Current Biology 23, 156-161, January 21, 2013 02013 Elsevier Ltd All rights reserved http://dx.dol.org/10.1016.cub.2012.11.048 Report Selective Attention in an Insect Visual Neuron Steven D. Wiederman and David C. O'Carroll 'Adelaide Centre for Neuroscience Research, School of Medical Sciences, The University of Adelaide, Adelaide, SA 5005, Australia local inhomogeneity in the receptive field (i.e., variable excit- atory and inhibitory synaptic inputs and local differences in spatiotemporal response tuning). Responses are strongest near a frontal "hot spot" 60 above the horizon but also depend on stimulus contrast and size (Figures 1C, 1E, and SIC). This is due in part to the optics of the eye, with a pronounced region of maximal acuity (<0.5') in the frontal- dorsal visual field, falling 3-fold by 40 away (14). The neuron is correspondingly more sensitive to small targets frontally and larger targets in the periphery (Figure S1C). Although in swarms comprising prey and conspecifics [1], a feat that CSTMD1 responds to targets of contrast below 25% (Fig- ure 1E), the receptive field is smaller than for higher contrasts (Figure 1C), with significant responses only in the vicinity of Summary Animals need attention to focus on one target amid alter- native distracters. Dragonflies, for example, capture flies requires neurons to select one moving target from com- peting alternatives. Diverse evidence, from functional imaging and physiology to psychophysics, highlights the importance of such "competitive selection" in attention for the hot spot. Receptive fields are similar in the same neuron in different vertebrates [2-5). Analogous mechanisms have been pro- dragonflies. They are also stable over prolonged recording periods, illustrated by the similarity in maps obtained by repeated stimulation of the ipsilateral receptive field (Figures 1B and 1D) and eight identical scans through the hot spot from an identified dragonfly visual neuron (11, 12] that over 15 hr (Figure 1F). Consequently, successive scans of perfectly match a model for competitive selection within identical targets are very strongly correlated with one another limits of neuronal variability (r - 0.83). Responses to indi- irrespective of their size, contrast, or location (- 0.76) posed in artificial intelligence (6) and even in invertebrates [7-9), yet direct neural correlates of attention are scarce from all animal groups (10]. Here, we demonstrate responses vidual targets moving at different locations within the recep- (Figure S2). tive field differ in both magnitude and time course. However, responses to two simultaneous targets exclusively track those for one target alone rather than any combination of The reproducible and unique time-varying response to single targets thus provides a characteristic temporal "finger- print" that allows us to test our hypothesis: if the neuron the pair. Irrespective of target size, contrast, or separation, selects one target, the response to two simultaneous targets this neuron selects one target from the pair and perfectly should resemble either one presented alone, not a blend, such as their sum or average. We tested this on unique tra- jectories T, and T, (Figure 1B), with either a single target, amenable to electrophysiological recordings, providing presented along each trajectory, or both targets presented neuroscientists with a new model system for studying selec- together ("Pair"). T, alone yields a strong response to 2.5", high-contrast targets (a near-optimal stimulus frontally) shortly after onset and passes through the hot spot, giving a maximal response late in the time course (Figure 1G). The more peri- pheral T2 yields a response that increases more gradually preserves the response, regardless of whether the "winner" is the stronger stimulus it presented alone. This neuron is tive attention. Results We recorded intracellularly from the "centrifugal small-target before declining (at least for the neuron shown in Figure 1H). motion detector" neuron CSTMD1 (13), a recently identified binocular neuron from the dragonfly midbrain. It responds selectively to small (1-3) targets moving across a large receptive field in either excitatory (psilateral) or inhibitory responses, which consistently resemble the responses for (contralateral) visual hemispheres (Figure 1 and see also Fig- ure $1 available online). CSTMD1's neuroanatomy (Figure S1A) one or the other single target. In Figure 2A, T, (red) and T, is consistent with a possible role in attention as targets move from one visual hemisphere to the other (12, 13]. To test its possible role in the competitive selection of targets, we compared CSTMD1's response to single and paired targets further neurons (N2 and N3 in Figure 2) for targets that are The time course depends also on the target size or contrast selected: smaller or lower-contrast targets yield weaker over- all responses. Our primary result is illustrated in Figure 2 by the Pair (blue) were small (1.25 square) targets 20 apart. After an initial lag in which the Pair response (black) is weaker than either single target, it closely follows the temporal fingerprint for T, alone. Figures 28 and 2C show examples from two (Figure 1). Because we cannot instruct a restrained dragonfly to "attend" to one target, we instead use inhomogeneity in the receptive field to determine which of two alternative targets the neuron tracks. When we stimulate CSTMD1 by drifting response more than T, (Figure 1C). Intriguingly, when we a small dark target at different locations across a bright LCD screen, differences in the response time course reflect both small (1.25) and low contrast. In both neurons, individual target responses are delayed, eventually responding robustly near the hot spot. Receptive field asymmetry delays the Ta present the Pair stimulus, the response appears to "lock" onto the T, fingerprint, even after T, passes out of the recep- tive field on that trajectory. The response falls to baseline levels, even though Tz is still within the receptive field. The Pair response thus appears to encode a single selected stimulus and ignore the other. "Correspondence: steven.wiederman@adelaide.edu.au (S.D.W.), david. ocarroleadelaide.edu.au (D.C.O.) CMa Selective Attention in an Inseet Visual Neuron 157 Figure 1. Receptive Fields of CSTMD1 in Hemi- cordulia tau and Response to Moving Targets A Dark targets drifted vertically (42 ) on a white background (315 cd/m, 120 Hz LCD display) within the contralateral field suppress intracel- hular responses to below spontaneous levels. Identical targets moved in the ipsilateral hemi field (T. T evoke excitatory responses with strength dependent on stimulus contrast (high 1, low 0.50, L d n (B) Target 1 (T)moves through the receptive field hot spot and Target 2 (T) is located 20 to the right. (C)A lower-contrast target maps a smaller recep- Contralateral (C) Ipsilateral (T1) Ipsilateral (T2) Low Contrast (T1) 300 250 200 150 100 50 tive fiekt. (D) Receptive field remapped as in (B), revealing consistent inhomogeneity in spatial structure. (D CSTMD1 centrifugal small-target motion detector" neuron) responses to targets of varying ntrast drifted horizontally through the recep- tive field hot spot (meana SEM, n neurons. dashed line mean spontaneous rate). 80 50 0 50 0 Azimuth (") 50 Azimuth (*) Contra Ipsi Azimuth (") 150 n Eight target scans over a 15 hr period reveal low neuronal variability (gray lines individual responses black line m (G) CSTMD1 response to three trials of the single T, stimulus (red). 0 single Ta blue), or ) simul- taneous presentation of both T, and T, (Pair black). mear 100 05s -T,AT, (Pair) Could the qualitative match between Pair and T, or T, be a chance observa- tion resulting from neuronal variability? Figure 3 shows scatter plots (color saturation indicates the density of multiple points; 25 ms bins) for responses within the receptive field from 72 trials at (Figures 2D and 2E). In the third trial, the response is initially 20 separation, pooled across all four combinations of target size and contrast, over nine neurons. We see a weak correla- T, (Figure 2F). In a further trial with smaller targets (1.25) tion when we plot responses for Pair against either T, . and two trials using lower contrasts, we see the opposite 0.58) or T2 (- 0.35) (Figures 3A and 3B). This confirms that result: Pair now resembles the initially stronger T, until the response to the Pair stimulus does not simply reflect the response to T, or T, alone. However, if we assume that switching behavior is not seen in every trial, most examples competitive selection operates to track either target at a given occur when responses to individual targets are equally strong, time point, by plotting Pair against either T, or T2. after suggestive of an underlying competitive mechanism. With computing whichever provides the least difference, we see a very strong correlation ( 0.83) (Figure 3C). Were T, and 2L), both T, and Ta yield very strong initial responses (>250 Ta similar to one another, some improvement in this correlation might be expected from neuronal variability, because this analysis compares Pair with two possible altematives at each time point. Our deliberate selection of different trajecto- Figures 2K and 2L), further suggesting initial competitive ries for T, and Ta, however, ensures that this is rarely the case, evidenced by both the raw data (Figure 2) and the much weaker correlation of T, with T2 (- 0.27) (Figure 3D). Indeed, the assumption of competitive selection yields a corre- Individually, these produce radically different response time lation as strong as for subsequent repetitions of identical trials at T, or Ta mean (- 0.76) (Figure S2). We conclude that, within limits of neuronal variability, the Pair response is somewhat independent of the potency of a stimulus, at least usually identical to that for one of the targets presented alone. We can further quantify whether Pair responses reflect competitive selection by considering differences between Pair and alternative combinations of T, and Tg. Figure 4A shows an example model for hypothetically "perfect" compet- itive selection based on the actual values of T, or Ta responses that correspond most closely to the Pair response. The close match between this model and the observed Pair response O 02 04 0.6 0.8 Contrast In two of three further trials from N2 with larger targets (2.5'), the Pair response follows T, despite this being weaker than T, identical but "switches" midway to closely track the stronger midway, before switching to T, (Figures 20-21). Although this near-optimal stimuli (2.5 targets, high contrast) (Figures 2J- spikes per second), a characteristic typically shared by Pair (e.g., Figure 2J). Rarely, however, there is a pronounced delay before Pair closely tracks an individual target (e.g., interactions. We tested stimuli, as illustrated combinations of size, contrast, or separation of target pairs. Figure 2, across varied courses for T, and Ta. The Pair response, however, consis- tently appears to select one target. Nevertheless, selection is as evidenced by the receptive field of CSTMD1. The selected target can be either T, or Ta, regardless of which one causes stronger CSTMD1 responses (Figure S3A). This variation in target choice suggests that selection involves a process akin to selective attention in vertebrates, a "cognitive" filter to focus on one particular target even in the presence of an equally (or more) salient distracter (15-17).