The Arabidopsis genome possesses genes that encode most of the polyadenylation factor subunits that have been described in other eukaryotes (Table 1; 33). Possible exceptions to this include the absence of orthologs to CFIm59/68 and Hrp1. However, this is probably due to an inability to identify, using BLAST, authentic orthologs in the large array of SR+RRM- or RRM-containing proteins encoded by the Arabidopsis genome. Many of these genes and their protein products have been studied previously. Moreover, with a few exceptions (discussed in the following), the expression of these genes can be seen in microarray studies. For the majority of these proteins, the sequence similarity with other eukaryotic counterparts (such as their human homologs) is extensive, suggestive of a conservation of function that has been preserved in the different eukaryotic lineages. However, a subset of the plant factors shares a more limited similarity with their eukaryotic counterparts. These proteins include AtCPSF30, AtCSTF64, and the FIPS and PCFS (Table 1). With these proteins, functional motifs are conserved, but other parts show sizable sequence divergence.

Although polyadenylation is expected to be essential for growth and development, the nature of some mutants impaired in Arabidopsis polyadenylation factor subunits 263134 raises the possibility that some plant polyadenylation-related genes are active at specific times during development, or in response to particular environmental cues. To explore this hypothesis, the expression of the set of Arabidopsis genes listed in Table 1 was studied using public domain microarray data. For this study, the data available from NASC (Nottingham Arabidopsis Stock Centre) was used; normalized expression values for most of the genes listed in Table 1 was extracted from the datasets listed in Additional file 1: microarray keys and data, and plotted so as to permit easy comparison. One of the Arabidopsis polyadenylation-related genes listed in Table 1 (At4g04885, AtPCFS4) is not represented in the Affytmetrix ATH probe set and was thus not included in this study. The complete results of this study are presented in Additional file 1. The most interesting and salient aspects of this study are discussed in more detail in the following.

As shown in Figure 1, the expression of most of these genes varied modestly at different stages of growth and development. The gene encoding AtPAPS3 was a pollen-specific gene (Figure 1C). Several genes showed elevated expression in developing seeds (this pattern is typified by the AtFIPS5 and AtCSTF77 genes, respectively) while others showed reduced seed expression (AtCPSF160 is an example). Curiously, a subset of these genes showed dramatically reduced expression in pollen; this set of genes includes those encoding AtCPSF160, AtCSTF77, and AtPABN3.

Of all the tissues and growth stages represented in Figure 1, pollen was the most different. To extend this observation, the expression of all of the genes listed in Table 1 (except for those not present in on the ATH chip) in pollen was plotted (Figure 2). This representation emphasizes the increased expression of AtPAPS3. As interesting, however, was the dramatic reduction in expression (more than 10-fold) of three other genes – AtCPSF160, AtCSTF77, and AtPABN3. Several other genes also had reduced levels of expression in pollen, suggestive of a tissue-specific gene expression program that may yield a modified polyadenylation complex.

A similar analysis of expression in response to various abiotic stresses revealed that most polyadenylation-related genes responded modestly, if at all, to the battery of stresses represented in the NASC dataset (Figure 3). For the most part, polyadenylation-related genes were unresponsive to chemical or hormone treatments (Figure 4). Cycloheximide, an inhibitor of protein biosynthesis, increased the expression of the AtPAPS1 and AtPAPS3 genes, suggesting that these mRNAs are relatively unstable 35. Many of these genes were affected by mutations in giberrellic acid-related pathways and were induced by imbibition, probably reflecting induction of expression upon germination. This was most predominant with the AtFIPS3 gene, the expression of which was rather GA and imbibition-dependent.

The responses of these genes to various pathogen-related stimuli (inoculation with bacterial of fungal pathogens, treatment with elicitors of defense responses) was modest, with no poly(A) – related gene showing more than 3-fold variation in response to the different treatments (Figure 5). Dark or different light treatments had little effects on the expression of these genes (sample 37–52 in Figure 5).

To better understand the functioning of the various plant polyadenylation factor subunits, a comprehensive set of pair-wise interaction assays was conducted. For this, a standard yeast two-hybrid approach was adopted. The protein coding regions for each of the Arabidopsis genes listed in Table 1 were cloned into the "AD" and "BD" yeast two-hybrid plasmids as described 3436. For most of these genes, the entire coding region was used. However, in some cases, the proteins were "broken" into domains, based on their predicted structures. This set of constructs (Additional file 2: Y2H constructs) was then used to collate an exhaustive pair-wise interaction map of the polyadenylation factor subunits. In these assays, both combinations of clones (e.g., AD-AtCPSF160 + BD-AtCPSF100 as well as the converse BD-AtCPSF160 + AD-AtCPSF100 combination) were tested whenever possible. Some combinations could not be tested, since several of the proteins possessed transcriptional activation domain activity in the yeast system (Additional file 2: Y2H constructs). Interactions were assessed by plating several double transformants from "non-selective" media (media that allows for identification of the double transformants) on which growth is possible only if there exists an interaction between the test subjects. All such tests included negative controls (cotransformation of the AD or BD recombinant with "empty vector" AD or BD plasmid) and positive controls [the SFN1/SFN4 combination 34, or the AtCSTF77-AtCSTF64 combination, reported as being positive 30, and confirmed in this study).

The results of this exercise are summarized in Additional file 3 (Yeast_2_Hybride_results). Of the 320 tested interactions, 56 (or 17.5%) proved to be positive. Limited confirmation tests suggest that these interactions are all authentic. Specifically, 15 independent tests, using in vitro or co-purification techniques, have confirmed the interactions (Table 2), and no tested two-hybrid interaction has been contradicted by other tests. Thus, the positive interactions listed in Additional file 3 are reliable.

The positive interactions (Additional file 3) were displayed using Cytoscape (Figure 6). From this exercise, it is apparent that the interaction network indicated by the two-hybrid study is extensively interconnected, as they are found to interact in the reciprocal yeast two-hybrid assays in most cases (e.g. AD-AtCPSF100 + BD-AtCPSF73-I, and BD-AtCPSF100 + AD-AtCPSF73-I; in some cases, due to self-activation of the BD constructs, such reciprocal tests were not possible). However, it does resolves itself into three hypothetical complexes, centered around AtFIPS5, AtCPSF100, and a putative CFIIm-like complex (consisting of AtCLPS and AtPCFS orthologs), respectively. The AtFIPS5 and AtCPSF100 subcomplexes are bridged by AtCPSF30, AtCFIS1, and three AtCSTF subunits. Additionally, AtCPSF30 links the CFIIm-like complex with the others. Interestingly, the four AtPAPS isoforms and the three AtPABN isoforms are all parts of the AtFIPS5 subcomplex, although one AtPAPS isoform (AtPAPS2) is also directly linked to the AtCPSF100 subcomplex. Also of interest, one CLPS and one CFIS isoform were positioned very differently from the other isoforms in the network. Thus, while AtCLPS3 was part of the CFIIm subcomplex, AtCLPS5 interacts independently with the two AtFIPS isoforms and with one (but only one) of the AtPAPS isoforms. While AtCFIS2 interacts with AtCPSF30, AtPAPS4, and AtFIPS5, the AtCFIS1 subunit interacts only with AtFIPS5.

The results of the meta-analysis of microarray data indicate that AtPAPS3 is a pollen specific gene, that AtPCPS1 and/or AtPCPS5 are probably restricted to small parts of the plants, and that pollen and seed have a reduced polyadenylation complex. When AtPAPS3 and AtPCFS1+AtPCFS5 are removed from the overall interaction network, very little changes as far as the overall topology is concerned (Figure 7). The CFIIm-like complex reduces to but two subunits and the FIPS-PAPS hub loses one PAPS, but the general layout and inferred functionalities are otherwise unchanged. This representation is the best estimate for the network that exists in most cells in the plant.

In contrast, the reduction of the pollen network is more substantial, as shown in Figure 8. This is apparent in the smaller size of the CPSF complex and FIPS hub. Of particular note is the absence of PABN and CFIS in the pollen network. However, these changes do not affect the overall topology of the network, which retains the CPSF and CFIIm complexes, the FIPS and CPSF30 hubs, and the bridging functions of two of the CSTF subunits.

As a general rule, the expression of polyadenylation-related genes in Arabidopsis is fairly consistent over a wide range of conditions (Figures 1, 3, 4, 5). However, some interesting exceptions to this rule exist (see Figure 1). The most interesting and striking exception is the pollen-specific expression of AtPAPS3; this gene encodes a putative cytoplasmic form of PAP, and the restriction of its expression to pollen is reminiscent of the involvement of cytoplasmic PAPs in spermatogenesis in animals 37. Interestingly, the protein encoded by AtPAPS3 is truncated with respect to the other three Arabidopsis PAPSs, as well as when compared with its eukaryotic nuclear counterparts. Moreover, this truncation leaves the protein without obvious nuclear localization signals. These observations suggest that AtPAPS3 is in fact a cytoplasmic enzyme, and plays functions during pollen development analogous to those fulfilled by the testis-specific cytoplasmic PAPs in mammals.

Two developmental states stand out when it comes to the expression of polyadenylation-related Arabidopsis genes. One of these is pollen. As noted above, one of the four Arabidopsis PAPs, AtPAPS3, is a pollen-specific gene. Remarkably, however, several other polyadenylation-related genes have normalized expression values in mature pollen that are less than 0.2 (Figures 1 and 2). These include the only genes for AtCPSF160, AtCSTF77, and the three PABN isoforms, as well as the AtCPSF73-I and AtCFIS1 genes. This observation suggests a different polyadenylation apparatus for pollen compared with other parts of the plant. Three of these subunits – AtCPSF160, AtCSTF77, and AtCPSF73-I – are core components of their respective complexes in mammals and yeast, and the prospect that polyadenylation can occur in their absence is surprising. However, removal of these seven nodes from the overall polyadenylation factor interaction network does not change the overall nature of the network in a fundamental way (Figure 8). The absence of AtCPSF160, which in mammals recognizes the AAUAAA hexamer, suggests that different polyadenylation signals are recognized in pollen compared with most other tissues in the plant. Regardless of the details, the tissue-specificity in gene expression suggests that the plant poly(A) apparatus is much more flexible than anticipated, capable of functioning with a reduced set of subunits. Of course, these considerations are predicated on the assumption that the diminished mRNA levels indicated by the microarray studies are reflected in reduced protein levels.

The other interesting developmental state is the seed. The genes encoding AtCPSF30 and AtCFIS1 have normalized expression values between 5 and 10-fold higher in seed; this is seen in several controls that study gene expression in the seed in response to ABA and imbibition (Additional file 1). This suggests a possible specialization of the polyadenylation apparatus in the seed. The possible significance of this is not clear; in other studies of the 3'-UTRs of seed-specific Arabidopsis genes, no clear nucleotide composition or sequence preference in these genes was seen (P. Thomas and A. G. Hunt, unpublished observations), apart from those that have been reported before 22. Thus, a possible link between polyadenylation complex architecture and novel poly(A) signal usage is not indicated. The significance of the distinctive expression pattern of these two genes will have to be established by additional studies.

The protein interaction network inferred from the yeast two-hybrid study resolves itself into three conceptual hubs. Two of these hubs recall biochemical studies of the polyadenylation apparatus in mammals and yeast. One hub is centered around AtCPSF100, and includes AtCPSF160, AtCPSF73-I, AtCPSF73-II, AtCPSF30, AtPAPS2, and FY. With the exception of FY (the mammalian counterpart of which has not been studied in this regard) and AtPAPS2, this hub corresponds to the classical CPSF complex, that in mammals includes CPSF160, CPSF100, CPSF73, and CPSF30. The two-hybrid results presented here are corroborated by other studies, providing a strong degree of confidence in this part of the network. Thus, the Arabidopsis CPSF subunit orthologs interact in vitro in a way that is consistent with the interaction network (Table 2; 34. The four canonical plant CPSF subunits (AtCPSF160, AtCPSF100, AtCPSF73-I, and AtCPSF30) as well as AtCPSF73-II (a relative of a subunit of the recently-characterized Integrator complex; 38) are present in nuclear extracts, indicative of their in planta expression and nuclear localization 31. These proteins reside in a protein complex, as demonstrated by coimmunoprecipitation studies (R Xu and QQ Li, unpublished data; 2731). Interestingly, FY is part of these complexes 31, lending support to the placement of this subunit as part of CPSF in plants.

From the protein-protein interaction patterns (Figure 6), it seems that both AtCPSF73-I and AtCPSF73-II interact only with AtCPSF100 among the polyadenylation factor subunits. Moreover, they do not interact with each other or form homodimers in the two-hybrid assays, and their in silico expression properties show some degrees of specialized expression (Figure 1; also, 34). These observations beg a question as to the relationship between the two AtCPSF73 orthologs. There are two possible models for their positions and functions in the complex. In one, in some tissues, both subunits are associated at the same time with AtCPSF100, in which case they are not competing for the same binding site on AtCPSF100. Alternatively, they could compete the same binding site on AtCPSF100, thus forming different complexes that exclude each other. This scenario should also apply to the tissues where these two genes are expressed differentially. Preliminary results of deletion experiments indicated that both AtCPSF73-I and II interact with the C-terminal quarter of the AtCPSF100 protein (R. Xu and QQ. Li, unpublished results), arguing for the existence of different complexes.

Symplekin was not included in the two-hybrid study, owing to some confusion at the outset of the project as to the nature of the apparent "split" gene encoding one of the two symplekin orthologs (this uncertainty remains, as discussed in 31. However, other studies have shown that symplekin resides in a complex that includes CPSF100, CPSF160, and FY 31; thus, symplekin would appear to be part of the CPSF complex indicated in the two-hybrid study.

Another hub that is indicated by the network analysis includes three of the PCFS isoforms and one of the CLPS orthologs. This hub corresponds to the yeast CFII complex that consists of Pcf11p and Clp1p, and to the corresponding mammalian CFII complex 2. In other eukaryotes, Pcf11p or its homologue bridges the polyadenylation apparatus and the C-terminal domain of RNA polymerase II, thereby promoting the polyadenylation-linked termination of transcription. The CTD-interacting domain found in other eukaryotic Pcf11p proteins is present in two of the Arabidopsis orthologs 39, suggestive of a similar bridging function. Likewise, the interaction between Rna14 and Pcf11p in yeast is recapitulated with one of the Arabidopsis Pcf11p orthologs (AtPCFS5; Figure 6). However, the expression studies indicate that this interaction may only apply to the hypothetical pollen polyadenylation complex; thus, in most parts of the plant, there may be no corresponding link between the plant CFII complex and CSTF77. This is also true for the Pcf11p-Rna15 interaction that has been seen in yeast. Whether this reflects a limitation of the two-hybrid assay or divergence in the sequence and function of Pcf11 proteins is not clear. In this regard, it is possible that the CLPS-CPSF30 interaction observed in this study may serve a similar bridging function between the polyadenylation complex and the CTD of RNA polymerase II.

In mammals, hClp1 appears to be a bridge between CFIm and CPSF 40. Such a role is not indicated by the results of this study. While the interaction of AtCLPS3 with AtCPSF30 is indicative of a link between the plant CFII complex and CPSF, there seem to be no direct physical links between either CFIS isoform and the plant CFII. Whether this discrepancy reflects limitations in the different approaches that have been used to assemble models for the polyadenylation complexes in different systems is not clear. However, resolution of the discrepancy with respect to the bridging functions of CLPS may reside in the absence of the larger CFIS subunit in the present study.

The third hub of plant polyadenylation factor subunits centers about AtFIPS5, and includes the four PAPS isoforms, the three PABN variants, and single members of the CFIS and CLPS subunit families. This hub has no obvious counterpart in the commonly-presented view of the mammalian polyadenylation apparatus (in which Fip1 is placed as part of CPSF) or in the yeast polyadenylation apparatus (in which Fip1 resides as a part of CPF). However, the interactions of the PAPS isoforms with AtFIPS5 in Arabidopsis recalls the function of Fip1 in yeast in recruiting PAP to the rest of the complex. The "FIPS hub" involves a number of proteins that are members of protein families – PAPSs and PABNs, to be specific. Moreover, with the exception of AtCSTF64, all of the interactions with AtFIPS5 involve the N-terminal 137 amino acids of the protein. It is unlikely that the sum total of interactions inferred from the two-hybrid analysis occur in a single complex; rather, a small subset of these interactions may be in force at a given moment.

Similar considerations factor into the discussion of the interactions involving AtCPSF30. While small in size, AtCPSF30 interacts with many proteins in the complex (Figure 6), including AtCPSF160, AtCPSF100, AtCFIS2, both FIPS orthologs, AtPCPS1, AtPCFS5, and AtCLPS3. It may be that AtCPSF30 is a hub around which these other subunits assemble in a large, static complex. However, AtCPSF30 seems to be too small for all these proteins to bind at once. An alternative model would involve a scenario whereby these various interactions reflect a progression through the steps of the polyadenylation reaction. These considerations reinforce those raised above, and lend themselves to a model of the plant polyadenylation complex as a dynamic system that changes its subunit composition, either as a means of recognizing different RNA substrates, interacting with other processes (such as small RNA biogenesis or transcription-related events), or progressing through the polyadenylation reaction. It is also possible that different complexes are involved alternative polyadenylation of mRNA.

Perhaps the most obvious possible difference between the predicted plant complex and the mammalian counterpart lies in the relationships of CstF subunits with other members of the complex. In mammals, CstF50 is part of an identifiable heterotrimeric complex and interacts physically with another subunit of the complex, CstF77. No such interaction is seen in the two-hybrid analysis nor in other in vitro studies 30, and the position of AtCSTF50 in the Arabidopsis network suggests that this protein is not a part of complex comparable to CSTF. AtCSTF50 does interact with CPSF, AtCSTF64, and PAPS, suggesting a novel bridging or assembly function. But such a role would seem to be different from that played by this protein in the mammalian polyadenylation complex.

The approach that we are taking can only identify the proteins that share homology with known mammalian and yeast polyadenylation factors. It is possible there are other proteins that may not share amino acid sequence homologies but functionally conserved. It is also equally possible that plants use additional proteins in the cleavage and polyadenylation process. These possibilities should be explored using different means, e.g. protein 3-D structure alignment search, proteomic and genetic approaches.