Introduction

The general transcription factor TFIID plays an important role in core promoter recognition and in nucleation of the preinitiation complex for RNA polymerase II (RNAPII) (Martinez et al. 1994; Verrijzer et al. 1995). Additionally, the subunits of TFIID also serve as recruitment targets for transactivator proteins bound to upstream promoter elements. As many as 15 different TAFs have been shown to associate with TBP to form the TFIID complex in eukaryotes. The existence of two Arabidopsis thaliana TBP proteins has been known for many years (Gasch et al. 1990); however, very little direct information is available regarding subunit composition and function of TFIID in plants. We (Lawit and Czarnecka-Verner 2004), as well as Lago et al. (2004), have reported the identification of a large family of TAF-like proteins predicted in the Arabidopsis genomic sequences. Two groups have purified TFIID from monocots. Wheat TFIID was isolated by coimmunoprecipitation with TBP to yield a protein complex that could support activated transcription in vitro (Washburn et al. 1997). Rice proteins were co-purified with TBP from suspension cells transformed with maize TBP engineered to have a biotin affinity tag (Zhong et al. 2003). A total of 86 peptides were identified by tandem mass spectrometry (Zhong et al. 2003). Of these, four were designated as TAFs; however, only two (OsTAF6 and OsTAF11) were related to TFIID subunits in other organisms. The remaining two proteins corresponded to Mot1 (OsTAF172), that is part of the B-TFIID complex (Pereira et al. 2001; Tora 2002; van der Knaap et al. 1997), and a TAF associated with RNA polymerase I. This finding, combined with the relative paucity of proteins then annotated as TAFs in the Arabidopsis genomic databases, originally led to speculation that plant TFIID contains few canonical TAFs (Zhong et al. 2003).

Mutant phenotypes have been linked to three plant TAFs. Of these, two are implicated in development, TAFs 6 and 10. Homozygous T-DNA insertion mutants of AtTAF6 are lethal, and heterozygotes show a markedly reduced rate of pollen tube growth (Lago et al. 2005). In Arabidopsis there are two alleles of TAF6 (TAF6 and 6b). Mutants of one of the alleles, AtTAF6, are lethal; however, this does not necessarily imply nonredundancy in function between AtTAF6 and 6b since these two alleles may not be co-expressed in all tissues. For example, only AtTAF6 is expressed at detectable levels in mature pollen and in tricell microgametes (https://www.genevestigator.ethz.ch, Zimmermann et al. 2004). In studies where TAF10 from Flaveria trinervia was overexpressed in Arabidopsis, two distinct phenotypes were evident (Furumoto et al. 2005). The first was a terminal flower-like phenotype displaying limitation of the indeterminate inflorescence, the presence of chimeric floral organs and dwarf curled leaves. The second was manifest in infertile plants with severely deformed, dark green leaves that resembled greening callus. In an independent study, overexpression of AtTAF10 in Arabidopsis improved seed germination rates under conditions of osmotic stress, whereas the T-DNA knocked-down mutant was more sensitive to salt stress (Gao et al. 2006). In addition to TAF involvement in plant development and osmotic stress, a mutation in Arabidopsis TAF1b (gene product of HAF2) is associated with an impairment in the integration of light signals into transcriptional regulation (Bertrand et al. 2005).

There are indications that plant TAFs play important roles in activated transcription and in the regulation of gene expression; however, the precise number of TAFs in plants and their associations within TFIID are still undetermined. Utilizing BLAST searches of the Arabidopsis genome, Lago and colleagues (2004) identified as many as 18 genes corresponding to 14 distinct RNAPII-associated TAFs. Phylogenetic analysis indicated that AtTAFs are most related to yeast TAFs rather than those in humans or Drosophila. Only AtTAF12 and AtBTAF1 are more related to metazoan TAFs. Four putative AtTAF genes encoding the two pairs, AtTAF14/14b and AtTAF15/15b, have very low overall similarity with possible homologs in yeast and metazoans (20–40%). Moreover, higher conservation is only seen in limited domains that are also present in non-TAF proteins making their assignment as bona fide TAFs difficult. However, evidence in the present study showing affinity of AtTAFs14/14b and 15/15b for a number of the more conserved TAF proteins of Arabidopsis supports their designation as true TAFs. With the exception of AtTAF1, AtTAF4 and AtTAF5, all Arabidopsis TAF genes map to regions of the genome duplicated in the past by events that affected approximately 70% of the Arabidopsis genome (Initiative 2000; Lago et al. 2004), but none of the TAF genes themselves were duplicated in this process. Of the seven TAF genes duplicated in Arabidopsis (TAFs 1, 4, 6, 11, 12, 14 and 15), all but TAF6/6b are represented by two genes located on different chromosomes.

In lieu of a direct demonstration of TAF integration into the TFIID complex via coimmunoprecipitation, our goal was to test protein interactions between putative TFIID subunit proteins and formulate a comprehensive map of the network of contacts between them. In the genome sequence database we identified Arabidopsis homologs of all canonical TAFs, excluding TAF3. TAF and TBP sequences were cloned from cDNA and analyzed utilizing the yeast two-hybrid system to generate a matrix of protein interactions. These experiments provide supporting evidence confirming the identity of putative TAFs originally indicated by sequence homology alone. The resulting interaction map of Arabidopsis TFIID is largely consistent with the trilobed topology determined for yeast TFIID (Leurent et al. 2002) suggesting that the major aspects of TFIID structure have been conserved among eukaryotes.

Materials and methods

Cloning of TAF cDNAs

Total RNA was extracted from A. thaliana suspension cells using a Plant RNeasy kit with on-column DNAse treatment (Qiagen, USA). First strand cDNA synthesis was performed with 1 μg of Arabidopsis total RNA as a template, anchored oligo-dT primers, and Superscript II reverse transcriptase (Invitrogen, USA). Primers listed in Supplementary Table 1 were used to PCR amplify the corresponding AtTAFs from the cDNA using Platinum Pfx (Invitrogen) and varying cycle parameters based on primer specifications. AtTAF12b was subcloned from the Arabidopsis Biological Resource Center (ABRC) clone U11077.

PCR products for TBPs, full length TAFs and their subfragments were cloned into pENTR/D-Topo (Invitrogen) according to the manufacturer’s instructions and transformed into chemically competent One Shot TOP10 E. coli cells (Invitrogen). All constructs were subsequently transferred to their respective destination vectors using Gateway protocols. Primers used to PCR amplify AtTAF1subfragments, AtTAF2 and AtTAF4 contained attB sites and were recombined by BP reactions into pDONR207 according to manufacturer’s protocol (Invitrogen). All AtTAF and TBP cDNA clones were verified by full insert-sequencing and are listed in Table 1.

Table 1 Arabidopsis TFIID subunit cDNA GenBank accession numbers and other characteristics
Full size table

Yeast complementation test

TAF mutant yeast strains carrying wild type copies of the corresponding yTAFs on URA3 plasmids (YSB373–TAF11, YSB377-TAF11, YSB590-TAF12 and YEK16-TAF10) were provided by Stephen Buratowski (Klebanow et al. 1996; Komarnitsky et al. 1999; Michel et al. 1998). AtTAFs 9–12, hTAF9 and yTAFs 9–12 were subcloned into pRS314 and transformed into the appropriate yeast strains as listed in Table 1. To test for complementarity, the strains carrying the pRS314 clones were streaked on SD -Trp, +0.1% 5′ fluoroorotic acid (5′FOA). The strains would grow only if the pRS314 clones were capable of complementing the yTAFs that were being lost with the URA3 (pJA73) plasmids.

Two hybrid analysis

Yeast two-hybrid vectors pGAD-T7 and pGBK-T7 (Clontech, USA) were adapted to the Gateway system by digesting both plasmids with SmaI and subsequent ligation with the Gateway conversion cassette B (Invitrogen). Sequence-verified pENTR/D-Topo constructs were recombined in vitro successively through pGAD-T7, pDONR207 and pGBK-T7 Gateway vectors using LR and BP Clonase reactions (Invitrogen) and transformed into electro-competent MACH1 E. coli (Invitrogen).

The resultant yeast two-hybrid bait (contains Gal4 DNA binding domain; Gal4 DBD) and prey (contains Gal4 activation domain; Gal4 AD) constructs were transformed into MaV204K provided by Dr. T. Ito, Kanazawa University, Japan (Ito et al. 2000) and AH109 (Clontech) yeast strains, respectively. All transformations of vector plasmids were conducted using the Frozen-EZ Yeast Transformation II system (Zymo Research, USA). Bait/MaV204K and prey/AH109 transformants were then mated and selected on SD -Trp, -Leu and SD -Trp, -Leu, -His, -Ade (high stringency) to test for mating efficiency and interaction, respectively. Yeast matings were performed similarly to the protocol in the Clontech Yeast Protocols Handbook (1999). Ten μl of overnight mating cultures was serially diluted 10-fold, three successive times. From each dilution, 3 μl of cell suspension were removed and spotted in a grid on both SD -Trp, -Leu and SD -Trp, -Leu, -His, -Ade screening plates (25 cm × 25 cm). The plates were then incubated at 30°C and monitored for growth over a period of 14 days.

Quantitative β-galactosidase (β-Gal) assays were performed using CPRG substrate (Roche, Germany) according to the Clontech Yeast Protocols Handbook. Colonies used to inoculate the cultures for this test were taken from the SD -Trp, -Leu plates, thus eliminating selection for false positive interaction. All values were expressed as Miller units (Miller 1972; Miller 1992). The β-Gal assay values were analyzed against their appropriate control values using a heteroscedastic, one-tailed distribution Student’s T-test.

Interactions initially designated as positive for growth on selective media (irrespective of normalized β-Gal values) were retested by co-transforming all identified pGADT7-AtTAF/pGBKT7-AtTAF interacting pairs into yeast strain AH109 and selecting on SD -Leu, -Trp. In addition, each construct was co-transformed with a non-interacting AtTAF and pGBKT7/pGADT7 empty vector as negative controls. Yeast co-transformants were picked and patched onto SD -Leu, -Trp and SD -Leu, -Trp, -His, -Ade, X-α-Gal (high stringency). Transformants that grew and turned blue on the high-stringency media within 3–4 days were scored as positive for interaction.

Results

AtTAF cloning and TFIID interaction mapping using yeast two-hybrid analysis

Using homology-based searches of the A. thaliana genomic sequence database, 21 loci encoding putative TAFs have been identified. Table 1 lists all identified cDNAs or genes used in this study and provides information regarding loci, accession numbers, CDS size, protein molecular weight, and presence of a histone-fold domain. Of the 21 loci that were identified, we isolated cDNAs for 19, being unable to clone AtTAF1b and AtTAF11b. We were, however, able to clone all four AtTAF6 splicing variants, and these were included in our analysis. Although we were unable to isolate a full-length cDNA of TAF1, we did observe a ∼6 kb PCR product in reactions targeting the full length CDS. This product was recalcitrant to further amplification and cloning in BP reactions (Gateway system, Invitrogen). We, therefore, cloned TAF1 in three separate overlapping fragments and smaller subfragments: TAF11–884, TAF1587–1417, TAF11326–1919 and specific regions corresponding to the ubiquitin domain (TAF1UbD; aa 587–884) and the bromodomain (TAF1BD, aa 1801–1919). The 22 cDNAs that we were able to isolate were incorporated into our yeast two-hybrid protein-protein interaction matrix.

The MaV204K strain was selected for bait transformations because it carries the SPLA10::URA3 reporter which allowed testing for spurious activation on 0.1% 5′FOA medium (Boeke et al. 1984). Thus, bait constructs in MaV204K were plated on SD -Trp, +0.1% 5′FOA. In the event that the bait protein contained an activation domain, the URA3 gene product would catalyze the formation of a toxic product from 5′FOA, preventing yeast growth and indicating an unsuitable bait construct for yeast two-hybrid analysis. The AtTAF7, AtTAF12, and AtTAF15 bait constructs were spurious activators defined by the lack of growth on 5’FOA containing plates. These bait constructs were thus excluded from further studies. It was also found that the TAF11326–1919 prey construct interacted with the empty vector Gal4 DBD control leading to the omission of this construct from the final matrix. The AtTAF12 prey construct also caused the activation of the reporter genes, and thus, AtTAF12 was cloned as N-terminal (aa 1–200), middle (aa 201–394), and C-terminal (aa 395–539) fragments into the pENTR/D-topo vector and subcloned into the bait and prey vectors as detailed above. The bait proteins that interacted with the Gal4 AD alone (AtTAF11–884, AtTAF121-200, AtTAF12395–538 and AtTAF15b) formed colonies with a frequency of over 70% of their matings and were, therefore, excluded from analysis.

The screening of bait and prey constructs was conducted in two steps. The first consisted of an initial screen based on the number of colonies per selection plate in a dilution series of matings coupled with enzymatic assays for β-Gal activity. Pair-wise interaction partners identified in the mating assays were doubly transformed into yeast strain AH109 and reassessed for protein interaction. This second test consisted of a simple positive or negative assessment of colony growth by the double transformants on selection plates and by detection of α-Gal activity visually by the formation of blue colonies. Table 2 shows all interactions and denotes positive tests as follows: (1) initial yeast mating-based two-hybrid growth; (2) β-Gal activity higher than controls with 95% confidence; (3) growth of doubly transformed AH109 on selective media; and (4) α-Gal activity visible from AH109 double transformants. More than 70% of the interactions identified in the first round of screening (matings) were reconfirmed by two or more of the subsequent tests. Of the interaction tests listed in Table 2, 506 were conducted in reciprocal fashion. Considering interactions in both bait/prey combinations, there were 82 positive interactions, with 48.8% being either reciprocal or representing homodimer formation. If reciprocal interactions are counted as one, a total of 65 discrete interactions were detected between the subunits of Arabidopsis TFIID.

Table 2 Yeast two-hybrid protein–protein interaction matrix between subunits of the Arabidopsis TFIID complex
Full size table

β-Gal activity from the initial mating screen provided a quantitative estimate of interaction strength. The averaged β-Gal assay values passing the Student’s T-test with 95% confidence are shown in Fig. 1. More than half of the 11 strongest interactions (activity >2 units) comprise histone-fold binding pairs. This supports a common theme with yeast and metazoan TFIID that the histone-fold binding pairs are key structural components of the complex.

Fig. 1
figure 1

Colorimetric assays of the β-Gal reporter levels in yeast diploids containing both bait and prey plasmids. Liquid culture enzyme assays were performed to obtain semi-quantitative data on the strength of identified protein–protein interactions. Striped bars indicate negative controls of empty bait and prey vectors (pGBKT7/pGAD-T7) and two control baits with Gal4-AD (LAM/pGAD-T7; 53/pGAD-T7). Black bars indicate the negative control for each bait construct with the empty prey vector (pGAD-T7). Grey bars are interactions yielding β-galactosidase activities with activity values significantly above appropriate controls at the 95% confidence level. Assay values are grouped by bait vectors denoted by brackets. Prey constructs are listed directly below each individual bar

Full size image

While the colony growth and β-Gal assays were found to be useful, control experiments were conducted in order to evaluate the significance of the results. For instance, the highest β-Gal activity that we observed was with an AtTAF11326–1919 homodimer (data not shown). AtTAF11326–1919 as a prey construct was originally found to interact with AtTAF11801–1919 and itself, and this was confirmed by the co-transformation test (second screening). However, Gal4AD-AtTAF11326–1919 was also found to interact with the empty pGBKT7 vector and weakly with AtTAF11, a protein that previously showed no interactions. Furthermore, the Gal4 DBD-AtTAF11326–1919 construct failed to interact with any prey constructs with the exception of the above mentioned Gal4AD-AtTAF11326–1919. This raises an additional question regarding the validity of the Gal4AD-AtTAF11801–1919 interaction. Because of the potential for self-activation or non-specificity, the results from the Gal4 AD-AtTAF11326–1919 assays were discounted as indications of biologically significant interactions and have not been included in Fig. 1 or Table 2. In addition, it must be noted that β-Gal activities of prey constructs may include an undetermined contribution from weak self-activation.

Our observed positive interaction rate of 11.8% is broadly in agreement with the results of Yatherajam et al. (2003) in their evaluation of yeast TFIID subunit interactions, where 17% of the potential contacts tested resulted in strong to intermediate interactions in the two-hybrid assay, and another 8% demonstrated possible weak interactions. In general, the relatively low number of interacting partners among the Arabidopsis TFIID subunits suggests that interactions were direct and not secondarily mediated by endogenous yeast general transcription factor components. Were molecular bridging occurring, interactions between two proteins that have a common interactor would be expected. For example, AtTAF4b interacted strongly with both AtTAF9 and AtTAF12b. If molecular bridging of AtTAF9 with AtTAF12b occurred through the endogenous yeast TAF4, then an AtTAF9-AtTAF12b interaction would have been observed, and it was not (Table 2).

Analysis of protein expression

Immunoblots to detect bait and prey proteins demonstrated a large variability in protein expression levels regardless of indications of interactions with other TAFs. Approximately half of the GAL4 AD-fusion proteins were readily detectable, but some appeared to have altered mobility forms (Supplementary Fig. 1A). In contrast, only a few GAL4 DBD-fusion proteins were detectable (AtTAFs 8, 10, 13, 14 and 14b; Supplementary Fig. 1B). No significant correlation between colony growth and the presence of detectable bait or prey proteins was evident. For instance, TAF5 was not detected in either blot; however, it readily interacted with several other AtTAFs in the matrix (Table 2).

It is possible that the lack of colony growth in the yeast two-hybrid assay was solely due to low levels of expressed protein in those constructs that showed no detectable protein by western analysis. However, it is unlikely that any of the bait or prey constructs expressed no protein at all, since each construct (with the exception of AtTAF2) supported growth in at least one mating combination tested here and in a larger study that included subunit proteins of other Arabidopsis general transcription factors (Lawit et al., manuscript in preparation).

Yeast complementation assay

In an attempt to test function of the putative AtTAFs, complementation tests were conducted to determine if Arabidopsis TAFs could substitute for their homologs in yeast. Four AtTAF cDNAs were cloned into a yeast expression vector (pRS314-derived, Stratagene) and introduced into TAF mutant yeast strains containing the corresponding yeast gene residing on a vector harboring the URA3 gene (Klebanow et al. 1996). Growth of transformants on 5′FOA plates resulted from exclusion of the yTAF-bearing plasmid and was indicative of complementation. Positive and negative controls were conducted using the appropriate yTAFs. The results shown in Table 3 demonstrate that none of the Arabidopsis TAFs tested (AtTAFs 9, 10, 11 and 12) were able to substitute for their yeast homologs. Although it is possible that none of the four putative AtTAFs are genuine subunits of TFIID, a more likely explanation is that despite much overall sequence conservation, TAFs in yeast and plants have diverged to the extent that critical functions have undergone organism-specific specialization. This is supported by the finding that human TAF9 could not complement the yeast TAF9 as well.

Table 3 Yeast TAF complementation tests
Full size table

Discussion

Although individual TFIID components may have diverged during the evolution of the major eukaryotic lineages, it appears that many of their protein-protein interactions have been conserved. The majority (39 of 65) of protein–protein contacts identified in this study have been described for homologs in other systems (Burley and Roeder 1996; Dubrovskaya et al. 1996; Gangloff et al. 2000, 2001a, b; Hisatake et al. 1995; Jacq et al. 1994; Klebanow et al. 1996; Kokubo et al. 1993a, b, 1994, 1998; Lawit and Czarnecka-Verner 2004; Mengus et al. 1995; Uetz et al. 2000; Verrijzer et al. 1994; Weinzierl et al. 1993a, b; Yatherajam et al. 2003). However, unique interactions have also been detected such as the potential for AtTAF5 and AtTAF8 homodimerization, and the affinity of AtTAF8 for AtTAF13 and AtTAF14. A prominent feature of the interaction matrix is the promiscuity of the histone fold domain (HFD)-containing TAFs, in particular AtTAF4b and AtTAF12b, which is greater than that typically seen in other eukaryotes.

In contrast, several other putative TFIID subunits exhibited little or no interactions within the matrix. For example, AtTBP (1 and 2) only showed reproducible affinity for TAF1 (AtTAF11–884 fragment), while in other eukaryotes TBP has been shown to bind a total of nine TAFs (Bai et al. 1997; Dubrovskaya et al. 1996; Dynlacht et al. 1993; Hisatake et al. 1995; Kokubo et al. 1994; Mengus et al. 1995; Tao et al. 1997; Verrijzer et al. 1994; Weinzierl et al. 1993a; Yatherajam et al. 2003; Yokomori et al. 1993). Limited interactions were also seen with AtTAF1, which only showed affinity for TBP (AtTAF11–884) and AtTAFs 7 and 14b (AtTAF1BD and AtTAF1UbD, respectively). However, as many as 11 TAFs have been reported in other organisms to show affinity for TAF1 (Dikstein et al. 1996; Dubrovskaya et al. 1996; Jacq et al. 1994; Kokubo et al. 1994; Lavigne et al. 1996; Mengus et al. 1995; Verrijzer et al. 1994; Weinzierl et al. 1993a; Yatherajam et al. 2003; Yokomori et al. 1993). A third example of a TAF with few interactions is AtTAF2, which only showed weak affinity for AtTAF11–884. Its homolog in yeast has been shown to bind five subunits (TAFs 3, 4, 7, 8 and 10), and the metazoan homolog has affinity for four (TBP, 4, 4b and 11) (Dikstein et al. 1996; Verrijzer et al. 1994; Yatherajam et al. 2003; Yokomori et al. 1993). The absence of conclusive interactions between AtTAF2 and other TFIID subunits in our study was unexpected since in metazoans a TAF2-TAF1 complex makes specific contact with the Initiator element (Inr) of the promoter (Chalkley and Verrijzer 1999). A recent survey in plants has identified what appears to be an Inr motif in some promoters (Shahmuradov et al. 2003; Shahmuradov et al. 2005). This suggests the potential for conservation of the one known function of TAF2; however, our yeast two-hybrid results failed to provide solid support for this prediction.

The structural and biological significance of these potentially absent interactions in Arabidopsis TFIID is not clear. It is also not known whether poor expression or instability of the AtTAFs in yeast contributed to these negative results, since western blots for many of the AtTAFs indicated that expression in yeast was either low or not detectable (Supplementary Fig. 1). At least one of the AtTAF subfragments (AtTAF1587–1417) that did not bind other TFIID subunits in our study was capable of protein interactions, however, since it scored positive for contacts with the small subunit of TFIIF (not shown).

No AtTAF6 interactions that have been previously described in other organisms (with TBP and TAFs 1, 5 and 9) were observed in our analysis; however, various isoforms of AtTAF6b interacted with AtTAFs 5 and 9. For AtTAF6, the lack of interactions in our analysis raises the possibility that this protein is not a bona fide subunit of TFIID, but may be a component of other TAF-containing complexes, analogous to the role played by TAF6L in metazoans (Tora 2002). TAF6 is frequently cited as the dimerization partner of TAF9 (Frontini et al. 2005; Xie et al. 1996), but our study showed that certain splicing variants of AtTAF6b (but not AtTAF6) bind AtTAF9. While both AtTAF6 and AtTAF6b are predicted to form HFDs in their N-termini, there are a number of amino acid substitutions between the two in this region that may account for lack of interaction between AtTAF6 and AtTAF9. Only three of the AtTAF6b isoforms we identified (1, 2 and 4) showed affinity for AtTAF9. The region predicted to form the HFD is spliced out in the non-interacting isoform (AtTAF6b-3), strengthening the idea that the other variants are interacting via their intact HFDs. Taken together, these data raise the possibility that AtTAF6b is the true TAF6 homolog in Arabidopsis.

TAF1 has been shown to interact with TAF7 in yeast and human cells (Lavigne et al. 1996; Yatherajam et al. 2003). Our studies also indicated that AtTAF1 has affinity for AtTAF7, but only with a small fragment containing the ubiquitin-like domain (AtTAF1UbD). Human TAF7 binds a region that is C-terminal to the hTAF1 acetyltransferase domain, a site that also has affinity for TFIIF (RAP74) (Gegonne et al. 2001). However, this region of TAF1 is poorly conserved between metazoans and Arabidopsis, and our two-hybrid results indicated that the site of AtTAF7 interaction is N-terminal to the acetyltransferase domain. The larger AtTAF1 fragment spanning amino acids 1–884 contains the UbD, but could not be directly tested against AtTAF7 since both were self-activators in yeast. Subclone AtTAF1587–1417, which also contains the UbD (Fig. 2), did not interact with AtTAF7 for unknown reasons, perhaps due to improper folding.

Fig. 2
figure 2

N-terminal domains of Arabidopsis (At, AF510669) and rice (Os, BAD37604.1) TAF1 proteins show similarity to the TAND-1 and TAND-2 regions of Drosophila and human TAF1. The TAND-1 similarity was only evident for the N-terminal region (A); however, alignment using a library of hidden Markov models (Gough et al. 2001) identified significant alignment between the N-terminus of Drosophila TAF1 and AtTAF1. Alignments shown were obtained using CLUSTAL W (Thompson et al. 1994). In the top panels, Arabidopsis and rice sequences are aligned with the TAND-1 (A) and TAND-2 (B) core sequences from Drosophila and human TAF1. (C) The shaded boxes represent homology to the Drosophila sequence, and the open boxes are the predicted TAND regions based on the conservation of amino acid sequences between Arabidopsis and rice. (D) Map represents the AtTAF1 fragments tested in the yeast two-hybrid assay. Legend: TAND, TAF N-terminal domain; UBQ, ubiquitin domain; AT, acetyltransferase domain; Zn, Zn finger; BD, bromodomain

Full size image

Is there a connection between levels of AtTAF7 in the nucleus and the response of plants to high temperature stress? Microarray analysis of Arabidopsis transcription (Genevestigator website, Zimmermann et al. 2004) showed that AtTAF7 is induced 6 to 9-fold at 1 h following heat shock. This response is most notable in cell cultures, but is also seen in aerial tissue and roots. Does this potential elevation in AtTAF7 protein play a role in the heat shock response? It has been shown that human TAF7 serves as a check-point in the process of transcriptional initiation (Gegonne et al. 2006). It is bound to hTAF1 during assembly of the preinitiation complex and inhibits the acetyltransferase activity of hTAF1. Upon arrival of RNAPII to the promoter, hTAF1 autophosphorylates. This causes a dissociation of hTAF7 and a resulting derepression of acetyltransferase activity as transcription is allowed to proceed. In transient assays, overexpression of human TAF7 inhibits transcription at hTAF1-dependent promoters (Gegonne et al. 2006). If this inhibitory role of hTAF7 is conserved in plants, increased expression of AtTAF7 is predicted to inhibit activity of Arabidopsis TAF1-dependent promoters in a similar manner. In yeast, 90% of the genes have promoters that are primarily dependent on TFIID, a group that is largely comprised of housekeeping genes that show basal expression and a limited capacity for induction (Huisinga and Pugh 2004). The remaining 10% have promoters that largely rely on SAGA and are often stress inducible (including heat inducible). This raises the interesting possibility that elevation in AtTAF7 expression may provide a mechanism for the global down-regulation of housekeeping genes in Arabidopsis during periods of heat stress.

Comparison of the Arabidopsis interactions with the yeast TFIID map

The protein–protein interactions observed in our experiments agree well with the trilobed structure of yeast TFIID based on electron microscopy and immunolocalization studies conducted by Leurent et al. (2002; 2004). Two copies of yTAF5 are postulated to form a scaffold for the complex (Tao et al. 1997), with lobes A and B representing the two C-terminal domains and lobe C resulting from the interaction of the N-terminal domains. Although yTAF5 did not homodimerize in the study by Yatherajam et al. (2003), AtTAF5 did in our analysis. That AtTAF5 interacts with itself and HFD AtTAFs 4b, 6, 6b-1, 8, 9 and 12b is consistent with its proposed role as a linker molecule between lobes A-C and C-B of TFIID (Leurent et al. 2004). All of the TAF5 proteins contain a C-terminal region consisting of a number of WD40 repeats and, additionally, a WD40-associated region in their N-termini. We propose that the N-terminal WD40-associated region represents a homodimerization motif for TAF5 contributing to its function as the major scaffold component of TFIID. This prediction is based on our interaction analysis, the evidence that TAF5 is present in two copies in yTFIID (Kirschner et al. 2002; Leurent et al. 2004) and the likelihood that the C-terminal repeats of TAF5 are involved in HFD-containing TAF interactions.

In Fig. 3, we have combined all but one of our interaction results with the yeast topological model (excluding homodimer results). Interactions between the Arabidopsis TFIID components are largely consistent with predicted subunit interactions, as expected if TFIID structure and function have been conserved between the eukaryotes. The AtTAF4b/AtTAF7 interaction we observed may be an exception to the yeast model, since yTAF7 is localized to lobe A, and yTAF4 is present only in lobes B and C. Alternatively, this may reflect a difference between Arabidopsis TAF4 and 4b (yeast has no 4b), or simply represent an interaction possible in the context of the yeast two-hybrid system, but not occurring in Arabidopsis. Consistent with the later interpretation is the finding by Yatherajam et al. (2003) that, contrary to the topological model, yTAF7 interacts with yTAF4 and yTBP (among other subunits) in the yeast two-hybrid system.

Fig. 3
figure 3

Arabidopsis interactions overlaid onto the topological model of yeast TFIID presented by Leurent et al. (2004). The protein–protein interactions of Arabidopsis TFIID subunits were determined by yeast two-hybrid analysis, α-Gal visual assays and β-Gal quantitative assays. The identified protein interactions are simplified to show only single AtTAF representations for both “a” and “b” forms. Interactions identified in our experiments are denoted by black arrows. Interactions of TAFs 14 and 15 observed in Arabidopsis were not examined by Leurent et al. (2004)

Full size image

Three TAFs were not included in the yeast model: TAF2, TAF14 and TAF15. Although no organism has previously been reported to contain both TAF14 and TAF15, Arabidopsis contains two homologs of each, based on domain homology (Lago et al. 2004; Lawit and Czarnecka-Verner 2004). The yeast genome does not encode a TAF15 protein, which has been studied only in mammalian systems. However, the interactions of AtTAF15/15b strongly suggest their localization to a lobe of TFIID containing the AtTAF4/12 HFD pair (B or C). We chose arbitrarily to locate AtTAF15 in lobe C (Fig. 3); however, since its stoichiometry is unknown, it may reside in either or both lobes.

Arabidopsis TAF14/14b appears to be present in each of the three lobes of TFIID. Unlike TAF15, TAF14 has been well described in yeast, and biochemical evidence has determined that it is present in three to four copies per TFIID complex (Sanders et al. 2002). TAF14 homologs have been tentatively identified in Arabidopsis, but because the homology is based on a YEATS domain, which is shared by a number of other proteins, assignment to TFIID was uncertain (Lago et al. 2004). We have demonstrated that these two putative TAFs interact with a number of HFD TAFs and with AtTAF5, which strongly suggests that these proteins are indeed subunits of Arabidopsis TFIID or other TAF-containing complexes like SAGA. The large number of interactions found for AtTAF14/14b (Table 2) and its stoichiometry of 3–4 proteins per complex in yeast (Sanders et al. 2002) led to our placement of AtTAF14/14b in each of the three lobes (Fig. 3). This was the only arrangement that could accommodate each of the observed two-hybrid interactions, except for possible AtTAF14 and/or 14b dimer formation. In yeast, TAF14 is a component of TFIID, TFIIF, and the chromatin remodeling complex SWI/SNF (Cairns et al. 1996). We have shown using yeast two-hybrid analysis that AtTAF14 and 14b interact with subunits of TFIIA, TFIIB and TFIIF, further strengthening our prediction that AtTAF14/14b are true Arabidopsis homologs of TAF14 (Lawit et al., manuscript in preparation).

In our study, 26 novel interactions were identified that may provide insight into the structural arrangement of TFIID. AtTAF8 and AtTAF10 both have HFDs and, as reported in yeast and Drosophila (Gangloff et al. 2001a; Hernandez-Hernandez and Ferrus 2001), interact strongly with each other. Both proteins also independently form dimers (which is a novel finding for TAF8). This suggests possible formation of a TAF82-TAF102 tetramer in TFIID. Several other Arabidopsis HFD TAFs also form dimers in the yeast two-hybrid system: TAFs 4b/4b (novel), 12/12b, and 12b/12b. Additionally, TAFs 4 and 12, and TAFs 8 and 10 have also been shown to form HFD binding pairs (Hernandez-Hernandez and Ferrus 2001; Hoffmann et al. 1996; Kokubo et al. 1994; Uetz et al. 2000; Yatherajam et al. 2003; Yokomori et al. 1993). These interactions suggest the possibility that heterooctamers of these subunits may be present in TFIID. However, Leurent et al. (2004) have argued against the existence of these types of structures as part of TFIID in vivo despite in vitro formation of a TAF6/9 and TAF4/12 octamer (Hoffmann et al. 1996; Xie et al. 1996). Their reasons are as follows: (1) yTAFs 4 and 12 co-localize to more than one lobe in immuno–mapping studies. The presence of a tetramer on either lobe would violate the stoichiometry previously determined (Sanders et al. 2002). (2) Additional HFD yTAFs are also present in these same lobes, which argues against a clean nucleosome-like octamer. Although analysis of pair-wise combinations in the yeast two-hybrid system reveals potential for interaction, these authors suggest that associations formed during the assembly of TFIID result in structures of higher complexity than the nucleosomal octamer.

AtTAF1

Arabidopsis is unusual in that two isoforms of TAF1 are present. AtTAF1 exhibits a domain structure similar to that present in other eukaryotes. AtTAF1b is closely related to AtTAF1, but lacks 91 amino acids at the N-terminus. AtTAF1 and AtTAF1b were originally identified in a data base search for histone acetyltransferase (HAT) containing proteins (Haf1 and Haf2, respectively) (Pandey et al. 2002). Expression analyses have demonstrated that both AtTAF1 and AtTAF1b transcripts are present in most tissues (Bertrand et al. 2005; Lago et al. 2004) (https://www.genevestigator.ethz.ch). Mutational studies suggest that a degree of functional redundancy may exist between AtTAF1 and 1b, since knockout mutants of AtTAF1b (haf2-1) showed only mild alterations in growth. These plants grew normally and were fertile; however, there were reductions in chlorophyll in the cotyledons and in expression of two light-responsive genes.

In our assay, both isoforms of AtTBP interacted only with the N-terminus of AtTAF1, a region of the protein predicted to contain a TAND domain (TAF1 N-terminal domain), which in yeast and Drosophila binds TBP (Kokubo et al. 1998; Liu et al. 1998). Although alignment at the amino acid sequence level of TAF1 from yeast, Drosophila, human, and Arabidopsis has been reported to show no strong conservation (Lago et al. 2004), we have identified what we believe to be similarity between a region in the N-terminus of AtTAF1 and rice TAF1 (OsTAF1) with TAND domains 1 and 2 of Drosophila TAF1 (Fig. 2). Sequence similarity between the N-terminus of Drosophila TAF1 (containing TAND-1) and the N-terminus of AtTAF1 was indicated using a program which utilizes a hidden Markov model to align a library of domains for which structure has been determined (http://supfam.org/SUPERFAMILY, Gough et al. 2001). The sequence similarity is strongest at the N-terminal portion of TAND-1 (designated TAND-1 core, Fig. 2A). A TAND-2-like region was identified using CLUSTAL W to align the isolated Drosophila TAND-2 domain to AtTAF1 (Thompson et al. 1994). The N-terminal region of AtTAF1b was then aligned with AtTAF1 for comparison (Fig. 2B). A limited amount sequence similarity to TAND-2 can be seen near the N-terminus of AtTAF1b, which implies that AtTAF1b is an N-terminally truncated form of AtTAF1 that begins near the start of the TAND-2-like domain. Although it has not been demonstrated that AtTAF1 inhibits AtTBP binding at the promoter, as is the case in Drosophila and yeast proteins (Banik et al. 2001; Kokubo et al. 1998), the interaction of the N-terminus of AtTAF1 with both AtTBPs, coupled with the strong sequence identity with TAF1 from Drosophila, suggest that the TAND-like region at the N-terminus functions in a manner analogous to its counterparts in metazoans and yeast. It should also be noted that the TAND-1 and 2 cores have been strongly conserved between Arabidopsis and rice (Fig. 2A, B). Although the TAND-1-deleted form is not present in rice, N-terminal truncations of TAF1 derived from transcriptional and posttranscriptional processes have been detected at low levels in yeast (Kasahara et al. 2004).

Summary

This work represents the first attempt to compare the structure of a plant TFIID complex with that determined for other eukaryotes. While no TAF3 homolog has been observed in plants, at least one homolog has been identified for each of the other 15 TFIID components. Among the proteins examined were two showing homology to yTAF14 and two to hTAF15. This combined with our interaction data suggest that Arabidopsis TFIID is unique in containing both TAF14 and TAF15. The presence of TAFs 14 and 15 in plants may indicate ancient roles for these proteins that were lost in metazoans and yeast, respectively.

A total of 65 discrete binary interactions were identified, including 26 novel protein interactions. With few exceptions, these interactions are consistent with the topological model of yeast TFIID (Leurent et al. 2002; Leurent et al. 2004). This finding strongly implies that the superstructure of TFIID has been evolutionarily conserved between yeast and plants even though the individual subunits have diverged. It is also likely that some combination of AtTAFs may co-localize to plant SAGA (Stockinger et al. 2001) or other transcriptional co-activators, as occurs in other eukaryotes. Thus, the apparent complexity of interactions revealed here by the yeast two-hybrid analysis may reflect a composite of protein contacts possible in a variety of in vivo regulatory complexes in plants.