We have used two complementary microarray-based strategies to obtain comprehensive gene expression profiles of developing C. elegans neurons. In the MAPCeL method, GFP-labeled embryonic neurons were isolated by FACS for microarray profiling . Because postembryonic neurons are not readily available for sorting , we used an alternative strategy, the mRNA-tagging method, to profile the larval nervous system . In this approach, neuronal mRNAs were purified by immunoprecipitation from transgenic animals expressing an epitope-tagged RNA binding protein (FLAG-PAB-1) in larval neurons. Together, these microarray datasets identify 2,488 transcripts that show elevated expression in the C. elegans nervous system relative to other tissues in at least one developmental stage (that is, embryonic or larval) (Additional data file 10). A bioinformatic query of WormBase confirmed enrichment of known neural transcripts in these datasets (Figure 3a). In addition, analysis of a representative group of newly constructed GFP reporters has confirmed in vivo neural expression of >90% of previously uncharacterized genes on these lists (Table 1). We therefore conclude that these 'panneural' profiles provide accurate representations of gene expression in the C. elegans embryonic and larval nervous systems. These transcripts encode proteins with a broad array of functions. For example, as expected, ion channels, neurotransmitter receptors and synaptic vesicle components are highly represented (Figure 7; Table 2; Additional data file 4). In a striking indication of the complex signaling capacity of the C. elegans nervous system, most of the known peptide neurotransmitter genes (for example, 20 of 23 FMRFamide genes or 'flps') are enriched in the larval pan-neural dataset (Figure 4; Additional data file 4) . Neural functions for previously uncharacterized members of these gene families can now be assigned by genetic or RNAi analysis. With this possibility in mind, we tested the applicability of these expression data for predicting in vivo functions for genes in this dataset that are also included in a genome-wide interaction map or 'interactome' for C. elegans proteins . This analysis revealed that proteins encoded by a subset of panneural transcripts are linked to identified components of the synaptic vesicle cycle and, therefore, predicts that genetic or RNAi perturbation of these genes should result in neurotransmitter signaling defects (Figure 10). In addition to finding transcripts that may have shared roles in both the embryonic and larval nervous system, these pan-neural profiles have also identified a significant number of genes (71%, 1,777/2,488) that are differentially enriched in either embryonic or larval neurons. In the future, it will be interesting to determine if these genes define stage-specific features of the developing nervous system.
The mRNA-tagging method can be used to generate gene expression profiles of specific neurons
In addition to detecting transcripts that are broadly expressed throughout the nervous system (that is, synaptic vesicle components), the pan-neural profiles also include genes that are selectively expressed in specific neurons. In most instances, these known assignments are based on promoter-GFP reporter constructs for a limited number of genes in a given neuron and are, therefore, incomplete. To test the applicability of the mRNA-tagging strategy for obtaining a comprehensive gene expression profile of a specific subset of neurons, we utilized this approach to fingerprint a group of 18 larval cells largely composed of A-type motor neurons [35,75]. This experiment revealed >400 transcripts with enriched expression in these cells (Additional data file 1). Although the majority (70%) of these transcripts also show elevated expression in the larval pan-neural profile (Figure 8), a significant fraction of these mRNAs are exclusively enriched in the A-class dataset in this comparison and are, therefore, likely to represent genes with limited expression in the nervous system. These results indicate that the mRNA-tagging strategy can now be applied to monitor gene expression in specific C. elegans neurons and that this approach should detect neuron-specific genes with potential key roles in the specification or function of individual neuron types. Our findings confirm an earlier study in which a neuron specific promoter was used in conjunction with the mRNA-tagging strategy to identify transcripts that are highly expressed in a group of approximately 50 sensory neurons from C. elegans . Our work provides the important technical advance, however, of substantially enhancing the sensitivity of this method; we show that reliable profiles can be obtained by amplifying nanogram quantities of mRNA whereas the method of Kunitomo et al.  required micrograms of starting mRNA.
Limitations of the mRNA tagging method
Despite the successful use of mRNA-tagging for these cell-specific profiling experiments, additional improvements in this method would be helpful. For example, with any given promoter, we sometimes observe FLAG-1::PAB-1 staining in the expected cell types as well as in additional ectopic locations (data not shown). This problem is unlikely to result from gene expression domains in the transgenic PAB-1 construct because the substitution of pab-1 cDNA to remove all possible genomic PAB-1 regulatory sites did not rectify this problem (Von Stetina et al., unpublished data). Our solution has been to generate multiple transgenic lines for each construct until we obtain at least one line in which FLAG-PAB-1 expression is limited to the cells of choice. A second problem with this method is pull-down of non-specific mRNA bound to the anti-FLAG sepharose beads. We have reduced this background by including a stringent wash step with a low salt buffer, but additional treatments to remove this extraneous mRNA would enhance the sensitivity of this method (see Materials and methods). Lastly, some promoters result in subviable transgenic lines or unpredictable genetic interactions that limit profiling experiments  (data not shown). The biological mechanisms of these effects are unknown but have also been observed for PAB-1 mRNA-tagging lines in Drosophila .
Applications of cell-specific microarray profiling methods
The mRNA-tagging strategy has been used to generate robust gene expression profiles of major C. elegans tissues (that is, muscles, intestine, nervous system) [11,13] (this paper). By exploiting promoter elements with more limited expression, it has also been possible to extend this approach to specific subsets of neurons. These results suggest that mRNAtagging can now be exploited to obtain gene expression profiles in a broad array of cell types at precisely defined developmental intervals. For example, mRNA-tagging profiles obtained during a critical larval period in which GABAergic motor neurons switch axonal versus dendritic polarity could potentially reveal genes that direct the remodeling process . The combined profiling results reported in this paper identify a set of 177 transcription factors showing enriched expression in neurons. Genetic analysis has established that many of these transcription factors regulate key aspects of neuronal differentiation and function [31,47,55-57,76,99,100]. Both the MAPCeL and mRNA-tagging approaches can now be utilized to generate comparisons of mutant versus wild-type profiles that should reveal transcription factor-regulated genes in specific neurons [9,37]. Microarray profiling of mutants for other classes of proteins could also be utilized to reveal unexpected gene regulatory roles. For example, a comparison of pan-neural mRNA-tagging datasets obtained from mutant versus wild-type animals indicates that the conserved synaptic protein RPM-1/Highwire regulates gene expression throughout C. elegans nervous system (JDW, SEV, DMM, unpublished results). The C. elegans nervous system is uniquely well-defined with a wiring diagram denoting chemical synapses and gap junctions among all 302 neurons. It should now be possible to exploit these cell-specific microarray profiling methods to define genes expressed in each type of neuron in this circuit. In turn, novel computational methods could be exploited to link specific subsets of these genes to roles in defining the connectivity architecture of this network [101,102].
Towards defining the transcriptome
In addition to transcripts showing elevated expression in neurons, our neural microarray profiles include a larger group of transcripts that are expressed in neurons and in other tissues at comparable levels. We refer to these transcripts as 'expressed genes'. A comparison of the three larval datasets described in this work (reference, larval pan-neural, larval A-class motor neuron) reveals that 1,424 EGs are shared and are, therefore, likely to represent transcripts that function in a broad array of cell types. In contrast, a smaller number of transcripts are uniquely detected in either the larval pan-neural (1,189) or larval A-class motor neuron (435) datasets. The three embryonic datasets (reference, embryonic pan-neural, embryonic A-class motor neuron) commonly express 4,995 EGs, with 280 EGs unique to embryonic A-class motor neurons and 480 mRNAs selectively detected in the embryonic pan-neural profile. These findings suggest that microarray-based strategies to confirm in vivo expression of all predicted C. elegans genes or to identify new, previously unknown transcripts (for example, tiling array profiles) , will require extraction of mRNA from a variety of specific cells and tissues with methods similar to those described here.