Positive Selection in Rapidly Evolving Plastid–Nuclear Enzyme Complexes

Abstract

Rates of sequence evolution in plastid genomes are generally low, but numerous angiosperm lineages exhibit accelerated evolutionary rates in similar subsets of plastid genes. These genes include clpP1 and accD, which encode components of the caseinolytic protease (CLP) and acetyl-coA carboxylase (ACCase) complexes, respectively. Whether these extreme and repeated accelerations in rates of plastid genome evolution result from adaptive change in proteins (i.e., positive selection) or simply a loss of functional constraint (i.e., relaxed purifying selection) is a source of ongoing controversy. To address this, we have taken advantage of the multiple independent accelerations that have occurred within the genus Silene (Caryophyllaceae) by examining phylogenetic and population genetic variation in the nuclear genes that encode subunits of the CLP and ACCase complexes. We found that, in species with accelerated plastid genome evolution, the nuclear-encoded subunits in the CLP and ACCase complexes are also evolving rapidly, especially those involved in direct physical interactions with plastid-encoded proteins. A massive excess of nonsynonymous substitutions between species relative to levels of intraspecific polymorphism indicated a history of strong positive selection (particularly in CLP genes). Interestingly, however, some species are likely undergoing loss of the native (heteromeric) plastid ACCase and putative functional replacement by a duplicated cytosolic (homomeric) ACCase. Overall, the patterns of molecular evolution in these plastid–nuclear complexes are unusual for anciently conserved enzymes. They instead resemble cases of antagonistic coevolution between pathogens and host immune genes. We discuss a possible role of plastid–nuclear conflict as a novel cause of accelerated evolution.

PLASTIDS carry reduced genomes that reflect an evolutionary history of extensive gene loss and transfer to the nucleus since their ancient endosymbiotic origin roughly 1 billion years ago (Timmis et al. 2004; Keeling 2010; Gray and Archibald 2012). Many of the proteins encoded by genes that have been transferred to the nuclear genome are trafficked back into the plastid (Gould et al. 2008), where they interact closely with proteins encoded by genes remaining in the plastid genome. These interacting proteins are key not only to photosynthesis, but also to transcription, translation, and critical nonphotosynthetic metabolic functions of the plastid. The interactions between these gene products create the opportunity for coevolution between plastid and nuclear genomes. Thus, studying the nuclear genes that contribute to plastid complexes is a valuable tool for understanding the processes underlying plastid genome evolution and cytonuclear coevolution.

Within angiosperms, most plastid genomes are highly conserved in sequence and structure (Jansen et al. 2007; Wicke et al. 2011), but multiple independent lineages have experienced accelerated rates of aa substitution in similar subsets of nonphotosynthetic genes (Jansen et al. 2007; Erixon and Oxelman 2008; Greiner et al. 2008b; Guisinger et al. 2008, 2010, 2011; Straub et al. 2011; Sloan et al. 2012a, 2014a; Barnard-Kubow et al. 2014; Weng et al. 2014; Dugas et al. 2015; Williams et al. 2015; Zhang et al. 2016). Several mechanisms have been hypothesized to explain these repeated accelerations including positive selection, reduced effective population size (N_e), altered DNA repair, changes in gene expression, and pseudogenization following gene transfer to the nucleus (see above citations). Distinguishing among these hypotheses has proved challenging, and the ultimate cause or causes of the extreme differences in rates of molecular evolution among genes within plastid genomes remain unclear.

In many cases of extreme plastid genome evolution, accelerations have disproportionately affected nonsynonymous sites, resulting in elevated ratios of nonsynonymous to synonymous substitution rates (d_N/d_S) (e.g., Erixon and Oxelman 2008; Guisinger et al. 2008; Barnard-Kubow et al. 2014; Sloan et al. 2014a), which indicates that changes in selection are likely involved. In addition, recent studies showed correlated increases in d_N/d_S between nuclear- and plastid-encoded subunits in ribosomal (Sloan et al. 2014b; Weng et al. 2016) and RNA polymerase complexes (Zhang et al. 2015), providing further evidence for changes in selection pressures. However, these studies could not confidently distinguish between two alternative explanations for increased d_N/d_S, positive selection and relaxed purifying selection, which can be difficult to disentangle based on sequence divergence data alone. Because these selection pressures can have very different effects on population genetic variation, analyses that combine data on intraspecific polymorphism and interspecific divergence (McDonald and Kreitman 1991) can detect positive selection even in cases where it is not readily identifiable based only on d_N/d_S (Rausher et al. 2008). However, most studies of accelerated plastid genome evolution and plastid–nuclear coevolution have not included the necessary intraspecific polymorphism data to perform these analyses.

In contrast to recent analyses of plastid genetic machinery (i.e., ribosomal and RNA polymerase genes; Sloan et al. 2014b; Zhang et al. 2015; Weng et al. 2016), the potential for molecular coevolution involving nuclear-encoded subunits in other plastid complexes remains largely unexplored. Two such complexes are the caseinolytic protease (CLP), which is an ATP-dependent protease required for proper plastid function (Nishimura and van Wijk 2015), and the heteromeric acetyl-coA carboxylase (ACCase), which is involved in fatty acid biosynthesis (Sasaki and Nagano 2004; Salie and Thelen 2016). The CLP complex and ACCase each contain a single plastid-encoded subunit (ClpP1 and AccD, respectively) and multiple subunits of nuclear origin. In most angiosperms, the sequences of the clpP1 and accD genes are generally conserved, but they are among the plastid-encoded genes that exhibit elevated rates of sequence evolution in multiple independent lineages. The clpP1 gene, in particular, exhibits recent and dramatically increased rates of nonsynonymous substitutions and indels (e.g., Erixon and Oxelman 2008; Sloan et al. 2014a).

In addition to phylogenetic and population genetic analyses, examining patterns of aa substitutions relative to protein structure can help distinguish between relaxed and positive selection. In the model angiosperm Arabidopsis thaliana, the CLP complex is made up of two stacked heptameric rings, comprising nine different types of paralogous and structurally related subunits that are derived from the single subunit found in the ancestral homotetradecameric form of this enzyme (Peltier et al. 2004; Yu and Houry 2007; Olinares et al. 2011). The P-ring is formed entirely of the nuclear-encoded subunits CLPP3, 4, 5, and 6 in a 1:2:3:1 stoichiometric ratio, and the R-ring contains the plastid-encoded subunit ClpP1 and the nuclear-encoded subunits CLPR1, 2, 3, and 4 in a 3:1:1:1:1 ratio. The CLPP subunits all contain a conserved catalytic Ser-His-Asp triad, which is lacking from the CLPR subunits (Peltier et al. 2004), meaning that ClpP1 is the only catalytic subunit within the R-ring. Other nuclear-encoded subunits such as CLPC, CLPD, CLPF, CLPS, CLPT1, and CLPT2 are physically associated with the core CLP complex and act as adapters, chaperones, and accessory proteins, helping to regulate the proteolytic activity of CLP (Peltier et al. 2004; Nishimura and van Wijk 2015; Nishimura et al. 2015).

Most flowering plants contain two different types of ACCase enzymes: a eukaryotic-like homomeric multidomain ACCase in the cytosol and a bacterial-like heteromeric ACCase within the plastids. The heteromeric ACCase consists of proteins encoded by four different genes (ACCA, B, C, and D; Sasaki and Nagano 2004; Salie and Thelen 2016). ACCC is a biotin carboxylase. In an ATP-dependent reaction, it carboxylates a biotin molecule attached to ACCB (a biotin carboxyl carrier protein), with bicarbonate serving as the donor of the carboxyl group (White et al. 2005). The nuclear-encoded ACCA and plastid-encoded AccD closely interact with each other and represent the α and β carboxyltransferase subunits, respectively. Each of these subunits first homodimerizes, and then they combine as a heterotetramer, forming the functional enzyme that transfers the carboxyl group from biotin to acetyl-CoA (Cronan and Waldrop 2002). Together, these enzymes convert acetyl-CoA to malonyl-CoA within the plastid, which is the first step in the fatty acid biosynthesis pathway (White et al. 2005). In some lineages, the homomeric ACCase has undergone a duplication, and one copy is targeted to the plastid while the other remains in the cytosol (Konishi and Sasaki 1994; Schulte et al. 1997; Babiychuk et al. 2011; Parker et al. 2014).

The angiosperm tribe Sileneae (Caryophyllaceae) has emerged as model for studying organelle genomes under divergent rates of sequence evolution (Mower et al. 2007; Erixon and Oxelman 2008; Sloan et al. 2012a,b, 2014a). This group contains multiple lineages with phylogenetically independent accelerations in rates of plastid genome evolution (Sloan et al. 2012a, 2014a). In contrast, closely related Sileneae lineages have largely maintained low ancestral rates of evolution. This rate variation among closely related species presents a powerful contrast to analyze the evolutionary mechanisms responsible for accelerated plastid genome evolution and test for correlated changes in nuclear-encoded counterparts. Here, we use transcriptome sequencing data coupled with structural information to identify variation in nuclear genes both within and among Sileneae species with highly divergent rates of plastid genome evolution. Specifically, we asked if there is evidence of selection on the sequences of nuclear-encoded subunits of the CLP and ACCase complexes in Silene species whose plastid-encoded counterparts have experienced recent accelerations in rates of evolution.

Materials and Methods

Taxon sampling, mRNA-seq, and transcriptome assembly

Silene conica, S. noctiflora, and S. paradoxa were all previously identified as having highly accelerated rates of nonsynonymous substitutions in a subset of plastid genes, with the most dramatic effects observed in clpP1 (Sloan et al. 2012a, 2014a). The accD gene also exhibited increased nonsynonymous substitution rates (albeit much less pronounced), as well as the accumulation of large indels in these species. In contrast, there was little or no evidence of accelerated sequence evolution in photosynthetic genes in these species. Silene latifolia, S. vulgaris, and Agrostemma githago were chosen as representatives of closely related lineages that have maintained low rates of evolution throughout their entire plastid genomes (Sloan et al. 2012a, 2014a). Transcriptomes for these six species (S. conica, S. latifolia, S. noctiflora, S. paradoxa, S. vulgaris, and Ag. githago) were taken from previously described data sets (Sloan et al. 2014b) that were each generated from a single individual and assembled with Trinity r20120608 (Grabherr et al. 2011). These data sets were used for all phylogenetic analyses and correspond to NCBI Sequence Read Archive (SRA) accessions SRX353031, SRX353047, SRX353048, SRX353049, SRX353050, and SRX352988. For two genes (CLPC and CLPP6), some sequences were extracted from separate SOAPdenovo-Trans v1.02 (Xie et al. 2014) assemblies of the same reads because the Trinity assemblies were fragmented or complex.

Seeds from 19 geographically dispersed S. conica collections, including ABR which was used for the original S. conica transcriptome referenced above and one collection of the close relative S. macrodonta (Supplemental Material, Table S1), were germinated on soil in either July or August 2014 (Fafard 2SV mix supplemented with vermiculite and perlite) and grown under a 16/8-hr light/dark cycle with regular watering and fertilizer treatments in greenhouse facilities at Colorado State University. Plants were grown for 7–9 weeks, and total RNA was extracted from 2 to 3 leaves of a single individual from each collection using an RNeasy Plant Mini Kit (QIAGEN, Valencia, CA). Rosette leaves were used for all individuals with the exception of the ARZ and PDA samples, for which cauline leaves were used. The resulting RNA was sent to the Yale Center for Genome Analysis for Illumina mRNA-seq library preparation and sequencing. For all but two samples, polyA selection was used during library construction, while for the ABR sample of S. conica and the single S. macrodonta sample, mRNA selection was performed using a Ribo-Zero Plant Leaf rRNA Removal Kit (Illumina) in an effort to capture more organellar transcripts (as part of an unrelated project). Resulting strand-specific Illumina libraries were sequenced on two lanes of an Illumina HiSeq 2500 to generate paired-end 151-bp (2 × 151) reads. Raw (i.e., nonnormalized and untrimmed) reads were then assembled using Trinity r20140717 with default parameters (note that the strand-specificity of the reads was not taken into account during assembly). Transcriptome assembly statistics and numbers of reads were similar among the 20 samples, except for an ∼50% reduction in the average and total length of assembled transcripts for samples where Ribo-Zero was used (Table S2).

Extraction and alignment of orthologous sequences from Sileneae species

The focus of our study was the nuclear-encoded components of the plastid CLP and ACCase complexes (Table S3). In addition, sets of gene sequences were obtained from photosystem I (PSI) and the mitochondrial-targeted CLP protease (mtCLP) to serve as a basis for comparison. PSI was selected because it contains subunits from both the nuclear and plastid genomes but, unlike CLP and ACCase, the plastid-encoded subunits have been highly conserved even in Silene species with accelerated rates of evolution in other plastid genes (Sloan et al. 2012a, 2014a). The mtCLP complex was chosen because it is homologous to the plastid CLP, but consists entirely of nuclear-encoded subunits and is targeted to a different cellular compartment (the mitochondria). The mtCLP complex has a homotetradecamer core consisting entirely of CLPP2 subunits that interacts with the chaperones CLPX1, 2, and 3 (van Wijk 2015). Additionally, 50 genes with a minimum coding sequence length of 600 bp were selected at random from a published list of single-copy nuclear genes in angiosperms to test for global increases in evolutionary rates within the nuclear genome (Table S4; Duarte et al. 2010). Genes that were annotated as being targeted to the mitochondria or plastids were excluded from this random set. The A. thaliana sequences for selected genes were obtained through the TAIR database (https://www.arabidopsis.org/) with accession numbers from the literature (Table S3 and Table S4; Peltier et al. 2004; Duarte et al. 2010; Olinares et al. 2011; van Wijk 2015).

BLAST+ 2.2.31 (Camacho et al. 2009) was used to run tblastn searches (default settings) with the selected A. thaliana aa sequences as queries against each of the assembled Sileneae transcriptomes. The top hit in each transcriptome was retrieved with a custom Perl script using BioPerl modules (Stajich et al. 2002). In cases where there were multiple paralogous genes, manual curation aided by exploratory tree-building was performed to identify orthologs. Genes that were absent from the transcriptomes or for which orthologs could not be confidently identified were excluded from further analysis. For the random set, genes were excluded and replaced with another randomly selected gene if one or more species lacked orthologous sequence or had a partially assembled transcript that was less than two-thirds the length of the coding sequence. Extracted sequences were aligned by nucleotide using the MUSCLE algorithm embedded within MEGA v6.0 (Tamura et al. 2013). The longest ORF in the Ag. githago sequence was predicted with the NCBI ORF Finder (http://www.ncbi.nlm.nih.gov/gorf/gorf.html), and other sequences were trimmed accordingly.

Sequences were then realigned by codon using MUSCLE. TargetP v1.1 (Emanuelsson et al. 2000) was used with translated ORFs from Ag. githago to predict the length of the N-terminal signal peptide (for plastid- and mitochondrial-targeted proteins), which was then removed from all sequences. In a few cases, TargetP was unable to predict a targeting peptide based on Ag. githago sequence, so S. latifolia, S. vulgaris, or A. thaliana was used instead to identify and remove signal peptides. Concatenated sequences for sets of genes in each complex were generated from final alignments with a custom Perl script using BioPerl modules.

Phylogenetic analysis of rates of sequence evolution

For each gene individually and for the concatenated sets of nuclear-encoded genes of each complex, we conducted multiple analyses of rates of synonymous and nonsynonymous substitutions by using the codeml program within PAML version 4.8 (Yang 2007). An F1 × 4 codon frequency model was applied in each analysis, and a constrained tree topology was used with the species in Silene subgenus Behenantha (S. conica, S. latifolia, S. noctiflora, and S. vulgaris) collapsed as a polytomy. First, we implemented a free branch model (model = 1 in PAML) to estimate d_N/d_S for each branch independently. Branches that were identified as having d_N/d_S values > 1 were then tested for significance using a LRT that compared the free branch model to a model that constrained d_N/d_S for the individual branch in question to a value of one. Finally, we classified species/branches into two groups, “fast” and “slow,” based on known rates of plastid genome evolution (Sloan et al. 2012a, 2014a) and estimated separate d_N/d_S values for each group (model = 2 in PAML). The terminal branches for S. conica, S. noctiflora, and S. paradoxa were assigned to the fast group, while those for S. latifolia, S. vulgaris, and Ag. githago were assigned to the slow group. The internal branch connecting the common ancestor of Silene to the base of Silene subgenus Behenantha was also included in the slow group. As above, we used an LRT to test for significance in cases in which individual genes or concatenated sequences for entire complexes had an estimated d_N/d_S value > 1 for the fast group (no such cases where identified for the slow group). For the constrained model in these LRT comparisons, the d_N/d_S value for the fast group was set to one.

McDonald–Kreitman (MK) tests

MK tests (McDonald and Kreitman 1991) were performed using sequences from the S. conica population genetic data set. Sequences were extracted, aligned, and trimmed following the same methodology described above for the phylogenetic analysis. The tests were implemented with the web server described by Egea et al. (2008). For each gene, the neutrality index (NI) was calculated by dividing the ratio of nonsynonymous to synonymous polymorphisms within S. conica (P_n/P_s) by the ratio of nonsynonymous to synonymous divergence from a closely related outgroup (see below) species (D_n/D_s) (Rand and Kann 1996). NI values < 1 are indicative of positive selection, with statistical significance assessed by a standard contingency-table χ² analysis. We also calculated the direction of selection (DoS) for each gene (Stoletzki and Eyre-Walker 2011). Positive DoS values are indicative of positive selection and an excess of nonsynonymous substitutions. We looked for evidence of selection in sets of related genes by summing polymorphism and divergence counts for genes belonging to the plastid CLP, ACCase, PSI, or mtCLP complexes, as well as for the set of random genes. Because summing across contingency tables can introduce statistical bias, we also calculated NI_TG for each combined set of related genes, which is an unbiased estimator of NI (Stoletzki and Eyre-Walker 2011). Two separate sets of analyses were carried out, using either S. latifolia or S. macrodonta as the outgroup. Extracted sequences from the S. conica and S. macrodonta transcriptome assemblies that could not be confidently identified as orthologous were removed from the analysis. Specifically, CLPR2, CLPX1, and one randomly selected gene (AXS2: AT1G08200) showed evidence of recent duplications in the S. conica lineage, leading to apparent chimeric assembly artifacts. Therefore, these genes were not used for MK tests. In addition, all three CLPX genes and 32 of the randomly selected nuclear genes had low coverage and fragmented assemblies in the S. macrodonta data set, so MK tests for these genes were only performed with S. latifolia as an outgroup. The low coverage for many nuclear genes in the S. macrodonta assembly was likely related to the use of Ribo-Zero in the construction of that library (Table S2).

Analysis of protein structure and position of substitutions

To gain insight into the functional consequences of aa changes observed in fast-evolving Silene species, we mapped substitutions onto plastid CLP and ACCase protein structures. Ancestral Silene sequences were inferred using codeml in PAML with the guide tree containing the five Silene species encoded as a polytomy, with Ag. githago and Escherichia coli as outgroups. Partial sequences were excluded when inferring ancestral sequences. For each CLPP and CLPR subunit (including the plastid-encoded ClpP1), changes that were inferred to have occurred in S. conica, S. paradoxa, or S. noctiflora from the ancestral Silene sequence were mapped onto the structure of an individual E. coli ClpP subunit (PDB accession 1YG6; Bewley et al. 2006; Yu and Houry 2007). Likewise, changes in ACCase subunits were also mapped onto solved E. coli structures (PDB accessions 4HR7 and 2F9Y: Bilder et al. 2006; Broussard et al. 2013). Template structures from E. coli were used because no plant CLP or ACCase structures have been solved. While it is likely that there have been structural changes within these complexes between bacteria and plants, most of these subunits are anciently conserved and can be reliably aligned (average aa identity = 43.3%).

Data availability

Raw Illumina reads and assembled transcriptome sequences are available via the NCBI SRA and Transcriptome Shotgun Assembly (TSA) database, respectively. Accession numbers are provided in Table S2. Sequence alignments used in PAML, MK, and phylogenetic analyses are provided in File S1.

Results

CLP and ACCase gene content in the tribe Sileneae

We were able to recover most of the expected genes from the Sileneae transcriptomes, as the gene content in these species was largely similar to that of Arabidopsis. However, we did find that some genes had experienced recent duplications or losses. We identified orthologs of all eight of the nuclear-encoded CLPP and CLPR genes that are targeted to the plastid in Arabidopsis (Figure S1), and we found that CLPP5 has been duplicated (with the resulting copies designated as CLPP5a and CLPP5b). The duplication appears to have occurred prior to the divergence between Agrostemma and Silene, but only one of these copies (CLPP5b) was recovered from the Agrostemma transcriptome (Figure S1). As described in the Materials and Methods, there may also have been more recent duplications of genes such as CLPR2 in individual species. In addition to the subunits that make up the core proteolytic rings of the plastid CLP complex, we also identified orthologs of the associated chaperones, adapters, and accessory proteins that have been described in Arabidopsis (Table 1; Nishimura and van Wijk 2015). The newly discovered CLPF adapter was also identified in our data set but was not included in the present analysis (Nishimura et al. 2015). Arabidopsis contains three paralogous chaperone genes that contribute to the plastid CLP complex (CLPC1, CLPC2, and CLPD). We found evidence that multiple copies of CLPC/D also exist in the Sileneae, but that the assemblies of these long genes were often fragmented, and we were only able to successfully extract one set of orthologs, which we refer to as CLPC. In addition to these plastid-targeted subunits, we also found orthologs of the mtCLP genes that have been identified in Arabidopsis (CLPP2, CLPX1, CLPX2, and CLPX3; van Wijk 2015).

Table 1. Summary of d_N/d_S estimates.

	Gene	Ag. githago	S. paradoxa	S. conica	S. noctiflora	S. latifolia	S. vulgaris
Plastid CLP	CLPP3	0.09	0.88	2.89	2.31	0.52	0.00
	CLPP4	0.02	0.36	0.62	1.24	0.21	0.00
	CLPP5a	0.03	0.52	1.55	0.53	0.00	0.03
	CLPP5b	0.02	0.94	2.03	1.77	0.13	0.17
	CLPP6	0.10	0.57	2.84	1.23	0.35	0.09
	CLPR1	0.04	1.79	8.86	2.40	0.46	0.10
	CLPR2	0.02	3.65	1.16	2.41	0.06	0.00
	CLPR3	0.03	8.64	1.24	2.29	0.22	0.07
	CLPR4	0.01	1.83	2.37	2.02	0.08	0.06
	CLPC	0.02	0.22	0.47	0.75	0.02	0.03
	CLPT1	0.10	1.66	1.31	7.31	0.27	0.10
	CLPT2	0.17	0.67	0.80	1.24	0.31	0.36
	CLPS	0.02	0.03	0.05	0.39	0.00	0.04
	Concatenated	0.05	0.87	1.30	1.55	0.16	0.06
ACC	ACCA	0.01	0.99	0.49	–	0.06	0.05
	ACCB1	0.17	–	0.45	–	0.07	0.14
	ACCB2	0.67	0.18	0.11	–	0.42	0.59
	ACCC	0.01	0.08	0.08	–	0.08	0.06
	Concatenated	0.13	0.33	0.21	–	0.16	0.13
PSI	LHCA2	0.03	0.00	0.04	0.04	0.09	0.02
	LHCA3	0.02	0.05	0.13	0.06	0.02	0.07
	PSAK	0.00	0.07	0.28	0.45	0.00	0.00
	PSAL	0.02	0.08	0.18	0.11	0.00	0.00
	PSAN	0.07	0.14	0.09	0.05	0.09	0.00
	PSAO	0.09	0.00	0.20	0.13	0.21	0.05
	PSAP	0.18	0.12	0.03	0.14	0.18	0.16
	Concatenated	0.05	0.05	0.14	0.11	0.08	0.04
mtCLP	CLPP2	0.05	0.48	0.09	0.00	0.12	0.21
	CLPX1	0.06	0.21	0.08	0.12	0.07	0.02
	CLPX2	0.16	0.13	0.26	0.20	0.22	0.18
	CLPX3	0.12	0.52	0.32	0.08	0.31	0.30
	Concatenated	0.10	0.28	0.18	0.11	0.16	0.15
Random	Concatenated	0.12	0.15	0.12	0.15	0.13	0.15

Values > 1 are highlighted in bold, with underlined text indicating statistical significance based on likelihood ratio tests. Cells containing “–” indicate that the particular gene was not found in the corresponding transcriptome. Only concatenated results are reported for the set of 50 random genes. Individual gene results for this random set are available in Table S5. Note that CLPP5a and CLPP5b are recently duplicated paralogs and that only one of these copies was recovered from the Agrostemma transcriptome. Therefore, this Agrostemma sequence was used in both CLPP5 analyses. CLP, caseinolytic protease; ACC, acetyl-coA carboxylase; PSI, photosystem I; mtCLP, mitochondrial-targeted CLP protease.

Sileneae genes were successfully identified from each of the three classes of nuclear-encoded subunits of the heteromeric plastid ACCase (ACCA, ACCB, and ACCC), including two divergent copies of ACCB. Two copies of this gene also exist in Arabidopsis (Fukuda et al. 2013), but it was not readily apparent from phylogenetic analysis if there is an orthologous relationship between Sileneae and Arabidopsis copies or if they are the product of independent duplication events (data not shown). Although all of the heteromeric ACCase genes were identified in this clade, we found evidence of recent gene loss in some of the Silene species, which is described in detail below (see “Gene Loss and Accelerated Evolution of Some Subunits in the ACCase Complex”).

With respect to the homomeric ACCase that is typically targeted to the cytosol, our transcriptome data indicated that Sileneae species express two distinct copies of this gene (Figure S2 and Figure S3), and that one of the resulting proteins has an N-terminal extension that is strongly predicted to act as a plastid-targeting peptide (with a specificity > 0.95 based on TargetP analysis). Duplication of the homomeric ACCase and retargeting to the plastids has occurred repeatedly and independently during angiosperm evolution (Figure S3). The observed duplication in our data set preceded the divergence between Agrostemma and Silene, but it was independent from similar duplications in grasses (Konishi and Sasaki 1994) and the Brassicaceae (Schulte et al. 1997; Babiychuk et al. 2011; Parker et al. 2014).

Nuclear-encoded components of the plastid CLP complex show elevated rates of aa substitution in species with rapidly evolving plastid genomes

We found that nuclear-encoded CLP genes have dramatically elevated d_N/d_S values in Silene species with recent accelerations in the evolutionary rates of plastid-encoded clpP1 (Table 1 and Table 2). When concatenated, all 13 nuclear-encoded CLP genes had a d_N/d_S value significantly > 1 for both S. conica and S. noctiflora, and nearly = 1 for S. paradoxa (Table 1). In contrast, concatenated CLP genes had d_N/d_S values between 0.05–0.16 for closely related species with typical rates of clpP1 evolution (Table 1). The extreme variance in d_N/d_S estimates resulted from elevated nonsynonymous substitution rates, whereas synonymous substitution rates were very similar across species (Figure 1). Rate differences were most pronounced for CLPR subunits, which occupy the same structural ring as the plastid-encoded CLPP1 subunit (van Wijk 2015). All 12 d_N/d_S estimates for CLPR genes within the “fast” species were > 1, with eight found to be significantly > 1 (Table 1). The d_N/d_S estimates for the CLPP genes within these fast species were also highly elevated, but only eight of the 15 were > 1, and only one was significantly so (Table 1). The adaptor gene CLPS was a noticeable outlier, being generally conserved in species regardless of their rates of plastid genome evolution (Figure 2).

Table 2. Summary of d_N/d_S for constrained sets of fast and slow lineages based on concatenations of genes in each functional complex.

	dN/dS
Genes	Slow Lineages	Fast Lineages	P-Value
Plastid CLP	0.07	1.25	0.000
ACC	0.14	0.26	0.014
Photosystem I	0.05	0.10	0.002
mtCLP	0.11	0.20	0.002
Random	0.13	0.13	0.894

The P-values indicate significant differences in rate between the two sets of lineages based on LRTs. d_N/d_S, nonsynonymous to synonymous substitution rates; CLP, caseinolytic protease; ACC, acetyl-coA carboxylase; mtCLP, mitochondrial-targeted CLP protease.

Figure 1.

Rates of sequence evolution in nuclear genes coding for subunits of the plastid and mitochondrial CLP complexes. Branch lengths are scaled to the amount of nonsynonymous (d_N) and synonymous (d_S) divergence per site. Species with rapid rates of plastid genome evolution (Sloan et al. 2014a) are highlighted in red. CLP, caseinolytic protease.

Figure 2.

Gene-by-gene comparison of d_N/d_S in fast vs. slow sets of lineages. Points are color-coded by functional complex, and the diameter of each point is proportional to gene length. All points represent individual genes except the “random” point, which is based on the concatenation of all 50 genes in that set. The size of that point is scaled to the average length of each gene rather than the total concatenation length. ACCA and CLPS, which exhibit distinct patterns from the other ACCase and CLP subunits, are labeled individually. ACCase, acetyl-coA carboxylase; CLP, caseinolytic protease; PSI, photosystem I.

Gene loss and accelerated evolution of some subunits in the ACCase complex

The three species with high rates of plastid-encoded accD evolution exhibited varied patterns with respect to nuclear-encoded ACCase genes, including some cases of accelerated evolution and other examples of outright gene loss. Notably, none of the nuclear-encoded ACCase genes were identified in the assembled S. noctiflora transcriptome. We confirmed the loss of these genes by searching a draft assembly of the S. noctiflora nuclear genome (D. B. Sloan, unpublished data). The nuclear genome assembly contained only pseudogenized fragments of ACCA, and none of the other ACCase genes were detected. The S. paradoxa transcriptome also appeared to lack a full complement of functional nuclear-encoded ACCase genes. Most of the species contained two ACCB paralogs, but we did not detect a copy of ACCB1 in the S. paradoxa transcriptome. In addition, the assembly of the S. paradoxa ACCA transcript was incomplete, covering only 543 nt of a 2133-nt alignment and exhibiting a d_N/d_S of 0.99 (Table 1). This partial ACCA transcript was aberrantly spliced, resulting in a 9-nt insertion that introduced a premature in-frame stop codon (Figure S4). Therefore, despite being transcribed, ACCA is likely a pseudogene in S. paradoxa. In contrast, ACCB2 and ACCC were both intact with very low d_N/d_S values in S. paradoxa. Finally, unlike in S. noctiflora and S. paradoxa, all four nuclear-encoded ACCase genes were present and intact in the transcriptomes of S. conica and all three “slow” species.

The ACCA subunit, which interacts directly with the plastid-encoded ACCD subunit, was highly divergent in aa sequence in S. conica. We initially estimated an elevated d_N/d_S value of 0.49 for ACCA in S. conica, and that value increased to 0.94 when we analyzed the full-length of the gene by excluding the partial S. paradoxa sequence from the alignment (PAML was run with the “cleandata” option, which ignores alignment positions for which gaps are present in any of the sequences). There was a striking difference between this elevated d_N/d_S in S. conica and the very low values (≤ 0.06) for ACCA in the slow group species (Table 1). The S. conica ACCB1 gene also exhibited substantially higher d_N/d_S values than in any of the other species, whereas d_N/d_S was very low in ACCB2 and ACCC (Table 1). In general, rates of nonsynonymous substitution in the slow species were very low (d_N/d_S < 0.2) for ACCase genes, with the exception of ACCB2. Interestingly, the d_N/d_S values for this gene showed a converse pattern, in which d_N/d_S was elevated in the slow species relative to S. conica and S. paradoxa (Table 1).

Low rates of nonsynonymous substitutions in mtCLP, PSI, and randomly selected genes

For both the PSI and mtCLP concatenated gene sets, there was significantly higher d_N/d_S in the fast group (Table 2). However, d_N/d_S values for both PSI and mtCLP were generally low in all species, and the differences between the two species groups were very small, especially in comparison to the differences observed for the plastid CLP complex and some ACCase genes (Figure S5 and Table 1). The lower overall d_N/d_S estimate for the slow species group appeared to be largely driven by the low values for the Ag. githago branch, which has a disproportionate effect on the estimation of d_N/d_S in this group because it represents more divergence time and a large fraction of total observed substitutions. The randomly selected nuclear genes showed no significant difference in d_N/d_S between the fast and slow species groups (Table 1, Table 2, and Table S5). Thus, it does not appear that there is a global elevation of d_N/d_S in the nuclear genomes of species with rapidly evolving plastid genomes.

MK tests reveal excess of nonsynonymous divergence between species

Despite high levels of observed nonsynonymous divergence in ACCA and the majority of the nuclear genes that encode components of the plastid CLP complex, most of the segregating variants within S. conica are synonymous. Thus, in ACCA and the concatenated set of nuclear-encoded CLP genes, there was a large and highly significant excess of nonsynonymous divergence from the outgroup S. latifolia relative to levels of nonsynonymous and synonymous polymorphism within S. conica (Table 3). The PSI genes had extremely low D_n/D_s values, but the P_n/P_s values were even lower, again resulting in a significant excess of nonsynonymous divergence for the concatenated gene set (Table 3). In contrast, there were no indications of a similar excess in the mtCLP genes or the set of randomly selected nuclear genes, as their concatenated sequences had NI values very close to one (Table 3 and Table S6). Repeating the MK analysis with a more closely related outgroup (S. macrodonta) produced similar results (Table S6 and Table S7).

Table 3. Summary of MK tests using intraspecific polymorphism within S. conica relative to interspecific divergence between S. conica and S. latifolia.

	Gene	Ps	Pn	Ds	Dn	NI	DoS	P-Value
Plastid CLP	CLPP3	8	3	16	83	0.07	0.57	0.000
	CLPP4	18	0	31	51	0.00	0.62	0.000
	CLPP5a	5	2	4	19	0.08	0.54	0.006
	CLPP5b	0	0	11	25	–	–	–
	CLPP6	0	2	11	59	–	−0.16	0.543
	CLPR1	1	1	22	144	0.15	0.37	0.256
	CLPR3	3	2	48	120	0.27	0.31	0.128
	CLPR4	1	1	22	80	0.28	0.28	0.337
	CLPC	4	1	55	50	0.28	0.28	0.226
	CLPT1	4	0	10	34	0.00	0.77	0.001
	CLPT2	7	1	19	39	0.07	0.55	0.002
	CLPS	3	0	11	2	0.00	0.15	0.467
	Concatenated	54	13	260	706	0.09 (0.11)	0.54	0.000
ACC	ACCA	13	7	42	111	0.20	0.38	0.001
	ACCB1	0	1	14	11	–	−0.56	0.270
	ACCB2	4	3	10	7	1.07	−0.02	0.939
	ACCC	10	3	22	18	0.37	0.22	0.160
	Concatenated	27	14	88	147	0.31 (0.31)	0.28	0.001
Photosystem I	LHCA2	4	0	19	6	0.00	0.24	0.271
	LHCA3	13	1	28	9	0.24	0.17	0.167
	PSAK	1	0	13	7	0.00	0.35	0.468
	PSAL	5	0	15	5	0.00	0.25	0.211
	PSAN	11	2	18	10	0.33	0.20	0.183
	Concatenated	34	3	93	37	0.22 (0.20)	0.20	0.011
mtCLP	CLPP2	1	1	21	4	5.25	−0.34	0.233
	CLPX2	4	5	26	23	1.41	−0.09	0.634
	CLPX3	9	3	12	8	0.50	0.15	0.387
	Concatenated	14	9	59	35	1.08 (1.05)	−0.02	0.866
Random	Concatenated	368	178	1297	570	1.10 (1.07)	−0.02	0.371

For each gene, the NI and the DoS were calculated, with NI values < 1 and DoS values > 0 indicative of positive selection (i.e., an excess of nonsynonymous divergence). For concatenations of genes within each complex, NI_TG (an unbiased estimator of NI) is reported in parentheses. Only concatenated results are reported for the set of random genes. Individual gene results for this random set are available in Table S6. MK, McDonald–Kreitman test; P_s, synonymous polymorphisms; P_n, nonsynonymous polymorphisms; D_s, synonymous divergence; D_n, nonsynonymous divergence; NI, neutrality index; DoS, direction of selection; CLP, caseinolytic protease; ACC, acetyl-coA carboxylase; mtCLP, mitochondrial-targeted CLP protease.

Substitutions in nuclear-encoded subunits preferentially occur at interfaces with plastid-encoded subunits within the CLP complex, but not within the ACCase complex

To investigate the effect of physical interactions between CLP complex subunits on substitution patterns, we mapped observed changes in CLPP and CLPR subunits onto the solved structure of the representative ClpP subunit from E. coli (Figure 3). This analysis showed that numerous substitutions have occurred throughout the entirety of the plastid-encoded ClpP1 subunit in fast-evolving Silene species. These include substitutions in: (1) the handle domain, which physically interconnects the two heptameric rings and likely stabilizes ring–ring interactions; (2) the head domain, which likely stabilizes interactions between subunits within a single ring; and (3) the axial loop regions, which form the axial pores and mediate interactions with associated chaperones (Figure 3C) (Yu and Houry 2007). Several individual residues that have been implicated in ring stability or substrate interactions (Wang et al. 1997) were also observed to have undergone changes in ClpP1 (Figure 3).

Figure 3.

(A) ClpP protease structure from E. coli (PDB 1YG6; Yu and Houry 2007) oriented with the two heptameric rings stacked on top of each other (left) or in front and behind (right). One of the 14 (identical) E. coli subunits is highlighted in blue. (B) A single E. coli ClpP subunit (as in part (A)), with the head, handle, and axial-loop domains highlighted. The surface of the catalytic triad is indicated in mesh and the individual residues (Ser, His, and Asp) are labeled. Other important residues are numbered and indicated with stick models, including those that interact with substrates (Phe-17 and Phe-49) or contribute to heptamer stability by forming hydrophobic bonds (Asn-41/Tyr-20, Asn-41/Thr-32, Asp-78/Asn-116, and Asp-171/Try-128) or ion pairs (Arg-118/Glu-141, Arg-170/Glu-134, and Asp-171/His-138). (C) Locations of substitutions in Silene species with fast-evolving CLP sequences (S. conica, S. noctiflora, and S. paradoxa). Residues with changes (relative to the inferred ancestral Silene sequence) in one, two, or three species are indicated in yellow, orange, and red, respectively. The black triangle highlights the ClpP1 site at which large insertions have occurred in S. conica (39 aa) and S. noctiflora (seven aa). CLP, caseinolytic protease.

Given the extreme levels of divergence in the plastid-encoded ClpP1 subunit, we reasoned that substitutions might be nonrandomly distributed within interacting nuclear-encoded subunits. Specifically, we predicted that the nuclear-encoded CLPR subunits would have an abundance of changes in the head domain because the ClpP1 subunit assembles with nuclear-encoded CLPR subunits to form the R-ring (Nishimura and van Wijk 2015), and the head domain contains residues that are likely to maintain intraring interactions (Yu and Houry 2007). Similarly, we predicted that nuclear-encoded CLPP subunits would have a disproportionate number of substitutions in their handle domains because these subunits form the P-ring, and their handle domains are likely involved in interactions with the highly divergent copies of ClpP1 in the R-ring. As predicted, aa substitutions were significantly overrepresented in the head domains of CLPR subunits and the handle domains of CLPP subunits in all three fast-evolving species (Table 4).

Table 4. The ratio of aa substitutions in the head vs. handle regions is higher in CLPR subunits than CLPP subunits.

	Head:Handle Ratio		Fisher’s Exact Test
Species	CLPP	CLPR	P-Value
S. conica	85:63	136:61	0.031
S. noctiflora	50:37	109:36	0.006
S. paradoxa	49:40	127:22	0.000

Values represent counts of substitutions summed across all nuclear-encoded genes in each category. CLP, caseinolytic protease.

In most plants, both the nuclear- and plastid-encoded CLPP subunits (but not the CLPR subunits) retain the Ser-His-Asp triad that confers protease activity (Peltier et al. 2004). However, we found that in fast-evolving Silene species, many of these subunits have experienced substitutions at the His or Asp positions within this highly conserved triad (whereas the catalytic Ser is universally conserved in our data set; Table S8).

Relative to the CLP complex, there were fewer changes in the ACCase subunits, and only a fraction of these could be analyzed in a structural context because several regions of ACCase proteins lack structural information (Broussard et al. 2013). In particular, there is a large N-terminal extension of AccD that is highly variable among angiosperms (Greiner et al. 2008a) and absent entirely from E. coli. Likewise, the C-terminal half of ACCA is also unique to plants. In contrast to the pattern observed in the CLP complex, the ACCase changes that were able to be mapped to the E. coli structure occurred away from protein–protein interfaces and generally did not involve functionally important residues (Figure 4). This was also true for a site at which large insertions are present in the AccD subunit in both S. conica and S. paradoxa (Figure 4B).

Figure 4.

(A) Structure of the carboxyltransferase component from the E. coli ACCase (PDB 2F9Y), oriented as in Figure 2C from Bilder et al. (2006). Residues of active sites (Gly-204, Gly-205, Gly-206, and Gly-207) are shown as spheres, the four cysteines that form cysteinyl zinc ligands (Cys-27, Cys-30, Cys-46, and Cys-49) are shown as stick models, and the helices and sheets that compose the catalytic platform are indicated. (B) The same structure as in (A) with inferred Silene changes mapped as described in Figure 3. The red arrow indicates the location of an AccD site with large insertions in both S. conica (26 aa) and S. paradoxa (58 aa). Importantly, this model contains only the portions of ACCA and AccD that can be aligned and mapped to E. coli, as the copies of these genes in plants are roughly twice the length of their counterparts in E. coli owing to large N- or C-terminal extensions of unknown function. (C) Structure of the biotin carboxylase (ACCC)-biotin carboxyl carrier protein (ACCB) complex from the E. coli ACCase (PDB 4HR7), as in Figure 2A from Broussard et al. (2013). (D) An ACCC monomer complexed with two ACCB monomers, as in Figure 2D in Broussard et al. (2013), indicating functional domains and ACCB–ACCC interfaces. The Lys-122 residues that bind biotin and the active site of ACCC (Arg-338) are shown as spheres, while residues that play important roles in ACCB-ACCC interactions are shown as sticks and highlighted with arrows. (E) The same structure as in (D) with inferred Silene changes mapped as described in Figure 3. (F) Two ACCC dimers and a single ACCB monomer, showing ACCC–ACCC interfaces as in (A) from Broussard et al. (2013). Residues that play important roles in ACCC–ACCC interactions are shown as sticks and highlighted with arrows. (G) The same structure as part (F), with inferred Silene changes mapped as described in Figure 3. ACCase, acetyl-coA carboxylase.

Discussion

In numerous angiosperm lineages, a subset of plastid genes, including clpP1 and accD, display accelerated evolutionary rates, but the causes of this recurring phenomenon have remained unclear (Jansen et al. 2007; Erixon and Oxelman 2008; Greiner et al. 2008b; Guisinger et al. 2008, 2010, 2011; Straub et al. 2011; Sloan et al. 2012a, 2014a; Barnard-Kubow et al. 2014; Weng et al. 2014; Dugas et al. 2015; Williams et al. 2015; Blazier et al. 2016; Zhang et al. 2016). We investigated the nuclear genes that contribute to the multisubunit complexes that include ClpP1 and AccD, and incorporated population genetic and structural data to distinguish between relaxed purifying selection and positive selection as drivers of elevated d_N/d_S values.

Our analysis revealed different patterns of selection on the nuclear-encoded CLP and ACCase genes, which may reflect the contrasting evolutionary histories of the plastid-encoded subunits in these two complexes. The patterns of clpP1 sequence divergence in some lineages are truly remarkable and include both structural changes (i.e., indels and loss of introns) and extreme increases in nonsynonymous substitution rates (e.g., Erixon and Oxelman 2008; Sloan et al. 2012a). For example, the aa identity between the clpP1 copies in the fast-evolving species S. conica and S. noctiflora is only 34%, and many Silene species have elevated d_N/d_S values (up to 5.9 in S. fruticosa; Erixon and Oxelman 2008). In contrast, slow-evolving species from the same genus, such as S. latifolia, retain up to 58% identity with free-living cyanobacteria, so these recent accelerations have led to far more divergence in the last few million years than has typically accumulated since the endosymbiotic origins of photosynthetic eukaryotes roughly one billion years ago. The contrasts between fast and slow lineages for accD are far less stark. The increased rates of aa substitution in fast lineages are only modest, and most of the sequence change is caused by indels (Sloan et al. 2012a, 2014a). Furthermore, even in species with typical, slow-evolving plastomes, it is primarily the catalytic C-terminal domain of AccD that is highly conserved, whereas the N-terminal domain, which is plant-specific and has an unknown function, accumulates substantial structural divergence (Greiner et al. 2008a). The evolution of accD is further complicated in some angiosperm lineages by functional transfer to the nucleus (Magee et al. 2010; Rousseau-Gueutin et al. 2013) or by functional replacement with a duplicated and retargeted copy of the homomeric ACCase (Konishi and Sasaki 1994). In contrast, there is no evidence (to our knowledge) of functional transfer of clpP1 to the nucleus in green plants.

The complex history of the plastid accD gene in angiosperms is mirrored by the varied evolutionary histories that we observed within Silene for the nuclear-encoded ACCase subunits. One lineage (S. noctiflora) has experienced the outright loss of the heteromeric ACCase complex, and another lineage (S. paradoxa) appears to be undergoing gene loss/pseudogenization with signatures of relaxed selection. However, in a third fast species (S. conica), all the ACCase subunits are retained and one gene shows clear evidence of positive selection. In contrast to this heterogeneity in the evolution of ACCase genes, we found a consistent signal of positive selection throughout nearly all the subunits in the plastid CLP complex in all three fast species. It was especially striking to find d_N/d_S values significantly > 1 when averaged over more than a dozen nuclear-encoded CLP genes.

Loss of plastid heteromeric ACCase

The finding that S. noctiflora has completely lost the nuclear-encoded heteromeric ACCase genes is consistent with previous observations that the copy of the plastid-encoded accD may be a pseudogene in this species (Sloan et al. 2012a). Although the accD reading frame is intact in S. paradoxa, the gene is highly divergent and contains multiple large insertions, raising questions as to its functionality (Sloan et al. 2014a). These questions extend to the entire ACCase complex, as we found evidence of gene loss/decay in nuclear-encoded S. paradoxa ACCase genes (i.e., the apparent loss of ACCB1 and pseudogenization of ACCA). In at least two angiosperm lineages, the plastid accD gene has been transferred to the nucleus (Magee et al. 2010; Rousseau-Gueutin et al. 2013). However, we found no evidence of a functional nuclear copy of accD in any Silene species examined. Instead, the presence of a duplicated homomeric ACCase that is predicted to be targeted to the plastids may be compensating for the lost or altered function of the heteromeric ACCase, as shown in grasses (Konishi and Sasaki 1994). Because the duplication of the homomeric ACCase appears to have happened long before the divergence of Agrostemma and Silene (Figure S3), the heteromeric and duplicated homomeric ACCases have remained conserved and expressed for more than 20 MY (Sloan et al. 2009) in many lineages within this clade. This raises intriguing questions for future investigation about the respective roles of these enzymes and why functional loss of the heteromeric ACCase has occurred in some lineages, while others have retained all of the subunits and even exhibit evidence of strong positive selection in one case (see below).

Positive selection acting on nuclear-plastid enzyme complexes

We found signatures of intense positive selection acting on the plastid CLP complex. In many genes within the fast species, d_N/d_S is > 1 (often significantly so; Table 1), which is an especially powerful signature of positive selection because any effects are averaged across the entire length of each gene and likely dampened by purifying selection acting on many residues. Although an increase in the rate of nonsynonymous substitutions can also be indicative of reduced functional importance or even a pseudogene, we can reject that hypothesis based on the population genetic data. The vast majority of the CLP sequence polymorphism that is segregating within S. conica is synonymous (Table 3), meaning that these genes are still functionally constrained and that purifying selection is purging most new nonsynonymous mutations from the population. Instead, the large observed excess of nonsynonymous divergence between species (relative to intraspecific polymorphism) is an indication that a specific subset of aa substitutions have been preferentially driven to fixation due to positive selection (McDonald and Kreitman 1991).

Interestingly, some of the observed sequence changes affected residues in the catalytic triad in Silene CLPP subunits (Table S8). Substitutions in the catalytic triad of the plastid-encoded ClpP1 subunit in Acacia have previously been interpreted as evidence of pseudogenization (Williams et al. 2015). However, the strong selection acting on the plastid CLP complex in Silene suggests that both plastid- and nuclear-encoded CLPP subunits may retain an important functional role despite changes in the catalytic triad (Table S8). Although the Ser-His-Asp catalytic site is a widely conserved feature across the diversity of life, some of the same substitutions in this triad have been observed in other atypical serine proteases and functionally related enzymes (Schrag et al. 1991; Ekici et al. 2008; Zeiler et al. 2013). Notably, substitutions in the catalytic triad were only observed in one nuclear-encoded CLPP gene in each of the fast species. It is possible that such changes are tolerable as long as some of the subunits in the P-ring retain the canonical catalytic residues.

Given the evidence of gene loss and relaxed selection on the heteromeric ACCase in S. noctiflora and S. paradoxa, we initially suspected that the elevated d_N/d_S for ACCA in S. conica (0.94 for the full-length gene) was also due to relaxed selection. However, the population genetic data demonstrated that ACCA and the heteromeric ACCase complex are still functionally constrained in S. conica, as most of the intraspecific polymorphisms were synonymous (Table 3). Therefore, the unusually high rate of fixed ACCA aa substitutions in this lineage are most likely the result of positive selection and adaptive evolution.

We randomly selected a set of nuclear genes that are not targeted to the plastids or mitochondria, as well as genes from the mtCLP and PSI complexes. These were chosen with the a priori expectation that they would be similar between the fast and slow species, because they either do not interact with organelle-encoded subunits (random genes and mtCLP), or they interact with plastid-encoded subunits that show typical, slow rates of evolution in all of the species (PSI). The mtCLP complex, which is comprised solely of nuclear subunits, and the set of random nuclear genes largely supported this expectation. The concatenated mtCLP genes in fast species exhibited very similar (albeit slightly higher) d_N/d_S levels compared to slow species, and there was no significant difference for the random genes (Table 1 and Table 2). Furthermore, the MK tests found no evidence of positive selection for either of these data sets (Table 3, Table S6, and Table S7). In contrast, the nuclear PSI genes did exhibit a significant excess of nonsynonymous divergence (Table 3 and Table S7), even though the absolute rate of aa substitutions was extremely low. This evidence suggests that, although rare, some of the nonsynonymous substitutions in PSI are adaptive changes that spread under positive selection rather than fixing by drift.

An alternate interpretation for the observed excess of nonsynonymous divergence between species is that there was an ancestral bottleneck, which could have led to an increased frequency of weakly deleterious alleles spreading to fixation because of the reduced efficiency of selection in small populations (Hughes 2007). However, we identified three lines of evidence that lead us to reject this possibility and conclude that our results are, in fact, indicative of strong positive selection. First, if a demographic bottleneck had occurred, we would expect an excess of nonsynonymous divergence across all genes, but we did not observe this in the mtCLP genes or the randomly chosen nuclear genes. Previous studies in Silene also support the conclusion that the observed changes are not the result of genome-wide demographic effects; analysis of 140 cytosolic ribosomal proteins and seven mitochondrial-targeted complex II genes in S. conica and S. noctiflora did not show elevated d_N/d_S relative to other Silene species (Sloan et al. 2014b; Havird et al. 2015). Second, we observed elevated levels of nonsynonymous divergence across different timescales in our MK tests (Table 3 and Table S7) by using two different outgroups: S. latifolia (∼5.7 MY divergence time) and S. macrodonta (∼1.8 MY divergence time; Rautenberg et al. 2012). Therefore, separate bottlenecks at a minimum of two different historical points would had to have taken place. Third, the magnitude of the observed effects is inconsistent with a bottleneck. The relaxed selection associated with a reduced N_e should not increase d_N/d_S to values significantly > 1, which were frequently observed in our data set (Table 1). Thus, we conclude that, although it is possible that an ancestral bottleneck in S. conica might have contributed to some minor increases in aa substitution rates, the massive rate increases (e.g., in CLP genes) are more likely to have been driven by positive selection than by a temporary reduction in N_e.

Antagonistic coevolution and plastid–nuclear conflict

The accelerated aa substitution rates in both the nuclear- and plastid-encoded components of CLP and ACCase are very unusual for anciently conserved enzyme complexes, but are similar in many ways to the patterns that result from antagonistic coevolution between pathogens and host immune genes (Hughes and Nei 1988; Borghans et al. 2004). Selfish interactions and “arms races” can occur within a cell (i.e., intragenomic conflict) when there is opportunity for genetic elements to enhance their own transmission at the expense of organismal-level fitness (Burt and Trivers 2006). Such conflicts are common between the nucleus and cytoplasmic genomes. For example, copies of mitochondrial genomes with large deletions can confer a replication advantage within the cell, even if they harm overall fitness by reducing or eliminating the cell’s ability to respire (Taylor et al. 2002; Clark et al. 2012; Phillips et al. 2015). In addition, because most cytoplasmic genomes are inherited maternally, they can benefit from manipulating sexual reproduction to increase female reproduction and fitness (Perlman et al. 2015). Examples of this phenomenon include chimeric ORFs in plant mitochondrial genomes that induce cytoplasmic male sterility (CMS; Ingvarsson and Taylor 2002; Touzet and Budar 2004; Fujii et al. 2011) and numerous bacterial endosymbionts that manipulate sexual reproduction in animal hosts (Werren et al. 2008).

Interestingly, some of the earliest hypotheses about cytonuclear conflict were developed based on observations of differential rates of replication of plastid genomes in heteroplasmic plants (Grun 1976; reviewed in Greiner et al. 2015). Since that point, however, research on cytonuclear conflict in plants has overwhelmingly focused on mitochondria, particularly their role in CMS. Although plastids are often viewed as being relatively benign, in principle, the same evolutionary pressures could apply to these maternally inherited organelles. One possibility is that there are limited pathways available for plastids to exploit in a selfish fashion. For example, the major role of plastids (specifically chloroplasts) is in photosynthesis, and male reproductive tissues are generally nonphotosynthetic. However, plastids also perform other important processes (including CLP and ACCase activity). Recent studies have provided support for the possibility of selfish plastid–nuclear interactions within complexes such as the heteromeric ACCase and the CLP complex. In particular, reproductive incompatibilities (including male sterility) between wild and domesticated lines of peas were recently attributed to variation in nuclear- and plastid-encoded components of the heteromeric ACCase (Bogdanova et al. 2015). Plastid–nuclear incompatibilities have also been implicated in male sterility in Oenothera (Stubbe and Steiner 1999). Disruption of the synthesis of fatty acids and their derivatives, such as jasmonic acid, has also been associated with sterility phenotypes (Park et al. 2002). In addition, a recent proteomic analysis in wheat found that plastid-encoded clpP1 was one of the most upregulated genes in the anthers of male-sterile individuals, suggesting that it may play important functional roles in male reproductive tissues (Li et al. 2015).

Although the correlated increases in evolutionary rates and signatures of positive selection in Silene plastid–nuclear complexes could indicate a history of genomic conflict, they are not conclusive evidence that plastid and nuclear genes are locked in an arms race, or even that they are coevolving in any fashion (Lovell and Robertson 2010). General changes in selection for CLP or ACCase function could simultaneously affect all subunits, without interactions among the subunits being a major source of selection. It is also possible that the overall structure or subunit composition of the CLP complex has been radically disrupted or reorganized. Furthermore, even if coevolutionary dynamics are at play, they could involve mutually beneficial changes rather than antagonistic interactions (Rand et al. 2004). Adaptive changes in one genome may alter fitness landscapes and facilitate subsequent adaptive changes in the other genome, though it is unclear what forces might trigger such runaway adaptive evolution in these systems. A more conventional model of compensatory change in the nucleus in response to accumulation of deleterious changes in asexual organelle genomes (Rand et al. 2004; Osada and Akashi 2012) seems less likely, particularly for the CLP complex, given the evidence that accelerated rates of sequence evolution in the plastid-encoded clpP1 evolution are driven largely by positive selection (Erixon and Oxelman 2008; Sloan et al. 2012a; Barnard-Kubow et al. 2014).

It is worth noting that the largest increases in substitution rates were found in the nuclear-encoded subunits that interact most directly with plastid-encoded subunits. Specifically, the greatest elevation of d_N/d_S in the CLP complex occurred in the CLPR subunits (Table 1), which assemble with plastid-encoded ClpP1 subunits to form the proteolytic R-ring (Nishimura and van Wijk 2015). More detailed structural analysis of the CLP complex (Figure 3 and Table 4) showed that, even within subunits, there was an enrichment for substitutions in domains that have intimate interactions with ClpP1 (i.e., the head domains of CLPR proteins and handle domains of CLPP proteins; Yu and Houry 2007; Nishimura and van Wijk 2015). For the heteromeric ACCase, we found evidence of positive selection (Table 3) only on the nuclear-encoded ACCA subunit in S. conica, which interacts directly with the plastid-encoded ACCD subunit to make up the carboxyltransferase (Sasaki and Nagano 2004). However, our structural analysis did not detect an enrichment of substitutions at the interface between ACCA and AccD (Figure 4), and it is difficult to draw firm conclusions about ACCase structure given the lack of knowledge about the functions and interactions of the plant-specific portions of ACCA and AccD.

Therefore, it appears more likely that plastid–nuclear interactions and coevolution have played a role in generating positive selection and the observed accelerations in rates of sequence evolution in the CLP complex than in the heteromeric ACCase. We speculate that the mysterious rate accelerations that have occurred repeatedly in clpP1 in Silene, and throughout the diversification of flowering plants (and possibly those that have occurred in other plastid genes as well), are the result of antagonistic coevolution between the plastid and nucleus. An important test of this hypothesis will be to functionally characterize the sequence changes from rapidly evolving lineages, particularly with respect to their phenotypic effects on plastid replication within cells and plant allocation to male vs. female reproductive output.