|Home | About | Journals | Submit | Contact Us | Français|
The diversity of protein isoforms arising from alternative splicing is thought to modulate fine-tuning of synaptic plasticity. Fragile X mental retardation protein (FMRP), a neuronal RNA binding protein, exists in isoforms as a result of alternative splicing, but the contribution of these isoforms to neural plasticity are not well understood. We show that two isoforms of D. melanogaster FMRP (dFMR1) have differential roles in mediating neural development and behavior functions conferred by the dfmr1 gene. These isoforms differ in the presence of a protein interaction module that is related to prion domains and is functionally conserved between FMRPs. Expression of both isoforms is necessary for optimal performance in tests of short and long-term memory of courtship training. The presence or absence of the protein interaction domain may govern the types of ribonucleoprotein (RNP) complexes dFMR1 assembles into, with different RNPs regulating gene expression in a manner necessary for establishing distinct phases of memory formation.
Protein isoforms coded by alternatively spliced pre-mRNAs expand the functional repertoire of genes. The importance of this mechanism for developing protein diversity is illustrated by the finding that most multi-exon human genes are subject to alternative splicing (Pan et al., 2008). Alternative splicing is common in neural tissue, where novel protein isoforms may function as on/off switches or graded modulators that fine-tune synaptic function (Lipscombe, 2005). The significance of alternative splicing to neurons is underscored by the variety of neurological diseases that result from its dysregulation (Licatalosi and Darnell, 2006).
Mutations in the fragile X mental retardation gene (FMR1) result in fragile X syndrome, a leading form of heritable mental retardation. The fragile X protein (FMRP) binds select RNA species, and the neural phenotypes of fragile X syndrome result in part from disruptions in trafficking of mRNA substrates through neuronal projections and translation regulation at the synapse (Bassell and Warren, 2008). Human FMRP may exist in at least 20 isoforms that arise from alternative splicing (Ashley et al., 1993; Eichler et al., 1993; Verkerk et al., 1993), but the contributions of FMRP isoforms to neural plasticity are largely unknown.
A Drosophila model to study FMRP function has been developed, with loss-of-function for dfmr1 leading to neural phenotypes that have significant parallels with those of fragile X patients and mice (Gatto and Broadie, 2009). Like its mammalian counterparts, dfmr1 pre-mRNA is subject to alternative splicing (FlyBase). In this study, we utilize Drosophila genetics to delve into the role(s) of dFMR1 isoforms in neural development and behavioral plasticity, and report on the role of a glutamine/asparagine (Q/N)-rich protein interaction domain present in the C-terminus of some dFMR1 isoforms through alternative splicing. The related C-terminal domain of human FMRP facilitates an interaction with kinesin and is needed for efficient FMRP-mediated dendritic RNA transport (Dictenberg et al., 2008). Deletion of the dFMR1 Q/N domain results in defects for a subset of neural development phenotypes associated with a null allele of dfmr1, demonstrating that it is not essential for all dFMR1-regulated processes. The ΔQ/N mutants display an intermediate level of performance in tests of short and long-term memory when compared to wild type and null alleles of dfmr1. A decline in memory performance is observed with the ΔQ/N mutants during the transition from immediate recall to one-hour short-term memory of training. This decline parallels the protein synthesis requirement for maintaining metabotropic glutamate receptor-dependent long-term depression. The dFMR1 ΔQ/N protein may be defective in associating with RNP complexes that promote RNA transport and/or regulate synaptic protein synthesis necessary for establishing memory. We find that expressing the dFMR1 Q/N- isoform is also needed for optimal performance in short and long-term memory of training. These observations may be explained by different isoforms of dFMR1 assembling into distinct RNP complexes that modulate patterns of protein synthesis required for establishing specific phases of memory. Our results indicate that unique biochemical properties associated with individual FMRP isoforms and the mechanisms that regulate their synthesis are necessary for normal cognition.
Except where indicated, all fly stocks are derived from a w1118 background. The dfmr13 allele is a deletion that removes the entire open reading frame (Dockendorff et al., 2002; Banerjee et al., 2007). A 14 kb genomic rescue fragment spanning the dfmr1 locus is described in Dockendorff et al. (2002). The terminal exon encoding the glutamine/asparagine domain was mutagenized as follows: PCR primers were devised to amplify a 2.0 kb fragment that resulted in a product that had NcoI and BspEI ends; these restriction sites are unique to the genomic rescue fragment. The amplification product was digested with these enzymes and ligated to the backbone of the rescue fragment that had been treated with the same enzymes. The substitution with the PCR product resulted in a construct where 90 codons of dfmr1 encompassing much of the Q/N domain were removed. The PCR amplification product was produced with a proofreading polymerase (Vent polymerase, New England Biolabs) and was sequenced to ensure the absence of secondary mutations arising from base misincorporations. Substituting an NcoI/BspEI fragment from dfmr1 genomic DNA with the corresponding fragment from dfmr1 cDNA created the dfmr1L allele, which does not produce the dFMR1 PB (Q/N-) isoform. The mutant transgenes were introduced to w1118 flies using standard transformation techniques. Resulting transgenic flies were then crossed into the dfmr13 background to produce fly stocks where the mutant rescue fragment was the sole source of dfmr1. Analysis of dFMR1 protein expression from the transgenes was measured by Western blotting, followed by quantification using ImageQuant (Molecular Dynamics) or ImageJ software from the National Institutes of Health. Stocks expressing the dfmr1L transgenes were crossed to each other to provide the reported doses of total dFMR1 protein. All dfmr1 stocks were balanced with the TM6C Tb Sb chromosome.
Larval neuromuscular junction (NMJ) type I boutons were detected by staining third-instar larval fillets with anti-horseradish peroxidase (Cappel, Aurora, OH) at a dilution of 1:200. Mushroom bodies were visualized by staining whole-mounts of brains with anti-FasII at a 1:10 dilution (mAb 1D4 obtained from University of Iowa Developmental Studies Hybridoma Bank). Rabbit polyclonal antiserum against GFP was obtained from Invitrogen and used at a 1:200 dilution. Secondary antibodies conjugated to either HRP or fluorochrome were obtained from Jackson ImmunoResearch (West Grove, PA) and used at a 1:200 dilution. Confocal images were collected on an Olympus FV500 microscope. Western blots were performed as described in Wan et al. (2000) using anti-dFMR1 antibody 5A11 at a 1:1000 dilution (ascites fluid) and anti-β-tubulin mAb E7 (both from the University of Iowa Developmental Studies Hybridoma Bank) at a 1:10,000 dilution (ascites fluid).
Strain W303 (matα, leu2-3,112 trp1-1 ura3-1 ade2-1 his3-11,15) was provided by Rodney Rothstein (Columbia University) and used for the yeast assays. Transformations were done using lithium acetate as described by Gietz et al. (1992). The 122 amino acids of the dFMR1 Q/N-rich domain were fused to a constitutively active form of the glucocorticoid receptor transcription factor (GR526; Schena and Yamamoto, 1988) expressed via a glyceraldehyde-3-phosphate dehydrogenase (GPD) promoter. Fusion of GR526 to yeast prions has been used to monitor prion stability and inheritance (Li and Lindquist, 2000). The dFMR1 Q/N::GR526 plasmid was co-expressed with a plasmid harboring a LacZ gene with GR binding sites upstream. Yeast cells carrying these plasmids were plated to X-gal media, and blue colonies selected. These colonies were grown overnight in selective media, diluted, and plated again to X-gal media to monitor colony color. GR526 by itself was used as a control for native expression of LacZ. Plates were incubated for 3 days at 30°C and then photographed.
GFP aggregation in Drosophila tissues was analyzed by driving expression of a dFMR1 Q/N::GFP fusion in third instar larval muscle tissues with the myosin heavy chain GAL4 driver. The larval fillets were stained with rabbit anti-GFP primary antibody (1:200) and FITC labeled anti-rabbit secondary antibody (1:200) to visualize GFP aggregates by confocal microscopy. Expression of native GFP was used as a control for any possible background aggregation.
For courtship behavior testing, males of the appropriate genotypes were collected within four hours of eclosion and kept in isolation prior to testing. Female targets were of the genotype XX, y, f (attached X) and collected as virgins for courtship testing. All flies were kept in 12:12 light/dark cycles at 25°C and 50-75% relative humidity, and were aged four to nine days prior to analysis. For the naive courtship analysis, the male and female were transferred via aspiration to a mating chamber 20 mm in diameter and 5 mm deep. Males were given a five-minute recovery period prior to addition of the female target. Males were monitored for courtship activity that included orienting, following of the female, wing extension and vibration, tapping of female with his foreleg and attempted copulation for a period of 10 minutes, or until copulation occurred. The percentage of time the male spent courting the female was recorded, and this percentage is the courtship index. Measurement of immediate recall was made by pairing a naive male with a non-receptive female for a single one-hour training session, and then placing him in a second chamber with a receptive female within 2 minutes of training. One-hour short-term memory was assessed by taking a male that had been trained with a non-receptive female for one hour and placing him in isolation for one hour prior to pairing with a receptive female. At least 17 animals were tested for each genotype during analyses of naive courtship, immediate recall, and short-term memory. The training paradigm for assessment of long-term memory is derived from McBride et al. (1999). Males were paired with a non-receptive female for seven continuous hours, and then kept in isolation for four days prior to testing. Sham-trained males were treated identically, except for the exposure to the training female. The observers were blind to the genotypes of the animals for all courtship studies.
Circadian behavior was tested as described in (Dockendorff et al. 2002). Flies were entrained to a 12:12 light/dark cycle, placed into activity monitors (Trikinetics, Waltham, MA) maintained in light/dark cycles, and then placed under constant darkness. Locomotion activity was collected in 30-minute bins. The percentage of flies judged to be rhythmic was assessed by Clocklab software (Actimetrics, Evanston, IL) as follows: Using a confidence level of 0.025, batch analyses were performed for the genotypes tested, monitoring the locomotion activity in constant darkness over seven days. The difference between the power(1) and significance(1) values was calculated for each fly, and a value of less than 10 was the basis for judging an arrhythmic phenotype. Visual analyses of periodograms and actograms were also conducted to confirm the results. Strength of rhythmicity was calculated by the difference between the power(1) and significance(1) values for individual flies of each genotype tested.
Courtship indices were arcsin transformed, then analyzed by one-way ANOVA, followed by a Tukey-Kramer post-test. NMJ bouton counts and strength of circadian rhythmicity were analyzed by one-way ANOVA with a Tukey-Kramer post-test, or by a Kruskal-Wallis test, followed by a Dunn post-test. The analyses of courtship indices, NMJ bouton counts, and strength of circadian rhythmicity were conducted using InStat or Prism software from GraphPad (San Diego, CA). Comparisons of mushroom body axon midline crossings and percentage of animals judged rhythmic were made by Fisher's exact test or a chi-square test for homogeneity.
Early studies of human FMR1 gene and transcript structure revealed the presence of multiple transcript isoforms resulting from exon skipping or the use of alternative splice sites during FMR1 pre-mRNA processing (Ashley et al., 1993; Eichler et al., 1993; Verkerk et al., 1993). The synthesis of at least 20 human FMR1 mRNAs is possible, but several of these transcripts comprise only a minor fraction of total FMR1 mRNA (Ashley et al., 1993; Verkerk et al., 1993; Xie et al., 2009). Most alternative splicing of FMR1 pre-mRNA occurs in regions of the transcript that code for the C-terminal domain of the protein (Figure 1A). FMRP isoforms can lack or have alterations in functional domains such as the second KH domain, a nuclear export sequence, and a major phosphorylation site (Sittler et al., 1996; Didiot et al., 2008). In addition to removing a nuclear export sequence in FMRP, skipping of exon 14 in human FMR1 pre-mRNA leads to a +1 frameshift that results in proteins with novel C-terminal peptides (Ashley et al., 1993; Sittler et al., 1996). Altered biochemical properties are ascribed to some FMRP isoforms (Sittler et al., 1996; Xie et al., 2009), and the switch between phosphorylation states is seemingly critical for synaptic plasticity (Pfeiffer and Huber, 2007; Narayanan et al., 2007). However, the contribution of most FMRP isoforms to neural function remains largely uncharacterized.
Transcript isoforms of the D. melanogaster fragile X gene (dfmr1) are present, based upon analyses of cDNAs (Figure 1A; FlyBase). Like human FMR1, dfmr1 pre-mRNA is subject to alternative splicing that results in dFMR1 isoforms with alterations in the second KH domain and composition of the C-terminal peptide. The C-terminal truncation associated with the PB isoform (FlyBase) of dFMR1 is derived from an alternative splicing event where much of a 541-nucleotide intron is retained (Figure 1B). The retained intron serves as a source for codons that specify a C-terminal peptide with the sequence DTN, a termination codon, and canonical hexanucleotide sequences (AAUAAA) that facilitate 3′ end formation of the transcript (Figure 1C). The PB isoform (referred to as Q/N- through the text) is 529 amino acids, and a protein corresponding to its predicted mass can be detected on Western blots of tissue extracts (Figure 1D). As is observed with several hFMRP isoforms, the Q/N- isoform constitutes a minor fraction of total dFMR1 protein.
While the presence of RNA-binding and protein interaction domains in the N-terminal amino acids of FMRP and the FXR paralogs have been noted for some time, a role for C-terminal amino acids had been less clear. Recent studies of the hFMRP C-terminal peptide strongly indicate that it has protein interaction capacity, as judged by yeast two-hybrid screening using the C-terminal peptide as a bait (Menon et al., 2004), and the finding that a C-terminal truncation of hFMRP inhibits its interaction with kinesin (Dictenberg et al., 2008). A comparison of the C-terminal 81 amino acids encoded by exons 16-17 of FMR1 and the terminal 81 amino acids of dFMR1 shows significant conservation, with 31 of 81 residues (38%) being identical or similar (Figure 1E). Two 3′ splice acceptor sites are present in exon 17 of human FMR1 pre-mRNA and dictate the presence or absence of 17 codons within the exon (Eichler et al., 1993; Figure 1E). In addition to FMR1, mammalian genomes encode the paralogous FXR1 and FXR2 genes (Siomi et al., 1995; Zhang et al., 1995). The C-terminal peptide of dFMR1 was compared to corresponding sequences from FXR1 and FXR2, and the ClustalW alignment reveals 33 and 43% amino acid identity/similarity, respectively (Figure 1E).
Examination of C-terminal amino acids that are coded by the terminal 3′ exon of dfmr1 reveal that this 122 amino acid peptide is unusually enriched in glutamine and asparagine (Q/N) residues (36% Q/N, Figure 1E). Protein domains enriched in Q/N residues are common in eukaryotic proteomes (Michelitsch and Weissman, 2000; Harrison and Gerstein, 2003) and where studied, act as modules that facilitate interactions with both homologous and heterologous proteins (Decker et al., 2007, Guo et al., 2007; Shewmaker et al., 2007; Kim et al., 2008). To test whether the dFMR1 Q/N domain has protein interaction properties, genes encoding GFP or a dFMR1 Q/N::GFP fusion were cloned behind a UAS sequence, and their expression was driven by myosin heavy chain-GAL4. Examination of larval muscles expressing either of the GFP isoforms shows that only the dFMR1 Q/N::GFP fusion exhibits aggregation of GFP antigen in these tissues (Figure 2A), showing that the dFMR1 Q/N domain can facilitate protein interactions in vivo. A protein interaction function is likely conserved between the C-terminal peptides of hFMRP and dFMR1.
The series of short glutamine repeats in the dFMR1 C-terminal peptide sequence resembles those present in prion-forming domains from fungi. Aggregation of S. cerevisiae Sup35 proteins to propagate a prion state is dependent upon protein interactions between their Q/N-rich domains (reviewed by Shorter and Lindquist, 2005; Wickner et al., 2007). Adoption of a prion-like state has been proposed as a means to maintain regulators of long-term memory in a self-sustaining active form (Si et al., 2003, 2010). Two independent algorithms devised to identify Q/N-rich domains with prion-forming potential both uncover dFMR1 (Michelistch and Weissman, 2000; Harrison and Gerstein, 2003). To test whether the Q/N domain of dFMR1 has such properties, we fused it to a reporter protein used to detect and monitor prion-like activity. The Sup35 NM and Aplysia CPEB Q/N domains confer a metastable state of activity when fused to an otherwise constitutively active glucocorticoid receptor (GR) transcription factor (GR526, Schena and Yamamoto, 1988; Li and Lindquist, 2000; Si et al., 2003). In this assay, yeast colonies plated on X-gal media only rarely switch between functional (blue colony) and non-functional (white colony) states of the GR526 transcription factor. When the 122 codons of the exon encompassing the dFMR1 Q/N domain are fused to GR526, we observe a high level of switching between active and inactive states of the GR526 transcription factor, as judged by scoring blue and white colonies that appear on X-gal plates (Figure 2B). This degree of instability contrasts with the high level of stability observed with the prion-forming Sup35 NM domain and the Q/N-rich domain of Aplysia CPEB protein (Li and Lindquist, 2000; Si et al., 2003), and indicates that the dFMR1 Q/N domain may not have true prion-forming properties. Studies have shown that less than 20% of proteins that are candidates for prion formation in yeast actually adopt such a state (Michelitsch and Weissman, 2000; Sondheimer and Lindquist, 2000; Alberti et al., 2009). Nonetheless, a protein interaction capacity for the dFMR1 Q/N domain may facilitate interactions with ribonucleoprotein (RNP) complexes that regulate neural function.
A 14 kb genomic rescue fragment encompassing the dfmr1 locus rescues phenotypes associated with the dfmr13 deletion null mutation (Dockendorff et al., 2002). We used this rescue fragment to create a 270 bp deletion derivative that removed most of the codons specifying Q or N from the 3′ exon (Figures 3A, B; see Materials and Methods). The mutant rescue fragment was introduced by P element transformation and two independent transgene insertions were subsequently analyzed. The mutant rescue fragments were crossed into a dfmr13 null background to generate stocks where the mutant transgene is the sole source of dFMR1 protein. The truncated proteins are stable in a steady state as judged by Western blotting (Figure 3C). Quantification of signals from Western blots shows that a single copy of the transgene results in a level of dFMR1 protein that is either equivalent to that observed in a w1118 control animal (ΔQ/N 8), or at about 1.8-fold overexpression (ΔQ/N 1B; Figure 3C). Whole-mount staining of adult brains with anti-dFMR1 antibody shows that mutant animals have a spatial expression pattern of dFMR1 in the central nervous system that is indiscernible from the pattern observed in brains of wild type animals (Figures 3D, E).
Deletion null alleles of dfmr1 elicit pleiotropic phenotypes that include defects in embryogenesis, neuronal development, and behaviors of larvae and adults (Zhang et al., 2001; Dockendorff et al., 2002; Morales et al., 2002; Lee et al., 2003; Michel et al., 2004; Deshpande et al., 2006; Monzo et al., 2006). We then set out to discern the phenotypic consequences of deleting the dFMR1 Q/N domain. Several possibilities exist for the effects of the Q/N deletion on dFMR1 function. As a likely protein interaction platform, the Q/N domain could be essential for all dFMR1 functions, and animals with the ΔQ/N allele would then exhibit phenotypes indistinguishable from those with a null allele. Conversely, the association of dFMR1 with other protein complexes might involve both the Q/N domain and a second protein interaction platform common to FMRPs (Figure 1A). The function of the other protein interaction domain might be enough to compensate for the loss of the Q/N domain, and thus result in no change from the wild type phenotypes. Finally, it is possible that the dFMR1 Q/N domain facilitates interactions with protein complexes that mediate select dFMR1 processes. If this were the case, variation in the relative strength of phenotypes resulting from the ΔQ/N allele would be predicted.
The development phenotypes of embryos produced by dfmr1 mutant females include defects in cellularization and cleavage furrow formation that are ascribed to misregulation of cytoskeletal dynamics. The breakdown of these embryonic development processes results in hatch rates of embryos from dfmr1 mutant females being about half that of embryos from wild type mothers (Monzo et al., 2006). To assess the effects of the Q/N deletion on embryonic development, we mated wild type males with wild type females, those homozygous for the dfmr13 null allele, and females expressing the transgene alleles of dfmr1 where the region coding for the Q/N domain is deleted, and measured the hatch rate of eggs. The results presented in Figure 4A show that both transgenic lines expressing the mutant dFMR1 protein provide a significant level of rescue of embryonic hatch rates over that observed with embryos from females homozygous for the dfmr13 deletion null allele.
In wild type flies, the β-lobe axons of mushroom bodies infrequently cross the midline of the central brain. However, flies homozygous for null alleles of dfmr1 display a midline-crossing defect in these axons with a high level of penetrance (Michel et al., 2004; Banerjee et al., 2007). We examined the brains of 2-day old adults harboring wild type, null, or Q/N domain deletion alleles of dfmr1 for this phenotype. Both transgenes expressing the dFMR1 ΔQ/N domain provide significant rescue of the phenotype observed with the null allele control (Figure 4B). These results, along with those observed with the embryonic hatch rate, show that functions and properties conferred by the Q/N domain are not essential for the ability of dFMR1 to execute at least a subset of its normal activities.
Larvae homozygous for a null allele of dfmr1 have elevated numbers of neuromuscular junction boutons, which are the swellings associated with synapse formation. The increase in boutons is due in part to excessive branching from the axons that innervate the larval muscles (Zhang et al., 2001). Using larvae expressing a wild type rescue fragment in an otherwise dfmr1 null background (WT rescue), larvae homozygous for a null allele of dfmr1, and larvae expressing dFMR1 with a deletion of the Q/N domain, we examined muscle 12 from larval segment A4 for numbers of type 1 boutons. The results from these experiments are presented in Figure 4C, and show that the loss of the Q/N domain results in a significant increase in bouton numbers when compared to larvae that are expressing a wild type allele of dfmr1. Larvae expressing both wild type and ΔQ/N alleles of dfmr1 had numbers of boutons like those of wild type controls, showing the ΔQ/N alleles are recessive. Similar phenotypes were observed with muscles 4 and 6/7 (Supplementary data Figure 1). That the ΔQ/N allele elicits bouton numbers that resemble those observed with the null allele (Figure 4C) indicates that the Q/N domain is a significant factor for dFMR1 function in this process of synaptic development.
A null mutation of dfmr1 results in loss of rhythmic circadian locomotion activity that likely occurs via misregulation of one or more clock-controlled output genes (Dockendorff et al., 2002). We thus tested the Q/N deletion transgenic stocks for their ability to maintain rhythmic locomotion activity in constant darkness. The results in Figure 4D show that about 60% of the animals from the ΔQ/N transgenic lines retain rhythmic locomotion in constant darkness, compared to about 20% of null mutants, and over 90% of dfmr13 homozygotes that express a wild type allele of dfmr1 (WT rescue). Strength of rhythmicity for the ΔQ/N flies, as measured by the difference in Power(1) and Significance(1) values, is intermediate between those with wild type or null alleles (Figure 4E). As was seen with the NMJ bouton numbers, the ΔQ/N mutation behaves as a recessive allele (Supplementary Figure 2). Two possibilities could explain the partial penetrance and expressivity of the circadian locomotion phenotypes displayed by animals with the ΔQ/N allele. First, both Q/N+ and Q/N- isoforms could contribute to the ability of the dfmr1 locus to control outputs from the circadian clock. Alternatively, the Q/N domain enhances, but is not essential for protein interactions that facilitate regulation of clock outputs by dFMR1.
Courtship in Drosophila involves a series of stereotyped behaviors between the courting male and the female mate (reviewed by Greenspan and Ferveur, 2000). Although conduct of these behaviors requires the function of several sensory modalities, their ethological relevance has made analysis of Drosophila courtship a common method to study genes that control behaviors and processes of learning and memory (reviewed by Mehren et al., 2004). When placed with a receptive virgin female, naive males will initiate courtship behaviors that include orienting to the female, following and tapping the female with his foreleg, extension and vibration of a wing, licking the female with his proboscis, and attempts at copulation. The percentage of time engaged in these behaviors over a defined period is referred to as the courtship index (CI).
Males homozygous for a null mutation of dfmr1 have a reduction in naive courtship activity compared to wild type flies (Dockendorff et al., 2002), indicating that dFMR1 protein is needed for establishment and/or function of neural pathways needed for efficient courtship behavior. However, dfmr1 mutant males can achieve copulation, indicating that their reduced courtship index is not simply the consequence of a total deficit in sensory detection or processing of courtship cues. We tested the dFMR1 ΔQ/N transgenic stocks for naive courtship behavior. Removal of the Q/N domain results in naive courtship levels that resemble those of the null allele control (Figure 5A). The Q/N domain of dFMR1 is thus necessary for development and/or function of neural circuitry that facilitates naive courtship behavior.
Conditioned courtship is a paradigm for memory where unreceptive Drosophila females present visual, behavioral, and pheromone-based cues towards a courting male that results in a reduction of his courtship activity. Moreover, the inhibition of the male's courtship activity carries over to receptive females, and thus represents a form of associative memory. Depending upon the training regimen with a non-receptive female, this memory can be maintained for periods of a few hours (short-term memory) to long-term memory that lasts several days (Siegel and Hall, 1979; McBride et al., 1999). Prior studies have established that like wild type males, the CI of dfmr1 null mutants declines during training, demonstrating that the mutant is able to detect and respond to the negative courtship cues displayed by the female (McBride et al., 2005). Flies homozygous for a null allele of dfmr1 are deficient in both immediate recall (0-2 minutes post-training) and one-hour short-term memory (STM) of conditioned courtship training (McBride et al., 2005).
The necessity for the dFMR1 Q/N domain on immediate recall and STM was tested with conditioned courtship. Males with wild type, null, or ΔQ/N alleles of dfmr1 were paired with unreceptive females for a single one-hour training session. Trained males were removed from the courtship chamber and placed with a receptive female (immediate recall), or left in isolation for one hour, and then paired with a receptive female to assess STM of the training. A comparison of the naive CI and the post-training CI shows that flies expressing the ΔQ/N protein have a significant decrease in courtship at the immediate recall stage, indicating that they have retained memory of the training (Figure 5B). In contrast, there is not a significant difference between the naive CI of ΔQ/N flies and the CI measured one hour after training (Figure 5C), while courtship levels of wild type controls remain suppressed.
Relative memory performance between genotypes was assessed through use of a memory index (reviewed by Kamyshev et al., 1999). We performed these analyses for the immediate recall and 1-hour STM, comparing wild type, ΔQ/N, and dfmr1 null flies. While flies expressing the ΔQ/N protein are judged to have immediate recall of training, the memory performance of these animals is intermediate to the wild type and null alleles (Supplemental data Figure 3). While the memory index in wild type flies does not decline from immediate recall to one-hour post-training, the memory index for ΔQ/N mutants and nulls declines during this interval. At one hour post-training the ΔQ/N mutants still retain a memory index intermediate to those of wild type and dfmr1 null flies (Supplemental data Figure 3). Taken together, these studies show that the Q/N domain is important for establishment of short-term memory.
To assess needs for the Q/N- isoform of dFMR1, we generated a derivative of the dfmr1 genomic rescue fragment where DNA encompassing the alternatively spliced intron was substituted with dfmr1 cDNA, resulting in an allele (dfmr1L) that is unable to produce the Q/N- isoform (Figure 6A, B). We analyzed flies expressing this construct as the sole source of dFMR1 protein for the same neural development and behavior phenotypes used to study the Q/N domain deletions. The developmental phenotypes of NMJ bouton overgrowth and midline crossing by mushroom body axons were absent in such animals, and no defects in rhythmic circadian locomotion were observed (Supplemental data Figure 4). We then tested flies for courtship behavior and memory of conditioned courtship training. Flies unable to produce the dFMR1 Q/N- isoform have a naive courtship level that is similar to that of wild-type controls and exhibit one hour STM of courtship training (Figure 6C). We then tested these flies for long-term memory (LTM) of courtship training using a 7-hour uninterrupted training regimen described in McBride et al. (1999). Both dfmr1L alleles and allele combinations tested reveal a deficit in 4-day long-term memory of courtship training (Figure 6D), showing that the function of the dFMR1 Q/N- isoform and the alternative splicing pathway that results in its production have a critical role in establishment of long-term memory.
The FMR1 pre-mRNA from mammals and D. melanogaster is alternative spliced to produce FMRP isoforms, but the necessity for these isoforms in promoting neural plasticity is largely uncharacterized. The Drosophila fragile X model facilitates a molecular genetics approach to probe the contribution of FMRP isoforms to neural function. We have explored the importance of dFMR1 isoforms that differ in the presence of a protein interaction domain related to those found in prions, and find that their expression is necessary for proper modulation of memory establishment pathways.
The genome of D. melanogaster encodes one FMR/FXR protein, while vertebrate genomes encode three. FMR/FXR proteins are highly conserved in their N-terminal RNA binding domains, but are more diverged in the C-terminal region (see Wan et al., 2000 for alignment). It was initially presumed that both redundant and distinct functions were possible among vertebrate FMR/FXR proteins, and recent studies now support this hypothesis. The KH2 domains from the mouse and fly FMR/FXR proteins recognize an RNA substrate with a kissing complex structure, but high-affinity recognition of a select G-quadruplex RNA substrate is unique to mammalian FMRP (Darnell et al., 2009). The highly related CYFIP1 and CYFIP2 proteins have differential binding profiles with FMR/FXR proteins (Schenck et al., 2001). Behavior and neurophysiology studies of FMR1 FXR2 double knockouts shows that while FMRP and FXR2 have overlapping function in controlling mouse circadian locomotion (Zhang et al., 2008), distinct synaptic plasticity phenotypes exist in the single mutants (Zhang et al., 2009). Muscle and cardiac development phenotypes arise from morpholino knockdown of FXR1 expression in zebrafish, and the function and/or low expression of FMRP or FXR2 in these tissues fails to compensate (van't Padge et al., 2009). It is thus likely that expression pattern, RNA binding, and protein interaction partners all factor into whether FMR/FXR proteins can or cannot provide compensatory function. The C-terminal peptides of hFMRP and dFMR1 are implicated as protein interaction modules (Menon et al., 2004; Dictenberg et al., 2008; this study), and the moderate degree of similarity between C-terminal peptides of FMR/FXR proteins (Figure 1E) suggest a related function. Divergence between C-terminal sequences of vertebrate FMRP, FXR1, and FXR2 may facilitate diversification and specialization of their function via the ability to interact with distinct subsets of proteins.
Protein domains with elevated levels of Q/N residues are quite common in eukaryotic proteomes (Michelitsch and Weissman, 2000; Harrison and Gerstein, 2003), and in fungi, a subset of Q/N-rich proteins form prions (Shorter and Lindquist, 2005; Alberti et al., 2009). Adoption of a prion-like state has been suggested as a means by which regulators of neural plasticity can be stabilized to maintain long-term memory (Si et al., 2003, 2010; Bailey et al., 2004). While our results indicate that the dFMR1 Q/N domain does not behave as a prion, it is important to note that most Q/N domains do not form prions (Michelitsch and Weissman, 2000; Sondheimer and Lindquist, 2000; Alberti et al., 2009). Nonetheless, Q/N domains have properties relevant to the roles that FMRP plays in dendrites and the post-synaptic compartment. High levels of Q/N residues promote protein-protein interaction through the hydrogen bonding that can occur via the side chains of these amino acids (Perutz et al., 1994) and mediate both homologous and heterologous protein interactions (Decker et al., 2007; Guo et al., 2007; Shewmaker et al., 2007; Kim et al., 2008). The Q/N domain of yeast Lsm4p is needed for assembly and aggregation of RNA-regulating P bodies (Decker et al., 2007). Dendritic RNA transport and regulated protein synthesis are necessary for mGluR-induced long-term depression and other forms of synaptic plasticity and memory (Huber et al., 2000; Costa-Mattioli et al., 2009). The memory deficits seen with the ΔQ/N mutants are consistent with the Q/N domain facilitating interactions with ribonucleoprotein (RNP) complexes that mediate RNA transport and translation regulation.
It is interesting to speculate that the ancestral FMR/FXR protein had a Q/N domain, as the high degree of conservation in terminal amino acids of dFMR1 and mammalian FMR/FXR proteins (LVNGVS consensus, Figure 1E) suggests that the exons encoding these peptides share a common ancestor. Duplication of the ancestral FMR/FXR gene would allow divergent evolution of codons that specify C-terminal peptides. A similar situation is possible with the RNA-binding Orb/CPEB (cytoplasmic polyadenylation element binding) protein family. While the Aplysia CPEB and D. melanogaster Orb2 proteins have a Q/N-rich domain, the four mammalian CPEB paralogs have comparatively modest levels of Q/N residues (Si et al., 2003; Theis et al., 2003; Keleman et al., 2007; Richter, 2007).
Although the dFMR1 Q/N+ isoform is important for acquisition of short- and long-term memory, it is by itself insufficient for development of long-term memory. Given the need for dFMR1 Q/N+ and Q/N- isoforms in short- and long-term memory, a question then is how these isoforms contribute to changes in neurons and synapses undergoing the transitions necessary to form and maintain LTM. The deficits in both short- and long-term memory observed with fragile X models arise at least in part from misregulated protein synthesis (Dölen et al., 2007; Bolduc et al., 2008; Liao et al., 2008). Thus, a possible explanation for the roles of dFMR1 Q/N+ and Q/N- isoforms to formation of short- and long-term memory is that there are differential patterns of gene expression that the two isoforms regulate. Multiple protein interaction domains are present in FMRP (Figure 1A), and absence of the Q/N domain may allow assembly of the Q/N- isoform into functionally distinct subsets of ribonucleoprotein (RNP) complexes that facilitate gene regulation events needed for LTM. The RNA-binding CPEB protein Orb2 from D. melanogaster has a glutamine-rich domain essential for LTM but not STM of conditioned courtship training, and it is suggested to function as an interaction domain for factors that modulate Orb2 activity (Keleman et al., 2007). Both transcription and local translation can provide the de novo protein synthesis required for longer-lasting forms of plasticity (reviewed by Kelleher et al., 2004; Hoeffer and Klann, 2007; Barrett and Wood, 2008; Costa-Mattioli et al., 2009). The activity of Aplysia CPEB protein is needed up to 48 hrs after serotonin stimulation to maintain long-term facilitation (Miniaci et al., 2008), and regulated translation of ATF-4 represses CREB-mediated LTP and LTM (Costa-Mattioli et al., 2007). While the role for FMRP in post-synaptic translation regulation is widely accepted, it may also regulate gene expression within the nucleus. Isoforms of hFMRP lacking exon 14 have a predominant nuclear localization (Sittler et al., 1996), and roles for FMRP in chromatin regulation, pre-mRNA splicing, and nucleocytoplasmic trafficking of mRNA are reported or proposed (Deshpande et al., 2006; Didiot, 2008; Lai et al., 2006). Figure 7 depicts possibilities for dFMR1-regulated pathways of gene expression that contribute to formation and consolidation of memory.
The need for specific FMRP isoforms in neural plasticity makes it important to consider the mechanisms that dictate alternative splice patterns of FMR1 pre-mRNAs. The nature of processes that control spatial and/or temporal expression patterns of FMRP isoforms are not known, but human FMRP binds to a G-quartet-forming segment of its transcript, where it is then proposed to autoregulate FMR1 pre-mRNA splicing (Didiot et al., 2008). Synaptic activity controls alternative splicing of the pre-mRNA encoding the NR1 subunit of the NMDA receptor (Mu et al., 2003), and it is interesting to note that components of the splicing machinery are detected in dendrites (Glanzer et al., 2005). FMR1 mRNAs are present in synaptoneurosomes and translated in response to neurotransmitter stimulation (Weiler et al., 1997). Given the proposed function of FMRP as a regulator of alternative splicing (Didiot et al., 2008), the possibility that synaptic activity could control a dendrite-localized splicing function for FMRP and/or FMR1 transcripts is intriguing. Changes in the profile of FMRP isoforms could also occur via selective protein turnover, and at least some of the FMRP pool is subject to ubiquitination and degradation within minutes after chemical stimulation of hippocampal slices (Hou et al., 2006). Future studies will reveal whether regulated synthesis and degradation of select FMRP isoforms arises from synaptic stimulation.
FMRP is implicated in regulating the expression of synaptic plasticity genes such as Arc/Arg3.1, CaMKIIα, and Map1b (reviewed by Bassell and Warren, 2008). The role for FMRP in both short and long-term memory of training is consistent with the possibility that it is a component of a synaptic tag and/or contributes to its synthesis (Martin and Kosik, 2002) and the protein interaction capacity of the FMRP C-terminal peptide may facilitate its capture by tagged synapses. It will be of interest to find if there are FMRP-regulated genes whose products promote a distinct phase of memory establishment. The availability of animal stocks that express a specific isoform of FMRP will be useful as biochemical tools to capture FMRP-containing RNP complexes. Such complexes might harbor RNA substrates that are unique to a particular FMRP-RNP and identification of these RNAs will help elucidate the patterns of gene expression that contribute towards consolidation of memory.
Supplementary Figure legends
Figure S1: Neuromuscular junction boutons in dFMR1 ΔQ/N mutants are increased over wild type in larval muscles 4 and 6/7. * indicates P < 0.05, and *** indicates P < 0.001, ANOVA.
Figure S2: Heterozygosity for the dfmr1 ΔQ/N allele does not elicit a dominant effect on rhythmic circadian locomotion patterns. Flies expressing a wild type and ΔQ/N allele of dfmr1 were tested for circadian locomotion. The percentage of such flies judged rhythmic (S2A), and the strength of the rhythmicity (S2B) are similar to those from wild type flies. N ≥ 17 for all genotypes examined.
Figure S3: Memory indices of flies expressing dfmr1 ΔQ/N or dfmr1L alleles. S3A) Formula used to calculate a memory index from the courtship indices of naive and trained flies from a genotype. S3B) While flies expressing dFMR1 ΔQ/N display immediate recall of courtship training, the relative strength of the memory is intermediate to that displayed by flies with wild type or null alleles of dfmr1. S3C) At one hour post-training, the memory index of wild type flies remains unchanged. In contrast, the memory index of flies with a null allele of dfmr1 has declined to zero, and the value for the ΔQ/N flies is reduced to half the value seen at immediate recall. S3D) The memory indices of flies expressing the dfmr1L allele at one hour post-training vary in a manner that may be dependent on the insertion site of the transgene.
Figure S4: The inability to express the dFMR1 Q/N- isoform does not significantly alter neural development or circadian locomotion phenotypes. Flies expressing a dfmr1L allele as the sole source of dFMR1 protein were tested for midline crossing of mushroom body β-lobe axons (S3A), numbers of larval neuromuscular junction boutons (S3B-D), and circadian locomotion (S3E, F) as described in the text and Materials and Methods. For the mushroom body and NMJ bouton phenotypes, N ≥ 30 for each genotype tested. N ≥ 25 for each genotype tested in the circadian analyses. NS = not significant. ** indicates P < 0.01, and *** indicates P < 0.001, ANOVA or Fisher's exact test.
We thank Neal Ferrick and Allison Terlizzi for technical assistance. Justin Fallon and Reed Wickner provided helpful discussions. Support from the faculty of the UT BCMB Dept. and Dr. Thomas McDonald (AECOM) was vital to this effort and is gratefully acknowledged. Mike O'Neil and the Research Computing Support group from the UT Office of Information Technology provided advice on statistics. Antibodies were obtained from The University of Iowa Developmental Studies Hybridoma Bank, and the labs of Susan Lindquist, Kausik Si, and Rodney Rothstein provided yeast strains and plasmids used for these studies. This work was funded by NIH grant GM068468 to TCD and by grants from the FRAXA Research Foundation to TCD, CHC, and SMB.