* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download A Cortical Substrate for Memory
Neuroeconomics wikipedia , lookup
Multielectrode array wikipedia , lookup
Holonomic brain theory wikipedia , lookup
Metastability in the brain wikipedia , lookup
Neuroanatomy wikipedia , lookup
State-dependent memory wikipedia , lookup
Premovement neuronal activity wikipedia , lookup
Biological neuron model wikipedia , lookup
Sparse distributed memory wikipedia , lookup
Neural correlates of consciousness wikipedia , lookup
Neuroanatomy of memory wikipedia , lookup
Stimulus (physiology) wikipedia , lookup
Synaptic gating wikipedia , lookup
Optogenetics wikipedia , lookup
Development of the nervous system wikipedia , lookup
Neural coding wikipedia , lookup
Nervous system network models wikipedia , lookup
Neuropsychopharmacology wikipedia , lookup
Neuron Article A Cortical Substrate for Memory-Guided Orienting in the Rat Jeffrey C. Erlich,1 Max Bialek,1 and Carlos D. Brody1,* 1Howard Hughes Medical Institute, Princeton Neuroscience Institute and Department of Molecular Biology, Princeton University, Princeton NJ 08544, USA *Correspondence: [email protected] DOI 10.1016/j.neuron.2011.07.010 SUMMARY Anatomical, stimulation, and lesion data have suggested a homology between the rat frontal orienting fields (FOF) (centered at +2 AP, ±1.3 ML mm from Bregma) and primate frontal cortices such as the frontal or supplementary eye fields. We investigated the functional role of the FOF using rats trained to perform a memory-guided orienting task, in which there was a delay period between the end of a sensory stimulus instructing orienting direction and the time of the allowed motor response. Unilateral inactivation of the FOF resulted in impaired contralateral responses. Extracellular recordings of single units revealed that 37% of FOF neurons had delay period firing rates that predicted the direction of the rats’ later orienting motion. Our data provide the first electrophysiological and pharmacological evidence supporting the existence in the rat, as in the primate, of a frontal cortical area involved in the preparation and/or planning of orienting responses. INTRODUCTION Behaviors that require the planning and execution of orienting decisions have long been investigated in rodents. A classic example is navigation through mazes (Tolman, 1938; Hull, 1932; Olton and Samuelson, 1976). Recordings from the rodent hippocampus and entorhinal cortex have led to important discoveries about the neural encoding of navigation and the representation of space (McNaughton et al., 2006; Moser et al., 2008). Navigation is composed of a sequence of individual orienting motions, but in contrast to rodent studies of spatial navigation, the neural control of individual orienting motions has been studied most thoroughly in primates, specifically with regard to the control of gaze by the frontal and supplementary eye fields (FEF and SEF) (Schall and Thompson, 1999; Schiller and Tehovnik, 2005). As a result of being separated by both different model species and by different behavioral paradigms, literature for the navigation system and literature for the orienting systems have remained far apart, making few references to each other (but see Arbib, 1997; Corwin and Reep, 1998; Kargo et al., 2007). Yet the two systems must necessarily interact (Whitlock et al., 2008). As part of bridging the gap between these two fields 330 Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. of research, we took a classic primate behavioral paradigm, memory-guided orienting (Gage et al., 2010; Funahashi et al., 1991), which is known to be FEF-dependent (Bruce and Goldberg, 1985; Bruce et al., 1985), and adapted it to rats. Then, in rats performing the task, we studied a rat cortical area that has long been suggested as homologous to the primate FEF. The area we studied appears in the literature under a large variety of names. These include M2 (Paxinos and Watson, 2004), anteromedial cortex (Sinnamon and Galer, 1984), dorsomedial prefrontal cortex (Cowey and Bozek, 1974), medial precentral cortex (Leichnetz et al., 1987), Fr2 (Zilles, 1985), medial agranular cortex (Donoghue and Wise, 1982; Neafsey et al., 1986), primary whisker motor cortex (Brecht et al., 2004), and rat frontal eye fields (Neafsey et al., 1986; Guandalini, 1998). A theme common to many studies of this area, and shared with the primate FEF, is a role in guiding orienting movements. We targeted a particular point at the center of the areas investigated in the studies cited above (+2 AP, ±1.3 ML mm from Bregma), and refer to the cortex around this point as the frontal orienting field (FOF). The homology between rat FOF and primate FEF was first proposed four decades ago by C.M. Leonard (1969), based on the anatomical finding that the FOF, like the FEF, receives projections from the mediodorsal nucleus of the thalamus (Reep et al., 1984), and projects to the superior colliculus (SC) (Reep et al., 1987). Later, Stuesse and Newman (1990) found that the rat FOF also projects to other oculomotor centers in the rat’s brainstem, in a pattern that mimics the oculomotor brainstem projections of the primate FEF. Also like the FEF, the FOF receives inputs from multiple sensory cortices, including visual, auditory, and somatosensory cortices (Condé et al., 1995), and has strong reciprocal connections with the prefrontal (Condé et al., 1995) and parietal cortices (Corwin and Reep, 1998). The rat FOF, like the primate FEF, is thus well-placed to integrate information from many different sources in the service of guiding orienting motions. Leonard’s proposal led to studies that found that unilateral lesions of the FOF produced effects consistent with contralateral neglect (Cowey and Bozek, 1974; Crowne and Pathria, 1982; Crowne et al., 1986), which is a classic symptom of FEF damage in humans and monkeys (Ferrier, 1875; Hebb and Penfield, 1940). Further support for Leonard’s proposal came from studies that revealed orienting motions in response to intracortical microstimulation of the FOF (Sinnamon and Galer, 1984). This parallels the orienting motions produced by stimulation of the primate FEF in headfixed (Bruce et al., 1985) as well as head-free animals (Monteon Neuron Cortical Basis of Memory-Guided Orienting in Rats et al., 2010). Neafsey et al. (1986) reported that stimulation of the FOF in anesthetized, head-fixed rats produced both eye and whisker motions and suggested it was an eye-head orientation cortex, homologous to the FEF. More recently, based on the whisker motions evoked by electrical stimulation of the FOF, the area has been studied as a whisker motor cortex (Brecht et al., 2004), with particular attention paid to its role in vibrissal active sensing (reviewed in Kleinfeld et al., 2006). To our knowledge, there are only a few electrophysiological studies recording single neurons of awake animals in this area (we are aware of only three, Carvell et al., 1996; Kleinfeld et al., 2002; Mizumori et al., 2005), and they have not focused on the FOF’s role in orienting motions. Kleinfeld et al. (2002) used head-fixed rats, precluding the study of head- or body-orienting movements. Carvell et al. (1996) recorded from awake rats that were whisking freely while being held in the experimenter’s hands, but orienting movements were not recorded, and the rats were not required to perform any task. Mizumori et al. (2005) reported head direction tuning (Taube, 2007) in the FOF. Mizumori et al. (2005) also mentioned observing neurons that encoded egocentric motions, including orienting movements, but they did not elaborate on this observation. To further investigate the role of the FOF in the control of orienting, we carried out unilateral pharmacological inactivations of the FOF and recorded extracellular neural spiking signals from the FOF, while rats were performing a memory-guided orienting task (Gage et al., 2010; Funahashi et al., 1991). Our findings provide the first pharmacological and electrophysiological evidence that the FOF plays an important role in the preparation (Riehle and Requin, 1993) of orienting movements. RESULTS The Memory-Guided Orienting Task We developed a computerized protocol to train rats to perform a two-alternative forced-choice memory-guided orienting task (Figure 1A). Training took place in a behavior box with three nose ports arranged side-by-side along one wall, and with two speakers, placed above the right and left nose ports. Each trial began with a visible light-emitting diode (LED) turning on in the center port. In response to this, rats were trained to place their noses in the center port, and remain there until the LED was turned off. We refer to this period as the ‘‘nose in center’’ or ‘‘fixation’’ period, and varied its duration randomly from trial to trial (range: 0.9–1.5 s). During the fixation period, an auditory stimulus, consisting of a periodic train of clicks, was played for 300 ms. Click rates greater than 50 clicks/s indicated that a water reward would be available on the left port; click rates less than 50 clicks/s indicated that a water reward would be available on the right port. On ‘‘memory trials,’’ the click train was played shortly after the rat placed its nose in the center port, and was followed by a silent delay period before the fixation period ended and the animal was allowed to make its response. On ‘‘nonmemory trials,’’ the click train ended at the same time as the fixation period, and the animal could respond immediately after the end of the stimulus. The two types of trials were randomly interleaved with each other in each session. For animals in behavioral and pharmacological experiments, we also interleaved, across trials within each session, six different click rate values, ranging from easy trials, with click rates far from 50 clicks/s, to difficult trials, with click rates close to 50 clicks/s. To maximize the number of identically prepared trials, animals in electrophysiological experiments were presented with only two click rates, 100 and 25 clicks/s, again randomly interleaved across trials (Figure 1C, filled circles). Here we present data from 25 male Long-Evans rats, five of which were implanted with bilateral FOF cannula for infusions, four of which were implanted with bilateral M1 cannula, and another five of which were implanted with microdrives for tetrode recording. Four of the five tetrode-implanted rats performed memory-guided click rate discrimination, as described in Figure 1. As a preliminary test of the effects of a different class of instruction stimulus, the fifth tetrode-implanted rat was trained on a memory-guided spatial location task, in which the click train rate was always 100 clicks/s, and the rewarded side was indicated by playing the click train from either the left or the right speaker. The behavioral performance and physiological results were similar for the two stimulus classes (i.e., click rate discrimination and location discrimination; see Figure S4 available online), and are reported together in the main text. Rats performed about 300 trials per 1.5 hr session each day, 7 days a week, for 6 months to 1.5 years. After each animal was fully trained, an average of !66,000 trials per rat were collected. Maintaining fixation is likely to require inhibitory control (Narayanan and Laubach, 2006; Munoz and Wurtz, 1992), and individual rats varied in the percentage of trials in which they broke fixation (range: 10%–50%). There were consistently more broken fixation trials for memory trials (mean ± standard error [SE], 37% ± 2%) than for nonmemory trials (mean ± SE, 29% ± 2%, paired t test, p < 10"5). Unless otherwise specified, all trials where rats prematurely broke fixation were excluded from analyses. For each rat, we combined the data across sessions and fitted four-parameter logistic functions to generate one psychometric curve for memory trials, and another curve for nonmemory trials (Figure 1C, thin lines). Percent correct on the easiest memory trials was similar to the easiest nonmemory trials (94% versus 95%, paired t test, p > 0.49). Click frequency discrimination ability, as assayed by the slopes of the psychometric fits at their inflection point, was also similar for memory and nonmemory trials ("2.3% versus "2.1% went-right per click/sec, paired t test, p > 0.35). This suggests that the two types of trials are of similar difficulty. We tested whether whisking played a role in performance of the memory-guided orienting task in three ways. First, we cut off the whiskers of three rats bilaterally. This manipulation had no statistically significant effect on psychometric function slopes or endpoints, although it did produce a small effect on overall percent correct performance (83% ± 1% without whiskers versus 87% ± 1% with whiskers, t test, p < 0.05). There was no differential effect on memory versus nonmemory trials (t test, p > 0.5; Figures 1D and 1F). Second, we probed whether asymmetric whisking played a role in task performance by using unilateral subcutaneous lidocaine injections to temporarily paralyze the whiskers on one side of the face of four rats. Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. 331 Neuron Cortical Basis of Memory-Guided Orienting in Rats A B C D E F Figure 1. Memory-Guided Frequency Discrimination Task and Behavioral Performance (A) Task schematic, showing a cartoon of a rat in the behavior box and the timing of the events in the task. Onset of the Center LED indicated to the rat it should put its nose in the center port, and remain there until the LED was turned off. During this variable-duration ‘‘nose fixation’’ or ‘‘nose-in-center’’ period, a 300 ms-long periodic train of auditory clicks was played. Click rates higher than 50 clicks/s indicated that a water reward would be available from the left port; click rates lower than 50 clicks/s indicated reward would be available from the right port. On memory trials (orange), the click train was played near the beginning of the fixation period, and there was a several hundred ms delay between the end of the click train and the end of the nose fixation signal. On nonmemory trials (green) the click train ended at the same time as the nose fixation signal. (B) An example of performance data for a single rat. Each circle indicates the percentage of trials in which the subject chose the right port for a given stimulus in a single session. There were six stimuli presented in each session. The thick line shows the psychometric curve, drawn as a 4-parameter sigmoidal fit to the circles. The left panel shows data from nonmemory trials, and the right panel shows data from memory trials. (C) Psychometric curves showing performance of 20 rats. Thin lines are the fits to individual rats, as in (B). Thick lines are the fits to the data combined across rats. The performance of electrode implanted rats (n = 5) is shown by the small filled circles at the two stimuli used with these animals (25 clicks/s and 100 clicks/s). (D) Bilateral whisker trimming (3 rats) has a minimal effect on performance. The gray line is the average of memory and nonmemory trials for control sessions before whisker trimming. Diamonds are data after trimming, solid lines are sigmoid fits. Memory trials are in orange, nonmemory trials in green. (E) Unilateral whisker pad anesthesia and paralysis (four rats) also has a minimal effect on performance. Open circles are data from lidocaine sessions. Color conventions as in (D). (F) Summary of effects of whisker trimming and lidocaine. See also Figure S1, Movie S1, Movie S2, and Movie S3. This manipulation did not generate any lateralized effects on performance, but led instead to a small bilateral effect, indistinguishable from that of bilateral whisker trimming (Figures 1E and 1F). Third, we performed video analysis of regular sessions (no drug, no whisker trimming), searching for differences in delay period whisking preceding leftward versus rightward movements. No significant differences were found (Figure S1). Furthermore, in the video analyzed, the whiskers were held still during the memory delay period (Movie S2, compare to exploratory whisking in Movie S1 and out-of-task whisking Movie S3). In sum, whisking appears to play a negligible role in the memory-guided orienting task. 332 Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. Muscimol Inactivation of the FOF Generates a Contralateral Impairment In contrast to the negligible effects found from manipulating the whiskers themselves, we found that manipulating neural activity in the FOF produced strong effects on memory-guided orienting. Unilateral inactivation of the FOF generated a clear impairment on trials where the animal was instructed to orient contralateral to the infusion site. (Figure 2, Contra trials). Performance on ipsilaterally-orienting trials was unaffected (Figure 2, Ipsi trials). Contralateral impairment was observed for both memory and nonmemory trials, which were randomly interleaved with each other. However, the effect was markedly stronger on memory Neuron Cortical Basis of Memory-Guided Orienting in Rats A B Figure 2. Unilateral Inactivation of FOF Generates a Contralateral Impairment that Is Larger for Memory Trials Compared to Nonmemory Trials (A) Behavioral performance on control and muscimol-infusion days. Top row: nonmemory trials. Bottom row: memory trials. Left column: muscimol infusions into left FOF. Right column: musicmol infusions into right FOF. Open circles, data from muscimol infusions. Closed circles: control data from days immediately preceding infusion days. Dashed lines: sigmoidal fits to muscimol data. Solid lines: sigmoidal fits to control data. Error bars are standard error of the mean. Error bars for control data were smaller than the marker in most cases. Underbraces at bottom indicate the sets of trials in which animals were instructed to orient ipsilaterally or contralaterally to the site of infusion. The percentages aligned to the dashed curves indicate the endpoint performance for the trials contralateral to the infusion. (B) Combined data from left and right infusion sessions and collapsed across all stimulus difficulty levels. The ‘‘No Drug’’ data come from the 20 sessions one day before infusion sessions. The Ipsi and Contra Muscimol data are the performance on ipsilateral trials and contralateral trials on infusion sessions (n = 20). See also Figure S2. trials (Figure 2A; compare top row to bottom row). Left infusions impaired rightward-instructed trials to the same degree that right infusions impaired leftward-instructed trials (four t tests: contra/mem p > 0.5, contra/nonmem p > 0.26, ipsi/mem p > 0.1, ipsi/nonmem p > 0.4). We therefore combined data from left and right infusion days for an overall population analysis, and confirmed that performance was worse for contralateral memory trials than nonmemory trials (Figure 2B, permutation test p < 0.001). Since memory and nonmemory trials are of similar difficulty (see above), the greater impairment on memory trials suggests that, in addition to a potential role in direct motor control of orienting movements, there is a memory-specific component to the role of the FOF. To test whether unilateral inactivation of primary motor cortex could produce a similar effect to inactivation of the FOF, we repeated the experiment, in the neck region of M1 (+3.5 AP, +3.5 ML). This is the same region in which Gage et al. (2010) recorded single-units during a memory-guided orienting task. Unilateral muscimol in M1 produced a pattern of impairment that was different, and much weaker, than that produced in the FOF. In particular, we found no difference in the impairment of contra-memory versus ipsi-memory trials (t test, p > 0.35) (Figures S2A–S2D). Neurons in the FOF Prospectively Encode Future Orienting Movements We obtained spike times of 242 well-isolated neurons from five rats performing the memory-guided orienting task. No significant differences were found across recordings from the left and right sides of the brain. Accordingly, we grouped left and right FOF recording data together. Below we distinguish between trials in which animals were instructed to orient in a direction opposite to the recorded side (‘‘contralateral trials’’) and trials in which they were instructed to orient to the same side (‘‘ipsilateral trials’’). We first analyzed spike trains from correct trials, with a particular interest in cells that had differential contra versus ipsi firing rates during the delay period, i.e., after the end of the click train stimulus but before the Go signal (see Figure 1A). We identified such cells by obtaining the firing rate from each correct trial, averaged over the entire delay period, and using ROC analysis (Green and Swets, 1974) to query whether the contra and ipsi firing rate distributions were significantly different. By this measure, we found that 89/242 (37%) of cells had significantly different contra versus ipsi delay period firing rates (permutation test, p < 0.05). We refer to these cells as ‘‘delay period neurons.’’ Examples of single-trial rasters for six delay period neurons are shown in Figure 3. For each cell, we then took the spike train from each trial and smoothed it with a half-Gaussian kernel to produce an estimated firing rate as a function of time (standard deviation [SD] of whole Gaussian = 200 ms; smoothing process is causal, i.e., looks only backward in time). At each time point, this gave us, across trials, a distribution of firing rates on contralateral trials and a distribution of firing rates on ipsilateral trials. We used ROC analysis to query whether the distributions were significantly Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. 333 Neuron Cortical Basis of Memory-Guided Orienting in Rats A B Figure 3. Upcoming Choice-Dependent Delay Period Activity in the FOF (A and B) (A) Three contralateral preferring cells and (B) three ipsilateral preferring cells that show delay period activity that is dependent on the upcoming side choice. The top half of each panel shows spike rasters sorted by the side of the rat’s response and aligned to the time of the Go cue. The pink shading indicates the time, for each trial, when the stimulus was on. The brown ‘‘+’’ indicates the time at which the rat placed its nose in the center port. The bottom half of each panel are PETHs of the rasters for ipsilateral (red) and contralateral (blue) trials. The two lines are indicate the mean ± SE. PETHs were generated using a causal half-Gaussian kernel with an SD of 200 ms. The thick black bar just below the rasters indicates the times when the cells response was significantly different on ipsi- versus contralateral trials (p < 0.01, ROC analysis). (C) Development of choice-dependent activity over the course of the trial. The lines indicate the % of cells (out of 242 neurons) that have significantly choice-dependent firing rate (p < 0.01) at each time point on memory trials (orange) and nonmemory trials (green). See Figures S3A and S3B for interspike interval histograms and waveforms for the example neurons. C different at each time point. By this assay, we found that (113/ 242) (47%) of cells in the FOF had significantly different contra versus ipsi firing rates at some point in time during memory trials (overall probability that a cell was labeled as significant by chance p < 0.05; time window examined ran from "1.5 s before to 0.5 s after the Go signal). The temporal dynamics of delay period neurons were quite heterogenous. Different cells had significantly different contra versus ipsi firing rates at different time points during the trial (indicated for each cell in Figure 3 by black horizontal bars). At each time point, we counted the percentage of neurons, out of the 242 recorded cells, that had significantly different contra versus ipsi firing rates, and plotted this count as a function of time for memory trials and for nonmemory trials (Figure 3C). For memory trials the population first became significantly active at 850 ms before the Go signal (Figure 3C, horizontal orange bar). For nonmemory trials the population became active 120 ms before the Go signal (Figure 3C, horizontal green bar). At the time of the Go signal on memory trials, 28% of cells had firing rates that predicted the choice of the rat. We labeled cells as ‘‘contra preferring’’ if they had higher firing rates on contra trials, and as ‘‘ipsi preferring’’ if they had higher firing rates on ipsi trials. When firing rates were examined across time (from "1.5 s before to 0.5 s after the Go signal), most cells 334 Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. had a label that was consistent across the duration of the trial: 82/89 (92%) of significant delay period neurons were labeled exclusively as either contra-preferring or ipsi-preferring. Seven of the 89 (8%) delay period neurons switched preference at some point during the trial, usually between the delay period and late in the movement period (data not shown). For our analyses below, we used labels based on the average delay period firing rate. Given the strong difference in contralateral versus ipsilateral impairment during unilateral inactivation (Figure 2), we were surprised to find no significant asymmetry in the number of contra-preferring versus ipsi-preferring delay period neurons: 50/89 cells (56%) fired more on contralateral trials (three examples are shown in Figure 3A), while 39/89 (44%) fired more on ipsilateral trials (three examples in Figure 3B). Although there were more contra preferring cells, the difference in number of contra versus ipsi-preferring cells was not statistically significant (c2 test on difference, p > 0.2). To perform population analyses of firing rates, we first Z-score normalized each cell’s perievent time histograms (PETHs) by subtracting their mean and dividing by their standard deviation, and then averaged across cells to obtain population normalized PETHs, shown in Figures 4A–4D. The early onset ramp we found in the count of cells with significantly different contra versus ipsi memory trial firing rates (orange line, Figure 3C) is paralleled in Figures 4A and 4B by an early onset in population firing rate difference for contra versus ipsi memory trials. Similarly, the late onset ramp in Figure 3C for nonmemory trials is paralleled in Figures 4C and 4D. We then turned to analyzing error trials. The activity on error trials (shaded pink for ipsi-instructed but contra motion, and blue for contra-instructed but ipsi motion; Figures 4A–4D) Neuron Cortical Basis of Memory-Guided Orienting in Rats A B C D E F Figure 4. Predictive Coding of Contra- and Ipsilateral Choice in the FOF (A–D) Each panel is a population PETH showing the average Z-score normalized response on correct (thick lines, mean ± SE across neurons) and error trials (shaded, mean ± SE across neurons) where the correct response was contralateral (blue) or ipsilateral (red) to the recorded neuron. PETHs are aligned to the time of the Go signal (center LED offset). (A) The average responses of memory trials for 50 contra-preferring neurons. Vertical axis tick marks indicate Z-score value. The average firing rate across all cells used for Z-score normalization is shown next to the Z = 0 mark (8.2 spikes/s). This overall mean ± the across-cell average of the PETH standard deviation are shown at the Z = ±1 marks. They indicate a typical firing rate modulation of 7.2 spikes/s. (B) The average responses of memory trials of 39 ipsi-preferring neurons. (C) Same as (A) but for nonmemory trials. (D) Same as (B) but for nonmemory trials. (E) Cells encode the direction of the motor response, not the identity of the cue stimulus. Scatter plots of the side-selectivity index for memory trials (orange) and nonmemory trials (green) (n = 89). (F) Histogram of the choice probability of neurons for trials where the rat was instructed to go in the cells’ preferred direction (n = 89). The dot and line indicate the mean ± 95% confidence interval of the mean. Black bars indicate individually significant neurons. White bars indicate neurons that were not individually significant. showed that, on average across the population, cells that fire more on correctly performed contra-instructed trials also fire more on erroneously performed ipsi-instructed trials; that is, these cells fire more on trials where the animal orients contralateral to the recorded side, regardless of the instruction. Similarly, ipsi preferring cells fire more on trials where the animal orients ipsilaterally, regardless of the instruction. This indicates that the firing rates of FOF cells are better correlated with the subject’s future motor response than with the instructing sensory stimulus. We quantified this observation on a cell-by-cell basis by generating a side-selectivity index (SSI) for each neuron (see Experimental Procedures for details). Positive SSIs mean that a cell fired more on contra-instructed trials. Negative SSIs mean that a cell fired more on ipsi-instructed trials. If cells encode the instruction we would expect SSIcorrect z SSIerror. But if cells encode the direction of the motor response, then we would expect SSIcorrect z "SSIerror. We first calculated the SSI focusing on the delay period of memory trials. We found that, over neurons, SSIcorrect correlates negatively with SSIerror (r = "0.42, p < 10"4), confirming that on memory trials, the delay period firing rates of FOF neurons encode the orienting choice of the rat, not the instruction stimulus. We then repeated this calculation for firing rates over the movement period (from Go signal to 0.5 s after the Go signal), for both memory (SSIcorrect and SSIerror correlation r = "0.59, p < 10"8) and nonmemory (r = "0.78, p < 10"17) trials. These negative correlations indicate Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. 335 Neuron Cortical Basis of Memory-Guided Orienting in Rats C A B D Figure 5. Trial-by-Trial Correlation of Neural and Behavioral Latency (A) Head angular velocity data from left correct memory trials in a single session. Each row is a single trial, showing head angular velocity (color-coded) as a function of time. The white dots indicate the time of the Go cue, and the green dots indicate the Response Onset time. Left panel: Trials are sorted by reaction time (Response Onset-Go cue). Right panel: same trials, after each trial has been time-shifted to maximize the similarity between the trial’s angular velocity profile and the average of all the other trials. (B) Same trials as in (A), but color code here indicates firing rate of a single neuron. Left panel is before alignment. Right panel is after time-alignment to maximize the similarity between each trial’s firing rate profile and the average of all the other trials. (C) Correlation between the angular velocity time offsets and the neural firing rate time offsets computed in (A) and (B). (D) Histogram of r values for 53 cells with significant delay period activity that were recorded during sessions where head-tracking was also recorded. Black bars indicate individual cells with correlations significantly greater than zero. The dot with the line through it shows the mean ± SE of r values for the population. See also Figure S5. that the FOF is again encoding the motor choice of the rat. We summarized the observations from both the delay and movement periods by calculating the SSI for the entire period, from "1.5 s before to 0.5 s after the Go cue. This again resulted in negative SSIcorrect and SSIerror correlations for both memory (r = "0.49, p < 10"5) and nonmemory (r = "0.59, p < 10"8) trials (Figure 4E). Overall, then, the firing rates of FOF neurons encode the orienting choice of the rat, not the instruction stimulus. If the delay period activity in the FOF subserves the planning of an orienting movement, then variation in that activity should lead to variation in behavior, even when the instruction stimulus is held constant (Riehle and Requin, 1993). One measure of trialto-trial covariation between neuronal signals and choice behavior is choice probability (Britten et al., 1996), which quantifies the probability that an ideal observer of the neuron’s firing rate would correctly predict the choice of the subject. We computed the choice probability for firing rates of delay period cells. For each cell, we focused on the last 400 ms of the delay period, using only memory trials in which the instruction was to orient to the cell’s preferred side. Consistent with the SSI delay period analysis, we found that an ideal observer would, on average, correctly predict the rat’s side port choice 64% of the time. The cell population is strongly skewed above the chance 336 Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. prediction value of 0.5, with 75% of cells having a choice probability value above 0.5 (Figure 4F). Twenty-seven percent of cells had choice probability values that were, individually, significantly above chance (permutation text, p < 0.05). We used red and blue LEDs, placed on the tetrode recording drive headstages of the electrode-implanted rats, to perform video tracking of the rats’ head location and orientation (Neuralynx; MT). Two thirds of the delay period neurons (53/89) were recorded in sessions in which head tracking data was also obtained. Figure 5A shows an example of head angular velocity data for left memory trials in one of the sessions, aligned to the time of the Go signal. There is significant trial-to-trial variability in the latency of the peak angular velocity as the animal responds to the Go signal and turns toward a side port to report its choice. As shown in data from the example cells of Figure 3, and an example cell in Figure 5B, many neurons with delay period responses also fire strongly during the movement period, and the latency of each neuron’s movement period firing rate profile can vary significantly from trial to trial. To quantitatively estimate latencies on each trial, we used an iterative algorithm that finds, for each trial, the latency offset that would best align that trial with the average over all the other trials (Figures 5A and 5B; see Experimental Procedures for details). Firing rate Neuron Cortical Basis of Memory-Guided Orienting in Rats A B Figure 6. Rats Plan Their Response During the Delay Period on Memory Trials (A) Movement times (MT) are faster for memory trials than nonmemory trials. Movement times are measured as median Response time " Response Onset time for each physiology session. The mean difference between memory and nonmemory trials is 47 ms (t test, t141 = 3.58, p < 10"5 from the five electrode implanted rats). The dot above the histogram indicates the mean ± SE of the distribution. (B) Average head-angle data from 84 recording sessions. Thin lines are 200 example trials randomly subsampled from all 84 sessions. The thick lines are the average across all trials across all sessions. In our coordinate system, 4 = 0# points directly toward the center port, positive 4 corresponds to rightward orientations, and negative 4 to leftward orientations. On memory trials one can observe a subtle but clear change in the head angle in the direction of the response during the delay period, starting around 500 ms before the end of the fixation period. See also Figure S6 for head-direction related neural activity. latencies and head velocity latencies were estimated independently of each other using this algorithm. We then computed, for each neuron, the correlation between the two latency estimates (e.g., Figure 5C). We focused this analysis on correct contralateral memory trials of delay period neurons (as in Riehle and Requin, 1993). Of 53 delay period cells analyzed, 23 of them (43%) showed significant trial-by-trial correlations between neural and behavioral latency (Figure 5D). Furthermore, as a population, the 53 cells were significantly shifted toward positive correlations (mean ± SE, 0.36 ± 0.05, t test p < 10"8). We concluded that a significant fraction of delay period neurons not only have firing rates that predict the direction of motion before it occurs (Figure 4F), but in addition, once the motion has begun, the timing of their firing rate profile is strongly correlated with the timing of the execution of the movement. Delay Period Firing Rates Cannot Be Explained As Encoding Head Direction On memory trials, the subject has many hundreds of milliseconds to plan a motor response in advance of the go signal. We examined the behavioral data for evidence of planning, and found it in two forms: faster reaction times on memory trials, and head angle adjustments during the fixation period. With respect to reaction time, we found that the time from exiting the central port until reaching the side port was, on average, 47 ms shorter on memory trials compared to nonmemory trials (t test,t141 = 3.58, p < 10"5; Figure 6A). This is consistent with the idea that prepared movements take less time to initiate and/or execute. We then asked whether there were any consistent head direction adjustments during the fixation period that would predict subsequent orienting motion choices. Figure 6B plots 4(t), the head angle as a function of time aligned to the Go signal, for both left-orienting and right-orienting trials. As can be seen from the average 4(t) for each of these two groups, during the delay period of memory trials, rats tended to gradually and slightly turn their heads toward their intended motion direction, even while keeping their nose in the center port. At the time of the Go signal, 4(t = 0), the rats’ heads had already turned, on average, !4# in the direction of the intended response. We used ROC analysis at each time point t to quantify whether the distribution of 4(t) for trials where the animal ultimately oriented left was significantly different from the distribution for trials where the animal ultimately oriented right. We found that, on average, 4(t) allowed a significantly above-chance prediction of the rat’s choice 444 ± 29 ms before the Go signal (mean ± SE) on memory trials, and 19 ± 26 ms before the Go signal on nonmemory trials. We also found that on some sessions (8/80, 10%) 4(t) was not predictive of choice at any time point before the Go signal, even while percent correct performance and neural delay period activity was normal in these sessions. This showed that preliminary head movements were not performed by all rats in all sessions, and suggested that preliminary head movements may not be necessary for performance of the task. Firing rates of some neurons in rat FOF have been previously described as encoding head-direction responses (Mizumori et al., 2005). That is, the firing rates of some FOF neurons were a function of the allocentric orientation of the animal’s head (Taube, 2007). Our recordings replicated this observation (Figure S6). Our data further revealed that head direction tuning in the FOF was significantly affected by behavioral context: for many cells the preferred direction depended on whether the animal was engaged versus not engaged in performing the task (Figure S6). Here, the observation of head direction tuning in the FOF, together with the data of Figure 6B, immediately raised the question of whether delay period firing rates could predict the rat’s choice merely by virtue of encoding the current head orientation 4 (that, as shown in Figure 6B, is itself predictive of the rat’s choice). To address this question in a quantitative manner that did not depend on an in-task versus out-of-task comparison or distinction, we took advantage of existing variability in 4 during the fixation period. We first reperformed the analysis of Figure 3A, but now restricting it to neurons recorded in sessions where head-tracking data was also recorded. We divided trials into two groups, based on the sign of 4 at t = +0.6 s after the Go Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. 337 Neuron Cortical Basis of Memory-Guided Orienting in Rats A B C Figure 7. Predictive Coding of Response Is Not a Simple Function of Current Head Angle (A) Plot of head angle as a function of time relative to the Go signal for memory trials. Thin blue lines are from a random subsample of trials where the head angle was >0 (oriented leftward from center port) at time t = +0.6 s relative to the Go signal, as indicated by the vertical dotted line. Thin red lines are from a random subsample of trials where the head angle was <0 at t = +0.6 s. Thick lines are the mean head angles for each group, averaged over all correct memory trials. (B) ROC plot (similar to Figure 5B) for the trial grouping defined in (A). (C) As in (A), but with groupings defined by the sign of the head angle at t = "0.9 s relative to the Go signal. (D) ROC plot for the trial grouping defined in (C). See Figure S7 for similar analyses using angular velocity and acceleration. D signal (shown in Figure 7A as traces in blue 4(0.6) > 0, and red 4(0.6) < 0). These two groups are essentially identical to the ‘‘ultimately went Left’’ and ‘‘ultimately went Right’’ groups of Figure 6B, but redefining them in terms of the sign of 4(t) will prove convenient below. We counted the percentage of neurons that had firing rates that significantly discriminated between these two 4(0.6) > 0 and 4(0.6) < 0 groups. The result, essentially replicating that of Figure 3A for the subset of sessions with head tracking data, is shown in Figure 7B. At the time of the Go signal (t = 0), 21% of cells significantly discriminated 4(0.6) > 0 versus 4(0.6) < 0 trials. At this same time point (t = 0), the mean difference in 4 for the two groups of trials was !8# . In other words, if FOF firing rates simply encode current head angle, an 8# head direction signal should produce a detectable firing rate change in !21% of cells. We then performed the same analysis, but this time based on the sign of 4 at t = "0.9 s before the Go signal (traces in blue for 4("0.9) > 0, and red for 4("0.9) < 0 in Figure 7C). At t = "0.9 s, the mean difference in 4 for this new grouping of trials was !8# , very similar to the difference at t = 0 s for the previous grouping (compare Figures 7A and 7C). However, only 5% of cells discriminated between the two groups at t = "0.9 s (Figure 7D). This is in strong contrast to the 21% that we would have expected if FOF neurons encoded head angle. We concluded that encoding of head angle was not sufficient to explain the FOF delay period firing rates that predict orienting choice. We repeated this analysis with angular head velocity 40 (t) (Figures S7A–S7D), and with angular head acceleration 400 (t) (Figures S7E–S7H) and found that, as with head angle, neither angular head velocity nor angular head acceleration could explain choice-predictive delay period firing rates. We also performed a regression analysis, fitting the firing rate of each cell on each trial, f(t), as a linear function of angular position, velocity, and acceleration (f(t) = b1 3 4(t) + b2 3 40 (t) + b3 3 400 (t) + r(t); 338 Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. see Supplemental Experimental Procedures for details). The residuals r(t) have had any linear effects of head angular position, velocity, and/or acceleration eliminated. At each time point, we used ROC analysis to test whether the distributions of residuals r(t) for ipsilateral versus contralateral trials were different, and as in Figure 3C, we counted the number of neurons for which this difference was significant. We found that only a small portion of the delay period activity could be accounted for by a combination 4(t), 40 (t), and 400 (t) (Figure S7I). DISCUSSION To investigate the contribution of the rat FOF (studies centered at +2 AP, ±1.3 ML mm from Bregma) to the preparation of orienting motions, we trained rats on a two-alternative forced-choice memory-guided auditory discrimination task. Subjects were presented with an auditory cue that indicated which way they should orient to obtain a reward. However, the subjects were only allowed to make their motor act to report a choice after a delay period had elapsed. The task thus separates the stimulus from the response in the tradition of classic memory-guided tasks (Mishkin and Pribram, 1955; Fuster, 1991; Goldman-Rakic et al., 1992). We carried out unilateral reversible inactivations of the FOF, M1, and the whiskers, recorded extracellular neural spiking signals from the FOF, and tracked head position and orientation, while rats were performing the task. The resulting data provide several lines of evidence supporting the hypothesis that the FOF plays a role in memory-guided orienting. First, unilateral inactivation of the FOF produced an impairment of contralateral orienting trials that was substantially greater for memory trials as compared to nonmemory trials (Figure 2). Control performance on both memory and nonmemory trials was very similar (Figure 1 and related text), suggesting that the differential impairment was not due to a difference in task difficulty, but instead reveals a memory-specific role of FOF activity Neuron Cortical Basis of Memory-Guided Orienting in Rats in contralateral orienting. Second, we found robust neural firing rates during the delay period (after the offset of the stimulus and before the Go cue) that differentiated between trials in which the animal ultimately responded by orienting contralaterally from those where it responded by orienting ipsilaterally (Figure 3 and Figure 4). Third, we found trial-by-trial correlations between neural firing and behavior, both for firing rates during the delay period (Figure 4H) and for neural response latency during periods that included the subjects’ choice-reporting motion. (Figure 5). Several groups studying the neural basis of movement preparation (Riehle and Requin, 1993; Dorris and Munoz, 1998; Steinmetz and Moore, 2010; Curtis and Connolly, 2008) have agreed upon three operational criteria for interpreting neural activity as being a neural substrate for movement preparation: (1), changes in neural activity must occur during the delay period, before the Go signal; (2), the neural activity must show response selectivity (e.g., fire more for contralateral than ipsilateral responses); (3), there must be a trial-by-trial relationship between neural activity and some metric of behavior (usually reaction time, but since our task was not a reaction time task we used choice probability). Our results satisfy all three of these criteria, so interpreting the activity in the FOF as ‘‘movement preparation’’ is, at least, consistent with prior work. There are several possible interpretations as to what component(s) of response preparation FOF neurons might encode: do they represent a motor plan? A memory of the identity of the motor plan? Attention? Intention? (Bisley and Goldberg, 2010; Glimcher, 2003; Goldman-Rakic et al., 1992; Schall, 2001; Thompson et al., 2005; Gold and Shadlen, 2001). Our data do not discriminate between these possibilities. Nevertheless, we conclude that, as in the primate, there exists in the rat frontal cortex a structure that is involved in the preparation and/or planning of orienting responses. An area with such a role may be conserved across multiple species, including birds (Knudsen et al., 1995). Since FOF delay period firing rates are better correlated with the upcoming motor act than with the initial sensory cue (Figure 4), our data do indicate that FOF neurons are not likely to encode a memory of the auditory stimulus itself. Furthermore, in memory trials, some form of memory is required immediately after the end of the auditory instruction stimulus. We did not observe a short-latency sensory response in the FOF, but instead observed a slow and gradual development of choicedependent activity during the delay period. This suggests that FOF neurons do not support the early memory the task requires. The FOF is strongly interconnected with the posterior parietal cortex (PPC) (Reep and Corwin, 2009; Nakamura, 1999) and with the medial prefrontal cortex (mPFC, Condé et al., 1995). We suggest both of these areas as candidates for supporting the early memory aspects of the task, perhaps even including the transformation from a continuous auditory signal (clickrate) to a binary choice (plan-left/plan-right). Based on data from an orienting task driven by olfactory stimuli, Felsen and Mainen (2008) recently proposed that the superior colliculus (SC) may play a broad role in sensory-guided orienting. Projections to the SC from the FOF (Leonard, 1969; Künzle et al., 1976; Reep et al., 1987), together with our current data, suggest that the FOF may be an important contributor to orienting-related activity in the SC. As in the primate, orienting behavior in the rodent is likely to be subserved by a network of interacting brain areas. The relative roles and mutual interactions between the FOF, PPC, mPFC, and SC (and possibly other areas, including the basal ganglia) during orienting behaviors in the rat remain to be elucidated. We focused our analyses here on the response-selective delay period activity of FOF neurons. However, we also found neurons carrying a wide variety of other task related neural signals, including ramping during the delay that was not responseselective (consistent with a general timing or anticipatory signal), sustained firing rate increases or decreases during the fixation period, and activity after the reward/error signal. Detailed descriptions of these neural responses are outside the scope of this manuscript and will be reported elsewhere. If we think of visual saccades as orienting responses, the results presented here from the rat FOF are, qualitatively speaking, consistent with results from monkey FEF studies of memory-guided saccades. Muscimol inactivation of FEF strongly impairs memory-guided contralateral saccades, but leaves visually guided and ipsilateral saccades relatively intact (Sommer and Tehovnik, 1997; Dias and Segraves, 1999; Keller et al., 2008). Similarly, we found that muscimol inactivation of rat FOF strongly impaired memory-guided contralateral orienting, had a weaker effect on nonmemory contralateral orienting, and spared ipsilateral orienting (Figure 2). However, FEF inactivation also increases reaction times of contralateral saccades and increases the rate of premature ipsilateral responses, two results that we failed to replicate. Recordings from monkey FEF show robust spatially selective delay period activity in memory-guided saccade tasks (Bruce and Goldberg, 1985; Schall and Thompson, 1999) for both ipsilateral and contralateral saccades (Lawrence et al., 2005), similar to the spatially-dependent activity we observed in rat FOF neurons (Figures 3 and 4). In typical visualguided saccade tasks a substantial portion of FEF neurons show responses to the onset of the stimulus (c.f. Schall et al., 1995), which we did not observe in our auditory-stimulus task. However, monkey FEF neurons also encode saccade vectors preceding auditory-guided saccades (Russo and Bruce, 1994), and show very little auditory-stimulus-driven activity. This again is similar to our observations in rat FOF (Figures 4A and 4B). We note that although we have focused here on similarities to the monkey FEF, which is a particularly well-studied brain area, we do not believe we have established a strict homology between rat FOF and monkey FEF. Similarities to other cortical motor structures may be greater, or it may be that the rat FOF will not have a strict homology with any one primate cortical area. We are aware of only one other electrophysiological study in rats during a memory-guided orienting task in which rats stay still during the delay period (Gage et al., 2010). In that study, Gage et al. (2010) recorded from M1, striatum, and globus pallidus. They found that, although a few response-selective signals in M1 could be observed many hundreds of milliseconds before the Go signal, maintained response selectivity in M1 neurons arose only !180 ms before the Go signal. In contrast, once neurons of the FOF start firing in a response-selective manner, they usually maintain their response selectivity throughout the rest of the delay period (Figure 3), even when their response selectivity arises many hundreds of milliseconds before the Go Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. 339 Neuron Cortical Basis of Memory-Guided Orienting in Rats signal. The population count of response selective FOF cells therefore starts rising very shortly after the end of the instruction signal, and rises continually until the Go signal (Figure 4F; compare to Figure 5B, top panel, of Gage et al., 2010). This suggests that orienting preparation signals are represented significantly earlier in the FOF than in M1. Consistent with the much weaker electrophysiological delay period signature found in M1, as compared to the FOF, unilateral pharmacological inactivations of M1 produced very different, and much weaker, behavioral effects than those found in FOF (Figure S2, compare to Figure 2). The difference is particularly strong for memory trials. FOF inactivation reduced contralateral memory trials to almost 50% correct performance (chance), but M1 inactivation impaired performance on these trials only to !75% correct. This was a saturated effect: doubling the dose of muscimol in M1 did not further impair performance (Figure S2). Much further work is required to draw and refine functional maps of the rat cortex during awake behaviors, but we do conclude that the role of the FOF in memory-guided orienting is not common across frontal motor cortex. We targeted the FOF based on previous anatomical, lesion, and microstimulation studies that suggested a role for this area in orienting behaviors (Leonard, 1969; Cowey and Bozek, 1974; Crowne and Pathria, 1982; Sinnamon and Galer, 1984; Corwin and Reep, 1998). However, a different line of research, observing whisker movements in response to intracortical microstimulation in head-fixed, anesthetized rats, has described the same area as whisker motor cortex (Brecht et al., 2004). Nevertheless, the functional role of the FOF in awake animals is not firmly established: single-unit recordings from the area in awake animals remain very sparse (Carvell et al., 1996; Kleinfeld et al., 2002; Mizumori et al., 2005). We asked whether whisking played a role in our memory-guided orienting task, and found that it did not: removing the whiskers had little effect on performance (Figures 1D and 1F and associated text), unilaterally paralyzing the whiskers did not produce a lateralized or memory specific effect (Figures 1E and 1F), and video analysis of regular trials did not find evidence of asymmetric or lateralized whisking during the memory delay period. The video showed instead that whiskers are held quite still during the delay period (Figure S1 and Movie S2). We speculate that well-trained animals that are highly familiar with the spatial layout of the behavior apparatus do not use whisking to guide their movements during the task. In particular, whisking appears to play no role in the short-term memory component of the task (Movie S2). The lack of whisker-related effects on task performance or task behavior contrasts with the strong pharmacological and electrophysiological correlates with behavior that form the basis of this report, and suggests that the FOF plays a role in orienting that is independent from any role in control of whisking. Previous single-unit studies of this area in awake animals, focusing on whisker motor control, have suggested that the FOF is not primarily involved in low-level motor control of whisking, but may instead play a more prominent role in longer timescale (!1 s or longer) control of whisking parameters (Carvell et al., 1996). More recent studies (D. Kleinfeld, personal communication) have identified some of the long timescale parameters as control of amplitude and offset angle of whisking; this last refers to the 340 Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. average orientation of the whiskers with respect to the head. Our data, by providing evidence that the FOF participates in the preparation of orienting movements many hundreds of milliseconds before these movements actually occur, is consistent with this view of the FOF as a high-level motor control area. A third line of research in this cortical area, represented so far only by a book chapter (Mizumori et al., 2005), has described finding head direction cells (Taube, 2007) in the FOF. Our recordings replicated this finding (Figure S6). We found no correlation between the strength of a neuron’s head direction tuning and the strength of its preparatory orienting signals (data not shown). The two types of signals coexist in the FOF, but are distinct from each other: a quantitative analysis showed that head direction tuning could not account for the preparatory orienting signals recorded during the delay period of memory trials (Figure 7). We found that head direction signals in the FOF are strongly modulated by behavioral context. That is, for many cells, tuning while animals were performing the task was very different to tuning while animals were not performing the task (Figure S6). The relationship between orienting preparation signals and head direction signals in the FOF is complex, and we will explore it in detail in a future manuscript. The confluence of three different types of signals (orienting, head direction, whisking) in a single area is remarkable. Although different, the signals are related: head direction information is important for making orienting decisions, whisking reaps information from the environment that can then be used to guide orienting decisions, and orienting movements themselves will have a direct effect on both head direction and whisker position. Having these three signals represented in a single area is consistent with the view of the FOF as an area that integrates multiple sources of information in the service of high-level control of spatial behavior. Elucidating the precise relationship between these signals, both in the FOF and in other brain areas, will require many further experiments that will bring together the orienting, navigation, and whisking literature. EXPERIMENTAL PROCEDURES Subjects Animal use procedures were approved by the Princeton University Institutional Animal Care and Use Committee and carried out in accordance with National Institutes of Health standards. All subjects were male Long-Evans rats (Taconic, NY). Rats were placed on a restricted water schedule to motivate them to work for water reward. Behavior Rats went through several stages of an automated training protocol before performing the task as described in the results (see Supplemental Experimental Procedures). All data described in this study were collected from fully trained rats. Sessions with poor performance (<70% correct overall or fewer than 8 correct memory trials on each side without fixation violations) were excluded from analyses. These sessions were rare (2.4% of all sessions from trained rats) and were usually caused by problems with the hardware (e.g., a clogged water-reward valve or a dirty IR-photodetector). To generate psychometric curves, we collected 12 data points: the % ‘‘Went Right’’ for each of six different click rates, separately for memory and for nonmemory trials. We then combined the data points across all sessions (total data points per fit = 6 3 # of sessions) and used MATLAB nlinfit.m to fit a 4-parameter sigmoid to the data. For these fits, x is the natural logarithm of clicks/sec, y is ‘‘% Went Right,’’ and the four parameters to be fit are: x0, Neuron Cortical Basis of Memory-Guided Orienting in Rats the inflection point of the sigmoid, b, the slope of the sigmoid, y0, the minimum % Went Right, and a + y0 is the maximum % Went Right. y = y0 + 1+e a "ðx " x0Þ b Data from memory and nonmemory trials were fit separately. Surgery All surgeries were done under isoflurane anesthesia (1.5%–2%) using standard stereotaxic technique (see Supplemental Experimental Procedures for details). The target of all FOF surgeries in our Long-Evans strain rats was +2 AP, ±1.3 ML (mm from Bregma). This location was chosen because it was the center of the distribution of stimulation sites that resulted in contralateral orienting movement in Sinnamon and Galer (1984). Infusions Dose and volume of muscimol infusions into FOF was 0.5 mg/mL and 0.3 ml, respectively. Infusions for M1 were done in two sets of experiments, first 0.5 mg/mL and 0.3 mL, then 1 mg/mL and 0.3 mL. See Supplemental Experimental Procedures for details. Recordings Recordings were made with platinum iridium wire (16.66 mm, California Fine Wire, CA) twisted into tetrodes. Wires were gold-plated to 0.5–1.2 MOhm. Spike sorting was done by hand using SpikeSort3D (Neuralynx). Cells had to satisfy several criteria to be included in the presented analyses: 1), zero interspike intervals <1 ms; 2), signal to noise ratio >4; and 3), at least one time point of a smoothed, response-aligned PETH had to have a firing rate of at least 3 spikes/s. We recorded 378 cells over 100 sessions that satisfied the first two criteria. A total of 242 cells (recorded from 91 sessions) satisfied all three criteria. Median number of cells per session was three. The maximum number of cells recorded in a session was 11. Neural Data Analysis We examined a 2 s window around the Go signal ("1.5 s pre, to +0.5 s post). Spikes from each trial were smoothed with a causal half-Gaussian kernel with a full-width SD of 200 ms—that is, the firing rate reported at time t averages over spikes in an !200-ms-long window preceding t. The resulting smooth traces were sampled every 10 ms. To determine whether cells were response-selective at any point between the stimulus and the rat’s choice, we divided correctly performed trials into contralateral-orienting and ipsilateral-orienting groups, and used ROC analysis at each time point to ask whether the firing rates of the two groups were significantly different for that time point. For each cell, we randomly shuffled ipsi and contra trial labels 2000 times and recomputed ROC values. We labeled individual time bins as significant if fewer than 1% of the shuffles produced ROC values for that time bin that were further from chance (0.5) than the original data was (i.e., p < 0.01 for each time bin). We then counted the percentage of shuffles that produced a number of significant bins greater than or equal to the number of bins labeled significant in the original data. If this randomly produced percentage was less than 5%, the cell as a whole was labeled significant (i.e., an overall p < 0.05 for each cell). To determine the time at which the population count of significant cells became greater than chance, we used binomial statistics. These indicate that with probability 0.999, at any given time point, an individual cell threshold of p < 0.01 would lead to fewer than 8/242 cells being labeled significant by chance. The population count was designated as significantly different from chance when it went above this p < 0.001 population threshold. In order to quantify whether neurons in FOF tended to encode the stimulus or the response we generated a stimulus selectivity index (SSI) from Go aligned PETHs for correct and error trials as follows: SSItt = 0:5 P t = "1:5 0:5 P t = "1:5 PETHcontra;tt " PETHipsi;tt PETHcontra;tt + PETHipsi;tt where tt indicates trial type (correct-memory, correct-nonmemory, errormemory, and error-nonmemory). If a cell fired only on contra and not on ipsi trials, then SSI = 1. If a cell fired on ipsi and not contra trials, then SSI = "1. If a cell fired equally for ipsi and contra trials then SSI = 0. For latency estimations, we used an alignment algorithm to find a relative temporal offset for each trial as follows. Given a signal as a function of time for each trial (either firing rate or head angular velocity), we computed the trial-averaged signal. For each trial we then found the time of the peak of the cross-correlation function between the signal for that trial and the trial-averaged signal. We then shifted each trial accordingly, and recomputed the trial-averaged signal after. We iterated this process until the variance of the trial-averaged signal converged, typically within fewer than five iterations. The output of this alignment procedure was an offset time for each trial, which indicated the relative latency for that trial. Histology In all cases, the electrode and cannula placements in FOF were within the borders of M2 and between 2 and 3 mm anterior to Bregma (Paxinos and Watson, 2004). In all cases the M1 placements were within the borders of M1 and between 2.5 and 3.5 mm anterior to Bregma (Paxinos and Watson, 2004). SUPPLEMENTAL INFORMATION Supplemental Information includes Supplemental Experimental Procedures, seven figures, and three movies and can be found with this article online at doi:10.1016/j.neuron.2011.07.010. ACKNOWLEDGMENTS We thank B.W. Brunton and J.K. Jun for contributions to software to obtain head direction data, D.W. Tank and J.P. Rickgauer for suggestions to improve whisker tracking, B.W. Brunton, J.K. Jun, C.D. Kopec, and T. Hanks for discussion and comments on the manuscript, A. Keller and D. Kleinfeld for discussions related to the role of the FOF in whisker control, and L. Osorio and G. Brown for technical assistance. This work was supported by the Howard Hughes Medical Institute. Accepted: July 7, 2011 Published: October 19, 2011 REFERENCES Arbib, M.A. (1997). From visual affordances in monkey parietal cortex to hippocampo-parietal interactions underlying rat navigation. Philos. Trans. R. Soc. Lond. B Biol. Sci. 352, 1429–1436. Bisley, J.W., and Goldberg, M.E. (2010). Attention, intention, and priority in the parietal lobe. Annu. Rev. Neurosci. 33, 1–21. Brecht, M., Krauss, A., Muhammad, S., Sinai-Esfahani, L., Bellanca, S., and Margrie, T.W. (2004). Organization of rat vibrissa motor cortex and adjacent areas according to cytoarchitectonics, microstimulation, and intracellular stimulation of identified cells. J. Comp. Neurol. 479, 360–373. Britten, K.H., Newsome, W.T., Shadlen, M.N., Celebrini, S., and Movshon, J.A. (1996). A relationship between behavioral choice and the visual responses of neurons in macaque MT. Vis. Neurosci. 13, 87–100. Bruce, C.J., and Goldberg, M.E. (1985). Primate frontal eye fields. I. Single neurons discharging before saccades. J. Neurophysiol. 53, 603–635. Bruce, C.J., Goldberg, M.E., Bushnell, M.C., and Stanton, G.B. (1985). Primate frontal eye fields. II. Physiological and anatomical correlates of electrically evoked eye movements. J. Neurophysiol. 54, 714–734. Carvell, G.E., Miller, S.A., and Simons, D.J. (1996). The relationship of vibrissal motor cortex unit activity to whisking in the awake rat. Somatosens. Mot. Res. 13, 115–127. Condé, F., Maire-Lepoivre, E., Audinat, E., and Crépel, F. (1995). Afferent connections of the medial frontal cortex of the rat. II. Cortical and subcortical afferents. J. Comp. Neurol. 352, 567–593. Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. 341 Neuron Cortical Basis of Memory-Guided Orienting in Rats Corwin, J.V., and Reep, R.L. (1998). Rodent posterior parietal cortex as a component of a cortical network mediating directed spatial attention. Psychobiology 26, 87–102. Künzle, H., Akert, K., and Wurtz, R.H. (1976). Projection of area 8 (frontal eye field) to superior colliculus in the monkey. An autoradiographic study. Brain Res. 117, 487–492. Cowey, A., and Bozek, T. (1974). Contralateral ‘neglect’ after unilateral dorsomedial prefrontal lesions in rats. Brain Res. 72, 53–63. Lawrence, B.M., White, R.L., 3rd, and Snyder, L.H. (2005). Delay-period activity in visual, visuomovement, and movement neurons in the frontal eye field. J. Neurophysiol. 94, 1498–1508. Crowne, D.P., and Pathria, M.N. (1982). Some attentional effects of unilateral frontal lesions in the rat. Behav. Brain Res. 6, 25–39. Crowne, D.P., Richardson, C.M., and Dawson, K.A. (1986). Parietal and frontal eye field neglect in the rat. Behav. Brain Res. 22, 227–231. Curtis, C.E., and Connolly, J.D. (2008). Saccade preparation signals in the human frontal and parietal cortices. J. Neurophysiol. 99, 133–145. Dias, E.C., and Segraves, M.A. (1999). Muscimol-induced inactivation of monkey frontal eye field: effects on visually and memory-guided saccades. J. Neurophysiol. 81, 2191–2214. Donoghue, J.P., and Wise, S.P. (1982). The motor cortex of the rat: cytoarchitecture and microstimulation mapping. J. Comp. Neurol. 212, 76–88. Dorris, M.C., and Munoz, D.P. (1998). Saccadic probability influences motor preparation signals and time to saccadic initiation. J. Neurosci. 18, 7015–7026. Felsen, G., and Mainen, Z.F. (2008). Neural substrates of sensory-guided locomotor decisions in the rat superior colliculus. Neuron 60, 137–148. Ferrier, D. (1875). Experiments on the brain of monkeys. Proc. R. Soc. Lond. 23, 409–430. Funahashi, S., Bruce, C.J., and Goldman-Rakic, P.S. (1991). Neuronal activity related to saccadic eye movements in the monkey’s dorsolateral prefrontal cortex. J. Neurophysiol. 65, 1464–1483. Fuster, J.M. (1991). The prefrontal cortex and its relation to behavior. Prog. Brain Res. 87, 201–211. Gage, G.J., Stoetzner, C.R., Wiltschko, A.B., and Berke, J.D. (2010). Selective activation of striatal fast-spiking interneurons during choice execution. Neuron 67, 466–479. Glimcher, P.W. (2003). The neurobiology of visual-saccadic decision making. Annu. Rev. Neurosci. 26, 133–179. Gold, J.I., and Shadlen, M.N. (2001). Neural computations that underlie decisions about sensory stimuli. Trends Cogn. Sci. (Regul. Ed.) 5, 10–16. Leichnetz, G.R., Hardy, S.G., and Carruth, M.K. (1987). Frontal projections to the region of the oculomotor complex in the rat: a retrograde and anterograde HRP study. J. Comp. Neurol. 263, 387–399. Leonard, C.M. (1969). The prefrontal cortex of the rat. I. Cortical projection of the mediodorsal nucleus. II. Efferent connections. Brain Res. 12, 321–343. McNaughton, B.L., Battaglia, F.P., Jensen, O., Moser, E.I., and Moser, M.B. (2006). Path integration and the neural basis of the ‘cognitive map’. Nat. Rev. Neurosci. 7, 663–678. Mishkin, M., and Pribram, K.H. (1955). Analysis of the effects of frontal lesions in monkey. I. Variations of delayed alternation. J. Comp. Physiol. Psychol. 48, 492–495. Mizumori, S.J.Y., Puryear, C.B., Gill, K.K., and Guazzelli, A. (2005). Head direction codes in hippocampal afferent and efferent systems: What function do they serve? In Head Direction Cells and the Neural Mechanisms Underlying Directional Orientation, W. Wiener and J.S. Taube, eds. (Cambridge: MIT Press), pp. 203–220. Monteon, J.A., Constantin, A.G., Wang, H., Martinez-Trujillo, J.C., and Crawford, J.D. (2010). Electrical stimulation of the frontal eye fields in the head-free macaque evokes kinematically normal 3D gaze shifts. J. Neurophysiol. 104, 3462–3475. Moser, E.I., Kropff, E., and Moser, M.B. (2008). Place cells, grid cells, and the brain’s spatial representation system. Annu. Rev. Neurosci. 31, 69–89. Munoz, D.P., and Wurtz, R.H. (1992). Role of the rostral superior colliculus in active visual fixation and execution of express saccades. J. Neurophysiol. 67, 1000–1002. Nakamura, K. (1999). Auditory spatial discriminatory and mnemonic neurons in rat posterior parietal cortex. J. Neurophysiol. 82, 2503–2517. Narayanan, N.S., and Laubach, M. (2006). Top-down control of motor cortex ensembles by dorsomedial prefrontal cortex. Neuron 52, 921–931. Goldman-Rakic, P.S., Bates, J.F., and Chafee, M.V. (1992). The prefrontal cortex and internally generated motor acts. Curr. Opin. Neurobiol. 2, 830–835. Neafsey, E.J., Bold, E.L., Haas, G., Hurley-Gius, K.M., Quirk, G., Sievert, C.F., and Terreberry, R.R. (1986). The organization of the rat motor cortex: a microstimulation mapping study. Brain Res. 396, 77–96. Green, D.M., and Swets, J.A. (1974). Signal Detection Theory and Psychophysics (Huntington, NY: R.E. Krieger Publishing). Olton, D.S., and Samuelson, R.J. (1976). Remembrance of places passed: spatial memory in rats. J Exp Psychol Anim Behav Process 2, 97–116. Guandalini, P. (1998). The corticocortical projections of the physiologically defined eye field in the rat medial frontal cortex. Brain Res. Bull. 47, 377–385. Paxinos, G., and Watson, C. (2004). The Rat Brain in Stereotaxic Coordinates (New York: Elsevier). Hebb, D.O., and Penfield, W. (1940). Human behavior after extensive bilateral removal from the frontal lobes. Arch. Neurol. Psychiat. 41, 421–438. Reep, R.L., and Corwin, J.V. (2009). Posterior parietal cortex as part of a neural network for directed attention in rats. Neurobiol. Learn. Mem. 91, 104–113. Hull, C.L. (1932). The goal gradient hypothesis and maze learning. Psychol. Rev. 39, 25–43. Reep, R.L., Corwin, J.V., Hashimoto, A., and Watson, R.T. (1984). Afferent connections of medial precentral cortex in the rat. Neurosci. Lett. 44, 247–252. Kargo, W.J., Szatmary, B., and Nitz, D.A. (2007). Adaptation of prefrontal cortical firing patterns and their fidelity to changes in action-reward contingencies. J. Neurosci. 27, 3548–3559. Reep, R.L., Corwin, J.V., Hashimoto, A., and Watson, R.T. (1987). Efferent connections of the rostral portion of medial agranular cortex in rats. Brain Res. Bull. 19, 203–221. Keller, E.L., Lee, K.M., Park, S.W., and Hill, J.A. (2008). Effect of inactivation of the cortical frontal eye field on saccades generated in a choice response paradigm. J. Neurophysiol. 100, 2726–2737. Riehle, A., and Requin, J. (1993). The predictive value for performance speed of preparatory changes in neuronal activity of the monkey motor and premotor cortex. Behav. Brain Res. 53, 35–49. Kleinfeld, D., Sachdev, R.N., Merchant, L.M., Jarvis, M.R., and Ebner, F.F. (2002). Adaptive filtering of vibrissa input in motor cortex of rat. Neuron 34, 1021–1034. Russo, G.S., and Bruce, C.J. (1994). Frontal eye field activity preceding aurally guided saccades. J. Neurophysiol. 71, 1250–1253. Kleinfeld, D., Ahissar, E., and Diamond, M.E. (2006). Active sensation: insights from the rodent vibrissa sensorimotor system. Curr. Opin. Neurobiol. 16, 435–444. Knudsen, E.I., Cohen, Y.E., and Masino, T. (1995). Characterization of a forebrain gaze field in the archistriatum of the barn owl: microstimulation and anatomical connections. J. Neurosci. 15, 5139–5151. 342 Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. Schall, J.D. (2001). Neural basis of deciding, choosing and acting. Nat. Rev. Neurosci. 2, 33–42. Schall, J.D., and Thompson, K.G. (1999). Neural selection and control of visually guided eye movements. Annu. Rev. Neurosci. 22, 241–259. Schall, J.D., Morel, A., King, D.J., and Bullier, J. (1995). Topography of visual cortex connections with frontal eye field in macaque: convergence and segregation of processing streams. J. Neurosci. 15, 4464–4487. Neuron Cortical Basis of Memory-Guided Orienting in Rats Schiller, P.H., and Tehovnik, E.J. (2005). Neural mechanisms underlying target selection with saccadic eye movements. Prog. Brain Res. 149, 157–171. Taube, J.S. (2007). The head direction signal: origins and sensory-motor integration. Annu. Rev. Neurosci. 30, 181–207. Sinnamon, H.M., and Galer, B.S. (1984). Head movements elicited by electrical stimulation of the anteromedial cortex of the rat. Physiol. Behav. 33, 185–190. Thompson, K.G., Biscoe, K.L., and Sato, T.R. (2005). Neuronal basis of covert spatial attention in the frontal eye field. J. Neurosci. 25, 9479–9487. Sommer, M.A., and Tehovnik, E.J. (1997). Reversible inactivation of macaque frontal eye field. Exp. Brain Res. 116, 229–249. Steinmetz, N.A., and Moore, T. (2010). Changes in the response rate and response variability of area V4 neurons during the preparation of saccadic eye movements. J. Neurophysiol. 103, 1171–1178. Stuesse, S.L., and Newman, D.B. (1990). Projections from the medial agranular cortex to brain stem visuomotor centers in rats. Exp. Brain Res. 80, 532–544. Tolman, E.C. (1938). The determiners of behavior at a choice point. Psychol. Rev. 45, 1–41. Whitlock, J.R., Sutherland, R.J., Witter, M.P., Moser, M.B., and Moser, E.I. (2008). Navigating from hippocampus to parietal cortex. Proc. Natl. Acad. Sci. USA 105, 14755–14762. Zilles, K.J. (1985). The Cortex of the Rat: A Stereotaxic Atlas (Berlin: SpringerVerlag). Neuron 72, 330–343, October 20, 2011 ª2011 Elsevier Inc. 343 Neuron, Volume 71 Supplemental Information A Cortical Substrate for Memory-Guided Orienting in the Rat Jeffrey C. Erlich, Max Bialek, and Carlos D. Brody Table of Contents Page Number Figure S1 - Spectral analysis of whisking 2 Movie S1 - Exploratory Whisking 3 Movie S2 - Delay Period Whisking 3 Movie S3 - Out of task Whisking in the Center Port 3 Figure S2 - Muscimol Inactivation of M1 4 Figure S3 - ISI histograms, waveforms, and electrode placement 7 Figure S4 - Population PETHs for frequency and location discrimination 8 Figure S5 - Latency correlations between cells 9 Figure S6 - Head Direction Tuning 10 Figure S7 - Velocity and Acceleration Control 12 Supplemental Experimental Procedures 14 1 B Whisking data from Movie S2: on-task, delay period center poking. 3 2 Power (a. u.) Power (a. u.) A right trials left trials 1 0 4 Whisking data from Movie S3: off-task, exploratory center poking. 3 2 1 0 6 8 10 Frequency (Hz) 4 6 8 10 Frequency (Hz) Figure S1, No significant difference in whisking between right choice and left choice trials when the animals are performing the task. Related to Figure 1. Figure S1, No of significant difference in whisking between right choice and left choice Quantification the whisking data collected in Movie S2 and S3; see caption to those videostrials when the are performing theand task. Related to Figure 1. The initiation time of each trial in for animals an explanation of the source form of the whisking data. Quantification of the orienting whisking task dataiscollected in by Movie S2 andRats S3; see caption those our memory-guided controlled the subject. do not initiatetotrials at videos for an consistentlyofidentical intervals, but instead perform the task in bouts, typically explanation the source and form of the whisking data. The initiation time ofperforming each trial many in our memorytens oforienting trials in quick regular succession call these time by short guided task is controlled by the(we subject placing its periods nose in“on-task”) the centerfollowed port. Rats do not initiate breaks (typically, a few minutes) of grooming and sniffing/whisking asinthey move aroundperforming the trials at consistently identical intervals, but instead perform the task bouts, typically many behavior boxin(“off-task”). (A) succession Data from Movie S2:these minimal motion, and no significant tens of trials quick regular (we call timewhisker periods “on-task”) followed by short breaks whisking a difference between left trials and trials during on-task (typically, few minutes) of grooming and right sniffing/whisking as theycenter move pokes. aroundThe the panel behavior box (“offshows a spectral analysis of whisking during the “nose in center” delay period from randomly task”). (A) Data from Movie S2: minimal whisker motion, and no significant whisking difference between selected, correctly completed left and right memory trials. (B) As in panel A, but from center left trialsduring and right trials during(Movie on-taskS3). center Theat panel shows a spectral analysis of whisking pokes off-task periods The pokes. clear peak the theta frequency (~ 6 Hz) during during the “nose in center” delay period from randomly selected, correctly completed left and these off-task center pokes demonstrates that animals are capable of producing measurable right memory (B) while As in their panelsnouts A, butare from center pokes during periods (Movie S3). The clear whiskingtrials. motions in the center poke. Dataoff-task in panels (A) and (B) suggest peak at the theta frequency (~ 6 Hz) during these off-task center pokes demonstrates that animals are that during on-task center poking, rats choose to not whisk. capable of producing measurable whisking motions while their snouts are in the center poke. Data in panels (A) and (B) suggest that during on-task center poking, rats choose to not whisk. 2 2 Movie S1. Strong, clearly visible whisking during exploration of the behavior box at the beginning of a session. To facilitate visualizing the whiskers, we painted several dots of titanium dioxide onto them, as well as affixing a small piece of aluminum foil to one of the right whiskers. At the start of each session, before performing any trials, rats typically sniff and whisk briskly around the behavior box. The video shows a rat during this initial exploration period, before performing any trials in the session. The video demonstrates that the titanium dioxide and aluminum foil do not affect the rat’s ability to rapidly whisk as it performs its exploratory motions. Movie S2. Minimal whisking during on-task center poking The video shows that when rats are performing the task, whiskers are held very still during delay period center poking. Data from this video is quantified in Figure S1A. Correct left or right memory trials were randomly selected for whisker position scoring. Trials were included if the rat’s head remained in a relatively horizontal position for the duration of the trial, so that whiskers were visible throughout the trial and their position could be scored. Scoring was performed blind as to whether a trial was a rightinstructed or a left-instructed trial. The video is 5x slower than real time to facilitate observation of the whiskers during the delay period. The rat in this video was cannulated, and the cannula sites were used as fiducial markers on the head. Two points on a right whisker were marked in every frame as fiducial points on the whisker. The distance from the distal whisker point to the right head fiducial point is marked with a red line. This distance (l, length of the red line) was used as a measure of whisker position. A spectral analysis of l(t) is shown in Figure S1A. Movie S3. Visible whisking during off-task center poking The video shows that rats are capable of whisking even when their snouts are in the center port. Data from this video is quantified in Figure S1B. During off-task pauses between runs of completed trials, rats will sometimes sniff and whisk around the behavior box, including occasionally nose poking into the center port. The video shows selected bouts of center poking that were defined as “off-task” in that no completed trials occurred closer than 5 seconds, neither before nor after the time period shown in the video. For example, in one of these periods the rat center poked during the inter-trial interval. In another, the rat initiated a regular center poke during the “nose-in-center” period, but then did not complete it and did not return to complete a trial until 5 seconds later. Scoring methods and conventions as in Movie S2. 3 M1, 0.5 mg/ml muscimol = Equal to the dose in FOF Left M1 Inactivations, n=8 Right M1 Inactivations, n=9 p<0.01 Non−Memory Trials 95 50 5 % Went Right 90 Memory Trials 95 B % Correct A 50 70 Non-Memory Trials Memory Trials 50 No Drug Ipsi Contra Muscimol 5 20 50 Clicks/Sec 125 20 50 125 ⇤⇥ ⌅⇤⇥ ⌅ ⇤⇥ ⌅ ⇤⇥ ⌅ Ipsi trials Contra trials Contra trials Ipsi trials M1, 1 mg/ml muscimol = Double the dose in FOF Left M1 Inactivations, n=8 Right M1 Inactivations, n=8 90 Non−Memory Trials 95 50 5 70 50 Memory Trials 95 % Went Right p<0.01 D % Correct C 50 Non-Memory Trials Memory Trials No Drug Ipsi Contra Muscimol 5 20 50 Clicks/Sec 125 20 50 125 ⇤⇥ ⌅⇤⇥ ⌅ ⇤⇥ ⌅ ⇤⇥ ⌅ Ipsi trials Contra trials Contra trials Ipsi trials 4 (continued on next page) Example FOF histology E G 10 7 6 0 +5 -5 -10 5 4 3 2 -15 Bregma 0 1 0 1 FOF histology 2 3 4 5 6 H 7 Figure 13 5 5 10 +15 +10 +5 0 M1 histology 6 0 10 0 Interaural Example M1 histology F 10 5 0 +5 -5 -10 4 3 2 1 0 1 2 3 4 5 6 -15 Bregma 0 Figure 10 -5 5 5 1 91 9 10 0 Interaural +15 +10 +5 0 -5 2 82 8 7 7 10 703 +5 6 -5 5 -10 +15 8 7 3 64 +10 9 4 2 1 0 1 2 3 4 5 6 7 3 Figure 12 AP=3.24 from Bregma 5 0 Interaural 5 3 -15 Bregma 0 5 10 6 4 4 10 +5 0 -5 0 55 5 46 6 37 7 1 2 3 AP=2.76 from Bregma 2 1 AP=2.52 from Bregma Interaural 11.52 mm 7 6 5 4 3 2 1 0 1 2 3 6 28 5 19 Bregma 2.52 mm 4 5 6 4 Interaural 12.24 mm 7 6 5 4 8 4 9 5 Bregma 3.24 mm 3 2 1 0 1 2 3 4 5 6 6 3 7 2 8 9 1 Interaural 11.76 mm 7 5 6 5 Bregma 2.76 mm 4 3 2 1 0 1 2 3 4 5 6 7 Figure S2, Unilateral inactivation of neck motor cortex (M1) produces a different and much weaker pattern of impairment than inactivation of the FOF. Related to Figure 2. (A,C) Behavioral performance in control and muscimol-infusion days for 0.5 mg/ml (A) and 1 mg/ml (C) muscimol. Top row: non-memory trials. Bottom row: memory trials. Left column: muscimol infusions into left M1. Right column: muscimol infusions into right M1. Open circles, data from muscimol infusions. Closed circles: control data from days immediately preceding infusion days. Dashed lines: sigmoidal fits to muscimol data. Solid Lines: sigmoidal fits to control data. Error bars are standard error of the mean across sessions (The number of sessions is indicated at the top of each plot). Error bars for control data were smaller than the marker in most cases. Underbraces at bottom indicate the sets of trials in which animals were instructed to orient ipsilaterally or contralaterally to the site of infusion. (B,D) Combined data from left and right infusion sessions at 0.5 mg/ml (B) and 1mg/ml (D) of muscimol and collapsed across all stimulus difficulty levels. The “No Drug” data come from the 20 sessions one day before infusion sessions. The Ipsi and Contra Muscimol data are the performance on ipsilateral trials and contralateral trials on infusion sessions (n=20). (E) Example of a nissl stained coronal section from a rat with cannula implanted in the FOF. The section is ~2.6 mm anterior to Bregma. (F) Example of an unstained bright field coronal section from a rat with cannula implanted in M1. The section is ~3.3 mm anterior to Bregma. (G) FOF cannula placements. Diamonds indicate the location of the scar left from the injection cannula for each rat. (H) M1 cannula placements. Diamonds indicate the location of the scar left from the injection cannula for each rat. 6 4000 2000 0 0 −3 −2 −1 0 ISI (log10 sec) 200 50 100 0 0 −50 −100 10000 1500 −100 Frequency 10000 5000 Frequency 0 µV 100 µV µV 2000 −3 −2 −1 0 ISI (log10 sec) 100 Frequency 4000 0 −3 −2 −1 0 ISI (log10 sec) B 0 6000 Frequency 5000 100 −100 6000 Frequency 10000 Frequency 200 200 100 0 −100 µV 100 50 0 −50 µV µV A 5000 1000 500 0 0 0 −3 −2 7 −1 06 ISI (log sec) 10 0 +5 -5 10 -10 5 4 −32 −2 −1 00 1 ISI (log10 sec) 3 -15 Bregma 0 1 2 3 −3 −2 −15 0 6 4 ISI (log10 sec) 5 5 7 Figure 13 0 10 C +15 D 10 0 Interaural +10 +5 0 7 -5 10 9 M2 6 0 +5 -5 -10 5 4 -15 Bregma 0 3 2 1 0 1 2 3 4 5 6 1 Figure 14 5 5 7 0 10 10 0 Interaural 8 +15 +10 +5 0 2 -5 1 9 3 7 2 8 4 6 3 7 5 5 4 6 6 4 5 5 7 3 6 4 2.52 mm 2 AP = 2.04from from Bregma Bregma AP=2.28 3 1 mm 8 7 9 1 Interaural 11.522 mm 7 6 Bregma 2.52 mm 5 4 3 2 1 0 1 2 2.28 mm4 3 5 6 8 7 9 1 Figure S3, Waveforms and Inter-Spike Interval histograms for example cells figure 3, and Interaural 11.28 mm Bregma in 2.28 mm histological placement of electrodes. Figure 3. for example cells in figure 3, Figure S3, Waveforms and Inter-SpikeRelated Intervaltohistograms (A) Waveforms (mean ± standard deviation) Related and inter-spike interval and histological placement of electrodes. to Figure 3. histograms for the three (A) Waveforms (mean ±neurons standard deviation) and inter-spike interval histograms the three contralateral preferring shown in Figure 3A. (B) Waveforms (mean ± for standard deviation) and contralateral preferring neuronsofshown in Figure 3A. (B) Waveforms (mean ± standard inter-spike interval histograms the three ipsilateral preferring neurons shown in Figure 3B. (C) deviation) inter-spiketrack interval histograms of the three ipsilateral neurons in and Example ofand an electrode in FOF. The thin black lines indicatepreferring the borders of M2 shown (Paxinos Figure 3B. (C) Example of an tracks electrode FOF. Therats. thin Green black lines indicateindicate the borders Watson, 2004). (D) Electrode for track the 5 in implanted diamonds the final location M2tips (Paxinos and Watson, 2004). Electrode for the 5 implanted rats. Green of the tips from ofofthe from electrodes in the left (D) FOF and red tracks diamonds indicate the final location diamonds indicate the final location of the tips from electrodes in the left FOF and red diamonds electrodes in the right FOF. indicate the final location of the tips from electrodes in the right FOF. 7 6 5 4 3 2 6 7 1 0 1 2 3 4 5 6 7 Population PETHs from rats performing frequency discrimination Population PSTH of Ipsi Preferring Cells (n=21) Population PSTH of Contra Preferring Cells (n=23) 8.2 Hz 0 4.6 Hz −1 1 Stimulus Ipsilateral Correct −1.5 −1 −0.5 0 0.5 E 0 0 0 −1 −1 −1.5 −1 −0.5 0 0.5 −1.5 D 1 1 0 0 −1 −1 −1 −0.5 0 0.5 F 1 1 −1 C −1.5 B −0.5 0 0.5 −1.5 Went-Contra Error −1 −0.5 0 0.5 −1 −0.5 0 0.5 −1 −0.5 0 0.5 H 1 1 Went-Ipsi Error −1.5 −1 G 0 0 −1 −1 −1.5 −1 −0.5 0 0.5 −1.5 Non-Memory Trials Firing Rate (Z-Score) Contralateral Correct 1 Population PSTH of Ipsi Preferring Cells (n=18) Population PSTH of Contra Preferring Cells (n=27) Memory Trials A 11.8 Hz Population PETHs from one rat performing spatial location discrimination Time from Go Cue (s) Figure S4, Comparison of electrophysiological results for frequency discrimination vs spatial location discrimination. Related to Figure 4. (A-D) Population PETHs for rats performing frequency discrimination (n=44). Each panel is a population perievent time histogram (PETH) showing the average response (across neurons) on correct (thick lines, mean±std.err. across neurons) and error trials (shaded, mean±std.err. across neurons) where the stimulus indicated that the correct response was contralateral (blue) or ipsilateral (red) to the Figure S4, Comparison electrophysiological recorded neuron. PETHsofare aligned to the timeresults of the for Go frequency cue (centerdiscrimination LED offset). vs (A) The average spatial location discrimination. Related to Figure 4. responses of memory trials for neurons that fired more on contralateral trials. (B) The average responses Population PETHs for performing frequency discrimination (n=44). panel of(A-D) memory trials of neurons thatrats fired more on ipsilateral trials. (C) Same as A but Each for non-memory trials. is aSame population perievent time histogram the PETHs averagefor response (across (D) as B but for non-memory trials.(PETH) (E-H) showing Population rats performing spatial neurons) discrimination on correct (thick (n=45) lines, mean±std.err. neurons) and errorshowing trials (shaded, mean response location Each panel across is a population PETH the average ±std.err. across neurons) where the stimulus indicated that the correct response was (across neurons) correct (thick lines, mean±std.err. and error trials contralateral (blue)on or ipsilateral (red) to the recorded neuron.across PETHsneurons) are aligned to the time of (shaded, mean±std.err. across neurons) where the stimulus indicated that the correct response was contralateral the Go cue (center LED offset). (A) The average responses of memory trials for neurons that (blue) or ipsilateral (red) to the recorded areofaligned to trials the time of the Go fired more on contralateral trials. (B) Theneuron. average PETHs responses memory of neurons thatcue (center LED The average responses trials for neurons thatSame fired as more onfor contralateral firedoffset). more on(E) ipsilateral trials. (C) Same asofAmemory but for non-memory trials. (D) B but trials. (F) The trials. average(E-H) responses of memory trialsfor of neurons that fired more on ipsilateral non-memory Population PETHs rats performing spatial location trials. (G) Same as E but for non-memory trials. Same asPETH F butshowing for non-memory trials. discrimination (n=45) Each panel is a(H) population the average response (across neurons) on correct (thick lines, mean±std.err. across neurons) and error trials (shaded, mean ±std.err. across neurons) where the stimulus indicated that the correct response was contralateral (blue) or ipsilateral (red) to the recorded neuron. PETHs are aligned to the time of the Go cue (center LED offset). (E) The average responses of memory trials for neurons that fired more on contralateral trials. (F) The average responses of memory trials of neurons that fired more on ipsilateral trials. (G) Same as E but for non-memory trials. (H) Same as F but for non-memory trials. 8 A B 15 40 7% p<0.05 % of Cells # of cell pairs All Cells, Contra Mem Trials 20 0 −1 1 C 30 10 20 5 10 0 0 r Onset of Selectivity Peak of Selectivity 0 −1 0 1 −1 0 Onset of Selectivity Time from Peak Head Velocity (s) 1 Figure S5. Neural latency is not correlated across pairs of simultaneously recorded neurons. Related to Figure 5. (A) Histogram of the correlations in neural latency (as computed in Figure 5) for 238 pairs of neurons. The population is not significantly different than zero, nor is the number of individually significant neurons (7%) significantly more than expected at p<0.05. (B) Histogram of the time difference between the onset of differential ipsi vs. contra firing rate (as measured by a sliding ROC, as in Figure 3A,B) and the time of peak head velocity. Each entry in the histogram is a neuron; the histogram shows all 166 neurons recorded during sessions withis head-tracking data.across One hundred forty-six (88%)recorded of the neurons had Figure S5. Neural latency not correlated pairs ofand simultaneously Related to Figure aneurons. differential firing rate onset5.that preceded the time of peak head velocity. (C) As panel (B), but now (A) histogram of the correlations in time neural latency (as differential computed ipsi in Figure 5) for 238rate pairs of the time of showing the difference between the of the largest vs contra firing and neurons. population is not significantly different than zero, is the number individually peak head The velocity. One hundred and twenty-four (75%) of thenor neurons had the of peak of their differential significant neurons (7%) more than expected at p<0.05. (B) Histogram of the time firing rate occur before thesignificantly time of peak head velocity. difference between the onset of differential ipsi vs. contra firing rate (as measured by a sliding ROC, as in Figure 3A,B) and the time of peak head velocity. Each entry in the histogram is a neuron; the histogram shows all 166 neurons recorded during sessions with head-tracking data. One hundred and forty-six (88%) of the neurons had a differential firing rate onset that preceded the time of peak head velocity. (C) As panel (B), but now showing the difference between the time of the largest differential ipsi vs contra firing rate and the time of peak head velocity. One hundred and twenty-four (75%) of the neurons had the peak of their differential firing rate occur before the time of peak head velocity. 9 Session 52767, Cell 1919 Session 52767, Cell 1922 0.5 10 −90 90 0 0.4 0.2 0 −200 10 −90 0.2 0.5 5 −90 0.2 0 −200 0 200 Head Angle (degrees) Session 51252, Cell 2920 0 −200 10 0 1 0.5 5 −90 Occupancy 0.1 0 200 Head Angle (degrees) 90 0 0.2 0.1 0 −200 0 200 Head Angle (degrees) 4 90 0 0.2 3 0 1 2 0.5 1 −90 0 0.2 0.1 0 −200 0 200 Head Angle (degrees) 0 200 Head Angle (degrees) C Out of Task 0 1 In Task 0 1 0.5 −90 In Task − Out of Task 0 1 0.5 90 0.5 −90 90 −90 90 D −90 0 10 0 10 5 5 90 90 0 0.4 Spikes/sec Spikes/sec Spikes/sec 0 1 0.5 10 15 10 0 1 90 0 0.4 0 −200 20 Session 51252, Cell 2922 15 Occupancy 0.5 5 Session 51252, Cell 2921 B 0 1 −90 0 200 Head Angle (degrees) Spikes/sec 0 1 30 Occupancy 20 Session 52767, Cell 1925 15 Spikes/sec In Task Out of Task Occupancy Occupancy Spikes/sec 30 Occupancy A 0 15 10 5 −90 90 10 −90 90 90 Figure S6. Head direction tuning in FOF. Related to Figure 6 (A) Head direction tuning and occupancy histograms for three cells recorded simultaneously in a single session. Tuning and occupancy were computed separately for in-task epochs (green; any timepoint within 10 seconds of a poke into any of the nose ports) and for out-of-task epochs (black; all other timepoints). Occupancy histograms indicate fraction of time spent with the head pointing in each direction. Error bars for firing rates are bootstrapped 99% confidence intervals of the firing rate at each of 18 bins of head angle. The center port is at 0 degrees. The compass plots to the right of each tuning histogram summarize the in-task and out-of-task tuning. Each cell’s summary vector is the weighted average of a set of unit length vectors pointing in each of the 18 head direction bins; the weighting was proportional to the cell’s firing rate in each direction bin. The summary vectors show how each cell contributes to the plots in panel C. (B) Same as A for three neurons recorded during a different session. (C) Summary of head-direction tuning for 166 cells recorded during sessions with head-tracking data. Each cell is represented by a summary vector, shown as a single arrow. The rightmost panel is the vector difference, for each cell, between its in-task summary vector minus its out-of-task summary vector. (D) Polar histogram of the angles of the arrows in C. The distribution of angles for In Task - Out of Task is significantly non-uniform (Omnibus Test, p<0.03) 11 !"’ ~ 10˚/s 0 50 100 C % Cells Selective 40 −1 −0.5 0 0.5 Time from Go Cue (s) 30 20 10 −1.5 40 −0.5 0 0.5 D 30 20 10 I 0 −1.5 −1 −0.5 0 0.5 Time from Go Cue (s) 2 −1 −0.5 0 0.5 Time from Go Cue (s) −1 E −1000 F !"’’ ~ 160˚/s2 !"’’ ~ 160˚/s2 −500 0 % Cells Selective −1.5 0 −1.5 Angular Head Acceleration s 2) Angular Head Acceleration (!ʼʼ(°, /°/s B !"’ ~ 10˚/s −50 % Cells Selective Angular Head Velocity (!ʼ ,(°°/s) Angular Head Velocity / s) A −100 40 20 Copy of data from Fig 3C After regressing out "(t), "’(t), "ʼʼ(t) 0 −1.5 −1 −0.5 0 0.5 Time from Go Signal (s) 500 1000 40 −1 −0.5 0 0.5 Time from Go Cue (s) G 30 20 10 0 −1.5 % Cells Selective % Cells Selective −1.5 −1 −0.5 0 0.5 Time from Go Cue (s) −1.5 40 −1 −0.5 0 0.5 H 30 20 10 0 −1.5 −1 −0.5 0 0.5 Time from Go Cue (s) Figure S7. Predictive coding of response is not a simple function of current angular head velocity or acceleration. Related to Figure 7 (A) Angular head velocity (φ’, deg/s) as a function of time relative to the Go cue for correct memory trials. Thin blue lines are from a random subsample of trials where the rat’s final choice was to move left; thin red lines are a random subsample of trials where the final choice was to move right. Thick lines are the mean φ’(t) for each group, averaged over all correct memory trials. At the time of the Go cue (t=0), the difference in head velocity between the two groups is approximately 10 deg/s. (B) As in panel A, but with the grouping defined by the sign of the angular head velocity at time t=−0.9 sec (yellow arrow; φ’(t=−0.9)>0 versus φ’(t=−0.9)<0). At t=−0.78 sec, the difference in head velocity between the two groups is approximately 10 deg/s. (C) Percentage of cells (out of 166 neurons) with significantly different firing rates for the red vs blue groups of panel A. At each timepoint, threshold for each cell being considered significant was p<0.01. At t=0, 23% of cells have firing rates that discriminate between the two groups. (Continued on next page) 12 (D) As in panel C, but for the φ’(−0.9)>0 versus the φ’(−0.9)<0 groups. At t=−0.78 sec, the difference in head velocity between the two groups is approximately 10 deg/s, but only 9% of cells discriminate between the two groups.(E, F, G, H) As in panels A,B,C,D, but for angular head acceleration (φ’’, deg/s2). (E) 200 ms after the time of the Go cue (t=0.2), the difference in φ’’ between the two groups is approximately 160 deg/s2. (F) As in panel E, but with the grouping defined by the sign of φ’’ at time t=−0.9 sec (φ'’(t=−0.9)>0 versus φ'’(t=−0.9)<0). At t=−0.9 sec, the difference in head velocity between the two groups is approximately 160 deg/s2. (G) Percentage of cells (out of 166 neurons) with significantly different firing rates for the red vs blue groups of panel E. At each timepoint, threshold for each cell being considered significant was p<0.01. At t=0.2, 23% of cells have firing rates that discriminate between the two groups. (H) As in panel F, but for the φ'’(−0.9)>0 versus the φ’'(−0.9)<0 groups. At t=−0.9 sec, the difference in head acceleration between the two groups is approximately 160 deg/s2, but only 9% of cells discriminate between the two groups. (I) Development of choice-dependent activity over the course of the trial after accounting for the effects of head angle variables. The orange line is copied from Figure 3C. The black line represents, at each timepoint, the percentage of cells (out of n=166) with significantly different ipsi vs contra residual firing rates. Residual firing rates are obtained after performing a linear regression to eliminate each cell’s firing rate dependencies on φ(t), φ’(t), and φ’’(t) (see Supplementary Experimental Procedures). The black and orange horizontal lines at the top of the panel indicate timepoints for which the percentage of significant cells is more than expected by chance. 13 SUPPLEMENTAL EXPERIMENTAL PROCEDURES Subjects Animal use procedures were approved by the Princeton University Institutional Animal Care and Use Committee and carried out in accordance with National Institutes of Health standards. All subjects were male Long-Evans rats (Taconic, NY). Rats were pair-housed during initial behavioral training and then single housed after being implanted with electrodes or cannula. Rats were kept on a reverse 12-hour light dark cycle and trained in their dark cycle, when they were more active. Rats were placed on a restricted water schedule to motivate them to work for water reward. Behavior Behavior took place in a custom training box (Island Motion, NY) inside a sound and light attenuating chamber (H10-24A, Coulbourn Instruments, PA). In each box there were three “nose ports”, conical openings into which the rats could poke their snouts. The three nose ports were arranged side-by-side along a curved wall. Each nose port had an infra-red (IR) beam across the front (used to detect when the rat’s nose was in the port), a visible white light emitting diode (LED), and a sipper tube connected to a water supply that was controlled by a computer controlled solenoid. In addition, there were two speakers mounted above each of the left and right nose pokes. All aspects of the task were computer controlled. Behavioral events were timestamped with greater than 1ms accuracy using a custom open-source software (Island Motion, NY; open-source code at http://code.google.com/p/rt-fsm/) on a computer running a realtime linux operating system. Rats were placed in and removed from the behavior box by technicians that were blind to the task. Rats were trained using an automated training protocol. First, rats were trained using a classical conditioning paradigm. On each trial a sound played out of one of the speakers and then water was delivered from the corresponding port. The goal of this stage was simply to teach the rat that the ports delivered water that was contingent on sounds. The trials during this training stage had an inter-trial interval of about 2 minutes. The side of the sound and water delivery was alternated each day. This stage lasted 8 days. The next stage involved the rats learning to poke in the center port in order to initiate a trial. A light would come on in the center port and when the rat poked that would trigger the presentation of sound. The sound was a click train, and the rats had to learn an association between the rate of the clicks and the associated side for reward. For example, 100 clicks/sec meant reward was on the left and 25 clicks/sec meant the reward was on the right. At first, trials were presented in blocks of right and left trials. As the rat showed that he had learned the basic task structure we moved to randomly interleaved right and left trials (about 1 week). The next stage of training was growing the 'fixation period'. At first, the rat could essentially just poke in the center and then respond in a side port. We incrementally grew the time that the rat had to stay in the center port before making his response. The end of the fixation period was indicated by extinguishing the center light. Once rats could maintain fixation for one second we moved to the next training stage. Rats were highly variable as to how long it took them to pass this stage (1-4 weeks). Until now the sound played for the entire duration of the trial. That is, there was no memory requirement. The next stage slowly shortened the duration of the stimulus to 300 ms. This stage took 2-5 weeks depending on the performance of the rat. At the end of this stage the rat was doing randomly interleaved, left and right memory trials. The next stage introduced no-memory trials where the stimulus was played at the end of the fixation period (Figure 1A). The final stage introduced intermediate stimuli in order to produce psychometric discrimination data. Rats learned that click trains < 50 clicks/sec meant reward on the right and > 50 clicks/sec meant reward on the left. In each session 6 discrete sounds were played (E.g. 100, 80, 60, 40, 30, 20 clicks/sec) and the values of the sounds were adjusted at the end of each session to better sample the psychometric curve. So, a rat with very good discrimination might end up with 100, 60, 52, 48, 40, 20 clicks/sec as the set of stimuli. The rats in the M1 muscimol experiment learned the opposite rule: click trains > 50 clicks/sec meant reward on the right and < 50 clicks/sec meant reward on the left. All data described in this paper was collected from rats in the final training stage. Sessions with poor performance (<70% overall or fewer than 8 correct memory trials without fixation violations per side) 14 were excluded from analyses. These sessions were rare (2.4% of all “final stage” sessions) and were likely caused by problems with the hardware (e.g. a clogged water-valve or a dirty IR-photodetector). Surgery All surgeries were done under isoflurane anesthesia (1.5-2%) using standard stereotaxic technique. Rats were given an injection of ketamine (10 mg) and buprenorphine (0.006 mg) to assist induction and provide analgesia. Five minutes later they were placed in an isoflurane induction chamber. We slowly increased the concentration of isolflurane from 0.5% to 4% over the course of 4 minutes. After induction, rats were moved to a stereotax (Kopf Instruments; CA) and their noses placed in a cone which provided 1.5-2% continuous isoflurane flow. After verifying surgical levels of anesthesia with pinch tests and eye blink tests, rats were secured in non-rupture ear bars (Kopf Instruments; CA). The scalp was shaved and cleaned with ethanol, betadine and ethanol and then a midline incision was made with a scalpel. A spatula was used to clean the skull of all overlying tissue before craniotomies were made with a dental drill over the FOF (AP +2, ML ±1.3 mm from Bregma). This location was chosen because it was the center of the distribution of stimulation cites that resulted in contralateral orienting orienting movement in Sinnamon & Galer (1984). Durotomies were then performed. After the durotomies were finished, saline soaked Gelfoam (Pfizer Injectables; NY) was placed in the craniotomies to protect the brain for the next step. Then a thin coat of C&B Metabond (Parkell, Inc; NY) was painted all over the skull. The Gelfoam was then removed from the craniotomies and the implant (either electrodes or cannula) was lowered onto the brain surface. For electrode implants only the tetrodes entered the cortex. For cannula implants (Plastics One, VA) the outer cannula was placed at brain surface and the injector (inserted only during infusions) extended 1.5 mm past the end of the guide. DuraLay (Reliance Dental, IL) cement was used to secure the implant to the Metabond coated skull. Rats were given buprenorphine and ketofen 24 and 48 hours post-operative and were allowed to recover on ad lib water for 5 days before returning to water restriction and behavioral training. Subsequent histological verification and comparison to Paxinos and Watson’s atlas, which is based on Wistar strain rats (Paxinos and Watson, 2004), indicated that our implants were at a position matching +2.4 AP, ±1.3 ML in the atlas. The target of the neck M1 surgeries was +3.5 AP, +3.5 ML (Gage et al., 2010). Infusions Five rats were used for the infusion experiments in FOF and 4 rats were used for the infusion experiments in M1. Rats were placed under a light non-surgical 1-1.5 % isoflurane anesthesia for infusions. The cap and dummy cannula were removed and replaced with an injector that extended 1.5 mm beyond the end of the guide cannula. For the FOF dose and volume of muscimol was 0.5 mg/mL and 0.3 uL respectively. For M1 we first used the same dose and volume as in the FOF (0.3 uL of 0.5 mg/mL muscimol) and we followed up that experiment with double the dose (0.3 uL of 1 mg/mL muscimol). The rate of infusion was 0.2 uL/min and we waited 4 minutes after infusions before removing injectors to allow the drug to spread. Rats were allowed to recover for 30 minutes from the anesthesia before beginning their training sessions for that day. Rats were placed in and removed from the behavior box by technicians that were blind to the task and the side of the infusion. Each infusion day was followed by a minimum of 3 days of running with no infusions. Left and right infusions were alternated. Each rat received at least two right infusions and two left infusions. For analysis of the effects of unilateral muscimol infusions we compared infusion days with the days immediately preceding the infusion. For example, if we did infusion on the 10th and 20th of the month, then we would use data from the 9th and the 19th as control days. We did not use saline infusions as a control because we are not specifically interested in the mechanism of the effect of muscimol, but rather whether perturbing unilateral function of the FOF has a lateralized effect on performance. In addition, we argue that the right and left infusions act as a control for each other, since in both cases the animals experience the same procedure, the only difference being the side of the infusion. Thus, any behavioral difference between left and right infusions would be due to the drug infusion and not some other aspect of the infusion procedure (a change in handling or the isoflurane). 15 Whisker Experiments In order to determine the contribution of whisking to memory-guided orienting we conducted three experiments: Unilateral lidocaine injections into the whisker pad; video analysis of the whiskers during the task; and removal of the whiskers. For the unilateral lidocaine experiments rats were anesthetized with isoflurane and then 0.2 cc of 1% lidocaine was injected subcutaneously either at the right or left whisker pad. The front and rear claws were blunted with a nail file to prevent self-injury from scratching at the anesthetized site. The effect of lidocaine was visually confirmed by unilateral absence of whisking on the injected side before training and every hour thereafter. In every case the effect of lidocaine lasted for at least 140 minutes. In some cases the whiskers were marked with titanium dioxide paint (SigmaAldrich, Product # 224227 mixed with clear nail polish) to facilitate visualization of the whiskers on video. The injections were performed on alternate sides with no-injection days in between each experiment: for example, a right injection on the 10th, a left injection on the 12th and a right injection on the 14th. For video analysis of whisking during memory guided orienting we used small pieces of aluminum foil glued to the whiskers as well as beads of titanium dioxide paint to make the whiskers more visible on video. Video was collected at 29.97 frames / sec using a Hamamatsu CCD (XC-77). Illumination was provided by an infra-red led 30Hz strobe light (custom made) with a 30% duty cycle to reduce motion blur. Then the position of the head and whiskers were marked by hand on each frame (assisted by a custom algorithm based on the local cross-correlation between subsequent frames). We used the distance of the whisker marker to the head marker as a measure of whisker position. We then used a multi-taper method (pmtm, MATLAB) with a WM (time-bandwidth product) of 2 to determine the spectral content of the whisker movements separately for exploratory whisking and delay period whisking on left versus right correct memory trials (Figure S1). Recordings Five rats were used to collect single-unit electrophysiology data. Recordings were made with platinum iridium wire (16.66 µm, California Fine Wire, CA) twisted into tetrodes. Each tetrode was threaded into an polyimide tube (34 AWG triple wall) which was part of a movable bundle of eight tubes. Two rats were implanted with a movable bundle of 8 tetrodes on each of the right and left FOF (16 tetrodes per rat). Three rats were implanted with an 8-tetrode bundle unilaterally. Implants were targeted to +2 AP ±1.3 ML (mm relative to Bregma). Within each bundle, tetrodes were spaced ~250 µm from each other; we estimate that neurons were sampled from within a radius of ~0.5 mm. Wires were goldplated to 0.5-1.2 MOhm. The tetrodes could be advanced by turning a nut against a spring on a 0-80 threaded rod so that a 1/8 turn drove the tetrodes down about 40 µm. The tetrodes were advanced at the end of sessions so that the brain tissue had time to stabilize before recording the next day. We used two electrophysiology systems to collect the data presented here. In both cases the reference selection, analog to digital conversion, time-stamping, filtering (600-9000Hz, FIR filter) and video tracking was done using a Digital Cheetah System (Neuralynx; MT). For some recording we used Neuralynx unity gain headstages (HS-36, Neuralynx). For other recordings we used 4 x 31 channel (124 channel total) time-division multiplexing headstages (Triangle BioSystems Inc, NC) with a 10 channel commutator (Dragonfly, WV), and then the signals were demultiplexed and then used as input to the Digital Cheetah system. Spike sorting was done by hand using SpikeSort3D (Neuralynx). Cells had to satisfy several criteria to be included in the presented analyses: 1) No inter-spike intervals < 1 ms; 2) Signal to noise ratio >4; 3) At least one bin of the perivent time histogram (PETH, aligned to the Response Onset time), had to have a firing rate of at least 3 spikes/sec. The PETH was a 2 second window (-1.5s before Response Onset to 0.5s after) with 10 ms bins and a causal half-gaussian smoothing window with a s.d. of 200 ms (effective smoothing was 100 ms since only half the gaussian was used). 378 cells recorded over 100 sessions satisfied the first two criteria (i.e. Well-isolated single units). 242 cells satisfied all three criteria. Median number of cells per session was three. The maximum number of cells recorded in a session was eleven. 16 Behavior Analysis Behavior Analysis To generate the psychometric curves curves (Figure(Figure 1b,c) we analyzed all sessions from the finalthe training To generate the psychometric 1b,c) we analyzed all sessions from final stage for each rat. Each session generates 12 behavioral data points: the % “Went Right” for each training stage for each rat. Each session generates 12 behavioral data points: the % “Went frequency (n=6) both memory (n=6) and non-memory trials.and Wenon-memory then combined theWe data across sessions Right” for for each frequency for both memory trials. then combined the (separately memory and non-memory trials) and fit that data with a trials) 4-parameter sigmoid data for across sessions (separately for memory and non-memory and fit that datawhere with ay, 4-“% parameter sigmoid y, “% went-right” is a function of x, the log(click frequency): went-right” is a function of where x, the log(click frequency): y = y0 + a 1+e (x x0) b The four parameters lower asymptotic performance; y0+a , theasymptotic upper asymptotic 0 , theasymptotic The four parameters are: y0 ,are: the ylower performance; y0+a , the upper performance; in log(frequency) of the inflection of the curve; b , theofslope 0 , the location performance; x0 , thexlocation in log(frequency) of the inflection point ofpoint the curve; b , the slope the If a rat ran 100 sessions would fit 4toparameters to 600fordata pointstrials for memory curve. of If athe ratcurve. ran 100 sessions we would fit 4we parameters 600 data points memory and and another 4 parameters 600 data points trials. for non-memory was done with anothertrials 4 parameters to 600 data pointstofor non-memory Fitting was trials. done Fitting with Matlab’s nlinfit. ForMatlab’s analysis nlinfit. of the effects of whisker trimming we included data from the day of trimming and the two For analysis the effects offor whisker trimming weback included data from the day ofWe trimming following days. (It takes of several weeks whiskers to grow to their normal length.) generated and the two following days. (It takes several weeks for whiskers to grow back to their normal psychometric fits to the whisker trimming data in the same way as we did for the behavioral data in length.) We generated psychometric fits to the whisker trimming data in the same way as we Figure 1b,c. Since the whisker trimming experiment immediately followed the lidocaine experiments we did for the behavioral data in Figure 1b,c. Since the whisker trimming experiment immediately used the same sessions as controls for the lidocaine and the whisker trimming sessions. followed the lidocaine experiments we used the same sessions as controls for the lidocaine and Wethe generated psychometric fits to the muscimol and control data in the same way as for Figure whiskerthe trimming sessions. 1. To generate statistics for the effect of muscimol wemuscimol combinedand left control and right infusion We generated the psychometric fits to the data in the days sameacross way asrats for and labeled trials asgenerate contra- orstatistics ipsi-muscimol. For example, if wewe infused into the trials Figure 1. To for the effect of muscimol combined leftleft andFOF rightthen infusion where the correct toward the would be labeled as contra-muscimol for infused that session. days acrossresponse rats andwas labeled trials asright contraor ipsi-muscimol. For example, if we into To determine statistical significance we bootstrapped the distribution for the mean of the differences the left FOF then trials where the correct response was toward the right would be labeled as between conditions. Wefor didthat 8 tests: memory vs. non-memory forsignificance ipsi-muscimol memory vs. contra-muscimol session. To determine statistical wetrials; bootstrapped the nonmemorydistribution for contra-muscimol trials; ipsi-muscimol vs. contra-musicmol for non-memory trials and ipsifor the mean of the differences between conditions. We did 8 tests: memory vs. muscimol vs. contra-muscimol for memory trial; ipsi-muscimol vs. control for non-memory trials and non-memory for ipsi-muscimol trials; memory vs. non-memory for contra-muscimol trials; ipsi-ipsimuscimol vs. contra-musicmol non-memory trials and ipsi-muscimol vs. contra-muscimol for muscimol vs. control for memory trials;for contra-muscimol vs. control for non-memory trials and contramemory trial; ipsi-muscimol vs. control for non-memory trialsthe and ipsi-muscimol vs.mean control for muscimol vs. control for memory trials. The p-values reported are probability that the memory trials; contra-muscimol control for non-memory andM1 contra-muscimol vs. difference is zero for each comparison. vs. The analyses and statisticstrials for the muscimol experiments control to forthe memory trials. The p-values reported are the probability that the mean difference is were identical FOF experiments. zero for each comparison. The analyses and statistics for the M1 muscimol experiments were to the FOF experiments. Neural identical Data Analysis To determine whether cells had upcoming-choice-dependent firing rates during the delay period on Analysis the firing rate of the cells during the delay by counting the number of memoryNeural trials, Data we determined To determine whether hadstimulus upcoming choice-dependent firing spikes fired after the offset of the cells auditory and before the Go cue; werates then during dividedthe by delay the period on memory trials, we determined the firing rate of the cells during the delay by counting duration of the delay period on that trial to obtain the firing rate. We sorted the trials into correct left and the number of spikes fired after the offset of the auditory stimulus and before the Go cue; we correct right trials. We then computed the area under the receiver operator characteristic curve (AUC) then divided by the duration of the delay period on that trial to obtain the firing rate. We sorted for the distribution ofcorrect firing rates on left vs. the distribution firingcomputed rates on right trials. To determine the trials into left and correct right trials. Weofthen the area under the receiver statistical confidence on the AUC value we for randomly relabeledofthe trials as left or right (keeping the #of of operator characteristic curve (AUC) the distribution firing rates on left vs. the distribution left andfiring right rates trials on theright same) and computed the AUC of the shuffled trials. We did this 2000 times. A trials. To determine statistical confidence on the AUC value we randomly cell was considered significantly selective if the AUC thesame) 95% confidence relabeled the trials asside left or right (keeping the #ofofthe leftdata andwas rightoutside trials the and computed intervals of the shuffled data. the AUC of the shuffled trials. We did this 2000 times. A cell was considered significantly side To determine which upcoming-choice-dependent firingofrates, we first data. selective ifat the AUCtimepoints of the datacells washad outside the 95% confidence intervals the shuffled generated To single trial firing rate traces by convolving the spike trains on each trial with a halfdetermine at which timepoints cells had upcoming choice-dependent firing causal rates, we first Gaussian (s.d. of 200 ms) smoothing At each timepoint, running in 10 bins from seconds generated single trial firing ratekernel. traces by convolving the spike trains onms each trial with1.5 a causal half-Gaussian (s.d.seconds of 200 ms) kernel. At each timepoint, running in distribution 10 ms bins of from before the Go cue to 0.5 aftersmoothing the Go cue, we then computed the AUC of the left seconds cue tothe 0.5significance seconds after Gocell cue, then computed the AUC of vs. right1.5 correct firingbefore rates.the To Go compute forthe each at we each time bin we randomly thethe distribution of left vs.and right correct firing rates.forTothe compute significance for each relabeled trials as left/right computed the AUC shuffledthe data 1000 times. If the cell AUCatof binneuron we randomly relabeled thewas trials as left/right andconfidence computedinterval the AUC the dataeach for atime given at a given time bin outside the 99% offor thethe shuffled shuffled databin, 1000 If the of the for a given at a given time bin was data then that time fortimes. that cell, wasAUC labelled asdata significant. Weneuron then determined, for each cell, the largest number of time bins labeled as significant in the shuffled data (n) that would result in 5% or more of the shuffled trials having n or more significant time bins. (The larger that n is, the fewer the shuffled 16 17 outside the 99% confidence interval of the shuffled data then that time bin, for that cell, was labelled as significant. We then determined, for each cell, the largest number of time bins labeled as significant in the shuffled data (n) that would result in 5% or more of the shuffled trials trials that are labeled as significant.) If the original data for the cell had more than n time bins labeled as having n or more significant time bins. (The larger that n is, the fewer the shuffled trials that are significant, then cell as a whole was labeled as the significant p<0.05. labeled asthe significant.) If the original data for cell hadatmore than n time bins labeled as Insignificant, order to quantify the timing of the recruitment of the FOF population over theused course of the trial we then the cell as a whole was labeled as significant at p<0.05 and in further used the same sliding ROC analysis of left vs. right correct memory trials with activity aligned tofor the time time-dependent analysis (e.g., Figure 3C). Cells labeled as not significant were not used of the time-dependent Go cue as above. According to binomial statistics, using a threshold of p<0.01, we would expect analysis. less than 8/242 cells to be significant byof chance 99.9% of the time. In order to quantify the timing the recruitment of the FOF population over the course of Tothe perform population of firing rates, we first normalized the perievent time histograms trial we used theanalyses same sliding ROC analysis of left vs. right correct memory trials with (PETHs) of each cell by computing mean deviation to (over time and over trial classes) of activity aligned to the time of the Go cueand as standard above. According binomial statistics, using a the cell’s PETHs,ofand then we subtracted that mean and divided by that standard deviation. The 99.9% resulting threshold p<0.01, would expect less than 8/242 cells to be significant by chance of ztime. were then averaged across cells to obtain z-scored population PETHs. scoredthe PETHs In order to quantify whether neurons FOF tended to encode the stimulus or the response In order to quantify whether neurons in FOFintended to encode the stimulus or the response we we generated a Stimulus Selectivity Index (SSI) from Go aligned PETHs for correct generated a Stimulus Selectivity Index (SSI) from Go aligned PETHs for correct and error and trialserror as trials as follows: follows: P0.5 P ET Hipsi,tt 1.5 P ET Hcontra,tt SSItt = Pt= 0.5 t= 1.5 P ET Hcontra,tt + P ET Hipsi,tt If a cell fired only on contra and not on ipsi trials, then SSI=1. If a cell fired on ipsi and not contra trials, If a cell fired only on contra and not on ipsi trials, then SSI=1. If a cell fired on ipsi and not then SSI=-1. If a cell fired equallyIf for ipsifired and equally contra trials then SSI=0. computed an SSI contra trials, then SSI=-1. a cell for ipsi and contraWe trials then SSI=0. Wefor the following four trial-types (tt):correct-memory, correct-non-memory, error-memory, and error-non-memory. computed an SSI for the following four trial-types (tt):correct-memory, correct-non-memory, The contra/ipsi label referred to the instructedThe response for that trial. So fortoerror trials the PETH ipsi was error-memory, and error-non-memory. contra/ipsi label referred the instructed response constructed where the rat was instructed to go ipsi but instead went contra. We also for thatfrom trial.trials So for error trials the PETH was constructed from trials where the rat was ipsi calculated delay to interval to 0 secwent relative to Go] response interval to 0.5 sec relative to Go] instructed go ipsi[-1.5 but instead contra. Weand also calculated delay[0interval [-1.5 to 0 sec SSIs. relative to Go] and response interval [0 to 0.5 sec relative to Go] SSIs. We computed the choice probability of neurons in order quantify the covariance of neural activity We computed the choice probability of neurons in to order to quantify the covariance of neural duringactivity the delay to variability counted spikes in a 400ms ending with the Go cue on during the delaytotobehavior. variabilityWe to behavior. We counted spikes in a 400ms ending with the Go cue on memory trials where the rat was instructed to respond towards the neurons memory trials where the rat was instructed to respond towards the neurons preferred side. We used preferredtoside. We used analysis determine how well the firing in that window ROC analysis determine howROC well the firing to rate in that window predicted therate future choice of the rat. predicted the future choice of the rat. Latency Analysis WeLatency used anAnalysis alignment algorithm to find a relative temporal offset for the neural and behavior data on Wefollows. used anSingle-trial alignmentPETHs algorithm to find a relative temporal offset for the Then neuralaand behavior each trial as were generated as described previously. trial-averaged data on each trial as follows. Single-trial PETHs were generated as described previously. Then PETH was generated for each cell. For each trial we found the time of the peak of the cross-correlation a trial-averaged PETH was generated for each cell. For each trial we found the time of peak function between the PETH for that trial and the trial-averaged PETH. We then shifted each trial of the cross-correlation function between the PETH for that trial and the trial-averaged PETH. We accordingly and iterated this process until the variance of the trial-averaged PETH converged. Usually then shifted each trial accordingly and iterated this process until the variance of the trialthis process required fewer than 5 iterations. The output of this alignment procedure was an offset time averaged PETH converged. Usually this process required fewer than 5 iterations. The output for each trial,alignment which indicated the relative latency for that trial. We indicated performedthe therelative same alignment of this procedure was an neural offset time for each trial, which neural procedure on head-velocity data acquired with the video-tracking system, which produced relative latency for that trial. We performed the same alignment procedure on head-velocity a data behavioral latency each trial. We then testedwhich whether the neural latency was correlated the acquired withfor the video-tracking system, produced a relative behavioral latencywith for each behavioral whether for the the average correlation significantly different trial. latency We thenand tested whether the population neural latency was correlated with was the behavioral latency andthan zero (Bootstrapped confidence of thecorrelation mean). We also compared,different in the same way, the neural whether for the populationintervals the average was significantly than zero latencies of pairs of simultaneously recorded neurons. (Bootstrapped confidence intervals of the mean). We also compared, in the same way, the neural latencies of pairs of simultaneously recorded neurons. Head Direction Tuning Analysis Direction Analysis WeHead determined theTuning relationship between firing rate and head-direction in and out of the task in the We determined the relationship between firing andsession head-direction and out of the task following way. “In-task” epochs were defined as any timerate in the within 10inseconds rat in the following way. “In-task” epochs were defined as any time in the session 10was not inpoking in any nose port. “Out-of-task” epochs were defined by exclusion as the timeswithin the rat seconds of the rat poking in epoch any nose “Out-of-task” epochs were defined by exclusion task. The head-direction for each wasport. divided into eighteen 20 degree bins, with the centeras port the times the rat was not in-task. The head-direction for each epoch was divided into aligned to 0 degrees. Each contiguous stretch of time, in which the head direction lay within eighteen a single direction bin, was taken as a single data point. Firing rate for that data point was simply the number of spikes fired divided by time duration. Average firing rate 17 for each direction bin was obtained by averaging the firing rates for all data points for that bin; error bars are the bootstrapped 99% confidence intervals of the mean. Occupancy was defined as the number of data points that fell within each bin. Having thus 18 rate for that data point was simply the number of spikes fired divided by time duration. Average firing rate for each direction bin was obtained by averaging the firing rates for all data points for that bin; error bars are the bootstrapped 99% confidence intervals of the mean. Occupancy was defined as the number of data points that fell within each bin. Having thus obtained firing rate as obtained firing rate as a function ofwe head direction, we athen computed a summary v for cell’s each cell. a function of head direction, then computed summary vector v for eachvector cell. Each summary vectorvector was the average of a set unitoflength vectors, each each of which pointed Each cell’s summary wasweighted the weighted average of aofset unit length vectors, of which head The weighting was proportional toto the cell’s pointedinineach eachofofthe theeighteen eighteendifferent different headdirection directionbins. bins. The weighting was proportional the cell’s firing each direction bin.the Thus the vector v forcell each cell was. firing rate inrate eachindirection bin. Thus vector v for each was v= 1 b fb b=18 ⇥ ûb = [cos( b ) sin( b )] fb ûb b=1 where fb is the firing rate for direction bin b, and θb is the angle that corresponds to direction bin b. For where f is the firing rate for direction bin b, and ! is the angle that corresponds to direction example, if a cell bfired equally at every head direction, then bv would be a vector with zero length. If a cell bin b. For example, if a cell fired equally at every head direction, then v would be a vector with fired only at 40˚ and nowhere else then v would be a vector with length 1 and an angle of 40˚. zero length. If a cell fired only at 40˚ and nowhere else then v would be a vector with length 1 and an angle of 40˚. Regression Analysis ToRegression asses the extent to which delay period activity was simply a function of movement parameters, we Analysis modeled the firing rate of eachtocell fromdelay [-1.5speriod to 0.5s] relative the Goacue on correct memory trials as To asses the extent which activity wastosimply function of movement a linearparameters, function of we head angle, angular acceleration: modeled the firingvelocity rate of and eachangular cell from [-1.5s to 0.5s] relative to the Go cue on f(t) =function β1⋅φ(t) of + βhead + β3⋅φ’’(t) + β4+r(t) 2⋅φ’(t)angle, correct memory trials as a linear angular velocity and angular Whereacceleration: f(t) is the PETH of each cell. If the cell’s firing rate were a linear function of movement parameters then the residuals r(t), would on average zero. if there firing rate encoded the future f(t) = !be + !2However, ⋅"’(t) + !3⋅"’’(t) + !the 1⋅"(t) 4+r(t) choiceWhere of the f(t) rat is onthe a trial, separately from Ifthe parameters, thenfunction the linear model would PETH of each cell. thehead cell’sdirection firing rate were a linear of movement parameters the residuals r(t),about would average be would zero. remain However, if there firing rate not fully account forthen f(t) and information theonfuture choice in r(t). We the performed ROC encoded the future choice of the rat on a trial, separately from the head direction parameters, analysis on the residuals r(t), at each timepoint for left and right trials in the same way as the analysis on thenforthe linear3C. model account for S7I. f(t) and information about the future choice firing rate Figure The would results not are fully shown in Figure would remain in r(t). We performed ROC analysis on the residuals r(t), at each timepoint for left and right trials in the same way as the analysis on firing rate for Figure 3C. The results are Histology shown in Figure S7I. on the cannula implanted rats (Figure S2E-H) and electrode implanted rats Histology was performed (Figure S3C,D) once all experiments were completed. In all cases the FOF placements were within the Histology borders of M2 and between 2 and 3 mm anterior to Bregma. In all cases the M1 placements were within Histology was performed on the the cannula implanted rats (Figure S2E,F) and electrode the borders of M1 and between 2.5 and 3.5 mm anterior to Bregma (Paxinos and Watson, 2004). implanted rats (Figure S3C,D) once all experiments were completed. In all cases the FOF placements were within the borders of M2 and between 2 and 3 mm anterior to Bregma. In all cases the M1 placements were within the borders of M1 and between 2.5 and 3.5 mm anterior to Bregma (Paxinos and Watson, 2004). 18 19