Shared Visual Attention and Memory Systems in the Drosophila Brain

Background Selective attention and memory seem to be related in human experience. This appears to be the case as well in simple model organisms such as the fly Drosophila melanogaster. Mutations affecting olfactory and visual memory formation in Drosophila, such as in dunce and rutabaga, also affect short-term visual processes relevant to selective attention. In particular, increased optomotor responsiveness appears to be predictive of visual attention defects in these mutants. Methodology/Principal Findings To further explore the possible overlap between memory and visual attention systems in the fly brain, we screened a panel of 36 olfactory long term memory (LTM) mutants for visual attention-like defects using an optomotor maze paradigm. Three of these mutants yielded high dunce-like optomotor responsiveness. We characterized these three strains by examining their visual distraction in the maze, their visual learning capabilities, and their brain activity responses to visual novelty. We found that one of these mutants, D0067, was almost completely identical to dunce1 for all measures, while another, D0264, was more like wild type. Exploiting the fact that the LTM mutants are also Gal4 enhancer traps, we explored the sufficiency for the cells subserved by these elements to rescue dunce attention defects and found overlap at the level of the mushroom bodies. Finally, we demonstrate that control of synaptic function in these Gal4 expressing cells specifically modulates a 20–30 Hz local field potential associated with attention-like effects in the fly brain. Conclusions/Significance Our study uncovers genetic and neuroanatomical systems in the fly brain affecting both visual attention and odor memory phenotypes. A common component to these systems appears to be the mushroom bodies, brain structures which have been traditionally associated with odor learning but which we propose might be also involved in generating oscillatory brain activity required for attention-like processes in the fly brain.


Introduction
We have previously shown that attention-like processes can be studied in Drosophila using both electrophysiological and behavioral approaches [1,2], and that mutations in learning and memory pathways modulate attention-like responses in flies [3]. Mutations in dunce (a cyclic AMP phosphodiesterase) and rutabaga (an adenylyl cyclase), which compromise memory formation via the cyclic AMP pathway [4], also produce defects in short-term processes relevant to selective attention. These defects include decreased responsiveness to visual novelty and decreased distractibility by competing visual stimuli. Interestingly, such attention defects were associated with increased behavioral responsiveness to moving gratings in an optomotor choice maze, where dunce and rutabaga mutants performed twice as strongly as wild-type flies. Why should learning and memory mutants display such strong visual reflexes?
Our interpretation of the stronger performance of these mutants in the optomotor choice maze is that these animals are less able to suppress prepotent, or reflexive, responses (to the moving grating in the maze). Thus, their attention-like defects (as assessed electrophysiologically [3]) are manifested behaviorally by a reduced ability to suppress a reflex such as the optomotor response. We suggested that such defective suppression abilities were a possible source of dunce mutant's learning deficits in olfactory [5] and/or visual [6] assays, where ongoing stimulus suppression dynamics may play a role in the normal acquisition or retrieval of memories. The reduced ability of dunce mutants to suppress prepotent reflexes is also evident in another visual paradigm, aversive phototaxic suppression, where wild-type flies learn to avoid a lit chamber associated with quinine whereas dunce fail to suppress their phototaxic reflex over time [7].
We therefore used optomotor responsiveness in the maze as a means of uncovering other pathways that might compromise visual attention in flies, with the prediction that strong optomotor reflexes in other mutants might be associated with distinct visual attention defects. Rather than performing a large-scale blind screen, we decided to focus on a panel of mutant strains found to be defective in olfactory long term memory (LTM) [8]. Our focus on LTM or plasticity genes stems from an earlier result showing that dunce attention defects could only be rescued by expression of the wild-type protein throughout development [3]. We hypothesize that LTM genes represent possible downstream targets of cyclic AMP signaling (i.e., in the same cells), and thus might be involved in establishing the neuronal architecture required for attention-like processes in the adult during brain development. We predicted that, like rutabaga 2080 and dunce 1 , some of these olfactory LTM mutants might also be characterized by distinct attentionlike defects in our behavioral and electrophysiological paradigms because LTM pathways are likely to modulate attention. Crucially, the genetic lesions in many of these LTM strains are caused by Gal4 constructs stably inserted into the genome [8]. These elements therefore also function as enhancer traps to allow for a complete examination of the neural architecture possibly compromised by the mutations. By examining the neuroanatomy, we could therefore uncover possible common substrates of visual attention and olfactory memory in strains where both phenotypes are affected. The Gal4 elements would then also serve as molecular 'handles' for active manipulation of circuits affecting attention and memory [9].
Among 36 LTM mutants, we found three strains with optomotor responsiveness as high as dunce mutants, namely D0264, D0067, and D0177 (named Rafael, Norka, and Toi, respectively, in the publication where they were first described [8]). Although these three strains produced similar levels of high optomotor responsiveness, they displayed strikingly different characteristics for attention-like processes, with D0067 most similar to dunce in both behavioral and electrophysiological paradigms. We also found all three strains to be defective in some aspects of visual learning. By using the inserted elements as Gal4 complements to either rescue dunce defects or silence synaptic activity, we found that central brain structures, namely the mushroom bodies, may be key to modulating visual attention phenotypes in Drosophila.

Optomotor Behavior
Mutations in the memory genes dunce and rutabaga increase visual responsiveness to moving gratings in an optomotor maze [3]. We hypothesized that defective attention-like processes in other genes associated with memory formation might also be uncovered by examining the behavior of other olfactory memory mutants in the optomotor maze. We therefore tested 36 long-term memory (LTM) mutants for visual responsiveness to moving gratings in our optomotor maze paradigm ( Figure 1A). These mutant strains were first isolated in a screen for defective LTM in an olfactory paradigm [8]. As shown in Figure 1B, the 36 mutants display a wide range of optomotor responsiveness levels, from zero responsiveness to very strong optomotor responses (as observed in dunce and rutabaga mutants) [3]. We focused on three olfactory LTM variants displaying significantly greater optomotor responsiveness than wild-type flies: D0264, D0067, and D0177 ( Figure 1B, black histogram bars). By selecting LTM mutants with dunce-like responsiveness in the optomotor maze, we attempted to uncover overlapping systems in the brain relevant to both memory formation and attention. Our investigation of these selected LTM variants progressed as follows: visual distraction effects in the maze, visual learning behavior, and neural correlates of visual attention. We then use the elements as Gal4 drivers to investigate their capacity for rescuing dunce attention defects and for active control of these neurons to modulate attention-like phenotypes. Our results are thus divided in two broad sections: first, an analysis of the LTM mutants for visual attention and memory phenotypes, and second an analysis of possible Gal4 networks affecting visual attention in flies.
To examine how attention-like processes might vary across the LTM mutants, we first tested the distractibility of the flies. Flies progressing through the optomotor maze can see objects surrounding the maze, and the position of these objects influences their optomotor response [1] (Figure 2A). For example, for wildtype flies, a black bar positioned on the opposite side of optomotor flow abolishes the flies' response to the moving grating, and this effect has been invoked as a measure of distractibility [1,3](GB-in Figure 2B). dunce mutants are unaffected by such static distracters, suggesting a defect in attention in these flies [3]. D0067 and D0264, which responded as strongly as dunce in the optomotor maze, displayed different behaviors when presented with distracters. Whereas D0067 was similar to dunce 1 in not being distracted by the competing visual stimulus, D0264, also a strong responder to the optomotor stimulus, was somewhat distractible, though not as strongly as wild type (a 37% attenuation versus 78% in wild type) ( Figure 2C). Optomotor responses in D0177 were also significantly decreased when the distracter was present (44% decrease, Figure 2C). Differences among these strains suggest that strong optomotor performance does not necessarily imply defects in perceiving static visual stimuli in competition with a moving grating. To compare this distraction phenotype in other LTM mutants, we randomly chose and tested three strains with wildtype optomotor response levels, namely A0023, D0940, and D0851 (see Figure 1B). We found that two of these LTM strains with normal optomotor responsiveness were also unaffected by the visual distracter, like dunce 1 and D0067 ( Figure 2D). This further supports our observation that optomotor performance and visual distraction effects are separable.

Visual Learning
Even though D0264, D0067, and D0177 were isolated for defective LTM in an olfactory paradigm, they display distinct behaviors in the optomotor maze, suggesting broader defects in short-term behavioral processes. These strains have never been tested for visual learning paradigms, partially because the best visual learning paradigms for Drosophila are single animal assays requiring flight (the advantages and disadvantages of the tethered flight paradigm are reviewed in [10]). We tested these strains for visual learning using a modified version of a walking paradigm that pairs visual stimuli (the conditioned stimuli) with shaking (the unconditioned stimulus, see Methods and Supporting Information Movie S1). Wild-type flies learn to avoid visuals associated with shaking, whereas dunce 1 fails to show learning [11]; Figure S1). Our three high optomotor performers (D0264, D0067, and D0177) all failed to show learning in this visual paradigm ( Figure S1). The fact that olfactory learning is normal in these mutants (only longterm memory is compromised) [8] highlights a difference between visual and olfactory paradigms with regard to classical conditioning.
The above learning paradigm reveals a performance index after training (see Methods), and thus may not best uncover subtle differences among mutants. Indeed, all three olfactory LTM mutants failed to learn, although differences in performance among these strains were evident during training ( Figure S1). Furthermore, different sensitivities to the unconditioned stimulus (shaking) might also influence the learning phenotype, as might other aversive cues accumulating in the tubes during training, such as CO 2 [12]. To further address visual learning through examining sensitization, we modified our optomotor maze paradigm: we ran flies through the maze multiple times to see if responses were affected by repeated exposure to the maze and visuals (see Methods). Interestingly, wild-type flies displayed stronger optomotor responses during repeated maze runs when they were performed in immediate succession ( Figure S2), resulting in optomotor scores as high as dunce 1 mutants by the third exposure to the optomotor maze. Re-running dunce flies in the maze slightly increased responses though the difference was not significant (1.6360.29 versus 1.2760.17 for baseline). With a strong performer such as dunce 1 one might suspect a ceiling effect for the behavior, where even higher responses might be unlikely in this 8-level choice design. Instead, we found that D0264 optomotor performance (which is already as high as dunce at baseline) was significantly increased by the third exposure to the maze, to 2.360.13. Neither D0067 nor D0177 displayed any such significant increases in optomotor responsiveness following repeated exposure ( Figure S2). These experiments revealed two valuable points: first, some mutants perform even better than dunce 1 in the maze under certain ''learning'' conditions (there is no ceiling at an optomotor score of ,1.2), and second, a strong optomotor performance does not always imply a lack of plasticity.  [1]. B. Data for wild-type flies. B, response (6s.e.m.) to the static bar alone; G, response to the moving grating alone; GB+, response when the bar is on the same side as the direction of the moving grating; GB-, response when the bar is on the side opposite to the direction of grating movement. The bar on the side opposite to grating movement (black histogram) significantly reduces the optomotor response. C. Distractibility profile for the three strongest optomotor performers, D0177, D0067, and D0264. D. Distractibility profile for three average optomotor performers, A0023, D0940, and D0851. doi:10.1371/journal.pone.0005989.g002 D0264, with a strong optomotor response, can improve its performance with repeated exposure. Interestingly, D0067, which was not distracted by a competing visual object ( Figure 2C) also showed no sensitization in the maze, making it most like dunce 1 .
Taken together, our behavioral experiments show that three olfactory LTM mutants, as well as dunce 1 , are significantly different than wild type in a variety of visual paradigms. This is consistent with the mutants having more generalized ''cognitive'' defects, which we now investigate further using electrophysiology.

Neural correlates
There are limits to how much behavioral assays can reveal about attention-like processes in animals. In contrast, electrophysiology can provide detailed insight with regard to the temporal dynamics of selection and suppression of brain responses to competing stimuli. We presented the mutants D0264, D0067, and D0177 with competing visual choices (a cross versus a square) while we recorded local field potentials (LFPs) from their brains using a previously reported paradigm to study visual selective attention in flies ( Figure 3A, and see Methods). In previous studies, LFP responses to visual stimuli using this paradigm revealed that certain frequency domains within the LFP, centered on the 20-30 Hz range, were correlated with visual salience [2]. In this paradigm, wild-type brains display increased LFP activity for novel objects while, at the same time, exhibited suppressed LFP activity for competing non-novel objects, following 100 s of visual training [3]. dunce mutants fail to show any such LFP selection/suppression in the 20-30 Hz range, but, instead, show a smaller effect only at a lower (,10-20 Hz) frequency range [3]. We first investigated baseline LFP responses to individual objects (a cross or a square) in our selected LTM mutants. The three strongest optomotor performers (D0264, D0067, and D0177) all revealed attenuated brain responses when exposed to the visual objects alone compared to wild-type flies ( Figure 3B). To compare these responses to another LTM mutant, we conducted the same experiments on A0023, which had wild-type optomotor respon- siveness and yet was not distractible ( Figure 2D). Surprisingly, A0023 also showed attenuated brain responsiveness compared to wild type ( Figure 3B). Thus, the four LTM mutants tested all appear to respond less strongly to visual objects (like dunce 1 [3]) regardless of their differences in visual behavior.
LFP responsiveness to novelty was investigated by presenting distinct objects together following a training period during which flies were exposed to identical objects ( Figure 3C). The three mutants D0264, D0067, and D0177 revealed different responses to novel objects at the level of brain LFP activity, and none exhibited wild-type responses ( Figure 3D, and see Figure S3 for baseline responses prior to novelty). D0264 responded strongly in the 10-20 Hz range, with some responsiveness in the 20-30 Hz range. D0067 responded weakly in the 10-20 Hz range only, resembling dunce 1 LFP effects. D0177 responded best in the 30-40 Hz range. The weakest effect was thus found in the mutant that proved least distractible in our behavioral assay, D0067. To test if the distraction phenotype ( Figure 2) might predict responsiveness to novelty in our brain recording preparation, we tested an additional LTM mutant, A0023, which was also found to be unaffected by a distracter in the behavioral paradigm. Indeed, LFP responsiveness to novelty in A0023 was also defective, most resembling D0067 by the absence of a strong frequency component in the LFP response ( Figure 3D). Thus, LTM mutations appear to influence the frequency of LFP oscillations in the fly brain as well as visual behavior.
The selective response to novelty depends on the amount of prior exposure to the non-novel object [3]. Wild-type flies require at least 50 s of visual training (or exposure) to identical squares to produce a selective response to a novel cross and a suppression of a response to the competing non-novel square. 25 s fails to generate such a response in wild type. To confirm the LFP frequency effects found in D0264, D0067, and D0177, we repeated the novelty experiments with 50 or 25 s of training. None responded significantly after 25 s training, while 50 s of training again yielded significant responses in the 10-20 Hz range for D0264 and in the 30-40 Hz range for D0177 ( Figure S4). The effect on LFP frequency ranges in these strains therefore appears to be robust, whereas, in D0067 (like dunce), the effect is weak.

Spatiotemporal rescue of dunce
We made use of the Gal4/UAS system [13] to explore the neuroanatomy of brain circuits which could be subserving the attention-like phenotypes described above. D0264, D0067, and D0177 are homozygous Gal4 inserts on standardized genetic backgrounds (see Methods), and as such can be used to explore the function of circuits impacted by the enhancer trap in addition to the mutagenic effects caused by the transposable element. We have previously shown that dunce 1 attention defects can be rescued by expressing wild-type dunce in the brain throughout development [3]. dunce-like optomotor phenotypes in D0264, D0067, and D0177 suggest that these mutations might compromise brain circuits relevant to dunce function (i.e., they may represent downstream plasticity genes expressed in the same cells as dunce). We thus sought to examine the brain regions required to rescue dunce 1 defects by expressing wild-type dunce in the D0264, D0067, and D0177 cells, in a mutant dunce 1 background. It is important to emphasize that in this heterozygous context, the LTM mutants become simple Gal4 drivers; the three inserts are recessive and have no abnormal phenotype as heterozygotes over wild type (data not shown). In the current UAS/Gal4 scenario, we are therefore testing effects on the brain cells controlled by these insertions, not the effects of the insertions themselves. Whether these Gal4expressing cells may be involved in olfactory learning and memory has not been formally investigated. However, the presence of mushroom body labeling for many of these Gal4 insertions [8] (and see Figure 4) suggests olfactory memory circuits may be involved.
Expressing wild-type dunce in the Gal4 cells defined by D0264 ( Figure 4A) rescued dunce 1 LFP defects: the dunce 1 ; UAS-dunce/ D0264-Gal4 strain responded strongly to novelty with a characteristic 20-30 Hz selection/suppression profile following 100 s of visual training ( Figure 5A, and see Figure S5 for baseline responses). In contrast, the dunce 1 ; UAS-dunce/D0177-Gal4 strain failed to rescue dunce 1 defects in this paradigm: we saw a small and insignificant response only in the 10-20 Hz range, like dunce mutants ( Figure 5B). Rescue experiments for D0067-Gal4 were not possible as the rescue strain dunce 1 ; D0067-Gal4/UAS-dunce was larval-lethal, possibly due to widespread expression in the brain beyond the mushroom bodies ( Figure 4B). The D0177 driver appears to not label in the Kenyon cell processes of the mushroom bodies ( Figure 4C), so it is perhaps not surprising that this enhancer trap fails to rescue dunce 1 . The enhancer traps D0067 and D0264 express in neurons throughout the Drosophila brain, with some overlap in the a and b lobes of the mushroom bodies ( Figure 4A, B). Most striking in D0264 is strong expression in the pars interecerebralis, with projections via the median bundle to the suboesophageal ganglion. A similar pattern of expression (minus the mushroom bodies) has been found in the c767-Gal4 driver for neurons controlling EGFR signaling and sleep in Drosophila [14].
To investigate more specifically whether the mushroom bodies might be involved in dunce-mediated visual attention responses, we used a Gal4 driver labeling most of the mushroom bodies, c309-Gal4 [15] ( Figure 4D). Expressing the wild-type dunce gene in the dunce 1 ; UAS-dunce/c309-Gal4 strain rescued LFP responsiveness to novelty ( Figure 5C). The involvement of the mushroom body was further supported by significant novelty detection in the strain dunce 1 ; UAS-dunce/A0023-Gal4 strain ( Figure 5D). In addition to widespread expression throughout the brain, A0023-Gal4 appears to also label a more restricted subset of cells in the a and b lobes of the mushroom bodies compared to D0264-Gal4 ( Figure 4E). However, dunce 1 rescue with A0023-Gal4 was only marginally significant (P = 0.05) compared to c309-Gal4 and D0264-Gal4 rescue. Consistent with this, optomotor behavior was rescued in dunce 1 ; UAS-dunce/c309-Gal4 (OI = 0.5560.14) but not in dunce 1 ; UAS-dunce/A0023-Gal4 (OI = 0.9560.25) or dunce 1 ; UAS-dunce/ D0177-Gal4 (OI = 1.0560.18). dunce 1 ; UAS-dunce/D0264-Gal4 was not testable for behavior in the maze as animals refused to enter.
Having spatially narrowed dunce effects on attention-like phenotypes to potentially the a and b lobes of the mushroom bodies, we next investigated the temporal window required for dunce function in our selective attention paradigm. We have previously shown that dunce was required throughout development in order to rescue attention-like defects in the adult by inducing wild-type dunce in a mutant background in larvae (induction in the adult did not rescue the defects [3]). We now asked whether attention-like defects could be rescued by expressing wild-type dunce only in the late pupal stage of development. To test this, we injected dunce 1 ; UAS-dunce/gsg-301 pupae with Mifepristone to induce wild-type dunce expression (see Methods) and then tested adults for brain responses to visual novelty. We found that expressing dunce at a late pupal stage was sufficient to rescue attention-like phenotypes in the brain ( Figure 6A, and see Figure  S6 for baseline responses). In contrast, injection of only the vehicle (DMSO) into late-stage pupae did not rescue dunce 1 attention defects ( Figure 6B). The dunce protein is thus required at earliest during pupal brain development for enabling subsequent attention-like responses in the adult. Taken together with our spatial rescue experiments, this result suggests that mushroom body Gal4 expression patterns in the pupal or early adult brain may be more relevant to understanding dunce rescue effects than patterns in the adult brain.

Active control of attention/memory neurons
If select mushroom body circuits rescue dunce 1 attention defects, then silencing synaptic output from these structures should modulate brain activity correlates of attention. Neuroanatomical localization of attention-like behavior was further explored by transiently silencing synaptic activity via the D0264, D0067, and D0177 Gal4 drivers, complemented by experiments more restricted to the mushroom bodies using other Gal4 drivers. LFP oscillations such as the 20-30 Hz response to visual salience are most likely generated by synchronization of synaptic potentials. This implies that ''upstream'' neurons are synchronizing their output in order to generate summed electrical effects near the recording site. Thus, modulating LFP activity by silencing distinct neuronal circuits should narrow the sources of neuronal synchronization resulting in the 20-30 Hz response to novelty. To attenuate synaptic output we expressed the shibire TS gene in the three Gal4 networks defined by D0264, D0067, and D0117, as well as in A0023, c305a, and c739, which appear to label subdivisions of the mushroom bodies, along with diverse other structures [16] (and see Figure S7). UAS-shibire TS expresses a thermolabile form of dynamin, involved in vesicle recycling at synapses, and causing rapid neurotransmitter depletion at the restrictive temperature (.30uC) [17]. To test the shibire TS effects in defined Gal4 expressing cells, we simplified the visual paradigms to a single moving bar ( Figure 7A), which evokes a baseline LFP response that is attenuated by c309 expression of shibire TS [2]. When shibire TS was expressed in the D0264 Gal4 cells ( Figure 4A), LFP activity was attenuated at the restrictive temperature for all frequency ranges between 10 and 40 Hz. However, only the effect at 20-30 Hz was significant ( Figure 7B, left panel). After this transient attenuation, LFP responses returned to baseline levels upon recovery. Interestingly, attenuating synaptic release in the D0264 cells also immobilized the flies (data not shown), presumably due to widespread expression with this driver (see Figure 4A), so we were not able to test these animals behaviorally. shibire TS effects on the D0067 cells were similar to D0264 with regard to brain responses, again revealing specificity for the 20-30 Hz frequency range which was attenuated at the restrictive temperature ( Figure 7B, middle panel). In contrast, expressing shibire TS in the D0177 cells did not compromise LFP activity when synaptic release was attenuated in those cells ( Figure 7B, right panel). The relevance of the mushroom bodies for this response was further investigated with three strains showing more restricted expression in these structures than c309. Restricting shibire TS effects to the A0023-Gal4 cells failed to attenuate the LFP response, but restricting the same effects to the c305a and c739 cells attenuated the brain response ( Figure S7). Finally, mushroom body effects on brain activity were matched by behavioral data: c309/UAS-shibire TS , c305a/UASshibire TS , and c739/UAS-shibire TS all displayed significantly in-   with our view that the mushroom bodies are involved in visual suppression; optomotor responsiveness increases (and 20-30 Hz brain responsiveness decreases) when mushroom body output is compromised.
The 20-30 Hz response has been associated with visual attention [2,3]. For flies to respond appropriately to novelty (as shown in Figures 3, 5 & 6) presumably requires a level of attention (as opposed to mere exposure) to the visual objects being presented during training (e.g., the two squares). Abolishing attention mechanisms during training should thus abolish the selective novelty response. This was indeed the case for two of the Gal4s described above that specifically abolish the 20-30 Hz response (D0264 and D0067). We exposed UAS-shibire TS /D0264-Gal4 and UAS-shibire TS /D0067-Gal4 animals to the restrictive temperature only during training in our visual novelty paradigm ( Figure 7C). For both lines, attenuating synaptic output during training resulted in a loss of the selective 20-30 Hz response to visual novelty, with flies responding to either competing object equally, as for naïve flies ( Figure 7D,E, right panels). Control experiments in the same animals done without heating showed the characteristic 20-30 Hz selection/suppression profile following novelty ( Figure 7D,E, left panels). This suggests a possible role in memory acquisition for the circuits embedded within these Gal4 expression patterns and further emphasizes that attention and memory are likely to involve interacting processes in the fly brain.

Discussion
To remember something well, it helps to pay attention. Studies of selective attention have traditionally been restricted to higher animals such as humans and monkeys, and as a result, our understanding of the phenomenon has often been confounded by our lack of understanding of the nature of consciousness and its relation to attention. Yet, attention may be better studied in other animals as essentially a suppression phenomenon, where the brain filters out most of the world except those stimuli immediately relevant or salient to the behavioral task at hand. In this view, in order to study attention we must measure suppression effects. This concept underpins the behavioral approach we used to uncover mutants with attention-like defects: specifically, we suggest that mutants such as dunce 1 display increased optomotor responsiveness because they are less able to suppress visual reflexes, and that this is in direct relation to their attention-like defects.
How an animal's capacity for visual suppression becomes a dynamic, experience-dependent phenomenon akin to selective attention seems to be a more difficult problem, requiring a parallel understanding of memory circuits and their interaction with suppression mechanisms. Understanding this interaction is possible through the use of model organisms, such as Drosophila melanogaster, where we can manipulate memory circuits to examine their effects on suppression. In this study, by progressing from behavioral screens to electrophysiology and active control of selected groups of neurons, we showed that 1) A sub-group of olfactory learning and memory mutants in Drosophila are also characterized by more general defects in visual attention-like processes, and 2) the mushroom bodies, structures traditionally associated with olfactory learning and memory, seem to also be involved in modulating visual attention.
Two LTM mutants clearly stood out by displaying optomotor response levels equivalent to dunce mutants: D0264 and D0067. Our strategy of screening for high optomotor levels was successful as both high optomotor performers revealed defects at the level of brain activity: LFP responsiveness was either severely compromised (in D0067) or LFP power was shifted to a lower frequency domain (in D0264). Since these mutants are also enhancer trap Gal4 strains [8], we were able to show that the cells subserved by these selected elements also attenuated the 20-30 Hz LFP response to visual salience when tested in combination with UAS-shibire TS at the restrictive temperature. The specificity of the effect for the 20-30 Hz range was striking (10-20 Hz and 30-40 Hz were only moderately affected), suggesting that this narrow frequency band at 20-30 Hz is functionally associated with activity in these networks of neurons. However, D0264 and D0067 seem to affect rather distinct groups of cells, with only the mushroom body (MB) a and b lobes representing a clear overlap between the two.
In a previous study, we found other MB-expressing Gal4 strains that, when driving UAS-shibire TS , attenuated the 20-30 Hz response, namely c309 and 121y [2]. To narrow down which parts of the MBs might be required for generating the LFP, we conducted additional UAS-shibire TS experiments in the present study on two more Gal4 strains shown to target the a and b and a' and b' lobes of the MB, c739 and c305a, respectively [16]. Both attenuated the 20-30 Hz response, but not as specifically as D0264 and D0067; neighboring frequency domains were also attenuated ( Figure S7). In contrast, the absence of any LFP attenuation with A0023 (which expresses only in a small subset of the MB Kenyon cells as well as much of the rest of the brain) suggested that a minimal portion of the MB is required for the LFP response. Thus, a directed approach to MB anatomy as well as an indirect approach based on olfactory LTM mutants all appear to point to the MB playing a key role in generating this visual salience-related LFP. The more focused effect of the D0264 and D0067 drivers to a narrow LFP frequency domain suggests that specificity in the LFP response is tied more to widespread circuits including much of the MB, possibly interacting with other brain regions, rather than to the sole contribution of individual MB lobes. Finally, our data showing that optomotor responsiveness increases when synaptic output is transiently blocked in c309, c305a, and c739 cells is consistent with our view that the MBs are involved in visual suppression mechanisms.
In this study we have thus narrowed the search space for control of visual attention-like phenotypes in the fly. First, it is clear that a proportion of learning and memory mutants found by completely different (olfactory) paradigms have relevant defects here. Therefore, the significant resources required for accomplishing a blind forward genetic screen may be sidestepped by exploring existing memory mutants. Again, a common feature of these mutations is the relevance of the MB to cognitive function. All the memory mutants we have found that compromise attention-like phenotypes target MB function: radish (van Swinderen and Brembs, submitted), dunce, rutabaga, D0067, and D0264. Consistent with this trend, synaptic silencing of the MB attenuates attentionlike responses in the brain, which suggests that the MB neurons are ''upstream'' of the circuits synchronizing to produce the LFPs detected in the brain. Since some Gal4 drivers were able to both rescue dunce 1 as well as attenuate 20-30 Hz responses, this suggests that dunce function overlaps to some extent with neurons causing the LFP oscillations (e.g., in D0264), but not always (e.g., in A0023). Immunohistochemical studies have shown that dunce is primarily expressed in the MB [19], which is consistent with our results showing rescue in these structures. Furthermore, Gal4 drivers such as c309 (which rescued dunce 1 function) are already active during metamorphosis [20], with strong expression in the MB of pupae (data not shown). Thus, we propose that dunce function during pupal development is required to generate the MB wiring necessary for attention-like oscillations. Whether the MB neurons themselves are oscillating (as opposed to causing oscillations downstream) is an open question. For the MBs to be required for generating oscillations while also generating the LFP responses themselves (as suggested by our dunce 1 rescue experiments) suggests a feedback system within these structures, as has been posited for other aspects of MB function [21]. Finally, our rescue of dunce 1 phenotypes by expressing wild-type protein in the pupal stage is consistent with a role for the MB here: substantial MB neuronal proliferation and organization occurs during this stage [22]. It is unclear why rescue of olfactory memory phenotypes can occur after this phase in development (albeit, for rutabaga mutants [23]) whereas visual attention phenotypes require pupal development (this study, and [3]). Future experiments investigating both visual attention and olfactory memory while controlling gene expression from pupae to adulthood (e.g., gene knockdowns or combined Gal4/Gal80 experiments [9] allowing for Gal4 induction only in the adult) should begin to unravel the level of overlap between visual attention and olfactory memory phenotypes with regard to mushroom body development and function.
Why might the MBs be relevant for selective attention in the fly? The consensus over several years of study is that these structures are crucial for olfactory learning and memory in flies [21]. Visual learning appears to reside in another brain area, the central complex [24]. However, while visual studies point to the MB as being dispensable for simple aversive visual learning [25] more recent studies have shown the MB to be required for contextual learning [26,27,28,29]. This ''more difficult'' form of learning includes context generalization, involving binding between modalities, disambiguating contradictory cues, and extracting salience from background. All of these phenomena are hallmarks of attention in human studies; context generalization, for example, may be re-interpreted as the suppression of non-predictive competing stimuli. The fact that the MB are required for these aspects of visual learning is consistent with their potential role in selective attention. Thus, we propose that these structures house a gating mechanism to memory formation, which could then be distributed throughout the brain (including the MBs themselves).
The gating mechanism would involve a dynamic selection of behaviorally relevant stimuli and suppression of competing stimuli, as we have been measuring in this study. Defective suppression mechanisms would thus lead to the dominance of certain prepotent behaviors (e.g., the optomotor response) and possibly learning defects when these prevent proper acquisition (or retrieval) of relevant events. Although we did not test visual learning in the Gal4 networks described in this work, a recent publication using aversive phototaxic suppression (APS) found that wild-type dunce expression in c309 cells rescued dunce learning defects in that visual paradigm [7]. The inability of dunce mutants to suppress a prepotent response, and rescue by wild-type dunce expression in the MB is entirely consistent with results from our study.
We propose that brain oscillations generated in part by the MB provide a mechanism of stimulus selection or suppression. Populations of neurons potentially firing in response to a stimulus (e.g., a moving grating) are either enabled or suppressed, depending on spatio-temporal characteristics of the 20-30 Hz oscillation. The connection between these attention-like oscillations and memory formation would be through spike timing dependent plasticity (STDP) in MB circuits, as has been demonstrated in another insect system [30]. In that study, oscillations in the locust MB were found to play a key role in synaptic plasticity mechanisms. This matches our observation that the three LTM mutants uncovered by our optomotor screen all display aberrant oscillatory behavior in the brain. Active control of such oscillations in Drosophila should further unravel the connection between attention and memory systems in the insect brain.

D. melanogaster strains and stocks
Flies were cultured at 22u, 50%-60% humidity, 12 hr:12 hr light:dark cycle on standard media. Wild-type flies are from the Canton-S strain; the dunce 1 mutant was obtained from Leslie Griffith (Brandeis University), the gsg-301 strain from Scott Waddell (University of Massachusetts), and the 36 LTM mutants from Josh Dubnau (Cold Spring Harbor Labs). The strains D0264, D0067, and D0177 were outcrossed to a white + background by standard procedures. Double mutants between these LTM strains and dunce 1 mutants were made by standard procedures using chromosomal balancers. Only 2-7 day-old female flies were tested, one day after having been anesthetized with CO 2 .

Optomotor maze and population responses
Visual responsiveness was tested behaviorally using an 8-choice maze placed over a CRT computer screen. The maze paradigm used was exactly as described previously [1]. Flies were collected under CO 2 anesthesia the day before an experiment and loaded (N = 25-30) into disposable polyethylene ''jumbo'' transfer pipettes (Fisher Scientific), where they were allowed to acclimatize 3 min in the dark before the pipette was inserted into to maze's starting position. After running the choice maze (2-5 min) in a darkened room, the flies' distribution among the nine collection tubes was scored as a weighted average ranging from 24 to +4 (see Figure 1). The Optomotor Index (OI) is the deviation of the weighted average from 0, the middle tube, where positive scores indicate optomotor responses in the same direction of image motion. All statistics were t-tests of experimental means (following tests for normality), unless otherwise specified (Wilcoxon rank sum tests were used for non-normally distributed data).
Visual stimuli presented to flies running the maze were exactly as described previously: 1 cm green/black gratings moving at 3 Hz, or single competing static black bars pasted onto the inside of the box covering the maze [1]. Visual distraction tests were conducted to assess how competing static images (such as the black bars) could compete with the moving green/black grating for perceptual resources. Tests for visual distraction required four separate experiments, each performed 8 times on separate fly groups within a genotype (see Figure 2B): Responsiveness to the black bar alone over a static green/black grating. G: Responsiveness to the moving grating alone without the black bar. GB+: Responsiveness to the moving grating when the bar is on the side towards where the grating is moving. GB-: Responsiveness to the moving grating when the bar is on the opposite side of grating movement. Significant distraction was determined by performing t-tests between the GB+ condition and the GB-condition, with the threshold level for significant differences where P,0.05. For experiments testing Gal4/UAS-shibire TS effects on optomotor responsiveness, we pre-incubated flies for 10 minutes at 37uC (already in their loading tubes) and then immediately ran them in the maze. Resulting optomotor scores were compared (by t-test set at P,0.05) to baseline (22uC) runs of the same genotypes. Wildtype optomotor responsiveness was unaffected by pre-incubation at 37uC (OI = 0.5760.17, compared to OI = 0.7860.09).

Visual learning paradigms
We employed two different paradigms to assay visual learning in walking flies. In the first paradigm, we used the optomotor maze to test for changes in performance when groups of flies re-ran the maze multiple times in immediate succession. Each individual fly was scored (24 to +4) as it completed the maze and then immediately collected into a loading tube. Upon collecting the requisite number of flies for an experiment (25)(26)(27)(28)(29)(30), these were then immediately re-tested in the same maze for up to four successive runs. We found that a form of learning (measured by increased optomotor responsiveness) reached a plateau after two re-runs in wild-type flies. Analyses of optomotor responsiveness were performed exactly as for regular maze experiments.
In the second visual learning paradigm, we tested associative conditioning by modifying a previously described protocol [11] where visuals are associated with aversive shaking. Flies (25)(26)(27)(28)(29)(30) were loaded into glass test tubes attached to a vibrating device (www.neutrogenawave.com). The tubes formed a ''V'' connected at the center by the vibrator, such that flies could travel into either tube but were shaken down to the middle upon vibration (see Figure S1). The ''V'' was centered over a computer monitor where visuals could be displayed on either half. Training involved five 1 min sessions of shaking associated with one color which then alternated with five 1 min sessions without shaking associated with another color. Shaking epochs consisted of six 5 s periods without vibration followed by six 5 s vibration periods. Upon the completion of training, flies remained in the dark for 1 min and then were shaken for 5 s (whereupon they tumbled to the center of the ''V''). They were then presented with both visual stimuli, one on either side. The experiment was filmed throughout under infrared illumination, and the proportion of flies choosing either arm of the ''V'' were counted at regular intervals: 1) pre-exposure to the visuals for 2 min, 2) after each 5 s epoch without shaking during training (and equivalent epochs for the alternate visual), and 3) for successive 10 s epochs upon completion when flies were presented with a choice of both visuals simultaneously. The colors used were violet versus cyan; all strains were tested for normal responsiveness to these colors in the maze (data not shown). All experiments were equally balanced and spatially alternated, with shaking associated with one color first, followed by association of the other color with the shaking (for a new set of flies). Four groups of 25-30 flies were tested per experiment (four glass ''V's'' attached to the vibrator), and a learning index was calculated as: # flies in unshaken side -# flies in shaken side)/total # flies. Control of visuals and shaking was accomplished automatically using LABVIEW software. A movie of the paradigm is available in Supporting Information (Movie S1).

Electrophysiology
Brain recordings were performed as described previously [2,3,31]. Briefly, two glass electrodes were implanted into the fly brain, one in the left optic lobe and the other in the central brain. A voltage differential between these, filtered between 1 and 100 Hz, was sampled at 300 Hz using Labview software (National Instruments). Visual novelty experiments and the analysis of local field potential (LFP) data were performed as described previously [3]. Two distinct visual objects were used for novelty experiments, a cross and a square. These were displayed inside a circular arena as dark objects moving on a green LED background (see Figure 3). The objects rotated around the fly once every 3 s. Training involved exposure for a set time (e.g., 100 s) to two identical squares, after which one of the squares changed to a cross (the novel object), which we refer to as a ''novelty transition''. Brain responsiveness in the epoch after a novelty transition was analyzed for selection or suppression of the competing distinct images. Briefly, a Fourier analysis was performed in Matlab (Mathworks) for a moving window resulting in power spectra assigned to 24 sequential sectors of a full panorama rotation. Significant responses to one or the other object was then determined by comparing summed power within a frequency domain between the opposing 6 sectors for either object (when it is in front of the fly). Significance was set at P,0.05 and tested by the Wilcoxon rank sum method for all comparisons, as sample sizes were small and often not normally distributed. Analysis of LFP data for synaptic silencing experiments involving shibire TS were performed as described previously [2]. The front-to-back difference in LFP activity (again, after calculating power across sequential 24 sectors) across the visual field determined the response to a single visual object, and this was quantified for heated epochs (100 s at 38uC) and contrasted to baseline (room temperature) for significance (P,0.05) by t-test and/or Wilcoxon rank sum.

Pupal injections
Late-stage pupae were injected with Mifepristone (Sigma) in experiments aimed at rescuing dunce 1 attention defects during development. 13.8 nL of 50 mg/ml Mifepristone (Sigma) in DMSO was injected into pupae using a Nanoject II system (Drummond). Pupae were stuck to a piece of tape on a glass slide, which was then transferred to a regular food vial. In control experiments 13.8 nL of only DMSO was injected.

Brain imaging
One week old female brains of Gal4/UAS-mCD8::GFP strains were removed by standard procedures in cold PBS and whole mounts were imaged immediately using a confocal fluorescence microscope. Images in Figure 5A-C were generously provided by Ann-Shyn Chiang (National Tsing Hua University, Taiwan); these were confirmed in our own laboratory strains.  Figure 1, were immediately collected into a loading tube and re-run through the same maze in batches of 20-30 animals. Tube score, the proportion of flies in each of the nine collection tubes, with tube +4 being most in the direction of optomotor flow. First, second, and third runs are shown in blue, red, and green, respectively, for wild-type flies. B. Average Optomotor Index for each of the runs in A. *, P,0.05, by t-test against means. C. Maze re-runs for D0264. *, P,0.05, by t-test. D. Maze re-runs for D0067. E. Maze re-runs for D0177. Found at: doi:10.1371/journal.pone.0005989.s002 (3.00 MB TIF) Figure S3 Baseline LFP responses with brain activity in the 10 s immediately preceding a novelty effect (as shown in Figure 3). In this case, LFP activity for 3 frequency domains are in response to two identical squares (shown in the schema at the bottom of each column). Comparisons of activity for the six sectors representing each object revealed no significant effects. D0177 showed a significant response (at 30-40 Hz), but it was not mapped to either object. Found at: doi:10.1371/journal.pone.0005989.s003 (3.00 MB TIF) Figure S4 Training requirements for novelty response in the brain. A. D0264 was trained (exposed to two identical squares, as in Figure 3) for 100 s, 50 s, and 25 s. * = significant response (6s.e.m.) to the novel object in the 10 s following a novelty transition (P,0.05, by Wilcoxon rank sum of the 6 sectors for the square versus the 6 sectors for the box). Results are shown for the 10-20 Hz frequency range only, where D0264 displayed greatest responsiveness (Figure 3). B. D0067 responses, for the 10-20 Hz domain, following the three different training regimes. C. D0177 responses, for the 30-40 Hz domain (see Figure 3), following the three different training regimes. Found at: doi:10.1371/journal.pone.0005989.s004 (3.00 MB TIF) Figure S5 Baseline LFP responses for spatial rescue of dunce1. Brain activity in the 10 s immediately preceding a novelty effect (as shown in Figure 5). In this case, LFP activity for 3 frequency domains is in response to two identical squares (shown in the schema at the bottom of each column). The data follows exactly the outline of Figure 5. A significant effect was found for the c309 circuit in the 20-30 Hz range (P,0.05, Wilcoxon rank sum), indicating that in this particular strain flies responded to one square more than the other, immediately preceding a choice between a square and a cross. Found at: doi:10.1371/journal.pone.0005989.s005 (3.00 MB TIF) Figure S6 Baseline LFP responses for temporal rescue of dunce1. Brain activity in the 10 s immediately preceding a novelty effect (as shown in Figure 6). M = Mifepristone. No significant effects were detected. Found at: doi:10.1371/journal.pone.0005989.s006 (3.00 MB TIF) Figure S7 Electrophysiology of synaptic silencing in 3 mushroom body circuits. A. Recording arena setup. The visual stimulus is a moving dark bar on a lit green background. B. Brain responses to the visual for the Gal4 drivers A0023, c305a, and c739 [16] complemented by UAS-shibire. LFP responses are calculated as the normalized maximum -minimum [2] for three successive temperature conditions, for three frequency domains. B = baseline response, at 22u, H = response during heating at 38uC, R = recovery response at 22u. Calculations were for 100 s at each condition, averaged from a triplicate test per fly. N = 3 flies per genotype. * = significantly different from baseline, set at 1 (P,0.05, by t-test). C. UAS-GFP expression of the Gal4 lines tested in B. Found at: doi:10.1371/journal.pone.0005989.s007 (3.00 MB TIF)

Supporting Information
Movie S1 Movie file of segments from visual learning experiment. Visual classical conditioning apparatus. Found at: doi:10.1371/journal.pone.0005989.s008 (40.34 MB AVI)