Evolutionary Dynamics of Floral Homeotic Transcription Factor Protein-Protein Interactions.

Protein-protein interactions (PPIs) have widely acknowledged roles in the regulation of development, but few studies have addressed the timing and mechanism of shifting PPIs over evolutionary history. The B-class MADS-box transcription factors, PISTILLATA (PI) and APETALA3 (AP3) are key regulators of floral development. PI-like (PI(L)) and AP3-like (AP3(L)) proteins from a number of plants, including Arabidopsis thaliana (Arabidopsis) and the grass Zea mays (maize), bind DNA as obligate heterodimers. However, a PI(L) protein from the grass relative Joinvillea can bind DNA as a homodimer. To ascertain whether Joinvillea PI(L) homodimerization is an anomaly or indicative of broader trends, we characterized PI(L) dimerization across the Poales and uncovered unexpected evolutionary lability. Both obligate B-class heterodimerization and PI(L) homodimerization have evolved multiple times in the order, by distinct molecular mechanisms. For example, obligate B-class heterodimerization in maize evolved very recently from PI(L) homodimerization. A single amino acid change, fixed during domestication, is sufficient to toggle one maize PI(L) protein between homodimerization and obligate heterodimerization. We detected a signature of positive selection acting on residues preferentially clustered in predicted sites of contact between MADS-box monomers and dimers, and in motifs that mediate MADS PPI specificity in Arabidopsis. Changing one positively selected residue can alter PI(L) dimerization activity. Furthermore, ectopic expression of a Joinvillea PI(L) homodimer in Arabidopsis can homeotically transform sepals into petals. Our results provide a window into the evolutionary remodeling of PPIs, and show that novel interactions have the potential to alter plant form in a context-dependent manner.

Introduction

Gene regulation during development involves a complex interplay between DNA, RNA, and protein, resulting ultimately in organismal phenotype. Altered gene regulation, over the course of evolution, is one important mechanism through which phenotypes change. Regulatory landscapes evolve because of both noncoding and coding changes in the genome. Although noncoding changes play important roles in the evolution of organismal form (Stern 2000; Wray et al. 2003; Carroll 2005; Wittkopp and Kalay 2012; Della Pina et al. 2014), coding changes can also effect phenotypic change (Stern and Orgogozo 2008; Bartlett and Whipple 2013). Altered protein–protein interactions (PPIs), particularly between transcription factors, represent an example of coding change that can alter phenotype (Löhr and Pick 2005; Veron et al. 2007; Lynch and Wagner 2008; Airoldi et al. 2010; Brayer et al. 2011; Baker et al. 2012). Although we have learned much about how PPIs can alter transcription factor function (Adhikary and Eilers 2005; Slattery et al. 2011; Kong et al. 2012; Sayou et al. 2014), few studies have addressed the precise course and potential of PPI evolution (but see Yasumura et al. 2007; Melzer et al. 2014). To understand how novel transcription factor PPIs arise, and how they change over time, PPIs need to be investigated in an explicitly evolutionary context.

One transcription factor family that has been extensively investigated in plants is the MIKC MADS-box transcription factors. MADS-box transcription factors are present in all eukaryotic lineages and named for the first cloned representatives: MEF2, AGAMOUS, DEFICIENS, and SRF. The plant-specific MIKC MADS-box proteins have a characteristic domain structure: A DNA-binding MADS-box, intervening loop domain, keratin-like coiled-coil domain, and a disordered C-terminus (reviewed in Immink et al. 2010). MIKC MADS-box proteins are important regulators of floral development. In the ABC(E) model of floral development, floral organ identity is determined by the combinatorial action of ABC(E) MADS-box genes. For example, B-, C- and E-class functions together confer stamen identity, whereas C- and E-class functions together confer carpel identity (Coen and Meyerowitz 1991; Pelaz et al. 2000; Honma and Goto 2001). The ABC(E) model was originally derived using genetic mutants of Arabidopsis and Antirrhinum majus (snapdragon), but some of its core components, particularly B- and C-class functions, are broadly conserved in the angiosperms (reviewed in Litt and Kramer 2010; Smaczniak, Immink, Angenent, et al. 2012; Ó’Maoiléidigh et al. 2014).

Interactions between floral MADS-box proteins are critical for function. MADS-box proteins only bind DNA as dimers (Pellegrini et al. 1995) and the “floral quartet” model predicts that floral MADS-box proteins function as tetramers in planta (Theissen and Saedler 2001). A growing body of evidence supports the floral quartet model, and shows that higher-order PPIs are fundamental to floral MADS-box protein function (Honma and Goto 2001; Kaufmann et al. 2009; Smaczniak, Immink, Muino, et al. 2012; Mendes et al. 2013; Puranik et al. 2014). Interactions between MADS-box transcription factors can also have more subtle roles in the regulation of target genes. For example, E-class SEPALLATA (SEP) proteins of Arabidopsis bind DNA cooperatively. Thus, MADS-box PPIs have the capacity to influence the kinetics of DNA-binding, which may, in turn, affect downstream gene regulation and functional diversification (Jetha et al. 2014).

Changes to MADS-box PPIs have been implicated in phenotypic change. In Eschscholzia californica and Thalictrum thalictroides, mutations that disrupt MADS-box PPIs result in altered floral morphologies (Galimba et al. 2012; Lange et al. 2013). In snapdragon, shifting PPIs contributed to subfunctionalization of the C-class proteins PLENA and FARINELLI (Airoldi et al. 2010). In the oil palm, Elaies guineensis, two independent mutations that affect MADS-box dimerization underlie a modified fruit morphology that has been a target of selection in breeding (Singh et al. 2013). In the genus Medicago, the evolution of coiled fruit form is associated with stronger MADS-box PPIs (Fourquin et al. 2013). The evolution of B-class MADS-box PPIs, in particular, correlates with the evolution of some floral traits (Hernandez-Hernandez et al. 2007; Theissen and Melzer 2007; Lenser et al. 2009; Liu et al. 2010; Melzer et al. 2014). Floral development is more canalized in the eudicots and in many monocots, and B-class dimerization patterns have been connected to this canalization, particularly in the lineage leading to the core eudicots (Winter et al. 2002; Lenser et al. 2009; Melzer et al. 2014).

The B-class proteins APETALA3 (AP3) and PISTILLATA (PI) control petal and stamen development and are encoded by genes that are the product of a gene duplication event that predates angiosperm diversification (Kim et al. 2004). B-class function appears to be deeply conserved. In monocots and eudicots, B-class mutants exhibit similar phenotypes in which second whorl (petal) and third whorl (stamen) development is disrupted (Ambrose et al. 2000; Nagasawa et al. 2003; Vandenbussche et al. 2004; Drea et al. 2007; Kramer et al. 2007; Roque et al. 2013; Sharma and Kramer 2013; Quinet et al. 2014; Zhang et al. 2014; Bartlett et al. 2015). Despite a high degree of sequence similarity and overlapping expression domains, AP3L and PIL proteins are broadly nonredundant (Jack et al. 1992; Goto and Meyerowitz 1994; Ambrose et al. 2000; Bartlett et al. 2015). This nonredundancy is due in part to their PPI landscape: In many lineages, AP3L and PIL proteins bind DNA as obligate heterodimers (summarized in Melzer et al. 2014). AP3L/PIL heterodimers in Arabidopsis, maize, and snapdragon also regulate their own expression through autoregulatory feedback loops (Schwarz-Sommer et al. 1992; Goto and Meyerowitz 1994; Jack et al. 1994; Riechmann, Krizek, et al. 1996; Bartlett et al. 2015).

Broad surveys of B-class dimerization across the angiosperms show that although obligate PIL/AP3L heterodimerization is pervasive in the eudicots, both PIL and AP3L homodimerization occur across the tree, particularly in monocots and in early-diverging angiosperms (Winter et al. 2002; Melzer et al. 2014). For example, PIL proteins homodimerize in both Amborella trichopoda and the magnoliid Liriodendron tulipfera (Melzer et al. 2014). In the monocots, PIL proteins from Lilium (Winter et al. 2002; Chen et al. 2011), Tulipa (Kanno et al. 2003), and the orchid Phalaenopsis equestris (Tsai et al. 2008) both homo- and heterodimerize with AP3L proteins (facultative homodimerization). The single identified PIL protein from the close grass relative Joinvillea (J-PI) also homodimerizes (Whipple and Schmidt 2006). Ancestral state reconstructions integrating all available interaction data indicate that AP3L/PIL heterodimerization evolved early on in angiosperm history from facultative AP3L and/or PIL homodimerization (Winter et al. 2002; Melzer et al. 2014). How variable B-class dimerization is on smaller evolutionary timescales (within individual families and orders) is still unclear.

Here, we report on the evolution of B-class dimerization in the Poales, and on our molecular characterization of obligate heterodimerization versus homodimerization in the order. Our survey of B-class dimerization revealed an unexpected level of evolutionary change. Our analyses indicate multiple transitions between facultative PIL homodimerization and obligate PIL/AP3L heterodimerization in the Poales. Ancestral state reconstructions and population genetic data show that the obligate heterodimerization characteristic of one maize B-class protein STERILE TASSEL SILKY EAR1 (STS1) is a derived state, evolved recently from PIL homodimerization, and fixed during maize domestication. Although STS1 forms obligate heterodimers with the single maize AP3 homolog SILKY1 (SI1), its ortholog in Brachypodium distachyon (BdSTS1) can homodimerize, as can the Joinvillea PIL protein J-PI (Whipple and Schmidt 2006; Bartlett et al. 2015). Thus, comparisons between BdSTS1, J-PI, and STS1 provide a framework to investigate the molecular underpinnings of B-class homo- versus heterodimerization. We identified domains and residues that mediate homo- versus heterodimerization of PIL proteins in the Poales. Finally, we show that altered B-class PPIs can affect phenotype in a heterologous system.

Results

PIL Dimerization Patterns Are Labile in the Poales

Cases of PIL homodimerization have been reported in the monocots (most recently summarized in Melzer et al. 2014), but it was not clear whether these represented isolated curiosities or were indicative of broader evolutionary trends. To address this question, we used gel shift assays to survey 31 PIL proteins from 18 taxa spanning the Poales. We incubated a labeled DNA probe containing a known PI/AP3 binding site (Riechmann, Krizek, et al. 1996; Riechmann, Wang, et al. 1996; Hill et al. 1998) and one or more B-class proteins. Given that MADS-box proteins bind DNA as dimers, if we observed a shift when a single protein was added, we concluded the protein could bind DNA as a homodimer. Our analyses represent the first biochemical assays for most proteins under investigation. We found incidences of obligate AP3L/PIL heterodimerization, PIL homodimerization, and of AP3L homodimerization (fig. 1A, complete results in supplementary fig. S1, Supplementary Material online). A SI1 C-terminal truncation (SI1-C) formed artifactual homodimers in gel shift assays (fig. 1A and supplementary fig. S1, Supplementary Material online). Our results were consistent with published work, except for a single report on Oryza sativa PI-1 (OsMADS4) (Ronai et al. 2003). OsMADS4 did not form homodimers in yeast two-hybrid assays (Moon et al. 1999), and did not form homodimers in our gel shift assays (supplementary fig. S1, Supplementary Material online). Thus, we scored OsMADS4 as an obligate heterodimer. A second anomaly involved B-class proteins from the grass Pharus virescens. Although neither Pharus PI-1 nor Pharus PI-2 formed detectable homodimers in our gel shifts, we did not detect reliable interactions with any AP3L protein. Therefore, we scored both Pharus PI-1 and PI-2 as equivocal in all downstream analyses.

F

PIL dimerization patterns are evolutionarily labile in the Poales. (A) Typha PI did not homodimerize, but could form heterodimers with SI1-C. (B) Tillandsia PI can form homodimers (AP3L protein = Ecdeiocolea AP3). (C) Neither PI-1 nor PI-2 from Streptochaeta can form homodimers, but both can form heterodimers with Streptochaeta AP3. Streptochaeta AP3 can form DNA-binding homodimers. (D) PI-2 from Eriachne formed DNA-binding homodimers, whereas Eriachne PI-1 did not (AP3L protein = SI1). (E) BiFC assays with B-class proteins from Joinvillea and Zea mays. (E) STS1-YFPN and STS1-YFPC, (F) STS1-YFPN and SI1-YFPC, (G) J-PI-YFPN and J-PI-YFPC, (H) J-PI-YFPN and J-AP3-YFPC, (I) SI1–YFPN and SI1-YFPC (J) J-AP3-YFPN and J-AP3-YFPC, (K) Yeast two-hybrid assays with full-length proteins and IKC constructs. (L) Ancestral state reconstructions (likelihood) of dimerization states plotted onto a Bayesian phylogenetic reconstruction of PIL genes from the Poales, as well as noneudicot taxa where PIL dimerization has been assessed. Numbers next to the tree indicate nodes discussed in the text. •, SI1-C homodimer band.

Select interactions were further examined using bimolecular fluorescence complementation (BiFC) and yeast two-hybrid assays (fig. 1E and supplementary fig. S2, Supplementary Material online). The yeast two-hybrid assays produced spurious interactions when both tested proteins were truncated (IKC constructs) (similar results in Leseberg et al. 2008), and full-length proteins did not interact in yeast (even at 22 °C), as reported for other B-class proteins (data not shown, Honma and Goto 2001; Yang, Fanning, et al. 2003; de Folter et al. 2005). Therefore, we tested interactions using a full-length protein fused to the DNA-binding domain and a truncated IKC protein fused to the activation domain. All BiFC assays were performed with full-length protein in both directions. All three assays produced similar results, although with some minor inconsistencies (fig. 1 and supplementary fig. S1, Supplementary Material online). We used gel shifts to further investigate B-class PPIs because we could consistently use full-length proteins and assay DNA-binding as a consequence of dimerization. As B-class proteins are DNA-binding transcription factors, electrophoretic mobility shift assays (EMSAs) provided the most robust and biologically relevant assay.

PIL dimerization patterns are unexpectedly labile across the Poales. We mapped known PIL EMSA dimerization states onto a Bayesian gene phylogeny, including interactions described here (fig. 1L and supplementary fig. S3, Supplementary Material online). Although many more monocot PIL sequences are available, we included only those assayed for dimerization. We reconstructed ancestral dimerization states using maximum-likelihood and parsimony methods, as implemented in Mesquite V2.75 (Maddison and Maddison 2009). The parsimony reconstructions also used synthetic species trees that reflect current best estimates of relationships in the order (Givnish et al. 2010; GPWG II 2012; Bouchenak-Khelladi et al. 2014). The state at the base of the Poales was reconstructed as more likely to be homodimerizing, with or without non-Poalean outgroups (node 1, fig. 1 and supplementary fig. S4, Supplementary Material online). Similarly, the most parsimonious state at the base of the Poales is reconstructed as either equivocal or PIL homodimerization (supplementary fig. S4, Supplementary Material online). These results suggest that the Joinvillea PIL protein has retained the ability to homodimerize (Whipple and Schmidt 2006) from an ancestral condition in the Poales. We cannot precisely reconstruct the ancestral state at the base of the Poales, or the precise history of changes in B-class dimerization. However, both PIL homodimerization and obligate heterodimerization occur, and B-class interactions have clearly been in flux across the order.

Obligate Heterodimerization and PIL Homodimerization Evolved Convergently in the Grasses

PIL dimerization has also been in flux within the grass family. Obligate heterodimerization is the more likely ancestral state at the base of the grasses (node 2, fig. 1L and supplementary fig. S4, Supplementary Material online), but homodimerization has emerged within the family. Within the grasses, PIL genes have duplicated, resulting in two gene clades, PI-1 and PI-2 (Munster et al. 2001; Whipple et al. 2007). PI-1 proteins have retained the most likely ancestral condition of obligate heterodimerization. In contrast, PIL homodimerization emerged within the PI-2 clade, prior to the diversification of the hyperdiverse BEP and PACMAD clades (fig. 1L, node 6). Although we cannot pinpoint exactly when PIL homodimerization emerged in the grass PI-2 gene clade, ancestral state reconstructions suggest that J-PI homodimerization and the homodimerization characteristic of many grass PI-2 proteins are most likely independently derived. Our analyses unequivocally reconstruct obligate heterodimerization between the maize PI-2 protein, STS1, and the AP3 ortholog SI1 as a derived character state, evolved from PIL homodimerization. Thus, the obligate heterodimerization characteristic of the PI-1 and PI-2 proteins of maize (Whipple et al. 2007; Bartlett et al. 2015) has arisen independently.

If PIL homodimerization and obligate heterodimerization did arise de novo in the grass PI-2 gene lineage, and if there is more than one path to each dimerization state, the molecular determinants of obligate heterodimerization versus homodimerization would likely be distinct between J-PI, PI-2 from B. distachyon (BdSTS1), and PI-2 from maize (STS1). In order to identify which protein domains were determining obligate heterodimerization versus homodimerization in the Poales, we performed domain swap experiments. Because the I and K domains are critical for the AP3/PI interaction in Arabidopsis (Riechmann, Krizek, et al. 1996; Krizek and Meyerowitz 1996b; Yang, Fanning, et al. 2003; Yang, Xiang, et al. 2003; Yang and Jack 2004), we first swapped the I and K domains between BdSTS1 and STS1, and between J-PI and STS1.

Our gel shift assays indicated that the I-domain mediates homodimerization when swapped between J-PI and STS1, but not when swapped between BdSTS1 and STS1. A chimeric J-PI protein that contained the STS1 I-domain dimerized very weakly with J-AP3 (fig. 2B-1), but showed robust dimerization with a similar AP3L protein from Ecdeiocolea macrostachya (fig. 2C). In the reciprocal experiment, a chimeric STS1 protein that contained the J-PI I-domain formed homodimers (fig. 2B-2). Swapping the K-domain between J-PI and STS1 did not affect dimerization (figs. 2B-3 and B-4). In the BdSTS1/STS1 swaps, neither the I-swaps nor the K-swaps were sufficient to disrupt BdSTS1 homodimerization, or STS1 obligate heterodimerization (fig. 2D). These results indicate that the J-PI I-domain, but not the BdSTS1 I-domain, is sufficient to confer homodimerization on STS1. The reciprocal swaps showed that the STS1 I-domain is sufficient to abolish J-PI homodimerization, but does not affect BdSTS1 homodimerization. Thus, the molecular determinants of J-PI and BdSTS1 homodimerization are distinct, lending support to the hypothesis that PIL homodimerization has evolved convergently in the Poales.

F

A single amino acid change in the I domain is sufficient to allow STS1 homodimerization. (A) Domain structure of BdSTS1, J-PI, STS1, and ZMM18. The amino acids that differ between the I domains of BdSTS1, J-PI, and STS are highlighted. (B) J-PI and STS1 domain swaps. The AP3L protein in all lane pairs is Joinvillea AP3 (J-AP3). The J-PI I-domain is sufficient to confer the ability to homodimerize on STS1 (lane pair 2), and the I-domain from STS1 abolishes the ability of J-PI to homodimerize (lane pair 1). J-PI/J-AP3 heterodimerization is also negatively affected (lane pair 1). Swapping the K domains between J-PI and STS1 has no effect on dimerization capacity (lane pairs 3 & 4). (C) J-PI with the I domain from STS1 can interact with an AP3L protein from Ecdeiocolea. (D) BdSTS1 (Brachypodium distachyon) and STS1 (maize) domain swaps. Neither the I nor the K domain from STS1 is sufficient to disrupt BdSTS1 homodimerization. The AP3L protein in all lanes is SI1-C. (E) STS1 mutagenesis at single sites. G81D (lane pair 3) is sufficient to allow STS1 homodimerization. AP3L in lane pair 1 is SI1-C. The AP3L protein in lanes pairs 2-4 is J-AP3. (F) The reciprocal change on J-PI (D81G) is sufficient to disrupt J-PI homodimerization in favor of obligate J-PI/J-AP3 heterodimerization. (G) T73I disrupts the ability of STS1(G81D) to form homodimers. The PIL protein in lane pair 1 is STS1(T73I), in lane pair 2 it is STS1(T73I + G81D). The AP3L protein in both lane pairs is SI1-C. (H) The G81D mutation does not promote ZMM18 (G81D) homodimerization. SI-C forms homodimers in gel shifts (lower band), but migrates faster than ZMM18/SI-C heterodimer (at arrowhead, top band in lane pairs 1 & 2). (I) STS1 mutagenesis in all possible combinations. AP3L construct in (F), and (G) is SI1-C. •, SI1-C homodimer band.

A Single Amino Acid Change Can Switch the Maize PIL Protein STS1 from an Obligate Heterodimer to a Homodimer

To determine precisely which I-domain amino acid(s) mediate differential dimerization activity between STS1 and J-PI, we compared their I-domain sequences. Four amino acids differ between the STS1 and J-PI I-domains: S63T, S65P, G81D, and E82D (all residue numbers refer to STS1, STS1 residue written first). We mutated each amino acid in STS1 to the J-PI variant individually and in all possible combinations, and tested these mutagenized STS1 variants for dimerization activity (fig. 2 and supplementary table S1, Supplementary Material online). A single amino acid change, from glycine to an aspartic acid at position 81 (G81D) was sufficient to confer the ability to homodimerize on STS1 (fig. 2D-3). STS1 could be induced to bind DNA as a homodimer when G81D was changed alone, in combination with S60P and/or S63T, and when all four amino acids were changed (fig. 2E-3 and I). Furthermore, when the reciprocal change (D81G) was made to J-PI, J-PI (D81G) no longer homodimerized in gel shifts, but instead formed obligate heterodimers with J-AP3 (fig. 2F-2). Thus, a single amino acid change (G81D) in the I-domain is sufficient to confer the ability to homodimerize on STS1.

Curiously, the BdSTS1 I-domain contains an aspartic acid at position 81, but the BdSTS1 I-domain was not sufficient to confer the ability to homodimerize on STS1 (fig. 2D). This led us to hypothesize that other residues in the BdSTS1 I-domain, in addition to residue 81, influence B-class dimerization patterns. BdSTS1, J-PI, and STS1 differ in that BdSTS1 contains an isoleucine at position 73 (threonine in J-PI and STS1), unique among grass PIL proteins. In order to determine whether the identity of the residue at position 73 had an effect on dimerization, we mutated it in STS1, and in STS1(G81D). STS1(T73I) and STS1(T73I + G81D) both formed obligate heterodimers with AP3L proteins (fig. 2G). Thus, it is only in combination with the precise background protein sequence of STS1 that G81D is sufficient to confer the ability to homodimerize on STS1.

Similarly, the ability of G81D alone to influence dimerization is specific to STS1, and does not extend to the maize PI-1 protein ZMM18. All maize PIL proteins contain a glycine at position 81, however ZMM18 (G81D) still bound DNA as an obligate heterodimer with SI1 in gel shifts, albeit somewhat weaker than with the unmutagenized proteins (compare top band in fig. 2H-1 and H-2). Taken together with our domain swap experiments, this demonstrates that the affect of G81D on B-class protein dimerization is context-dependent. In addition, these results show that the molecular determinants of homodimerization versus obligate heterodimerization in STS1 and ZMM18 are distinct. Along with our ancestral state reconstructions, this indicates that STS1 obligate heterodimerization evolved independently from the obligate heterodimerization characteristic of the maize PI-1 proteins.

Obligate Heterodimerization of STS1 with SILKY1 Evolved Very Recently

The question remained whether obligate heterodimerization evolved along the branch leading to STS1 (maize), or whether obligate PIL/AP3L heterodimerization was characteristic of B-class proteins from a number of taxa, more closely related to maize than what we included in our assays. The glycine at position 81 that mediates STS1 hetero- versus homodimerization is present in all maize PIL proteins, but not in any other PIL proteins in our data sets (supplementary figs. S3 and S5, Supplementary Material online). Although this result suggested that obligate heterodimerization evolved recently, our sampling did not allow us to assess exactly when obligate heterodimerization arose in the PI-2 clade. To address this question, we took advantage of the extensive genetic and genomic resources in maize. The maize HapMap2 project characterized genetic variation across 60 improved maize lines, 23 maize landraces, 19 wild relatives (teosintes), and 1 representative of the sister genus, Tripsacum dactyloides (Chia et al. 2012). The nucleotide change that ultimately results in a transition from D (Asp) to G (Gly) at position 81 in STS1 was characterized by the HapMap2 project (G/A, Chr3, position 171,472,161, AGPv3). This causative single nucleotide polymorphism (SNP) is one of four STS1 missense polymorphisms in the HapMap2 data set, but the other three are all toward the C-terminus of the K domain, and not likely contributors to hetero- versus homodimerization. The G81 allele (G at nucleotide position 241) was fixed in 78/78 maize lines genotyped at this position. In teosinte the D81 allele (A at nucleotide position 241) was found in the heterozygous condition in one line, whereas the remaining 16 of 17 teosinte lines genotyped at this position had the G81 allele (G). The tested accession of T. dactyloides was homozygous for the D81 allele (A). These results indicate that the D81G change that contributed to the evolution of obligate heterodimerization of STS1 occurred along the branch leading to maize and teosinte, and became fixed in maize, possibly during the bottleneck of domestication. Thus, phylogenetic, biochemical, and population genetic data combined all indicate that obligate heterodimerization of STS1 evolved very recently, in large part because of a shift from aspartic acid to glycine at residue 81.

Positive Selection Acting on PI-2 Residues Predicted to Be Involved in PPIs

Gene duplications, such as the duplication that led to the PI-1 and PI-2 gene clades, are hypothesized to precede gene neo- or subfunctionalization in some cases (Lynch et al. 2001). Neo- or subfunctionalization may leave a signature of codon diversification or directional evolution that can be detected using computational tests (Yang and Bielawski 2000; Kosakovsky Pond and Frost 2005b). In the case of the evolution of PIL dimerization, we expected short bursts of diversifying or directional selection acting on individual branches and sites, rather than persistent, diversifying selection (e.g., Hughes and Nei 1988; Renner and Specht 2012). In particular, we hypothesized that positive selection might have acted along the branch leading to the PI-2 clade of genes in the grasses, or along the branch leading to node 6 (fig. 1L). We employed multiple tests to detect either periodic diversifying (branch-sites random effects likelihood [REL], branch-sites Bayes Empirical Bayes [BEB] tests) or directional selection acting on specific branches and sites (model of episodic directional selection [MEDS]), or directional selection acting on sites in a phylogeny (directional evolution of protein sequence [DEPS] and FUBAR approach to directional evolution [FADE]) (Pond et al. 2005; Kosakovsky Pond and Frost 2005a; Yang 2007; Kosakovsky Pond et al. 2008, 2011; Murrell et al. 2012).

We identified residues under either directional or diversifying selection acting on the branch leading to the PI-2 clade, or within the PI-2 clade, in all but one test (branch-sites REL) (fig. 2M and supplementary fig. S5 and table S2, Supplementary Material online). Our results were robust to differing sampling and multiple runs. Our selection test results likely differ from previous reports (Mondragon-Palomino et al. 2009; Wei and Ge 2011) because of different sampling, different tests employed, and differing model choices (Zhang et al. 2005). We focused on six sites identified as experiencing positive selection, particularly directional selection, by at least two tests (table 1). Five of these six sites occur in interaction motifs (fig. 3A), previously identified as key mediators of Arabidopsis MADS-box PPIs (van Dijk et al. 2010). Thus, we detected a signal of positive selection acting on amino acid residues in PIL proteins, preferentially located in motifs known to influence MADS-box PPI specificity, associated with the gene duplication event that led to the PI-1 and PI-2 gene clades.

Table 1.

Amino Acid Residues under Selection, as Detected Using Multiple Methods.

Residue (STS1)a	BEBb	MEDSb (aa)	DEPSb (aa)	FADEb (aa)	Domain	PPIc	Motif
47(V)		0.00089 (V)	489 (V)	1 (V)	M	Dimerization	11A, 1A, 2B
63 (S)			1792.2 (S)	1 (S)	I	N/A	8A
65(S)		0.00167 (S)	102.6 (S)	0.99 (S)	I	N/A	8A
123(M)	0.89–0.99	0.00871 (M)			K	None predicted	10A, 8B, 3A
135(L)		0.00087 (V)	192.2 (V)		K	Dimerization	9A
147(V)	0.94–0.99		172.2 (V)	0.99 (V)	K	Tetramerization	None

aPosition and amino acid identity are in parentheses all for STS1.

bMEDS analysis, P values are shown. DEPS analysis, Bayes Factors are shown. Bayes Empirical Bayes and FADE analyses, posterior probabilities are shown. Target amino acids in tests for directional selection are shown in parentheses.

cPredicted to contact another chain in a PPI in homology models. Residues 63 and 65 are outside of regions that can be reliably modeled.

F

Positively selected residues are preferentially clustered in interaction specificity motifs, and the identity of residue 47 can affect PIL dimerization. (A) Residues predicted to be under positive selection are preferentially clustered in interaction motifs (Van Dijk et al, 2010), indicated by black dotted lines. The position of G81, key in mediating STS1 dimerization, is shown in blue. (B) Homology model of a dimer of proteins from node 5. The highly conserved valine at position 43 is shown in green, the more variable residue at position 47 (valine or isoleucine) is pink. (C) The isoleucine at position 47 in the node 5 ancestral protein is close enough to interact with val43 in B-class dimerization. (D) The identity of residue 47 can effect PIL dimerization. When the STS1 I-domain is combined with I47V mutagenesis, BdSTS1 homodimerization is disrupted (lane 1-1). The chimeric protein can still form heterodimers with Brachypodium AP3 (lane 1-2) and SI1-C (lane 1-3). BdSTS1 homodimerization is also disrupted if only the I47V and D81G changes are made (lane 2-1). BdSTS1(I47V + D81G) forms obligate heterodimers with SI1-C. Arrowhead points to artIfact band in the lysate control. (E) Conservation (ConSurf) scores of PIL proteins from the Poales, shown on a model of MEF2. Scores range from 0 to 9. u, undefined. Green square = conserved residue 43, Pink square = variable residue 47.

To address whether any residues under positive selection were likely involved in PPIs, we used homology modeling. No crystal structures of full-length MIKC MADS-box protein have been reported; however, crystal structures of (nonplant) MADS-domain dimers and K-domain tetramers from SEP3 have been determined (Pellegrini et al. 1995; Tan and Richmond 1998; Santelli and Richmond 2000; Puranik et al. 2014). We used the SWISS-MODEL server to identify the best templates for modeling, and built MADS-domain and K-domain homology models for STS1 and the predicted ancestral sequences at nodes 2, 4, and 6 (Arnold et al. 2006; Biasini et al. 2014).

Four of the six positively selected residues identified by more than one test were in protein regions that could be modeled with high confidence, and three of those four were predicted to be involved in PPIs (table 1). The closest homolog with resolved MADS domain dimer crystal structures is mammalian MEF2 (55.88% homology). We used two independent MEF2 crystal structures in our homology modeling experiments (Santelli and Richmond 2000; Jayathilaka et al. 2012;). Our MADS-domain homology models predict that residue 47 is one site of interchain interaction in PI-2 MADS dimer formation. The amino acid at position 47 is consistently involved in a hydrophobic interaction with a highly conserved valine at position 43 of the other monomer (fig. 3B and ). Thus, the identity of the residue under selection at position 47 might affect B-class PPIs. Similarly, when we used a recent SEP3 crystal structure (28–35% homology) (Puranik et al. 2014) to model the K-domains of both extant and reconstructed ancestral proteins from the grasses, residue 135 was consistently involved in dimerization. In addition, residue 147 was predicted to be involved in tetramerization of the ancestral protein reconstructed at node 4, and of STS1. Thus, three of four positively selected amino acids that could be reliably placed in a structural homology model are likely directly involved in interactions between MADS monomers and dimers. These data, taken together, indicate that the signal of positive selection we detected acting on PIL proteins in the grasses may indeed be associated with the evolution of PPIs.

Altering a Positively Selected MADS-Domain Amino Acid Residue Can Change B-Class Dimerization

To test whether the identity of positively selected residues could affect dimerization in vitro, we performed further site-directed mutagenesis experiments with PI-2 proteins. We chose to investigate residue 47 because it was predicted to be under directional selection, could be modeled reliably, and changed within the PI-2 clade. STS1 has a valine at residue 47 in all genotyped maize and teosinte accessions, and in T. dactyloides (Chia et al. 2012). In contrast, the B. distachyon STS1 ortholog (BdSTS1), and the predicted ancestral sequences at nodes 2, 3, 4, 5, and 6 all have an isoleucine at position 47. The STS1 I-domain was not sufficient to disrupt BdSTS1 homodimerization (fig. 1D-1). However, when we changed I47 to valine in addition to the I-domain, this chimeric protein no longer formed homodimers (fig. 3D-1). Similarly, when BdSTS1 residues 47 and 81 were changed to the maize variants (BdSTS1 [I47V + D81G]), the mutagenized protein could no longer form homodimers (fig. 3D-2). Both mutagenized proteins still formed heterodimers with AP3L proteins. Similar to G81D, although the identity of residue 47 was critical in determining homodimerization potential in the context of BdSTS1, a valine at position 47 is not completely linked to obligate heterodimerization. Val47 is found in a number of PIL proteins that form homodimers (e.g., Avena sativa PI-2; supplementary fig. S5, Supplementary Material online). However, our results show that the identity of the positively selected residue at position 47 contributes to the dimerization landscape of grass PIL proteins.

Patterns of Sequence Conservation Differentiate Modules within PIL Proteins

Changes to protein-coding regions are thought to have a higher potential for negative pleiotropic effects than changes to noncoding regions (Carroll 2000; Stern 2000). Nonetheless, we found evidence for coding evolution in the Poales, under positive selection, and with demonstrated consequences to PPIs (fig. 3). We hypothesized that substitutions affecting core protein function (i.e., DNA binding) have the potential to severely impact protein function, and therefore may have experienced different selective pressures than substitutions affecting PPIs. To explore variable signatures of selection acting on PIL proteins in the context of protein structure, we used the ConSurf webserver. ConSurf can be used to assess patterns of sequence conservation within an alignment, and transfer that information to a homology model (Landau et al. 2005). ConSurf sequence conservation scores for each position in an alignment range from 1 (extremely variable) to 9 (highly conserved). We found that amino acids potentially important for B-class dimerization specificity and/or under positive selection occurred in weakly to moderately conserved regions of PIL proteins. For example, residue 47, which is under positive selection and involved in variable dimerization, had a relatively low ConSurf score of 4. Furthermore, most amino acids predicted to be critical for DNA binding are deeply conserved in PIL proteins with high Consurf scores between 7 and 9 (fig. 3E). This result underscores the modular architecture characteristic of the MIKC MADS-box transcription factors, and indeed of many proteins (Kaufmann et al. 2005; Hoekstra and Coyne 2007; Bartlett and Whipple 2013). Changes to the DNA-contacting amino acid residues that would presumably be detrimental to core protein function are selected against. Consequently, these residues are more likely to remain invariant over evolutionary time and have high conservation scores. In contrast, the regions of PIL proteins that modify PPIs are under less stringent selection, possibly allowing for shifting interactions over evolutionary time. This modularity and variability in selective pressure means that some classes of structural change have less capacity for negative pleiotropy than others (Lynch and Wagner 2008; Bartlett and Whipple 2013).

J-PI and STS1 Show Similar Rescue of B-Class Mutants

The question remained as to what function, if any, PIL homodimers from the Poales have in planta. We turned to a rapid, heterologous transformation system to assess the function of a PIL homodimer in planta. Specifically, we expressed J-PI under the control of the AP3 promoter in Arabidopsis using the pOp/LhG4 transactivation system (Baroux et al. 2005), and assayed the ability of J-PI to rescue putative null alleles of AP3 and PI. Arabidopsis AP3 expression is confined to whorls 2 and 3 early in development, and is expressed at low levels in adult sepals. Arabidopsis PI expression is similar, except that it is also expressed in whorl 4 early on in floral development (Goto and Meyerowitz 1994; Schmid et al. 2005). When J-PI was expressed in pi-1 mutants, (pi-1; pAP3::Lhg4 ≫ 10-Op::J-PI), first whorl sepals were petaloid (fig. 4C), and second whorl petals appeared normal. Third and fourth whorl organ rescue was more variable. The J-PI rescue lines had a mix of hybrid stamenoid–carpeloid organs, bare filaments, carpels, small abortive stamens, and almost completely normal stamens in the third whorl. The increase in total organ number in J-PI rescue lines was statistically significant (t-test, P < 0.05). J-PI also rescued other pi-1 phenotypes including gynoecium defects and third and fourth whorl fusion defects (fig. 4I and ). J-PI, expressed under the pAP3 promoter, did not rescue ap3-4 mutants (seven independent transgenic lines).

F

Heterologous expression of J-PI homodimers and obligate STS1 heterodimers in Arabidopsis thaliana: (A) Wildtype Arabidopsis flower, (B) pi-1 mutant, (C) J-PI under the AP3 promoter (pAP3::LhG4 ≫ 10-Op::J-PI) showing partial rescue of the pi-1 mutant phenotype. Sepals (marked with *) are petaloid (closest sepal removed). (D) The same J-PI construct under the AP3 promoter (pAP3::LhG4 ≫ 10-Op::J-PI) does not rescue the ap3-4 mutant phenotype. (E, F) Transformants expressing STS1 under the AP1 promoter (pAP1::LhG4 ≫ 10-Op::STS1) exhibit petaloid sepals. Transformants expressing J-PI under the AP1 promoter (pAP1::LhG4 ≫ 10-Op::J-PI) either show petaloid sepals (G), or the complete homeotic transformation of sepals into petals (H). (I, J) Quantification of J-PI rescue of ap3-4 and pi-1 mutants. J-PI, under the AP3 promoter, partially rescued the club-shaped gynoecium phenotype, as well as the fusion and organ number defects typical of the pi-1 mutant. Neither club-shaped gynoecia nor fusion defects were observed in wildtype flowers.

We attained very similar results in a pi-1 and ap3-4 mutant lines expressing STS1 under the AP3 promoter (supplementary fig. S6, Supplementary Material online). Our STS1 pi-1 rescue results recapitulated those of Whipple et al. (2004), where STS1 was also expressed under the Arabidopsis AP3 promoter, and caused petaloid sepal margins. Both the J-PI and the STS1 pi-1 rescue lines strongly resembled those where Arabidopsis AP3 or PI was ectopically expressed in flowers, under the 35S promoter (Krizek and Meyerowitz 1996a; Yang, Xiang, et al. 2003). STS1, expressed under the pAP3 promoter, did not rescue ap3-4 mutants (two independent lines). In summary, both J-PI and STS1 showed similar rescue of the pi-1 mutant phenotype, where they were presumably acting as part of complexes with AP3 (Whipple et al. 2004; Whipple and Schmidt 2006). In addition, neither J-PI nor STS1 conferred B-class function without an AP3 partner as neither J-PI nor STS1 could rescue ap3-4 mutants.

The J-PI Homodimer Can Transform Arabidopsis Sepals into Petals

Altered interactions between MADS-box proteins from snapdragon, in conjunction with variable expression patterns, have the capacity to effect phenotypic change in Arabidopsis (Airoldi et al. 2010). This led us to ask whether the J-PI homodimer, when expressed at high levels in a new domain, could induce novel phenotypes in Arabidopsis, distinct from those induced by STS1. To this end, we again used the pOp/LhG4 transactivation system (Baroux et al. 2005), this time to drive high expression of J-PI and STS1 in developing sepals, using the Arabidopsis AP1 promoter. AP1 is initially expressed throughout the floral meristem, but is only detected (using in situ hybridization) in the first (sepal) and second (petal) whorls by the time sepal primordia become evident (Mandel et al. 1992).

Ectopic expression of STS1 in developing sepals (4/4 lines) resulted in first whorl sepal margins that became slightly petaloid, similar to what was observed under the AP3 promoter (fig. 4E and ). In contrast, ectopic expression of J-PI in Arabidopsis sepals resulted in the complete transformation of the first whorl organs into petals indistinguishable from second whorl petals (6/11 independent transformation lines) (fig. 4H). These homeotically transformed organs resembled sepals ectopically expressing both AP3 and PI (Krizek and Meyerowitz 1996a). Similarly, when a version of AP1 that has strong transactivation potential is ectopically expressed in Arabidopsis sepals, they are also transformed into petals, probably because of the ectopic upregulation of both AP3 and PI (Ng and Yanofsky 2001; Lamb et al. 2002). In 5 of the 11 J-PI lines, sepals resembled those of pAP1 ≫ STS1 transformants with stronger phenotypes. In strong pAP1 ≫ STS1 lines (2/4 independent lines), sepals had petaloid margins but were not completely transformed (fig. 4G). Our results with STS1 were similar to those seen with ectopic expression of Arabidopsis PI, or the Petunia hybrida AP3 gene green petals in all four whorls of the Arabidopsis flower (under 35S, both obligate heterodimers; Halfter et al. 1994; Krizek and Meyerowitz 1996a; Yang, Xiang, et al. 2003). These results show that J-PI, when expressed at a high level in a novel domain, differs from STS1 in its effects on floral development.

Discussion

The PPI landscape of any gene product is a critical component of function. In plant development, there are myriad examples of regulatory PPIs central to the development of individual traits and organs (e.g., Sabatini et al. 2003; Cui et al. 2007; Kanaoka et al. 2008; Lau et al. 2014). Changing these interactions over evolutionary time is one hypothesized way to achieve regulatory diversity, without negatively affecting core protein function (Lynch and Wagner 2008; Tuch et al. 2008). We identified a pattern of unexpected evolutionary lability in B-class MADS-box PPIs in the Poales. Within the grasses, obligate AP3L/PIL heterodimerization has evolved from PIL homodimerization at least twice. The evolution of obligate heterodimerization of the maize protein STS1 appears to have occurred very recently: The causative SNP was fixed during domestication. We detected a signal of positive selection acting on residues predicted to be involved in PPIs, one of which has an effect on PIL protein dimerization. We also showed that heterologous expression of Poalean MADS-box proteins, which differ in their PPI spectra, can affect floral form in Arabidopsis.

Shifting B-Class PPIs and Evolutionary Change in the Poales

Given our sampling in an extremely speciose family and order, the examples of shifts described here are almost certainly an underestimate. This is in stark contrast to what has been reported in the core eudicots, where B-class homodimers are rare (summarized in Melzer et al. 2014). In the monocots outside of the Poales, in the early-diverging angiosperms, and in the noncore eudicots there appears to be more lability and more frequent B-class homodimerization (summarized in Melzer et al. 2014), but it is hard to assess the degree of change without more fine-grained sampling. Obligate B-class heterodimerization has been connected to the evolution of the highly canalized flowers typical of the core eudicots, where there is almost no intergradation between floral organs of different types (Winter et al. 2002; Lenser et al. 2009; Specht and Bartlett 2009; Melzer et al. 2014). The evolutionary lability we have found in the Poales, as well as the presence and maintenance of PIL homodimerization in lineages like the grasses, which on the whole possess highly canalized flowers (Kellogg 2015), argues against this idea. Instead, our results suggest that obligate B-class heterodimerization in particular is not necessary for the tight maintenance of separate floral whorls, and may serve another function, or indeed no function at all, particularly in the Poales.

Within the Poales, there are suggestive associations between dimerization state and floral and inflorescence morphology. In nongrass Poaleans, we observed obligate heterodimerization more often in taxa that are wind pollinated. Ecdeiocolea, Elegia, and Typha are all wind pollinated, and all show obligate B-class heterodimerization. In contrast, Flagellaria, Joinvillea, Tillandsia, and Xyris are all probably animal pollinated, and all of their PIL proteins can form homodimers (pollination states summarized in Givnish et al. 2010). The morphological wind pollination “syndrome” often includes a reduced perianth, anthers on long flexuose filaments, and/or unisexual flowers (Linder 1998; Culley et al. 2002). In addition, wind pollination evolves in a correlated way with traits that could be linked to B-function, including floral showiness (often associated with perianth characteristics) and floral sexuality (Friedman and Barrett 2008). Thus, some of the morphological traits thought to favor wind pollination include modifications to second and third whorl organs, which are often specified by B-class gene function. Within the grasses, PIL homodimerization may have emerged coincident with the origin of lodicules and the specialized grass spikelet (node 4), or coincident with a major diversification event that led to the hyperdiverse BEP and PACMAD grass clades (node 6). Although our sampling is far from complete, and it is unwarranted to connect cause and effect, these patterns are intriguing and may be indicative of a connection between morphology and B-class dimerization in the Poales.

Just as the shifting PPIs we have uncovered may be connected with morphological evolution in the Poales, they may also underlie phenotypic conservation. For example, in “developmental systems drift,” the regulatory architecture underlying a particular phenotype may change considerably, but the phenotype itself may remain stable (True and Haag 2001). An excellent example of this phenomenon is the nematode vulva. Although formed from the same three cells in both Caenorhabditis elegans and Pristionchus pacificus, vulva development is controlled by divergent genetic pathways in the two species (Sommer 2012). In the Poales, the regulatory architecture of floral development may differ drastically between species, but floral development may ultimately proceed along similar lines across the order. In an allied concept, floral development in the Poales may be robust to underlying genotypic change, and shifting B-class PPIs. Robustness allows for the accumulation of cryptic genetic variation, which in some cases can serve as raw material for natural selection, resulting ultimately in rapid phenotypic change (Wagner 2011, 2005; Paaby and Rockman 2014). For the B-class genes in particular, robustness has been suggested to be derived from partial to complete redundancy between duplicates in a genome, particular PPIs, and regulatory relationships between B-class genes (Lenser et al. 2009; Geuten et al. 2011). In the Poales, the lability we have uncovered may represent a small fraction of the molecular divergence underlying robust floral developmental pathways.

PI Gene Function Has Changed in the Grasses, Concomitant with Shifting PPIs

The signature of positive selection we detected acting on residues that mediate PPIs of one PIL paralog in the grasses may indicate sub- or neofunctionalization of the two grass PIL genes. There is evidence for functional differentiation of the two PI homologs in rice (Kang et al. 1998; Kyozuka et al. 2000; Yadav et al. 2007), which differ in their ability to homodimerize (fig. 2L and supplementary fig. S1, Supplementary Material online). The two rice PIL homologs, OsMADS2 and OsMADS4 are expressed in essentially the same domains, but OsMADS2 (or the OsMADS2/SPW1 heterodimer) has a particular role in second whorl organ (lodicule) development (Kang et al. 1998; Nagasawa et al. 2003; Prasad and Vijayraghavan 2003; Yao et al. 2008). Gene function has also diverged within the PI-2 clade, again associated with changing PPIs. In contrast to rice OsMADS2, STS1 function is not limited to lodicule identity (Bartlett et al. 2015). OsMADS4 and STS1 have very similar expression patterns (Kyozuka et al. 2000; Munster et al. 2001; Whipple et al. 2004, 2007). Thus, the functional differences between OsMADS4 and STS1 may be due to altered PPIs. In addition, the regulatory relationships between rice B-class genes are distinct from those seen between the B-class genes of maize. In maize, STS1 shows no redundancy with ZMM18 or its close paralog, ZMM29, likely because ZMM18 and ZMM29 require the STS1/SI1 heterodimer to initiate their expression (Bartlett et al. 2015). In contrast, knockdowns of rice OsMADS2 and OsMADS4 do not affect expression of other B-class genes (Prasad and Vijayraghavan 2003; Yao et al. 2008). This suggests that STS1 and OsMADS2 have diverged not only in dimerization potential and homeotic function but also in their roles within the floral development gene regulatory landscape. Functional differentiation of B-class proteins is associated with altered PPIs in a number of lineages (Kang et al. 1998; Prasad and Vijayraghavan 2003; Vandenbussche et al. 2004; de Martino et al. 2006; Geuten and Irish 2010). In the grasses, we have found a similar relationship between altered B-class PPIs, and changing gene function.

Changing PPIs and Functional Change

Evolutionary changes to core protein function, like altered DNA-contacting or active-site residues, have the potential to eliminate protein function. Regulatory proteins often have multiple roles in different developmental contexts, and altering DNA-binding has the potential to change all of these roles (negative pleiotropy). Thus, coding changes to DNA-binding domains are thought to be, and are in our analyses, under strong negative selection (Stern and Orgogozo 2008). In contrast, evolutionary changes in residues affecting PPIs may be under less stringent selection and allow for the evolution of new protein function (Lynch et al. 2008; Tuch et al. 2008). Consistent with this hypothesis, we found evolutionary change in B-class dimerization across the Poalean phylogeny (fig. 2L). Residues in Poalean PIL proteins likely directly contacting DNA are deeply conserved, whereas residues involved in dimerization show more evolutionary variability (fig. 3E). Not all coding changes have the same potential for negative pleiotropy or for affecting functional divergence. Changing PPIs, in particular, may be one class of structural change that allows for novel protein function, while avoiding some of the pitfalls inherent to changing deeply conserved protein function, like DNA-binding specificity.

The J-PI Homodimer Differs from STS1 in Its Effects on Floral Development in Arabidopsis

Although the J-PI homodimer could not rescue the ap3-4 mutant, ectopic expression of J-PI in sepals (using the AP1 promoter) was sufficient to promote the homeotic transformation of sepals to petals. In contrast, we never observed this complete transformation with STS1, which forms obligate heterodimers with AP3L proteins. Instead, ectopic STS1 expression resulted in slight petaloidy of the sepals, similar to that observed when PI is ectopically expressed in Arabidopsis thaliana sepals (fig. 5; Krizek and Meyerowitz 1996a; Yang, Xiang, et al. 2003). In contrast to our results, the ectopic expression of a homodimerizing PIL protein from Lilium longiflorum resulted in slight petaloidy at sepal margins, rather than complete transformation of sepals into petals (Chen et al. 2011). These differing results may be due to different promoters used (pAP1 vs. 35S), or the homeotic transformation we observed may not be the direct result of hetero-versus homodimerization. In addition to differential dimerization patterns, J-PI and STS1 may bind different DNA sequences, and/or J-PI and STS1 may differ in their ability to interact with other Arabidopsis MADS-box proteins, including AP3. As our homology models suggest that the J-PI and STS1 residues most likely to contact the DNA are identical, we favor the hypothesis that any potential differences in DNA-binding activity between J-PI and STS1 are because of divergent PPIs or protein–cofactor interactions. These results support our overall hypothesis that differing PPIs, in conjunction with a novel expression domain, can alter protein function and organismal phenotype (fig. 4).

F

A model of STS1 and J-PI action in Arabidopsis flowers expressing either STS1 or J-PI under the AP1 promoter. (A, C) STS1, either alone or in complex with other Arabidopsis proteins, does not bind the AP3 promoter or activate high levels of AP3 or PI expression in sepals. (B, D) J-PI, as a homodimer and/or in complex with other Arabidopsis proteins, can bind the AP3 promoter and, in turn, activate high expression of both AP3 and PI. This ectopic B-class expression triggers the homeotic transformation of the sepals into petals in J-PI expression lines, but not STS1 expression lines.

We integrated our results into a model to explain J-PI’s function in Arabidopsis (fig. 5). The J-PI homodimer does not have the capacity for full homeotic function and cannot rescue the ap3 mutant (fig. 4D). Although STS1 cannot bind the AP3 promoter without an AP3 partner, the J-PI homodimer can bind the AP3 promoter in vitro (the probe used in our EMSAs is derived from pAP3). This J-PI binding may upregulate AP3 expression in the sepals of the pAP1≫ J-PI lines. This ectopic AP3 can form heterodimers with J-PI (Whipple and Schmidt 2006), or PI itself, which is weakly expressed in older sepals (Schmid et al. 2005), and go on to homeotically transform the sepals into petals. Thus, a novel protein complex that includes the J-PI homodimer, expressed in a new domain, could have the capacity to affect floral morphology in significant ways.

Coding and Noncoding Changes Both Impact the Evolution of Gene Function

It was only in the context of high expression in Arabidopsis sepals that we saw a difference between heterologous expression of J-PI and STS1 (under the AP1 promoter). This interplay between PPIs and gene expression patterns has been demonstrated in studies of both evolutionary and developmental biology (Melzer and Theißen 2009; Smaczniak, Immink, Muino, et al. 2012; Davies and Bergmann 2014; Jetha et al. 2014; Furumizu et al. 2015). For example, altered PPIs overlaid on existing interactor expression domains contribute to the functional divergence of snapdragon C-class function (Airoldi et al. 2010). A changing PPI does not occur in isolation. Changes to PPIs are always overlaid on the expression pattern of the protein in question, and the expression patterns of all of its interactors. Although this layering of complexity can make dissecting the functional effects of evolutionary change challenging, it also provides a mechanism for amplifying small changes that might otherwise be neutral. The J-PI homodimer on its own is not sufficient to confer homeotic B-class function in Arabidopsis, but a novel protein complex in a novel domain can effect phenotypic change.

Conclusion

We have shown evolutionary lability in PIL dimerization in the Poales, associated with functional differentiation and positive selection. We identified individual amino acids that change across the tree and mediate PIL dimerization, and we have shown that altered PPIs have the potential to effect phenotypic change under certain circumstances. The precise role of changing B-class PPIs in the regulation of floral development and the evolution of morphological diversity in the Poales remains an open question. Our results highlight the need for functional characterization of B-class hetero- versus homodimers in the organisms in which they occur.

Materials and Methods

Assessment of PPIs

Full-length PIL genes were amplified from taxa spanning the Poales using a combination of primers in conserved gene regions, primers in conserved upstream noncoding sequences, and 3′-RACE (Rapid Amplification of cDNA Ends) (supplementary table S3, Supplementary Material online). The full-length cDNAs were subcloned into the expression vector pSPUTK for use in EMSAs. EMSAs were performed as previously described (Whipple et al. 2007). Clones for Y2H assays were generated by amplifying full length or truncated cDNAs and inserting into the pENTR d-TOPO entry vector (Life Technologies), followed by recombination into GAL4 DNA binding domain (pGBKT7) or activation domain (pGADT7) converted to Gateway destination vectors. Correct clones were transformed into the PJ69-4a (pGBKT7) and pJ69-4∝ (pGADT7) yeast strains, and mated to generate diploid yeast. For BiFC assays, full-length cDNAs lacking a stop codon were cloned into a pBJ36 derivative containing either the N- or the C-terminal of YFP, then subcloned into the Not1 site of the pMLBART27 binary vector. BiFC experiments were performed as described in Gallavotti et al. (2011), and imaged 60 h after initial infection using a Olympus FluoView FV1000 confocal laser scanning microscope. Constructs with only the N- or the C-terminal of YFP were coinfected as negative controls.

Phylogenetic Analysis, Ancestral State Reconstructions, and Tests for Selection

Full-length PIL coding sequence alignments were generated using MUSCLE, as implemented in Geneious, and refined by hand. Two matrices were generated: One that had all full-length PI-like sequence data for the monocots and select outgroups, and another that included only those sequences for which dimerization data existed. Sequences we generated have been deposited on GenBank KU740296-KU740323. Model and partitioning scheme selection was performed using PartitionFinder. Bayesian reconstruction of phylogeny was conducted using MrBayes v3.2.1. Maximum-likelihood reconstruction was conducted using RAxML, as implemented on the CIPRES server. Ancestral sequence reconstruction was performed using the DataMonkey webserver, and Lazarus. Ancestral dimerization states were reconstructed using both maximum likelihood and parsimony in Mesquite v2.75. The tests for selection included branch-sites REL (Kosakovsky Pond et al. 2011), MEDS, fixed effects episodic diversifying selection (Murrell et al. 2012), and DEPS (Kosakovsky Pond et al. 2008) tests in Hyphy v2.2 (Pond et al. 2005) or using the DataMonkey webserver (Kosakovsky Pond and Frost 2005a), and the branch-sites test in PAML v4.8 (Yang 2007).

Domain Swaps and Site-Directed Mutagenesis

Chimeric Joinvillea-PI (J-PI), Zea PI-2 (STS1) proteins were constructed using overlap extension followed by high-fidelity polymerase chain reaction (PCR) (Warrens et al. 1997; Atanassov et al. 2009). All primers used are listed in supplementary table S3, Supplementary Material online. Initial PCR reactions contained 0.75 units of Taq polymerase (NEB), 10% dimethyl sulfoxide (DMSO), forward and reverse primers (0.5 μM each), dNTPs (0.25 mM each), MgCl2 (1.7 mM), and 50 ng of template plasmid DNA in a 1× buffered solution. Individual amplified domains were gel extracted and purified using the Promega Wizard SV Gel Clean-Up system and quantified using a NanoDrop spectrophotometer (ThermoFisher) before overlap extension. Overlap extension reactions contained 0.6 units of Phusion polymerase (ThermoFisher), dNTPs (0.2 mM each), and template DNA (approximately 270 ng of PCR product for each domain). Following ten cycles of overlap extension (cycling conditions: 98 °C for 10 s, 60 °C for 30 s, 72 °C for 1 min for ten cycles), the polymerase was replenished and amplification primers added to a final concentration of 0.4 μM. Chimeric DNA fragments were amplified in a final touchdown PCR reaction. Resulting PCR products were cloned into pJET1.2 for Sanger sequencing and then subcloned into pSPUTK.

Site-directed mutagenesis was performed using the Quikchange II or the Quikchange II XL kits from Agilent using primers designed using Agilent’s online primer design tool (supplementary table S3, Supplementary Material online; http://www.genomics.agilent.com/primerDesignProgram.jsp, last accessed March 3, 2015). Mutagenesis was conducted on 50 ng of plasmid DNA, for ten cycles, according to manufacturers’ instructions. Correct clones were verified using PCR and Sanger sequencing.

Homology Modeling and Assessment of Sequence Conservation

The SWISS-MODEL server was used to search the protein databank for appropriate structural templates, generate pairwise alignments, and model B-class protein structure (Arnold et al. 2006; Biasini et al. 2014). B-class proteins from the Poales were modeled against protein databank templates (1egw and 3mu6) for comparative analyses (Berman et al. 2000; Santelli and Richmond 2000; Jayathilaka et al. 2012). A ConSurf analysis was conducted using an alignment of all available PI-like genes from the monocots (Landau et al. 2005).

Arabidopsis Transformations

cDNA sequences of Joinvillea-PI and STS1, both under the synthetic 10-Op promoter (Baroux et al. 2005), were cloned into pBJ36. The promoter-gene constructs were then subcloned into the Not1 site of the pMLBART27 binary vector. Clones were transferred into Agrobacterium tumefasciens, which was subsequently used to transform A. thaliana Ler containing the synthetic RNA polymerase Lhg4 under either the AP1 or AP3 promoter from Arabidopsis. The resulting Arabidopsis transformants were crossed to either the ap3-4 (Jack et al. 1992) or pi-1 (Goto and Meyerowitz 1994) mutants and F2–F5 progeny from these crosses were characterized.

Supplementary Material

Supplementary figures S1–S6 and tables S1–S3 are available at Molecular Biology and Evolution online (http://www.mbe.oxfordjournals.org/).