Neuronal mRNAs travel singly into dendrites

Abstract

RNA transport granules deliver translationally repressed mRNAs to synaptic sites in dendrites, where synaptic activity promotes their localized translation. Although the identity of many proteins that make up the neuronal granules is known, the stoichiometry of their core component, the mRNA, is poorly understood. By imaging nine different dendritically localized mRNA species with single-molecule sensitivity and subdiffraction-limit resolution in cultured hippocampal neurons, we show that two molecules of the same or different mRNA species do not assemble in common structures. Even mRNA species with a common dendritic localization element, the sequence that is believed to mediate the incorporation of these mRNAs into common complexes, do not colocalize. These results suggest that mRNA molecules traffic to the distal reaches of dendrites singly and independently of others, a model that permits a finer control of mRNA content within a synapse for synaptic plasticity.

A number of mRNA species are selectively enriched in neuronal dendrites (1–3). It is believed that they are transported there in a translationally repressed state (4, 5). At synapses that receive sustained stimulation, these mRNAs are locally translated into proteins needed for synaptic differentiation (4–6). This phenomenon is important for the establishment of long-term potentiation and is considered an important cellular correlate of memory (7).

The transport of mRNAs into dendrites is controlled by proteins that bind to RNA motifs called dendritic targeting elements (DTEs), which are often present in the 3′-UTR of these mRNAs (4, 5, 8). The importance of DTEs is underscored by the example of calmodulin-dependent protein kinase IIα (CaMKIIα) mRNA, in which the deletion of the DTE causes the loss of its dendritic localization (9), a drop in the expression of the protein in the dendrites, and an impairment of synaptic plasticity in the organism (10).

It is believed that dendritically localized mRNAs travel as components of ribonucleoprotein complexes (often referred to as RNA transport granules) that serve to keep the mRNA translationally repressed within the soma and during transit (11–13). Distinct from other RNA granules such as stress granules and processing bodies, these granules interface with the cell's active transport mechanisms (14) and carry components of the translational machinery (11, 15). RNA granules contain more than 40 different proteins, some of which are known to bind to the DTEs of transported mRNAs, some of which are part of the translation apparatus, and some of which are associated with the cell's active transport machinery (14, 16). The granules appear to be heterogeneous, as each protein is not found in all of the granules (16).

A number of studies suggest that several different mRNA species can coassemble within the same RNA granule (11, 17–20). Pairs of in vitro-transcribed mRNAs labeled with distinct fluorophores and coinjected into neurons were found to colocalize in common particulate objects (18–20). Endogenous mRNAs encoding CaMKIIα, neurogranin (Nrgn), and activity-regulated cytoskeleton-associated protein (Arc) were found to be colocalized in the same set of granules by fluorescence in situ hybridization (FISH) (18). Also, the fact that individual RNA granules can be imaged with the RNA staining dye SYTO 14 implies that these granules possess a high RNA content (21). Coassembly of more than one mRNA species into the same RNA granule has also been observed in other biological contexts. For example, ASH1 and IST2 mRNAs colocalize in the same structure in the bud tips of yeast (22), and wingless and pair-rule mRNAs travel together to the apical surface of syncytial nuclei in early Drosophila embryos (23).

In contrast, a pair of recent studies examined several dendritic mRNAs by FISH and found that some were located in common granules and others were not (20, 24). Another study performed on fibroblasts found that, although the mRNAs encoding arp2, arp3, and β-actin are all concentrated at the leading edges, they do not colocalize with each other (25).

We studied the colocalization of mRNA molecules within neurons, using an imaging technique in which endogenous mRNAs are detected in situ with single-molecule sensitivity (26). Our results demonstrate that the dendritic mRNAs in cultured hippocampal neurons occur singly; i.e., two or more molecules of the same or different mRNA species do not occur together in common complexes.

Results

Different Species of Dendritic mRNAs Are Not Confined in Common Complexes.

We obtained single-molecule sensitivity by using ∼50 different hybridization probes, each labeled with a single fluorescent dye, that simultaneously bind to the same mRNA target at different sites within the molecule. Binding of these probes render each mRNA molecule so intensely fluorescent that it can be seen as a bright diffraction-limited spot, whereas background signals from unhybridized probes are of lower intensity and are diffused. Therefore, image processing programs can readily identify and count these spots and are able to determine the location of their centers with great precision (26). Evidence for extremely high specificity and single-molecule sensitivity of this approach and of the accuracy of the spot-detection algorithm is provided in SI Text and in Figs. S1 and S2.

We imaged pairwise combinations of eight mRNA species (Arc, β-actin, CaMKIIα, Ef1α, MAP2, Nrgn, PKMζ, and Ube3a) that have all previously been characterized as being dendritically localized (4). One member of each pair was detected with probes labeled with tetramethylrhodamine, and the other member was detected with probes labeled with Alexa 594. For each mRNA species, we observed discrete spots corresponding to individual mRNA molecules. Fig. 1 shows representative results obtained from 4 of the 28 pairwise probe combinations that were studied. Each image is composed of the merged optical slices of an entire cell (z-stack). The third panel from left in each pairwise combination represents the color-coded merges of the first two panels. A cursory examination of the merged images indicates that the molecules of the two mRNA species in each pairwise combination do not colocalize with each other.

Fig. 1.

Individual molecules of different mRNA species do not colocalize with each other. Single-molecule FISH was performed on cultured hippocampal neurons in pairwise combinations of indicated mRNA targets. Gray-scale fluorescence images in the two Left column panels are merged images (z-stacks) obtained from each fluorescence channel. The color images in the third column were obtained by coding the images in the first column green, the images in the second column red, and then merging them. The z-stacks in both channels were analyzed using an algorithm that finds spot-like signals in each image, determines their 3D coordinates, and then looks for spots in one channel that have a counterpart in the other channel that lies within a distance of 250 nm; and if a pair of spots is found to meet that criterion, the algorithm classifies them as being colocalized. The locations of these colocalized molecules (displayed as yellow circles) along with the locations of solitary molecules (displayed as red or green circles) are overlaid onto DIC images of the neurons in the Right column panels. Although all 28 pairwise combinations were imaged (Table 1), only 4 of the 28 pairwise combinations are shown here, as these pairs provide representative results for each of the eight mRNAs that were studied. (Scale bars, 5 μm.)

To explore the degree of colocalization between different mRNA species quantitatively, spots corresponding to the individual molecules of mRNA labeled in each color were computationally identified, and the 3D coordinates of their centers were determined (Fig. S1). For each spot of a given color, we measured the distances between that spot and all other spots labeled in the other color, thereby determining the distance between that spot and the nearest spot of the other color. The distributions of these “nearest neighbor” distances are presented Fig. S3. If a significant fraction of the two mRNAs in each pair were to be confined within objects as small as an RNA granule, we would have seen a peak corresponding to the size of an RNA granule. Instead, these distributions have means that range from 800 to 1,300 nm. Almost the same distributions are obtained between different molecules of the same mRNA species (Fig. S3 C and D), indicating that they relate merely to the population density of the mRNAs and to the dimensions and geometry of the neurons. The distributions of sparsely expressed mRNA exhibit longer tails, again suggesting that the distributions relate to the density. These results indicate that the mRNA molecules of two species in a pair do not have a strong tendency to cluster together in objects smaller than the cell size.

Size of RNA Granules.

Estimates of the size of neuronal RNA granules range from 100 to 700 nm (11, 15, 16, 21, 27). This uncertainty reflects the variations in the preparation and sources of neurons and homogenates used in different studies. Electron microscopic observations of fractionated homogenates of whole brain and cultured neurons indicate a size range of 100–250 nm (11, 16, 27), whereas a cross-correlation analysis of FISH images in oligodendrocytes suggests a range of 600–700 nm (15). Because the former estimate is more direct, we consider the diameter of RNA granules to be about 250 nm. Because the limit of optical resolution is about the same, a priori, one may expect the spots in two different channels produced by two molecules that are confined within a granule to precisely overlap with each other. However, whereas, on one hand, the centers of diffraction-limited spots can be determined within a few nanometers (28, 29), on the other hand, spectral shifts and other microscopic aberrations may cause the two spots to diverge from each other.

To determine these distances empirically, we imaged each half of the molecules of the same mRNA species (MAP2) with a differently colored set of probes (Fig. 2A). The merged raw images in each color (Fig. 2B) indicate that most mRNA molecules yield a spot labeled in both colors. The distances from the center of each green spot (corresponding to the left half of MAP2 mRNA) to the center of the nearest red spot (right half of MAP2 mRNA) were determined, and their distributions are shown in Fig. 2C. In sharp contrast to the distributions of nearest neighbor distances between two different mRNA species, whose means range from 800 to 1,300 nm, and whose half-maximum heights spread over a range of 1,100–1,900 nm, the distances between the two differently colored spots labeling the same mRNA molecule cluster in a narrow range that centers at 110 nm and spread only as far as 250 nm. An identical distribution was obtained when differently colored probes for mRNA were intermixed by targeting them to alternate sites along the entire length of the mRNA, indicating that the distances measured are not intramolecular distances but instead represent the accuracy of measurement of the same diffraction-limited object in two different channels.

Fig. 2.

Determination of empirical distances between centers of spots produced by two probe sets bound to different regions of the same target mRNA. (A) Two adjacent regions of MAP2 mRNA were probed with two probe sets labeled with distinct fluorophores. (B) Merged raw images of MAP2 mRNA molecules from two fluorescence channels were created while coding the image from one channel with green color and the other with red color. (Scale bar, 5 μm.) (C) Distribution of distances from the centers of the green spots to their nearest-neighbor red spots. (D) Classification of spots in B based on the colocalization distance limit of 250 nm. We found that 73% of all molecules were colocalized, indicating that each probe set may be able to detect 85% of the molecules (SI Text).

Because, fortuitously, the upper limit of the granule size determined from electron microscopy and the empirical limit obtained above are the same, we classified spots whose centers were less than 250 nm apart as being colocalized. The locations of these colocalized pairs, along with the locations of solitary molecules, are overlaid on differential interference contrast (DIC) images (Fig. 1, Right panels). An enlarged image of a pair of spots that are identified as colocalized and another that has closely placed spots, but were considered not to be colocalized by our algorithm, are shown in Fig. S2C.

The percentage of colocalization obtained by dividing the number of colocalized mRNAs by the total number of mRNAs in each of the 28 pairwise combinations is presented in Table 1. These low frequencies of colocalization indicate that different mRNAs in the 28 pairwise combinations that were studied have little or no tendency to colocalize. Increasing the limit of colocalization to 300 nm did not change the frequencies of colocalization appreciably. In addition, we present for four pairs of mRNAs, how the frequency of colocalization changes as the colocalization distance is changed continuously up to and beyond the size of the cell in Fig. S3. These distance distributions suggest that the mRNAs have no clear tendency of clustering at any subcellular distance.

Table 1.

Percentage of total mRNA molecules that were found colocalized within 250 nm of each other in pairwise combinations

	Arc	MAP2	CaMKIIα	Nrgn	PKMζ	Ube3a	Ef1α
β-Actin	1.80 ± 0.62	2.25 ± 0.33	1.56 ± 0.65	2.46 ± 0.60	1.19 ± 0.39	2.18 ± 0.52	1.60 ± 0.39
Arc		2.09 ± 0.67	3.35 ± 0.82	0.42 ± 0.26	1.13 ± 0.44	0.75 ± 0.53	2.48 ± 0.60
MAP2			3.06 ± 0.54	3.36 ± 0.58	1.82 ± 0.54	3.46 ± 0.97	1.79 ± 0.31
CaMKIIα				1.87 ± 0.56	1.74 ± 0.62	1.97 ± 0.63	2.61 ± 0.86
Nrgn					2.23 ± 0.39	0.33 ± 0.28	0.54 ± 0.16
PKMζ						2.59 ± 0.90	1.37 ± 0.54
Ube3a							0.62 ± 0.21

The percentage of colocalization was calculated by dividing the number of molecules of the lower-abundance mRNA species that were colocalized with the other mRNA species by the total number of mRNA molecules (within a ±95% confidence interval, where at least 10 neurons were examined). The number of spots analyzed ranged from 1,148 to 12,298.

Significantly, the frequencies (obtained using the 250-nm limit) were lower in the dendritic compartments than in the bodies of the neurons, where the RNAs are relatively more crowded, indicating that, to the extent that we observe any colocalization, it reflects the likelihood of finding the two molecules in close proximity by chance, rather than due to any specific coassembly into a molecular complex (Tables S1 and S2). Stimulation of neurons by the addition of bicuculline or dihydroxyphenylglycine into the culture medium did not change the frequencies of colocalization appreciably. Similarly, increasing the duration of neuronal culture from 7 to 10 d to 15 to 18 d did not change the frequencies of colocalization substantially.

It can be argued that mRNAs within RNA granules can be so tightly sequestered that they are not accessible to the probes. To address this concern, we imaged one of the mRNAs along with the RNA granule marker protein Staufen 1, which has been shown to associate with the CaMKIIα mRNA (30). Consistent with previous observations, more than 60% of CaMKIIα mRNA molecules were colocalized with Staufen 1 signals, which also appeared as particulate (Fig. S4). Interestingly, the sizes and intensities of the Staufen 1 spots varied much more than the sizes and intensities of the mRNA spots and, consistent with previous studies, other mRNAs did not colocalize with Staufen 1 (30). Furthermore, mRNAs sequestered within the stress granules have been detected using our approach (30), suggesting that mRNA in the granules is accessible to the probes under hybridization conditions.

Dendritic mRNAs Show No Propensity for Self-Oligomerization.

To explore whether multiple copies of the same mRNA species are present within RNA granules, we first measured the average number of mRNA molecules per neuron for one of the dendritic mRNA species, MAP2, by FISH and then compared that result to the number of MAP2 molecules per neuron measured by real-time RT-PCR. The number of MAP2 mRNA molecules obtained by the two methods was similar (90 ± 15 molecules per neuron by RT-PCR versus 109 ± 44 molecules per neuron by single-molecule FISH) (Fig. 3). Furthermore, bicuculline stimulation resulted in similar increases in the two measurements (Fig. 3). This analysis was repeated for β-actin mRNA, which resulted in a similar correspondence between the two measurements (Fig. S5). The concordant results obtained by these independent measurements indicate that very few, if any, of the spots that were identified by imaging represent granules possessing multiple copies of the same mRNA.

Fig. 3.

The number of molecules of MAP2 mRNA per neuron obtained by counting spots in the FISH images is very similar to the number obtained by real-time PCR performed on RNA extracted from the neurons. (A) Merged z-stacks of neurons imaged after hybridizing with probes for MAP2 mRNA. (Scale bar, 5 μm.) (B) Identified spots are overlaid onto the image shown in A to demonstrate that almost all MAP2 mRNA molecules are counted by our algorithm. (C) Quantification of mRNA copy number by real-time RT-PCR. A standard curve obtained from serial dilutions of full-length in vitro-transcribed MAP2 RNA is shown. The cyan dot identifies the result obtained from a reaction initiated by RNA isolated from resting neurons, and the pink dot identifies the result obtained from a reaction initiated by RNA isolated from the same number of neurons that had been exposed to bicuculline. (D) Comparison of the mRNA molecules per neuron determined by single-molecule FISH with the mRNA copy number per neuron determined by real-time PCR. The error bars represent 95% confidence intervals.

Any tendency to multimerize would manifest itself in spots of higher fluorescence intensity. If multiple mRNA molecules are confined within an area smaller than the diffraction limit, the apparent fluorescence intensity at that spot should be the sum of the individual intensities of the mRNA molecules present. We measured the maximum intensities within the spots for each dendritically localized mRNA species. The resulting distributions of spot intensities for dendritic populations of MAP2, β-actin, and brain-derived neurotrophic factor (BDNF) mRNAs are shown in Fig. 4. The distributions of intensities of each mRNA species fit well to a unimodal Gaussian distribution (Fig. 4, blue curves). Because the same numbers of singly labeled oligonucleotides (48) were used to image each of these three mRNAs, the mean intensity of their distributions should be, and were, similar. Furthermore, the mean intensity does not change for a given mRNA species when the number of molecules expressed in the cell is increased substantially by up-regulation (Fig. S2B). Although these unimodal Gaussian distributions indicate a priori that most of the granules contain only one “kind” of object, we realized that it was desirable to demonstrate that these optical measurements would indeed be able to identify complexes containing two or more mRNA molecules. We therefore prepared a unique control for each of these three mRNA species.

Fig. 4.

The maximum intensities of spots produced by three different dendritic mRNAs follow unimodal distributions. (A–C) Upper panels show the data obtained with a single set of 48 probes against native mRNAs, and Lower panels are different controls designed to show how the distributions would have looked if the population contained a significant proportion of two or more mRNA molecules within a single granule. To the right of each histogram is a representative image of one of the cells from the population. (A) MAP2 mRNA either bound to one set of probes that is specific for one region of each mRNA molecule (Upper) or bound to two sets of probes that are each specific for different regions of the same mRNA molecule (Lower). (B) A set of probes that bind specifically to natural β-actin mRNA was used to image uninfected neurons (Upper), and the same set of probes was used to image β-actin mRNA in neurons that had been infected with a lentiviral construct that expresses a tandem dimer of β-actin mRNA that was fused to a sequence encoding GFP (Lower). Approximately half of the spots in the images of neurons that contain mRNA molecules possessing the β-actin dimer were twice as intense as the other spots (which arise from endogenous β-actin mRNAs). (C) Two different sets of probes were bound to naturally occurring isoforms of BDNF mRNA in neurons. One set of probes bound to a region of the BDNF mRNA that is common to both isoforms, and the other set of probes bound to a region of the BDNF mRNA that occurs only in the longer isoform. Histogram and images resulting from the first set of probes and from both sets of probes. The images to the right of B and C are overlaid with circles that identify spots of unit intensity (red) and of double intensity (green). The intensities of all of the spots were fitted to a unimodal Gaussian distribution (blue curves) and to a mixture of two Gaussian distributions (red curves). The number of spots analyzed ranged from 446 to 2,306. (Scale bar, 5 μm.)

In the case of MAP2 mRNA (Fig. 4A), we prepared a second set of 48 probes labeled with the same fluorophore that bind to a different region of MAP2 mRNA, and we used both sets of probes simultaneously to image native MAP2 mRNAs. Analysis of the resulting images shows that the intensity of the entire population of MAP2 mRNAs is about twice as great as the mean intensity obtained with the first set of probes alone (after subtracting the background intensity from both). To determine the fraction of the spots obtained with the first set of probes alone that are bright enough to be considered as “dimers” and the fraction of spots obtained with both sets of probes simultaneously that are dim enough to be considered as “monomers,” we deconvolved the distributions into two different populations (red curves in Fig. 4) with the aid of a “two-component mixed-Gaussian” analysis. In this procedure, the proportions, the means, and the SDs of two Gaussian distributions (five free parameters) are iteratively varied until the two distributions that “best fit” the population distributions are found. The results indicate that “dimers” are a very small fraction (5%) of the spots obtained with the single probe set and that “monomers” are an extremely small fraction (1%) of the spots obtained with the two probe sets combined, indicating that MAP2 mRNA almost always occurs as single molecules in the dendrites.

In the case of β-actin mRNA (Fig. 4B), we constructed an artificial gene that contained two tandem copies of β-actin mRNA fused to the coding sequence of GFP. This construct was expressed in the neurons via a lentiviral vector. We probed neurons expressing both the dimeric β-actin mRNA from this construct and the natural β-actin mRNA of the cell, using the same set of 48 probes. Images obtained from these infected neurons showed two kinds of spots: (i) spots displaying lower intensity, similar to the intensity of native β-actin mRNA expressed in uninfected cells, and (ii) brighter spots due to the presence of the engineered β-actin dimers that were expressed from the lentivirus. The deconvolution of spot intensity distributions indicated that 52% of the mRNA molecules in the infected cells were twice as intense as the natural β-actin. By comparison, the same deconvolution algorithm found that only 2% of the mRNA molecules in the uninfected neurons were twice as intense. Using the intersection of the two Gaussian fit curves as a “cut-off intensity,” we classified the spots (Fig. 4B, Right panels) into low-intensity native β-actin mRNA molecules (red circles) and high-intensity dimeric lentiviral β-actin mRNA molecules (green circles). The infected neurons had many spots of high intensity, whereas the uninfected neurons had only occasional high-intensity spots. Significantly, when the RNAs were also probed with GFP-coding–specific probes, the higher intensity mRNAs were more likely to show colabeling with those probes than mRNAs with unit intensity. These observations indicate that naturally occurring β-actin mRNA is almost always present as a monomer within the neurons.

In the case of BDNF mRNA (Fig. 4C), nature provides an appropriate control. BDNF mRNA exits in two isoforms that differ in the length of their 3′-UTR (Fig. S6). Usually about 15–20% of BDNF mRNA has the longer 3′-UTR (31), which is thought to enable this isoform to travel further within a dendrite. We imaged neurons with two sets of probes: one probe set specific for the coding sequence that is common to the two isoforms, and the other probe set specific for the portion of the 3′-UTR that is found only in the second isoform. Both probe sets were labeled with the same fluorophore. We found that 17% of the BDNF mRNA spots in neuronal images displayed a higher intensity than the rest. On the other hand, when only the probe set complementary to the common coding region of the two isoforms was used, only the lower intensity population could be detected. We also imaged BDNF mRNAs with a pair of probe sets that had differently colored fluorophores to distinctly label the longer isoforms. The results of these experiments indicated that 16% of the spots were due to the presence of the longer isoforms (Fig. S6).

These results indicate that BDNF mRNA is also not found as multimers. Indeed, the other six species of dendritic mRNAs that we imaged also displayed unimodal distributions in their intensities, confirming that, in general, dendritic mRNAs do not exist as multimers in neurons (Fig. S7).

mRNAs with a Common DTE Do Not Coassemble into the Same RNA Granules.

A rationale proposed for the coassembly of multiple mRNAs into the same transport granule is that transacting proteins bind to common sequence elements present in different mRNAs (17, 20) that are linked together by multivalent scaffolding proteins (17). One such motif, the A2 response element that binds to the heterogeneous nuclear ribonucleoprotein A2, is so short and degenerate that it is found in many mRNAs (18). The dendritic mRNAs CaMKIIα, Nrgn, and Arc contain this sequence element, and it has been reported that more than 70% of these mRNAs are coassembled within common granules (18–20). However, our pairwise colocalization analysis indicates that these mRNAs colocalize with a frequency of 3.35% or less (Table 1 and Table S1). To further explore this issue, we performed FISH simultaneously with three differently colored sets of probes and identified all seven possible combinations of solitary and colocalized mRNA molecules (Fig. S8A). The results of these experiments demonstrate that these mRNAs do not coassemble into common granules. The frequency of colocalization of all three mRNAs within the same granules was found to be less than 1% and is probably an artifact of crowding.

Would longer, well-characterized DTEs mediate coassembly of mRNAs into common granules? The DTEs in several mRNAs have been identified, but none is shared among two or more different mRNA species (4, 20). We therefore created an artificial mRNA encoding GFP and containing the 54-nucleotide-long DTE that occurs in the 3′-UTR of β-actin mRNA (the “β-actin zip code”). Neurons were infected with a lentiviral construct expressing this artificial GFP mRNA, and the mRNA in these cells was then simultaneously hybridized in situ to a probe set that is specific for the coding sequence of β-actin mRNA and to a differently colored probe set that is specific for the coding sequence of GFP mRNA (Fig. S8B). The results indicate that, even though both of these mRNAs possess an identical β-actin zip code, and both of these mRNAs migrate to dendrites, they do not coassemble into common RNA transport granules.

Discussion

Several hundred different species of mRNAs have been identified to be in the dendritic compartment in different genome wide screens (1–3). However, the mRNA sets identified in different screens have little overlap with each other. For example, less than 2% of the 260 mRNA species identified in the dendritic compartment by Eberwine et al. (1) were also present in the set of 177 species identified by Poon et al. (2). Therefore, rather then selecting RNAs from these sets, we chose only those that have been individually characterized to be present within dendrites. However, our results suggest that, absent any specific mechanism that brings together mRNAs of a particular pair of species, neuronal mRNAs usually do not cluster together.

How can we reconcile these results with earlier observations in which several different mRNA species were seen to coassemble in common granular structures? The key difference between our method and the methods used in earlier studies is that we use directly labeled probes, which tether the signal to the target. On the other hand, previous methodologies use probes labeled with haptens, which bind to antibodies linked to enzymes that, in turn, convert a small diffusible nonfluorescent substrate (usually tyramide) into a fluorescent precipitate (18, 20). While giving strong fluorescent signals, this technique leads to the spread of the fluorescence signal away from the original location of the target molecule. Consistent with this hypothesis, the spots corresponding to the target mRNA in the previous works appear larger than what we observed and could have led to the incorrect conclusion that mRNAs of different species colocalize with each other. A second line of evidence supporting the coassembly of multiple mRNA molecules within common granules comes from injection (usually into the cytoplasm) of in vitro-produced and -labeled mixtures of two different mRNA species (18–20). Under these artificial conditions, the amount of mRNA is much greater than what is naturally found in the cell, and the mRNAs are not bound to the retinue of proteins that normally bind to the mRNAs during their synthesis in the nucleus, making aberrant assemblages of mRNAs possible. Furthermore, movement of mRNA molecules in live neurons can cause the dispersal of the fluorescence signals, resulting in an overestimation of colocalization frequency.

Would a larger fraction of mRNAs be found colocalized if a larger limit for colocalization is used? We determined the colocalized fractions for all distances up to the size of neurons (Fig. S3) and found that the number of colocalized molecules increased substantially. However, the distributions of these distances do not show any specific tendency of coassembly into subcellular objects of fixed sizes; instead, these distributions are expected from molecules that are scattered randomly within the confines of neurons.

mRNAs bind to a number of proteins in the nucleus that determine their cytoplasmic fates (32). They are produced at different times at disparate gene loci where they bind cotranscriptionally to their entourage of proteins and are then exported rapidly to the cytoplasm after their release from the chromosome (for most mRNA species, we see only a few molecules in the nucleus at any given moment). Thus, there is little opportunity for them to form multimeric complexes in the nucleus. Furthermore, there is evidence for some mRNAs, such as β-actin mRNA, that transacting proteins that bind to them in the nucleus stay bound to them until they reach their cytoplasmic destination (33). To the extent that these mRNA–protein complexes are remodeled in the cytoplasm of neurons to yield functional RNA transport granules, our results indicate that multiple mRNA molecules are not coassembled within common granules. On the other hand, in the case of oskar mRNA in fruit fly oocytes—a system in which mRNAs have clearly been shown to assemble into multimeric complexes—there exists a specific protein that mediates the process of multimerization (34).

The dendritic mRNAs serve diverse, independent, and still poorly understood roles in the process of synaptic differentiation. They are needed at the activated synapses in unique stoichiometries and at unique times. If mRNA granules were composed of multiple mRNAs, we would have to consider highly complex models according to which a granule assembly system in the soma is able to select and package multiple sets of mRNAs into common granules, transport them through dendrites, and then release individual mRNA species from the multiplex granule at the activated synapse (17). The results obtained from our single-molecule imaging studies clarify the picture by suggesting a simpler model in which solitary mRNA molecules bound to their characteristic set of proteins travel to distal reaches of the neurons and respond to the activated synapses individually, providing a finer control to the synapse.

Materials and Methods

We detect mRNA molecules by simultaneously hybridizing about 50 probes to each mRNA species. Hippocampal neurons were cultured on coverslips. After 10 d in culture, they were fixed, permeabilized, and hybridized with probe mixtures. Imaging was performed in oxygen-depleted mounting media. The images were analyzed as described in SI Materials and Methods.

The supporting information includes SI Text, detailed SI Materials and Methods, Tables S1 and S2, Dataset S1, and Figs. S1–S8.