Functional dissection of the neural substrates for sexual behaviors in Drosophila melanogaster.

The male-specific Fruitless proteins (FruM) act to establish the potential for male courtship behavior in Drosophila melanogaster and are expressed in small groups of neurons throughout the nervous system. We screened ∼1000 GAL4 lines, using assays for general courtship, male-male interactions, and male fertility to determine the phenotypes resulting from the GAL4-driven inhibition of FruM expression in subsets of these neurons. A battery of secondary assays showed that the phenotypic classes of GAL4 lines could be divided into subgroups on the basis of additional neurobiological and behavioral criteria. For example, in some lines, restoration of FruM expression in cholinergic neurons restores fertility or reduces male-male courtship. Persistent chains of males courting each other in some lines results from males courting both sexes indiscriminately, whereas in other lines this phenotype results from apparent habituation deficits. Inhibition of ectopic FruM expression in females, in populations of neurons where FruM is necessary for male fertility, can rescue female infertility. To identify the neurons responsible for some of the observed behavioral alterations, we determined the overlap between the identified GAL4 lines and endogenous FruM expression in lines with fertility defects. The GAL4 lines causing fertility defects generally had widespread overlap with FruM expression in many regions of the nervous system, suggesting likely redundant FruM-expressing neuronal pathways capable of conferring male fertility. From associations between the screened behaviors, we propose a functional model for courtship initiation.

GENETICS offers powerful approaches for (1) identifying the neural circuitry underlying complex behaviors, (2) elucidating how such neural circuits are organized during development or modified by experience, and (3) understanding how such circuits function in behaving animals. Indeed, distinct genetic elements may regulate the neural substrates of behaviors as diverse as courtship and mating, aggression and avoidance, speech, language, and social behavior (Juntti ; Robinson ; Fisher and Scharff 2009; Siwicki and Kravitz 2009; Wu ; Juntti ; Robinett ).

As reproductive behaviors are often innate, they provide excellent systems for genetic approaches (Manoli ; Kimchi ; Portman 2007; Juntti ; Villella and Hall 2008). Neuronal circuits mediating reproductive behaviors must act to discriminate relevant from nonrelevant stimuli, integrate information across multiple sensory modalities, and generate appropriate behavioral output. Understanding how sex-specific genetic functions organize this circuitry, as well as how that circuitry functions, will hopefully elucidate how complex behaviors are generated by the nervous system. Here, we use molecular tools derived from the Drosophila melanogaster sex determination gene to dissect the behavioral components of male courtship behavior.

The D. melanogaster male courtship ritual is an extensively studied, complex innate behavior that can be executed by males reared in isolation (Hall 1994; Greenspan and Ferveur 2000). Information perceived via multiple sensory modalities is integrated to direct both the initiation and progression of courtship. The initial identification of appropriate female targets is via visual and olfactory cues (Sturtevant 1915; Hall 1994; Greenspan and Ferveur 2000; Stockinger ; Kurtovic ; Krstic ). Subsequent steps are mediated via contact-mediated chemosensory or mechanosensory cues perceived during tapping, licking, and attempted copulation (Acebes ; Bray and Amrein 2003; Lacaille ; Moon ; Koganezawa ). In addition, auditory cues (song) generated by males enhance male courtship drive and stimulate female receptivity (Kowalski ; Ejima ). Although most steps of courtship are innate, some are experience dependent (Griffith and Ejima 2009). Thus studies of male courtship behavior can contribute to understanding sensory processing and integration, coordination of motor programs, and motor output, as well as experience-dependent behavioral modifications.

In Drosophila a regulatory gene hierarchy governs all aspects of somatic sexual differentiation, including the potential for male courtship behavior. This cascade directs the synthesis of the sex-specific transcription factors encoded by the () and () genes (Baker ; Manoli ; Dickson 2008; Villella and Hall 2008; Yamamoto 2008; Siwicki and Kravitz 2009). is the key effector through which the nervous system is sculpted for male behavior. Transcripts derived from the distalmost promoter (P1) of the locus (fru) are sex-specifically spliced to produce mRNAs that encode Fru proteins (FruM) in males and are untranslated in females (Ryner ; Heinrichs ; Lee ; Usui-Aoki ). FruM transcription factors are expressed in ∼2% of neurons in the brain and ventral nerve cord (VNC), as well as in sensory components of the PNS (Ryner ; Lee ; Manoli ; Stockinger ; Cachero ; Yu ). FruM expression in the appropriate neurons is necessary and sufficient to generate the potential for nearly all aspects of male courtship behavior (Manoli ; Stockinger ; Manoli ). Although the fru expression pattern is grossly similar between males and females (Manoli ; Stockinger ), visualization of subsets of fru-expressing neurons revealed several fru-dependent sexual dimorphisms (Kimura ; Rideout ; Datta ; Kimura ; Koganezawa ; Mellert ). More recent systematic anatomical characterizations of individual fru neurons in males and females have revealed extensive sexual dimorphism in the circuitry (Cachero ; Yu ).

The presence or absence of fru products also governs other sex-specific Drosophila behaviors, which include male- and female-specific aggression behaviors (Chen ; Nilsen ; Vrontou ; Chan and Kravitz 2007), and at least some aspects of female reproductive behaviors (Kvitsiani and Dickson 2006; Yapici ; Häsemeyer ; Yang ). fru-expressing neurons appear to be dedicated to mediating these social interactions, since silencing these neurons via expression of the neural silencer shi does not affect other general behaviors (Manoli ; Stockinger ), and activating them triggers courtship behavior (Kohatsu ; Pan ; von Philipsborn ).

plays an important role in the production of courtship song and the generation of sexually dimorphic numbers of neurons in parts of the central nervous system (CNS) and peripheral nervous system (PNS) (Demir and Dickson 2005; Manoli ; Rideout ; Kimura ; Sanders and Arbeitman 2008; Mellert ; Rideout ). However, the overall effects of null mutants on courtship behavior are subtle compared to those of .

The broad pattern of fru expression and the potent effects of impaired fru function on sexual behavior raises the question of which neurons regulate particular aspects of the behavior. Mosaic mapping experiments using XO//XX mosaics identified regions of the fly that need to be male for different steps of courtship to occur (Hotta and Benzer 1972; Hall 1977, 1978, 1979). Although these studies have inherent limitations that restrict their resolution (Kankel and Hall 1976), these results are consistent with the proposition that subsets of the CNS appear to function in specific steps of sexual behaviors, a key premise of the current study.

More recently, many studies focused on select groups of fru neurons have associated particular behavioral phenotypes with defined subsets of fru-expressing neurons (Ferveur ; O'Dell ; Ferveur and Greenspan 1998; Lee and Hall 2001; Lee ; Manoli and Baker 2004; Couto ; Fishilevich and Vosshall 2005; Chan and Kravitz 2007; Kurtovic ; Clyne and Miesenböck 2008; Datta ; Kimura ; Koganezawa ; Kohatsu ; von Philipsborn ). These studies provide entry points for the further understanding of the circuitry underlying male sexual behavior.

To more systematically assess the functional roles of subsets of FruM-expressing neurons, we have used >1000 random GAL4 enhancer traps to target RNAi-mediated inhibition of FruM expression in reproducible, restricted subsets of FruM neurons (Figure 1). Our primary screen used assays for male fertility, elevated levels of male–male interactions, and perturbations in easily scored aspects of male–female courtship to dissect the functional components of Drosophila sexual behavior. From these assays, we identified lines producing altered latencies to courtship initiation, increased male chaining, male–male courtship or aggressive behaviors, and defects in male fertility. The GAL4 lines identified in our initial screen were subjected to secondary screens that sought to differentiate between various neural processes that may have contributed to similar behavioral phenotypes, as well as to assess the contribution of certain types of neurons to the phenotypes observed. Finally, as an initial attempt to correlate these functionally defined phenotypic classes with the manipulation of particular subsets of –PI-expressing neurons, we determined which neurons express the GAL4 drivers that produced deficits in male fertility.

Figure 1 

Regulation and targeting of FruM-specific elements of Drosophila sexual behavior. (A) Expression of UAS-fru produces effects only in those cells where enhancer-trap GAL4 expression (green) intersects (white) with endogenous FruM expression (magenta). (B) Schematic illustrating behavioral screening for phenotypes arising from directed inhibition of FruM expression. GAL4/UAS-fru males were screened with three primary assays for (1) male fertility, (2) abnormal male–male interactions, and (3) courtship behavior. Lines that produced phenotypes in any of these screens were subjected to specific secondary screens in each primary phenotypic class.

Materials and Methods

Drosophila stocks and cultures

Flies were raised on standard dextrose media. Crosses were performed at 29° to optimize GAL4 and RNAi function. CO2 was used to anesthetize flies. The UAS-fru, UAS-GFP-IR, and fru-P1-LexA stocks have been previously described (Manoli ; Mellert ). The GAL4 enhancer trap collection was generated in the laboratory of Ulrike Heberlein (personal communication; University of California, San Francisco) and was produced via the mobilization of the pGAWB element into a w Berlin strain. The Canton-S strain (CS) used was the CS-A strain obtained from the laboratory of J. Hall (Brandeis University, Boston). The UAS-fru, Cha-GAL80 stock was made by recombining the Cha-GAL80 (Kitamoto 2001) transgene with the chromosome III UAS-fru. The fru allele was obtained from the Dickson lab (Demir and Dickson 2005). We used a nuclear GFP reporter for fru-P1-LexA, LexAop-Stinger-GFP, made by replacing the UAS element of UAS-Stinger-GFP with four LexAop binding sites (Barolo ; Lai and Lee 2006).

Primary assays

F1 males containing each GAL4 enhancer trap and two UAS-fru (experimental) or two UAS-GFP-IR (control) transgenes were collected within 12 hr of eclosion from 29° crosses and stored individually at 29° for the times indicated. To verify behavioral phenotypes, genotypes displaying altered behavior were produced from new crosses and subject to identical retests.

Fertility:

Three to 5 days after collection, 5–10 F1 males were crossed individually to a pair of ∼1-week-old CS virgins. Crosses were carried out at 29° and the number of pupae counted once the first pharates appeared (7–8 days). GAL4/UAS-GFP-IR males were used as controls and produced an average of 45.7 pupae (standard deviation of 12.9). Genotypes with an average fecundity of <25 pupae were retested.

General courtship:

F1 males were entrained for 3 days in a 12-hr light:dark (12 l/d) cycle at 29°. Following entrainment, typically four males and four 2-day-old CS virgin females were anesthetized with CO2 2–4 hr after lights on [zeitgeber time (ZT) 2–4] and placed into individual wells of a 24-well tissue culture plate (Falcon). Animals were allowed to recover for 3–5 hr at 25° following anesthesia. For courtship assays, plates containing males were inverted onto those containing females, with a thin plastic barrier between the chambers at 25°. The barrier was moved allowing a row of males to be simultaneously introduced into the female chambers below and the barrier replaced, giving each well one male/female pair. Latency to courtship initiation (first wing extension) was measured, and any gross aberrations in courtship behavior were noted. Retests were performed on genotypes in which two or more males initiated courtship in <60 sec or failed to initiate courtship within 2 min.

GAL4 lines showing defects in wing extension and vibration were reexamined in detail on video recordings. Pairs were observed in a standard Plexiglas mating wheel with cylindrical chambers (diameter 10 mm, height 6 mm) (Villella ) at 25° and videotaped for 10–15 min or until copulation. Observations were conducted between ZT 2 and ZT 7. At least four males from each line were videotaped and for some assays a program called LifesongX programmed by J. Reiffel was used to count the number of abnormal behaviors, as well as to record time spent courting. Both CS and GAL4/UAS-GFP-IR males were used as controls.

Male–male interaction:

Following 3–5 days at 29°, males were entrained at 29° for 3–4 days in a 12 l/d cycle. A total of seven to eight males per line were then anesthetized and placed together into a 24-well tissue culture plate (Falcon) that contained ∼1 mL of standard dextrose media at 29°. Males were observed for several minutes twice daily (ZT 2–3 and ZT 8–10) for 3 days and chaining behavior and male–male courtship and aggressive behaviors were scored qualitatively on a 0–3 scale, ranging from no aberrant behavior to very high levels of abnormal behavior. Chaining was defined as three or more males showing persistent courtship in a series for at least 5 sec, whereas male–male courtship was between pairs of males (supporting information, File S1). Aggressive behaviors were defined as previously described (File S2) (Chen ; Nilsen ).

Secondary Assays

Male–male habituation:

Males from lines exhibiting chaining behavior or male–male courtship were isolated and entrained as in the male–male interaction assay. Typically 20 males were anesthetized using CO2 in the circadian morning (ZT 2–4) and placed individually in wells of a 48-well tissue culture plate (Falcon). Animals were allowed to recover for 4–5 hr following anesthesia at 25°. Plates containing males were inverted onto those containing sibling males, with a thin plastic barrier between the chambers. For a given line (10 sibling pairs), the barrier was partially moved, the male from the upper chamber was introduced into the lower chamber, and the barrier replaced. Males were videotaped for 7 min immediately after pairing and for 5 min beginning 55 min after pairing, at 25°. Courtship indexes (time courting/observation period) were determined for minutes 2–7 (initial) and 55–60 (final).

Mate preference:

Males from lines exhibiting chaining behavior or male–male courtship were tested for their sexual preference in a triad assay in which one test male was placed together with a target female and a target (decoy) male. Decoy males were genotypically c155-GAL4/UAS-fru and thus phenotypically male externally, but unable to court due to the expression of RNAi throughout their nervous system. For these assays, test males were isolated and entrained as in the male–male interaction assay. CS virgins were 4–6 days old at testing, not entrained, and kept at room temperature following collection. c155-GAL4/UAS-fru decoy males were 18 hr to 3 days old at the time of testing and were maintained at 29° until testing. Test males were anesthetized with CO2 for loading and allowed to recover for at least 3.5 hr at 25°. All behavioral tests took place in the circadian afternoon (ZT 7–9). Chambers were 13 mm in diameter and 18 mm deep and split into three vertical sections by removable thin plastic dividers. Flies were separated until testing, then introduced together by removal of the dividers. Following divider removal and tapping down, trials were recorded for 5 min.

To calculate the courtship preference index (Cpi), test males were scored for 5 min or until copulation. The courtship index was calculated as follows: CI = (time spent courting)/(duration of trial); CIf and CIm represent the fraction of time courting the female and decoy male, respectively; Cpi = (CIf − 0.5)/(CIf + CIm). Mean Cpi for each line was compared to mean wild-type Cpi and to zero by a one-sample t test.

Cha-GAL80 suppression:

Enhancer-trap lines exhibiting chaining behavior or reduced fertility were tested as to whether the neurons through which these phenotypes were generated were cholinergic. F1 GAL4/UAS-fru males with and without a Cha-GAL80 transgene were collected and isolated within 12 hr of eclosion from parallel crosses raised at 29°. Fertility and male–male interactions were assayed as in Primary Assays above.

Female fertility rescue:

Enhancer-trap lines that had reduced male fertility in the primary assay were subsequently examined for their ability to rescue female infertility induced by expression of the -masculinizing fru allele (Demir and Dickson 2005). Males containing UAS-fru and the fru allele were crossed to females from each GAL4 line, and their ; UAS-fru; UAS-fru, fru male (as a control for the fru) and female offspring were tested for fertility as previously. Males were crossed to two CS females, whereas females were crossed to two CS males.

Expression

To visualize fru expression independently of GAL4, we inserted the LexA::VP16 transcriptional activator into the locus (Mellert ). This transactivation system makes use of the Escherichia coli LexA gene fused with the VP16 activation domain, which targets specific LexAop binding sites (Lai and Lee 2006).

We used a nuclear GFP reporter, LexAop-Sti-GFP with fru-P1-LexA. As has been reported (Lai and Lee 2006), LexAop reporters often suffer from leaky expression. We used an insertion (E) with the least leakage for examination of fru and GAL4 overlap in the CNS and genitalia, and that insertion in combination with a second one (F, which does leak in the CNS and genitalia) for examination of expression in the antenna, maxillary palp, proboscis, and legs, where the stronger expression provided by two insertions is useful. We used fru-P1-LexA, LexAop-Sti-GFP in conjunction with an optimized nuclear DsRed, UAS-Red-Sti (Barolo ) to simultaneously visualize LexA and GAL4 expression.

For the visualization of FruM protein we used rat anti-FruM antibody at 1:300 dilution and Cy3 antirat secondary antibody from Jackson ImmunoResearch at a 1:800 dilution, as described (Lee ). Samples were imaged at ×20 magnification on one of the following confocal microscopes: BioRad MRC 1024, Zeiss 510 Meta, or Zeiss 710, and processed with ImageJ software.

Statistical analysis

Statistical analyses were performed with Apple Numbers 2009, Microsoft Excel 2008, and SAS JMP 7. Multivariate regression analyses were performed relating expression and infertility, including survival analysis and logistic regression. As quantitative fertility data were not collected for lines with high fertility, we used survival analysis to account for the right censored data, treating fertility as the time variable. In parallel, treating fertility as a binary result, we performed nominal logistic regression on the same dataset.

Results

Screen design

A total of 1057 independent GAL4 enhancer-trap lines were crossed to a stock containing both second and third chromosome inserts of a GAL4-responsive RNAi transgene that targets male-specific Fru isoforms (UAS-fru) (Manoli and Baker 2004 ), and their male progeny were screened for alterations in sexual behaviors. Given the specificity of the RNAi for fru isoforms, only fru-expressing neurons in which GAL4 is expressed should be affected by UAS-fru expression in these males (Figure 1A). To minimize background effects the GAL4 lines were in a common Berlin genetic background and the UAS-fru transgenes were introgressed into the same genetic background.

Three high-throughput behavioral assays were used to screen multiple GAL4; UAS-fru male progeny from each GAL4 line for courtship defects (Figure 1B). Males were tested for (1) reduced fertility, (2) the occurrence of chaining behavior (in which multiple males form a chain, each male courting another male), male–male courtship or aggression between pairs of males, and (3) overt courtship defects and alterations in the time to initiation of courtship (courtship latency). Following initial testing, all putative positive lines were retested in the same assays using progeny from independent crosses. To control for nonspecific effects of RNAi expression, a randomly selected subset of GAL4 lines (see below) were crossed to an RNAi targeting GFP transcripts (UAS-GFP-IR). Behavioral defects seen with UAS-fru were not observed or were at a much lower rate with UAS-GFP-IR.

Male fertility

Previous studies have demonstrated that FruM is required for fertility, and that FruM-dependent reductions in male fertility can result from overall decreases in courtship, inability to initiate or successfully complete copulation, defective innervation of male internal genitalia, as well as defects in the transfer of sperm and seminal contents (Villella ; Lee ; Manoli and Baker 2004). Thus, we postulated that perturbations of courtship behavior at multiple stages might manifest as male infertility.

To measure fertility 5–10 GAL4/UAS-fru male progeny of each GAL4 line were crossed individually with two CS virgin females and the number of offspring measured as described (see Materials and Methods). A total of 235 lines showing fertility defects were initially identified and through retesting, 19 lines were verified as sterile and 90 as having substantially reduced fertility.

Male–male interactions

Social behaviors require that an animal correctly identify a conspecific and its sex, as well as respond appropriately to both sexes in different ethological contexts. Intrasexual interactions occur both in the context of reproduction, as with courtship, as well as in aggressive interactions. As wild-type male reproductive and aggressive interactions are both dependent on (Hall 1978; Ito ; Ryner ; Villella ; Goodwin ; Anand ; Vrontou ), we assayed for abnormalities in interactions within groups of GAL4/UAS-fru males.

For male–male interaction assays, seven to eight GAL4/UAS-fru males were grouped and interactions scored as described (see Materials and Methods). Wild-type males displayed little in the way of male–male courtship interactions, likely due to inhibitory cues from males and habituation to positive cues (Miyamoto and Amrein 2008; Lacaille ). Similarly, control males showed little aggression, likely due to our chambers having a large available food surface, as limited and localized resources appear to evoke territorial/aggressive behaviors in wild type (Chen ; Nilsen ).

From 378 lines initially selected as exhibiting elevated levels of male–male interactions, retesting using the same experimental regime identified 120 lines showing chaining behavior, 53 lines showing high levels of male–male courtship (between pairs of males), and 112 lines showing aggression (File S1, File S2, and see Materials and Methods). Lines showing both male–male courtship and chaining at different times were scored as chaining lines. Interestingly, for the aggression lines, only a single male per chamber showed aggressive behavior at each observation, often controlling the entire food surface. Because males were unmarked and only periodically observed rather than constantly monitored, it was not possible to determine whether a single male was aggressive across all observations (which might suggest the establishment of dominance hierarchies) (Nilsen ; Vrontou ). At different times of observation, 33 lines showed chaining and aggressive behaviors, whereas 20 other lines showed both male–male courtship and aggression, often contemporaneously. The high incidence of lines displaying male–male interactions is notable: of the 1057 GAL4 lines tested, 288 (27%) had reproducibly elevated levels of one or more types of male–male interactions, with significant associations between these and other behavioral phenotypes (see Discussion).

Overt courtship defects

FruM function is required for all steps of courtship from its initiation through copulation and ejaculation (Villella ; Lee ; Manoli and Baker 2004). We thus assayed abnormalities in courtship latency (time to the first wing extension) and the more visually overt courtship behaviors (i.e., wing extension, wing vibration, and copulation) of individual GAL4/UAS-fru males paired with a CS female.

Initial screening of four males per GAL4 line yielded 260 lines that showed either reduced or extended courtship latencies. After retests, 86 lines with a short courtship latency (fast courtship initiation, median latency <30 sec), and 24 lines with a long latency (slow courtship initiation, median latency >4 min) were selected. Control flies expressing UAS-GFP-IR under control of randomly selected GAL4 lines had a latency of 157 ± 97 sec. To determine whether the observed courtship latency phenotypes were due to changes in the overall activity of mutant males, we assayed locomotive activity of the fast and slow lines (using a Trikinetics DAM 5). The strongest fast and slow lines did not dramatically differ in activity from GAL4/UAS-GFP-IR and CS controls (data not shown).

For ∼40 lines, GAL4/UAS-fru males exhibited defects in wing extension or vibration (on the basis of visual examination). In addition, one line exhibited defective copulatory behavior (File S3). These lines were examined in greater detail in smaller, standard courtship chambers and videotaped for 10 min or until copulation. Lines in which at least 75% of males tested demonstrated a given phenotype were selected for further analysis. Nine lines showed abnormal wing extension, in which the male rotates the wing upwards to ∼30° above the horizontal plane, instead of the usual extension parallel to horizontal. While the wing is held at this upward angle, no vibration is observed. Seventeen lines showed simultaneous extension of both wings, varying between 30° and 80° in extent, instead of the usual unilateral wing extension (File S4). Seven lines showed scissoring behavior, in which both wings are rapidly extended somewhat outward and then retracted again (File S5). Five of the lines showing scissoring also showed double-wing extension, suggesting a relationship between the two behaviors. Eight lines showed permanent wing extension, in which one or both wings were permanently extended perpendicular to the body, and not visibly vibrated during courtship. This phenotype was exhibited even before the male was presented with the female, in contrast to the previous phenotypes. Control flies expressing GFP-IR under the control of a subset of these GAL4 drivers did not show these defects.

From these three parallel behavioral screens of 1057 GAL4 lines (2833 initial assays, as not every line was tested in all three screens, and ∼6000 total assays with retests), we obtained 373 lines with robust behavioral changes, 83 of these with defects in more than one assay.

Secondary assays

We used a battery of secondary assays to examine whether each phenotypic class of GAL4 lines could be divided into subgroups on the basis of additional neurobiological and behavioral criteria. That the groups of lines with similar behaviors in our primary assays were divisible into different subgroups on the basis of secondary assays suggests that the phenotypic classes identified in the primary screens were generated by multiple mechanisms.

Mechanisms underlying male–male courtship:

We sought to determine whether chaining behavior resulted from a shared disruption or whether subsets of the chaining lines showed the same overall phenotype as the consequence of perturbations in different neural processes. We therefore examined whether chaining behavior in our lines resulted from: (1) a defect in habituation, by which naïve D. melanogaster males learn not to court other males or (2) from a change in mate preference, such that males are competitive with or preferred over females as courtship targets (Gailey ; Griffith and Ejima 2009).

We examined habituation in the lines showing chaining or high levels of male–male courtship in a modified assay in which pairs of GAL4/UAS-fru males from each line were placed together and the amount of courtship measured both immediately after they were placed together and again after they had been together for 1 hr. Although control GAL4/UAS-GFP-IR and CS males, as well as the majority of GAL4/UAS-fru males showed robust decreases in courtship over 1 hr, we found that 15 of the 109 lines assayed initially had at least a moderate level of courtship (courtship index >0.05, Figure 2A), which was unchanged or increased after 1 hr. Thus, increasing attraction toward males or increasing indiscriminate levels of courtship drive may play a role in this behavior. As pairs of mutant males were used rather than one mutant and one wild type, it is also possible that the rejection behavior of the courted male has changed rather than that of the courting male. Thus, although we term the sustained courtship a failure to habituate, other mechanisms are not excluded.

Figure 2 

Male–male habituation and changes in mate preference in males that display chaining behavior. (A) Male–male habituation. GAL4/UAS-fru males from lines that produced male-chaining behavior were paired and observed over a 1-hr period. Shown are CI averages ± SEM from minutes 2–7 (blue) and 55–60 (red) of the observation period. The lines shown are the subset with final CI > 0.05 and no significant decrease in CI. Figure S1A shows all tested lines. (B) Mate preference. GAL4/UAS-fru males from lines that produced chaining behavior were presented with a virgin Canton S female and a C155/fru male simultaneously and courtship preference index (Cpi) values are defined as Cpi = (CIf − 0.5)/(CIf + CIm), where CIf and CIm represent the fraction of time courting the female and decoy male, respectively (Villella ). Shown are Cpi values ± SEM from n = 10 assays for the 14 lines with lowest Cpi and control lines. Figure S1B shows all tested lines.

To examine whether chaining behavior reflects a change in sexual orientation (Villella ) we carried out a mate preference assay, in which a GAL4/UAS-fru test male was put together with a wild-type virgin female and a noncourting male, and the fraction of time the GAL4/UAS-fru test male spent courting each of the other two flies was measured. To create noncourting males that would not compete with the experimental male yet retained male somatic identity, we used the panneuronal C155(elav)-GAL4 driver to drive UAS-fru, thus eliminating courtship by these males.

To quantify the data from these assays, a courtship preference index based on three parameters is calculated for each line from the analysis of video recordings (see Materials and Methods). The Cpi calculates the total relative courtship directed at each target and is defined as Cpi = (CIf − 0.5)/(CIf + CIm), where CIf and CIm represent the fraction of time courting the female and decoy male, respectively (Villella ). Many lines exhibited a reduced preference for females, with ∼30% of the 46 tested lines showing similar courtship of both males and females (Figure 2B). That many chaining lines also show reduced female preference suggests that some fru neurons contribute to sexual preference by inhibiting inappropriate courtship of males. Most chaining males are willing to court males and females, but either retain a female preference or have no preference for either sex, rather than a distinct bias for courtship toward males. Thus, it appears that inhibition of FruM expression is capable of reducing female preference while maintaining significant levels of courtship, supporting the separation of the mechanisms underlying arousal and courtship drive from those mediating the specificity of courtship targets.

Phenotypes involving cholinergic fruM-expressing neurons:

To begin to elucidate the neural mechanisms underlying the behavioral phenotypes we observed, we examined the contribution of cholinergic fru neurons to fertility defects and chaining behaviors. Inhibition of neural activity in a subset of primarily cholinergic neurons has previously been associated with chaining behavior (Kitamoto 2001). A role in fertility was suggested by the finding that male-specific cholinergic neurons in the abdominal ganglion regulate the release of sperm and seminal fluids (Acebes ). Thus, we examined the effects of rescuing FruM expression in cholinergic neurons on chaining behavior and fertility. We used the GAL4 repressor GAL80 to selectively prevent the inhibition of FruM expression (by GAL4 driven UAS-fru) in the subset of neurons that express choline acetyltransferase () (Figure 3A), thus allowing us to determine the contribution of these neurons to the phenotypes we have identified. Thus we combined the cha-GAL80 transgene (Kitamoto 2001) with UAS-fru and each GAL4 line to observe the extent of rescue.

Figure 3 

Subsets of fertility defects and male–male courtship phenotypes depend on cholinergic fru neurons. (A) Expression of GAL80 in cholinergic neurons by cha-GAL80 (blue) prevents GAL4-directed, fru-mediated inhibition (green) of FruM expression (red) in cholinergic fru neurons that are labeled by the enhancer trap (white). (B) Inhibition of fru-mediated disruption of FruM expression in cholinergic neurons suppresses male sterility or fertility defects in a subset of lines producing fertility defects. Shown are mean fertility counts ± SEM for GAL4/fru (x axis) and GAL4/fru/cha-GAL80 (y axis) males for individual lines. The yellow region indicates lines with substantially increased fertility with cha-GAL80. (C) Inhibition of fru-mediated disruption of FruM expression in cholinergic neurons prevents various aspects of male–male interaction phenotypes. Shown are relative changes in severity of behavioral phenotypes in GAL4/fru/fru/cha-GAL80 males for individual phenotypic classes. The top row of graphs includes lines with decreased severity of phenotypes, and the bottom row includes lines with combinations of increases and decreases.

Of 99 retested lines originally showing fertility defects, 44 lines more than doubled their fecundity when inhibition of FruM expression was prevented with cha-GAL80 compared to simultaneous controls lacking the GAL80 (Figure 3B and data not shown). Thus perturbation of FruM expression in cha–GAL80-expressing neurons in these 44 lines substantially contributed to their decreased fertility. The failure to rescue the other infertile lines indicates that, consistent with previous studies, FruM function is also required in noncholinergic neurons for full fertility (Lee ).

Of the 123 retested lines showing high levels of chaining because of GAL4-directed fru expression, 8 lines showed a complete suppression of the aberrant behavior with cha-GAL80 and 17 showed a substantial reduction (Figure 3C). Thus, similar to the fertility rescue results, it appears that both cholinergic and noncholinergic neurons contribute to the regulation of male–male interactions.

Rescue of female fertility:

The general similarity of fru neuronal expression in males and females (Manoli ; Stockinger ) raised the question of whether the labeled neurons share homologous functions in both sexes. Masculinization of fru neurons in a female, via the use of a dominant male fru allele, is sufficient to produce sterility (Demir and Dickson 2005). Furthermore, UAS-shi-based silencing of the fru neurons in females caused virgins to exhibit mated female behavior, including reduced courtship receptivity and increased egg laying, even though unmated (Kvitsiani and Dickson 2006). While these findings show that some fru-expressing neurons are necessary for female fertility, they do not address whether these are the same subset of fru-expressing neurons that are required for male fertility. As our screen identified a subset of GAL4 lines that produced male infertility when driving UAS-fru, we asked whether the fru-expressing neurons in these lines are also important for female fertility.

We assayed whether the female infertility produced by fru expression can be suppressed by UAS-fru expression driven by the GAL4 lines that caused male fertility defects in our screen. We examined the female fertility of 75 GAL4 lines carrying UAS-fru and fru by crossing them individually to two CS males and counting their offspring (Figure 4). Eighteen of the lines showed significant rescue (P < 0.05, one-way ANOVA comparison to a negative control GAL4 line). Male siblings of the test females generally showed fertility levels similar to what was observed in previous tests of the lines without the fru, indicating that the UAS-fru is able to overcome any dominant effects of the fru allele (data not shown). However, analysis revealed no significant correlation between the level of reduction of male fertility and rescue of female fertility in this assay (Figure 4), suggesting that although some neurons may be important for fertility in both sexes, there are likely also neurons that are more important for fertility in just one sex. More detailed examinations of fru expression in males and females have also identified several sexually dimorphic patterns of cell survival and projections, which could also explain the lack of correlation (Kimura , 2008; Cachero ; Yu ).

Figure 4 

Rescue of fru-induced female infertility by GAL4 lines with reduced male fertility. Mean fertility ± SEM of GAL4; UAS-fru, fru males (x axis) is compared to females (y axis). The yellow region indicates the 18 lines (and positive controls) with substantially increased fertility in females compared with controls, highlighting that several lines did rescue female fertility. Comparison of male vs. female fertility with the same GAL4 driver does not show a clear relationship, however (R2 = 0.03).

As described, we used a large collection of GAL4 enhancer-trap lines to inhibit fru expression in subsets of neurons in males, who were then screened for courtship defects. Beyond the immediate value of these behavioral genetic studies as discussed below, our larger goal has been to identify the particular neurons affected in each GAL4 line and thus delimit the neurons that contribute to different aspects of sexual behaviors. Our general anatomical approach has been to fluorescently label fru and GAL4 neurons using separate reporters and determine the neurons in which the reporters’ expression overlapped. The neurons expressing both labels represent those affected by the RNAi against fru and thus the maximal set of neurons implicated in the observed behavioral phenotypes, as the UAS-fru has no known effect outside of fru-expressing neurons (Manoli and Baker 2004).

When we initiated studies to identify the overlaps in expression between fru and these GAL4’s we used anti-FruM antibodies and fluorescent-labeled secondary antibodies together with a UAS-driven nuclear GFP (UAS-Stinger) (Barolo ) to detect fru and GAL4 expression, respectively, in the brain and VNC, as fru expression was believed at that time to be restricted to the CNS (Ryner ; Lee ). We examined 183 lines for the concordance of nuclear GFP and FruM expression in 22 regions of the CNS using the FruM clusters described by Lee , as the basis for this CNS analysis. At the 0–24 hr posteclosion time examined, cluster 11 was not reliably labeled by our antibody and was excluded. We additionally subdivided VNC clusters 17–19 into dorsal (D) and ventral (V) subclusters due to their separation along the D–V axis.

The discovery of fru expression in extensive sets of peripheral sensory neurons (Manoli ; Stockinger ) significantly complicated initial attempts to associate the behavioral phenotypes identified in our screen with particular sets of neurons and substantially increased the scope of the project. Because of the tough cuticle surrounding peripheral sensory neurons, antibodies do not reliably penetrate to label these neurons, one of the reasons fru expression had not been identified there previously. Furthermore, although fru-P1-GAL4 beautifully labels these PNS neurons, our screen of GAL4 lines uses the same driver system, preventing us from unambiguously identifying the overlap between the two expression patterns.

This discovery of peripheral fru expression provided a potential explanation for our results up to this point, as our analyses had revealed only limited correlations of CNS expression and behavior (Figure S4 and see below). Primarily, lines with delayed courtship and/or chaining behavior tended to have broad CNS overlap with FruM. But from that dataset, we were unable to tie a behavior to expression in any particular FruM cluster.

To label fru expression independently of GAL4 in both the CNS and PNS, we used homologous recombination to insert the LexA::VP16 transcriptional activator into the fru-P1 locus (Lai and Lee 2006; Mellert ), which allowed us to follow fru expression via expression of LexAop-GFP. To detect patterns of GAL4 expression, we used an optimized nuclear DsRed, UAS-Red-Stinger (Barolo ).

As a test case for this approach, we examined the concordance of nuclear GFP and RFP expression in lines that had fertility defects. The examined set included 96 lines with fertility defects and 48 control GAL4 lines selected on the basis of normal behavior in our screen. Although we focused on lines with fertility defects, many lines had phenotypes in other assays as well, allowing a limited examination of other phenotypic classes. The results for these other assays may be biased by this selection, however.

We used the clusters described by Lee as a basis for recording expression in the CNS, additionally noting Kenyon cell expression and subdividing the VNC clusters 16–19 into dorsal and ventral subsets. Cluster 15 could not be distinguished from cluster 14 with our reporter and the two were considered together as 14. We additionally scored expression in the following regions of the PNS: the second and third antennal segments, maxillary palp, and proboscis of the head; the five tarsal segments, tibia, and femoral chordotonal organ of the foreleg; and the anal plate, lateral plate, and claspers of the external genitalia. Together these observations identified the fru neurons in which each GAL4 line is expressed.

Using this approach, we found that the vast majority of lines causing behavioral phenotypes had extensive, distributed overlap with different FruM clusters (Figure 5, Figure S2, File S6, File S7). As a result, direct comparison of the GAL4 and fru expression with observed behavioral defects has yet to indicate an individual cluster where expression is necessary or sufficient to cause fertility defects. Nonetheless, we did observe a general trend toward lines showing behavioral defects, and especially delayed courtship initiation or chaining, tending to have increased average expression compared to controls (Figure 5). Clustering by expression allowed for the visualization of the individual lines’ expression and phenotypes but again did not indicate an obvious association beyond the general trend seen in the averaged data (Figure S3).

Figure 5 

Average GAL4 overlap with fru for each courtship phenotype. Phenotypes are sorted by total average expression, with control lines having lowest expression and lines with delayed courtship initiation having highest. Kenyon cell and leg chordotonal expression were scored on a qualitative 0–3 scale for strength of expression. R1–R20, regions 1–20 of the CNS as per Lee ; Ant 2 and 3, antennal segments 2 and 3; Prob, proboscis; T1–T5, tarsal segments 1–5 of the forelimb; G_CL, genital claspers; G_LP, genital lateral plate; G_MAN anal plate; KC, Kenyon cells; LegChord, foreleg chordotonal neurons of the femur. See Figure S2 for expression scaled by cluster size, Figure S4 for similar analyses of antibody data, and File S6 for an animation stepping through each phenotype.

We used multivariate regression to look for an expression:phenotype relationship that eluded direct comparison. As fertility was not counted for vials with >50 offspring, thus right censoring the data, we first treated fertility as a time variable in multivariate parametric survival analysis, examining the contribution of each expression cluster to the likelihood of the observed fertility (Table S1). The analysis identified expression in foreleg tarsal segments 4 and 5 and anterior brain cluster 6 as significantly likely to result in lowered fertility. We separately treated fertility as a binary result and analyzed the same dataset with logistic regression. This approach also identified tarsal segment 4 and brain cluster 6 (but not tarsal segment 5) as significant drivers of infertility, supporting the result from the survival analysis. We are nonetheless hesitant to draw strong conclusions about these clusters without further experimental data to directly test their behavioral effects, as basic assumptions of regression models—especially independence of the variables—are unlikely to be met in this complex biological system, but these neural populations remain potential targets for further study.

One explanation for the difficulty of relating expression and behavior from these results is that expression in some of the larger clusters was present in almost all of the lines, and such large clusters likely represent multiple functional classes of neurons, as suggested by their divergent projection patterns and lineage relationships (Cachero ; Yu ). For example 146 out of the 153 examined lines had overlap with fru expression in region 20, the abdominal ganglion, including 54 of 58 lines with normal fertility. Similarly, 123 lines (and 20 out of 58 with normal fertility) overlapped with fru in the third antennal segment, and 117 lines (and 24 out of 58 with normal fertility) overlapped with fru in region 14 on the posterior brain. This breadth of expression precluded drawing a direct relationship between expression and phenotype, as effects of expression in other clusters cannot be ruled out with the existing dataset.

Discussion

The studies presented here use an intersectional approach to probe the functional organization of the neural substrates underlying male sexual behaviors in Drosophila. The results of our screens—considered both with respect to the phenotypes we recovered and their underlying causes, as well as the relative frequencies with which certain phenotypic classes were recovered—provide insights into the organization of the fru-specified circuitry underlying courtship behaviors in D. melanogaster.

The frequency with which abnormalities in courtship behavior were detected in our three primary screens is at first glance surprisingly high. Of the GAL4 lines successfully tested in all three primary assays, 44% (345/777) showed reproducible courtship phenotypes in at least one screen. When considered at the level of individual primary tests, 18% (482/2704) revealed reproducible behavioral defects. There are several factors that likely contribute to this high incidence of courtship phenotypes. First, the progenitor cells that give rise to the CNS are derived from most if not all segments of the body, and fru-expressing neurons comprise a widely dispersed set of ∼2–3% of CNS neurons. Second, fru neurons make up significant proportions of the primary neurons of the olfactory, gustatory, auditory, and (to a lesser degree) mechanosensory systems, which are derived from imaginal discs. Thus collectively, fru neurons provide a broad set of targets across the fly for potential overlap with the expression of GAL4 enhancer traps. In this context, it is reasonable that the screen we conducted identified a high frequency of lines in which GAL4-driven UAS-fru expression perturbed male courtship.

That the groups of lines with similar behaviors in our primary assays were divisible into subgroups on the basis of the secondary assays suggests that the phenotypic classes identified in the primary screens were generated via multiple mechanisms. Being able to make such subdivisions is important not only for specifying the functions perturbed in individual lines, but also for identifying subgroups of lines that are more likely to have a homogenous mechanism of action and thus a common set of affected fru-expressing neurons.

Comparisons of the individual classes of courtship behavior defects detected in our screens and secondary tests reveal that there are substantial differences in the relative frequencies with which different phenotypic classes were detected. As the same set of GAL4 lines was used in all screens, the different frequencies with which particular courtship behavior defects are detected reflect properties of different parts of the courtship circuitry. Following discussion of the individual assays, we discuss associations between phenotypes and propose a simple functional model for courtship initiation that builds upon our observations.

Sterility

Reduced fertility was chosen as one of the diagnostic screen phenotypes because behavioral male sterility is a phenotype associated with many mutants and mutant combinations. Further, mutants defective at different steps of courtship were associated with male sterility (Hall 1978; Villella ; Manoli and Baker 2004). We thus anticipated that male sterility would be associated with the GAL4/UAS-fru silencing of FruM expression in multiple different sets of neurons and hence be found as a relatively common phenotype in our screens. Against this background, we were surprised that only 24 of the 911 lines tested produced sterile males. This relative refractoriness of those aspects of the fru circuitry necessary for fertility to RNAi perturbations suggests the possibility that there is functional redundancy in these aspects of the circuitry or that fru neurons essential for fertility are restricted with regard to gene expression and thus only rarely overlap with GAL4-expression patterns. From an evolutionary perspective, functional redundancy in the circuitry required for fertility makes sense. It is notable that this robustness of male fertility is not likely due to simple bilateral redundancy in the nervous system, since all the GAL4-expression patterns that we observed were bilaterally symmetrical. That many more lines showed perturbed behavior than sterility indicates that the observed insensitivity of fertility to GAL4/UAS-fru manipulations is also not simply due to a general ineffectiveness of the RNAi, although it is nonetheless possible that the neurons important for fertility are particularly refractive to the RNAi. In addition, the result that many lines altered aspects or specificity of sexual behaviors but only partially impaired these males’ fecundity suggests many fru neurons may play a modulatory role in these behaviors rather than being strictly necessary for fertility.

Altered courtship latency

Our general courtship assay revealed GAL4 lines with altered courtship initiation latencies toward females. More than three times as many lines were recovered with a shortened courtship latency (95 “rapid” lines), than lines with lengthened courtship latency (25 “delayed” lines). The isolation of rapid courtship lines per se suggests that the fru-specified courtship circuitry contains components that actively inhibit the initiation of courtship. Additionally, the relatively high incidence of rapid courtship lines may suggest that the mechanisms inhibiting courtship in inappropriate contexts are less robust to disruption compared to mechanisms driving courtship, perhaps due to there being more of a cost to delaying courtship than to triggering it inappropriately. Encouragingly, one group of fru neurons in the median bundle that appear to restrain courtship has been identified (Manoli and Baker 2004). That components of the fru circuitry function to slow progression through courtship is consistent with arguments that ancestral Dipteran courtship was rapid, similar to what is seen in solitary flies like Musca and Calliphora, where the strategy for courtship appears to be to identify an object of near optimal size and speed, grab it, then determine if it is an appropriate mating target (Spieth 1974; Collett and Land 1975). Under this view, during the course of evolution from rapid mating to expanded courtship, D. melanogaster would have acquired functions in its courtship circuitry that generate the relatively complex, ordered process seen today. Such a protracted courtship may have an evolutionary advantage in allowing the male to demonstrate his virility and both the female and male to receive sufficient information to evaluate a potential mate. It remains to be determined whether lines that demonstrate rapid courtship onset also display specific deficits in discriminating appropriate courtship targets. It should be noted, however, that the previously characterized group of fru-expressing neurons that function to restrain courtship (Manoli and Baker 2004) do not have a fru-dependent role in the discrimination of females from males as appropriate courtship targets.

The most common phenotype detected in our screens was an elevated level of male–male interactions: 26% (232/908) of the GAL4 lines tested had reproducibly elevated levels of one or more types of male–male interaction. There are several elements of note with respect to the high incidence of lines producing a male–male interaction phenotype.

First, male–male interactions encompasses at least two classes of male behavior that likely have different etiologies. The category includes 117 lines showing aggressive male–male interactions, 121 lines showing chaining behavior (involving three or more males), and 49 lines showing high levels of male–male courtship (between pairs of males). It is currently unclear whether the latter two categories are distinct phenotypic classes or just differ in the level of male–male courtship, although some lines showed very high levels of male–male courtship without chaining, perhaps due to differences in rejection behavior.

Second, even with this subdivision of male–male interacting lines the incidence of lines showing male–male courtship is quite high. Contributing to the high incidence of male–male courtship lines in our screen is the fact that this phenotype itself can be potentially further subdivided into classes of lines having different etiologies. A priori these include: (1) a change in sexual preference so that males are the preferred courtship targets, (2) an inability to distinguish females from males as appropriate courtship targets resulting in both being courted, (3) a failure of males to reduce initial courtship of other males via habituation (i.e., learning not to court males; Gailey ), or (4) an increase in sensitivity to general conspecific cues such that inhibitory cues that typically prevent male–male courtship are insufficient to do so. For example, there have been a number of reports of flies with altered gustatory perception showing male–male chaining behavior (Lacaille ; Miyamoto and Amrein 2008; Moon ). Our tests to investigate these possibilities revealed that in some lines increased male–male courtship may be attributed to a lack of male habituation or increased sensitivity to general cues. Many other lines showed reductions in their preference of females over males as courtship targets. However, none of the lines we examined showed a clear preference for males over females.

Cholinergic fru-expressing neurons

For the GAL4 lines that produced chaining or fertility defects when driving UAS-fru we extended our intersectional strategy to examine the contribution of cholinergic fru-expressing neurons to these chaining and fertility defects. In particular, we used Cha-GAL80 to selectively restore FruM expression in the subset of GAL4/UAS-fru-expressing neurons that also likely express choline acetyltransferase () (Figure 3A). The expression of Cha-GAL80 at least doubled the fecundity of ∼40% (44/99) of lines with fertility defects, whereas it substantially or completely suppressed chaining in ∼20% (25/123) of lines originally showing chaining. As cholinergic neurons are found in the CNS and also make up most if not all, olfactory, gustatory, auditory, and chordotonal sensory neurons (Salvaterra and Kitamoto 2001), the most informative aspect of these results are the findings of GAL4 lines with phenotypes that are not suppressed by the restoration of FruM function in cholinergic neurons, as these likely identify cases where fru CNS expression is important for the observed phenotype. It is worth noting that the nine chaining lines in which restoration of FruM function produced the greatest suppression of chaining (and 20 of the top 25 such lines) showed normal behavior in the original courtship and fertility assays and retests. This suggests that FruM function in specific cholinergic neurons may be necessary to prevent chaining or the courtship of other males but that these neurons do not play a role in the proper performance of other courtship behaviors. Thus it appears that some elements underlying the discrimination of a courtship target are distinct from those subserving arousal or the drive to display specific sexual behaviors and appear to be mediated by separable subsets of fru-expressing neurons. That FruM expression in cholinergic neurons is important for selection of a courtship target is in part not surprising, as chemosensory cues contribute significantly to mate preference (Stockinger ; Kurtovic ; Datta ; Krstic ) and most if not all chemosensory neurons are cholinergic (Salvaterra and Kitamoto 2001). Similarly, that fertility in part was restored by rescue of expression in cholinergic neurons is consistent with the importance of cholinergic and noncholinergic neurons in copulation (Lee ; Acebes ; Acebes ).

Behavioral associations

In one sense, our screens can be viewed as having isolated an allelic series of fru mutants, in each of which only some subset of fru-expressing neurons are mutant in phenotype. By examining the behavior of many lines with altered patterns of FruM expression, we can begin to gain insights into the functional units governing the interactions of male flies with other flies. For example, we observed that specific behavioral phenotypes tended to co-occur among sets of enhancer-trap lines (Figure 6), which could indicate that reduced FruM expression in neurons shared by the lines produces both behavioral outcomes. Alternatively, some enhancer traps may be expressed in several groups of neurons that independently give rise to subsets of the correlated phenotypes. Of the 343 lines showing robust behavioral changes, 84 lines showed changes for more than one type of behavior, providing a substantial group for comparison. The two most striking correlations we found (1) that lines with longer courtship latencies tended to show chaining behavior, whereas (2) lines with shorter latencies tended to show aggression, are discussed below.

Figure 6 

Associations between behavioral assays. (A) If chaining, aggression, and courtship behaviors were independent, the indicated number of GAL4 lines would show each phenotype or combination thereof. (B) The observed associations significantly (P values indicate Pearson’s chi square significance) deviate from the expected values. In particular we observed an increase in lines with both long courtship latency and chaining behavior, as well as lines with both short courtship latency and aggressive behavior.

Delayed courtship associates with chaining:

We initially hypothesized that FruM likely functions in some neurons to repress courtship, such that increased male–male courtship or chaining resulting from the suppression of FruM function might also be associated with rapid initiation of female courtship. We found in fact the opposite: a positive correlation between delayed courtship initiation toward females and chaining behavior. Whereas a minority of lines (120 out of 835, or 14.4%) showed chaining behavior, over half of the lines (14 of 24, or 58.3%) with slow courtship initiation also showed chaining (P < 0.001, Pearson’s chi square test; Figure 6). No lines showed delayed courtship and high male–male courtship without also exhibiting chaining behavior.

Four differences between the male–male interaction assay and the courtship assay may contribute to this association: (1) the target of courtship being male or female, thus changing the nature of the stimulus; (2) the number of flies in the chamber (two vs. eight), thus changing the overall level of general conspecific cues in the environment; (3) the time together being minutes vs. days, thus allowing for increased stimulation after longer, sustained interactions; and (4) the presence of food in the chamber, thus altering the environmental cues and perhaps arousal. Our examination of male habituation in the chaining lines also addresses whether lengthened exposure could stimulate courtship of males. Our results suggest that this may be the case in a subset of lines, as 15 of the 109 tested lines showed persistent or increasing male–male courtship over an hour. In general however, the lines showing both delayed courtship and chaining did not show defects in habituation. Similarly, the majority of lines that showed chaining behavior did not have a significant change in sexual orientation in the mate preference assay but, rather, appeared to have an overall decrease in the sex specificity of their courtship. Thus, while we cannot rule out a stimulatory role for the presence of food or overall increases in social behaviors or arousal due to cues from conspecifics, we hypothesize that these lines have a decreased sensitivity to stimulation, which results in courtship only in the presence of multiple conspecifics, rather than when a single female or male was present.

Rapid courtship initiation associates with aggression:

Lines showing rapid courtship initiation of females were more likely to show aggressive behavior toward males (20 out of 86 lines or 23.2%) than expected, on the basis of the fraction of total lines showing aggression (112 of 835 lines, 13.4%; P < 0.01, Pearson’s chi square test; Figure 6). A supportive, but borderline significant trend was seen with lines with delayed courtship, none of which showed aggression (0 of 24, P = 0.0504, Pearson’s chi square). Thus, it appears that the inhibition of FruM expression in regions sufficient to cause rapid courtship initiation toward females can be sufficient to increase other sexual behaviors in different contexts. This suggests that common mechanisms may control the stimulus thresholds necessary to elicit appropriate sexual behaviors in specific contexts, while others mediate the discrimination necessary to specify which behaviors occur in distinct ethological situations.

Mechanisms underlying the specificity and sensitivity of courtship initiation:

On the basis of the assays described, the relative frequencies of lines producing distinct phenotypic classes, and the correlations described between different behaviors, we have developed a relatively simple functional model for the mechanisms underlying the initiation of sexual behaviors (Figure 7). Although other models can be arrived at from our data, this one is concise and has several attractive features that make it integrate well with our knowledge of fru circuitry as well as observations made by many groups regarding various aspects of Drosophila sexual behaviors. Notably, the model does not distinguish between different sensory modalities. We hope that it will prove useful in interpreting our data and understanding functional elements from expression studies (see File S8, Supplemental Model Discussion).

Figure 7 

Functional model of mechanisms mediating sex specificity and stimulation during courtship initiation. (A) In a wild-type male, arousal cues common to both sexes act via FruM-independent mechanisms to increase the likelihood of either aggressive or courtship behaviors, but are insufficient to elicit behavioral programs themselves (black). Female-specific cues act and are enhanced via FruM-dependent mechanisms to drive the initiation of courtship and also likely maintain courtship drive until successful mating (blue). Furthermore, distinct mechanisms modulated by FruM regulate the threshold of stimulation necessary to elicit courtship. Thus, we propose that FruM functions to allow the detection of female-specific cues to increase the strength of signaling downstream of this perception to drive courtship and to modulate the activation threshold for initiation of courtship. The detection of male-specific cues via FruM-dependent mechanisms additionally provides stimulus toward both the initiation of aggression and the inhibition of courtship initiation (green). FruM-dependent mechanisms distinct from the detection of and response to male-specific cues act to modulate the threshold of stimulation necessary to elicit aggressive or courtship behaviors when such cues are perceived. Thus, in addition to regulating the processing of sex-specific cues, distinct FruM-dependent mechanisms modulate the activation thresholds for specific behavioral programs, enabling the alteration of thresholds for sexual behaviors without affecting the specific contexts within which these behaviors occur. (B) In males lacking FruM function, FruM-dependent mechanisms are either eliminated or exist in a default state (red). In the absence of FruM function, these pathways may represent female-like mechanisms of stimulus detection or processing.

General considerations:

We have presented an unbiased approach to attempt to discern correlations between specific populations of fru neurons and distinct aspects of sexually dimorphic behaviors. However, it has proven challenging to correlate common sets of fru neurons with specific aspects of sexual behavior. Factors likely contributing to these difficulties include: the nature of the GAL4 enhancer-trap lines, functional diversity between neighboring populations of fru neurons, and redundancy in neural circuitry underlying sexual behaviors.

Functional considerations:

Several factors have directed our thinking as to how best to use this data to identify the fru-expressing neurons responsible for the various courtship phenotypes we observed. Our initial observations with the FruM antibody, that GAL4 expression commonly overlapped many FruM clusters per line, agreed with our expectation that some overlapping neurons may not be involved in the specific behavioral phenotype(s) we observed. There are several ways in which we can envisage expressing UAS-fru in a subset of fru neurons might not elicit a behavioral phenotype in our assays. First, a set of fru neurons might be functionally redundantly with some other subset of fru neurons. Second, subtle phenotypes in sexual behaviors may not have been detected in our assays. Third, some fru neurons may be involved in conveying information from the circuitry mediating sexual behaviors to parts of the nervous system involved in other behavioral processes. Taken together these thoughts suggest two approaches that might be used to associate particular behavioral phenotypes with individual groups of fru neurons.

First, the neurons most likely to contribute to a given behavioral phenotype are those that are common to multiple drivers that produce that behavioral deficit. Thus, to narrow down the set of neurons to those most likely responsible for an observed behavior, we took advantage of the many lines that showed similar phenotypes. We assume that the neurons not responsible for the phenotype will be labeled in some but not other GAL4 lines producing the same phenotype, whereas the essential neurons will be more consistently labeled in multiple GAL4 lines producing the same phenotype. By examining the neurons affected in each such line and comparing them between the lines, one expects to find the shared subset responsible for the observed phenotype.

However, our studies revealed some important challenges to this initial theoretical framework. One is that, as we observed from subdividing the initial behavioral classes by our secondary assays, similar phenotypes can have multiple underlying causes. Thus, multiple independent sets of neurons may be able to cause these phenotypes, preventing isolation of a single group of neurons responsible for a given phenotype. We hope to have reduced this concern by increasing our behavioral resolution with the secondary assays, but likely have not eliminated it. Thus, our analysis must allow for multiple foci for a phenotype rather than seeking a single neural population. Second, as described above, while covering several ethological contexts, our assays by no means exhaust the myriad of conditions in which sexual dimorphisms in behaviors may occur. Thus, neuron populations that overlap with regard to phenotypic class in our assays may do so secondary to deficits in pathways that function in contexts outside of those tested, thus preventing their separation.

An alternative to the approach just outlined picks out the GAL4 line(s) with a given phenotype that have the most restricted overlap with fru expression. Fewer potential neurons responsible for a phenotype are naturally more straightforward to interpret. Both of these approaches have proved useful in delimiting groups of neurons important for particular behaviors (O'Dell ; Broughton ; Manoli and Baker 2004; Stockinger ; Luan ; Chan and Kravitz 2007; Kimura ; Luo ; Häsemeyer ; Yang ).

Neuroanatomical considerations:

Perhaps unsurprisingly, screening for behavioral defects without regard to expression patterns of their drivers has yielded many GAL4 lines with interesting and informative behaviors but with complex expression patterns. The challenge then is to design constraints upon the manipulators either into the initial screen or efficient secondary tests. The primary constraint upon which we relied when we began the screen in 2003 was the restriction of GAL4/UAS-fru effects to fru-expressing neurons. Discovery of fru expression in the peripheral nervous system substantially expanded the scope of this set of neurons and required new tools to efficiently label fru expression (Manoli ; Stockinger ; Mellert ). Additionally, we did not anticipate the breadth of typical GAL4-driver expression across the nervous system, resulting in simultaneously altering FruM levels in diverse sets of neurons. Finally, while the initial neuroanatomical analysis of fru expression revealed likely lineage-related populations based on the proximity of nuclei, recent analyses of the morphology of fru neurons suggests that even developmentally related populations appear functionally diverse on the basis of their projection patterns and presumed connectivity (Datta ; Kimura ; Cachero ; Yu ; and data not shown). These observations suggest more refined neuroanatomical parameters are likely also necessary to functionally subdivide populations of fru neurons and establish behavioral correlations.

Conclusion

We anticipate that the set of lines we have isolated and the richness of behavioral data based on phenotypic classes that we have generated presents a resource that will advance our understanding not only of the neural substrates of sexual behavior in the fly, but perhaps also the fundamental principles that underlie the specification and function of circuitry that subserves complex behaviors in general. Once identified, the characterization and manipulation of these distinct sets of neurons will allow us to begin to understand the representation of information relevant to specific behavioral states within the nervous system and how such information is processed to generate and execute innate behavioral programs.

Finally, it is also worth noting that the formal logic of our approach—carrying out a large-scale screen for genetic variants that have phenotypes perturbing a process of interest, validating those phenotypes by retests, and then sorting them into classes and subclasses on the basis of subsequent secondary assays—is the basic paradigm that has been used across all model organisms to dissect developmental and physiological processes of interest. Our experiments establish that this approach can be extended to the analysis of an innate behavior and provide guides for designing similar studies of other behaviors and their underlying genetic and neural substrates.