Bidirectional control of orienting behavior by distinct prefrontal circuits

10 The prefrontal cortex (PFC) orchestrates voluntary behavior by biasing activity in 11 downstream structures to promote actions consistent with current task goals while 12 inhibiting inconsistent ones. PFC circuits comprise of vastly converging inputs and 13 diverging outputs, but how this anatomical diversity allows it to bidirectionally control 14 specific behaviors remains unclear. Here we use multiple approaches to show that a 15 subdivision of the mouse PFC, the anterior cingulate cortex (ACC), integrates and routes 16 discrete sensory inputs to anatomically segregated populations of projection neurons in 17 order to promote and inhibit goal-directed visual orienting responses. Surprisingly, ACC 18 outputs to the superior colliculus principally inhibit incorrect orienting movements. 19 Optogenetic analyses and a projection-based activity model make the unexpected 20 prediction that feedback from the ACC to the visual cortex is critical for correct orienting, 21 which we confirm. Integrating anatomically non-overlapping but functionally 22 complementary projections for bidirectional control may be a general organizing principle 23 for PFC circuits. 24 Animals respond to their environments using complex and diverse motor movements, but 25 are limited by being able to enact only single actions at a time. Hence, voluntary control 26 over behavior requires context-dependent mechanisms that select appropriate actions 27 and suppress complementary but inappropriate ones. Such duality of behavioral control 28 is readily apparent in sets of commonly displayed opposing behaviors, such as 29 freeze/flight, approach/avoidance, and exploration/exploitation1–6. The prefrontal cortex 30 (PFC) has been widely implicated in dynamically coordinating behavior by biasing the 31 flow of activity in downstream cortical and subcortical structures7–11, but a fundamental 32 outstanding question is how the anatomical organization of inputs to and outputs from the 33 PFC enables its proposed role. An emerging theme from recent studies is that the PFC 34 uses distinct output circuits to promote and suppress innate behaviors such as 35 conditioned fear responses12,13, reward-seeking14–16, and social interactions17. The 36 behavioral repertoire of animals is highly enriched by their ability to learn arbitrary 37 associations between environmental sensory cues and motor movements18–21. Bilateral 38 orienting (such as turn left or right) to sensory cues is a fundamental action at the core of 39 a wide range of such learned sensorimotor behaviors22–27 and involves mechanisms that 40 select and suppress specific orienting movements28. Unlike other dual behaviors that 41 often consist of asymmetric action-inaction pairs14,17,29, orienting involves a symmetrical 42 pair of actions (expression of each action has a counter action that was suppressed), 43 making it an attractive model for studying bidirectional control of behavior. Orienting 44 behavior also requires activity in the PFC30–34, but the circuit-level mechanisms by which 45 the PFC generates signals for bilateral control are unclear. Here, we use a combination 46 of virus-mediated anatomical tracing, two-photon imaging of inputs/outputs, and 47 projection-specific optogenetics to show that a subdivision of the mouse PFC, the anterior 48 cingulate cortex (ACC), uses distinct projection neuron populations to promote and 49 suppress goal-directed orienting. 50 Cortico-cortical and cortico-collicular circuits for bilateral visual orienting 51 Bilateral visually-guided orienting crucially requires integration of inputs from the two 52 halves of the visual field. Recent anatomical and functional studies in mice suggest a role 53 for the ACC in coordinating visual behavior30,31,35,36. However, it is unclear if the ACC 54 contains a subregion anatomically specialized for the control of bilateral orienting. 55 Injection of rabies viruses encoding GFP and tdTomato into caudal and rostral ACC (Fig. 56 1A) showed that although both compartments received inputs from medial higher visual 57 cortex, corresponding to functionally-defined anteromedial and posteromedial areas37, 58 the caudal ACC also received inputs from the primary visual cortex (Fig. 1B-C). Each 59 compartment also received prominent inputs from the contralateral hemisphere spanning 60 the entire extent of the ACC (Fig. 1D), suggesting that visual information relayed from the 61 visual cortex (VC) may be exchanged between ACC hemispheres via callosal projections. 62 We tested this possibility by using two-photon microscopy to image visually-evoked 63 GCaMP6s responses of VC and callosal axons in the ACC (Fig. 1E). While VC axons 64 responded nearly exclusively to stimuli presented in the contralateral visual field (relative 65 to the site of recording), callosal axons responded preferentially to ipsilateral stimuli (Figs. 66 1F, G; Extended Data Figs. 1, 2). In addition, we found that the caudal ACC sends outputs 67 to the ipsilateral motor-related layers of the superior colliculus (SCm; Fig. 1H-J), a 68 midbrain structure that coordinates visually-driven contraversive orienting movements38. 69 These anatomical and functional observations, together with previous studies 70 demonstrating that visual cortical areas have largely lateralized representations of stimuli 71 and callosal connectivity limited to the midline of the visual field37,39, suggest that the 72 caudal ACC is uniquely positioned to integrate extended visual inputs presented in either 73 hemifield and may bidirectionally control behavioral responses to them by influencing 74 activity in SCm (Fig. 1K). 75 To test this possibility, we designed a task for head-fixed mice, modified from a 76 previous design27, that allowed us to dissect the role of ACC in promoting and inhibiting 77 visual orienting behavior. Head-fixed mice are unable to orient with head or whole-body 78 movements; hence, we trained mice to orient to visual cues with their forepaws. We 79 trained mice to simultaneously learn two forepaw orienting movements so that no 80 particular action is reflexive and the selection of each requires simultaneous suppression 81 of the other. Mice had to orient to the spatial location of a visual cue presented in either 82 hemifield by rotating a trackball along a fixed axis with their forepaws to the left or right 83 (Fig. 2A, B). We provided visual feedback by coupling the rotation of the trackball to the 84 location of the stimulus on the screen in closed-loop such that correct performance 85 required mice to move the presented visual cue to the center of the screen. Importantly, 86 the use of lateralized visual cues allowed us to make precise predictions about the flow 87 of sensory information along anatomical pathways revealed by our tracing experiments 88 (Fig. 1K). Trials in which mice rotated the trackball in the direction opposite to that 89 instructed by the cue were considered incorrect and allowed a dissociation of neural 90 signals that process sensory or movement information. We used a reaction time task91 design that allowed mice to move as soon as they could after stimulus onset. This 92 minimized potential confounds of short-term memory and extensive movement planning 93 associated with delay tasks40. We first tested trained mice with stimuli of varying 94 luminance and found that behavioral performance was dependent on stimulus strength 95 (Extended Data Fig. 3A). We increased task difficulty by simultaneously presenting two 96 visual cues with different luminance intensities and requiring mice to select the cue with 97 the higher luminance. We found that performance was perceptually limited by the 98 difference in luminance of the two stimuli (Extended Data Fig. 3B). Hence, we reduced 99 perceptual ambiguity by presenting single, high luminance visual cues. Under these 100 conditions, performance errors are unlikely to be due to perceptual errors and likely reflect 101 the failure to execute the correct or suppress the incorrect movement. Experienced mice 102 performed well on this task (~90% accuracy), moving the ball rapidly at short latencies 103 (~200ms) after stimulus onset (Extended Data Fig. 3C-G). 104 Next, we conceptualized a circuit diagram for an orienting system in which the ACC 105 promotes and inhibits behavior by sending bias signals that directly or indirectly modulate 106 activity in SCm (Fig. 2C). We reasoned that VC and callosal inputs allow ACC neurons in 107 each hemisphere to represent both contralateral and ipsilateral stimuli and coordinate 108 orienting responses to them. In our task, an ipsiversive orienting movement requires 109 concurrent suppression of the contraversive movement; hence, ACC neurons active on 110 ipsiversive orienting trials may recruit pathways that suppress activity in SCm to inhibit 111 erroneous contraversive orienting. Similarly, other ACC neurons may recruit pathways 112 that enhance activity in SCm to promote correct contraversive orienting. 113 VC provides sensory inputs to, and SCm provides motor outputs of, the orienting 114 system 115 We used optogenetics to first determine the contribution of the input and output structures 116 of the orienting system, the VC and SCm, to task performance. Illuminating the cortex or 117 SCm with blue or orange light in the absence of an opsin did not produce significant 118 changes in behavior (Extended Data Fig. 4). In contrast, unilaterally inactivating the VC 119 by delivering orange light onto Jaws-expressing neurons increased the error rate on 120 contraversive trials and decreased contraversive orienting (Fig. 2D, G). While this 121 manipulation did not change the movement start time or velocity, there was an increase 122 in the miss rate (i.e., trials in which no response is made) on contralateral stimulus trials 123 (Extended Data Fig. 5A). The VC projects to various brain areas in addition to the ACC. 124 Hence, we used projection-specific optogene

Animals respond to their environments using complex and diverse motor movements, but 25 are limited by being able to enact only single actions at a time. Hence, voluntary control 26 over behavior requires context-dependent mechanisms that select appropriate actions 27 and suppress complementary but inappropriate ones. Such duality of behavioral control 28 is readily apparent in sets of commonly displayed opposing behaviors, such as 29 freeze/flight, approach/avoidance, and exploration/exploitation [1][2][3][4][5][6] . The prefrontal cortex 30 (PFC) has been widely implicated in dynamically coordinating behavior by biasing the 31 flow of activity in downstream cortical and subcortical structures [7][8][9][10][11] , but a fundamental 32 outstanding question is how the anatomical organization of inputs to and outputs from the 33 PFC enables its proposed role. An emerging theme from recent studies is that the PFC 34 uses distinct output circuits to promote and suppress innate behaviors such as 35 conditioned fear responses 12,13 , reward-seeking [14][15][16] , and social interactions 17 . The  Having established the inputs to and outputs of the orienting circuit, we asked how 145 its key node, the ACC, conveys critical signals for the task. Our observation that the ACC  1,2) suggests that it integrates this information to differentially modulate 148 activity in SCm through distinct output pathways. Hence, we used projection-specific 149 recordings and activity manipulations to critically evaluate the contribution of discrete 150 ACC output pathways to SCm.

151
Activity of ACC-SC neurons predicts ipsiversive orienting 152 We first evaluated the role of direct ACC outputs to SCm to orienting in our task. We between ACC-SC and unlabeled neurons, it is likely that additional ACC output pathways 168 also contribute to orienting behavior in our task. 169 Next, we investigated whether task-responses of ACC-SC neurons represent the 170 location of the stimulus or the direction of the orienting movement. We reasoned that 171 since the same visual cue is presented on either correct or error trials, neurons 172 representing the stimulus would respond similarly under both conditions. Likewise, since 173 animals make opposite orienting movements on these trials, we would expect neurons

193
ACC-SC neurons are active on contraversive and ipsiversive trials (Fig. 3C), suggesting 194 that they modulate both orienting movements. We tested this by expressing Jaws or 195 eNpHR3.0 in excitatory ACC neurons and locally inactivating ACC outputs to SCm during 196 the task (Fig. 4A, B). In surprising contrast to our original hypothesis (Fig. 2C), inactivating 197 the ACC-SC output pathway decreased performance on ipsiversive trials but had no 198 effect on contraversive trials (Fig. 4C). Thus, the ACC-SC pathway is principally engaged 199 during ipsiversive trials to suppress erroneous contraversive orienting. We previously 200 showed that activity in SCm promotes contraversive orienting (Fig. 2F, G). Hence, this 201 result suggests that the ACC-SC pathway specifically inhibits the SCm to suppress 202 incorrect orienting (Fig. 4D).

203
Projection optogenetics non-specifically inhibits ACC outputs to SCm, making it 204 difficult to isolate the contribution of ipsiversive responses of ACC-SC neurons to activity 205 in SCm. We reasoned that since callosal axons convey sensory information for ipsiversive neurons that suppress activity in SCm. We unilaterally expressed ChR2 in excitatory ACC 208 neurons and placed an optic fiber cannula in the opposite ACC to target callosal inputs. 209 We made extracellular recordings of spontaneous activity with a 16-channel silicone 210 probe in the SCm ipsilateral to the site of the implanted fiber and activated callosal inputs 211 with blue light (Fig. 4E, F). Laser activation of callosal ACC axons modulated the activity 212 of 32.4% (24/75) of isolated single SCm units (Wilcoxon signed-rank test, p < 0.05).

213
Inspection of single-neuron responses showed a time-locked decrease in activity with 214 laser activation (Fig. 4G). Quantifying the effect of callosal photostimulation on the activity 215 of all laser-modulated SCm neurons showed that this manipulation led to a significant net 216 decrease in activity (Fig. 4H). As a comparison, directly activating ChR2-expressing ACC  tdTomato and GABA showed that ~30% of SCm neurons that receive inputs from the ACC 232 contained GABA and hence are likely to be inhibitory neurons (Extended Data Fig. 9D).

233
Therefore, recruiting local inhibition in SCm is a possible mechanism by which ACC-SC 234 neurons inhibit activity in SCm. to suppress erroneous contraversive orienting (Fig. 4D). 236 Although the ACC-SC pathway is primarily engaged during ipsiversive orienting, calcium 237 imaging of non-SC projecting ACC neurons revealed robust responses on contraversive 238 trials (Extended Data Fig. 6B), suggesting that the ACC also contributes to correct 239 contraversive orienting. We tested this by photostimulating ChR2-expressing inhibitory 240 neurons in the ACC with blue light to inactivate it during the task. ACC inactivation 241 produced similar changes in behavior as observed with inactivation of SCm (Fig. 2F, G) 242 and led to an overall decrease in contraversive orienting (Fig. 5A). Furthermore, this 243 manipulation did not change the miss rate or movement start time, but reduced movement 244 velocity on contraversive orienting trials (Extended Data Fig. 10). Thus, activity in ACC is 245 necessary for correct contraversive orienting.

246
This finding is surprising because if ACC-SC neurons suppress the SCm during 247 ipsiversive orienting, inactivating the ACC should disinhibit SCm activity on these trials 248 and increase contraversive orienting. Hence, regional inactivation of the ACC, which 249 simultaneously manipulates activity across a range of output pathways that provide 250 extrinsic inputs to SCm 43 , must exert its effect through another ACC output that promotes 251 contraversive orienting. Regional inactivation of the VC showed that it promotes 252 contraversive orienting (Fig. 2D). Moreover, the ACC sends dense outputs to the in activity between the two hemispheres of SCm such that the side with higher activity 269 leads to its preferred contraversive orienting movement (Extended Data Fig. 11B). We

351
Similar to previous studies 20,22 , we also found that non-discriminate, regional

animals). E) ACC was injected with an AAV virus expressing
CaMKII-Jaws-GFP and its outputs to the VC were inactivated with orange light through a chronic window. F) Error rates for contraversive and ipsiversive trials with (filled) and without (unfilled) photoinhibition of ACC outputs to the VC (n = 12 sessions, 4 mice). G) Comparison of normalized laser-induced change in contraversive bias for inactivation of ACC outputs to VC (green) or SCm (red). *p < 0.025, **p < 0.01, ***p < 0.005 (Wilcoxon signed rank or rank sum test). Significance evaluated at Bonferroni-adjusted p-value of 0.025.

Prediction of behavioral deficit with ACC-VC inactivation
Use to simulate ACC-VC inactivation with E ACC-VC-SC set to 0 (eq. 5) For each unilateral optogenetic experiment, set corresponding pathway variables to 0. For example, ACC inactivation on a contra trial yields:  507  508  509  510  511  512  513  514  515  516  517  518  519  520  521  522  523  524  525  526  527  528  529  530  531  532  533  534  535  536  537  538  539  540  541  542  543  544  545  546  547  548  549 Extended Data Figure 11. Linear activity model for predicting the effect of inactivation of ACC outputs to the VC. A) An anatomically-inspired linear activity model was used to predict the contribution of ACC-VC to orienting in the task (see Methods for complete details). The schematic assumes that the left side of the brain was inactivated and accordingly designates stimuli as contralateral or ipsilateral. Activity in SCL (i.e., left SCm) and SCR was expressed as the linear sum of activity from four excitatory input pathways: 1) direct ACC-SC pathway that excites the SC on contraversive trials (E ACC-SC), 2) direct ACC-SC pathway that inhibits the SC on ipsiversive trials (I ACC-SC), 3) ACC-VC pathway that excites the SC on contraversive trials (E ACC-VC-SC), and 4) a pathway representing the baseline, task-independent activity of the SC (D). B) Equations relating contributions of these pathways to activity in the two SCm hemispheres (SCL, SCR ). We parameterized the contribution of the ACC-SC and ACC-VC-SC pathways by variables c and i representing the observed proportion of ACC-SC neurons preferentially active on contraversive and ipsiversive trials, respectively (c = 0.65 and i = 0.35; see Methods). Each trial type is described by two equations that express activity in SCL and SCR on that trial; eq. 1 shows a pair of example equations for a contraversive trial. We assume that contraversive (or ipsiversive) movements result when SCL -SCR > 0 ( or SCL -SCR < 0) for a given trial (eq. 2). C) We derived a qualitative estimate (see below) for the contribution of each of the four inputs to activity in SCm by using the average behavioral performance from control conditions and three crucial inactivation experiments (of the ACC, SC or ACC-SC). Optogenetic inactivation is represented in the model by setting the contribution of the corresponding pathway(s) to 0. Eq. 3 illustrates the representation of unilateral inactivation of the left ACC on a contraversive trial. Three of the four input pathways are dependent on the ACC and are set to 0 (red strikethroughs) for the SCL equation. Note that the inhibitory I ACC-SC pathway in the right ACC is zeroed because it relies on callosal inputs from left ACC, which is inactivated in this example. Eq. 4 shows these equations derived in a similar way for control and optogenetic inactivation of the ACC, SC, and ACC outputs to the SC, and expressed in matrix form, for contraversive trials. Each pair of rows corresponds to a specific experiment, wherein the first row is SCL -SCR (as described for eq. 2) for contra and the second row for ipsi trials (Matrix A). The unknown variables are in vector , and the observed behavioral performance for contralateral and ipsilateral stimulus trials for each experiment is given by vector b. Generically, this relationship is expressed as A = b. Determining the contribution of each pathway to the difference in activity between the two SCm hemispheres, and hence contraversive or ipsiversive orienting, requires solving for the unknown variables in vector . We derived an estimate for this vector, termed , using a common linear algebra technique (least-squares approximation) from each of the three optogenetic experiments (denoted by the subscript j; see Methods). Note that estimates the contribution of the ACC-VC pathway (E ACC-VC-SC), but the behavioral results from its optogenetic inactivation were not included in the derivation of . D) We simulated inactivation of the ACC-VC pathway on the left side by setting E ACC-VC-SC to 0 and derived an equation for the difference in activity between the two SCm hemispheres (SCL -SCR) for contralateral and ipsilateral trials (eq. 5), similar to that done for deriving matrix A in eq. 3. Multiplying this matrix P by gave three predictions of behavioral results expected from inactivation of ACC-VC pathway, one from each of the three optogenetic experiments considered in the model. E) Schematic summary of behavioral results observed with inactivation of the SC, ACC, and ACC-SC pathway. The model predicts that inactivation ACC-VC pathway decreases contraversive orienting.

707
The following surgical procedures were performed for optogenetics or imaging 708 experiments. For experiments requiring the use of dental acrylic, we removed a portion 709 of the scalp using spring scissors, scraped away the periosteum membrane overlying the 710 skull, and used a dental drill to abrade the skull to improve adhesion. Same animals were 711 used for inactivation of the visual cortex or its projection to the anterior cingulate cortex. 712 We drilled a 3mm craniotomy over the visual cortex and made 8-12 injections (100nL  anterior, relative to Bregma). 806 We injected rabies viruses encoding GFP or tdTomato into the visual cortex or the 807 superior colliculus to identify ACC neurons projecting to these structures. We counted the 808 total number of back-labeled neurons in the ACC (AP range 0 to 1mm), every 200 µm in each animal and quantified the proportion of neurons that were labeled with GFP, 810 tdTomato, or both out of all labeled cells. 811 We performed anterograde transsynaptic tracing experiments 42 to identify SC 812 neurons that receive inputs from the ACC. We produced the Flp-dependent tdTomato   Mice were taken through successive stages of training until they became proficient 867 at the task. Once mice recovered from the surgery, they were water restricted for 5-7 days 868 (≥ 1ml/day) and then trained to lick the metal spout to obtain small water rewards (3-6 869 µL). If mice did not receive their water allotment during training, they were given the 870 remaining amount as hydrogel (Clear H2O) in their home cage. After mice reliably licked the water spout, they earned water rewards by using the trackball to move the presented 872 stimulus to the center of the screen. To discourage spontaneous trackball movements, 873 mice were required to hold the ball still for 1s to trigger trial start, which was signaled with 874 an auditory tone (0.5s, 1 KHz); the visual cue appeared with a 1s delay after trial start.

875
During early stages of training, only movements in the correct direction contributed to 876 movement of the stimulus. Once mice reliably moved the ball in either direction on >90% 877 of trials, this condition was removed and the movement of the stimulus was fully coupled 878 to the movement of the trackball. In the next stage of training, we used an anti-bias 879 algorithm in which the same stimulus was repeated on consecutive trials if mice made an 880 error until they performed the trial correctly; stimulus location on trials following correct 881 trials were randomly chosen. Once performance reached 70%, the anti-bias algorithm 882 was turned off and stimuli were presented in a randomized manner. Throughout all 883 stages, white noise was used to signal miss trials if mice failed to move the ball to 884 threshold before expiration of the response window. As mice progressed through the 885 training stages, we gradually decreased the response time from 10s to 1-2s. Correct and 886 incorrect trials were signaled with an auditory tone (0.2 kHz and 10kHz, respectively), 887 followed by an inter-trial delay of 2.5s. In a subset of mice, incorrect trials were punished 888 with a brief timeout (3s), lengthening the inter-trial delay to 5.5s.

889
Animals trained for two-photon imaging experiments were taken through two 890 additional stages of training. First, we turned off the closed-loop coupling between the 891 trackball and the stimulus, and flashed stimuli for 100-200ms. Second, we introduced 892 uncertainty in temporal expectancy for stimulus onset by randomizing the period 893 (exponential distribution with mean of ~1.8 seconds, min and max delay of 1 and 5s, 894 respectively) between the auditory cue signaling trial start and the onset of the visual 895 stimulus. Once animals achieved criterion performance of 70% correct with randomized 896 trials, we commenced optogenetic or imaging experiments. 897 We also trained a subset of mice on two variants of the task to interrogate their 898 psychometric performance. In the first variant, we presented a single visual cue as before 899 but of varying luminance values. In the second variant, mice were presented with two 900 cues simultaneous (target and distractor). The luminance of the target cue was set to 64% 901 while that of the distractor cue ranged from 0 to 48%; animals were rewarded for orienting 902 to the target cue. We quantified performance (excluding miss trials) for each target-  Optogenetic manipulation of behavior: Photostimulation was provided with a solid 929 state 473nm blue laser for experiments using ChR2 and a 593nm orange laser for 930 experiments using Jaws or eNpHR3.0 (OptoEngine). Laser stimulation was triggered from 931 the behavior control computer and lasted from 0.3s before to 1s after visual cue onset.