The mutant alleles used in this study: ttg1-1 (Ler) (Koornneef, 1981; Walker et al., 1999). gl3-3 (Col) background (Jakoby et al., 2008); egl3-77439 (Col) background (TAIR accession 1008704039); tt8-048673 (Col) background (TAIR accession 1005848854); gl3 egl3 (Col) (Friede et al., 2017); gl3/egl3/tt8 (homozygous progeny by crossing gl3/egl3 with tt8-048673); gl1 (Col) (Oppenheimer et al., 1991); pap1 (pst16228) (No-0; CS28564) (Kuromori et al., 2004); pap2 (Col) (salk_093731, TAIR accession 1005457343). Plants were grown on soil at 24°C with 16 h of light per day. All transgenic plant lines used and created in this work are listed in Supplementary Table S2. Transgenic lines were generated using the floral dip method (Clough and Bent, 1998). Transgenic T1 plants were selected on ½ MS agar plates containing 10 mg/L BASTA [glufosinate ammonium, Sigma-Aldrich (Munic, Steinheim)].

All CDS entry clones were generated by amplifying the CDSs from start to stop codon from A. thaliana (At), Arabis alpina (Aa), G. hirsutum (Gh), Petunia hybrid (Ph), and Zea mays (Zm) followed by BP recombination in pDONR201/207.

Three different destination vectors were used for subsequent LR reactions. pcDNA3-Rluc-GW and pTREX-dest30 (Invitrogen) for the N-terminal fusion of Renilla reniformis and Staphylococcus aureus protein, respectively, were described before (Pesch et al., 2015).

Genes were N-terminal fused to the S. aureus protein A sequence in pTREX-dest30-ntProtA by LR reaction. For the negative control, the vector pTREX-dest30-ntProtA was recombined with pENTR1A-w/occdB.

The full-length Renilla luciferase sequence was N-terminal fused to the coding sequences in pcDNA3-Rluc-GW. Also pENTR1A-w/o-ccdB was recombined to this vector as a negative control.

All CDS of MBW homologs were cloned into Donor vectors by BP reactions (Invitrogen). Then recombination of the corresponding entry clones with the 35S promoter containing vector pAMPAT-35S-GW [GenBank accession no. AY436765 (Pesch et al., 2015)].

The presence of trichomes was scored on leaves 1–6 on 2-week-old seedlings. For the analysis of root hairs H-files were identified microscopically by their position over cortical cell boundaries in roots of 7-day-old 1/2MS plate-grown T2 progeny seedlings The number root hairs in the flanking N-files was roughly determined in 1-week-old basta-resistant T2 seedlings under a Leica stereomicroscope (MZ FLIII).

Anthocyanidin was analyzed in T2 seedlings grown on 1/2MS germination medium containing 3% sucrose, and analyzed using a Leica stereomicroscope (MZ FLIII) with the Multi-Focus and Montage option of the Leica Application Suite V3 (Leica Microsystems, Wetzlar, Germany). Seed coat color was observed in T1 progeny seeds and pictures were taken with a Canon EOS 5D Mark (Canon, Krefeld, Germany). For the analysis of seed mucilage T2 seeds were imbibed in 0.2% w/v aqueous ruthenium red (Sigma) solution with 0.5% agar and monitored after 3–4 h by light microscopy and pictures taken using the DISKUS software (Carl H. Hilgers Technisches Büro, Germany) (Modified from Penfield et al., 2001).

For LUMIER assays, each protein was transiently expressed in HEK293TM cells (BioCat/SBI: LV900A-1) as hybrid proteins either with the S. aureus protein A (ProtA) or with the R. reniformis luciferase (Rluc) fused to their amino N termini. The LUMIER assay is based on the technique established for high throughput screens (Blasche and Koegl, 2013) and was optimized for the proteins used in this study (Pesch et al., 2013; Pesch et al., 2015). Constructs (1.5 μg plasmid DNA) were single-transfected into 1 × 10⁶ HEK293TM cells in 6-well plates using 10 μl Lipofectamine 2000 (Thermo). After 48 h, the medium was removed and cells were washed with PBS. The cells were collected by centrifugation (600 g, 15 min) and lysed on ice in 50 μl ice-cold lysis buffer [20 mM Tris pH 7.5, 250 mM NaCl, 1.1% Triton X-100, 10 mM EDTA, 10 mM DTT, Protease Inhibitor Cocktail (Roche, 1836170) and benzonase (Novagen, 70746; 0.0125 units per μl final concentration)] for 1 h. Subsequently, lysates were centrifuged at 15,000 rpm (21,380 g) for 15 min at 4°C, and the supernatant was mixed with 10 μl PBS-prewashed sheep anti-rabbit IgG-coated magnetic beads (Invitrogen, Dynabeads M280; 2 mg/ml final concentration) and incubated for 1 h on ice. The beads were collected using a magnetic holder (Neodym-Magnets, 1.3 Teslar) and luminescence measured in a microtiter plate reader (BMG Fluostar Optima). The pulldown was also performed with untransfected cells and with cells solely expressing Luciferase-protein to exclude any non-specific signal from the cell lysate and non-specific binding of Luciferase-protein to the beads, respectively.

The percentage of Rluc on the beads compared with the lysate was calculated by dividing the Rluc activity on the beads by the Rluc activity in the same amount of lysate used in the pull-down assay (Input). Pulldown efficiency was calculated by: [Rluc pulldown/Input] × 100.

Unrooted Maximum Likelihood Phylogenetic Trees for WDR, bHLH, and R2R3MYB was done with MEGA6 (Tamura et al., 2013) using entire amino acid sequences based on the JTT matrix-based model (Jones et al., 1992). In addition to the proteins used in this study we added proteins from other species also discussed to play a role in anthocyanidin regulation (Zhang and Schrader, 2017).

In order to analyze whether the protein function of homologous TTG1 genes in other species is conserved we performed rescue experiments in A. thaliana ttg1-1 mutants (Koornneef, 1981; Walker et al., 1999). Toward this end we selected TTG1 homologs from several species including A. thaliana (AtTTG1), A. alpina (AaTTG1), G. hirsutum (GhTTG1, GhTTG2, GhTTG3, GhTTG4), P. hybrida (PhAN11) and Z. mays (ZmPAC1, ZmMP1). A. thaliana TTG1 served as a control for the rescue efficiency.

Arabidopsis thaliana and A. alpina TTG1 showed efficient rescue of all traits in most T1 lines (Table 1). The cotton GhTTG1 and GhTTG3 proteins also rescued all traits in more than 50% of the T1 lines [see also (Humphries et al., 2005)]. GhTTG2 and GhTTG4 did not show rescue which is consistent with the previous result that transient expression cannot rescue the anthocyanidin phenotype of the ttg1 mutation in M. incana (Humphries et al., 2005). Petunia AN11 efficiently rescued all traits in most T1 lines [see also (Payne et al., 2000)]. Consistent with previous results, the maize PAC1 also rescued most T1 lines (Carey et al., 2004). MP1 expression did not rescue the ttg1 phenotypes.

bHLHs proteins involved in MBW functions in the context of the five traits analyzed are typically important for some but not all traits. In order to study the function of bHLH homologs in Arabidopsis we used the gl3 egl3 tt8 triple mutant in Col background. This triple mutant exhibits defects in all five traits similarly as ttg1 mutants (Zhang et al., 2003).

In a first step, we studied the rescue ability of the Arabidopsis bHLH proteins AtGL3, AtEGL3, and AtTT8 (Table 2). The genetic analysis had revealed partially overlapping functions of these genes such that trichome and root hair formation are controlled by AtGL3, AtEGL3, anthocyanidin production by AtGL3, AtEGL3, and AtTT8, seed color by AtTT8, seed coat mucilage by AtEGL3 and AtTT8 (Zhang et al., 2003). To distinguish, whether this differential function is due to differential expression or different protein functions, we expressed the respective CDS in the gl3 egl3 tt8 mutant. In addition, we included AtMYC1 which is involved in trichome and root hair development but has a molecular function distinct from AtGL3 and AtEGL3 and cannot rescue the gl3 egl3 mutant trichome phenotype (Zhao et al., 2012; Pesch et al., 2013). We noted that none of the genes rescued seed coat color. We reasoned that the 35S promoter might not be sufficient for driving the expression in the seed coat. We therefore performed a second set of experiments in which the bHLH were expressed under the TT8 promoter in tt8 mutants and used these results for the interpretation (Table 3).

Consistent with the genetic results, AtGL3 rescued trichome and root hair formation and anthocyanidin production but not seed color or mucilage release (Supplementary Figures S3–S5 and Tables 2, 3). AtEGL3 rescued all four traits in which it is also genetically relevant and in addition shows a weak rescue of seed coat color. Consistent with its genetic function, TT8 expression led to a rescue of seed color and mucilage release (Supplementary Figures S5, S6). Thus, the differential function of the analyzed bHLH proteins can be explained at the protein level. The MYC1 protein, could rescue root hair and seed mucilage release.

To analyze the function of putative orthologous bHLH proteins we focused on another Brassicaceae species, A. alpina. Here, orthologous genes can be identified by sequence similarity and the chromosomal context of the respective genes by studying synteny. We found similar rescue properties for most bHLH proteins with one notable exception (Tables 2, 3). In contrast to AtGL3, the AaGL3 protein was unable to rescue the seed mucilage phenotype. This indicates that the function of GL3 has diverged between the two species at the protein level.

In cotton, we chose to analyze GhDEL65 and GhDEL61 that are linked to fiber formation and that can functionally replace AtGL3 and AtEGL3 in Arabidopsis when expressed under the Arabidopsis GL3 upstream sequences (Wang et al., 2013; Shangguan et al., 2016). We found rescue of the root hair, anthocyanidin and mucilage phenotype (Tables 2, 3). It is difficult to explain why trichome rescue was observed when the two cotton genes were expressed under the control of the AtGL3 upstream sequences but not under the control of the 35S promoter in our studies. As the AtGL3 gene requires regulatory intron sequences for the correct expression (Friede et al., 2017), it is possible that the different results can be explained by unknown differences in the temporal, spatial or levels of expression. However, our data clearly show that the two bHLH proteins from cotton can regulate root hair, anthocyanidin and mucilage release. Hence, by their protein function they are most similar to Arabidopsis EGL3 and clearly distinct from TT8.

We analyzed two bHLH genes from Petunia, PhJAF13 and PhAN1. Mutations in PhAN1 lead to defects in anthocyanidin production, seed color and seed coat cell differentiation (Spelt et al., 2000, 2002). PhJAF13 was shown to activate anthocyanidin production upon overexpression (Quattrocchio et al., 1998). Our rescue experiments revealed different properties (Tables 2, 3). While PhAN1 expression in the gl3 egl3 tt8 mutant rescues the anthocyanidin, seed color and mucilage phenotypes, PhJAF13 overexpression resulted in a rescue of the root hair and seed coat mucilage phenotypes. Thus, the rescue ability of PhAN1 corresponds well with the mutant phenotype of the respective Petunia mutant. By contrast, PhJAF13 appears to have also the property to promote root hair formation (Supplementary Figure S3).

From Maize we selected three bHLH genes for our analysis, ZmR(Lc), ZmR(S), and ZmB. Mutations in all three genes result in severe anthocyanidin phenotypes (Dooner, 1979; Ludwig et al., 1989; Patterson et al., 1991; Stadler and Neuffer, 1953). Overexpression of both ZmR genes was sufficient to rescue all five traits whereas ZmB exhibited no rescue of any trait (Tables 2, 3).

We focused our analysis of the rescue ability of R2R3MYB proteins on the rescue of an anthocyanidin mutant as an example for the most ancient trait and trichome mutants as an example for a more recently evolved trait (Table 4). For the analysis of the anthocyanidin phenotype we used the Arabidopsis pap1 pap2 double mutant. AtPAP1 and AtPAP2 are close homologs and both involved in the regulation of anthocyanidin (Teng et al., 2005; Gonzalez et al., 2008). The trichome phenotype was assessed in Arabidopsis gl1 mutants. The R2R3MYB proteins were expressed under the control of the 35S promoter. As expected both, AtPAP1 and AtPAP2, rescued the double mutant efficiently. However, none of the other Arabidopsis R2R3MYB proteins including AtGL1, AtWER, AtTT2, and AtMYB61 were able to rescue the anthocyanidin phenotype. The gl1 trichome phenotype was rescued only by GL1 but not by any of the other proteins (Supplementary Figure S7). This suggest that also the R2R3MYB protein have evolved trait specific properties. A similar result was obtained for the Arabis R2R3MYB proteins. Only the PAP-like protein AaPAPL revealed rescue of the anthocyanidin phenotype and only AaGL1 the gl1 phenotype. In cotton we chose four R2R3MYB proteins. GhMYB2 is expressed in fibers and can rescue the Arabidopsis trichome phenotype (Wang et al., 2004). GhMYB25 is expressed in the seed coat and mutations in the gene lead to slower fiber growth (Machado et al., 2009). GhRLC1 was shown to be involved in the regulation of anthocyanidin in cotton (Gao et al., 2013). In addition we added GhMYB3 which we found as a close homolog of GhMYB2. Consistent with the mutant phenotype, GhRLC1 rescued the anthocyanidin phenotype (Table 4). GhMYB2 and GhMYB3 rescued the gl1 trichome phenotype, GhMYB25 exhibited no rescue. In Petunia we analyzed three genes. PhAN2 and PhAN4 are important for anthocyanidin production (Quattrocchio et al., 1993, 2006b; Spelt et al., 2002). PhPH4 regulates the acidification of the vacuole in petals (Quattrocchio et al., 2006b). All three R2R3MYB proteins rescued the anthocyanidin phenotype but not the trichome phenotype. Finally, we studied the ZmC1 gene from maize that was described to act as a regulator of anthocyanidin production (Sainz et al., 1997; Grotewold et al., 2000; Hernandez et al., 2004). Consistent with this function in maize we found that ZmC1 rescued the pap1 pap2 mutant but not the gl1 mutant.

It is generally thought that the specificity of MBW function is due to the formation of specific combinations of particular components. In this light, the differential rescue behavior of MBW proteins from different species might be explained by differences in the protein-protein interactions with the proteins in A. thaliana. Interactions of MBW proteins in each species were already extensively studied (Supplementary Table S3). To analyze the interactions of MBW proteins from other species with the Arabidopsis proteins we performed LUMIER assays [LUminescence-based Mammalian IntERactome (Blasche and Koegl, 2013)]. In LUMIER assays proteins of interest are expressed in human HEK293TM cells with one protein fused to the S. aureus protein A (ProtA) and the other with a R. reniformis luciferase (Rluc) followed by pulldown experiments and a quantification of the precipitated luciferase activity (Supplementary Figure S8 and Table 5).

In a first step we explored the binding behavior of TTG1 homologs to Arabidopsis bHLH proteins (Table 5). AaTTG1, GhTTG1, PhAN11, and ZmPAC1 exhibited interactions with all four Arabidopsis bHLH proteins. GhTTG3 interacted with AtGL3, AtEGL3, and AtMYC1 but not with AtTT8. GhTTG2, GhTTG4, and ZmMP1 proteins did not interact with any of the bHLH proteins. This correlates well with our finding that the latter four TTG1 homologs did not rescue ttg1 mutants in Arabidopsis.

In a next step we tested the binding of homologous bHLH proteins to Arabidopsis TTG1 and MYB proteins (Table 5). All bHLH proteins tested showed an interaction with the Arabidopsis AtTTG1 protein. Binding of bHLH proteins from other species with the Arabidopsis R2R3MYB proteins varied. Most bHLH proteins showed interaction with GL1, WER, PAP1, PAP2, and TT2 but not with MYB61. Arabis EGL3 exhibited a differential binding behavior such that it interacts with GL1, WER, PAP1, and TT2 but not PAP2. We found no interactions of ZmB with any R2R3MYBs. This is consistent with the fact that ZmB possesses no rescue ability.

Finally, we studied the binding behavior of R2R3MYB from other species with Arabidopsis MBW proteins (Table 5). As expected, we generally found no binding to AtTTG1. Only ZmPL exhibited weak binding to AtTTG1. Except for GhMYB25 all R2R3MYB proteins interacted with all four bHLH proteins from Arabidopsis.

Taken together the analysis of the interactions of MBW proteins from other species with the Arabidopsis proteins can explain those cases in which we found no rescue of Arabidopsis mutants as the respective proteins do not interact with the relevant partners. However, the binding patterns do not explain any differential rescue of traits.

In Arabidopsis, the MBW complex regulates five different traits by involving different bHLH and R2R3MYBs in a trait-specific manner. This raises the question, whether trait specificity is due to transcriptional regulation or different properties of bHLH and R2R3MYB proteins. There are examples available suggesting that both aspects are relevant (Lai et al., 2013; Lloyd et al., 2017). To address this question, we expressed the MBW proteins under the 35S promoter in appropriate A. thaliana mutants and analyzed, which traits were rescued in the T1 generation. This enabled us to judge the ability of MBW proteins to function in the regulation of the different traits. Rescue experiments with A. thaliana bHLH proteins GL3, EGL3, and TT8 revealed that the rescue ability of the different traits matches their genetic requirements (Supplementary Table S1) (Zhang et al., 2003). This suggests that these bHLH proteins possess properties mediating trait specific regulation events. By contrast, the bHLH protein MYC1 is important for trichome development but cannot rescue the trichome phenotype of the gl3 egl3 ttg8 mutant. This is consistent with previous data (Zhao et al., 2012) and can be explained by a different cellular localization and molecular function (Pesch et al., 2013). Similarly, we found some trait specificity for the R2R3MYB proteins. However, we found no rescue of gl1 mutants with overexpressed WER. This is in conflict with previous findings where WER expression under the GL1 promoter showed rescue (Lee and Schiefelbein, 2001). This may be considered to show the limitations of the overexpression approach taken in this study. It may also reflect that an overexpression approach is more stringent such that only GL1 can rescue the gl1 mutant but not WER.

Because a rescue approach considering all five traits does not allow a systematic analysis of evolutionary conserved MBW proteins, we selected genes and species for which genetic or molecular data are available suggesting a functional relevance in one or several of the pathways under consideration. As a consequence, some of our experiments are repetitions and can serve as references and controls for comparison with previous results. Most WDR homologs studied here were able to complement ttg1 A. thaliana mutants. Only the two cotton proteins GhTTG2 and GhTTG4 and the maize MP1 protein did not show any rescue. This observation correlates with the phylogenetic classification that places MP1 and PAC1 in different clades (Carey et al., 2004). All WDR proteins including AtTTG1, AaTTG1, GhTTG1, GhTTG2, GhTTG2, and GhTTG4 are most similar to AtAN11, which belongs to MP1 clade (Carey et al., 2004; Humphries et al., 2005). The rescue ability of bHLH protein from other species correlates reasonably well with the function of those for which mutant data are available. This is true for the A. alpina proteins TT8, EGL3, petunia AN1 and maize R(LC) and R(S). A notable exception is the maize ZmB protein as the corresponding maize mutants show an anthocyanidin phenotype. As the ZmB protein does not interact with any of the A. thaliana R2R3MYB proteins, it is conceivable that ZmB cannot rescue because of this lack of interactions. We found a similar situation for the R2R3MYB proteins. Here all R2R3MYBs for which a function in the regulation of anthocyanidin was demonstrated by mutant phenotypes in the respective species specifically rescued the anthocyanidin phenotype but not the trichome phenotype. This includes the cotton proteins GhRCL1 (Gao et al., 2013), the petunia proteins PhAN2 and PhAN4 (Quattrocchio et al., 1993, 2006b; Spelt et al., 2002) and the maize protein ZmC1 (Sainz et al., 1997; Grotewold et al., 2000; Hernandez et al., 2004).

Previous experiments aiming to understand whether changes in the temporal-spatial regulation or changes in protein functions are relevant to adopt trait specific functions revealed examples for an importance of gene regulation. GL1 and WER can equally well rescue the wer root hair phenotype when expressed under the WER promoter (Lee and Schiefelbein, 2001). Similarly, the gl1 mutant phenotype can be rescued by the MYB82 protein when expressed under the GL1 promoter (Liang et al., 2014). Taken together, our data suggest that also changes in protein functions of MBW proteins are important. The specificity for the regulation of the five the different TTG1-dependent traits can be explained to a large degree by the protein behavior of the bHLH and R2R3MYB proteins. It will be interesting to see in future experiments to understand which biochemical properties of the bHLH and R2R3MYB proteins might be relevant for trait specific regulation.