Network activity-independent coordinated gene expression program ...

Network activity-independent coordinated gene expression program for synapse assembly Luis M. Valor, Paul Charlesworth, Lawrence Humphreys, Chris N. G. Anderson, and Seth G. N. Grant* Genes to Cognition Programme, The Wellcome Trust Sanger Institute, Hinxton, Cambridge CB10 1SA, United Kingdom Edited by Joshua R. Sanes, Harvard University, Cambridge, MA, and approved January 20, 2007 (received for review October 13, 2006)

Global biological datasets generated by genomics, transcriptomics, and proteomics provide new approaches to understanding the relationship between the genome and the synapse. Combined transcriptome analysis and multielectrode recordings of neuronal network activity were used in mouse embryonic primary neuronal cultures to examine synapse formation and activity-dependent gene regulation. Evidence for a coordinated gene expression program for assembly of synapses was observed in the expression of 642 genes encoding postsynaptic and plasticity proteins. This synaptogenesis gene expression program preceded protein expression of synapse markers and onset of spiking activity. Continued expression was followed by maturation of morphology and electrical neuronal networks, which was then followed by the expression of activity-dependent genes. Thus, two distinct sequentially active gene expression programs underlie the genomic programs of synapse function. synaptogenesis 兩 transcriptome 兩 microarray 兩 multielectrode array

D

espite intensive efforts, little is known about the cellular programs underlying mammalian synapse formation (1, 2). A large number of studies, primarily focused on disrupting synapse proteins, have failed to find factors that are necessary for synapse formation, although their genetic manipulation has led to changes in synapse morphology or function (2). Recent descriptions of the composition of mammalian synapses from proteomic studies provide an opportunity to examine synaptogenesis from a global point of view. More than 1,000 synapse proteins have been reported in mammalian synaptosomes, postsynaptic densities (PSDs), and synaptic complexes (3). Because only a small proportion of these have been tested in synaptogenesis assays, it remains possible that the key factor will be found among these proteins. Alternatively, there may be a coordinated gene expression program for initiating synapse formation that is enacted before the expression of synapse proteins. Single-gene studies indicate that transcriptional mechanisms are of importance in synaptogenesis. For example, mRNA levels of the amyloid precursor protein have been correlated with both transcription factor binding and activation of its promoter in hippocampal cultures (4); the mutation of methyl-CpG-binding protein 2 produced a delayed neuronal maturation of the cortex and a decrease in excitatory synaptic transmission in hippocampal neurons (5, 6); and synaptic remodeling and vesicle accumulation of the zebra-fish olfactory bulb in the critical period of synaptogenesis is dependent on CREB and NFAT (7). Most importantly, the transcription factors involved in neurogenesis and neuronal fate specification, including the bHLH proneural factors, Tbr-1 and -2, Pax6 and Sox proteins (8–11), exert their influence as part of the neural differentiation program. In addition, several transcription factors can control dendritic morphology, including CREST, NeuroD, distal-less homeobox (Dlx) 1, Neurogenin2, and MEF2 (12–16). A genome-wide analysis performed in Drosophila suggests a large number of transcriptional regulators influence morphology (17). In addition to a role in development, synaptogenesis has been implicated in activity-dependent behaviors including learning (18,

4658 – 4663 兩 PNAS 兩 March 13, 2007 兩 vol. 104 兩 no. 11

19). Activity-dependent gene expression is thought to contribute to synaptogenesis, although blockade of neuronal activity or synaptic transmission does not prevent synaptogenesis in cultured primary neurons (20, 21). The overlap between transcriptional programs for both synaptogenesis and neuronal activitydependent changes remains poorly understood. Given the complexity of synaptogenesis, we reasoned that global approaches may provide appropriate coverage of this phenomenon. To examine the features of a possible synaptogenesis program and its relationship to activity-dependent gene regulation, we have integrated the information from both gene expression and multielectrode arrays on primary neuronal culture preparations. Primary neuronal cultures from an embryonic hippocampus have been extensively used as a model of synaptogenesis, in which cells from different stages of development become resynchronized after dissociation (22), and electrical activity is dependent on the connectivity of the neurons (i.e., network activity), rather than the properties of individual cells (23). Multielectrode arrays provide an efficient, spatially comprehensive, and noninvasive method of recording the development of network spiking activity over time, which has been applied successfully in hippocampal cultures (24). Gene expression arrays can determine the simultaneous profiling of thousands of transcripts. Because transcript expression correlates with protein expression in murine tissues (25–27), the transcriptome results may be used as a general indicator of protein levels. Using these approaches, we describe the relationship between network neural activity and genome-wide expression profile, and hence the temporal sequence of events that underlie synaptogenesis. Results Properties of Primary Hippocampal Cultures During Synaptogenesis.

We first characterized the morphology, electrical activity, and gene expression profiles of primary neuronal cultures from dissociated mouse E17.5 hippocampal cells. The temporal profile of synapse formation was monitored at the morphological level by using immunocytochemistry [supporting information (SI) Fig. 6]. As observed in the phase-contrast images (SI Fig. 6B), our cultures showed a typical pattern of development. Neurites were first observed 6 h after plating [0.25 days in vitro (DIV); data not shown], followed by a progressive profusion and enlargement of soma and branches (SI Fig. 6B), eventually resulting in a dense neuronal network after ⬇16-DIV. At the Author contributions: L.M.V., P.C., and S.G.N.G. designed research; L.M.V., P.C., L.H., and C.N.G.A. performed research; L.M.V. and P.C. analyzed data; and L.M.V., P.C., and S.G.N.G. wrote the paper. The authors declare no conflict of interest. This article is a PNAS direct submission. Freely available online through the PNAS open access option. Abbreviations: DIV, days in vitro; GO, gene ontology; MEA, multielectrode array; PCA, principal component analysis; PSD, postsynaptic density; TTX, tetrodotoxin. Data deposition: The microarray data have been deposited in the ArrayExpress database, www.ebi.ac.uk/arrayexpress (accession no. E-MEXP-878). *To whom correspondence should be addressed. E-mail: [email protected]. This article contains supporting information online at www.pnas.org/cgi/content/full/ 0609071104/DC1. © 2007 by The National Academy of Sciences of the USA

www.pnas.org兾cgi兾doi兾10.1073兾pnas.0609071104

synapse level, it was clear that expression of typical synaptic markers (postsynaptic: GluR1, PSD-95; presynaptic: synaptophysin) occurred early in development (SI Fig. 6A). Even as early as 4-DIV, some clusters of colocalized presynaptic and postsynaptic markers were observed, indicating the formation of assembled synapses as previously described (28). Then there was an increase in the staining patterns of the synaptic markers concomitant with the growth and an increase in arborization of dendritic processes. The timing of both phenomena was consistent with previously reported studies of dendritic spine density (29, 30). Primary neuronal cultures were plated onto multielectrode arrays (MEAs) containing 59 electrodes, which allowed the recording of action potentials in individual neurons as well as network activity. Action potentials (spikes) were first observed in MEA cultures at 4- to 5-DIV, in keeping with the results of immunostaining (see above). At this time point, activity was restricted to isolated spikes, or very brief bursts of spikes, and was only observed in a few (⬍10%) of the 59 electrodes of the MEA (Fig. 1). With increasing time in culture, spiking activity was detected at more electrodes until a plateau was reached at maturity (⬎21-DIV; Fig. 1) when, on average, about two-thirds of the array recorded responses (mean 63.7 ⫾ 5.9% SEM). This activity profile reflects the expansion of the functioning neuronal network as more neurons became synaptically connected. Over this time period, spiking activity at individual sites also increased due to an increase in the number of spikes within bursts and the frequency of bursts. Although both the number of active channels and the array-wide spike activity reached a plateau after 21-DIV, the expansion in the dimensions of the net occurred more rapidly than the growth in the magnitude of the array wide spiking response; half-maximal values were reached at 8.8- and 13.9-DIV, respectively (Fig. 1). This difference suggested that new synapses strengthened over time; i.e., the initial weak contact increased the extension of the net, and then subsequent maturation increased the spike number. Raster plots of spike activity illustrate the increase in the number of active electrodes and the activity within individual electrodes with time in culture (SI Fig. 7). To define the transcriptome profile in the development of hippocampal cultures, triplicate samples of total RNA were isolated from multiple time points (0.25-, 1-, 2-, 4-, 8-, 12-, and 16-DIV) and hybridized to MG-430 2.0 arrays (Affymetrix, Santa Clara, CA) that contained 45,037 probe sets excluding controls. With ⬎39,000 transcripts represented, we had maximal coverage of the transcriptome currently available in a commercial format. The data were normalized and filtered (see Materials Valor et al.

NEUROSCIENCE

Fig. 1. Network activity profile recorded with MEAs. Mean values of the number of active channels (squares) and the array-wide spike activity (circles) from five independent recordings. Data are expressed as mean ⫾ SEM and were fitted to sigmoidal curves (lines).

Fig. 2. Gene expression profile of the significantly changing probe sets. (a) Principal component analysis (PCA) of the samples used in the microarray experiments. The three principal components were represented by each axis and individual samples (squares) plotted (see Materials and Methods). The variance of the data that corresponds to each component was 60.6% for X, 10.8% for Y, and 4.9% for Z. The variance between sequential time points is indicated (%). Two velocities in the gene expression profile were observed (Fast and Slow). (b) Hierarchical clustering of the samples used in the microarray experiments. Red lines delimitate the main clusters (solid) and subclusters (dashed). (c) Gene expression levels (normalized signal intensities of probe sets) are represented by the mean ⫾ SD on a logarithmic scale. Coloring is based upon the expression at 0.25-DIV: red, overexpressed (⬎1); green, lowexpressed (⬍1). The arrow depicts the switch point at 4-DIV (see text).

and Methods). Linearity of the signal intensities between samples and conditions was verified (SI Fig. 8). The principal component analysis (PCA) revealed that replicates for each time point were spatially clustered as expected for similar samples (Fig. 2a). In addition, quantitative RT-PCR assays for selected transcripts confirmed the expression profiling observed in the microarray analyses (SI Fig. 9). Finally, a combination of k-means clustering and gene ontology (GO) analyses agreed with a previous functional classification of gene expression reported in developing hippocampus (31) and hippocampal (32) cultures, in which enrichment of gene products involved in general biosynthesis of DNA, RNA, and proteins (replication, transcription, RNA splicing, and translation) was found in the cluster containing the down-regulated genes, and enrichment of gene products involved in synaptic functions (receptors, channels and transporters, cell adhesion and vesicle-trafficking molecules, etc.), signal transduction (e.g., Ras-related components), and energetic metabolism (mitochondrial enzymes) was present in the cluster containing the up-regulated genes (Fig. 3; see SI Tables 1–3). Plotting the expression of the changing (filtered) probe sets versus time in culture revealed that most of the probe sets that PNAS 兩 March 13, 2007 兩 vol. 104 兩 no. 11 兩 4659

Fig. 3. Summary of the GO classification of the gene expression profile. Pie charts show the percentage of significant terms in each category (molecular function, biological process, and cellular component) and subcategories (in brackets) for the two main trends in gene expression (up- and downregulated; see SI Methods). The GO analysis was performed by using eGOn version 2.0 (www.genetools.microarray.ntnu.no) and is explained in SI Methods and SI Tables 1–3.

were expressed at low levels at the early time points became expressed at high levels at the later time points, and vice versa, through a major reversal or switch point at 4-DIV (Fig. 2c), indicating that changes in gene expression take place very early in development of the cultures. As shown in Fig. 2c, the most dynamic period of gene expression occurred before 8-DIV, which was confirmed by using PCA and hierarchical clustering. The PCA showed that the variance of gene expression during the first 8 days in culture (0.25- to 8-DIV) was ⬇65%, in contrast to the later 8 days (8- and 16-DIV) that showed only ⬇11% (Fig. 2a). Hierarchical clustering of the time points showed four main branches (Fig. 2b); one corresponding to 0.25-DIV and another from 1- to 2-DIV, a third to 4-DIV, and finally samples from 8to 16-DIV that were indeed not ordered, indicating these later stages were very closely related. Thus, changes in gene expression occurred more rapidly at early stages of development. Overlaying the three profiles (morphology, electrical activity, and gene expression), it was found that the morphological and electrical profiles coincided and that the maturation of both synapses and neural activity were thus correlated. In contrast, the gene expression profile did not fit with these phenotypic profiles because its most prominent changes occurred before the maturation of synapses (Fig. 2). The simplest hypothesis that could explain this separation is that the observed gene expression changes are responsible for synaptogenesis. Transcriptome of the Synaptic Machinery. We next focused our attention on the expression levels of synapse genes. Although GO terms are useful for global analyses, a major limitation is the state of annotations for gene products. We therefore compiled three subsets of genes (642 in total) with structural and functional relevance to synapses: 458 genes as the core postsynaptic components (core PSD genes), defined by more than one proteomic study (3) to be physically found in the PSD; 137 synaptic plasticity genes (phenotype genes), defined by mouse genetic studies of loss-offunction alleles resulting in altered paradigms of cellular memory, including long-term potentiation (LTP) and long-term depression (LTD); and 109 classical neurotransmitter receptor subunits (NT receptors genes) (SI Table 4). For all three lists of genes, we first identified the time point in development where they showed the most significant change. The percentage of genes in any given set that showed a signif4660 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0609071104

Fig. 4. Gene expression of synaptic constituents in neuronal hippocampal cells. (a) Proportion of significantly changing genes (percentage of significant genes) during time in culture (DIV) for three lists of genes relevant to the synapse (see text). Data are expressed as mean ⫾ range. The network electrical activity is depicted in dashed line as percentage of total mean frequency. (b) Ratio between adjacent time points (next time point value/previous time point value) of the number of significantly up-regulated genes (solid line) or mean frequency of firing events (dashed line). (c) Quantitative RT-PCR for activity-dependent genes in cultures incubated in 2 ␮M TTX from 1-DIV. The inhibition of gene expression [% of inhibition (TTX/Control)] is shown for representative mRNAs at four time points. Data were collected from four independent cultures and normalized by GAPDH reactions. GAPDH was not affected by activity, as shown when normalizing the data by using tubulin as endogenous control (data not shown). Data are expressed as mean ⫾ SEM. Asterisks depict significant changes (P ⬍ 0.05, Student’s t test). Other activitydependent genes tested did not show a change in expression (c-Fos, Egr-1, tPA; SI Fig. 12).

icant change at each time point in culture when compared with the values of 0.25- or 1-DIV was calculated (P ⬍ 0.05, Student’s t test; SI Table 4). As seen in Fig. 4a, an abrupt increase in the number of significantly up-regulated genes in all of the lists was Valor et al.

Network Activity Is Not a Key Factor in the Synaptic Transcriptome Profile. The most prominent changes in the expression of synaptic

components occurred between 2- and 8-DIV, during which time the network electrical activity increased by only ⬇5% (Fig. 1), thus the main regulatory events in gene expression preceded the onset of high rates of neural activity. Moreover, the rate of change in neuronal activity showed a peak 4 days after the peak in the up-regulated genes (Fig. 4b). As neuronal activity increased (4- to 16-DIV), the proportion of up-regulated genes remained unchanged (Fig. 4a), consistent with the model that network activity was not involved in the initiation of expression of synapse proteins. To test this model, we measured gene expression in cultures that were chronically treated with 2 ␮M tetrodotoxin (TTX), which prevented all action potential firing (data not shown). Previous reports show that chronic exposure to TTX does not affect dendritic formation and synapse density (33, 34). In addition to studying the effect of activity on synaptic genes, we also examined the expression of well characterized activitydependent genes (35–41) by using quantitative RT-PCR assays (Fig. 4c). The effect of TTX was first visible at 8-DIV in the activity-dependent genes and became more prominent by 16DIV (Arc, BDNF, Cox6a2, Kv1.4–Kcna4, and Narp–Nptx2; Fig. 4c). In contrast, TTX had no effect on the expression of the tested synapse proteins at any time point (CaMKII␣, PSD-95, GluR1, NR2B, or Synaptophysin, among others; SI Fig. 12). Together these results support the conclusion that network activity was not a major force in the synaptogenesis program, but plays a role in subsequent activity-dependent gene regulation. Discussion To our knowledge, this is the first report in which both genomewide transcriptome and network electrical activity profiles have been integrated into a functional genomic approach to study synapse assembly and neural activity. We used two array platforms to monitor both profiles: multielectrode arrays and Affymetrix GeneChips. Although gene expression profiling in hippocampal neurons has been reported (32), we extended the transcriptome analysis further in time to be able to correlate this information with functional data from neural activity. We found evidence that embryonic neurons from the hippocampus utilize a program that results in the coordinated transcription of a diverse set of synapse proteins during development (Fig. 5). This coordinated expression preceded the morphological and electrical features of synapses in neuronal networks. The functional importance of this program was indicated by the large subset of regulated genes that, if mutated, result in impaired LTP and LTD (SI Table 4). The synaptogenesis program also preceded the activity-dependent gene expression program, which is known to involve many of these LTP and LTD genes. The overall network electrical activity occurred in parallel with the morphological changes. However, the gene expression Valor et al.

Fig. 5. Phases of events in hippocampal-cultured neurons. This scheme summarizes the periods of changes in gene expression and network activity during the development of hippocampal cultures. (Upper) Ages (days in culture) at which samples were prepared. (Lower) Periods of maximum values (block arrows) and periods of increase (dashed lines).

profile revealed that changes in transcript levels began to occur before the onset of spiking activity based upon the following observations: (i) the gene expression profile exhibited a switch point ⬇4-DIV, when only sporadic spikes were recorded; (ii) the PCA and hierarchical clustering showed two velocities in the gene expression profile, the faster one occurring between 0.25and 8-DIV and the slower one between 8- and 16-DIV; (iii) the most significant changes in the expression of known synaptic components occurred between 2- and 4-DIV, which affected not only the levels of transcript (measured as signal intensity in SI Fig. 10 or as fold change in SI Fig. 9), but also in the number of genes that were up-regulated (Fig. 4 a and b) when only a very subtle increase in activity was found; and (iv) cultures chronically exposed to TTX (i.e., ‘‘silent cultures’’) exhibited only a significant effect from 8-DIV. These four observations indicate that there was a network activity-independent genetic program that allowed the transcriptional synthesis of the synapse proteome, and this program appeared fully committed ⬇4-DIV. The existence of preformed complexes containing scaffolding proteins in young neurons (5- to 6-DIV) (28) or even earlier (SI Fig. 6A) is in agreement with our observations. It is likely that this general activity-independent program is not restricted to hippocampal cells maintained in vitro, but also exists in the development of the whole brain (42) and in the formation of the neuromuscular junction (43). Despite the absence of a detectable role for network activity in the gene expression events preceding synapse formation, the specific functions of spontaneous neurotransmitter release, specific receptor activation, and calcium transients at the earliest stages remain to be elucidated in our system. Because mutations in essential components of the synaptic vesicle machinery showed formation of normal synaptic connections despite an absence of spontaneous release (21, 42), it seems unlikely that the gene expression events will be triggered by these forms of activity. Other models of neural activity and gene expression (44), including homeostatic plasticity (45), deserve investigation by using the array-based methods reported here. The key signals that initiate or trigger the synaptogenesis gene expression program remain unknown. Apart from other forms of activity as mentioned above, it is possible that there may be a cell-autonomous role for proneural genes involved in the early stages of specifying neural cells (10, 11), or alternatively, extrinsic factors such as ligands or growth factors (including those from glia) may trigger synaptogenesis. Although neuroligins appear to initiate synapse formation in vitro (46), the recent triple knockout mice for neuroligins were still capable of forming synapses in both cultured neurons and intact brain (47). These results support the model that these ligands and factors are involved in PNAS 兩 March 13, 2007 兩 vol. 104 兩 no. 11 兩 4661

NEUROSCIENCE

observed between 2- and 4-DIV and was followed by a plateau after 8-DIV, comprising ⬇15–20% of the genes. This change was not only observed in the number of significantly up-regulated genes, but also in the signal intensity values, where the most prominent change between adjacent time points was between 2and 8-DIV (SI Fig. 10). These data indicate that synaptic proteins in different functional classes are coordinately upregulated (SI Fig. 11). These results were only partially due to the overlapping genes in the lists (data not shown). Not surprisingly, the most significantly down-regulated genes were fewer in number (⬍5% of the genes) and did not show the abrupt change seen in the up-regulated genes (dotted lines in Fig. 4a, SI Fig. 10). These observations are in agreement with the GO analysis, in which the enrichment of synaptic genes was observed in the up-regulated genes (Fig. 3 and SI Tables 1 and 2).

synapse maturation, rather than synapse initiation (1, 2). We therefore reasoned that the coordinated set of genes identified in the present study could provide tools for identifying putative coordinating transcription factors, which may in turn lead to the trigger mechanisms. It is also possible that microRNAs may play a role in regulating the levels of mRNAs. In addition to finding regulators of synaptogenesis, the dataset could be used to identify novel candidate genes encoding synapse proteins. In Drosophila, it has been estimated that ⬇40% of the entire transcriptome is required to orchestrate synaptogenesis at the neuromuscular junction (48). Because we focused on ⬇6% of the transcriptome, there remains considerable potential for discovering novel synaptogenic genes. As the cultures mature, the switch from synaptogenesis to activity-dependent gene expression is a reflection of the function of the synapse–genome signaling pathways, which are themselves a product of the synaptogenesis program. Given that synapse formation is a feature of learning and memory, and thus likely to reutilize the synaptogenesis program, it will be of interest in future studies to examine the interplay of these two transcriptional programs within single neurons in the mature nervous system. Materials and Methods Primary Neuronal Cultures. Hippocampi were dissected from E17.5

C57BL/6 c/c mouse embryos and transferred to papain (10 units/ml) for 22 min at 37°C. All mice were treated in accordance with the U.K. Animals Scientific Procedures Act of 1986, and all procedures were approved through the British Home Office Inspectorate. Cells were manually dispersed in PBS and centrifuged twice at 400 ⫻ g for 3.5 min. The final pellet was resuspended in Neurobasal/B27 0.5 mM Gln, and dissociated cells were plated onto poly(L-lysine) and laminin-coated substrates at a density of 500 cells/mm2. To either introduce exogenous or deplete endogenous trophic factors, media were not replaced during culture. In the chronic blockade of neural activity experiments, 2 ␮M TTX was introduced at 1-DIV. Immunocytochemistry. See SI Methods. Electrophysiology. Cells were plated directly onto coated MEA that

consisted of 8 ⫻ 8 grids of titanium nitride electrodes of 30-␮m diameter and interelectrode spacing of 200 ␮m (Multi Channel Systems, Reutlingen, Germany). Spontaneous action potentials were recorded (30-min epochs at 2- to 3-day intervals) in Mg2⫹-free Locke solution (140 mM NaCl, 3 mM KCl, 15 mM Hepes, 10 mM glucose, 1 mM CaCl2). Cultures were maintained in Mg2⫹containing medium (Neurobasal), but recordings were made in the absence of this ion to allow full activation of NMDA receptors (49). Spikes were detected and analyzed by using MC Rack (Multi Channel Systems) and Neuroexplorer (Nex Technologies, Littleton, MA). To construct the sigmoidal curve, normalized total spike count was fitted to the equation y ⫽ 1 ⫺ 1/(1 ⫹ exp((X⫺X0)/dX)), in which X is time of recording, X0 is the time for half of the maximum number of spikes, and dX is a time constant. In parallel, the number of active electrodes, i.e., the electrodes able to detect at least one action potential per min, was also plotted against time in culture (Fig. 1). In Fig. 4, the percentage of total mean frequency is the average of the normalized spike count expressed as a percentage of the maximum value. RNA Isolation and Microarray Hybridization. Cells were lysed at

different time points (0.25-, 1-, 2-, 4-, 8-, 12-, and 16-DIV) by using 0.5–1 ml of TRIzol (Invitrogen, Carlsbad, CA) per well. RNeasy Mini columns (Qiagen, Valencia, CA) were used to purify the RNA, and an on-column DNase treatment was carried out to remove DNA. During all of the purifications, manufacturers’ instructions were followed. Samples were concentrated with the addition of 1/10 volume of 3 M NaOAc (pH 6) and 3 4662 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0609071104

volumes of 100% EtOH and precipitated at ⫺20°C. Once resuspended in 12 ␮l of RNase-free H2O, an aliquot was used to check the RNA integrity in 1% formaldehyde-agarose gels. For the transcriptome profiling, MG-430 2.0 arrays were used that contain 45,101 probe sets, including controls. Three replicates were assayed by using one array per sample. Briefly, 1.5 ␮g of total RNA was reverse transcribed, labeled, and hybridized by using One-Cycle Labeling Kit instructions (Affymetrix, Santa Clara, CA). Fluidics Station 450 and GCS3000, both from Affymetrix, were used for the washing and scanning steps, respectively. Microarray Data Acquisition and Statistical Analyses. Signal intensity values and detection flags were extracted with GCOS (Affymetrix) by using a target value of 500 in the scaling step. Scaling did not significantly affect the signal intensities for the spike-in controls (coefficient of variation ⫽ 0.142 ⫾ 0.038). A first filter was applied, in which all probe sets that were not detected (as present) in at least one time point (in the three replicates) were removed, leaving 26,259 probe sets (58.2% of the array). Subsequent analyses were performed by using GeneSpring version 7 (Silicon Genetics, Redwood City, CA). Data were normalized by the median per gene and per array, and a second filter was applied by using an ANOVA test (P ⬍ 0.05) without assuming equal variances, adjusted with the Benjamini–Hochberg false discovery rate to minimize the number of false positives: 11,701 probe sets (25.9% of the array) passed the filter. PCA was performed, in which expression data were decomposed into singular values or principal components (50); in other words, these components represent the behavior of all genes in a data set measured as variance of the whole data. The three most important components for a particular sample defined the coordinates on a 3D graph, with each component corresponding to one dimension in space (x, y, z). Hence, the degree of similarity between samples was represented by their spatial location in the graph. The hierarchical analysis was used to classify the samples according to their gene expression profiles by using Spearman correlation. Both analyses produced the best results when working with ANOVA-filtered data (data not shown), and therefore we only worked with these filtered data. Because data followed a normal distribution after median normalization (data not shown), one-tail Student’s t tests were performed to estimate the number of genes in a particular list (core PSD, phenotype, NT receptors, and functional subgroups therein; see text) that were significantly changed during development of the culture. An ANOVA test was not used because we analyzed pairs of genes as independent events. The normalized signal intensity values that passed the presence filter for each time point were compared with 0.25-DIV, and the number of P values ⬍0.05 was counted independently in up- and downregulated genes. We repeated that process, taking the 1-DIV values as reference and averaging both results. To avoid any bias from genes represented by more than one probe set, the values were simply averaged because no unbiased criteria were available to truly discriminate isoforms. For the k-means clustering and GO terms analyses, see SI Methods. Quantitative RT-PCR. Typically, 0.5 ␮g of total RNA was used for

the retrotranscription, following the recommendations of Invitrogen (SuperScript II Reverse Transcriptase). PCRs were set up per sample by using 1 ␮l of cDNA, a concentration of 0.3 ␮M for each primer, and 25 ␮l of 2⫻ Quantitect SYBR Green PCR Master Mix (QIAGEN) to a final volume of 50 ␮l. The sequences of the primers are provided in SI Table 5. Reactions were run out in the 7500 Real-Time PCR System (Applied Biosystems, Foster City, CA) with the following conditions: 1 cycle of 95°C for 15 min; 35–40 cycles of 94°C for 30 sec, 55°C for 30 sec, and 72°C for 50 sec; and 1 cycle of 72°C for 5 min. The resulting amplicons were checked for specificity and size in agarose gels. The data Valor et al.

We thank Mark O. Collins and Alex Howell for the data curation of the lists of genes, and Esperanza Ferna´ndez, Frederic Laumonnier, and Julia Wright for their technical support. This work was supported by grants from Genes to Cognition, The Wellcome Trust. L.M.V. was supported by a Fundacio ´ n Ramo ´ n Areces Fellowship, and L.M.V., P.C., L.H., C.N.G.A., and S.G.N.G. were supported by The Wellcome Trust.

1. Waites CL, Craig AM, Garner CC (2005) Annu Rev Neurosci 28:251–274. 2. Craig AM, Graf ER, Linhoff MW (2006) Trends Neurosci 29:8–20. 3. Collins MO, Husi H, Yu L, Brandon JM, Anderson CN, Blackstock WP, Choudhary JS, Grant SG (2006) J Neurochem, 97(Suppl):16–23. 4. Yang Y, Quitschke WW, Vostrov AA, Brewer GJ (1999) J Neurochem 73:2286–2298. 5. Fukuda T, Itoh M, Ichikawa T, Washiyama K, Goto Y (2005) J Neuropathol Exp Neurol 64:537–544. 6. Nelson ED, Kavalali ET, Monteggia LM (2006) Curr Biol 16:710–716. 7. Yoshida T, Mishina M (2005) J Neurosci 25:3067–3079. 8. Lee JE (1997) Curr Opin Neurobiol 7:13–20. 9. Wegner M (1999) Nucleic Acids Res 27:1409–1420. 10. Guillemot F, Molnar Z, Tarabykin V, Stoykova A (2006) Eur J Neurosci 23:857–868. 11. Hevner RF, Hodge RD, Daza RA, Englund C (2006) Neurosci Res 55:223–233. 12. Aizawa H, Hu SC, Bobb K, Balakrishnan K, Ince G, Gurevich I, Cowan M, Ghosh A (2004) Science 303:197–202. 13. Gaudilliere B, Konishi Y, de la Iglesia N, Yao G, Bonni A (2004) Neuron 41:229–241. 14. Cobos I, Calcagnotto ME, Vilaythong AJ, Thwin MT, Noebels JL, Baraban SC, Rubenstein JL (2005) Nat Neurosci 8:1059–1068. 15. Hand R, Bortone D, Mattar P, Nguyen L, Heng JI, Guerrier S, Boutt E, Peters E, Barnes AP, Parras C, et al. (2005) Neuron 48:45–62. 16. Flavell SW, Cowan CW, Kim TK, Greer PL, Lin Y, Paradis S, Griffith EC, Hu LS, Chen C, Greenberg ME (2006) Science 311:1008–1012. 17. Parrish JZ, Kim MD, Jan LY, Jan YN (2006) Genes Dev 20:820–835. 18. Kandel ER (2001) Science 294:1030–1038. 19. West AE, Griffith EC, Greenberg ME (2002) Nat Rev Neurosci 3:921–931. 20. Rao A, Craig AM (1997) Neuron 19:801–812. 21. Varoqueaux F, Sigler A, Rhee JS, Brose N, Enk C, Reim K, Rosenmund C (2002) Proc Natl Acad Sci USA 99:9037–9042. 22. Fletcher TL, Banker GA (1989) Dev Biol 136:446–454. 23. Siebler M, Koller H, Stichel CC, Muller HW, Freund HJ (1993) Synapse 14:206–213. 24. Arnold FJ, Hofmann F, Bengtson CP, Wittmann M, Vanhoutte P, Bading H (2005) J Physiol 564:3–19. 25. Cardozo AK, Berthou L, Kruhoffer M, Orntoft T, Nicolls MR, Eizirik DL (2003) J Proteome Res 2:553–555. 26. Mijalski T, Harder A, Halder T, Kersten M, Horsch M, Strom TM, Liebscher HV, Lottspeich F, de Angelis MH, Beckers J (2005) Proc Natl Acad Sci USA 102:8621–8626.

27. Kislinger T, Cox B, Kannan A, Chung C, Hu P, Ignatchenko A, Scott MS, Gramolini AO, Morris Q, Hallett MT, et al. (2006) Cell 125:173–186. 28. Gerrow K, Romorini S, Nabi SM, Colicos MA, Sala C, El-Husseini A (2006) Neuron 49:547–562. 29. Papa M, Bundman MC, Greenberger V, Segal M (1995) J Neurosci 15:1–11. 30. Fletcher TL, De Camilli P, Banker G (1994) J Neurosci 14:6695–6706. 31. Mody M, Cao Y, Cui Z, Tay KY, Shyong A, Shimizu E, Pham K, Schultz P, Welsh D, Tsien JZ (2001) Proc Natl Acad Sci USA 98:8862–8867. 32. Dabrowski M, Aerts S, Van Hummelen P, Craessaerts K, De Moor B, Annaert W, Moreau Y, De Strooper B (2003) J Neurochem 85:1279–1288. 33. Kossel AH, Williams CV, Schweizer M, Kater SB (1997) J Neurosci 17:6314– 6324. 34. Harms KJ, Craig AM (2005) J Comp Neurol 490:72–84. 35. Rao VR, Pintchovski SA, Chin J, Peebles CL, Mitra S, Finkbeiner S (2006) Nat Neurosci 9:887–895. 36. Lein ES, Shatz CJ (2000) J Neurosci 20:1470–1483. 37. Wong-Riley MT, Huang Z, Liebl W, Nie F, Xu H, Zhang C (1998) Vision Res 38:1455–1477. 38. Platenik J, Kuramoto N, Yoneda Y (2000) Life Sci 67:335–364. 39. Galvan CD, Wenzel JH, Dineley KT, Lam TT, Schwartzkroin PA, Sweatt JD, Swann JW (2003) Eur J Neurosci 18:1861–1872. 40. Tsui CC, Copeland NG, Gilbert DJ, Jenkins NA, Barnes C, Worley PF (1996) J Neurosci 16:2463–2478. 41. Baranes D, Lederfein D, Huang YY, Chen M, Bailey CH, Kandel ER (1998) Neuron 21:813–825. 42. Verhage M, Maia AS, Plomp JJ, Brussaard AB, Heeroma JH, Vermeer H, Toonen RF, Hammer RE, van den Berg TK, Missler M, et al. (2000) Science 287:864–869. 43. Kummer TT, Misgeld T, Sanes JR (2006) Curr Opin Neurobiol 16:74–82. 44. Spitzer NC (2006) Nature 444:707–712. 45. Turrigiano GG, Nelson SB (2004) Nat Rev Neurosci 5:97–107. 46. Scheiffele P, Fan J, Choih J, Fetter R, Serafini T (2000) Cell 101:657–669. 47. Varoqueaux F, Aramuni G, Rawson RL, Mohrmann R, Missler M, Gottmann K, Zhang W, Sudhof TC, Brose N (2006) Neuron 51:741–754. 48. Liebl FL, Werner KM, Sheng Q, Karr JE, McCabe BD, Featherstone DE (2006) J Neurobiol 66:332–347. 49. Mangan PS, Kapur J (2004) J Neurophysiol 91:946–957. 50. Sherlock G (2001) Brief Bioinform 2:350–362.

Valor et al.

PNAS 兩 March 13, 2007 兩 vol. 104 兩 no. 11 兩 4663

NEUROSCIENCE

generated were analyzed in the 7500 System SDS version 1.2.2 software, which calculated the cycle threshold (CT) and the fold change (⫽ 2⫺⌬⌬CT). GAPDH, tubulin, and Atf2 were used as endogenous controls. Reactions for each pair of primers were run at least in triplicates. To analyze statistically the quantitative PCR results, Student’s t test was used.