In the study of the molecular changes underlying adaptive evolution, there is debate as to whether regulatory or structural changes are of greater importance. Regulatory changes, especially those affecting where and when a transcriptional regulator is expressed, are thought to predominate. Structural changes are thought to have a higher degree of negative pleiotropy, and are probably not tolerated to the same degree as regulatory changes (Carroll, 2000, 2005, 2008; Stern, 2000). Despite this prevailing view, structural changes have been shown to have had a noteworthy role in the evolution of some key adaptive traits (Hoekstra and Coyne, 2007). In the evolution of plant form and function in particular, examples of both regulatory (Arnaud et al., 2011) and structural (Airoldi et al., 2010) changes exist. With time and more data, the argument may be resolved, but the point that every trait is different may be key (Wessinger and Rausher, 2012). In all likelihood, in most cases there is no single quantitative trait nucleotide (QTN), but rather a collection of myriad small changes, both regulatory and structural, that have contributed to the evolution of a novel phenotype (Rockman, 2012).

Regardless of which class of changes predominates, both regulatory and structural mutations have happened through the course of evolution. We have chosen to review those cases where structural mutations have had demonstrably functional consequences. Interpreting a mutation as either regulatory or structural is not always straightforward. We use the definition proposed by Hoekstra and Coyne (2007), with some modifications. They propose that mutations that occur in the coding sequences of genes are structural, and all other mutations, including those that occur in introns, are regulatory. This definition includes nonsense null mutations, altered miRNA-binding sites, and silent mutations affecting transcription and translation dynamics as “structural” (Hoekstra and Coyne, 2007). We prefer a more narrow definition of “structural mutation,” and include only those examples where amino acid sequence is changed and protein function is not completely lost. This circumscription of structural mutations thus includes mostly missense mutations, but also insertions and deletions, frameshifts, and premature stop codons that produce proteins with altered functions. We have chosen this definition out of expediency. Compelling arguments exist for putting all excluded mutations (e.g., miRNA-binding site mutations) back into the set of structural mutations, and then for taking them right back out again. For the purpose of investigating the evolution of protein function, we feel that our narrow definition best frames the discussion.

Our review focuses on those cases where protein function has been altered, in turn altering some aspect of phenotype. It is important to highlight that many of the phenotypes we discuss may or may not be adaptive, but that is not the focus of this review. It is not trivial to unambiguously determine the molecular cause of a phenotypic adaptation, or even to confirm some phenotype as adaptive (Barrett and Hoekstra, 2011). Moreover, there are few studies that have explicitly investigated the quantitative trait loci (QTL) underlying natural variation in an evolutionary framework, and as a consequence it is hard to determine their adaptive significance. A more widespread genotype might hint at some adaptive value, but for many of the examples we cite the responsible protein change is only found in a single accession where it may be deleterious and/or of short duration. These isolated QTL are not inherently less interesting, however, because they reveal the scope of molecular diversity to be found in natural environments, diversity that selection may ultimately act upon.

Another important preliminary consideration is that protein diversification through deep time can only be discussed in a framework of gene birth. In plants in particular, and in eukaryotes in general, a major source of new genes is gene duplication. Most gene families have expanded considerably through gene duplication and divergence, and often these expansions show lineage-specific patterns (Flagel and Wendel, 2009). The new gene duplicates are thought to have one of three fates. Formally, duplicate genes may divide up the functions of the progenitor gene between them (subfunctionalization), one of the duplicates may gain an entirely new function (neofunctionalization), or one of the duplicates may decay into a non-functional pseudogene (Ohno, 1970; Lynch et al., 2001). These categories are often difficult to assign, but where they are relevant, most of the examples we will discuss are of neofunctionalization.

In our review of the literature, we found that functional protein changes, the result of underlying structural mutations, fell into six broad, non-mutually exclusive categories. We have divided up our discussion according to these categories: (I) altered active or binding sites; (II) altered protein–protein interactions; (III) altered domain content; (IV) altered activity as a transcriptional activator or repressor; (V) altered protein stability; or (VI) hypomorphic and hypermorphic alleles (Figure 1 and Table 1).

Amino acid replacement in the active sites of enzymes, or the DNA-binding sites of transcription factors, is perhaps the most easily understood mechanism of protein evolution. Changes to the core functional domain of a protein, either through gradual replacement of many amino acids over the course of time (Zhao et al., 2008), or through the replacement of a few key residues (Greenhagen et al., 2006; O’Maille et al., 2008), has the potential to generate novel protein function. Active and binding site changes also have the greatest potential to be deleterious if they destroy a protein’s primary function (Carroll, 2008). Despite this potential for negative effects, numerous examples (outlined below) have been uncovered where active and binding site evolution has been tolerated and led to neofunctionalization.

Protein–protein interactions are of prime importance in plant development. There are many examples of particular interactions regulating key developmental and physiological processes (e.g., Riechmann et al., 1996a; Kim et al., 1997; Cui et al., 2007). Altered PPIs may be one way to generate functional diversity without negatively affecting core protein function. The DNA-binding domain of a protein may stay intact, but an interaction domain may change to interact with a new partner, perhaps expressed in a discrete domain. In this way novel functions can emerge, while the protein’s original functions are preserved (Lynch and Wagner, 2008). Despite this compelling argument for investigating PPI evolution, and despite their integral role in development, few studies have tackled PPIs in an evolutionary framework.

One interaction that has been studied in an evolutionary context occurs between the gibberellin phytohormones (GA), GID1-like proteins (GLP1), and the DELLA transcriptional repressors. In A. thaliana DELLA proteins, as part of GLP1–GA–DELLA complexes, are polyubiquitinated and recruited to the 26S proteasome for destruction, releasing DELLA targets from repression (reviewed in Sun, 2011). The GLP1–GA–DELLA interaction is deeply conserved in angiosperms (Sun, 2011), and appears to have been acquired gradually through the course of land plant evolution (Yasumura et al., 2007). The results of mutant analyses and heterologous transformation experiments suggest that DELLA’s acquired their characteristic growth-repressive function after the divergence of the lycophytes from the rest of the land plants, perhaps through cis-regulatory changes. The GA-stimulated GLP1–DELLA interaction appears to have arisen after the divergence of the bryophytes from the remainder of the land plants, probably through structural alterations to DELLA proteins. Thus DELLA protein changes that facilitate the GLP1–DELLA interaction, together with the evolution of an altered GA response, allowed for the emergence of the GLP1–GA–DELLA module characteristic of flowering plants (Yasumura et al., 2007).

In the study of plant development, the network of interactions between the ABC(E) MADS box proteins has been extensively investigated. The ABC(E) class MADS box genes, and the single non-MADS A class gene AP2 (APETALA2), control floral organ identity in a combinatorial manner. In Arabidopsis and Antirrhinum the A class genes control sepal identity. The A and B class genes together control petal identity, B and C class genes together confer stamen identity, and the C class genes control carpel identity (Coen and Meyerowitz, 1991). The E class genes are needed in all four whorls of the flower for proper organ identity specification (Pelaz et al., 2000; Honma and Goto, 2001). The ABC(E) MADS box proteins are known to dimerize, but probably function as part of tetramers (“floral quartets”). These proteins have four domains: the DNA-binding MADS domain, an Intervening domain (I), a keratin-like coiled coil (K), and a disordered C-terminal domain. The I, K, and C-domains have been implicated in mediating PPIs amongst MADS box proteins (reviewed in Immink et al., 2010).

There are a few examples where novel mutant phenotypes are probably caused by disrupted PPIs of MADS box transcription factors. The fast neutron induced seirena mutant of the California poppy, Eschscholzia californica (Ranunculaceae), shows a B class mutant phenotype, and may result from compromised interactions between the B class, C class, and E class MADS box proteins. Site-directed mutagenesis experiments revealed that the B–C–E interaction in Eschscholzia may be mediated by the PISTILLATA (PI) motif, missing from sei-1 mutant protein (Lange et al., 2013). The PI motif is conserved, but not universally present, in PI homologs. Although the PI motif may well have a role in MADS box complex formation wherever it is found, distinct interaction motifs may have evolved convergently in lineages where the PI motif is missing or altered, but higher order complexes still form (Lange et al., 2013). The double-flowered mutant phenotype of an ornamental cultivar of Thalictrum thalictroides (Ranunculaceae) may also be the result of disrupted PPIs between C and E class MADS box proteins (Galimba et al., 2012).

APETALA3 (AP3)-like and PI-like genes comprise the two main lineages of B class MADS box genes. In all core eudicots investigated thus far, B class proteins bind DNA as obligate heterodimers: AP3-like proteins cannot bind DNA without PI-like proteins and vice-versa (Riechmann et al., 1996a,b). The AP3–PI heterodimer in Arabidopsis goes on to autoregulate late PI and AP3 expression (Honma and Goto, 2000). This obligate heterodimer relationship is uncommon in the large MADS box gene family (Riechmann et al., 1996a), and obligate heterodimerization coupled with autoregulation is a rare, if not unique regulatory mechanism. All angiosperms investigated thus far have at least one AP3-like and one PI-like gene, and AP3-like and PI-like proteins bind DNA as obligate heterodimers in distantly related angiosperms, including the grass Zea mays (Vandenbussche et al., 2004; Whipple et al., 2004; Drea et al., 2007; Kramer et al., 2007). The only characterized B class proteins isolated from a gymnosperm thus far (the Gnetalean Gnetum gnemon) bind DNA as homodimers (Winter et al., 2002; Wang et al., 2010). These data, taken together, suggested that the obligate heterodimerization relationship evolved from a homodimerizing ancestor shortly after the duplication event that led to the AP3 and PI gene lineages, and prior to the diversification of the angiosperms. However, PI homologs from Lilium were found to be capable of homodimerizing and heterodimerizing (Winter et al., 2002), but with no other data points, it was unclear whether this was an autapomorphy or indicative of a broader evolutionary trend. The single PI-like protein (J-PI) in Joinvillea, a close grass relative, can homodimerize (Whipple and Schmidt, 2006). PI-like homodimerization has also been observed in Chloranthus (Chloranthaceae; Li et al., 2005) and Eschscholzia (Lange et al., 2013). Together with the data from Lilium, these data imply the intriguing convergent evolution of obligate heterodimerization both in the monocots and in the lineage leading to the core eudicots. What remains to be deciphered is why obligate B class heterodimerization evolved at least twice. What, if any, is the functional difference between B class homodimers and heterodimers? One hypothesis suggests that the convergent evolution of obligate AP3–PI interaction is not the result of drift, but rather because the AP3–PI heterodimer confers a selective advantage: a robust switch in floral development (Lenser et al., 2009). It must be stated that all investigations into B class homo- vs. heterodimerization have been conducted in vitro. There is no evidence as of yet that PI-like homodimers function in planta.

The C class genes (PLENA and FARINELLI) of Antirrhinum have subfunctionalized, in part because of shifting PPIs. PLENA controls both male and female organ identity (stamens and carpels), while FARINELLI confers only male organ identity, both in A. majus and when overexpressed in A. thaliana flowers (Davies et al., 1999; Causier et al., 2005; Airoldi et al., 2010). When ectopically expressed, PLE, like AG, is capable of specifying both male (stamen) and female (carpel) organ identity, but FAR confers only stamen identity. This functional divergence has been traced to a single glutamine insertion in FAR, the result of an altered splice site. This amino acid insertion affects PPIs with the E class SEPALLATA (SEP) proteins: FAR can only interact with SEP3, while AG can interact with SEP1, 2, and 3. This change in PPIs, overlaid on SEP homolog expression patterns, has resulted in the subfunctionalization of FAR and PLE. Structural and regulatory changes have acted in concert to effect functional differentiation (Airoldi et al., 2010). In the genus Medicago (Fabaceae), a major difference in fruit morphology is correlated with a similar single amino acid insertion into SHATTERPROOF (SHP)-like MADS box proteins. Rather than disrupting PPIs, however, the amino acid insertion may strengthen the interaction between Medicago SHP and SEP3 homologs (Fourquin et al., 2013).

Outside of the MADS box genes, there is evidence that PPIs affect natural variation in altered trichome density (Symonds et al., 2011) and light response in A. thaliana (Filiault et al., 2008), domestication traits in wheat (Simons et al., 2006), and flowering time in barley (Turner et al., 2005). Trichome density, in particular, changes in response to herbivore pressure, and has a fitness effect (Mauricio, 1998). The bHLH transcription factor ATMYC1 was found to underlie a QTL for trichome density in four separate A. thaliana mapping populations. A single amino acid change (P189A) was sufficient to abolish binding of atmyc1 to TTG (TRANSPARENT TESTA GLABRA) and GL1 (GLABROUS1) in yeast two hybrid assays (Symonds et al., 2011). Both TTG and GL1 are essential for trichome initiation in A. thaliana (reviewed in Balkunde et al., 2010). Presumably it is this altered interface with the trichome initiation pathway that results in reduced trichome initiation in plants with the Ler atmyc1 allele. In a cautionary tale for evolutionary biologists, positive selection acting on the ATMYC1 coding sequence was detected, but the region under selection was downstream of the trichome-reducing P189A substitution (Symonds et al., 2011).

Protein domains have been described that target proteins to particular cellular compartments [e.g., nuclear localization signals (Lange et al., 2007)]; that act as repressor or activator domains [e.g., the EAR repression domain (Ohta et al., 2001)]; that function in mediating the assembly of protein complexes [e.g., the PDZ domain (Kennedy, 1995)]; that act as post-translational modification (PTM) sites (Lusser et al., 2001); and that target proteins for destruction [e.g., the D box, (Ho et al., 2008)], to name a tiny subset of the existing diversity. The evolutionary origin of many characterized protein domains is often unclear or unexamined, except in a few cases. In a study of the evolution of plant protein domain gain and loss, Kersting et al. (2012) showed that new, plant-specific domains have emerged throughout plant history, but the highest rate of novel domain emergence was detected on the branch leading to the seed plants. This study also demonstrated that the arrangement of domains in individual proteins varies considerably, particularly at shallower phylogenetic levels. Lineage-specific domain architectures are not uncommon (Kersting et al., 2012).

Plant-specific gene lineages may possess domains present in all eukaryotes, but in land-plant-specific combinations (Xing et al., 2013). For example, the F-box and the tubulin DNA-binding domain are both found in all eukaryotes, but they are found adjacent to one another only in plants (Charoensawan et al., 2010). Similarly, HMG-box and AT-rich interaction domains are found in combination only in plants (Hansen et al., 2008). To catalog all characterized protein motifs and domains, and their occurrence in plant genomes, is beyond the scope of this paper. Instead, we have chosen to discuss examples where new functional domains in plant proteins have arisen through defined mechanisms, and to discuss examples where domain loss has been shown to have some defined functional consequence.

FT (FLOWERING LOCUS T) and TFL (TERMINAL FLOWER) are distantly related paralogous regulators of flowering in Arabidopsis. FT is a floral integrator, and FT expression induces flowering. TFL is a floral repressor and maintains indeterminate growth of the shoot apical meristem. This functional distinction between FT and TFL has been separately traced to a single amino acid difference in the predicted anion-binding pocket (Y85 in FT and H88 in TFL; Hanzawa et al., 2005) and to differences in an external protein loop termed “segment B” (Ahn et al., 2006). There is evidence that FT and TFL exert their respective functions as part of transcriptional activator and repressor complexes (reviewed in Taoka et al., 2013). Y85 in FT and H88 in TFL may be working to recruit transcriptional coactivators or corepressors, either alone or in concert with “segment B” (Ahn et al., 2006; Taoka et al., 2013).

Similarly, two FT homologs in Beta vulgaris (sugarbeet) show antagonistic functions in the regulation of flowering. BvFT2 function is conserved with FT and acts as a floral promoter while BvFT1 represses flowering. The antagonistic functions of BvFT1 and BvFT2 have been traced to differences at three amino acid residues in “segment B.” BvFT1 and BvFT2 appear to be the products of a relatively recent gene duplication event: BvFT2 homologs have not been found outside of the genus Beta (Pin et al., 2010).

Some soybean (Glycine max, Fabaceae) cultivars display a narrow leaflet phenotype, long been known to be controlled by a single gene, ln. Ln has been mapped to a genomic region that includes a single gene – Gm-JAG1– a homolog of the A. thaliana zinc-finger gene JAGGED. A single amino acid substitution (D9H) in the transcriptional repressor EAR motif of Gm-JAG1 is likely to be the causal ln mutation, rendering Gm-JAG1 non- or hypofunctional (Jeong et al., 2012). In addition to altering leaf morphology, the ln mutation affects the number of seeds per fruit (You et al., 1995; Dinkins et al., 2002). This example highlights how pleiotropic protein mutations may be tolerated and maintained in populations, possibly because of some fitness advantage. In this case, a fitness advantage may be conferred by the higher seed set of the Ln/ln heterozygote (Dinkins et al., 2002).

Teosinte glume architecture1 (tga1), an SBP-domain transcription factor, has been identified as a key locus in the domestication of maize from its wild progenitor, teosinte (Wang et al., 2005). Morphological differences between maize and teosinte ears are probably caused by a single coding change (K6N) in Tga1. This single amino acid change alters the biochemical function of TGA1, but the exact mechanism of this change remains unclear (Preston et al., 2012). Given the degree of morphological change associated with this single amino acid change, it is reasonable to hypothesize that TGA1 is a transcriptional activator, activating the set of genes responsible for the development of teosinte-like glume and inflorescence morphology. The single amino acid change observed in maize was sufficient to abolish, or significantly alter, this role of TGA1 (Wang et al., 2005).

Protein degradation is one common mechanism of post-translational gene regulation. In plants, polyubiquitylation of proteins, followed by proteolysis mediated by the 26S proteasome, is a particularly prevalent mechanism of post-translational regulation (Vierstra, 2003). Examples of altered protein stability, possibly because of altered polyubiquitylation and degradation, have been observed in the light-sensing cryptochromes and phytochromes, known to be degraded in a light- and ubiquitin-dependent manner (El-Assal et al., 2001; Maloof et al., 2001; Filiault et al., 2008).

Light responses, such as flowering time, vary considerably amongst A. thaliana accessions (Maloof et al., 2001). Multiple independent inactivations of FRIGIDA and FLOWERING LOCUS C have been identified in the study of natural variation in flowering time (reviewed in Alonso-Blanco et al., 2009), but structural changes in light-sensing cryptochromes and phytochromes have also been implicated. For example, a novel allele of CRYPTOCHROME-2 (CRY2) underlies a large-effect QTL controlling daylength sensitivity (El-Assal et al., 2001). A single missense amino acid substitution in CRY2 (V367M) results in a more stable protein as compared to the more common Ler allele (El-Assal et al., 2001). The same amino acid substitution in CRY2 (V367M) is also associated with shorter fruits, and decreased ovule number (El-Assal et al., 2004). A single amino acid (M548T) substitution in the phytochrome protein PHYA underlies reduced far-red light sensitivity in the Lm-2 accession of A. thaliana (Maloof et al., 2001). The substituted amino acid is able to affect multiple aspects of PHYA function: the photochemical properties of Lm-2 PHYA are affected by the M548T substitution; Lm-2 PHYA levels remained high in the light; and Lm-2 PHYA showed reduced autophosphorylation activity (Maloof et al., 2001). It is conceivable that the observed amino acid substitutions in both CRY2 and PHYA are interfering with some aspect of the phosphorylation, polyubiquitination, and 26S-mediated protein degradation pathway.

Hilscher et al. (2009) surveyed naturally occurring A. thaliana accessions for variation in trichome density. A single amino acid change, K19E, in the MYB domain transcription factor gene ENHANCER OF TRY AND CPC 2 (ETC2), underlies one large effect trichome density QTL. K19, although highly conserved in single-repeat R3 MYB proteins, is not in a characterized protein domain, but may represent an ubiquitination site. In the low-density accessions, where this lysine is replaced with a glutamate, ubiquitination of the ETC2 repressor may have been reduced or lost, resulting in higher stability of ETC2 and, ultimately, fewer trichomes (Hilscher et al., 2009). An interesting point arising from this study is the relationship between trichomes and root hairs. ETC2 is the only characterized single-repeat R3 MYB gene family member that affects trichome density, but not root hair density. The K19E replacement, found at a relatively high frequency in naturally occurring accessions, may be tolerated because it occurs in a gene with low pleiotropy (Hilscher et al., 2009).

Mutations that either decrease or increase protein function can be termed hypomorphs or hypermorphs, respectively (Muller, 1932). Examples of both hypomorphic and hypermorphic alleles in natural variation in a number of A. thaliana phenotypes have been described.

Hyperaccumulation and salt tolerance have repeatedly been associated with altered functionality of transporters and biosynthetic enzymes. Amino acid substitutions in conserved domains of HMA3 and HMA5 underlie A. thaliana QTL for Cd accumulation (Chao et al., 2012) and Cu tolerance (Kobayashi et al., 2008), respectively. The amino acid substitutions in HMA3 result in a hypofunctional translocator and, ultimately, higher Cd accumulation. Similarly, high sulfate accumulation in the Shahdara accession of A. thaliana (Loudet et al., 2007) and differences in salt tolerance between rice accessions (Ren et al., 2005) have been separately associated with hypomorphic alleles.

The late flowering Sy-0 accession of A. thaliana is distinctive in its morphology. The basal rosette is enlarged, aerial rosettes form in the axils of stem leaves, and early floral meristems revert to indeterminate growth (Poduska et al., 2003). A single amino acid replacement in the pre-mRNA processing factor, HUA2, is responsible for the majority of the Sy-0 aerial rosette phenotype. HUA2 has been shown to positively regulate the flowering genes AG (floral patterning, floral determinacy) and FLC (flowering time). In the Sy-0 accession, AG function is attenuated, and FLC expression is enhanced. Thus, the single Sy-0 amino acid replacement in HUA2 (K525E) is a partial loss-of-function (hypomorphic) allele with respect to its effects on AG, and a gain-of-function (hypermorphic) allele with respect to FLC expression. Although the morphological phenotype exhibited by the Sy-0 accession is not rare, the nucleotide polymorphism that causes the K525E amino acid replacement is rare. In a survey of 113 A. thaliana accessions, only Sy-0 was found to possess the causative single nucleotide polymorphism (SNP; Wang et al., 2007).

Naturally occurring accessions of A. thaliana exhibit considerable diversity in the rate of leaf production. One accession, Est-1, shows both slower leaf production, as well as extensive necrosis on older leaves. Both slower leaf production and late onset leaf necrosis in Est-1 are due to gain of function (hypermorphic) mutations in a single gene, ACCELERATED CELL DEATH6 (ACD6). ACD6 encodes a transmembrane protein involved in the regulation of salicylic acid accumulation and the defense response. The increased activity of ACD6 observed in Est-1, and 14 other A. thaliana accessions, may confer enhanced pathogen resistance, but with costs. Enhanced pathogen resistance comes at the price of reduced biomass (fewer, smaller leaves), which in turn is associated with fitness costs (Abreu and Munné-Bosch, 2009; Todesco et al., 2010).

We have divided our discussion into six broad categories of protein change, but we could also have divided the examples according to the evolutionary scale at which the change was predicted to occur (Table 1). Evolutionary change can be considered microevolutionary (occurring within a single population or species) or macroevolutionary (transcending species boundaries; Gould, 2002). When protein evolution is considered with these categories in mind, do certain changes occur preferentially on a micro- or macroevolutionary scale? It must be stated that all evolutionary events probably happen at a microevolutionary scale, within a population, but the scale at which we observe these events changes. Some categories of change were detected at both micro- and macroevolutionary scales, including active site evolution of enzymes, altered activity as a transcriptional activator or a repressor, and the evolution of PPIs. The evolution of competitive inhibition appears to occur primarily on macroevolutionary time scales, while dominant negatives were detected exclusively at a microevolutionary scale. Dominant-negative alleles and competitive inhibition are similar in character, and it is conceivable that dominant-negative alleles might represent the first step on one pathway to the evolution of competitive inhibition. Domain loss, observed at both micro- and macroevolutionary scales, may represent another pathway leading to competitive inhibition.

The existing examples of DNA-binding domain evolution occur on very deep, macroevolutionary time scales. Similarly, there were no examples of novel domains at microevolutionary timescales. Are these events so rare, and so often deleterious, that they are seldom uncovered in the study of population-level natural variation? Or, would systematic analysis of DNA-binding or protein domain architecture at a population-level reveal microevolutionary examples?

At the opposite side of the spectrum, but similarly illuminating, lie changes that were detected predominantly on microevolutionary scales. In addition to dominant negatives, hypo- and hypermorphic alleles and altered protein stability were detected almost exclusively on microevolutionary, or intrageneric, time scales. These examples may suggest where to look for innovation on macroevolutionary scales. These changes, sometimes causing drastically altered phenotypes, are tolerated in natural environments. Often it is difficult to distinguish functional and phylogenetic signal from the noise in evolutionary analyses of gene families. Perhaps looking for altered stability of evolutionary variants, for example, might yield insight into the functional consequences of molecular evolution. Altered protein stability, in particular, may represent one way in which a protein’s function might stay intact, but the protein may persist for a shorter or longer period of time. This could conceivably result in a heterochronic shift (Klingenberg, 1998) in a particular trait.

One interesting point arising from our survey of the existing literature is that proteins can change in a number of ways that were not uncovered here. One class of changes, in particular, that remained elusive was PTMs. The examples of altered protein stability may have ultimately been because of altered PTMs, but that remains to be determined. The absence of altered PTMs in the study of protein evolution is perhaps because many of the PTMs of individual extant proteins are still incompletely understood, so assessing PTMs in an evolutionary context remains extremely challenging. In the case of QTL cloning, PTM alterations may not be tolerated very often, and will therefore vary only very rarely on microevolutionary scales. Examples do arise in mutant analyses (Soppe et al., 2000; Kim et al., 2006), so more cases of natural variation in PTMs may be forthcoming. PTMs have clearly arisen and diversified in proteins and the study of their evolution represents an interesting area of future exploration.

Although many of the discussed changes primarily affect transcription factors, the phenotypic outcomes of these changes are often vastly different. Even within one class of change, altered PPIs, one altered interaction affects trichome density in Arabidopsis, another affects floral morphology in Thalictrum. Although similar biochemical changes might have occurred, the ultimate phenotypes on which natural selection might act are distinct and not evolutionarily equivalent.

Genetic analysis (QTL cloning) has deepened our understanding of the molecular underpinnings of phenotypic diversity to a considerable degree. As more QTL are uncovered and cloned, no doubt this understanding will grow ever deeper. But systematically cloning QTLs will not tell us everything there is to know about the evolution of plant form and function. It remains important to combine all of the strategies available to us, including phylogenetic analyses of gene families, structural analyses, and functional analyses of proteins in an evolutionary context, in order to gain a more complete picture of protein evolution. It would also be extremely informative to know how many of the QTL that have been cloned confer adaptive phenotypes, or have the potential to be adaptive under certain conditions. Although challenging, field and laboratory selection tests on some of the more promising accessions would no doubt yield fascinating results.

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