent phenotypes whose molecular underpinnings have been investigated include adaptations in snake and agamid lizard mitochondria (Castoe et al., 2009), echolocation in mammals (Parker et al., 2013; Thomas & Hahn, 2015; Zou & Zhang, 2015; Storz, 2016), and hemoglobin function in birds (Natarajan et al., 2016). Convergent traits can evolve via changes toward the same derived state (similar phenotype) from the same initial state, which is known as parallelism, or through changes of different initial states, referred to as convergence (Zhang & Kumar, 1997; Storz, 2016). For the sake of simplification, we will refer to these two processes using the general terminology ‘convergence’ and ‘convergent replacements’ throughout the manuscript, unless differently stated.

Several traits are also known to have convergently evolved in land plants (Li et al., 2018; Lu et al., 2018; Preite et al., 2019). One of the most notable examples is represented by the repeated evolution of the C₄ photosynthetic pathway in flowering plants. The C₄ pathway is a complex functional adaptation that allows for better photosynthesis efficiency under certain environmental conditions, such as dry and warm climates, high light intensity, low CO₂ concentration, and limited availability of nutrients (Knapp & Medina, 1999; Long, 1999). The C₄ pathway involves cytological, anatomical and metabolic modifications thought to have evolved multiple times independently in various lineages from the C₃ type (Kellogg, 1999; Sage, 2004; Sage, Christin & Edwards, 2011). According to phylogenetic, anatomical and biochemical evidence, the few slightly different variants of the C₄ photosynthesis type evolved more than 60 times in angiosperms (Sage, Sage & Kocacinar, 2012; Heyduk et al., 2019). In grasses (family Poaceae) alone, the C₄ pathway has evolved independently ∼20 times (Grass Phylogeny Working Group II, 2012).

Transitions from C₃ to C₄ plants resulted from genetic changes that include nonsynonymous substitutions, gene duplications and gene expression alterations (Christin et al., 2007; Christin et al. 2013a; Christin et al., 2015; Goolsby et al., 2018; Heyduk et al., 2019). It has been suggested that the evolution of the C₄ pathways proceeded throughout a series of evolutionary steps wherein the Kranz leaf anatomy typical of this pathway originated first, followed by changes in the expression patterns of key genes and finally by adaptive modifications of protein sequences (Sage, Sage & Kocacinar, 2012; Christin et al. 2013b; Williams et al., 2013). A model of the adaptive steps leading to C₄ photosynthesis showed that key biochemical components of this pathway evolved modularly along a trajectory that was likely very similar across lineages with C₃ to C₄ transitions (Heckmann et al., 2013). Overall, these scenarios suggest that enzymes involved in C₃ to C₄ transitions experienced similar selective pressures that resulted in the convergent evolution of the same amino acid replacements across C₄ lineages.

Evidence of convergent changes in proteins associated with photosynthetic processes has steadily accumulated since genomic data from multiple C₄ lineages have become available in the past couple of decades. Most of these studies have focused on the ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO), a large multimeric enzyme that catalyzes the carboxylation of ribulose-1,5-bisphosphate (RuBP), allowing plants to fix atmospheric carbon (Andersson & Backlund, 2008). RuBisCO also catalyzes oxygenation of RuBP, which leads to loss of carbon in the process of photorespiration (Andersson & Backlund, 2008; Maurino & Peterhansel, 2010). RuBisCO’s limited ability to discriminate between CO₂ and O₂ has been attributed to the much higher CO₂ to O₂ atmospheric partial pressure until ∼400 million years ago (Sage, 1999; Sage, 2004; Sage, Sage & Kocacinar, 2012).

Previous studies have revealed multiple convergent amino acid replacements in the large RuBisCO subunit in C₄ lineages, encoded by the chloroplast gene rbcL (Kapralov & Filatov, 2007; Christin et al., 2008; Kapralov et al., 2011; Kapralov, Smith & Filatov, 2012; Piot et al., 2018). Some of these convergent replacements have been associated to positive selection of the corresponding codons in C₄ monocot and eudicot lineages (Kapralov & Filatov, 2007; Christin et al., 2008; Kapralov, Smith & Filatov, 2012; Piot et al., 2018). Notably, biochemical analyses have demonstrated that some recurrent amino acid changes in the large RuBisCO subunit of C₄ plants critically alter the kinetics of RuBisCO, resulting in an accelerated rate of CO₂ fixation at the beginning of the Calvin-Benson cycle (Studer et al., 2014; Bouvier et al., 2021). Convergent amino acid changes have also been described in enzymes that are encoded by nuclear genes and play a primary role in the C ₄pathway, including the phosphoenolpyruvate carboxylase PEPC (Christin et al., 2007; Besnard et al., 2009), the NADP-malic enzymes NADP-me (Christin et al. 2009b), the phosphoenolpyruvate carboxykinase PEPCK (Christin et al. 2009a) and the small RuBisCO subunit (Kapralov et al., 2011).

Given the number of biochemical, physiological and anatomical traits that were affected in each evolutionary transition from C₃ to C₄ photosynthesis (Heyduk et al., 2019), it is likely that many genes experienced analogous selective pressures across taxa that include C₄ plants. This has been shown to be the case by Huang et al. (2017), who have developed an approach to identify potential genes involved in the transition to C₄ photosynthesis using a genome-wide scan for selection along a phylogeny of PACMAD grasses. Of the 88 genes showing signatures of positive or relaxed selection in C₄ species, several were not previously known to have a role in C₄ photosynthesis. Although this study did not focus on finding convergent replacements, it provided a comprehensive strategy and statistical testing framework to identify novel genes that have likely played a role in the evolution of C₄ grasses. It is possible that a significant fraction of these genes accumulated convergent amino acid replacements during C₃-to-C₄ transitions.

Another recent, important work has produced the first analysis of convergent replacements across multiple proteins involved in the metabolism of C₄ and crassulacean acid metabolism (CAM) among species belonging to the portullugo clade (Caryophyllales). Goolsby and colleagues (2018) compared evolutionary patterns in 19 gene families with critical roles in metabolic pathways of both C₄ and CAM plants, also known as carbon-concentration mechanisms (CCMs) genes, and in 64 non-CCM gene families. They found convergent replacements in proteins from C₄ and CAM lineages, as well as higher levels of convergent replacements in CCM vs. non-CCM gene families (Goolsby et al., 2018). Additionally, several amino acid replacements that are prevalent among C₄ and CAM taxa compared to C₃ lineages were identified in this study (Goolsby et al., 2018).

Altogether, the results of this and other studies demonstrated that convergent molecular evolution occurred across multiple genes in both C₄ and CAM groups. While significant progress has been made towards the detection of signatures of selection associated to the evolution of CCMs (Huang et al., 2017; Piot et al., 2018), a rigorous framework to assess the full extent of molecular convergence in C₃ to CCMs transitions has yet to be presented. For example, analyses of convergent evolution should include null hypotheses that assume no differences between taxa with and without convergence. In the case of CCMs evolution, a plausible null hypothesis consists in statistically equivalent numbers of convergent replacements between C₄ (or CAM) lineages and C₃ lineages.

Additionally, nonadaptive replacements should be used to normalize convergent replacements, in order to account for variation in the rates of nonsynonymous substitutions across lineages. This approach has been successfully applied in studies of molecular convergent evolution in vertebrates by assessing both convergent replacements and protein sequence changes that result in different amino acids, or divergent replacements (Castoe et al., 2009; Thomas & Hahn, 2015; Zou & Zhang, 2015). A broader definition of the latter group incorporates all replacements leading to different amino acids, regardless of their ancestral state. We refer to such changes as non-convergent replacements.

Furthermore, testing hypotheses about the extent of convergent molecular evolution remains particularly challenging for many nuclear genes, because of the prevalence of duplicated copies, particularly in plants (Christin et al., 2007; Goolsby et al., 2018). Single-copy nuclear or organelle genes allow to more easily recognize convergent changes and overcome possible confounding compensatory effects due to the presence of paralogous copies.

Given these premises, we sought to test if convergent amino acid changes occur more frequently in proteins encoded by chloroplast genes in a taxon that includes multiple well-characterized lineages of C₄ and C₃ grasses. Chloroplast proteins represent an ideal set of targets to study the role of convergent evolution in C₃ to C₄ transitions for a variety of reasons. First, most chloroplast proteins are involved in biochemical and biophysical processes that are critical to photosynthesis. For instance, out of ∼75 functionally annotated protein-coding genes in the maize chloroplast genome, 45 genes are implicated in photosynthesis-related processes, including rbcL, 17 genes coding for subunits of the photosystems I and II (PS I and PS II), 12 genes coding for subunits of the NADH dehydrogenase complex, 6 genes coding for chloroplast ATPase subunits, 4 genes coding for cytochrome b6f complex subunits, and a few more genes implicated in the assembly of other protein complexes (Maier et al., 1995). Second, nonannotated orthologous copies of chloroplast genes can be readily identified across plants through sequence homology searches, taking advantage of the thousands of complete chloroplast genome sequences currently available for green plants. Third, comparative studies of convergent evolution in C₄ photosynthesis are facilitated by detailed reconstruction of phylogenetic relationships within groups with both C₄ and C₃ lineages. Fourth, signatures of positive selection have been found in multiple chloroplast genes in taxa that contain both C₃ and C₄ plants, although only the genes rbcL and psaJ, which encodes a small subunit of the Photosystem I complex, showed evidence of adaptive changes exclusively in C₄ lineages (Christin et al., 2008; Goolsby et al., 2018; Piot et al., 2018). Finally, most chloroplast genes occur as single copy loci, as opposed to the multiple paralogs typically present for plant genes encoded in the nucleus.

In this study, we analyzed 67 chloroplast genes from 64 grass species, including 43 C₄ and 19 C₃ species belonging to the PACMAD clade, named after six of its most representative subfamilies: Panicoideae, Arundinoideae, Chloridoideae, Micrairoideae, Aristidoideae and Danthonioideae. Using published information, we placed thirteen known independent C₃ to C₄ transitions in the reconstructed phylogeny of these 64 species. We applied a series of tests based on convergent vs. non-convergent amino acid replacements and determined that convergent molecular evolution occurred at a higher rate in chloroplast genes of C₄ lineages compared to C ₃lineages, a pattern that remained largely unchanged after excluding the RbcL protein from the convergence analyses. Our findings suggest that the evolutionary trajectories of multiple chloroplast genes have been affected during the emergence of the C₄ adaptation in the PACMAD clade, a result that has significant implications for our understanding of C₄ photosynthesis evolution.

Methods

Definition and characteristics of “reference branches”

In the reconstructed PACMAD phylogeny, we identified the branches including the most recent common ancestors of C₄-only clades and C₃-only clades. We refer to these branches as “C₄ reference branches” and “C₃ reference branches”, respectively (see Figs. 1 and 2). We then compared the inferred protein sequence of each reference branch with the inferred sequence in their ancestral branch (next branch toward the root), in order to identify individual site changes that occurred along reference branches.

To assess the number of convergent and non-convergent replacements, amino acid changes were compared in all possible pairs of reference branches. Replacements in two reference branches that resulted in the same state (amino acid) at a given site were considered convergent, regardless of whether the corresponding ancestral states were the same or different (Castoe et al., 2009). After identifying convergent replacements, we separated them into parallel and convergent changes (Zhang & Kumar, 1997; Storz, 2016). Likewise, two replacements were considered non-convergent if states at the descendant orthologous sites were different, regardless of the corresponding ancestral states (Castoe et al., 2009).

The pairwise comparisons between reference branches are akin to the phylogenetically independent contrast (PIC) method developed by Felsenstein (Felsenstein, 1985). In the PIC approach, the values to compare are represented by differences between branches. The differences between two branches are independent of the differences between two other branches. Therefore, pairwise comparisons of these values are independent and can be tested using 2 × 2 contingency table tests (see also below). In our study, pairwise comparisons are independent from each other, i.e., replacements in each pair of branches are independent of replacements in each other pairs of branches. The difference from the PIC method is that we compare both differences (non-convergent replacements) and similarities (convergent replacements). A similar approach have been used in studies of convergent amino acid replacements (Castoe et al., 2009; Foote et al., 2015; Thomas & Hahn, 2015).

Reference branch lengths were extracted from the RAxML phylogeny obtained on the AIC partitioning scheme (Fig. S2). Testing was performed on the sum of pairs of branch lengths for each photosynthesis type using the R package exactRankTests (Table S2).

Results

Convergent and non-convergent amino acid replacements across chloroplast proteins

We assessed the level of molecular convergence in C₃ to C₄ transitions by quantifying convergent and non-convergent amino acid replacements across the PACMAD phylogeny by performing pairwise comparisons of reconstructed sequences in reference branches (Figs. 2 and 3, Table S3; see Methods). A total of 217 sites showed at least one convergent replacement: 104 in C₄:C₄, 120 in C₃:C₄ and 34 in C₃:C₃ pairs. A further 201 sites exhibited one or more non-convergent replacements: 96 in C₄:C₄, 121 in C₃:C₄, and 39 in C₃:C₃ pairs (Table 1). The difference in convergent/non-convergent site distributions between the three photosynthesis types was not statistically significant (P ≥ 0.05, Boschloo’s test; Table 1). The vast majority of convergent replacements shared the same ancestral state and should thus be considered parallel replacements according to widely accepted definitions of convergence (Zhang & Kumar, 1997; Storz, 2016). Only two sites, one in MatK (T205S/K205S in two C₄ reference branches) and the other in NdhF (L636I/K636I in one C₄ and three C₃ reference branches), shared replacements with different ancestral states, representing true convergent sites (Table S3).

To control for possible biases in the counting of convergent replacements due to branch length variation, we tested whether reference branch lengths in the three photosynthesis types C₄:C₄, C₃:C₄ and C₃:C₃ were different (Table S2). We found no significant difference among types (P > 0.5 for each of the three pairwise comparisons, Mann–Whitney U test). We performed the same test only on branches with convergent and non-convergent replacements and found no significant difference between categories (P > 0.5, Mann–Whitney U test; Table S2). Therefore, branch length variation between the three types is not expected to affect our results.

Among the C₄ reference branches, several individual sites showed high contrast in the number of branches involved in convergent and non-convergent replacements (Fig. 3, Tables S3 and S4). For example, seven C₄ branches (54%) shared the H18Q replacement in the product of ndhH, with no non-convergent replacements. Six, five, and four C₄ branches (46%, 38%, and 31%) showed convergent replacements at three sites in the RbcL protein (V101I, M309I, and A328S, respectively). Furthermore, six C₄ branches shared the S25G replacement in the product of ndhI and four L204F changes in the protein encoded by matK. In all these cases, there were no other convergent or non-convergent replacements in C₃:C₃ or C₃:C₄ branch comparisons, except for one H18Q change in NdhH in a C₃:C₃ branch. Two sites with convergent replacements in the proteins encoded by ndhF (L557F) and rpoC2 (H875Y) were found uniquely in C₃:C₃ pairs, and only one site in the protein Rps3 showed convergence independently in C₄:C₄ and C₃:C₃ pairs (Fig. 3).

Table 1:

Numbers of amino acid sites and genes with convergent and non-convergent replacements in reference branch comparisons.

Comparisons were made between pairs of C₄:C₄, C₃:C₃ and C₃:C₄ branches. Numbers of replacements unique to a given category (*), and the corresponding ratios Con:NC (Ratio). Differences between the C₃:C₃ and C₄:C₄ categories are not statistically significant (P ≥ 0.05, Boschloos test).

	C4:C4			C3:C4			C3:C3
	Con	NC	Ratio	Con	NC	Ratio	Con	NC	Ratio
Sites	104	96	1.08	120	121	0.99	34	39	0.87
Sites*	80	64	1.25	82	69	1.19	17	16	1.06
Genes	24	23	1.04	26	32	0.81	13	17	0.76
Genes*	24	20	1.2	25	29	0.86	9	10	0.9

DOI: 10.7717/peerj.12791/table-1

Notes:

Con

convergent

NC

non-convergent

We then searched for convergent replacements that occurred along more than two C₄ branches at sites that remained otherwise conserved in C₃ and C₄ lineages, arguing that such changes could result from selective pressure rather than drift. We identified twelve C₄-specific convergent sites in proteins from 7 genes: matK, ndhF, ndhG, ndhI, rbcL, rpoC1 and rpoC2 (Table S4). Five of these sites were found in RbcL, whereas two sites were identified in NdhI. We also observed two convergent sites NdhF and one in RpoC2 that were uniquely found in three C₃ branches.

Molecular convergence in individual chloroplast proteins

Convergent and non-convergent amino acid replacements were detected in the products of 45 chloroplast genes, thirteen of which had at least one site with four or more replacements (Fig. 4, Tables 1 and S3). Twenty-four genes had convergent changes in C₄:C₄, 26 in C₃:C₄, and 13 in C₃:C₃ types of pairs (Table 1). Although the convergent/ non-convergent replacement ratio was higher in C₄:C₄ pairs than C₃:C₄ and C₃:C₃ pairs, the differences between the three photosynthesis types were not statistically significant (P ≥ 0.05, Boschloo’s test; Table 1). The lack of replacements was the single most common state for chloroplast proteins across photosynthesis types; however, in C₄:C₄ there were more genes with a higher number of convergent vs. non-convergent replacements (Fig. 4 and Table S5).

Overall, 26 proteins showed a higher number of convergent vs. non-convergent sites, of which 16, 13 and 10 were found in C₄:C₄, C₃:C₄ and C₃:C₃ pairs, respectively (Fig. 5 and Table S5). We found statistically significant differences in the number of convergent vs. non-convergent replacements between C₄:C₄ and C₃:C₄ pairs, but not C₃:C₃ pairs, in the products of the genes rbcL, rpoC1 and rpoC2 (P < 0.05, Boschloo’s test; Table S5). In RbcL and RpoC1, C₄:C₄ pairs shared much higher proportion of convergent vs. non-convergent replacements, whereas the opposite was true in RpoC2. RpoC1 was also the only protein showing more convergent than non-convergent replacements in C₄:C₄ pairs compared to C₃:C₃ and C₃:C₄ pairs. In C₄:C₄ pairs, RpoC1 shared 4 convergent and 1 non-convergent replacement, compared to 1 and 2 in C₃:C₃ pairs and 1 and 5 in C₃:C₄ pairs, respectively. Additionally, the proteins NdhG, NdhI, PsaI, RpoA, Rps4 and Rps11 exhibited convergent replacements only in C₄:C₄ pairs (Table S5). When considering the number of affected sites rather than the number of replacements, no genes showed a significantly different pattern between photosynthesis types (P ≥ 0.05, Boschloo’s test; Table S5).

The proteins encoded by matK, rpoC2 and ndhF shared much higher numbers of both convergent and non-convergent replacements than other chloroplast proteins across all photosynthesis type comparisons (Table S5). Both matK and ndhF are known to be rapidly evolving and have been consistently used in low taxonomic level phylogenetic studies in flowering plants (Patterson & Givnish, 2002; Barthet & Hilu, 2008). The gene rpoC2 has also been recently described as a useful phylogenetic marker in angiosperms (Walker et al., 2019).

Molecular convergence across reference branches

The comparison of reference branch pairs with convergent and non-convergent replacements revealed remarkable differences between photosynthesis types. Overall, C₄:C₄ pairs of reference branches showed a distribution skewed toward more convergent and non-convergent replacements than the two other categories (Fig. 6). There were significantly fewer pairs of C₄:C₄ reference branches with no replacements and with no convergent replacements than C₃:C₄ and C₃:C₃ pairs (P < 0.05, Boschloo’s test; Table 2). Conversely, significantly more C₄:C₄ pairs shared more convergent than non-convergent replacements, and at least two convergent changes compared to C₃:C₄ and C₃:C₃ pairs (P < 0.05, Boschloo’s test; Table 2). No significant difference was observed between pairs of C₃:C₄ and pairs of C₃:C₃. We found identical patterns when the same analyses were performed after excluding all replacements in the RbcL protein, except for the lack of a significant difference between C₄:C₄ and C₃:C₃ in the proportion of pairs with non-convergent replacements and pairs with more convergent than non-convergent changes (Table S6).

Table 2:

Number of reference branches with convergent and non-convergent replacements.

Comparisons were made between pairs of C₄:C₄, C₃:C₃ and C₃:C₄ branches. Proportions of pairs of reference branches over all branches by category are shown in parenthesis. The total number of pairs of reference branches are 78, 36 and 117 for C₄:C₄, C₃:C₃ and C₃:C₄ comparisons, respectively. All comparisons between C₄:C₄ pairs and both C₃:C₃ and C₃:C₄ pairs were statistically significantly different (P < 0.05, Boschloos test). No comparison between C₃:C₃ and C₃:C₄pairs was statistically significant (P ≥ 0.05, Boschloos test).

	C4:C4	C3:C4	C3:C3
No replacements	6 (.08)	30 (.26)	12 (.33)
No Con	12 (.15)	48 (.41)	16 (.44)
w/Con	66 (.85)	69 (.59)	20 (.56)
w/NC	63 (.81)	67 (.57)	18 (.50)
Con > NC	40 (.51)	36 (.31)	10 (.28)
Con > 1	49 (.63)	39 (.33)	8 (.22)

DOI: 10.7717/peerj.12791/table-2

Notes:

Con

convergent

NC

non-convergent

Con > NC

pairs of branches with more convergent than non-convergent replacements

Con > 1

pairs of branches with more than one convergent replacement

Discussion

The recurrent emergence of carbon-concentration mechanisms (CCMs) across multiple angiosperm clades in the past 35 million years represents one of the most striking examples of convergent evolution of a complex phenotypic trait (Sage, Christin & Edwards, 2011; Heyduk et al., 2019). Several investigations have shown that the phenotypic parallelism across C₄ lineages is to some extent mirrored by convergent changes in the sequence of proteins with key metabolic roles in the biochemistry of C₄ photosynthesis, both in monocots and eudicots (Christin et al., 2007; Besnard et al., 2009; Christin et al. 2009a; Christin et al. 2009b; Kapralov et al., 2011; Goolsby et al., 2018). Furthermore, biochemical analyses have determined that some of these changes reflect adaptive shifts, as in the case of the increased availability of CO₂ at the RuBisCO site (Studer et al., 2014). Substantial changes in several RuBisCO kinetic traits associated to C₃ to C₄ transitions have recently been described (Bouvier et al., 2021). Further evidence of changes in the selective pressure associated to the C₃ to C₄ transitions have emerged from the detection of several positively selected sites in multiple genes associated with photosynthetic processes (Christin et al., 2008; Studer et al., 2014; Goolsby et al., 2018; Piot et al., 2018). These and other discoveries have paved the way to a more nuanced understanding of the molecular basis of phenotypic convergence in CCM plants and may accelerate the development of crop varieties with augmented resistance to high temperature and low water availability.

For these aims to be fully realized, a robust framework to assess the extent and phenotypic impact of convergent molecular changes is necessary. Along the lines of strategies applied in vertebrates research (Castoe et al., 2009; Foote et al., 2015; Thomas & Hahn, 2015; Zou & Zhang, 2015), we presented here the results of a novel methodological approach to the study of molecular convergence in C₄ grasses. We investigated patterns of convergent and non-convergent amino acid changes in nearly 70 chloroplast proteins across multiple C₄ and C₃ lineages in the PACMAD clade, with the goal of testing a specific hypothesis: is the evolution of chloroplast proteins showing stronger signatures of convergent amino acid replacements in C₄ lineages compared to C₃ lineages? This analysis also allowed us to establish if proteins other than enzymes involved in the CCM biochemistry underwent parallel amino acid changes in C₄ lineages. Our reasoning is that many proteins expressed in the chloroplast could have experienced similar selective pressure across multiple C₃ to C₄ transitions and might have accumulated convergence replacements as a result. In agreement with our expectation, dozens of nuclear genes sharing signatures of positive or relaxed selection and likely associated with the evolution of C₄ PACMAD grasses have been recently described, albeit these analyses relied on a limited number of species (Huang et al., 2017).

We based our analysis on the identification of amino acid replacements shared by pairs of reference C₄ branches, defined here as branches corresponding to C₃ to C₄ transitions in the PACMAD phylogeny. We compared these changes to those identified in reference C₃ branches, namely all C₃ lineages that include only C₃ species (Figs. 1 and 2), and to changes found between reference C₃ and C₄ branches. For each of the three possible pairs of photosynthesis types, C₄:C₄, C₃:C₄ and C₃:C₃, we determined the number of amino acid sites, genes and pairs of reference branches with convergent replacements.

We detected signatures of convergent evolution in all types of datasets. First, we identified many individual replacements that emerged repeatedly and uniquely in C₄ reference branches, particularly in the proteins RbcL, NdhH, NdhI and MatK. We also observed C₃-specific convergent replacements in NdhF and RpoC2, and a case of multiple C₄ and C₃ convergent changes in Rps3. Additionally, we identified 7 chloroplast genes with one or more C₄-specific convergent sites and 3 chloroplast genes with at least one C₃-specific convergent site. Second, we found evidence of significantly higher rates of convergent replacements in C₄ lineages in both RbcL and RpoC1, and several convergent replacements that occurred exclusively in C₄:C₄ pairs in proteins encoded by ndhG, ndhI, psaI, rpoA, rps4 and rps11. These genes are involved in a variety of biological processes in the chloroplast, from the cyclic electron transport in (ndhG and ndhI) and the stabilization of (psaI) the photosystem I, to transcription (rpoA and rpoC1), translation (rps4 and rps11) and CO₂ fixation (rbcL). Third, we identified statistically significant differences in pairs of C₄ branches with convergent replacements (Table 2). Crucially, we observed more pairs with higher convergent than non-convergent replacements in C₄:C₄ compared to both C₃:C₃ and C₃:C_4, even after removing replacements identified in the RuBisCO large subunit, RbcL.

Altogether, these findings suggest that multiple biochemical processes occurring in the chloroplast might have experienced recurrent adaptive changes associated with the emergence of C₄ photosynthesis. Notably, some of these proteins are not directly involved in the light-dependent or light-independent reactions of the photosynthesis, implying that processes such as regulation of gene expression and protein synthesis in the chloroplast are also experiencing significant selective pressures during the transition from C₃ to C₄ plants. These results should motivate further studies to determine the prevalence of convergent amino acid replacements in transitions to CCMs among the thousands of proteins encoded by nuclear genes but expressed in the chloroplast (Jarvis & Lopez-Juez, 2013). Although such analyses are currently hindered by the limited number of sequenced nuclear genomes in taxa with multiple C₃ and C₄ lineages, including the PACMAD clade, genome-wide investigations of convergent replacements will be possible in the near future given the current pace of DNA sequencing in plants.

A further important conclusion drawn from these results is that convergent replacements are not uncommon between C₃:C₃ and C₃:C₄ lineages. This is possibly due to some environmental factors affecting the evolution of chloroplast genes that are shared across grass lineages regardless of their photosynthesis type.

The analysis of individual convergent replacements in the RuBisCO large subunit both confirmed previous findings (Christin et al., 2008; Studer et al., 2014; Piot et al., 2018) and highlighted novel potentially adaptive changes among PACMAD species. Importantly, these novel convergent replacements are known to evolve under positive selection in non-PACMAD seed plants (Kapralov & Filatov, 2007; Sen et al., 2011). This underscores the potential of our approach to identify novel changes with functional significance in the transition to CCMs in grasses, as opposed to standard statistical tests of positive selection. Alternatively, some RbcL sites could experience convergence across a variety of seed plants because of selective pressure other than those associated with C₃ to C₄ transitions.

Overall, our results are robust to several possible confounding factors. First, we analyzed branches that are strongly supported in our phylogeny reconstruction. The phylogenetic tree built using the 67 chloroplast genes is well supported, with the exception of three branches with fairly low bootstrap support. However, all three branches are short and have minimal impact upon our conclusions regarding C₄ evolution (Fig. 1 and Figs. S1–S3). Moreover, the tree is largely consistent with a comprehensive recent study of 250 grasses based on complete plastome data (Saarela et al., 2018). Second, by focusing only on reference branches and ignoring amino acid replacements that may have occurred after the divergence of species within a given C₄ clade, our strategy provided a conservative estimate of the number of convergent changes that could have occurred during the evolution of PACMAD grasses. Third, we eliminated genes with possible paralogous copies, which could have introduced false positive replacements.

We recognize some potential caveats in our approach. By relying on a relatively small sample of PACMAD species, our statistical power to detect signatures of convergent evolution was limited. Increasing the number of reference C₄ and C₃ lineages should provide a broader representation of convergent replacements in C₄ clades. Furthermore, we applied a strict definition of convergence that ignores changes to amino acids with similar chemical properties. We think that a conservative approach was necessary given that amino acids with similar chemical properties might have a very different functional effect on protein activity given their size and tridimensional interactions with nearby residues. Third, we assumed that all the observed convergent replacements were the result of convergent phenotypic changes, which fall under the general category of homoplasy (Avise & Robinson, 2008). However, some of these replacements could instead represent hemiplasy, or character state changes due to introgression between different C₄ lineages, incomplete lineage sorting (ILS) of reference alleles or horizontal gene transfer (Avise & Robinson, 2008). Recombination between chloroplast genomes, which is required for introgression to occur, has been documented but appears to be rare (Carbonell-Caballero et al., 2015; Greiner, Sobanski & Bock, 2015; Sancho et al., 2018). Introgression or horizontal gene transfer between congeneric species has been associated to the acquisition of part of the C₄ biochemical pathway in the PACMAD genus Alloteropsis (Christin et al., 2012; Olofsson et al., 2016). However, these transfers were limited to a few nuclear genes. Moreover, only a very few cases of horizontal transfer between chloroplast genomes have been reported in plants (Stegemann et al., 2012). Therefore, the contribution of hemiplasy to the observed pattern of convergent replacements in C₄ lineages is likely to be minimal. Finally, we treated C₄ species regardless of their photosynthesis subtype (NAPD-ME, NAD-ME and PEPCK), which is known to vary among PACMAD subfamilies (Taylor et al., 2010). We argue that our results are conservative with regard to this aspect because convergent replacements should be expected to occur more often between C₄ groups sharing the same photosynthesis subtype.

Conclusions

In this study, we showed that molecular convergent evolution in the form of recurrent amino acid replacements affected multiple chloroplast proteins in C₄ lineages of the PACMAD clade of grasses. This finding significantly broadened the number of genes known to have evolved convergently in C₄ species. We observed for the first time that genes not directly involved in photosynthesis-related processes experienced convergent changes, suggesting that future efforts should rely whenever possible on genome-wide analyses of amino acid changes rather than focus primarily on candidate key metabolic genes, similarly to previous investigations on gene expression patterns in C₄ and CAM plants. Our methodological approach based on the comparison of convergent and non-convergent replacements among photosynthesis types underscores the importance of a more rigorous hypothesis-based testing of convergent evolution signatures in C₄ plant evolution. Our results should inform more nuanced approaches to introduce CCM-like processes in C₃ crops.

Supplemental Information