Monosynaptic trans-collicular pathways link mouse whisker circuits to integrate somatosensory and motor cortical signals

The superior colliculus (SC), a conserved midbrain node with extensive long-range connectivity throughout the brain, is a key structure for innate behaviors. Descending cortical pathways are increasingly recognized as central control points for SC-mediated behaviors, but how cortico-collicular pathways coordinate SC activity at the cellular level is poorly understood. Moreover, despite the known role of the SC as a multisensory integrator, the involvement of the SC in the somatosensory system is largely unexplored in comparison to its involvement in the visual and auditory systems. Here, we mapped the connectivity of the whisker-sensitive region of the SC in mice with trans-synaptic and intersectional tracing tools and in vivo electrophysiology. The results reveal a novel trans-collicular connectivity motif in which neurons in motor- and somatosensory cortices impinge onto the brainstem-SC-brainstem sensory-motor arc and onto SC-midbrain output pathways via only one synapse in the SC. Intersectional approaches and optogenetically assisted connectivity quantifications in vivo reveal convergence of motor and somatosensory cortical input on individual SC neurons, providing a new framework for sensory-motor integration in the SC. More than a third of the cortical recipient neurons in the whisker SC are GABAergic neurons, which include a hitherto unknown population of GABAergic projection neurons targeting thalamic nuclei and the zona incerta. These results pinpoint a whisker region in the SC of mice as a node for the integration of somatosensory and motor cortical signals via parallel excitatory and inhibitory trans-collicular pathways, which link cortical and subcortical whisker circuits for somato-motor integration.


Introduction
The superior colliculus (SC) is part of a phylogenetically ancient brain network that directs quick motor actions in response to ascending sensory signals [1,2]. As such, the SC is a central hub for multiple sensory-motor arcs linking sensory information to motor actions. From an PLOS Biology | https://doi.org/10.1371/journal.pbio.3002126 May 19, 2023 1 / 29 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 evolutionary perspective, much of the SC's function in mammalian brains has been taken over by the neocortex, via parallel, cortically controlled sensory-motor arcs [3]. Intriguingly, recent work in the visual and auditory systems demonstrate that SC-mediated behaviors are modulated by cortical inputs [4][5][6], raising the question of how cortex-and SC-mediated sensorymotor arcs interact to organize desirable behavior. More specifically, it is not well understood how cortico-collicular pathways engage with SC microcircuitry and, in turn, generate SC output signals to the brainstem and diencephalon (including thalamus). This information is essential not only to understand how cortico-collicular pathways may coordinate between the "new" cortical and the "old" collicular arcs, but also to answer principal questions concerning the identity of cortical-recipient (i.e., targeted by cortex)-SC neurons: to where do they project, are they excitatory or inhibitory, and do they receive sensory and/or motor signals? SC circuits have mostly been studied in relation to vision [2,[6][7][8], i.e., looking at functions of the "visual SC." In contrast, somatosensory functions of the SC are less studied and less understood, even though somatosensory functions such as whisker sensation are vital for rodents and other animals [9][10][11][12]. This study set out to address the involvement of "somatosensory SC" circuits in the whisker system and focused on 3 principal questions about the organization of trans-collicular pathways that mediate cortical input to SC downstream targets (Fig 1).
(1) To where do cortical-recipient SC neurons project? Cortico-collicular signals may be routed directly to SC downstream targets (i.e., in the diencephalon and brainstem) via monosynaptic trans-collicular pathways and/or more indirectly, via intracollicular connections to SC output neurons [2] (Fig 1, left panel). While intracollicular circuits are well described [13], evidence for monosynaptic trans-collicular pathways is limited because relatively few studies have delineated the precise input-output connectivity of defined subpopulations in SC [2,14]. Indeed, input-output connectivity of the SC is mostly considered at the level of SC layers [2,15] and radial cortico-collicular input zones [16]. However, these mesoscopic modules comprise intermingled populations of input and output neurons as well as input neurons with different input identities [16][17][18], for example, retinal and cortical inputs [19]. Therefore, dissection of input-defined cell types and analysis of their outputs is critical to unambiguously link specific SC input pathways to specific SC downstream targets and to determine whether cortical input signals can be transformed directly into SC output to downstream SC target circuits.
(2) Are cortical-recipient SC neurons excitatory or inhibitory? The SC comprises excitatory and inhibitory neurons; for example, in the visual SC, approximately one-third of the neurons are GABAergic [15,20]. However, very little is known about GABAergic neurons in the somatosensory SC (Fig 1, center panel), including their proportions, their inputs and outputs, and whether they are classic local interneurons or projections neurons. Determining how cortical pathways engage specific excitatory or inhibitory SC circuits and identifying which downstream pathways emanate from these circuits is indispensable to make circuit-mechanistic predictions about recently discovered top-down modulation of SC-mediated innate escape and defense behaviors [4][5][6].
(3) Do pathways from motor and somatosensory cortices converge at the level of individual SC neurons? The multisensory nature of the SC has motivated a large body of studies demonstrating the integration of visual, auditory, somatosensory, and motor signals in the SC [17,18,[21][22][23][24][25]. Nevertheless, whether sensory and motor cortex pathways converge in individual SC neurons remains untested (Fig 1, right panel). The lateral SC (LSC) has been identified as a putative point of convergence of somatosensory and motor cortical efferents [16], making it a promising target to test this hypothesis.
Here, we mapped the input-output connectivity of the mouse SC on the single-cell level with trans-synaptic tracing, intersectional viral approaches, and optogenetic-assisted electrophysiology. In awake animals, we identify a whisker-sensitive region in the SC, which receives trigeminal and cortico-collicular input from motor and somatosensory cortex. We then targeted specific subsets of input-defined pathways from the brainstem and motor and somatosensory cortices to the SC and traced their axonal outputs to downstream targets, revealing direct trans-collicular pathways, which provide disynaptic long-range links between cortical pathways and SC target regions. Trans-synaptic labeling in combination with optogenetic input mapping reveals long-range input convergence on the level of individual SC neurons, with approximately one-third of the cortical recipient SC neurons receiving convergent input from both motor and somatosensory cortex. We find that long-range input pathways extensively target inhibitory SC neurons, which, in turn, give rise to long-range GABAergic projections to thalamic nuclei and the zona incerta (ZI). In sum, this study pinpoints a "whisker SC," in which converging cortical and brainstem inputs innervate GABAergic and non-GABAergic SC neurons, which, in turn, provide parallel trans-collicular pathways to downstream whisker circuits in the diencephalon and brainstem. These results suggest that this cortical control of SC directly affects the brainstem-SC-brainstem sensory-motor arc, as well as the outputs from SC to diencephalic stations.

The whisker-sensitive LSC receives input from the motor cortex and barrel cortex and from the brainstem
In awake mice, we tested neuronal responses to whisker deflections by targeting silicon probes to different locations in the lateral SC (Fig 2A and 2B), which has been shown in anesthetized animals to be whisker sensitive [26]. Whisker deflections were induced with an airpuff to whiskers contralateral to the recording sites. We found a varying degree of whisker-modulated collicular neurons across 12 recordings from 8 mice with approximately 30% of the units with significant modulation (325/1,005 units modulated, 302/325 positive, 23/325 negative; Figs 2C and S10, respectively). Whisker responses were clearly bimodal, with fast and slow components corresponding to "trigeminotectal" and "corticotectal" SC drive, respectively [25]. The locations of the probes were determined by post hoc localization of dye replacement. To visualize the location of whisker-sensitive units in the LSC, we then pooled and mapped all recorded units onto LSC outlines in coordinate space and computed their modulation indices ( Fig 2C).
Having identified the whisker-sensitive region in the LSC, we next wanted to determine the long-range inputs to that region. To do so, we targeted a retrograde virus (rAAV-ChR2-tdTomato; [29,30], Fig 3A) to the whisker-sensitive region in LSC, based on stereotaxic coordinates estimated from the recordings (Fig 2). LSC-projecting neurons were found in multiple cortical and subcortical areas (Fig 3B), including whisker-related areas in the ipsilateral whisker motor cortex (MC), in the barrel cortex (BC), and in the contralateral trigeminal complex in the brainstem (Bs), (Fig 3C-3E; whisker motor cortex [31] MC: M1 and M2; see S1 Fig). In addition to these whisker-related projections to SC, we found SC-projecting neurons in the auditory, insular, and ectorhinal cortices as described before [16].
The laminar profiles of LSC-projecting neurons in MC and BC illustrate that in both cortices, cortico-collicular neurons originate in layer 5 (L5) (Fig 3C and 3D). The L5 origin of cortico-collicular projections was further confirmed by registering LSC-projecting somata with layer 6 (L6)-specific EYFP fluorescence in the Ntsr1-EYFP reporter mouse line, showing segregation of these 2 subcortical projection neuron types ( Fig 3C). Moreover, virus-mediated labeling of cortical boutons revealed dense innervation of LSC by MC and BC, with moderately sized cortico-collicular boutons (median diameters: BC 1.15 μm, MC 1.38 μm; S2 Fig), consistent with recent reports [32,33]. In summary, the whisker-sensitive region of the LSC receives direct and dense input from whisker areas in the brainstem and from L5 neurons in the barrel and motor cortices.

Cellular organization of somatosensory and motor pathways in the lateral SC
We next sought to directly target MC-, BC-, and Bs-recipient neurons (RNs) in the LSC to study their organization, cellular identity, and projection patterns. To do so, we employed a trans-synaptic anterograde approach, which is based on the ability of the AAV1 serotype to trans-synaptically jump to postsynaptic neurons [4]. We injected AAV1-Cre in combination with AAV2-DIO-mCherry into MC, BC, and Bs to simultaneously label the injection sites and their axonal projections with mCherry (red) and to express cre in the synaptically connected postsynaptic target neurons in LSC. Finally, Cre-expressing RNs were revealed by injecting AAV2-DIO-EGFP into LSC (green) (Figs 4A, 4B, S3 and S9). This strategy allowed us to visualize and reconstruct the 3 RN populations (MC-RNs, 6 mice; BC-RNs, 6 mice; Bs-RNs, 7 mice) and register them to standard anatomical borders within SC. This revealed that the 3 whisker-related input pathways predominantly target neurons in the intermediate layer of the LSC (Fig 4C). The RN types largely overlap along the antero-posterior axis ( Fig 4D) but are more segregated in the dorsal-ventral and medial-lateral dimensions. Here, the organization shows a lateral Bs-recipient zone, adjacent to a medial cortical-recipient zone with considerable overlap between MC-and BC-RNs (Fig 4E and 4F). We overlaid these recipient zones on      [16]. The comparison shows that recipient zones fall into the radial zones as follows: BS ! lateral / centrolateral; BC ! centrolateral; MC ! centrolateral / centromedial ( Fig 4E).
Together, the anterograde and retrograde tracing experiments identify a region within the intermediate layers of the LSC, which spans approximately 1.2 mm in the rostro-caudal axis (approximately 85% of SC's extent) and which receives input from main whisker circuits in the cortex and brainstem. This region is highly whisker sensitive (Fig 2), suggesting that the intermediate layer of the LSC is a "whisker SC" located ventrally to the "visual SC."

Cortical and brainstem pathways directly target GABAergic neurons in the LSC
Are the neurons targeted by cortical and brainstem long-range input to the LSC excitatory or inhibitory? To address this question, we first estimated the proportion of GABAergic neurons in the LSC. Using a GAD-GFP mouse, in which GABAergic neurons express GFP, we colabeled all neurons using the pan-neuronal marker NeuN-Alexa 647 ( Fig 5A) and estimate that approximately 23% of the LSC neurons in the intermediate layer are GABAergic (Figs 5F and S6), which is slightly lower than in the superficial SC (approximately 30%; [15]).
We then tested for monosynaptic innervation of GABAergic LSC neurons by employing a trans-synaptic intersectional strategy that allowed us to separate GABAergic RNs (iRNs) from the population of RNs, for each long-range input pathway ( Fig 5B). In GAD-Cre mice, which express Cre in GABAergic neurons [34], we injected anterograde virus (AAV1-Flpo) into MC, BC, or Bs, as well as conditional reporter viruses (AAV-ConFon EYFP, AAV-fDIO mCherry) into LSC. This intersectional approach differentially labeled RNs with mCherry and iRNs with EYFP, demonstrating direct innervation of GABAergic neurons by MC, BC, and Bs input pathways (Figs 5C and S4 and S5 for controls). Reconstructions of the 2 populations show that iRNs and RNs are intermingled for all 3 input pathways ( Fig 5D). Our estimates of the iRN/ RN proportions suggest that each pathway targets between 34% and 37% GABAergic neurons along the full extent of the recipient antero-posterior axis (Figs 5E-5G and S11).
In summary, long-range input pathways from MC, BC, and Bs extensively target GABAergic neurons in the LSC. The proportion of iRNs exceeded the proportion of GABAergic neurons (>34% versus 23%; Fig 5G), suggesting a preferential targeting for GABAergic neurons by these long-range input pathways.

Convergence of somatosensory and motor cortex in LSC neurons
The considerable overlap between recipient zones in LSC observed in Fig 4 suggests possible convergence of long-range input pathways on the level of individual LSC neurons. To test for monosynaptic input convergence in single LSC neurons (CVG-RN), we employed an intersectional strategy as follows: In the same mouse, we injected 2 projection sites-for example, MC and BC-each with a different trans-synaptic variant-AAV1-Cre or AAV1-Flpo. Injections of the double-conditional (Cre-and Flpo-dependent) reporter virus in SC indeed revealed CVG-RNs for all 3 pathways (Fig 6A and 6B). MC and Bs convergence is highest in the anterior LSC, while MC and BC and BC and Bs convergences peak in the center and posterior SC, respectively ( Fig 6C). mean ± SEM. The data for Fig 4D-

PLOS BIOLOGY
Trans-collicular whisker circuits    To quantify the degree of convergence, we compared (1) z-scored mean CVG-RN counts per input pair and (2) the proportion of CVG-RNs with respect to RNs between all input pairs. Both analyses revealed that MC and BC convergence stands out compared to the other convergence pairs. Notably, the z-scored MC and BC CVG-RN estimates a 2.33-fold higher deviation compared to Bs and cortex convergence ( Fig 6D).
To estimate the proportion of convergence neurons relative to all recipient neurons, we counted mCherry-labeled RNs and EYFP-labeled CVG-RNs (Figs 6A and S12). Cortico-cortical convergence was estimated to be about 2.3-fold higher than brainstem-cortical convergence (approximately 23% CVG-RN for the MC and BC input pair versus approximately 10% CVG-RN for both the BC and Bs and MC and Bs input pairs) (Fig 6E). Fitting a two-component Gaussian Mixture Model to the coordinates of MC and BC CVG-RNs localized the cortico-cortical convergence zone to the dorsal portion of the intermediate layer ( Fig 6F).
Thus, motor and barrel cortex pathways converge on single neurons in the LSC, highlighting the whisker SC as a node for the integration of somatosensory and motor cortical signals. This convergence motif could potentially support fast, temporally precise integration of signals from different cortical regions; therefore, we next tested this possibility at the functional level.

Single LSC units integrate sensory-and motor cortex inputs
To validate functional convergence of MC and BC in LSC, we made electrophysiological recordings in LSC in combination with optogenetic stimulation of L5 neurons in MC and BC in anesthetized mice. We used Rbp4-Cre-ChR2-EYFP mice (n = 4), in which L5 neurons express ChR2 [35,36] and successively stimulated MC-L5 and BC-L5 during the same experiment by using a movable fiber optic placed above the pial surface ( Fig 7A).
By characterizing the spiking activity of individual SC units in response to successive optogenetic stimulation of L5 in MC or BC (Fig 7B), we identified 30 units that responded to either MC-L5 or BC-L5, from which 9 responded to both ( Fig 7C). Unit spike latencies in response to optogenetic stimulation of L5 were approximately 9 to 12 ms (Fig 7D, exact values in S3 Table), comparable to the latencies of optogenetically evoked spiking in L5 and L5-innervated downstream neurons in the thalamus [37,38], suggesting monosynaptic activation of LSC neurons via L5-SC synapses.
Thus, together with the anatomical tracing, these results show that the LSC integrates dual functional input from both motor and barrel cortex layer 5 neurons in a subpopulation of about 20% to 30% of the recipient neurons.

Trans-collicular pathways via cortico-and trigemino-collicular projections to the diencephalon and brainstem
It is not known whether cortico-collicular signals can be routed directly to LSC downstream targets via monosynaptic trans-collicular pathways. If this is the case, cortico-collicular input neurons in the LSC would be required to form long-range pathways that leave the SC. To test this possibility, we leveraged the trans-synaptic labeling approach to search for axonal signals from RN labeling experiments (3 mice) throughout the brain. Indeed, we observed abundant RN axons and varicosities in multiple target areas in the brainstem and diencephalon (reconstructions in Fig 8A; high magnification examples in S7 Fig).
To estimate relative innervation strengths of the target nuclei, we computed RN-output maps from reconstructed axon counts (Fig 8B) to visualize relatively enriched or sparse innervation. All 3 trans-collicular pathways innervated multiple diencephalic nuclei, with the highest axonal density in ZI and sensory-motor-related nuclei in the thalamus, i.e., the posterior (PO), ventro-medial (VM) and parafascicular (PF) nuclei. In the brainstem, the strongest

PLOS BIOLOGY
innervation was found in the pontine reticular (PnC; caudal part), the gigantocellular reticular (Gi), and the facial nucleus (7N). The innervation strength of a given target was often similar for the 3 trans-collicular pathways, particularly for brainstem targets. The output maps differed more strongly in the thalamus, for example, in the lateral-posterior nucleus (LP), which received relatively enriched innervation by Bs-RN axons.
We next analyzed the inhibitory RN pathways and asked if they solely resemble classic interneurons, which only project within intracollicular circuits, or if they also give rise to longrange projections outside the SC (GABAergic projection neurons). Unexpectedly, we found that iRNs from all 3 pathways give rise to high density innervation of the ZI as well as projections to PO and PF (Figs 8C and S8). In contrast, no iRN projections were found in the brainstem. Thus, trans-collicular pathways extend to long-range GABAergic output with hitherto unknown inhibitory functions in thalamic nuclei and the ZI. Based on the observed axonal projections, both RNs and iRNs innervate the dorsal and ventral ZI (Figs 8A, 8C, S7A, S7B, S8A and S8B). Since the ZI is a main source of GABAergic projections to higher-order thalamus [39], this connectivity motif suggests inhibitory and disinhibitory gate control of higherorder thalamus via RN and iRN pathways, respectively.
In summary, we found excitatory and inhibitory monosynaptic trans-collicular pathways (Fig 9), suggesting that at least part of the input signals from cortex and brainstem are directly transformed into SC output to rapidly route cortical and brainstem information to diverse points along the neural axis.

Trans-collicular pathways
The superior colliculus-a highly conserved sensory-motor structure in vertebrates-is said to be one of the best-characterized structures in the brain [2,14,15]. However, precise understanding of how particular input pathways to the SC are synaptically connected to particular SC output targets has been hampered due to the lack of tools to determine the connectivity features of input-defined SC neurons. Moreover, such anatomical constraints are necessary to develop functional predictions and testable hypotheses for sensory-motor computations in the SC and, in a broader context, to understand how cortex-mediated and SC-mediated sensorymotor arcs relate to each other.
The present study elucidated the input-output organization of the mouse LSC by tracing input-defined pathways from the brainstem and motor and somatosensory cortices to the LSC and further on to collicular downstream targets in the diencephalon and brainstem (Fig 9). We thereby identified a connectivity scheme with significant implications for collicular computations: trans-collicular pathways, monosynaptically traversing through the LSC and directly linking cortical and brainstem whisker circuits to downstream targets in the brainstem and diencephalon. The connectivity scheme followed by these trans-collicular neurons differs from that found in other layered circuits-for example, compared to the canonical circuit framework of the cortex, in which layer 4 neurons are considered input neurons that form only intracortical connections. In contrast, the SC input neurons identified here directly provide long-range output axons to multiple subcortical targets, suggesting that at least part of the input signals are directly transformed into SC output via trans-collicular pathways.
In awake animals, we found that single neurons in the intermediate layer of the LSC are highly sensitive to whisker stimulation, in accordance with earlier work in anesthetized rodents that showed whisker sensitivity in the LSC [26] and trigeminal and cortical input from the somatosensory barrel and motor cortex [16,25]. Our results expand on this work by demonstrating that the whisker-sensitive LSC is densely interconnected with whisker circuits throughout the brain, both in terms of inputs and outputs, together suggesting that the intermediate layer of the LSC serves as a whisker SC.
The microanatomical analyses and mapping data we present here allow several functional predictions for whisking behavior. Firstly, we found that brainstem long-range inputs to the LSC turn into direct excitatory outputs to the 7N, a brainstem motor nucleus for whisker movements [40]. This finding predicts a short sensory-motor loop that could mediate the "minimal impingement" phenomenon, which animals use to optimize tactile sensitivity by minimizing the force of tactile sensors upon surface contact [41]. Our data suggest that whisker touch signals ascending from the brainstem trigeminal nucleus to the LSC are monosynaptically transformed into motor output signals to the 7N, as a direct sensory-motor loop to stabilize whisker movement force, while animals move relative to objects.
Furthermore, in the thalamus, the 3 investigated trans-collicular pathways from MC, BC, and Bs target higher-order thalamic nuclei, such as PO, but not first-order thalamic nuclei (Fig  8). This finding is particularly relevant for higher-order thalamic computations of cortical signals because it suggests parallel entry routes of cortical L5 signals to the higher-order thalamus: an indirect L5 trans-collicular pathway and a direct cortico-thalamic pathway [33,[42][43][44].
Another significant outcome of our mapping data is the link between the whisker and the visual system: whisker trigemino-collicular pathways send abundant projections to the visual LP-the rodent pulvinar in the thalamus (Fig 8). This result provides an anatomical substrate for the recent finding that the LP integrates visual and tactile information [45]. The integration of trans-collicular whisker signals in the LP may be an important mechanism underlying visuo-tactile functions such as hunting behavior, in which the SC has been shown to play a key role [8].

Long-range input convergence in the LSC
The multimodal nature of the SC is well established and functional observations of multisensory SC neurons [18] consistently support multimodal convergence in the SC. However, only few studies directly demonstrated long-range convergence at the level of individual SC neurons using anatomical means [19,46,47], and, to the best of our knowledge, the possibility of specific convergence between motor and sensory cortical pathways in the SC had not yet been tested. Using anatomical and functional tools to test convergence of input pathways from MC, BC, and Bs in individual LSC neurons, we found subpopulations of LSC neurons, receiving monosynaptic input from 2 input pathways. While the estimated proportion of brainstem-cortical convergence neurons was approximately 10%, the proportion of cortico-cortical convergence neurons (MC and BC) was more than twice as high (approximately 23%). We demonstrate that convergence between motor and sensory cortex is functional by showing that approximately 30% of the individual LSC units receiving functional cortical input are responsive to optogenetic stimulation of L5 neurons in both MC and BC. The integration of convergent L5 signals from sensory and motor cortices by single SC neurons suggests a low-level neuronal substrate for sensory-motor integration in the SC from which functional predictions in the context of the execution and control of motor plans [24] can be derived. In this framework, inputs from MC inform the SC about an intended motor action, i.e., an "efference copy" of the motor command [48], while BC provides information on the resulting sensory status, thereby enabling mechanisms to initiate small motor command readjustments and/or context-dependent dampening or enhancement of sensation, possibly analogous to the SC mechanisms proposed for the stabilization of visual perception [49]. However in the context of this proposition, it also needs to be noted that L5 neurons of both S1 and M1 convey motor-related signals, i.e., are not fully independent [31,50].

Long-range innervation of GABA neurons in the LSC
The SC contains substantial populations of GABAergic neurons with reported numbers ranging from 30% to 45% [15]. We estimate that the LSC contains approximately 23% GABAergic neurons. Rather unexpectedly, we found that GABAergic neurons are directly innervated by long-range pathways from MC, BC, and Bs. Interestingly, the estimated proportion of innervated GABAergic neurons was between 34% and 37% and thereby significantly exceeded the estimated proportion of GABAergic neurons in the LSC (approximately 23%), suggesting an important functional role for the recruitment of inhibitory SC-circuits by cortical and brainstem long-range inputs. Furthermore, a recent report using anterograde trans-synaptic viral tracing found that nearly half of the SC neurons downstream of the secondary motor cortex are GABAergic [24]. Together, these findings suggest that the dual recruitment of inhibitory and excitatory circuits in the SC provides cortico-collicular pathways with the ability to flexibly dampen or enhance SC-mediated innate behaviors.
Our mapping data demonstrate that the populations of targeted GABAergic neurons contain projection neurons. Notably, these GABAergic projections target PO and the ZI but not the brainstem. Thus, the PO and the ZI, the latter of which is a potent source of inhibition for higher-order thalamic nuclei [51], receive both GABAergic and non-GABAergic trans-collicular input, therefore predicting bidirectional control of these LSC targets. Future investigations of how the ZI is modulated by whisker signals from parallel GABAergic and non-GABAergic trans-collicular pathways will likely generate substantial insight into the role of whisking on ZI's myriad functions, which include locomotion, feeding, hunting, pain regulation, and defense behaviors [7,8].
What is the relation between the "older" collicular and the "newer" cortical arcs of sensorymotor loops? In the mammalian brain, in which both arcs are densely interconnected [52], and as suggested by the present study, one possible distinguishing function is the primary foci of their corresponding "brain-world" loops: self-motion for SC-mediated loops and external objects for cortical-mediated loops. However, it should also be noted that the mammalian SC maintains a direct leg to external sensory signals (for example, the brainstem inputs described here) upon which quick motor actions can most likely be computed independently of the cortex. The cortex may thus provide a dynamic interference conduit to these SC-generated sensory-motor functions to enhance behavioral flexibility-for example, by "vetoing" SCmediated behaviors via selective recruitment of the GABAergic pathways depicted in Fig 9. Achieving a functional understanding of the interactions between cortical and collicular sensory-motor loops is a valuable challenge for the future. The reagents and resources used in this study are summarized in Table 1.

Animals
Mice (male and female, 10 to 20 weeks of age at the time of virus injection) were housed with food and water ad libitum on a 12-h light/dark cycle.

Virus-mediated labeling and intersectional strategies
Retrograde labeling: To identify input regions to the LSC (Fig 3A), we targeted the retrograde rAAV-TdTomato [30] into the whisker-sensitive region of the LSC (Fig 2). To delineate SC projection neurons (Td-Tomato) from L6 neurons, these injections were done in Ntsr1-EYFP mice, in which L6 but not L5 neurons express EYFP [54,55].

Trans-synaptic anterograde and intersectional labeling:
To label input-defined recipient neurons, we made use of the ability of AAV1-Cre and AAV1-Flpo particles to jump to postsynaptic neurons via vesicle release [4,57,58]. We adopted this strategy to reveal different subsets of SC neurons defined by their monosynaptic input from whisker-related projection sites of interest (MC, BC, Bs) and to characterize intersections between overlapping cell populations, by using different combinations of anterograde and reporter viruses ( Table 2). For example, the RNs contain a subpopulation of inhibitory neurons (iRNs), and we employ intersectional approaches based on the differential labeling of neurons, which fulfill different conditions. For RNs to be labeled, they need to receive input from an area of interest (for example, BC). For the iRNs to be labeled in a different color, they need to fulfill both, the previous condition, AND the additional condition to express cre (from the GAD-Cre line). We used the following strategies to label: (1) Recipient neurons (RNs); (2) Cortical and brainstem RNs; (3) Inibitory recipient (iRNs); and (4) Convergence neurons (CVG-RNs) as follows:

Stereotaxic surgeries
Mice were placed in a stereotaxic frame (Kopf Instruments) and anesthetized with 1.2 to 2 vol % isoflurane in medical oxygen at a flow rate of 0.8 l/min while keeping body temperature at 39˚C. Carprofen (CP-Pharma) was administered subcutaneously (5 mg/kg) and lidocaine (Xylocaine 1%, Aspen Pharma) was injected under the scalp and around ear bars for local anesthesia. Eyes were covered with Bepanthen ointment (Bayer) to keep them lubricated during the surgery. After ensuring the absence of tail-and toe-pinch reflexes, the skin was opened with a midline incision, the periosteum and aponeurotic galea were removed to visualize the reference points bregma and lambda, and the head was aligned to the stereotaxic frame. Small craniotomies were drilled above the injection sites and viral particle solutions were injected with calibrated glass micropipettes (Blaubrand; IntraMARK) at the following coordinates relative to bregma (AP, ML) and pia (DV), summarized in Table 3.
After viral particle injections, the pipette was left in place for 10 min to allow diffusion and was then slowly retracted. Finally, the skin above the skull was sutured with silk (PermaHand, 6-0; Ethicon), and the animal was transferred to its home cage.

Slicing and immunolabeling
After virus expression, mice were exposed to a lethal dose of Ketamine (80 mg/kg) and Xylazine (10 mg/kg) and transcardially perfused with 4% paraformaldehyde (PFA) in PBS. The brain was removed and postfixed for 12 h in 4% PFA at 4˚C. The same procedure was applied for GABA immunofluorescence labeling except using 2% glutaraldehyde and 2% PFA in PBS. Brains were sliced (50 μm) with a vibratome (Thermo Scientific Microm HM650V).

Microscopy
Injection sites, labeled cells in SC, and axonal outputs were imaged with a fluorescent microscope (Leica DM6000) outfitted with an automated stage to generate mosaics (Andor Neo camera, SCC02104) and laser lines 365, 470, 505, 530, and 625 nm. The objective HC PL APO 10×/0.40Dry was used to image the injection site and labeled cells in SC (voxel size 2 × 2 μm). We used the objective HC PL APO 20×/0.70Dry (voxel size 1 × 1 μm) for imaging RN axonal outputs. Varicosities, retrogradely labeled cells, and immunolabeled cells were imaged with a confocal microscope (Nikon A1R or C2, Nikon Imaging Center Heidelberg, Germany) outfitted with an automated mosaic stage and laser lines 405, 488, 561, and 640 nm. Objectives used: Nikon Plan Apo 20× NA 0.75 Nikon Plan Fluor 40× NA 1.30 oil immersion, Nikon Apo λS 60× NA 1.40. Slices were scanned in sequential mode; scanner zoom: 1× to 3×; resolution of individual tiles: 2,048 × 2,048 pixels. The laser power, photomultiplier gain, and pinhole (20 to 60 μm) were adjusted so that the signal was not saturated.

Cell counting
Fluorescence image series of SC were loaded into Fiji and manually rotated to align all slices along the midline. From each image series, we selected the section that matched the most rostral part of SC (AP: −3.08 mm) in Paxinos' atlas [59], so that all series contained the same domain (AP: −3.08 to −4.28 mm). To select the most rostral section, we compared the shape and size of the hippocampus, optic tract, anterior pretectal nucleus, medial geniculate nucleus, and SC and its brachium anatomical references against Paxinos' −3.08 mm slice. The resulting SC series were processed in Fiji as follows. First, as a measure of background, we took the average pixel intensity of 3 different regions in the non-recipient area and subtracted this average value from each section in the series. Images were then contrast enhanced with an exponential transformation (Fiji "Exp" function), and brightness was manually restored by scaling intensity values by a factor between 1.2 and 2. Labeled neurons were detected using Fiji 3D Object Counter with 2 user parameters: gray-value threshold and a minimum-size filter, which we set to 30 pixels to exclude non-somatic signals. The distributions of detected neurons across the rostro-caudal (AP) SC axis are shown in 13 bins of 100 μm from −3.08 to −4.28 mm.
To estimate the proportion of GABAergic neurons in the intermediate layer of LSC, confocal image series of double-immunolabeled slices (GAD67 and NeuN) were analyzed with Fiji's "Cell Counter" plugin. GAD67-and NeuN-labeled somata were counted in 3 segments of 200 × 200 μm per image within the intermediate layer (n = 10 slices). We excluded marginal slices when the cell number did not satisfy a minimum N sample criterion to detect the proportion difference with a statistical power of 0.8 β.

RN distribution analysis
All analyses were done using MATLAB (R2018b). For comparisons across different RNs, neuron counts were normalized to the maximum value across sections within the same brain.
Normality of the distributions was tested with one-sample Kolmogorov-Smirnov test and differences between the distributions were tested with Wilcoxon rank sum test. Mean values were reported with SEM (error bars or plus/minus) and median values with Q1 and Q3, or interquartile range (IQR). To compare distributions among different RN groups, a non-parametric analysis of variance was performed (Kruskal-Wallis and multiple comparisons test, using the Bonferroni method).

Axon maps
Fluorescent axonal projections from RNs were reconstructed in Canvas X (version 10, ACD system). Coronal images were desaturated, inverted, and manually registered to the corresponding atlas section [59]; see schematics in Figs 8A, 8B and S7. We considered axonal signals in the AP regions between −1.22 and −2.54 mm (diencephalon) and −5.4 and −6.4 mm (brainstem). The resulting axon counts were organized into 2 matrices (for diencephalon and brainstem, respectively; Table 4) per input-defined RNs. Each matrix was m×n, where m is the number of 100-μm bins in the AP dimension and n is the number of nuclei analyzed; 15 for diencephalon and 7 for brainstem.
Relative axon counts (RACs, normalized counts) were calculated for each pathway across the AP axis based on their maximum count as follows: RAC maps were visualized using MATLAB's built-in function contour ("Fill", "on") and are shown in descending order, taking MC-RNs as reference. Finally, to compare the projection patterns among different RN groups, we summed the axonal counts per nuclei across the antero-posterior (m) dimension and calculated the z-score of the total axonal count per nucleus (n) (Fig 8B).

In vivo electrophysiological recordings
Awake recordings. Mice between 8 and 12 weeks (n = 12) were recorded on a cylindrical treadmill, based on the model established in [60], consisting of a 15-cm diameter foam roller mounted on a custom built low friction rotary metal axis, attached to 2 vertical posts. Prior to whisker stimulation experiments, a head plate was implanted through stereotaxic surgery following the preparation procedures described in the section for viral injections. After exposing the skull, a small craniotomy was made above the recording site in LSC (AP: −3.6; ML: 1.25 to 1.3; DV: −1.65 to −2.1). A plastic ring was cemented (Paladur, Kulzer, GmbH) around the craniotomy to create a small ringer reservoir for the reference electrode. The well was covered with silicone elastomer (Kwik-Cast, World Precision Instruments) until the experiment. Next, a polycarbonate two-winged head plate was cemented onto the skull with dental cement (Super-bond, Sun Medical). Habituation began 5 days after the surgery and lasted for 3 days. Animals were head-fixed on the treadmill apparatus with the head plate. Habituation sessions lasted for approximately 60 min during which mice freely walked on the cylindrical treadmill and were fed sweetened condensed milk as reward. Recording sessions were done on the following day and lasted between 20 and 30 min. The protective silicone was removed and silicon probes were lowered to the SC. Activity was recorded using an electrophysiology system composed of a multielectrode silicon probe (sharpened ASSY-77 E1, Cambridge Neurotech); a 64-channel amplifier chip (RHD2164) to a USB-2 interface board (RHD-EVAL, Intan Technologies, California, US); and signals were fed into Bonsai [61] Intan RHD library. Neural activity was recorded at 30 kHz using a 0.1 Hz to 15 kHz bandpass filter. The resulting binary file was fed to Kilosort, a MATLAB-based semi-automated spike-sorting software [62]. The. bin files and the E1-probe channel map in μm were input to Kilosort with the default parameters and the results were curated with Phy. Single units with <3% refractory period (1.5 ms) violations and a baseline spike rate >0.1 Hz were included for subsequent analysis. Whisker stimulation. Whiskers were stimulated with airpuffs, delivered via an air tube placed behind the whiskers of the animal to prevent stimulation of any other body part (face, ears, or eyes). The tube was connected to an air regulator module from Modular Electronics for Cell Physiology (Max-Planck-Institute for Medical Research, manufactured by Sigmann Elektronik, GmbH, Germany) and set to 2 Bar. Airpuffs were controlled with Bonsai [61] and consisted of 100 ms airpuffs at random intervals between 1 s and 100 s.
Anesthetized recordings. Preparations and recordings were done in 8-to 12-week-old Rbp4-Cre-ChR2-EYFP mice (n = 4), which is crossbred between "Rbp4-Cre" × "Ai32" to specifically express ChR2 in L5 PT neurons [63]. Mice were anesthetized with a mixture of 5% urethane solution (IP, 1.3 g/kg body weight) and 1% isoflurane in medical degree oxygen, applied via an inhalator mask. After incision, bregma and lambda were revealed and aligned using a micro-manipulator (Luigs-and-Neumann) to drill craniotomies above SC (AP: −3.6, ML: 1.3), BC (AP: −1, ML: +3), and MC (AP: +1, ML: +1). SC activity was recorded using an electrophysiology system composed of a multielectrode silicon probe in AP: −3.6, ML: +1.3 DV: −1.6 to −2 mm (sharpened ASSY-77 E1, Cambridge Neurotech); a 64-channel amplifier chip (RHD2164) to a USB-2 interface board (RHD-EVAL, Intan Technologies, California, US); and the Spike2 (version 9.14) recording suite with Intan Talker module (Cambridge Electronic Devices, Cambridge, UK). Neural activity was recorded at 30 kHz using a 100 Hz to 10 kHz bandpass filter. The resulting file (.smrx) was converted into a binary file (.bin) to feed it to Kilosort, a MATLAB-based semi-automated spike-sorting software [62]. The file conversion consisted of reading the channels in the.smrx file, transforming them into unsigned 16-bit integers (uint16) values from the 16-bit depth analog-to-digital converter (ADC), and writing them into a.bin file. The.bin files and the E1-probe channel map in μm were input to Kilosort with the default parameters, and the results were curated with Phy. Single units with <3% refractory period (1.5 ms) violations and a baseline spike rate >0.1 Hz were included for subsequent analysis.
Optogenetic stimulation. An optical fiber was placed at an angle of approximately 45˚, at approximately 100 μm from the pia to optogenetically activate L5 neurons in MC or BC. Optogenetic stimulation pulses were controlled by a stimulation protocol prepared in Spike2 through interface hardware (Power1401, Cambridge Electronic Design, Cambridge, UK). Single 5 ms laser pulses (20 to 50 mW/mm 2 at the fiber tip) were delivered every 5 s using a custom-build laser setup containing a 488-nm solid-state laser (Sapphire, Coherent, Germany, maximum power 22 mW); an ultrafast shutter (Uniblitz, Rochester, NY, USA); and a collimator to focus the laser beam into the optical fiber (inner diameter 400 μm, NA 0.48, Thorlabs, USA). After stimulating L5 in one cortex, the optical fiber was moved to the other, for example, MC and then BC.
Cell responsiveness. Whisker responses: We considered a single unit to be whisker responsive if the median spike count across trials within a 30-ms window (−20 to −50 ms) prestimulus was significantly different from the median spike count within an identical-length window (20 to 50 ms, α = 5%, two-sample Wilcoxon test). The modulation index (MI) was computed using Cell Explorer [27] as follows: MI ¼ ðR þ CÞ=ðR À CÞ, where R and C are firing rates in the response and spontaneous windows, respectively [64]. A given unit could have MI values ranging from −1, when R = 0 & C 6 ¼ 0; to 0, when R = C; and to 1, when Optogenetic responses: We considered 2 equal-length time windows in which we counted spikes before the onset of the optogenetic stimulus (spontaneous, −18 to −6 ms) and after the optogenetic stimulus onset (evoked, 6 to 18 ms) for each trial and for each unit. We computed the z-scores of evoked counts using the mean and standard deviation of spontaneous counts and compared evoked z-scores against a 1.96 z-threshold (α = 5%). Trials that crossed the zthreshold were considered significantly responsive and were assigned a logical true value. We then computed the proportion of responsive (successful) trials per unit. In order to state significance for the response proportion, we estimated the necessary trial number (N) to achieve a statistical power of 0.8 with proportion values p and hypothesis p 0 = 1-p using MATLAB's test of the "N" parameter (trials) for a binomial distribution (sampsizepwr). For p ≂ 1/3, N ≂ 40, which is our experimental trial number. Additionally, we compared the unit's trial crossing proportions with increasing thresholds (α = 5%, one-tailed t test, ttest in MATLAB) and visually searched for a region where the cardinality of included units stabilized. With an increasing threshold, the number of included units decreases until the responsive units would be found. We chose p 0 = 1/3 as an optimal inflection point, minimizing the threshold and maximizing the responsive unit cardinality.