Acute Genetic Manipulation of Neuronal Activity for the Functional Dissection of Neural Circuits — A Dream Come True for the Pioneers of Behavioral Genetics

This review summarizes technical development of the functional manipulation of specific neural circuits through genetic techniques in Drosophila. Long after pioneers ’ efforts for the genetic dissection of behavior using this organism as a model, analyses with acute activation of specific neural circuits have finally become feasible using transgenic Drosophila that expresses light-, heat-, or cold-activatable cation channels by xxx/upstream activation sequence (Gal4/UAS)-based induction system. This methodology opened a new avenue to dissect functions of neural circuits to make dreams of the pioneers into reality.


INTRODUCTION
From late 1960s, Hotta and Benzer started genetic dissection of behavior controlling mechanisms with classic genetic methodology, that is, mutant analyses through behavior observation. Such mutant analyses had made great success in genetics of bacteria and phages. Likewise, mutant analyses using Drosophila made a substantial success to understand functions of the nervous system. However, there was a certain caveat in this approach. Complex behavior of a given multicellular animal is regulated by a neuronal network consisting of thousands of neurons connected with each other. We cannot expect that the function of a single gene should regulate the entire pattern of certain behavior. Thus, disruption of a single gene in all the neurons would not provide enough insights for understanding the mechanisms understanding the control of behavior through complex neuronal network. Therefore, success in Drosophila ' s behavioral genetics was restricted to only a limited repertoire of genes, whose specifi c defects result in a global behavioral disorder. One great example of such success was the analysis of the period ( per ) gene, which was identifi ed as causing defects in circadian rhythm by Konopka in Benzer ' s lab (Konopka & Benzer, 1971). Another representative example, Shaker ( Sh ), was identifi ed for its phenotype of ether-induced tremor (Kaplan & Trout, 1969) and turned out to be a structural gene of potassium channel (Kamb et al., 1987;Papazian et al., 1987;Tempel et al., 1987). In both per and Sh , phenotypes of the mutant genes were clearly recognizable as behavior defects of a whole animal. Although mutants isolated from behavior screening gave certain useful information, we had some kind of frustrated feeling that behavior genetics had not given fundamental mechanistic insights into the regulation mechanisms of complex behavior. Attempts for genetic dissection of behavior had not really dissected the role of the complex neural networks in the brain. Rather, they tend to result in the dissection of other issues such as intracellular signal transduction or fate determination during development.
Recently, a technical breakthrough that enables acute activation, i.e., " remote control " of specifi c neurons, was developed using Drosophila genetics (Lima & Miesenbock, 2005), which subsequently applied in other model organisms such as mice (Zhang et al., 2006). The new approach has opened up a way to dissect functions of Gregor Johann Mendel, former student of a physicist Christian Andreas Doppler at the University of Vienna, adopted quantitative methodology of physics into biology to analyze the mechanisms of inheritance by counting populations of peas (Mendel, 1866), leading to the establishment of " genetics. " In a similar quantitative strategy, Seymour Benzer at the California Institute of Technology (Caltech), also a former physicist who then performed a historical study in bacteriophage genetics to establish the concept of " cistron, " began genetic analyses of Drosophila behavior by observing a population of organisms rather than individual animals to quantify phenotypes. The fi rst example of Benzer ' s challenge was the quantifi cation of phototaxis behavior with the " counter-current apparatus " (Benzer, 1967). As the Benzer ' s fi rst postdoctoral fellow for Drosophila research, Yoshiki Hotta isolated many mutants in visual behavior using this technique. To characterize those visual mutants, Hotta employed electroretinogram (ERG), which is the extracellular potential recording on the surface of Drosophila compound eyes (Hotta & Benzer, 1969). Their approach was proven to be quite powerful to dissect various phenomena in neuroscience. For example, Hotta and Benzer ' s screening has led to the fi nding of many important mutants in signal transduction of Drosophila visual system such as norpA , which encodes phospholipase C protein (Bloomquist et al., 1988;Masai & Hotta, 1991).

Ikeda ' s Dream -" Command Neuron " in Neuroethology
Interestingly, another Japanese scientist, Kazuo Ikeda, also started genetic dissection of Drosophila behavior at the same time. He approached in the opposite direction to that of Hotta and Benzer. Whereas the latter moved from genetics to behavior, Ikeda originally worked in the fi eld of neurophysiology and neuroethology, from which he sought involvement of genetic factors in behavior. Before using Drosophila as a model, Ikeda had made a historically important study in neuroethology: the fi nding of " command neurons. " In Wiersma ' s lab at Caltech, Ikeda recorded periodic bursting pattern of ventral ganglion of crayfi sh that correspond to the movement of its swimmeret . Then, Ikeda physically dissected a crayfi sh neuropil connecting ganglions, and stimulated neurons one by one to fi nd neurons that can change the bursting pattern. After tiresome efforts, Ikeda found specifi c neurons that can turn on or off the bursting pattern, which he named the " command neurons " . It was the birth of the concept of command neurons, which was revived by a recent technical breakthrough we explain later.
After this study Ikeda embarked on the study of command neurons in Drosophila to seek genetic regulation of commanded behavior. For this purpose, He chose the fl y ' s fl ight system as a model (Ikeda et al., 1980;Koenig & Ikeda, 1980a, 1980b. Although physiological studies of the Drosophila fl ight system, together with those by Wyman ' s group (Harcombe & Wyman, 1977;Tanouye & Wyman, 1980), gave certain insights on the understanding neuronal networks, unfortunately we cannot say that genetic methods have provided suffi cient data to understand the functional regulation of the fl ight system. Because genetic methodology had not yet developed suffi ciently at that time, Ikeda and his colleagues of the period could not make full use of genetic technique for dissecting neural networks. Rather, because fruit fl ies are so small, fi ne neurophysiology at the same level as those performed at crayfi sh proved to be diffi cult.
Ikeda ' s interest shifted towards synaptic function using neuromuscular junctions of fl ight muscles as model synapses, leading to the series of works on shibire mutant. This mutant was isolated from a series of forward genetic screening of temperature-sensitive paralysis performed by David Suzuki and colleagues (Grigliatti et al., 1973). The temperature-sensitive shibire ts mutant enabled acute temporal controlling of its phenotype, providing clear data for their physiological experiments. This study led to the fi rst genetic dissection of synaptic function (Ikeda et al., 1976) and a critical fi nding that shibire mutant has defects in the recycling step of synaptic vesicles (Kosaka & Ikeda, 1983). Together with the cell biological analysis of Dynamin, the protein coded by the shibire gene (Takei et al., 1995), these analyses greatly contributed to the understanding of vesicle traffi cking.
Thus, Ikeda ' s shibire study started from the dissection of specifi c behavior but resulted in the understanding of general cell biology. Interestingly, though, the shibire ts mutant became a critical tool for acute temporal regulation of specifi c neurons more than 30 years later, bringing a methodological breakthrough for the functional dissection of neural circuits that Ikeda originally intended to address (see below).

Mosaic Analyses by Ikeda and Hotta
Hotta ' s forward genetics approach and Ikeda ' s physiological approach paved the way to understanding the relationship between genes and physiological phenomena. For example, Ikeda and Kaplan made the fi rst extracellular and intracellular recording from the Drosophila central nervous system to show abnormal bursting patterns of action potential in the thoracic ganglion of Hyperkinetic ( Hk ) mutant fl ies (Ikeda & Kaplan, 1970a). This study ties with the fi rst genetic dissection of Drosophila electrophysiology using ERG recording by Hotta and Benzer (Hotta & Benzer, 1969).
To ask which cells are actually responsible for the defect, Ikeda and Kaplan (Ikeda & Kaplan, 1970b) employed a genetic tool called mosaic analyses. The technique takes advantage of the ring X chromosome, which is lost during the fi rst nuclear division to make a gynandromorph: an animal one half of which is made from male cells and the other half from female cells. A heterozygous egg with a mutant gene on the X-chromosome and otherwise wild-type ring X chromosome, after loss of the ring X, becomes an adult gynandromorph fl y with mutant hemizygous male cells and heterozygous female cells, the latter showing wild-type phenotype. By correlating bursting phenotype and external markers in gynandromorphs, they found that abnormal bursting activity by Hk in one side of the thoracic ganglion corresponds to the external markers linked to Hk mutation, on the closest leg, suggesting that the bursting phenotype is cell autonomous to the Hk mutation. Hotta and Benzer (Hotta & Benzer, 1970) also performed mosaic analyses on an ERG mutant, tan , and suggested that the abnormal ERG is cell autonomous to the mutation.
In these mosaic studies, the observed behavior can be explained by the function of single cells because of their cell autonomous nature, and it is the reason why the mosaic analyses worked nicely. It is more diffi cult to utilize single gene mutation to dissect complex behavior that depends on the synergistic activity of complex neural networks, because it would not be common that a single gene is specifi cally responsible for the expression of a complex behavior in which various types of neurons are involved. More neurophysiological genetic tools are required to address these issues.

Hotta ' s Courtship Analyses -A Milestone of Classical Genetic Dissection
Hotta proceeded one-step forward by plotting the responsive locus of various behavior mutants onto the blastoderm map they newly established (Hotta & Benzer, 1972). In 1976, Hotta and Benzer made the fi rst success in dissecting complex behavior (Hotta & Benzer, 1976). Though they employed the same mosaic analyses technique, their approach this time was different from the previous ones; the study did not deal with any mutant gene or mutant phenotype, instead it dealt with the behavior that is different between male and female, namely the malespecifi c courtship behavior. Instead of using fl ies with any mutant gene on the X-chromosome, they just made gynandromorphs and observed their courtship. If the cells responsible for male courtship behavior are male cells in a gynandromorphous fl y, it should behave as a male, and vice versa. These analyses allowed them to determine which region in a fl y ' s body is responsible for courtship behavior. Through the mosaic analyses, they found that the region responsible for the male ' s wing vibration behavior to attract a female is located in the brain, whereas the region responsible for the next step of courtship, attempting copulation, resides in the thoracic ganglion.
In this special case, male/female difference in certain parts of the body caused clear behavioral differences, even though underlying structural/functional changes in the neural network should be very complex. Their courtship mosaic analysis was extremely cool and unique in this regard, and has a historical importance as the fi rst clear genetic dissection of a network-dependent complex behavior.
However, because this mosaic analysis maps the focus of responsible cells by measuring the correlation with the morphological phenotype of the external markers, the spatial resolution was rather low. Though it was able to map the responsible sites of wing vibration and attempting copulation behaviors onto the brain and the thoracic ganglion, it was not possible to locate which neurons in these ganglia are actually involved in courtship control. We had to wait until recent technical advances for enabling more detailed analyses at the network level. In the following sections, we explain technical advancements that fruited into the present genetic manipulations of behavior at single-cell level.

Development of Gene Expression System in Drosophila Molecular Genetics
One of the most infl uential breakthroughs in fl y genetics was the transformation technique of ectopic genes using the P element transposon, developed by Rubin and Spradling (Rubin & Spradling, 1982;. It enabled researchers to introduce any gene of interest into fl y genome to drive its expression. As an additional benefi t, the technique turned out also to be highly useful for mutagenesis, because the P element transposon is randomly inserted into the fl y genome to cause an insertional mutation. Intentional mobilization of P element transposon by supplying the transposase source was attempted for the purpose of mutagenesis, followed by the cloning of genes adjacent to the P-element insertion (Cooley et al., 1988;Yoshihara et al., 1988). A transformant strain often called " delta2 -3, " in which a P element transposase gene without the intron between open reading frames ORF2 and ORF3 (Laski et al., 1986) was stabilized in the genome (Robertson et al., 1988), turned out to be the most stable transposase source, which is used as a common tool for P element -mediated transformation and mutagenesis until today.
The next important step in the history of Drosophila genetics appeared soon afterwards: the " enhancer trap " technique (O'Kane & Gehring, 1987). Because the original P element -mediated mutagenesis mobilized a transposon that carries only a marker gene to isolate the transgenic fl ies (e.g., rosy or white driven by a constitutive promoter), the recombinant line remained useless unless the transposon was inserted right into or near certain gene to induce detectable phenotype. O ' Kane put one more gene construct called " reporter " in the transposon. The reporter is a gene whose expression can easily be detectable -e.g., the β -galactosidase gene lacZ -, under the control of a weak promoter. The reporter gene will be expressed under the infl uence of a nearby enhancer. Thus, the activity of the enhancer near the insertion locus can be " trapped " by visualizing the reporter protein. If the enhancer regulates the expression of an endogenous gene A, the expression pattern of the reporter gene is supposed to mimic the expression of gene A. The enhancer-trap strains provided an effi cient way to isolate genes with specifi c expression pattern regardless of whether the transposon caused insertional mutation.
Another big advancement came several years later by Brand and Perrimon (Brand & Perrimon, 1993), who introduced a yeast transcription factor, Gal4 , and its target, the upstream activation sequence (UAS), into fl y genome. Unlike reporter genes such as lacZ , Gal4 does not visualize the cells by itself but drives expression of another gene that is put downstream of UAS. By crossing a single strain carrying a Gal4 insertion with the strains carrying various genes under UAS, these genes can be expressed in the same set of cells reproducibly. Likewise, by crossing a strain carrying a UAS construct with various GAL4carrying strains, the same gene can be expressed specifi cally in numerous ways. The separation of the tools to specify the gene expression pattern (by Gal4 strains, called drivers) and those for selecting the genes to be expressed has dramatically enhanced the way to visualize and manipulate specifi c cells in the nervous system, fulfi lling a requirement for the modern approach towards circuit dissection.

Effectors for Neuronal Function: Spatially Regulated Neural Manipulation
The UAS-conjugated genes are called either a reporter or effecter, depending on whether it is used just to visualize the cells or to alter their fate or function. Taking the advantage of the Gal4/UAS system, many effecter genes were developed. For example, Sweeney et al. (Sweeney et al., 1995) made UAStetanus toxin light chain ( TNT ), which cleaves a synaptic vesicle protein Synaptobrevin that is essential for nerve-evoked synaptic transmission (Yoshihara et al., 1999). There were also attempts to suppress or enhance activity of neurons. To suppress activity of neurons by enhancing potassium conductance, White et al. (White et al., 2001) made UAS-EKO (electrical knock out) by genetic engineering of the Shaker potassium channel (Kamb et al., 1987;Papazian et al., 1987;Tempel et al., 1987) to deprive its inactivation property. Baines (Baines et al., 2001) made UAS-Kir , which encodes an inward rectifi er potassium channel that regulates the resting potential of neurons. Overexpression of Kir channel is also supposed to increase potassium conductance, leading to the inactivation of neurons. In the opposite direction, UAS-NaChBac , which is a bacterial sodium channel, was introduced into fl y genome to enhance neural activity by increasing sodium conductance (Luan et al . , 2006a). Using these effectors, researchers can now block synaptic transmission (by UAS-TNT ) or suppress or enhance neural activity (by UAS-driven EKO and Kir or NaChBac , respectively) of specifi c sets of neurons.

UAS-shibire: Breakthrough for Temporary Regulated Neural Manipulation
However, the effect of these manipulations are chronic; the nervous system may resort to physiological or developmental compensatory effects to minimized the defects caused by the malfunctioning neurons (Turrigiano & Nelson, 2000). Thus behavioral phenotypes by expression of these effecters are somewhat tricky to be interpreted.
After a series of seminal study on the shibire mutant in synaptic transmission by Kazuo Ikeda, Kim and Wu (Kim & Wu, 1990) found antimorphic effect of the mutation, shibire ts1 , that is, even heterozygous shibire ts1 / ϩ fl ies show weak paralytic phenotype. It suggested that the expression of the mutant gene works dominantly to suppress synaptic transmission. To take advantage of this dominant effect, Toshi Kitamoto, who was working in the next lab to Kazuo Ikeda, simply combined UAS with shibire ts1 , which had been cloned by Meyerowitz ' s lab (van der Bliek & Meyerowitz, 1991). Overexpressed mutant shibire protein caused no effect when the fl ies are kept in low temperature, but quickly suppressed synaptic transmission when the ambient temperature was raised to ca. 30 ° C (Kitamoto, 2001). This system provided researchers with a highly effi cient tool to manipulate neural function both cell-specifi cally and temporally. The system became immediately popular as a genetic tool for memory studies to prove its effi ciency (Waddell et al., 2000;Dubnau et al., 2001;McGuire et al., 2001). The ease of temporal regulation just by temperature shift allowed researchers to ask when synaptic transmission from specifi c groups of neurons is required for each step of memory formation and retrieval.
(Note: we should be careful when interpreting the results of shibire experiments. Because shibire is a mutant form of Dynamin GTPase, a crucial protein for vesicle formation, its malfunction not only affects synaptic vesicles but also intracellular vesicle traffi cking in other parts of the cells. For example, effect of shibire is also observed in the postsynaptic compartment, which may well play important roles in memory formation [Yoshihara et al., 2005], and expression of shibire in the glial cells blocks their phagocytic action to affect axon pruning [Awasaki & Ito, 2004]).

Temperature-Sensitive Gal80: An Alternative Approach for Temporary Regulated Neural Manipulation
As discussed before, there are many effecter genes that can block or alter neural function, but unfortunately temperature-sensitive mutant variants have not been identifi ed for them. To compensate this problem, another method for spaciotemporal neural manipulation was developed using a temperature-sensitive allele of Gal80 , which inhibits Gal4 function by its binding to Gal4 in yeast (Ma & Ptashne, 1987) . The ability of Gal4 as cell-specifi c expression driver can be blocked by expressing Gal80 ts , which is usually driven generally by constitutive promoter such as Tubulin promoter. When the temperature is raised to ca. 30 ° C, the temperature-sensitive GAL80 ts protein can no longer suppress GAL4, allowing the expression of UAS-conjugated effecter genes (McGuire et al. , 2003).
When the fl ies are kept at high temperature for a relatively long period (e.g., overnight), the effecter protein accumulated in the cells can continue affecting the neural function for several hours after the fl ies are moved back to the ordinary temperature. This is convenient in certain experiments in which normal behavior is affected simply by raising the temperature (e.g., auditory courtship response; Kamikouchi et al., 2009). In those experiments, chronic temporal regulation using Gal80 ts has advantage over acute manipulation using UASshibire .

Acute Activation of Ion Channels: The Biggest Breakthrough to " Remote Control " Neurons
Although effecters such as shibire ts , TNT , EKO , Kir , and NaChBac block or modify neural function, it is not possible to activate specifi c neurons acutely like stimulating electrode can do. A new series of technical breakthrough to address this issue came in three fl avors: activation of cation channels by caged compound, light, and temperature shift.
The fi rst approach was developed by the laboratory of Gero Miesenbock (Lima & Miesenbock, 2005). They made a transgenic fl y carrying the UAS construct with the mammalian adenosine triphosphate (ATP) receptor channel gene, whose homologue does not exist in fl y genome. To activate the channel, they injected caged ATP into fl y ' s body. Flashing light uncaged the caged molecule to release ATP, which then binds to the ATP receptor to open the channel, causing infl ux of cations into neurons to depolarize them. To test the system, they expressed the ATP receptor channel in the giant fi ber neurons, which are known to trigger the jump muscle of the legs to induce escape response. As expected, light illumination triggered fl ies to suddenly jump and beat wings. As they called the methodology " remote control, " the new technique can activate neurons in free-running animals, a great advantage over the conventional stimulation methods with electric wires used for mammals. The dramatic success of Lima and Miesenbock revived Ikeda ' s concept, the " command neuron, " which triggers a stereotypic pattern of behavior such as the escape response commanded by the giant fi ber neurons. However, administration of caged ATP into tiny fl y bodies is not very practical, making it diffi cult to apply this technique to a large number of animals.
Another approach made use of a bacteria-derived Channelrhodopsin 2 (ChR2), a sodium channel that opens when illuminated by blue light (Nagel et al., 2003). Unlike ATP receptor channel, injection of additional chemical is not required. A fl ush of blue light is enough to activate the ChR2-expressing neurons. The system was successfully used in fl y larvae to activate different types of monoamine neurons to induce reward and avoidance learning (Schroll et al., 2006), showing that these neurons convey essential signal for memory formation. Because light stimulation is easily applicable to vertebrates, the technique is now commonly used by mammalian researchers (Zhang et al., 2006). However, remote control with ChR2 is rather difficult for adult fl ies because of the two reasons. First, unlike larvae, the adult fl y is covered by brownish cuticle, which prevents effi cient transmission of blue light into the brain.
And second, even though ChR2 does not require injection of caged chemicals, it does require retinal for its activation (Nagel et al., 2003). Feeding fl ies and worms with retinal is therefore necessary Schroll et al., 2006), whereas supplying retinal is not necessary for mammalian neurons (Boyden et al., 2005). However, whereas larvae feed ravenously, adults live on tiny amount of daily food, making it diffi cult to administrate suffi cient amount of retinal.
The third approach utilizes heat or cold stimulus to activate neurons. Drosophila TrpA1 (Hamada et al., 2008) and mammalian TRPM8 (Bautista et al., 2007) are members of the ubiquitous transient receptor potential channels that respond to specifi c ranges of temperature to be used for fl y thermoenetics. In Drosophila , TrpA1 is activated at high temperature ( Ͼ 28 ° C; Hamada et al., 2008) and TRPM8 at low temperature ( Ͻ 15 ° C; Peabody et al., 2009). Remote control can be realized by expressing these channels in specifi c neurons of the brain and shifting the temperature of a small chamber housing the fl ies. Because of the small thermal capacity of the fl ies due to their featherweight body mass (ca. 1 mg), heat or cold stimulation can activate neurons quickly even when they are embedded deeply in the fl y brain. Thus, the temperature-induced Trp channels have become the most popular method for remote controlling neurons in the adult fl ies (Aso et al., 2010;Krashes et al., 2009). The technique, however, is not easily applicable to mammals because of their homeothermic nature and large thermal capacity.

Methodology to Restrict Expression Beyond Enhancer Trapping
Whereas the temporal control technique of effecters has dramatically been improved during the last decade, the resolution of the spatial control remains largely unchanged since early 1990s, when the Gal4 enhancer-trap system was developed. To drive expression in a smaller numbers of neurons, a group at the Janelia Farm research campus established a large number of transformant strains in which the Gal4 fused with a small genomic fragment from an upstream or downstream region, or an intron of various genes are inserted at a fi xed position of the genome (Pfeiffer et al., 2008). The expression pattern of Gal4 is expected to mimic subset of that of the endogenous gene.
However, in situ RNA hybridization and antibody labeling of many genes show a general tendency that the expression patterns of endogenous genes are by themselves not highly restricted. Rather, a gene is often expressed in a variety of cells scattered in the brain (the law of low specifi city), and not all the cells in a defi ned brain region express the same gene (the law of low ubiquity; Ito et al., 2003). To improve the specifi city of driver expression in such circumstances, so called " intersection " approach is being developed. This method makes use of two different enhancer trap patterns to induce expression only in the overlapping regions.
One method toward this aim is called the split Gal4 system (Luan et al., 2006b), in which two major domains of the Gal4 protein -the Gal4 DNA-binding domain (GAD) and Gal4 transcription -activation domain (GTA) -are put into deferent P-element constructs to make independent enhancer-trap strains. Gal4-mediated expression of UAS-linked genes should occur only in the cells where GAD and GTA are co-expressed. Thus, by crossing the GAD-enhancer trap and GTA-enhancer trap strains, specifi c expression can be induced only in the intersection of the expression patterns of the two lines.
Another intersection approach uses aforementioned Gal80 to inactivate Gal4 in certain cells. Enhancer-trap strains were generated with Gal80 under control of a weak promoter (Bohm et al., 2010). If one of these lines is crossed to a Gal4 line, Gal4 cannot drive expression in the cells that express Gal80. Thus, the collection of the cells that express Gal80 will be subtracted from that of the Gal4-expressing cells to attain a more specifi c pattern. Both split-Gal4 and Gal80-subtraction methods requires generation of a large number of new enhancer-trap strains to make the approach suffi ciently versatile.
Random loss of Gal80 suppression using MARCM (Mosaic analysis with a repressible cell marker) (Lee & Luo, 1999) or fl ip-out Gal80 (Struhl & Basler, 1993) technique can also limit the expression pattern of Gal4. Both features yeast-derived fl ippase gene to induce somatic recombination or fl ipping-out to remove Gal80 from the DNA. If fl ippase is expressed only mildly, the loss of Gal80 occurs only in a few cells, in which GAL4mediated effecter expression should occur. Unlike other methods, expression of the effecter genes is not reproducible, because the loss of Gal80 occurs only randomly. In spite of this, the approach has a potential to dissect a given neural network at the level of single neurons.

Foci and Clusters: Courtship Analyses From Hotta to Yamamoto and Dickson
Following the pioneering study on Drosophila courtship by Hotta and Benzer (1976), Daisuke Yamamoto and his colleagues performed a large-scale screening of Pelement insertion lines to isolate mutants in sex-specifi c courtship behavior. One of the isolated lines caused interesting homosexual behavior, which later turned out to be an allele of the fruitless ( fru ) gene, a critical regulator of male/female courtship (Ito et al., 1996). Because the original strain featured the fi rst-generation P-element con-struct without any reporter or Gal4, advanced analysis of expression pattern was diffi cult. In what can be called bonanza, when a large scale generation of the Gal4 enhancer-trap strains by the Japanese NP (Nippon, Japan) consortium was performed with the crucial assistance of the Yamamoto lab (Yoshihara and Ito, 2000), the 21st strain out of over 4500 lines generated hit the locus of fru , paving the way for detailed visualization and manipulation of fru -expressing neurons.
Further analysis of fru -expressing neurons led Yamamoto ' s group to identify the fi rst sexually dimorphic neurons in the Drosophila brain (Kimura et al., 2005). The fru -expressing neurons form several clusters. To ask which cluster is involved in triggering male courtship behavior, they employed the remote control approach to activate one of the fru -expressing neurons randomly by restricting the expression of TrpA1 effecter to singles cells by MARCM technique (Kohatsu et al., 2011) and isolated the most likely neuron clusters, P1 and P2b, which contains the sexually dimorphic neurons identifi ed earlier (Kimura et al., 2008).
Approaching from a different starting point, the group led by Barry Dickson employed the intersection method to restrict fru -expressing neurons and identifi ed P1 and another cluster similar to P2b as the command centers for courtship song (von Philipsborn et al., 2011). They also identifi ed neurons in the thoracic ganglion as parts of the central pattern generators for wing vibration, whose existence had been suggested by Miesenbock ' s group (Clyne & Miesenbock, 2008).
Hotta and Benzer (Hotta & Benzer, 1976) identifi ed various regions of the brain and the thoracic ganglion that are responsible for the phenotypes of each step of Drosophila courtship behavior and called them foci. More than 30 years later, these recent studies appear to be on the extrapolation of Hotta ' s mosaic analyses of courtship behavior, because the clusters of neurons identifi ed in these studies are likely to correspond to the foci identifi ed by their predecessor.

Fulfi lling Ikeda ' s Dream: The First Systematic Screening of Command Neurons
Though Ikeda established the concept of command neurons, technical limitation made it diffi cult to isolate such neurons in the insect nervous system. With the array of advanced molecular genetic techniques at hand, we are now in a position to make his dream reality.
The analysis of fru -regulated courtship circuitry is a sort of reverse-genetics approach, in which both the genes to be analyzed and the phenotype to be observed are defi ned beforehand. Considering the highly limited level of our current knowledge and understanding about brain and behavior, forward genetics approach may also provide important insights.
To this aim, we used the collection of Gal4 enhancertrap lines established by the NP consortium (Yoshihara & Ito, 2000) to systematically seek command neuron circuits. We crossed Gal4 driver lines to cold-or heatactivatable TRPM8 (Peabody et al., 2009) or TrpA1 (Hamada et al., 2008)) channels, and video-recorded the behavior of fl ies after acute temperature shift to activate the channels in a small temperature-controlled chamber. Though the screening is still continuing, various elements of behavior, such as jumping, grooming, wing beating, egg laying, etc., have already been observed.
Stimulation of one strain evoked the entire sequence of feeding behavior, complete with proboscis extension, expansion of labellum, movement of pharyngeal pumps, and retraction of the proboscis. Narrowing down the responsible neuron by the " fl ip-out Gal80 " led us to identify a single pair of interneuron neurons, named " feeding neurons, " whose activation alone can trigger the entire feeding sequence. Because this command neuron is likely to integrate information of food taste with other inputs to trigger feeding conditionally, analysis of these neurons would serve as a model system expand our knowledge of the synaptic plasticity revealed at the neuromuscular junction (Yoshihara at al., 2005) to understand the role of central synapses in the plastic neural task such as Pavlov ' s classical conditioning at central synapses.
Similar large-scale screening of command neurons is currently going on in various research groups. Forty-eight years after his study of crayfi sh, Ikeda ' s dream not only came reality but also became the main stream of modern neuroscience in the 2010s.

CONCLUSION
Yoshiki Hotta preached us at his retirement at the University of Tokyo, based upon his deep insight: " We cannot change the past, but we can build our future. " Hotta and Benzer ' s approach made a conceptual breakthrough to understand cellular functions of neurons as the basis of behavior, which has guided scientists from various backgrounds during the last several decades. However, Hotta ' s dream and Ikeda ' s dream -to dissect behavior from the viewpoint of individual neural circuits -were somewhat too advanced at that period. To enable effi cient genetic dissection of behavior, they -and we -had to wait until today ' s technical development. We are now thrilled to witness the dramatic future of behavioral genetics. Their dreams have just started to bloom.

Declaration of interest:
The authors report no confl icts of interest. The authors alone are responsible for the content and writing of the paper.