Neuron-Specific Feeding RNAi in C. elegans and Its Use in a Screen for Essential Genes Required for GABA Neuron Function

Forward genetic screens are important tools for exploring the genetic requirements for neuronal function. However, conventional forward screens often have difficulty identifying genes whose relevant functions are masked by pleiotropy. In particular, if loss of gene function results in sterility, lethality, or other severe pleiotropy, neuronal-specific functions cannot be readily analyzed. Here we describe a method in C. elegans for generating cell-specific knockdown in neurons using feeding RNAi and its application in a screen for the role of essential genes in GABAergic neurons. We combine manipulations that increase the sensitivity of select neurons to RNAi with manipulations that block RNAi in other cells. We produce animal strains in which feeding RNAi results in restricted gene knockdown in either GABA-, acetylcholine-, dopamine-, or glutamate-releasing neurons. In these strains, we observe neuron cell-type specific behavioral changes when we knock down genes required for these neurons to function, including genes encoding the basal neurotransmission machinery. These reagents enable high-throughput, cell-specific knockdown in the nervous system, facilitating rapid dissection of the site of gene action and screening for neuronal functions of essential genes. Using the GABA-specific RNAi strain, we screened 1,320 RNAi clones targeting essential genes on chromosomes I, II, and III for their effect on GABA neuron function. We identified 48 genes whose GABA cell-specific knockdown resulted in reduced GABA motor output. This screen extends our understanding of the genetic requirements for continued neuronal function in a mature organism.


Introduction
In C. elegans, there are two basic ways to generate mosaic gene expression: knocking gene function down in specific cells of an otherwise normal animal; or rescuing wild type gene function in a mutant animal. Examples of the first method include triggering local RNAi by targeted expression of hairpin or double-stranded RNA [1,2]; examples of the second include the use of unstable DNA elements or the targeted expression of wildtype coding sequences [3]. However, all of these methods require the construction of a new transgenic animal for each gene of interest. The requirement for one strain per gene limits the usefulness of these techniques for questions involving many genes, because it is impractical to construct so many transgenic animals.
Gene knockdown in C. elegans can be induced by feeding RNAi [4], and the development of whole-genome feeding RNAi libraries means that RNAi can be used for large-scale genetic analysis, up to and including whole-genome screens [5][6][7]. Moreover, many of the cellular mechanisms that mediate feeding RNAi have been described. In particular, interfering RNA species enter cells using the dsRNA channel SID-1 [8][9][10], while within each cell the Argonaute protein RDE-1 is required to achieve gene knockdown [11]. This molecular understanding has been used to develop a new method for mosaic gene expression that is generated by feeding RNAi, and thus is compatible with the study of many genes. For example, a muscle-specific rde-1 mosaic enables musclespecific knockdown in response to feeding RNAi [12]. Similarly, manipulating sid-1 expression can alter the response of touch neurons to feeding RNAi [13].
Neurons, however, present particular problems for feeding RNAi. Most C. elegans neurons are resistant to feeding RNAi [14][15][16]. Genetic backgrounds have been developed that enhance the sensitivity of neurons to feeding RNAi, such as the lin-15B; eri-1 mutant [17] and neuronal expression of sid-1 [13]. However, these same genetic backgrounds can also result in increased transgene silencing in the nervous system [17][18][19]. Such transgene silencing could limit expression of the transgenes driving mosaic rescue of RNAi, even while these mutations enhance RNAi sensitivity. Thus, an approach that allows feeding RNAi to generate tissuespecific gene knockdown in neurons that can be generalized to a variety of neuronal subtypes is not currently available. Here, we describe a strategy in C. elegans that allows feeding RNAi to generate cell-specific knockdown in a wide variety of neuronal subtypes. We use this method to examine the genetic requirements of mature GABA motor neurons.

An approach to neuron-specific feeding RNAi
We chose to restrict RNAi sensitivity to selected neurons using rde-1 mosaic animals [12]. At the same time, we also sought to increase RNAi sensitivity in the selected neurons, since many neurons are resistant to RNAi. We used two complementary techniques to increase neuronal sensitivity. First, we used a genetic background (lin-15B; eri-1) that enhances the sensitivity of all neurons to RNAi [17]. Second, we overexpressed the doublestranded RNA transporter sid-1 only in the selected neurons [13]. Thus, our strain carries three background mutations (lin-15B; eri-1; rde-1) and expresses two transgenes in select neurons (rde-1(+) and sid-1(+)) ( Fig. 1A). We initially found that our approach was subject to significant transgene silencing effects, likely caused by the combination of the lin-15B; eri-1 background and rde-1 overexpression (see Fig. 2A) [17,18]. To avoid transgene silencing, we combined the rde-1(+) and sid-1(+) rescue fragments into a single transcriptional unit, separated by an SL2-specific trans-splice site [20]. This artificial operon was placed in a MosSCI-compatible [21] MultiSite Gateway vector for easy manipulation, single-copy integration, and expression under cell-specific promoters.

GABAergic neuron-specific RNAi
We first targeted GABA-releasing neurons. We used Gateway recombination to place the rde-1(+); sid-1(+) operon behind the Punc-47 promoter, which drives expression exclusively in the GABA motor neurons (Fig. 1B) [22]. This construct was inserted into the genome as a single copy using the MosSCI technique [21], and this transgene was crossed into the lin-15B; eri-1; rde-1 mutant background to generate a strain of animals in which the interfering response of exogenous dsRNA is limited to GABA neurons.
To determine the effectiveness of our approach, we expressed nuclear-localized GFP in all somatic cells of our GABA-specific RNAi strain and also marked the GABA neurons with mCherry. We fed these animals RNAi against GFP and found that, compared to control RNAi, GFP was efficiently knocked down in GABA neurons but was still present in other cells -including muscles, intestine, skin, and non-GABA neurons ( Fig. 2A, B). This suggests that the RNAi response is limited to the tissue in which our artificial operon is expressed and does not spread to other adjacent cells or tissues.
We also tested the viability of our strain when challenged with RNAi against an essential gene. ama-1 encodes the large subunit of C. elegans RNA polymerase II [23], and RNAi against ama-1 results in a larval arrest phenotype with penetrance approaching 100%. By contrast, the GABA-specific strain, and the others described below, were completely resistant to ama-1 RNAi-induced arrest, demonstrating that gene function can be studied in the GABA neurons of an otherwise normal animal even when these genes have essential functions in other tissues.
Next, we sought to determine the effectiveness of our approach against endogenous, single-gene targets. We took advantage of the robust and specific behavioral 'shrinker' phenotype generated by lack of either pre-or post-synaptic components of GABA neurotransmission. The GABA-specific shrinker phenotype is readily distinguished from the paralyzed phenotype caused by loss of basal neurotransmission components. We compared the effect of feeding RNAi between a standard neuron-sensitized strain (lin-15B; eri-1) and our GABA-specific strain. We performed RNAi against two GABA-specific genes: unc-25, which encodes glutamic acid decarboxylase [24] and is required in GABA neurons for GABA neurotransmission; and unc-49, which encodes the GABA A receptor [25] and is required in muscles for GABA neurotransmission. We also targeted two components of basal neurotransmission, both of which are required in all neurons for synaptic vesicle release: unc-13, which encodes UNC-13 [26], and snb-1, which encodes synaptobrevin [27]. We found that, as expected, knockdown of GABA genes in the neuron-sensitized strain resulted in a GABA-specific shrinker phenotype, while knockdown of basal neurotransmission genes resulted in an uncoordinated phenotype. By contrast, in our GABA-specific strain, knockdown of the neuronal GABA gene unc-25 resulted in a shrinker phenotype, while knockdown of the muscle GABA gene unc-49 had no effect. Further, in our GABA-specific strain, knockdown of basal neurotransmission genes also resulted in a shrinker phenotype (Fig. 2C). To quantify behavioral changes due to GABA-specific knockdown, we utilized an aldicarb-sensitivity assay to indirectly measure GABA output. Aldicarb is an acetylcholinesterase inhibitor that causes acute paralysis due to accumulation of acetylcholine at neuromuscular junctions (NMJs). Loss of inhibitory GABA input leads to hypersensitivity to aldicarb, causing more rapid paralysis [28]. As expected, GABA-specific knockdown of unc-25 as well as the basal neurotransmission genes unc-13 and snb-1 led to hypersensitivity to aldicarb, while RNAi against unc-49 had no effect (Fig. 2D).
In addition to synaptic genes, we sought to target a broadlyexpressed gene involved in maintenance of the nervous system. unc-70 encodes b-spectrin, a component of the plasma membrane skeleton that is expressed in all cells [29]. Animals lacking unc-70 generate spontaneous breaks in their neurons as a result of mechanical stress [30]. Neuron-sensitized (lin-15B; eri-1) animals fed dsRNA against unc-70 are slow to grow, dumpy, and paralyzed. Knockdown of unc-70 in the GABA-specific strain, however, caused an aldicarb hypersensitivity phenotype (Fig. 2D), with no other obvious phenotypes. When the GABA neurons of these worms were examined, we observed defects consistent with lack of unc-70, including branched processes, broken axons -some of which terminated in regenerative growth cones -as well as substantial degeneration of the dorsal nerve cord, especially where disconnection of distal fragments was apparent (Fig. 2E). Together, these data demonstrate that our approach allows knockdown of endogenous genes within selected neurons, while preventing knockdown of those genes in other cells. Moreover, this method

Author Summary
Living organisms often reuse the same genes multiple times for different purposes. If one function of a gene is essential, death or arrest of the mutant masks other functions. Understanding the functions of essential genes is particularly critical in the nervous system, which must maintain plasticity and fend off disease long after development is complete. However, current strategies for generating conditional knockouts rely on making a new transgenic animal for each gene and thus are not useful for forward genetic screens or for other experiments involving a large number of genes. We have developed a technique in C. elegans for generating gene knockdown in selected neuron sub-types in response to feeding RNAi. Using this technique, we performed a screen aimed at identifying essential genes that are required for the function of mature GABAergic neurons. By knocking these genes down in only GABAergic neurons, we can circumvent the muddying effects of pleiotropy and find essential genes that function cell intrinsically to promote GABA neuron function. The genes we identified using this method provide a more complete understanding of the complex genetic requirements of post-developmental neurons.
allows for dissection of the site of gene action, easily separating preand post-synaptic functions, as well as neuron sub-type-specific effects of gene knockdown.

Other neuron sub-types
To determine whether our system for controlling feeding RNAi was adaptable to other sets of neurons, we used other promoters to drive expression of our artificial rde-1(+); sid-1(+) operon, made single-copy MosSCI integrations of each construct, and placed each resulting MosSCI transgene in the lin-15B; eri-1; rde-1 mutant background. Next, we characterized the resulting strains by challenging them with ama-1 RNAi. Three of these new strainsthose using the Pdat-1, Punc-17, and Peat-4 promoters -satisfied the test of ama-1 RNAi resistance in this context, and further experiments with these three strains are discussed below. However, with two other promoters -Prab-3 and Pmig-13 -we found that the resulting strain was not resistant to ama-1 RNAi. Thus, these promoters are not suitable for the analysis of essential genes using our system. By contrast, the Pdat-1, Punc-17, and Peat-4 promoters appear to be tightly controlled, suggesting that the three strains using these promoters can be used for neuron-specific RNAi.

Dopaminergic neuron-specific RNAi
Pdat-1 drives expression in the dopaminergic neurons [31], which comprise eight cells in adult hermaphrodites [32]. One function of dopamine release from these cells is to control a behavioral response to food called ''basal slowing,'' in which animals slow their rate of locomotion when they encounter a bacterial lawn [33]. We used a basal slowing assay to evaluate the ability of our Pdat-1 strain to restrict RNAi to the dopaminergic neurons. The cat-2 gene encodes tyrosine hydroxylase and is required for dopamine synthesis [34]. Mutant animals that lack cat-2, or animals in which all eight dopamine neurons have been ablated, do not exhibit basal slowing [33]. Basal slowing response is also absent in mutants that lack dop-3, which encodes a D2 dopamine receptor and is not expressed in the dopaminergic neurons [35]. Thus, basal slowing requires factors both intrinsic and extrinsic to the dopamine neurons. We found that both a standard sensitized RNAi strain (lin15B; eri-1) and the Pdat-1 strain exhibited a robust basal slowing response on control RNAi (Fig. 3A, B). Further, the basal slowing response was completely blocked in both strains on RNAi against cat-2, which is required in the dopamine neurons themselves. By contrast, RNAi against dop-3 blocked basal slowing in the control strain but did not affect basal slowing in the Pdat-1 strain. We also tested basal slowing following feeding RNAi directed against unc-13 or snb-1. In these experiments, the standard lin-15B; eri-1 strain could not be tested because knockdown of these genes resulted in an uncoordinated behavioral phenotype (Fig. 2C). By contrast, knockdown of unc-13 and snb-1 in the Pdat-1 strain eliminated the basal slowing response (Fig. 3B). Further, although the basal slowing response was eliminated by knockdown of unc-13 or snb-1, these knockdowns did not affect the rate of locomotion (p = 0.0816 and p = 0.5566 respectively, compared to control off food). These data demonstrate that feeding RNAi in the Pdat-1 strain is effective in dopamine neurons but is blocked in other cells.

Glutamatergic neuron-specific RNAi
Next, we targeted glutamatergic neurons by driving expression of our artificial operon under the Peat-4 promoter. In C. elegans, one behavior mediated by glutamatergic neurotransmission is reversal in response to nose touch [36]. Glutamatergic control of this behavior requires the gene eat-4, which encodes VGLUT, the glutamate synaptic vesicle transporter [37] that functions intrinsically in glutamatergic neurons. Glutamate control of the nose touch response also requires the gene glr-1, which encodes an AMPA-type ionotropic glutamate receptor [38,39] and is required in the post-synaptic cells that respond to glutamate. We fed neuron-sensitized (lin15B; eri-1) animals dsRNA against eat-4 and glr-1. Under both conditions, these animals were deficient in their  RNAi strain expressing ubiquitous, nuclear-localized GFP and GABA mCherry after control or GFP RNAi feeding. Arrows mark GABA neurons with nuclear GFP. Stars mark GABA neuron cell bodies. GFP RNAi eliminates GFP in GABA cells, while GFP remains in other cell types. In control RNAi animals, GFP in GABA neurons is reduced relative to other cells due to increased transgene silencing in these neurons (see text). Scale bars = 5 mm. (B) Quantification of the percent of GABA cell bodies with nuclear-localized GFP under control or GFP RNAi conditions. Error bars are SEM, n = 10 worms, *** p,0.0001 (C) Locomotion behaviors in response to RNAi against genes known to ability to respond to nose touch when compared to controls (Fig. 3C). Thus, feeding RNAi can target both pre-and postsynaptic components of glutamatergic neurotransmission to generate a glutamate-specific behavioral defect. We then tested the glutamatergic neuron-specific RNAi strain. On control RNAi, these animals exhibited a normal nose touch response, similar to the standard sensitized strain ( Fig. 3C and D, gray bars; p = 0.3421). Feeding RNAi against eat-4 resulted in a loss of nose touch response compared to empty vector fed controls, similar to the loss in the standard strain (Fig. 3D). Thus, RNAi of an endogenous gene in glutamatergic neurons is effective in the glutamate-specific strain. By contrast, RNAi against glr-1 did not affect the ability of the glutamate-specific strain to respond to nose touch. This result suggests that unlike the standard sensitized strain, the glutamate-specific strain is insensitive to RNAi outside the glutamate neurons. In support of this, we found that feeding RNAi against the basal neurotransmission genes unc-13 and snb-1 resulted in a glutamate-specific behavioral defect in nose touch, rather than a general defect in movement. In addition, glutamate-specific RNAi worms fed dsRNA against unc-70 were also impaired in their response to nose touch when compared to controls (Fig. 3D), suggesting that unc-70 is important for maintaining the integrity of glutamatereleasing neurons.

Cholinergic neuron-specific RNAi
We also targeted cholinergic neurons using the Punc-17 promoter, which drives expression in acetylcholine-releasing neurons, including the excitatory motor neurons that innervate the body wall muscles and are required for proper locomotion [40,41]. We first measured the thrashing rate in liquid of neuronsensitized (lin-15B; eri-1) worms and found that they exhibited a significant decrease when fed bacteria producing dsRNA against snb-1 or unc-13 compared to empty vector control (Fig. 2E). We then knocked these genes down in the acetylcholine-specific RNAi strain and observed a similar decrease in the rate of thrashing (Fig. 2F). We also fed the acetylcholine-specific RNAi worms dsRNA against unc-70. These worms were impaired in their thrashing rate but displayed no other phenotypes indicative of systemic RNAi of unc-70, suggesting that RNAi in this strain is restricted to the acetylcholine neurons.
A screen for essential gene function in mature GABA neurons Neurons are complex cells, and synaptic transmission and maintenance of normal neuronal function requires the concerted action of a large number of genes. Although many such genes have been identified by forward genetic screens, we hypothesized that these screens may have missed important requirements in neurons for essential genes. Essential genes -those required for the growth of an organism to a fertile adult -are difficult to recover in screens for neuronal function because of death, arrest, or sterility of the mutant. Yet such genes might have critical roles in neurons. Accordingly, we sought to expand our ubderstanding of the genetic requirements for proper GABA neuron function by screening our GABA-specific RNAi strain against a large number of essential genes.
We began by curating a list of all essential genes by reported RNAi phenotype of lethal, arrested, or sterile using WormMart (wormbase.org), and we arrayed corresponding Ahringer RNAi [6] clones into a custom essential gene RNAi library. Our primary screen consisted of 1,782 essential RNAi clones from chromosomes I, II, and III. Using the GABA-specific RNAi strain described above, we screened animals fed these clones for hypersensitivity to aldicarb-induced paralysis. Due to the strict experimental control needed for proper neuronal RNAi and phentoyping, such as log-phase culture and age-matched progeny, we were successfully able to screen 1,320 clones (outlined in Table  S1) targeting essential genes for their effect on GABA output. From the primary screen, we identified 79 clones (,6%) that produced aldicarb hypersensitivity of at least two standard deviations above the mean ( Figure 4A).
We also examined the morphology of the GABA nervous system in each experiment for branching, degeneration, or cell death. We found that in contrast to the functional defects, we observed no morphological defects in any of the 1,320 RNAi experiments. Thus, the functional deficits we observe in response to RNAi are the result of altered neurotransmission rather than cell death or degeneration. However, our screen was conducted on young adult animals, so it is possible that longer-term knockdown would result in morphological phenotypes for some genes.
To validate the primary screen hits, we retested each of the clones in at least three independent trials. We selected those that retested above the cutoff from the primary screen, which was well above that needed for statistical significance (p,0.0001 for all selected clones compared to control). These clones were then sequenced to confirm the targeted gene. We discarded any clone that could not be mapped to a single gene target, including clones that mapped to more than one gene, intergenic region, or intron. We identified 48 genes ( Figure 4C, Table 1) whose knockdown in GABA neurons led to aldicarb hypersensitivity, and thus decreased GABA motor output. Eighty-three percent (40) of these genes have predicted human orthologs [42], suggesting that we have identified a largely conserved set of genes that are important for postdevelopmental GABA neuron function.

Discussion
The technique and strains presented here enable cell-specific knockdown in designated neurons simply by performing feeding RNAi. Our results in GABA, dopamine, glutamate, and acetylcholine-releasing neurons suggest that the technique can be used to limit feeding RNAi to any neuron or group of neurons. However, since RDE-1 acts at a rate-limiting step early in the exogenous RNAi pathway, small amounts of misexpression can trigger an amplified interference response -hence, tightly controlled promoters are required to drive expression of the transgene to ensure specificity.
One major use of this technique will be to rapidly determine the site of action of particular genes. For example, our results demonstrate that the dopamine receptor dop-3 does not function in the dopamine neurons, and the glutamate receptor glr-1 does not function in the glutamate neurons.
Another major use will be to easily determine the function in specific neurons of genes that are ubiquitously expressed. For function in GABA neurotransmission, in animals sensitive to RNAi in all cells (lin-15B; eri-1) and in the GABA-specific strain (XE1375). In GABA-specific RNAi strain, a GABA-specific shrinker phenotype is observed when pre-but not post-synaptic GABA signaling genes are targeted. (D) Percent of paralyzed animals after 100 min of acute exposure to 750 mM aldicarb. Error bars are SEM, n = 3 trials of ,25 animals each, ** p = 0.0023, *** p#0.0003. (E) GABA commissures after control or unc-70 RNAi. In unc-70 RNAi, the axon has broken and initiated a regenerative growth cone (arrow), and the dorsal nerve cord (DNC) has degenerated. Scale bars = 10 mm. doi:10.1371/journal.pgen.1003921.g002 Finally, our technique enables new kinds of forward genetic screens, as we have demonstrated with our GABA-specific RNAi screen of essential genes. Essential genes make up a substantial portion of the genes in the genome, but are virtually inaccessible to traditional genetic screens. These genes, however, are some of the most conserved -while only ,38% of all C. elegans genes have human orthologs [42], ,76% of the genes we selected for their essential RNAi phenotype have predicted orthologs. By using a strain that limits RNAi to non-essential neurons (such as our GABA-specific strain), it is now possible to screen for neuronal functions of genes that normally have lethal, sterile, or other pleiotropic phenotypes.
We have identified 48 genes whose cell-specific knockdown lead to deficits in GABA neurotransmission. As expected, we found components of essential cellular processes such as energy metabolism, transcription, and translation. Additionally, components of several important and conserved signaling and gene regulation pathways were identified, such as miRNA (drsh-1), Wnt signaling (mig-5), and Hpo signaling (wts-1). These pathways have been studied extensively for their role in development, but our glutamate neurotransmission (eat-4, unc-13, snb-1) as well as the structural protein unc-70 inhibit the nose touch response, while RNAi of a postsynaptic component (glr-1) does not. Error bars are 95% confidence interval, n = 5 trials/animal, 10 animals, *** p#0.0006. (E) In standard neuronsensitive strain, RNAi of presynaptic components of general neurotransmission (unc-13 and snb-1) slows rate of thrashing. Error bars are SEM, n = 10 worms, *** p,0.0001 (F) In acetylcholine-specific strain, RNAi of presynaptic components of general neurotransmission (unc-13 and snb-1) as well as unc-70 slows rate of thrashing. Error bars are SEM, n = 10 worms, *** p,0.0001. doi:10.1371/journal.pgen.1003921.g003  data suggest that these pathways may be important postdevelopmentally for maintenance of GABA neuron function.
The genes identified in this study provide a more complete understanding of the complex genetic requirements of postdevelopmental neurons. Additional studies will be required to determine the mechanism through which these genes act to promote GABA neuron function, whether through specific modulation of neuronal functions such as neurotransmitter release, or general cellular health and metabolism. The four strains presented here enable the rapid knockdown of many single gene targets in a given neuron sub-type. The efficiency of RNAi in each strain varies, possibly due to differences in expression levels of our bi-cistronic transgene when driven by various promoters. In the case of the GABAergic neuron-specific strain, we are able to recapitulate the null phenotype of unc-25(e156) with unc-25 RNAi ( Figure S1A, p = 0.0994). The efficiency of the dopaminergic-specific RNAi strain is comparable to that of the standard sensitized RNAi strain -when fed RNAi against cat-2 ( Figure 3A, B), basal slowing response is abolished to a similar level in both strains. The efficiency of RNAi in the glutamatergic neuron-specific RNAi strain is also comparable to the standard sensitized strain when nose touch behaviors are compared after feeding with eat-4 RNAi ( Figure 3C, D, p = 0.5205). Finally, RNAi in the cholinergic neuron-specific RNAi strain is slightly less efficient than the standard sensitized strain -there is a small, but significant difference in the thrashing rates when these strains are fed RNAi against unc-13 ( Figure 3E, F, p = 0.0430) or snb-1 (p = 0.0478). Additionally, RNAi of unc-13 in the cholinergic-specific RNAi strain is unable to fully recapitulate the slowed thrashing rate of unc-13(e51) mutants ( Figure S1B, p = 0.0003), due to a combination of decreased penetrance and effect size. All the strains presented, however, show dramatic behavioral defects when fed RNAi against genes that are required for those neurons to function.
An interesting feature of our system is that we do not observe effects for knockdown of gene function during development. For example, although GABA neurotransmission is required at all developmental stages for normal movement, our GABA-specific strain does not exhibit behavioral defects until the L4 and adult stage. Similarly, no developmental defects were observed during our GABA-specific screen of essential genes. A likely reason for this is that in our system, RNAi is not initiated until rde-1 is expressed, and expression from the promoters we use does not begin soon enough to affect behavior at earlier stages. Although this delay means that the developmental functions of genes cannot be studied with our system, it also allows bypassing developmental effects for genes that function both during development and afterward. For example, if a neuronal gene functions in axon guidance and also functions in neurotransmission, our system will allow specific analysis of this later role.
To our knowledge, our approach is the first in any metazoan that enables cell-specific knockdown in any chosen neuron subtype (with an available promoter), in a way that is high throughput enough to be compatible with questions involving large numbers of genes -including whole-genome screens. As additional specific sensitized strains are developed in addition to the four presented here, it will be possible to combine analysis of neural circuits with genetics, knocking down specific genes in specific parts of circuits and determining the effect on output. In general, we expect this technique will be useful for two major classes of applications: first, to rapidly determine the site of gene function by knocking down a gene of interest in specific neurons; and second, to perform forward genetic screens in a mosaic context, defeating the muddying effects of pleiotropy and biasing the hits toward genes that function intrinsically in the neurons of interest.

Strains and transgenics
All mutant C. elegans strains were provided by Caenorhabditis Genetics Center and maintained at 20uC as previously described [44]. Transgenic C. elegans lines carrying the transgenes as single copy insertions were created as described [21,45] using insertion site ttTi5605, then verified by PCR and Sanger sequencing. These insertions were then crossed into lin-15B(n744); eri-1(mg366); rde-1(ne219) mutant animals and genotyped by PCR, Sanger sequencing, and resistance to ama-1 RNAi. XE1583 was created by microinjection of XE1375 with 15 ng ml 21 of pTG96 [46] as

Microscopy
Young adult hermaphrodites were mounted in a slurry of 0.1 mm diameter polystyrene beads (Polysciences) on a 5% agarose pad and imaged using a UltraVIEW VoX (Perkin Elmer) spinning disc confocal microscope using a 606 CFI Plan Apo VC, NA 1.4, oil objective. Images were pseudo colored in Fiji [48].

Behavioral assays
Aldicarb hypersensitivity was measured as described [28]. Briefly, NGM agar was poured into 12-well plates and allowed to dry for 14 days at RT. Plates were then weighed, top-spread with 30 mM aldicarb (Ultra Scientific) to a final concentration of 750 mM, and allowed to dry for 6 hrs. Plates were then seeded with 5 ml of OP50 culture and allowed to grow overnight at RT. 25 young adult worms were then transferred to each well, and after 100 min, the number of paralyzed (defined as the cessation of all spontaneous movement) worms was counted. Each experiment was performed in triplicate.
Basal slowing was measured as described [33]. Locomotion for 20 worms was measured for each RNAi and each treatment.
Thrashing was measured by picking a single young adult worm into a drop of M9 buffer [44] on a glass slide at RT. After equilibrating for 30 sec, the number of body bends (complete movement of the anterior of the worm from one extreme to the other and back) was counted for 30 sec. Rates were measured for 10 worms for each RNAi treatment.
Response to nose touch was measured as described [36]. 10 animals were tested for each condition, 5 trials/animal. The percentage of reversals per total trials was calculated.

Screen
RNAi was performed as described above, with cultures grown in 96-deep-well plates. Two positive (unc-25) and two negative (L4440) controls were included for each 96 clones screened. Only cultures grown to log-phase were seeded onto plates as measured by OD600 using a Perkin Elmer Victor 2 plate reader. Aldicarb hypersensitivity was performed as above. Morphology of GABA neurons was examined using a Leica M165FC epi-fluorescent dissecting microscope under 5006 magnification.

Statistical analysis
Two-tailed, unpaired Student's t-tests were used to compare GFP RNAi and aldicarb hypersensitivity data in Figures 2 and 4, as well as basal slowing and thrashing in Figure 3. Two-tailed Fisher's exact test was used to compare nose touch behaviors in Figure 3. Figure S1 A comparison of neuron-specific RNAi phenotypes to mutants. (A) Aldicarb sensitivity of the GABAergic neuron-specific RNAi strain (grey bars) and isogenic strains (black bars). unc-25 RNAi recapitulates the null phenotype of unc-25(e156). Error bars are SEM, n = 3 trials for RNAi, 5 trials for isogenic strains of ,25 animals each (B) Thrashing rates of the cholinergic neuron-specific RNAi strain (closed circles) and isogenic strains (open circles). Each circle represents an individual worm, line is mean, error bars are SEM, n = 10 worms. (TIF)

Supporting Information
Table S1 Essential genes screened in primary GABAergic neuron-specific RNAi screen. 1,320 essential gene RNAi clones and the resulting aldicarb sensitivity when fed to the GABAergic neuron-specific RNAi strain. (XLS)