Molecular Basis of RNA Recognition by the Embryonic Polarity Determinant MEX-5*

Embryonic development requires maternal proteins and RNA. In Caenorhabditis elegans, a gradient of CCCH tandem zinc finger (TZF) proteins coordinates axis polarization and germline differentiation. These proteins govern expression from maternal mRNAs by an unknown mechanism. Here we show that the TZF protein MEX-5, a primary anterior determinant, is an RNA-binding protein that recognizes linear RNA sequences with high affinity but low specificity. The minimal binding site is a tract of six or more uridines within a 9–13-nucleotide window. This sequence is remarkably abundant in the 3′-untranslated region of C. elegans transcripts, demonstrating that MEX-5 alone cannot specify mRNA target selection. In contrast, human TZF homologs tristetraprolin and ERF-2 bind with high specificity to UUAUUUAUU elements. We show that mutation of a single amino acid in each MEX-5 zinc finger confers tristetraprolin-like specificity to this protein. We propose that divergence of this discriminator residue modulates the RNA-binding specificity in this protein class. This residue is variable in nematode TZF proteins, but is invariant in other metazoans. Therefore, the divergence of TZF proteins and their critical role in early development is likely a nematode-specific adaptation.

may be governed by differences in the RNA-binding specificity of each protein. Consistent with this model, TTP is an exquisitely specific RNA-binding protein; it recognizes nonameric UUAUUUAUU sequences present in the 3Ј-untranslated region (UTR) of its targets (19,20,25,26). An NMR structure of the mammalian TTP homolog ERF-2 (also known as Tis11D) demonstrates that each finger individually recognizes a UAUU repeat (27). In contrast, an in vitro interaction between any of the TZF proteins from C. elegans and RNA has not been demonstrated, and as such their mRNA target specificity has not been explored.
MEX-5 and MEX-6 diverge from TTP in a few notable ways ( Fig. 1B): (i) nine amino acids rather than eight separate the first two cysteines in each zinc finger, (ii) the spacing between fingers is lengthened, and (iii) several highly conserved amino acids that contribute to RNA binding in mammalian TZF pro-teins are not conserved in MEX-5 and MEX-6. These differences could impact the ability of MEX-5 and MEX-6 to bind to RNA. Moreover, MEX-5 has been shown to interact with ZIF-1 protein in a yeast two-hybrid assay (16), suggesting that it may not regulate this factor at the RNA level. We set out to describe the RNA-binding properties of MEX-5 to probe its role in patterning the anterior-posterior axis.
Purification of Recombinant Proteins-TZF domains from MEX-5, MEX-6, and mutants thereof were expressed and purified from Escherichia coli JM109 cells as C-terminal fusions to maltose-binding protein. Liquid cultures grown at 37°C were induced in mid-log phase with 0.1 mM isopropyl 1-thio-␤-Dgalactopyranoside. Zinc acetate was added to a final concentration of 100 M at the time of induction. Harvested cells were resuspended in lysis buffer (50 mM Tris, pH 8.8, 200 mM NaCl, 2 mM dithiothreitol, EDTA-free protease inhibitor tablet (Roche Applied Science), 100 M Zn(OAc) 2 ) and lysed by sonication. Soluble protein was purified over an amylose column (New England Biolabs). Fractions containing the fusion protein were pooled and dialyzed into Q buffer (50 mM Tris, pH 8.8, 20 mM NaCl, 2 mM dithiothreitol, and 100 M Zn(OAc) 2 ) and then further purified over a Hi-trap Q HP column (GE Healthcare). Final purification was achieved by combining fractions containing the protein and dialyzing them into S buffer (50 mM MOPS, pH 6.0, 20 mM NaCl, 2 mM dithiothreitol, 100 M Zn(OAc) 2 ) before running them over a Hi-trap S HP column (GE Healthcare). Care was taken to minimize the amount of time the protein was exposed to pH 6.0 buffer. Pure fractions as determined by Coomassie-stained SDS-PAGE were combined and dialyzed into storage buffer (20 mM Tris, pH 8.0, 20 mM NaCl, 100 M Zn(OAc) 2 , 2 mM dithiothreitol). After dialysis, the protein concentration was determined using Beers law by measuring absorbance at 280 nm and a calculated extinction coefficient determined using the ProtParam server (28). The protein was concentrated to ϳ50 M before storage at 4°C.
Analytical Gel Filtration Chromatography-A Superdex 200 10/300 GL column (10 ϫ 300 mm, GE Healthcare) was used to determine the apparent molecular weight of MBP-MEX-5-(236 -350) and its mutants. The column was equilibrated on an AKTA FPLC with 2 column volumes of filtration buffer (50 mM Tris, pH 8.0, 300 mM NaCl) at a flow rate of 0.5 ml min Ϫ1 prior to loading the protein sample. Approximately 50 l of sample (17 M) was loaded on to the column and eluted with 1.5 column volumes of filtration buffer. Retention time was determined in relation to standards (Bio-Rad).
RNA Labeling Protocol-All of the RNA sequences used in this work were prepared by chemical synthesis and deprotected/lyophylized as the manufacturer directed (Dharmacon or Integrated DNA technologies). Lyophilized samples were  (15) with GFP reporters and by Priess and colleagues (14) using immunofluorescence in fixed embryos. The protein accumulates predominantly in the cytoplasm of the anterior blastomeres (a, ABa and ABp), but is also in P-granules (p) and on the centrosome of P 2 (c). The cells are labeled in the DIC image. b, domain structure of MEX-5. The location of the two CCCH zinc fingers (ZF1 and ZF2) is shown in blue. The numbers represent the primary amino acid sequence. Alignment of C. elegans zinc finger domains MEX-5, MEX-6, and POS-1 with mammalian TTP and ERF-2. The C. elegans proteins are boxed. Gray bars denote the three cysteines and one histidine that coordinate the zinc ion. The asterisks represent points of contact with RNA observed in the NMR structure of ERF-2 (27). The adenosine recognition pocket is boxed. The discriminator position is colored red for acidic side chains, blue for basic side chains, and green for hydrophobic side chains. Amino acids in italics were mutated in the present work. resuspended in 300 l of TE buffer, pH 8.0, and the concentration was measured by determining the absorbance at 260 nm using a calculated extinction coefficient based on the nucleotide content.
Fluorescein 5-thiosemicarbazide (Invitrogen) was used to 3Ј-end label each RNA via the method of Reines and Cantor (29). A typical 50-l reaction consisted of 0.5 nmol of RNA, 100 mM NaOAc, pH 5.1, and 5 nmol of NaIO 4 . After a 90-min incubation at room temperature, the sample was ethanol precipitated with 1 l of RNase-free glycogen (Invitrogen 20 g/l), 5 M NaCl (1/20 the volume), and 2 volumes of 100% ice-cold ethanol. The resulting pellet was resuspended in 50 l of 100 mM NaOAc, pH 5.1, containing 1 mM fluorescein-5-thiosemicarbazide. This reaction was incubated overnight at 4°C and unreacted label was removed using a Roche G-25 spin column. The labeling efficiency was determined by calculating the ratio of fluorescein absorbance at 490 nm to RNA-fluorescein absorbance at 260 nm. Typical efficiencies were 60 -80%.
Electrophoretic Mobility Shift Assays-Electrophoretic mobility shift assays were used to measure the binding activity of recombinant MEX-5 and MEX-6 to fluorescein-labeled RNA oligonucleotides. Typical reactions consisted of 2-4 nM labeled RNA equilibrated with varying concentrations of protein in equilibration buffer for 3 h. Equilibration buffer is 0.01% IGEPAL CA630 (a mild detergent used to prevent adhesion of protein and RNA to tubes, microplates, and gel wells), 0.01 mg/ml tRNA (a polyanionic nonspecific binding inhibitor), 10 mM Tris, pH 8.0, 100 M Zn(OAc) 2 , and 100 mM NaCl. The RNA was heated to 60°C and allowed to cool to room temperature before use. Immediately prior to loading, one-fifth volume of 30% (v/v) glycerol, 0.01% (w/v) bromcresol green was added to each reaction as a dye marker. A 40-l sample of each reaction (100 l total) was loaded onto a 1% agarose gel (EMD Biosciences, some lot to lot variability was observed) in 1ϫ TB buffer. The gels were run for 40 min at 120 volts then immediately scanned using a fluor-imager (Fujifilm FLA-5000) with a blue laser at 473 nm. The fluorescence intensity of unbound RNA was determined as a function of protein concentration using ImageGauge software. The data were fit to a sigmoidal dose-response function (Equation 1) to determine the halfmaximal saturation point (K d,app ), where is the fluorescence intensity, m is the maximal signal, b is the minimal signal, P is the protein concentration, and n is the apparent Hill coefficient. It is important to note that K d,app is not equivalent to the thermodynamic equilibrium dissociation constant for RNA sequences that contain multiple overlapping binding sites. In all cases, the reported value is the average of at least three experiments and the reported error is the standard deviation. Competition assays were performed as above except a constant concentration of subsaturating recombinant MEX-5 was used in the equilibration, whereas varying concentrations of unlabeled competitor RNA were added to the reactions. The apparent dissociation constant of the competitor RNA was determined by a fit of the data to a quadratic solution of the Lin and Riggs equation (30,31) as described (32).
Fluorescence Polarization Assays-Equilibration reactions (100 l volume) were set up using the same conditions as the electrophoretic mobility shift experiments above in 96-well black plates (Greiner). The apparent fluorescence polarization was determined using a Victor 3 plate reader (PerkinElmer Life Sciences) equipped with fluorescein-sensitive filters and polarizers. A total of five reads were measured for each experiment and the average and standard deviation of the millipolarization value (mP) were calculated for each protein concentration. The data were fit to Equation 1 (where represents polarization rather than intensity) to extract the apparent dissociation constant. The reported value is the average of at least three experiments and the error is the S.D.
Stoichiometric binding experiments were performed as above except the reactions were supplemented with unlabeled RNA to a final concentration of 1.5 M. The elevated concentration of RNA enables determination of the apparent stoichiometry by measuring the equivalent saturation point. This value was estimated by plotting polarization as a function of molar equivalents of protein to RNA and performing linear fits to pre-and post-saturation data. The equivalence point was determined by the intersection point of the two lines, and separately by a fit of the data to the quadratic equation as described (33).
UTR Sequence Analysis-C. elegans 3Ј-UTRs were retrieved from Wormbase release WS165. To determine the frequency of each possible octamer sequence in 3Ј-UTR-space, a Ruby script was written to enumerate each possible octamer and send it to the pattern searching tool PATSCAN (34). The PATSCAN output files were analyzed using standard UNIX text processing tools. The total number of occurrences of each class of A-, C-, G-, or U-containing octamers was determined by summing up the number of occurrences of each octamer in that class. The theoretical distribution of each class was determined by a binomial distribution weighted by the fractional proportion of each base in all C. elegans 3Ј-UTRs.
RNAi-C. elegans strains expressing GFP-MEX-5 (JH1448) or GFP-PIE-1 (JH1327) were obtained from the Caenorhabidits Genetics Center and cultured by propagating animals with the roller phenotype. Embryos were harvested from young adult hermaphrodites grown on OP50 food by bleach treatment and then deposited on NGM plates seeded with OP50 or NGMisopropyl 1-thio-␤-D-galactopyranoside plates seeded with mex-3 RNAi food (generously provided by Dr. Craig Mello) and cultured as described (35). Embryos were collected from gravid adult hermaphrodites by dissecting the worms in M9 on a 4% agarose pad with a fine gauge needle. DIC and GFP images were collected with live specimens using a Zeiss Axioskop microscope with ϫ40 or 100 objectives.

MEX-5 and MEX-6 Bind to ARE Repeat
Elements-To clarify the role of MEX-5 and MEX-6 in development, we set out to test whether these proteins, like TTP, bind with high affinity and specificity to RNA. A recombinant fragment of each pro-tein comprising the TZF domain was purified from bacteria and tested for the ability to interact with an established TTP binding sequence, the AU-rich element (ARE) of TNF-␣ mRNA (26). Two approaches were employed to measure affinity to this RNA: electrophoretic mobility shift (EMSA) and fluorescence polarization (FP) assays. Both methods reveal that MEX-5 and MEX-6 bind to TNF-ARE RNA. The EMSA experiments show that multiple binding sites are present in this sequence. The apparent dissociation constant (K d,app ) of MEX-5 for TNF-ARE RNA is 17 Ϯ 1 nM by EMSA and 14 Ϯ 4 nM by FP (Fig. 2, A and B, and Table 1). Similarly, MEX-6 binds to this RNA with an affinity of 4 Ϯ 3 nM by EMSA and 12 Ϯ 3 nM by FP (Fig. 2a). Both proteins are capable of binding to RNA with high affinity. Furthermore, because the EMSA and FP results are nearly equivalent, it is clear that both assays can effectively monitor RNA binding by these proteins.
TNF-ARE RNA contains several UUAUUUAUU repeat sequences and therefore can bind multiple molecules of TTP (26). To determine the affinity of MEX-5 for a shorter RNA variant containing only one TTP site, we repeated the binding analyses with ARE13 RNA (AUUUAUUUAUUUA). The apparent dissociation constant for this sequence is 55 Ϯ 15 nM by EMSA and 97 Ϯ 4 nM by FP (Fig. 2). In contrast, TTP binds to ARE13 RNA with 5-10-fold tighter affinity (26). Together, the results show that MEX-5 binds to UUAUUUAUU repeat sequences, but with an overall affinity that is weaker than TTP.
MEX-5 Binds to Regulatory Elements in glp-1 and nos-2 3Ј-UTRs-Prior work demonstrates that several factors are aberrantly expressed in mex-5 mutants (14). Ectopic expression of five proteins (SKN-1, PIE-1, MEX-1, POS-1, and PAL-1) and reduced levels of two others (GLP-1 and MEX-3) result from mex-5 mutation. Furthermore, MEX-5 is required to activate zif-1, which in turn targets TZF proteins for degradation (16). Last, recent studies reveal that MEX-5 is required for anterior degradation of mex-1, nos-2, and pos-1 transcripts (21,36). Although the results are consistent with MEX-5 regulating a network of maternal genes, this regulation has not been shown to be a direct result of MEX-5 binding to target mRNAs.
Extended UAUU sequence repeats similar to the TNF-ARE are not present in the 3Ј-UTR of any candidate MEX-5 regulatory target. However, several functional regulatory elements have been identified in the 3Ј-UTR of glp-1 and nos-2 mRNAs (Fig. 3, A and B). Two translational control elements are present in the 3Ј-UTR of glp-1 mRNA, a spatial control region, and a temporal control region (TCR) (37). Furthermore, five elements (subA-E) found in the 3Ј-UTR of nos-2 mRNA coordinate translational silencing, mRNA localization, and 3Ј-end formation (21). To determine whether MEX-5 binds these functional elements, we performed EMSA and FP experiments with 30 overlapping nucleotide fragments of the TCR and all five elements from the nos-2 3Ј-UTR (Fig. 3, Table 1). Surprisingly, MEX-5 binds with high affinity to all of the TCR fragments and three of the elements from nos-2 mRNA (TCR1-4, subA, subC, and subE, K d,app ϳ 25-100 nM). MEX-5 binds moderately to nos-2 subD (K d,app ϭ 200 Ϯ 20 nM) and poorly to nos-2 subB (K d,app 400 Ϯ 50 nM by FP, Ͼ1 M by EMSA).
Because a single shifted species is observed with TCR2 RNA, we decided to investigate the stoichiometry of the complex by repeating the FP experiments with elevated TCR2 RNA concentration (Fig. 3, c and d). The apparent stoichiometry is approximately 1 to 1 (equivalence point N is 0.9 Ϯ 1 by a quadratic fit), demonstrating that there is only one binding site in this RNA and that the recombinant protein is nearly 100% active (Fig. 3d). Consistent with these results, analytical gel filtration chromatography reveals that the TZF domain is predominantly monomeric at concentrations well above the apparent dissociation constant for this RNA sequence (17 M, K d,app ϭ 31 Ϯ 9 nM, Fig. 3d).
Inspection of the RNA fragments reveals that all of the interacting sequences contain a tract of 6 -8 uridines within an 8-nucleotide window. This feature is absent in nos-2 subB, suggesting that MEX-5 requires this element to bind. To test this model, a mutant version of the second TCR fragment where all of the uridines are replaced by adenosine was prepared. As expected, this sequence does not bind to MEX-5 ( Table 1). The results show that MEX-5 does not require UAUU repeats to bind to RNA with high affinity, and suggest that MEX-5 can bind to sequences harboring an extended uridine tract.
MEX-5 Binds to Polyuridine-TTP displays an 80-fold preference for UAUU repeat sequences over polyuridine (26). In contrast, our data show that MEX-5 binds with high affinity to uridine-rich RNAs lacking canonical ARE motifs. To test whether uridine nucleotides are sufficient to promote MEX-5 binding, EMSA and FP experiments were performed with a 30-nucleotide polyuridine sequence. MEX-5 binds to this RNA with an apparent K d of 29 Ϯ 6 nM by EMSA and 23 Ϯ 2 nM by FP (Fig. 4A). The data show that MEX-5 can bind to polyuridine with affinity similar to that of TNF-ARE. This demonstrates that MEX-5 binds with less specificity than TTP.
To further probe MEX-5 specificity, EMSA and FP experiments were performed with 15-nucleotide polyuridine, polyadenosine, polycytidine, and polyguanosine sequences. As before, MEX-5 binds to polyuridine with slightly weaker affinity than a similar length AUUUA repeat RNA (ARE13), but does not bind to polycytidine, polyadenosine, or polyguanosine ( Fig.  4a and Table 1). Together, the results show that uridine tracts are both necessary and sufficient to promote MEX-5 binding.
To identify the shortest RNA fragment that can bind to MEX-5, competition EMSA and FP experiments were performed with even shorter polyuridine sequences (U6 and U9).
a NA, not applicable. MARCH  Competition experiments were favored with shorter RNA sequences to prevent the possibility that steric hindrance by the 3Ј-fluorescein label would influence the apparent affinity. Disruption of the complex between MEX-5 and labeled ARE13 RNA was used to probe the affinity of unlabeled competitor sequences. The apparent dissociation constant of the competitor (K c,app ) was determined by a fit of the data to the Lin and Riggs equation (30,31). By self-competition, K c,app for ARE13 is equivalent within error to the K d,app determined by direct titration (Fig. 4B). In contrast, U6 and U9 are unable to compete for MEX-5 binding (K c,app Ͼ 1 M). Together, the results demonstrate that the minimal MEX-5 binding site is greater than 9 nucleotides in length. Uridine Tracts Are Remarkably Abundant in C. elegans 3Ј-UTR Sequences-It is possible that MEX-5 specifically regulates only those maternal genes that contain uridine tracts in their 3Ј-UTRs. To identify potential MEX-5 targets, we searched every annotated 3Ј-UTR in Wormbase release WS165 for octamer sequences containing at least 6 uridines using the PATSCAN algorithm (34). Amazingly, 91% of 3Ј-UTRs harbor at least one potential MEX-5 binding site. This demonstrates that C. elegans 3Ј-UTRs are remarkably rich in uridine tracts, and implies that MEX-5 alone cannot specify mRNA targets for regulation.

MEX-5 RNA Recognition
To further explore this hypothesis, we repeated the PATSCAN search with every possible octamer sequence (4 8 sequences) in the C. elegans 3Ј-UTR data base. There are 11,938 unique 3Ј-UTRs annotated in Wormbase. These UTRs are rich in uridine and adenosine but are relatively poor in guanosine and cytidine. To determine whether tracts of uridine are overrepresented, we determined the theoretical distribution of octamer units weighted by the relative base content in 3Ј-UTRs using a binomial distribution. If a given base has a propensity to segregate into tracts of high and low base content, the fre-quency of octamers containing six to eight occurrences of the base will be greater than expected from the random distribution. Similarly, the frequency of octamers containing 0 -3 base occurrences will also be greater than the expected amount. Therefore, a plot of octamer frequency versus the number of base occurrences will appear more broad and shallow than the theoretical random distribution (Fig. 5A). In C. elegans 3Ј-UTRs, the observed distribution of uridine and to a lesser extent adenosine reveals a propensity to segregate into high and low base content tracts, whereas the observed distributions of guanosine and cytidine very closely match the random distribution (Fig. 5A). This is consistent with a selective pressure that favors runs of uridine and adenosine. When the distribution of observed octamer sequences is plotted as a function of relative frequency, we find that the most frequently observed octamer is U8, and 29 of the top 30 octamers contain at least 6 uridines (Fig. 5B). This highlights the preponderance of uridine tracts in the 3Ј-UTRs of C. elegans genes, and demonstrates that MEX-5 does not bind with sufficient specificity to select mRNAs for regulation on its own.

MEX-5 and PIE-1 Localization Do Not Depend on MEX-3-If
MEX-5 regulates specific maternal transcripts, it must do so as part of a complex with a more specific RNA-binding protein. If so, a likely candidate is MEX-3. MEX-3 is a KH domain RNAbinding protein that is also required for anterior development (38). Mutation of mex-3 results in the same terminal phenotype as mex-5, and the protein displays a similar distribution throughout the 1-4 cell stages of development. Moreover, MEX-6, which is 70% identical to MEX-5, was identified in a yeast two-hybrid screen for MEX-3 interacting proteins (39). Finally, the localization pattern of MEX-3 protein is perturbed in mex-5 mutants (14).
Because the localization pattern of MEX-3 depends on MEX-5 (14), it is possible that the two proteins localize in a complex. If so, then anterior MEX-5 accumulation might depend upon the presence of MEX-3. In addition, because MEX-5 activation of zif-1 drives posterior accumulation of PIE-1 (16), localization of this protein should be altered in the absence of MEX-3 if it serves as a zif-1 co-activator. To test these hypotheses, we knocked down MEX-3 levels using RNAi in worms expressing GFP-MEX-5 or GFP-PIE-1 by feeding them E. coli expressing double-stranded RNA targeted against mex-3 transcripts. The cellular distribution of GFP-MEX-5 and  GFP-PIE-1 in embryos was determined by wide field fluorescence microscopy. No difference is observed in the expression pattern of GFP-MEX-5 or GFP-PIE-1 in mex-3 RNAi embryos (n ϭ 135, n ϭ 127, respectively; only embryos in the 1-8 cell stage were counted) compared with control embryos (n ϭ 101, n ϭ 111, Fig. 6). More than 90% of the embryos laid onto mex-3 RNAi plates failed to hatch (n Ͼ 500), indicating that RNAi was disrupting mex-3 function. We conclude that MEX-3 is not required for MEX-5 accumulation in the anterior or for activation of zif-1. The results argue against a functional role for a MEX-3⅐MEX-5 complex. However, we cannot rule out the possibility that they co-regulate other maternal transcripts.
A Single Residue in Each Finger Defines MEX-5 RNA-binding Specificity-The NMR structure of the mammalian TTP homolog ERF-2 reveals that each zinc finger recognizes adjacent UAUU repeats (27). To understand the molecular basis for differential MEX-5 RNA-binding specificity, we prepared a homology model of MEX-5 based on the ERF-2 structure using SWISS-MODEL (40) (Fig. 7a). The model reveals several amino acid differences in the RNA-binding interface that may contribute to the difference in specificity. Most notably, three adjacent residues (Glu, Leu, and Cys) in  each finger of ERF-2 combine to form an adenosine recognition pocket. In MEX-5, the glutamate residue is replaced with arginine (Arg-274) in the first finger and lysine (Lys-318) in the second (Fig. 1c). In the NMR structure, the glutamate side chain forms a hydrogen bond with the exocyclic amine of the adenosine base. In the homology model, the basic residues rotate away from the adenosine and form backbone contacts with adjacent nucleotides (Fig. 7b). The model predicts that loss of base-specific hydrogen bonds and formation of additional backbone contacts contributes to the relaxed specificity of MEX-5. If so, mutation of the basic residues to glutamate might confer TTP-like specificity to this protein.
To test this hypothesis, we prepared mutant variants of MEX-5 where either or both of these basic residues were replaced with the glutamate residue present in TTP (R274E, K318E, and R274E/K318E). A substitution at an unrelated position in the RNA interface (M288Y) was prepared as a control. First, we examined the ability of the mutants to interact with the TNF-ARE sequence. The R274E and K318E mutations do not affect the interaction, whereas the double mutation binds with 4-fold reduced affinity to this sequence ( Table 2). A small loss in affinity is expected due to increased electrostatic repulsion from the glutamate residues. The M288Y mutation has no effect on TNF-ARE affinity.
In contrast, the effect of glutamate mutations on binding to the 30-nucleotide polyuridine RNA is significantly more drastic. The individual glutamate mutations reduce binding by 2-4fold, whereas the double mutation significantly reduces binding Ͼ15-fold by EMSA and 7.5-fold by FP (Fig. 7c, Table 2). As before, the M288Y mutation has no effect. Because all variants retain the ability to bind to TNF-ARE RNA, and because all are soluble and monomeric as determined by analytical size exclusion chromatography (supplementary materials Fig. 1), we conclude that reduction in polyuridine binding results from a change in binding specificity rather than from protein misfolding effects. The data demonstrate that a single glutamate in each finger confers the ability to discriminate between UAUU repeat RNA and polyuridine, thus defining the molecular basis for the difference in RNA discrimination between MEX-5 and TTP/ERF-2.

DISCUSSION
Accurate post-transcriptional regulation of gene expression requires the ability to distinguish specific transcripts from the total pool of cellular RNA. Regulatory factors target mRNA through sequence-specific or structure-specific interactions (41). In the case of micro-RNAs, the mechanistic basis for mRNA recognition is easy to conceptualize; complimentary base pairing of a 7-or 8-nucleotide seed drives mRNA discrimination (42). In contrast, the sequence code recognized by eukaryotic RNA-binding proteins usually includes a greater degree of degeneracy.
In C. elegans, early embryogenesis requires a class of divergent CCCH-type tandem zinc finger proteins. They segregate to opposite poles of the 1-cell embryo in response to PAR protein activity (14 -16, 43). Because they are homologous to mammalian TTP, most are thought to regulate maternal gene expression at the RNA level. Our data show that two of the C. elegans TZF proteins, MEX-5 and MEX-6, can in fact bind to RNA with high affinity but with relaxed specificity compared with TTP. The results are consistent with the hypothesis that they directly regulate maternal transcripts, but also suggest that accessory factors are required for mRNA targeting.
Implications for the Role of MEX-5 in Development-Uridine is by far the most common nucleotide present in the 3Ј-UTRs of C. elegans genes. We show that tracts of six or more uridines are present in the 3Ј-UTR of 91% of annotated transcripts, which is more frequent than expected from a random distribution. Therefore, contrasting with previous hypotheses, MEX-5 does not bind to RNA with sufficient specificity to drive regulation of a subset of maternal transcripts.
The relaxed RNA-binding specificity of MEX-5 has several implications for its function in development. It is possible that it is a general mRNA repressor, silencing all transcripts in the oocyte and the anterior lineage until it is destroyed after the 8-cell stage. This is unlikely because translation of glp-1 mRNA occurs in the anterior and appears to require MEX-5 (14,37). Alternatively, MEX-5 may be an affinity adapter for a more specific RNA-binding protein, regulating expression from just a few maternal transcripts as part of a cooperative complex. If so, then MEX-5 could behave like Drosophila melanogaster Nanos protein, which works in complex with Pumilio to regulate maternal mRNA expression (44 -46). Finally, MEX-5 RNAbinding activity may not contribute to maternal RNA regulation. Instead, RNA-binding activity may be required to target MEX-5 to the posterior centrosome and to P-granules (Fig. 1A), two subcellular organelles that contain RNA (14,15,47,48). If so, then MEX-5 RNA-binding activity may play a role in germline maintenance in addition to its role in germline formation.
Implications for RNA Recognition by Other CCCH-type TZF Proteins-The difference in specificity between MEX-5 and TTP is governed by a single amino acid substitution within a highly conserved region of each finger. Our data indicate that the amino acid identity of this discriminating position defines a code that predicts the specificity of other TZF proteins. We predict that glutamate residues encode selectivity for UUAUUUAUU RNA, whereas basic residues lead to promiscuous binding to uridine-rich RNA sequences including, but not limited to, UAUU-repeat sequences.
To determine the variability of the discriminator residue in TZF proteins, we performed a BLAST search of the C. elegans genome using the TZF domain of MEX-5 as a query (Fig. 8). Sixteen hits result from this analysis. Variability is observed at the discriminator position, but the other two residues that comprise the adenosine recognition pocket are conserved. Of the 16, only MEX-5 and MEX-6 contain basic amino acids in both fingers at the discriminator position. One protein, F38B7.1, has a glutamate at both positions similar to TTP and ERF-2. We predict that this protein binds specifically to UAUU repeat RNAs. Interestingly, five proteins (POS-1, Y116A8C.19, Y116A8C.20, Y57G11C.25, and F38C2.5) have small hydrophobic residues in both discriminator positions, an alanine in finger one and a valine in finger two. Experiments with recombinant POS-1 protein indicate that it binds to RNA with specificity that is different from both MEX-5 and TTP. 3 Therefore, small hydrophobic residues at the discriminator position may define a third specificity class for TZF proteins. The remaining TZF proteins have a combination of basic, small hydrophobic, or acidic/polar amino acids at the discriminator position. Three proteins required for oocyte maturation (OMA-1, OMA-2, and MOE-3) have a valine in the first finger and a lysine in the second. We predict these proteins will display hybrid specificity between POS-1 and MEX-5. In contrast, PIE-1 contains an arginine in finger 1 and a glutamine in finger 2. Like glutamate, glutamine can accept a hydrogen bond from adenosine, but it can also donate a hydrogen bond to the O-6 carbonyl of a guanosine base as well. F38C2.7, C35D6.4, and MEX-1 have the combination of a small hydrophobic residue in one finger and a polar amino acid (serine or asparagine) that could theoretically accept and/or donate a base-specific hydrogen bond in the RNA complex. The variations of the discriminator residue imply differences in RNA-binding specificity by the divergent C. elegans TZF proteins. Further experiments will determine the possible RNA interactions allowed by each discriminator amino acid within this class of proteins.
Remarkably, the variations in the TZF protein discriminator residue found in C. elegans are absent in higher eukaryotes. Blast analysis reveals that every vertebrate CCCH-TZF protein in GenBank TM has a glutamate residue at both discriminator positions. In contrast, homologs with variable discriminators are present in Caenorhabditis briggsae and Caenorhabditis remanei, and in more divergent Nemata including parasitic species. Because the identity of the discriminator correlates with its RNA-binding specificity and thus its molecular function, the critical role of divergent TZF proteins during early development must be a special adaptation of this phylum.
The role of the PAR proteins in establishing cell polarity is conserved from worms to flies to mammals (49). Asymmetric expression of signaling proteins and transcription factors from maternal mRNAs is also highly conserved (2). Yet the cassette of RNA-binding proteins that connect these two layers is clearly highly divergent (50). Defining the basis of RNA-binding protein function in early development will provide a framework by which mechanistic differences in the regulation of maternal mRNAs contribute to variability of metazoan body plan.