A Differential Circuit via Retino-Colliculo-Pulvinar Pathway Enhances Feature Selectivity in Visual Cortex through Surround Suppression

Summary

In the mammalian visual system, information from retina streams into parallel bottom-up pathways. It remains unclear how these pathways interact to contribute to contextual modulation of visual cortical processing. By optogenetic inactivation and activation of mouse lateral posterior nucleus of thalamus (LP), a homologue of pulvinar, or its projection to primary visual cortex (V1), we found that LP contributes to surround suppression of layer (L)2/3 responses in V1 by driving L1 inhibitory neurons. This results in subtractive suppression of visual responses and an overall enhancement of orientation, direction, spatial and size selectivity. Neurons in V1-projecting LP regions receive bottom-up input from superior colliculus (SC) and respond preferably to non-patterned visual noise. The noise-dependent LP activity allows V1 to “cancel” noise effects and maintain its orientation selectivity under varying noise background. Thus, the retina-SC-LP-V1 pathway forms a differential circuit with the canonical retino-geniculate pathway, to achieve context-dependent sharpening of visual representations.

eTOC Blurb

Fang et al. show that the lateral posterior nucleus of thalamus, the rodent homologue of pulvinar, enhances feature selectivity in primary visual cortex via subtractive surround suppression mediated by inhibition from cortical layer 1 driven by the bottom-up retina-SC-LP-V1 pathway.

Introduction

In the mammalian visual system, sensory information from the retina undergoes a hierarchy of processing by streaming into two major bottom-up pathways (Nassi and Callaway, 2009). One is through the dorsal lateral geniculate nucleus (dLGN), the first-order thalamic nucleus, to reach the primary visual cortex (V1) and subsequent higher visual cortical areas to generate visual perception. The other is through the superior colliculus (SC) in the midbrain to generate certain visually guided behaviors (Basso and May, 2017; Shang et al., 2015; Wei et al., 2015; Yilmaz and Meister, 2013). While visual cortical feedback modulates SC activity via corticocollicular projections (Liang et al., 2015; Zhao et al., 2014) and thus influences visually induced behaviors such as looming induced freezing (Yilmaz and Meister, 2013; Zingg et al., 2017), SC activity in turn may have impacts on visual cortical processing (Ahmadlou et al., 2018; Beltramo and Scanziani, 2019; Ogino and Ohtsuka, 2000; Tohmi et al., 2014). The lateral posterior nucleus (LP) of thalamus, which is generally considered as the rodent homologue of the primate pulvinar (Zhou et al., 2017), appears well-poised to bridge SC and visual cortex and to play a role in mediating the SC impacts on visual cortex. This possibility is strongly suggested by the anatomical evidence that among thalamic nuclei, LP receives the strongest projection from SC (Gale and Murphy, 2014; Redgrave et al., 1993; Stepniewska et al., 2000; Wei et al., 2015; Zingg et al., 2017), and that it projects to both primary and secondary visual cortices (Beltramo and Scanziani, 2019; Bennett et al., 2019; Juavinett et al., 2019; Kaas and Lyon, 2007; Nakamura et al., 2015; Petry and Bickford, 2018; Roth et al., 2016; Shipp, 2001; Wong et al., 2009; Zhou et al., 2018).

Previous studies of pulvinar have mostly been focused on higher visual functions such as visual attention, perceptual suppression as well as planning and selection of visually guided eye and hand movements (Dominguez-Vargas et al., 2017; Grieve et al., 2000; Saalmann et al., 2012; Soares et al., 2017; Wilke et al., 2009, 2010; Zhou et al., 2016). Recently, there has been evidence that LP/pulvinar transmits moving visual information as well as motor-related signals to secondary or higher visual cortices, contributing to the visual response properties there (Beltramo and Scanziani, 2019; Bennett et al., 2019; Berman and Wurtz, 2011; Tohmi et al., 2014). However, the role of LP/pulvinar in fundamental information processing in V1 has remained largely unclear. Different from the dLGN which relays ascending visual information mainly to layer (L)4 of V1, the excitatory projections of LP to V1 primarily terminate in L1 and deep layers, while its projections to secondary visual cortices (V2) mainly terminate in L4 (Herkenham, 1980; Roth et al., 2016; Zhou et al., 2017). Cortical L1 is predominantly populated by inhibitory neurons (Jiang et al., 2013) and has been implicated in the modulation of V1 responses in superficial layers by long-range projections from other cortical areas (Jiang et al., 2013; Zhang et al., 2014; Zhou et al., 2014; Ibrahim et al., 2016). This innervation pattern of LP axons in V1 raises a hypothesis that LP might play a modulatory role in influencing V1 responses, rather than an excitatory “driver” role as suggested in a previous study of pulvinar in galagos (Purushothaman et al., 2012). Consistent with this idea, functional studies of LP neurons and their axon terminals in L1 of V1 have shown that these neurons exhibit large receptive fields (RFs) (Allen et al., 2016; Bender, 1982; Berman and Wurtz, 2011; Chalupa et al., 1983; Durand et al., 2016; Roth et al., 2016), suggesting that LP may be able to provide contextual information to modulate V1 responses (Roth et al., 2016).

In this study, by applying pharmacological and optogenetic manipulations in awake mice, we explored the functional contribution of LP to V1 processing and the underlying circuit basis. We demonstrate that LP activity improves fundamental visual processing functions through a feedforward, surround-suppression mechanism mediated by L1 inhibitory neurons. Our study further suggests that the retina-SC-LP-V1 pathway together with the parallel retino-geniculo-V1 pathway forms a bottom-up “differential” circuit by which V1 selectivity can be largely maintained despite varying visual noise background.

Results

LP silencing impairs visual discrimination task performance

To study potential functional contributions of LP, we first employed a visual discrimination task. We trained mice to lick water for reward upon detection of a Go signal (a visual grating pattern of vertical orientation) but refrain from licking upon detection of a No-Go signal (grating of horizontal orientation) (Figure 1A). The animals could learn this task over one-week training, as manifested by the increasing correct performance rate (hit + correct rejection) over training sessions before reaching a plateau level (Figure 1B, top). In the well-trained animals, silencing LP bilaterally either by local infusion of muscimol or by expressing inhibitory Designer Receptors Exclusively Activated by Designer Drugs (DREADDi) and administering the DREADD ligand, Clozapine-N-oxide (CNO) (see STAR Methods), significantly lowered the correct performance rate, whereas injections of CNO in GFP control mice had no effect (Figure 1B). The poorer performance resulted from an increase in the false alarm rate, while the miss rate was not significantly changed (Figure S1A). While the overall impairment of performance could be attributed to deficits in visual attention after silencing LP (Zhou et al., 2016), the increased false alarm rate together with the unchanged miss rate suggests that the animal’s ability to distinguish visual patterns was compromised after silencing LP while the animal was still engaged in the task. Thus, it is likely that LP activity directly contributes to normal information processing in visual cortex.

Figure 1. Change of V1 response properties by pharmacologically silencing LP.

(A) Paradigm of visual discrimination task.

(B) Top, correct rate (percentage of hit and correct rejection trials) over training days for two example animals. Arrows indicate bilateral infusion of muscimol or CNO on day 8, and of control saline on day 7. Bottom, summary of correct rates. Data points for the same animal are connected by a line. ***p < 0.001; N.S., not significant, paired t-test, n = 4 for muscimol silencing, 4 for DREADDi, and 5 for GFP control.

(C) Left, awake recording with the animal implanted with a cannula for drug infusion (top) and an example fluorescence image showing the spread of drug infusion (bottom). Scale bar: 500 μm. Right, visually evoked (top) and spontaneous (bottom) firing rates of recorded LP neurons before and 30 min after drug infusion. ***p < 0.001, paired t-test, n = 12 cells from 2 mice. Bar = SD.

(D) Polar plots of evoked firing rates at different directions of moving gratings (full-field) for two example L2/3 neurons (top inset, axis limit: 10 Hz) and average normalized orientation tuning curves (with Gaussian fit) of the recorded L2/3 population before (black) and after (red) silencing LP with bupivacaine. Bar = SEM. n = 28 cells from 5 mice (same cells for E-H)

(E) The gOSI values of recorded L2/3 neurons after vs before silencing LP. Dashed line is the unity line. Solid symbols label cells showing significant changes (permutation test with bootstrapping, p < 0.05). p < 0 .001, paired t-test (if not otherwise indicated).

(F) The gDSI values after vs before silencing LP (p < 0.001).

(G) Preferred orientation (p = 0.37).

(H) Change of evoked firing rate for preferred and orthogonal orientations (p = 0.40). Bar = SD.

(I) Response levels to flash white-noise patterns after vs before LP silencing (p < 0.001, n = 27 cells from 6 mice).

(J) Firing rates (averaged by trials) to different stimulus sizes (in radius) before (black) and after (red) silencing LP for an example L2/3 neuron. Bar = SD.

(K) Surround suppression index (SSI) for recorded L2/3 neurons after vs before silencing LP (p = 0.002, n = 18 cells from 4 mice). Inset, population average of normalized size tuning curves aligned by the peak response before silencing LP. ***p < 0.001, **p < 0.01, *p < 0.05, paired t-test. Bar = SEM.

(L) RF subfield of dominant contrast for an example L2/3 neuron before and after silencing LP. Left, raw RF subfield. Right, subfield fitted with 2D elliptical Gaussian (oval, white for before and red for after). Scale bar: 8°.

(M) RF subfield sizes (in radius) of recorded L2/3 neurons after vs before silencing LP (p < 0.001, n = 22 cells from 3 mice).

(N-W) For L4 neurons. Data are displayed in same manners as in (D-M). Axis limit in (N) (top inset): 20 Hz. Statistics: (O-R), p = 0.61, 0.68, 0.85, and 0.25 respectively, n = 19 cells from 5 mice; (S), p = 0.87, n = 18 cells from 6 mice; (U), p = 0.65, Wilcoxon signed-rank test, n = 17 cells from 4 mice; (W), p = 0.51, n = 18 cells from 3 mice.

LP silencing reduces functional selectivity in L2/3 of V1

To understand how LP influences visual information processing, we carried out in vivo single-cell loose-patch recordings and examined visual responses of the same V1 neurons before and after manipulating LP activity in awake head-fixed mice (Figure 1C, top left), following our previous studies (Chou et al., 2018; Xiong et al., 2015). We first silenced LP by infusing 0.5% bupivacaine (Lee et al., 2008) via an implanted cannula (Figure 1C, bottom left; Figure S1B), which effectively eliminated both evoked and spontaneous spikes of LP neurons for at least 30 min (Figure 1C, right) (see STAR Methods).

For each recorded neuron, we applied full-field moving gratings of 12 directions (0° to 330°, 30° step) to measure orientation selectivity and direction selectivity. As shown by the polar graphs for two example L2/3 neurons and the average normalized orientation tuning curves (Figure 1D), after silencing LP orientation selectivity was weakened due to increases of evoked response at all orientations. This change was evident in most of L2/3 neurons we recorded, as shown by a reduced global orientation selectivity index (gOSI) (Figure 1E; Figure S1F) and conventional orientation selectivity index (OSI) (Neill and Stryker, 2008; Tan et al., 2011) (Figure S1F). Direction selectivity was also weakened, as shown by a lowered global direction selectivity index (gDSI) (Figure 1F; Figure S1G) and conventional direction selectivity index (DSI) (Figure S1G). Notably, orientation preference was not altered after LP silencing (Figure 1G). In the meantime, changes in evoked firing rate after LP silencing were the same for preferred and orthogonal orientations (Figure 1H), suggesting that the change of orientation tuning can be described as a subtractive effect.

We also measured the overall visual response level by presenting a series of full-screen, flashing dense white-noise patterns. The visual response level was significantly increased after silencing LP (Figure 1I; Figure S1E; see control in Figure S1C), consistent with the grating results. The spontaneous firing rate was increased from 2.07 ± 1.18 Hz to 3.03 ± 2.15 Hz (mean ± SD, n = 27 cells from 6 mice, p = 0.029, paired t-test) after silencing LP. Therefore, our results suggest that the overall effect of LP activity on V1 L2/3 responses is suppressive, resulting in an enhancement of orientation and direction selectivity (Figure S1F-G), i.e. improving visual processing.

In another cohort of animals, we examined size tuning of V1 neurons (see STAR Methods). As shown by an example L2/3 neuron (Figure 1J), before silencing LP the cell had an optimal response to 10° size and its responses to larger stimuli were suppressed. In other words, the cell exhibited size selectivity and surround suppression (Adesnik et al., 2012; Nienborg et al., 2013). After silencing LP, the responses were increased, especially for larger stimulus sizes (Figure 1J), suggesting weakened surround suppression. We quantified the strength of surround suppression with a surround suppression index (SSI) (Adesnik et al., 2012), which ranges from 0 (no suppression) to 1 (complete suppression by the largest stimulus). The SSIs of recorded L2/3 neurons were significantly reduced after silencing LP (Figure 1K; Figure S1H). In particular, in several cells tested, size tuning was completely lost, i.e. SSI became zero (Figure 1K). As shown by the population average of size tuning curves from all the cells, the increase of evoked firing rate after silencing LP was relatively higher for larger than smaller stimulus sizes (Figure 1K, inset). At the population level, the distribution of optimal sizes was consistent with that reported previously (Adesnik et al., 2012; Nienborg et al., 2013; Vaiceliunaite et al., 2013) (Figure S1D). We also directly measured spatial receptive fields (RFs) using a reverse correlation method. The size (in radius) of the RF subfield of dominant contrast (On or Off) increased after LP silencing (Figure 1L–M; Figure S1I). These results indicate that LP activity contributes to surround suppression and increases the size and spatial selectivity of V1 L2/3 neurons.

Interestingly, all the above effects observed in L2/3 were absent in L4. Basically, there was no change in visual selectivity, response level or RF subfield size in L4 neurons after silencing LP (Figure 1N–W; Figure S1J–N). This absence of effect on L4 neurons argues against the possibility that the drug infusion had affected the dLGN, which would result in a reduction of response level in L4 (Reinhold et al., 2015). Rather, it suggests that the influence of LP on V1 response properties was specific to superficial layers.

Optogenetic silencing of LP reduces visual selectivity in L2/3

To have a better spatial and temporal control of LP activity, we used an adeno-associated viral (AAV) vector to express an inhibitory opsin, archaerhodopsin (ArchT), in LP (Figure 2A–B; Figure S2A). Green LED light (530 nm) was applied to silence LP neurons (see STAR Methods), which covered the entire duration of visual stimulation. Responses of the same neurons were compared between interleaved LED-on and LED-off (control) trials. First, we found that visual responses in LP were nearly abolished by the optogenetic stimulation, whereas those in dLGN were not affected (Figure 2C), verifying the effectiveness and spatial specificity of LP silencing. Next, we recorded from V1 neurons. Essentially, the optogenetic silencing of LP reproduced the results of pharmacological silencing with bupivacaine: orientation and direction selectivity was weakened without changing preference (Figure 2D–H; Figure S2D–E), the overall visual response level was increased (Figure 2I; Figure S2C; see control in Figure S2B), and size tuning/surround suppression was weakened (Figure 2J–K; Figure S2F). These effects were observed only in L2/3, but not in L4 (Figure 2L–S; Figure S2G–J). In addition, spontaneous firing rate was increased from 2.00 ± 1.10 Hz in the LED-off condition to 3.56 ± 2.51 Hz in the LED-on condition (n = 26 cells from 4 mice, p = 0.0097, paired t-test). The optogenetic silencing experiments further demonstrate that LP activity improves V1 processing in L2/3.

Figure 2. Change of V1 response properties by optogenetic silencing of LP.

(A) Awake recording setup.

(B) Left, recording in V1 while optogenetically silencing LP. Right, expression of ArchT in LP. Scale bar: 500 μm.

(C) Visual responses in LP (***p < 0.001, paired t-test, n = 13 cells from 2 mice) and dLGN (p = 0.36, n = 13 cells from 2 mice) and without (OFF) and with (ON) LED illumination.

(D) Polar plots of orientation tuning for two example L2/3 neurons without (black) and with (green) optical inactivation of LP (top, axis limit = 20 Hz) and average tuning curves for the recorded L2/3 population. n = 26 cells from 4 mice (same cells for D-H).

(E) The gOSI values of recorded L2/3 neurons with versus without LED stimulation. Solid symbols mark neurons with significant changes (permutation test with bootstrapping, p < 0.05). p < 0.001, paired t-test (if not otherwise indicated).

(F) The gDSI values (p < 0.001).

(G) Preferred orientation (p = 0.16).

(H) Change in evoked firing rate for preferred and orthogonal orientations (p = 0.27). Bar = SD.

(I) Response levels to flash noise patterns (p < 0.001, n = 30 cells from 6 mice).

(J) SSI values (p < 0.001, n = 27 cells from 6 mice).

(K) Population average of normalized size tuning curves of L2/3 neurons without (dark gray) and with (green) LED stimulation. ***p < 0.001, **p < 0.01, *p < 0.05, paired t-test, n = 27 cells from 6 mice. Bar = SEM.

(L-S) For L4 neurons. Statistics: (M-P), p = 0.11, 0.85, 0.49, 0.084, respectively, n = 19 cells from 4 mice; (Q), p = 0.54, Wilcoxon signed-rank test, n = 18 cells from 6 mice; ®, p = 0.25, Wilcoxon signed-rank test, n = 18 cells from 6 mice.

Activation of LP improves visual processing in L2/3

To further understand the direct effect of LP activity on V1 responses, we injected AAV encoding channelrhodopsin2 (ChR2) into LP, and applied blue (473 nm) LED light pulses to activate LP (Figure 3A; Figure S3A). The optogenetic stimulation effectively evoked spiking responses of LP neurons while left the dLGN unaffected (Figure 3B). In V1 L2/3, effects opposite to inactivating LP were observed: orientation and direction selectivity was strengthened (Figure 3C–G; Figure S3D–E), the visual response level was reduced (Figure 3H; Figure S3C; control in Figure S3B), and size tuning/surround suppression was enhanced (Figure 3I–J; Figure S3F). In addition, spontaneous firing rate was reduced from 2.58 ± 1.71 Hz in the LED-off condition to 1.42 ± 0.68 Hz in the LED-on condition (n = 26 cells from 6 mice, p = 0.0043, paired t-test). Furthermore, similar to inactivating LP, effects on V1 neurons were only observed in L2/3 but not L4 (Figure 3K–R]; Figure S3G–J). The optogenetic stimulation did not result in significant eye movements (Figure S3K). These activation experiments further support the conclusion that LP activity exerts a suppressive effect on responses of V1 L2/3 neurons, resulting in enhanced surround suppression and improvements of tuning selectivity of these neurons. The overall suppressive effect of LP could be described as subtractive rather than divisive, as the orientation tuning curve of L2/3 neurons appeared to be shifted down when activating LP (Figure 3C,G) while shifted up when inactivating LP (Figure 2D,H).

Figure 3. Change of V1 response properties by optogenetic activation of LP.

(A) Left, recording in V1 while optogenetically activating LP. Right, expression of ChR2 in LP. Scale bar: 400 μm.

(B) Firing rates of neurons in LP (***p < 0.001, paired t-test, n = 12 cells from 2 mice) and dLGN (p = 0.18, n = 16 cells from 2 mice) without (OFF) and with (ON) LED stimulation.

(C) Polar plots of orientation tuning for two example L2/3 neurons without (black) and with (blue) LED stimulation (top inset, axis limit is 15 Hz) and average tuning curves for the recorded L2/3 population (n = 26 cells from 6 mice).

(D) The gOSI values of recorded L2/3 neurons with versus without LED stimulation. Solid symbols mark neurons with significant changes (permutation test with bootstrapping, p < 0.05). p < 0.001, paired t-test, n = 26 cells from 6 mice (same cells for E-G).

(E) The gDSI values (p < 0.001).

(F) Preferred orientation (p = 0.21).

(G) Change in evoked firing rate for preferred and orthogonal orientations (p = 0.14). Bar = SD.

(H) Response levels (p = 0.014, n = 20 cells from 4 mice).

(I) SSI values (p = 0.0032, n = 34 cells from 8 mice).

(J) Population average of normalized size tuning curves of L2/3 neurons without (dark gray) and with (blue) LED stimulation. ***p < 0.001, **p < 0.01, *p < 0.05, paired t-test, n = 34 cells from 8 mice. Bar = SEM.

(K-R) For L4 neurons. Axis limit in (K): 15 Hz. Statistics: (L-O), p = 0.95, 0.91, 0.64, 0.89, respectively, n = 16 cells from 6 mice; (P), p = 0.84, n = 18 cells from 4 mice; (Q), p = 0.25, n = 18 cells from 8 mice.

LP axons innervate L1 inhibitory neurons in V1

We next examined LP projections to V1, by injecting a retrograde dye CTB in V1 (Figure 4A). Labeled neurons were distributed nearly across the entire LP in the anterior-posterior axis, although more neurons in the rostral than caudal part of LP projected to V1 (Figure 4B). This rostral-caudal gradient is consistent with recent anatomical results (Beltramo and Scanziani, 2019; Bennett et al., 2019). LP contains predominantly excitatory neurons (Evangelio et al., 2018). Since these LP projections to visual cortex are excitatory (Roth et al., 2016; Zhou et al., 2018), how do they produce suppressive effects on V1 L2/3 neurons? By injecting AAV-GFP in LP, we observed that in V1 the anterogradely labeled LP axons were densely distributed in L1 besides some distributions in deep layers (Figure 4C), consistent with previous results (Bennett et al., 2019; Roth et al., 2016; Zhou et al., 2018). This suggests that LP axons may directly innervate L1 neurons, which are inhibitory (Jiang et al., 2013). In slice preparations, we made whole-cell recordings from V1 pyramidal (PYR) or L1 GABAergic neurons while optically stimulating ChR2-expressing LP axons (Figure 4D). With TTX and 4-AP present in the bath solution, blue light evoked monosynaptic excitatory postsynaptic currents (EPSCs) in L1 GABAergic neurons, which could be blocked by CNQX (Figure 4E). Such EPSCs with a relatively large amplitude were observed in a majority of recorded L1 inhibitory neurons (Figure 4F, red), indicating that indeed LP axons can directly innervate L1 inhibitory neurons. In comparison, most pyramidal neurons in L2/3 or L4 were not innervated by LP axons, and those innervated by LP axons only exhibited weak EPSCs (Figure 4F, black). In addition, we found that parvalbumin (PV) and somatostatin (SOM) inhibitory neurons in L2/3 and L4 were essentially not innervated by LP axons (Figure 4G).

Figure 4. LP axons innervate L1 inhibitory neurons in V1.

(A) Retrograde tracing with CTB.

(B) Distribution of CTB-labeled neurons at different coronal sectional levels. Scale bar: 200 μm.

(C) Expression of AAV-encoded GFP in LP (left) and GFP-labeled LP axons in V1 (right). Scale bar: 500 (left) / 200 (right) μm.

(D) Slice recording. Red dots represent tdTomato-labeled inhibitory neurons.

(E) Average light-evoked EPSC traces recorded in an example V1 L1 neuron before (top) and after (bottom) applying CNQX. Blue arrow marks the onset of blue light pulse.

(F) Peak amplitudes of light-evoked monosynaptic EPSCs vs cell depths. Red and black dots represent L1 GABAergic and pyramidal neurons, respectively.

(G) Probability of innervation of V1 neurons of different types in different layers by LP axons. n = 5 GAD2-Cre, 4 PV-Cre and 4 SOM-Cre mice.

(H) Left, superimposed traces of hyperpolarizing membrane potential responses in an example L2/3 pyramidal neuron to activation of LP-V1 axons. Right, average hyperpolarizing voltage (mean ± SD) (n = 7 L2/3 pyramidal cells from 3 mice).

(I) Left, traces of membrane potential responses to a square current injection without (upper) and with (lower) stimulation of LP-V1 axons (10 pulses) in a L2/3 pyramidal neuron. Different trials are labeled by different colors. Right, average number of action potentials per trial without and with optical stimulation. ***p < 0.001, paired t-test, n = 6 cells from 3 mice.

In the normal bath solution without TTX and 4-AP, blue light activation of LP axons in V1 induced a hyperpolarizing membrane potential response in L2/3 pyramidal neurons, which reduced the number of action potentials induced by current injections (Figure 4H–I). These results suggest that activation of LP-V1 axons can generate a net suppressive effect on L2/3 pyramidal cells through driving L1 inhibitory neurons.

LP-V1 projection can modulate V1 processing

We further examined whether the effect of LP activity on V1 observed in vivo could be mediated by the LP to V1 projection. To address this issue, we injected AAV encoding Cre-dependent halorhodopsin (NpHR3.0) in LP of Vglut2-Cre mice (Figure 5A), and applied yellow (589 nm) LED light to the surface of V1 to suppress presynaptic release at LP axon terminals (Mahn et al., 2016). This light stimulation reduced visual responses of L1 neurons, as shown by the loose-patch recording from L1 neurons (Figure 5B–C), demonstrating the effectiveness of the terminal silencing. Consistently, we observed effects similar to silencing LP neuron cell bodies: the visual response level was elevated and orientation/direction/size selectivity was weakened in L2/3 (Figure 5D–H). Spontaneous firing rate was increased from 2.03 ± 0.96 Hz in the LED-off condition to 2.83 ± 2.09 Hz in the LED-on condition, but the increase did not reach a significant level (n = 10 cells from 4 mice, p = 0.27, paired t-test). No effects were observed in L4 (Figure 5I–L).

Figure 5. LP to V1 projection primarily accounts for the LP modulation of V1 responses.

(A) Top, activation of eNpHR3.0 with yellow LED light applied to the surface of V1. Bottom, expression of eNpHR3.0-EYFP in LP. Scale bar: 400 μm.

(B) Peristimulus spike time histograms (PSTHs) of an example V1 L1 neuron to flash noise patterns without (left) and with (right) inactivation of LP-V1 axons. Shaded area marks the duration of LED stimulation.

(C) Visually evoked firing rates of L1 neurons without (OFF) and with (ON) the optical stimulation. **p = 0.0039, Wilcoxon signed rank test, n = 9 cells from 2 mice.

(D) Response levels of L2/3 neurons. ***p < 0.001, paired t-test, n = 10 cells from 4 mice (same cells for E-G).

(E) Polar plots of orientation tuning for two example L2/3 neurons (top inset, axis limit = 10Hz) and average tuning curves of the recorded L2/3 population without (black) and with (orange) the optical stimulation.

(F) The gOSI values (***p < 0.001, paired t-test).

(G) The gDSI values (**p = 0.0072).

(H) The SSI values (***p = 0.0010, n = 9 cells from 4 mice).

(I-L) For L4 neurons. Axis limit in (J): 20Hz. Statistics: p = 0.79 (I), 0.90 (K) and 0.96 (L), n = 10 cells for 4 mice.

(M) Activation of LP-V1 axons.

(N) PSTH of an example V1 L1 neuron. Blue bar marks the duration of LED stimulation.

(O) Firing rates of V1 L1 neurons without and with LED stimulation (***p < 0.001, n = 8 cells from 3 mice).

(P) Visually evoked firing rates of L2/3 neurons to gratings. ***p < 0.001, n = 12 cells from 5 mice (same cells for Q-S).

(Q) Polar plots of orientation tuning for two example L2/3 neurons and average tuning curves without (black) and with (blue) optical activation of LP-V1 axons. Axis limit: 20Hz.

(R) The gOSI values (***p < 0.001, Wilcoxon signed rank test).

(S) The gDSI values (**p = 0.0067).

(T) The SSI values (**p = 0.0072, n =8 cells from 5 mice).

(U-X) For L4 neurons. Axis limit in (V): 10Hz. Statistics: p = 0.70 (U), 0.18 (W) and 0.63 (X), n = 10 cells from 5 mice. Bar = SD, except for (E), (J), (Q), (V).

Conversely, we activated the LP-V1 projection by expressing ChR2 in LP and shining blue LED light on the surface of V1 (Figure 5M). In line with the slice recording data showing monosynaptic innervation of V1 L1 neurons by LP axons, LED stimulation effectively evoked spiking responses of V1 L1 neurons in vivo (Figure 5N–O). Opposite to the effects of suppressing LP-V1 axon terminals, the visual response level was reduced, and orientation/direction/size selectivity was enhanced in L2/3 (Figure 5P–T). Spontaneous firing rate was reduced from 2.44 ± 1.53 Hz in the LED-off condition to 1.44 ± 0.64 Hz in the LED-on condition (n = 12 cells from 5 mice, p = 0.040, paired t-test). Again, no effects were observed in L4 (Figure 5U–X).

As a control, we recorded from the posteromedial (PM) region, which is the most proximate to the monocular zone of V1 among V2 areas (Wang et al., 2011). The LED stimulation of LP-V1 axons did not affect either the spontaneous or evoked activity of superficial-layer PM neurons (Figure S4). This suggests that the effects of the LED stimulation observed in V1 L2/3 are unlikely attributed to feedback input from V2 to V1. Together, our results indicate that the LP to V1 projection can primarily account for the LP modulatory effect on V1 L2/3 neurons.

The sSC drives LP to modulate V1 feature selectivity

LP receives strong input from the superficial layer of SC (sSC) (Baldwin et al., 2011, 2013; Masterson et al., 2009). Using an anterograde transsynaptic labeling approach we recently developed (Zingg et al., 2017), we further demonstrated that the retinorecipient SC neurons, which were located in sSC (Figure 6A), projected to LP across the rostral-caudal axis (Figure 6B). The projection to the caudal part of LP was apparently stronger than the rostral part, reminiscent of recent results (Beltramo and Scanziani, 2019; Bennett et al., 2019). In LP regions both receiving prominent SC input and projecting to V1 (between bregma −2.0 mm and −2.5 mm, see Figure 4B, 6B), response onsets to flash stimuli were slower than SC but faster than V1 (Figure S5A), consistent with a bottom-up SC-LP-V1 hierarchy. When sSC was silenced with muscimol (Figure S5B), visual responses of LP neurons were reduced by about 75% whereas those of dLGN neurons were not significantly changed (Figure S5C, left). In comparison, LP responses were reduced by only about 15% after silencing V1 (Figure S5C, right). Thus, neurons in V1-projecting LP regions can be strongly driven by bottom-up visual input from SC.

Figure 6. SC input drives LP to modulate V1 responses.

(A) Left, transsynaptic labeling of retinorecipient SC neurons. Right, GFP-labeled retinorecipient SC neurons. Scale bar: 500 μm.

(B) Projections of retinorecipient SC neurons at different coronal sections. Scale bar: 500 μm.

(C) Polar plots of orientation tuning for two example L2/3 neurons (axis limit = 20Hz) and average tuning curve of the recorded L2/3 population before (black) and after (red) silencing sSC with bupivacaine. n = 24 cells from 4 mice (same cells for D-G).

(D) The gOSI values (p = 0.015, paired t-test).

(E) The gDSI values (p = 0.0036).

(F) Preferred orientation (p = 0.86).

(G) Change in evoked firing rate for preferred and orthogonal orientations (p = 0.88).

(H) Visual response levels (p < 0.001, n = 26 cells from 4 mice).

(I) The SSI values (**p = 0.0023, n = 20 cells from 6 mice).

(J) Population average of size tuning before (dark gray) and after (red) silencing sSC. ***p < 0.001, **p < 0.01, *p < 0.05, paired t-test, n = 20 cells from 6 mice.

(K) RF subfield of an example L2/3 neuron before and after silencing sSC. Scale bar: 8°.

(L) RF subfield sizes after versus before silencing sSC (**p = 0.0038, Wilcoxon signed-rank test, n = 19 cells from 4 mice).

(M-V) For L4 neurons. Axis limit in (M): 20Hz. Statistics: (N-Q), p = 0.21, 0.30, 0.68, 0.45, respectively, n = 16 cells from 4 mice; (R), p = 0.62, Wilcoxon signed-rank test, n = 18 cells from 4 mice; (S), p = 0.86, n = 18 cells from 6 mice; (V), p = 0.74, n = 21 cells from 4 mice. Bar = SD except in (C), (J), (M), (T).

To further test whether silencing sSC could lead to effects similar to silencing LP, we recorded single-cell responses in V1 before and after silencing sSC with bupivacaine (Figure S6A). Indeed, we observed effects similar to silencing LP: orientation and direction selectivity was weakened without changing preference (Figure 6C–G; Figure S6C–D), the visual response level was enhanced (Figure 6H; Figure S6B), size-tuning/surround suppression was reduced (Figure 6I–J; Figure S6E), and the dominant RF subfield was enlarged (Figure 6K–L; Figure S6F). Similarly, these effects were observed only in L2/3, but not in L4 (Figure 6M–V; Figure S6G–K). Together, our results indicate that the LP modulation of V1 L2/3 responses can be driven by the bottom-up visual input from sSC.

LP helps to maintain V1 orientation selectivity in visually noisy background

The above experiments have shown that LP enhances selectivity in V1. We wondered under what circumstances such LP modulation might be particularly useful. Previously, it has been shown that visually evoked responses in V1 are reduced when stimuli are presented with a noisy background (Macknik and Livingstone, 1998). Here, we noticed that LP neurons responded more strongly to noise patterns than gratings (Figure 7A–B). In addition, they exhibited much weaker orientation and direction selectivity than V1 neurons (Figure 7C–D), consistent with previous reports (Durand et al., 2016; Roth et al., 2016). These observations prompted us to speculate that LP might be more engaged when visual background is noisy. To test this idea, animals were presented with full-screen gratings of different orientations embedded in ever-changing white-noise background of different contrasts (Figure 7E, insets). We recorded spike responses of LP neurons as well as of V1 L2/3 neurons before and after silencing LP. In the control condition, we observed that LP neuron responses to gratings increased (Figure 7E,G), whereas V1 neuron responses decreased (Figure 7F,H), with increasing noise contrasts. Within a range of low to moderate noise levels, V1 orientation selectivity was largely unchanged, but with the noise level further increased it was then gradually reduced (Figure 7I, black). After LP was silenced, orientation selectivity of L2/3 neurons reduced at all noise levels (Figure 7I, red), accompanied by increases of firing rate to both preferred and orthogonal orientations (Figure S7). In particular, selectivity was most severely compromised at a mid-level noise contrast, so that increasing noise level had an accelerated damaging effect on selectivity. Thus, LP plays a role in maintaining V1 orientation selectivity in the face of varying visual background noise.

Figure 7. Maintaining V1 orientation selectivity by LP in background noise.

(A) Recording in LP.

(B) Firing rates to moving gratings of optimal orientation/direction vs. to flash noise patterns (full screen). p < 0.001, Wilcoxon signed-rank test, n = 16 cells from 2 mice.

(C) Polar plot of orientation tuning for an example LP neuron and a V1 L2/3 neuron (top inset, axis limit = 8 Hz and 15 Hz, respectively) and average normalized tuning curves for recorded LP (light grey, n = 13 cells from 2 mice) and V1 L2/3 (dark, n = 16 cells from 2 mice) neurons.

(D) Comparisons of gOSI and gDSI between LP and V1 neurons. ***p < 0.001, Mann-Whitney rank sum test.

(E) Top inset, sample stationary grating patterns (full screen) embedded in a noise background at different contrasts. Bottom, PSTHs of responses of a LP neuron.

(F) PSTHs of responses of a V1 L2/3 neuron to the same stimuli (0–200 ms).

(G) Firing rates of LP neurons to gratings with increasing noise levels. n = 13 cells from 4 mice. Bar = SEM.

(H) Firing rates of V1 L2/3 neurons. n = 15 cells from 3 mice.

(I) The gOSI values of V1 L2/3 neurons at different noise contrasts before (black) and after (red) silencing LP with muscimol. Star indicates a significant difference in gOSI as compared with the one-level lower noise contrast in the same condition. ***p < 0.001, **p < 0.01, paired t-test. The pound sign indicates a significant difference between conditions at the same noise contrast. ^###p < 0.001, ^##p < 0.01, ^#p < 0.05, paired t-test, n = 13 cells from 4 mice. Bar = SD.

Discussion

In this study, we have discovered a functional contribution of LP to V1 processing: LP provides a net suppressive signal to V1 superficial layers via the feedforward activation of V1 L1 inhibitory neurons, which reduces the visual response level of L2/3 pyramidal neurons and significantly enhances their feature selectivity including orientation, direction, spatial and size tuning selectivity. This suppressive effect contributes to surround suppression in V1 and can be described as “subtractive”, analogous to a thresholding effect (Liang et al., 2018; Priebe and Ferster, 2008). Such modulation of orientation/direction and size tuning is consistent with the functional properties of LP neurons: they are coarsely retinotopic, exhibiting large receptive fields, and mostly not orientation/direction selective (Allen et al., 2016; Bender, 1982; Berman and Wurtz, 2011; Chalupa et al., 1983; Durand et al., 2016; Roth et al., 2016).

A bottom-up colliculo-pulvinar visual pathway

LP receives top-down inputs from various cortical regions (Bennett et al., 2019; Kamishina et al., 2009; Petry and Bickford, 2018; Roth et al., 2016; Shipp, 2001; Tohmi et al., 2014; Ungerleider et al., 2014; Zhou et al., 2017), suggesting that top-down inputs may be able to modulate V1 processing through LP. On the other hand, LP also receives input from SC, a route likely conveying bottom-up visual information (Baldwin et al., 2011, 2013; Bennett et al., 2019; Gale and Murphy, 2014; Masterson et al., 2009; Roth et al., 2016; Zingg et al., 2017), and some direct input from melanopsin-expressing retinal ganglion cells (Allen et al., 2016). Recently, an extensive anatomical and functional study demonstrates that SC projects preferentially to the posterior compartment of LP (pLP) and that pLP preferentially projects to ventral-stream V2 areas, while an anterior compartment of LP (aLP) preferentially projects to V1 and dorsal-stream V2 areas (Bennett et al., 2019). Consistently, visual responses in the postrhinal cortex (POR), a ventral-stream V2 area, have been found to be largely dependent on inputs from SC (Beltramo and Scanziani, 2019). In this study, we are interested in LP regions that receive input from the visual SC and also project to V1. Based on the coordinates, it appears that our recording sites covered a central region of LP which impinges on both the pLP and aLP described in the above study (Bennett et al., 2019). The visual responses of LP neurons in this region can be strongly driven by the bottom-up input from SC, and silencing SC results in similar changes in response properties of V1 L2/3 neurons to silencing LP. This indicates that visual input from SC can drive LP to modulate V1 responses. Furthermore, optogenetic manipulations of LP-V1 axon terminals suggest that the modulation can be largely mediated by the direct LP-V1 projection, although they do not exclude the possibility that LP can also modulate V1 responses indirectly through feedback projections from higher cortical areas such as V2. Together, we have identified a colliculo-pulvinar pathway for modulating visual response properties in V1, which has been poorly studied before.

Since LP neurons do not exhibit specific feature selectivity (Figure 7D; Durand et al., 2016; Roth et al., 2016), they do not directly contribute to V1 selectivity, but rather their activity sharpens V1 selectivity by generating broad and unselective net suppression in L2/3 neurons (Li et al., 2012; Liu et al., 2011). Since LP neurons are quite responsive to common visual stimuli, their modulation of V1 responses could be ubiquitous. In other words, LP participates in normal V1 processing and likely contributes to visual perception. Thus, it can be viewed as an important component of visual processing centers. Our behavioral experiments showing that silencing of LP impairs visual discrimination performance support the idea that LP activity in the normal condition contributes to visual discrimination. It is worth noting that the LP effect on V1 responses may be highly dependent on anesthesia/wakefulness states, since the type of modulation we have observed here was not reported in previous studies in anesthetized animals (Ahmadlou et al., 2018; Purushothaman et al., 2012; Tohmi et al., 2014).

Contextual modulation of V1 processing

Our results in this study suggest that LP can contribute to contextual modulation of V1 processing through its direct impact on V1 circuits. Contextual modulation, in particular when surround suppression is concerned, is thought to be mediated by combined feedforward, intracortical horizontal, and feedback circuit mechanisms (Angelucci et al., 2017). Subcortical neurons such as dLGN cells already exhibit a certain level of surround suppression (Alitto and Usrey, 2008; Bonin et al., 2005), which can be relayed to the cortex via feedforward thalamocortical connectivity. In L2/3 of V1, SOM inhibitory neurons strongly integrate horizontal inputs from L4-driven excitatory neurons (Adesnik et al., 2012), allowing their responses to grow with increasing stimulus sizes and their contributing to surround suppression of the excitatory neurons (Adesnik et al., 2012; Nienborg et al., 2013). Finally, feedback inputs from higher cortical areas contribute to surround suppression of V1 neurons (Hupé et al., 1998; Nassi et al., 2013; Nurminen et al., 2018), and SOM neurons could be involved (Angelucci et al., 2017; Zhang et al., 2014). These different circuit mechanisms may operate at different time and spatial scales (Angelucci et al., 2017). Our findings in this study have thus revealed a previously unrecognized source for driving surround suppression in V1 L2/3, LP/pulvinar, which is independent of the canonical retinogeniculate pathway. Since this extrageniculate pathway is mainly driven by feedforward input from SC, it may contribute to surround suppression in V1 L2/3 at a faster time scale than feedback projections.

A differential circuit

Compared to patterned visual stimuli, LP neurons seem to be more sensitive to non-patterned visual noise and their firing rates are monotonically modulated by the noise level. This property would lead to noise-dependent LP-driven suppression in V1 L2/3, which helps to maintain V1 orientation selectivity within a range of low to moderate background noise levels. When LP is silenced, increasing noise level much more quickly impairs V1 orientation selectivity. In other words, via the LP-mediated feed-forward suppressive modulation, the detrimental effect of background noise on V1 selectivity can be “cancelled” within a range of noise levels. Such “noise-cancelling” effect may be important for animals to quickly detect predators in visually noisy environments. Together, our results suggest that a bottom-up retina-SC-LP-V1 visual pathway, running in parallel with the canonical retino-geniculate-V1 pathway, forms a “differential” circuit which provides a suppressive but sharpening effect in V1 superficial layers (Figure 8). This may represent a critical circuit architecture in sensory systems to enhance information processing and achieve contextual modulations in general.

Figure 8. A circuit model for LP modulation of V1 processing.

The retina-SC-LP-V1 pathway drives “feedforward” suppression of L2/3 pyramidal neurons via L1 inhibitory neurons, and forms a differential circuit with the retino-geniculo-V1 pathway to “cancel” noise effects. Arrows represent excitatory projections. Red bar represents a suppressive effect.

STAR Methods

Contact for Reagent and Resource Sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Huizhong W. Tao (htao@usc.edu).

Experimental Models and Subject Details

The surgeries and experiments were performed in the Zilkha Neurogenetic Institute (ZNI) at the University of Southern California (USC). Animal Care and Use Committee of USC approved all procedures used in this study. Male and female wild-type C57BL/6J and transgenic Vglut2- Cre::Ai14, GAD2-Cre::Ai14, PV-Cre::Ai14 and SOM-Cre::Ai14 mice (The Jackson Laboratory) of 2–4 months old were used in this study. The animals were housed in the ZNI vivarium with 12h light/dark cycles.

Method Details

Head fixation and surgery

Three days before electrophysiological recordings, a screw was glued to the skull surface of the mouse with acrylic dental cement under the anesthesia with 1.5% isoflurane (v/v). The screw was clamped tightly with a metal head post on the recording setup to achieve head fixation. After recovery from anesthesia, the mouse was trained to get accustomed to the head fixation and to run freely on a plastic rotatable plate (Liang et al., 2015). On the day of recording, the mouse was again anesthetized with isoflurane. A craniotomy window of 0.2 mm × 0.2 mm was made over V1 (ML +2.7 mm, AP −3.45 mm), PM (ML +1.5 mm, AP −4.00 mm), LP (AP −2.2 to −2.3 mm, ML +1.4 mm, DV −2.35 mm), dLGN (AP −2.15 mm, ML +2.5 mm, DV −2.8 mm), or SC (AP −3.75 mm, ML +0.6 mm, DV −1.35 mm). A durotomy was further performed to allow the insertion of a glass pipette or silicon probe, or the implantation of a drug cannula or optic fiber. The recorded V1 was on the same side of LP or SC being manipulated. After the surgery, the exposed cortex was covered with a silicon elastomer (Kwik-CAST, WPI). The mouse was fully recovered from anesthesia before recording sessions.

Behavioral task

To assess the functionality of visual perception, mice were trained to perform a head-fixed Go/No-Go visual discrimination task (Lee et al., 2012). The visual cues are two full screen static sinusoidal grating patterns (0° and 90° in orientation, spatial frequency = 0.04 cpd, full contrast). The inter-stimulus-interval was randomly chosen between 10–12 s. The cues were presented for 0.5 s according to a pseudorandom sequence, followed by a delay period of 0.5 s. The reward (30 nl water) was given after the delay period when the 90° (vertical) grating (Go signals) was presented, and no reward was delivered when the 0° (horizontal) grating (No-Go signals) was present. After initial two days of water deprivation, mice were trained to lick to the Go cues but not to lick upon the No-Go cues. A trial was terminated 1 s after the water reward. No punishment was given when the animals licked to the No-Go cues. The licking behavior was videotaped by a camera placed closed to the mouth of the animal for online monitoring and offline analysis. Behavioral performance was quantified with a correct rate (the percentage of hit plus correct rejection trials). The animals were considered to have learned the task when the correct rate had been above 75% for two concessive days. One day before the last training day, 100 nl saline was infused into LP, and on the last day 100 nl of muscimol was bilaterally infused into LP. In another cohort of mice, AAV encoding inhibitory DREADD receptors (hM4Di) was bilaterally injected into LP two weeks before the beginning of training sessions. On the last training day, the DREADDi agonist, CNO, was administered (i.p., 1mg/kg) 40 min before the behavior test to chemogenetically silence LP. The correct rates one day before (saline) and on the day of LP silencing (either through muscimol injection or chemogenetic inactivation) were compared. For control, we injected CNO in well-trained GFP-expressing animals and compared their performance before and after the CNO injection.

In vivo electrophysiological recording

All recordings were performed in a dark room with weak background illumination. A single recording session was limited to at most 4hr. Between sessions, the mouse was allowed to rest and receive drops of 5% sucrose. The silicon seal was removed before recordings. Loose-patch recordings were performed following our previous studies (Ibrahim et al., 2016; Liang et al., 2015). A patch pipette (5–7 MQ) was filled with artificial cerebral spinal fluid (ACSF; 126 mM NaCl, 2.5 mM KCl, 1.25 mM Na₂PO₄, 26 mM NaHCO₃, 1 mM MgCl₂, 2 mM CaCl₂, and 10 mM glucose). A seal of 0.1–0.5 GQ was achieved on the soma. Spikes from the patched neuron were recorded with an Axopatch 200B amplifier (Molecular Devices) under voltage-clamp mode, with clamping voltage adjusted to achieve a zero baseline. The electric signals were sampled at 20 kHz sampling rate and passed through a 300–3000 Hz band-pass filter. The relatively large opening of the pipettes (2 μm) highly biased the recordings in favor of excitatory pyramidal neurons (Liang et al., 2015; Liu et al., 2010). The location of V1 was verified from the correct pattern of a retinotopic map in the exposed region, and all the recordings were performed in the monocular region of the V1 (Drager, 1975). The cortex was pre-penetrated before recording to minimize the tissue dimpling. The cortical layers were assigned according to the travel distance of the pipette orthogonal to the pia surface of V1, which was found to match well with the actual cortical depth determined form post hoc histological results in experiments where the pipette was coated with Dil (Figure S8A). The recorded L2/3 neurons had a range of depths, 170–350 μm from the pia, and L4 neurons, 375–500 μm from the pia (Figure S8B). This layer assignment was based on our previous observation in a L4-specific line, Scnn1a-Cre, that Scnn1a expressing neurons were located at from 374 ± 14 μm to 510 ± 16 μm below the pia (Li et al., 2012). To record from neurons in LP, dLGN or PM, a 64-channel silicone probe (NeuroNexus) was used. Signals were recorded by an Open-Ephys system (Open Ephys) at 30 kHz sampling rate. Raw unfiltered traces were saved for offline spike sorting and analysis. Typically, each animal went through 1–2 recording sessions. Recordings were performed during quiet wakefulness, and pharmacological or optogenetic manipulations of brain activity did not change the general locomotion status of the animals. We also monitored the pupil position during the recording sessions and found that optical stimulation did not trigger significant eye movements (Figure S3K).

Viral injection

The injection of virus was performed on a stereotaxic apparatus as previously reported (Chou et al., 2018; Liang et al., 2015; Xiong et al., 2015). The mouse was anesthetized with 1.5% (v/v) isoflurane. A 0.2 mm × 0.2 mm craniotomy was performed to expose the cortex above LP (AP −1.5 mm, ML +1.4 mm, DV −2.35 mm) or SC (AP −3.75 mm, ML +0.6 mm, DV −1.35 mm). The dura mater was then removed. The virus was delivered via a beveled glass micropipette with a tip opening 20–30 μm in diameter, which was attached a microsyringe pump system. The following adeno-associated viruses (AAVs) encoding ChR2, ArchT, eNpHR3.0 and h4MDi were used to optogenetically or chemogenetically manipulate neuronal activity: AAV1-CaMKIIa-hChR2-EYFP (1.7×10¹³ GC/ml, UPenn vector core, Addgene 26969), AAV9-EF1a-DIO-eNpHR3.0-EYFP (1.7×10¹³ GC/ml, UPenn vector core, Addgene 26966) and AAV1-CAG-ArchT-GFP (1.7×10¹³ GC/ml, UNC vector core, Addgene 29777), and AAV5-hSyn-hM4Di(Gi)-mCherry (3×10¹² GC/ml, UPenn vector core, Addgene 50475). 50 nl of the viral solution was injected into LP at a rate of 15 nl/min. AAV9-EF1a-DIO-eNpHR3.0-EYFP was injected in the Vglut2-ires-Cre knock-in mice (The Jackson Laboratory) to specifically silence the axonal terminals of excitatory LP projection neurons in V1. The pipette then stayed in the LP for 5 min before withdrawal to prevent leakage. For anterograde tracing of neural projections and control experiments for chemo- or optogenetics, a volume of 50 nl AAV1-CB7-CI-eGFP-WPRE-rBG solution (1.7×10¹³ GC/ml, UPenn vector core, Addgene 105542) was injected to SC and LP. To anterogradely label the axonal outputs of SC neurons that receive direct retina inputs, we adopted a two-step viral tracing approach (Zingg et al., 2017). 50 nl of AAV2/1-hSyn-Cre-WPRE-hGH (UPenn Vector Core, 2.5×10¹³ GC/mL, Addgene 105553) was first injected into the eye of Ai14 mice. After 3 days, a second load of AAV2/1-CAG-FLEX-eGFP-WPRE-bGH (UPenn vector core, 1.7×10¹³ GC/mL, Addgene 51502) was injected into the contralateral SC (50 nl). The spacing between two injections allowed sufficient time for the clearance of any remaining AAV-Cre virus that may have spread across the pial surface. For retrograde labelling of V1-projecting LP neurons, 50 nl of cholera toxin subunit B, Alexa 488 (CTB-488, 0.5% solution in PBS, ThermoFisher) was injected into V1. The scalp was re-sutured, and the mouse was administrated with 0.1 mg/kg buprenorphine subcutaneously and sent back to home cages. The recordings and imaging were performed 3 weeks after viral injection to allow the expression of the virus and the recovery of the mouse from surgery.

Histology and imaging

After experiments, the mouse was deeply anesthetized and transcardially perfused with 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS). The brain tissue was then harvested and sliced into 150 μm coronal sections using a vibratome (Leica, VT1000s). Fluorescence images were acquired using a confocal microscope (Olympus FluoView FV1000) to verify the location of virus expression and drug injection. Images of both the injection site and projection from SC to LP or LP to V1 were collected and imaged under a 4× objective. Regions with axonal labeling were further imaged under a 10× objective to get clearer projection patterns.

Visual and LED stimulation

Visual stimuli were generated in Matlab (MATLAB) with the Psychophysics Toolbox Version 2 (Brainard, 1997) and were presented on a ViewSonic VA705b monitor (1920 × 1440 pixels, 33.9 cm wide, 27.2 cm high, 60 Hz refresh rate, mean luminance 41 cd/m²) mounted on a flexible arm. The monitor was adjusted to be 20 cm away from the eye, at 45° azimuth, 25° elevation, and thus subtending 80° azimuth × 70° elevation of mouse’s visual field, corresponding to the monocular region of V1. The monitor was gamma-corrected to achieve linear luminance. To pre-map V1 or measure the receptive field (RF) center of a patched neuron, static bright (58 cd/m²) and dark (24 cd/m²) squares (each square 10° × 10°) of an 8 × 6 grid covering the entire screen were presented individually in a pseudorandom sequence on a gray background of mean luminance. Each square was displayed for 200 ms with 5–10 repeats, and the inter-stimulus interval was 440 ms. To measure the overall visual response level, a set of dense white-noise stimuli were presented. Each frame of the stimuli consisted of a grid of 20 × 20 squares (each square 4° × 3°) intensities of which were determined by an m-sequence. Each pattern was presented for 200 ms and 20–50 patterns were presented according to response level and fidelity. To measure orientation and direction tuning, a set of drifting sinusoidal gratings were presented in pseudorandom order. The stimulus set consisted of gratings of 12 directions (0°−330°, 30° per step, 6 orientations). To drive as many V1 neurons as possible, the spatial frequency of the gratings was chosen to be 0.04 cycles per second (cpd) and temporal frequency to be 2 cycles per second (Hz) based on previous research (Niell and Stryker, 2008). Each grating drifted for 1.5 s and another grating appeared, which remained to be static for 3 s before moving. To measure size tuning, drifting gratings of the optimal orientation but of varying sizes were presented at RF center. The size (diameter) of each grating stimulus was chosen from an array of values as 6°, 9°, 14°, 20°, 30°, 46°, 68° and 102° according to a pseudorandom sequence. To prevent from direction adaptation, gratings with preferred and null directions were interleaved. To precisely map spatial RFs, sparse noise stimuli, composed of static bright and dark squares (each square 3° × 3°) was used. The squares were presented individually on a gray screen within a 16 × 16 grid centered at RF center according to an m-sequence. Each square was displayed for 32 ms (i.e. updated every other frame), with zero inter-stimulus interval between 2 consecutive squares. To test V1 processing with a visually noisy background, we generated a series of visual noise patterns. As control, static sinusoidal grating patterns of 4 phases (0, π/2, π, 3π/2) and 6 orientations ranging from 0° to 150° (step = 30°) with 50% contrast (pixel values ranging evenly from 0.25 to 0.75, with 0 means black and 1 white, respectively) was generated. A series of noise squares with increasing contrast (17%, 33%, 50% and 75%, pixel values evenly distributed in ranges = [−0.08, +0.08], [−0.17, +0.17], [−0.25, +0.25], [−0.375, 0.375]) were added onto the grating patterns, so that the grating patterns were increasingly blurred and invisible among the noise. The pixel values of the merged images were cutoff at [0.25, 0.75] to ensure the mean luminance and the range of luminance of the pixels were the same as the original grating pattern. Each pixel of the noise patterns was 4° in visual degree, and each pattern was presented 200 ms for 10–20 repetitions in a pseudorandom order. For optogentic manipulations, two sequences were generated independently for drifting grating stimuli, with trials with and without LED light stimulation interleaved. The inter-stimulus interval between consecutive LED-ON and LED-OFF trials was 10 s to allow light-gated channels fully recover from desensitization.

In vivo optogenetic manipulation

To photo-manipulate LP neurons, an optic fiber (200 μm, Thorlabs) was inserted through the cortex close to the surface of LP (ipsilateral to the V1 recorded) at the depth of 2.1 mm. The fiber was then secured with acrylic dental cement. Right before recording sessions, the fiber was connected to a LED source (Thorlabs) to deliver blue (473 nm, 10 mW), green (530 nm, 10 mW) or yellow light (589 nm, 10 mW). No labeled neural structures other than LP were observed within 800 μm depth from the tip of the fiber. In the LED-ON sessions, light stimulation preceded the onset of visual stimulation by 500ms, covered the entire duration of visual stimulation and terminated at 300ms after the offset of visual stimuli. Blue light pulses (20 ms in duration, 20 Hz) and continuous green light were applied to examine the effect of activating (via ChR2) and silencing (via ArchT) LP on cortical visual processing, respectively. For terminal manipulation, optic fiber was implanted over V1. Blue light pulses and continuous yellow light were given to activate (via ChR2) and silence (via eNpHR3.0) LP axonal terminals in V1, respectively.

Pharmacological silencing of brain regions

Bupivacaine (a voltage-gated sodium channel blocker, 0.5%, dissolved in ACSF containing Alexa conjugated dextran) was used to silence targeted brain regions. To effectively silence LP (ipsilateral to the V1 recorded), a cannula was implanted 2.5 mm below the pia and 100 nl bupivacaine was slowly infused. For silencing superficial SC (ipsilateral to the V1 recorded), a cannula was implanted at the depth of 0.75 mm and a volume of 100 nl the drug was slowly administrated. The silencing effect was verified by comparing spontaneous and evoked spiking responses before and after drug infusion. We found that spiking activity was silenced at 5 min after the beginning of the infusion for at least 30 min and was recovered after 1.2 hr (data not shown). This method allowed us not only to temporally and reversibly silence LP when recording from one neuron, but also to record from multiple neurons during the entire recording session. In another set of multi-channel recording experiments, we infused 100 nl of fluorescence conjugated muscimol (1.5 mM, Life Technologies), an agonist of GABA_a receptors, into sSC and V1 for silencing while recording from LP or dLGN. In behavioral tests, muscimol was bilaterally infused into LP for silencing for a longer period of time (> 2hr).

Slice preparation and recording

To confirm the functional connectivity from LP to V1, GAD2-Cre, PV-Cre or SOM-Cre crossed with Ai14 mice and injected with AAV1-CaMKIIa-hChR2-EYFP in LP were used for slice recording. Three weeks following the injections, animals were decapitated following urethane anesthesia and the brain was rapidly removed and immersed in an ice-cold dissection buffer (composition: 60 mM NaCl, 3 mM KCl, 1.25 mM NaH₂PO₄, 25 mM NaHCO₃, 115 mM sucrose, 10 mM glucose, 7 mM MgCl₂, 0.5 mM CaCl₂; saturated with 95% O₂ and 5% CO₂; pH = 7.4). Coronal slices at 350 μm thickness were sectioned by a vibrating microtome (Leica VT1000s), and recovered for 30 min in a submersion chamber filled with warmed (35 °C) ACSF (composition:119 mM NaCl, 26.2 mM NaHCO₃, 11 mM glucose, 2.5 mM KCl, 2 mM CaCl₂, 2 mM MgCl₂, and 1.2 mM NaH₂PO₄, 2 mM Sodium Pyruvate, 0.5 mM VC). Cuts were made in the slices along the boundaries of secondary visual areas flanking V1 (identified based on fluorescence pattern of LP axons) before recording to prevent potential feedback inputs to V1. EYFP⁺ fibers, L1 inhibitory neurons or PV/SOM inhibitory neurons with tdTomato expression were visualized under a fluorescence microscope (Olympus BX51 WI). Patch pipettes (~ 4–5 MQ resistance) filled with a cesium-based internal solution (composition: 125 mM cesium gluconate, 5 mM TEA-Cl, 2 mM NaCl, 2 mM CsCl, 10 mM HEPES, 10 mM EGTA, 4 mM ATP, 0.3 mM GTP, and 10 mM phosphocreatine; pH = 7.25) were used for whole-cell voltage-clamp recordings. Synaptic currents were recorded with an Axopatch 200B amplifier (Molecular Devices) under voltage clamp mode at a holding voltage of −70 mV for excitatory currents or 0 mV for inhibitory currents, filtered at 2 kHz and sampled at 10 kHz. Tetrodotoxin (TTX, 1 μM) and 4-aminopyridine (4-AP, 1 mM) were added to the external solution for recording monosynaptic responses only to blue light stimulation (10 ms pulse, 3 mW power, 10–30 trials, delivered via a mercury Arc lamp gated with an electronic shutter). A neuron was considered as being innervated if blue light pulses consistently evoked monosynaptic responses in the presence of TTX and 4AP with an average amplitude larger than 3 standard deviations of baseline fluctuations and the response onset latency within a 3-ms time window after the onset of light pulses. To test the excitatory nature of the LP to V1 input, a glutamate receptor antagonist, 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX, 20 μM) was further added to the bath solution. For current clamp recordings, the potassium-based internal solution was used (130 mM K-gluconate, 2 mM KCl, 1 mM CaCl₂, 4 mM MgATP, 0.3 mM GTP, 8 mM phosphocreatine, 10 mM HEPES, 11 mM EGTA, pH = 7.25). Membrane potentials were recorded under current-clamp mode. To examine the effect of LP-V1 terminal activation on the postsynaptic membrane potential of V1 L2/3 neurons, a brief blue light pulse (10 ms pulse duration, 3 mW power, 10–30 trials) was applied. A positive current (100 pA, 500 ms) was injected into the cell to induce spiking. A train of blue light pulses (10 pulses at 20 Hz, 25 ms pulse duration, 10–30 trials) was applied concurrently with the current injection. Trials with and without blue light pulses were interleaved. The postsynaptic membrane potentials and firing rates without and with LP-V1 terminal activation were compared.

Quantification and Statistical Analysis

Data analysis

Data analysis was performed using customized scripts written in Matlab (MATLAB) and Labview (National Instruments). Raw waveforms were saved online, and spikes were detected offline using a thresholding algorithm.

Response level

To quantify the overall response level of V1 neurons, the number of spikes evoked by dense white noise flashes was counted within the 70–270 ms window after the onset of the visual stimulus. The spike number was averaged across trials to derive firing rate. The spontaneous firing rate was calculated from the time window 200 ms before the stimulus onset when a static grating pattern was present on the screen, and then subtracted from the evoked firing rate. The response levels before and after drug application or between LED-ON and LED-OFF trials were compared.

Orientation and direction tuning

To derive tuning curves, the spike number evoked by drifting gratings was counted within the 70–1570 ms window after stimulus onset. The responses were organized according to the direction of moving stimuli and averaged across repetitions. Spontaneous firing rate calculated within a 500 ms window before the stimulus onset was subtracted. Orientation tuning curves were obtained by averaging responses to two opposite directions (i.e. 0° and 180°, 30° and 210°, etc). To further quantify tuning strength, a global orientation (gOSI) and direction selectivity index (gDSI) were determined by dividing vector summation with the sum of scalar values as follows:

$$\begin{gathered} gOSI=\frac{\mathrm{\Sigma }\Vert R\left(\theta \right)e^{2i\theta }\Vert }{\mathrm{\Sigma }R\left(\theta \right)} \\ gDSI=\frac{\mathrm{\Sigma }\Vert R\left(\theta \right)e^{i\theta }\Vert }{\mathrm{\Sigma }R\left(\theta \right)} \end{gathered}$$

θ is the direction of moving grating. R(θ) is the response to direction θ. $i=\sqrt{-1}$. Both gOSI and gDSI ranges from 0 to 1. gOSI = 0 (or gDSI = 0) means that the neuron responds equally to all orientations (or directions). gOSI = 1 (or gDSI = 1) implies the neuron responds exactly to just one orientation (or direction). gOSI and gDSI values before and after drug application or between LED-ON and LED-OFF trials were compared. The conventional OSI was calculated as:

$$OSI=\frac{R_{pref}-R_{orth}}{R_{pref}+R_{orth}},$$

where R_pref and R_orth are the responses to the preferred and the orthogonal orientation (90° from the preferred orientation), respectively. Similarly, the conventional DSI was calculated as:

$$DSI=\frac{R_{pref}-R_{null}}{R_{pref}+R_{null}},$$

where R_pref and R_null are the responses to the preferred and the null direction (180° from the preferred direction), respectively.

Size tuning

To characterize the neuronal response to gratings of increasingly larger sizes, responses to gratings of optimal direction within a window of 70–1570 ms after the stimulus onset were organized by grating size and averaged across trials. A neuron was defined as size-tuned if the largest stimuli did not elicit the most robust response. The tuning strength was further quantified by surround suppression index (SSI), which determined as (Adesnik et al., 2012):

$$SSI=\frac{R_{pref}-R_{large}}{R_{pref}},$$

where R_pref is the response to the grating of the optimal size, R_large is the response to the largest grating. SSI = 1 suggests zero R_large (i.e. complete surround suppression), while SSI = 0 indicates no surround suppression at all. SSIs before and after drug application or between LED-ON and LED-OFF trials were compared.

Spatiotemporal receptive field (STRF)

To map the fine structure of RF, the spike train to the sparse noise stimuli was reversely correlated with the stimulus sequence to derive the spike-triggered average response of the stimuli (Jones and Palmer, 1987). A 2D standard deviation was calculated at each time lag from 0–200 ms as the strength of RFs. The RF showing strongest response was used to determine the size of RF. The background response level was determined by averaging the spike numbers at the 4 borders of RF and subtracted from RF. The RF map was further divided by its 2D standard deviation to obtain Z scores. Pixel value with Z score < 4 was set to be 0. The area of connected pixels was calculated and transformed to a circle with the same area. The radius of the circle was defined as the RF size. To better visualize the boundary, the thresholded RF map was fitted with a 2D elliptical Gaussian function Φ(X, Y). 2σ_X and 2σ_Y were used to outline the boundary of RF.

Spike sorting

Spike sorting was performed following our previous study (Zhang et al., 2018). The raw signals from the 64-channel silicon probe were filtered through a 300–6000 Hz band-pass filter. The spatially varying motion artifacts were removed by applying a local common average referencing (L-CAR) scheme. The nearby four channels of the silicon probe were grouped as tetrodes, and semi-automatic spike detection and sorting was performed using the Plexon offline sorter (Dallas, Texas). Clusters with isolation distance > 20 was considered as separate clusters. Spike clusters would be classified as single units only if the waveform SNR (Signal Noise Ratio) exceeded 4 (12 dB) and the inter-spike interval was longer than 1.2 ms for more than 99.5% of the spikes.

Statistics

The Shapiro-Wilk test was performed to test the normality of the data set. If the data were normally distributed, parametric paired t-test or two-sample t-test was used. Otherwise, the nonparametric Wilcoxon signed rank test or Mann-Whitney rank sum test was used. Two-sample Kolmogorov-Smirnov test was used to determine whether the data from two groups were from the same distribution. Bootstrapping approach was used to estimate the standard error of the mean (SEM) of gOSI and gDSI, and permutation test is used to determine whether gOSI or gDSI values between two conditions are significantly different. The significance level of the tests was set as 0.05. Unless mentioned in the text, the data were reported as mean ± SEM.

Key Resources Table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Fluorescent Nissl Stain	Invitrogen	RRID: AB_2572212
Bacterial and Virus Strains
AAV1-CaMKIIa-hChR2-EYFP	UPenn vector core	Addgene 26969
AAV9-EF1a-DIO-eNpHR3.0-EYFP	UPenn vector core	Addgene 26966
AAV1-CAG-ArchT-GFP	UNC vector core	Addgene 29777
AAV5-hSyn-hM4Di(Gi)-mCherry	UPenn vector core	Addgene 50475
AAV1-CB7-CI-eGFP-WPRE-rBG	UPenn vector core	Addgene 105542
AAV2/1-hSyn-Cre-WPRE-hGH	UPenn vector core	Addgene 105553
AAV2/1-CAG-FLEX-eGFP-WPRE-bGH	UPenn vector core	Addgene 51502
Biological Samples
N.A.
Chemicals, Peptides, and Recombinant Proteins
DiI	Invitrogen	D282
Paraformaldehyde	Alfa Aesar	10194340
NaCl	OmniPur	UI27FZEMS
KCl	Mallinckrodt	7447–40–7
NaHCO3	EMD Chemicals	48204847
MgCl2	J.T. Baker	7791–18–6
CaCl2	EMD Chemicals	41046444
NaH2PO4	EMD Chemicals	SX0320–1
Sodium Pyruvate	J.T. Baker	3354–04
Cs-Gluconate	Sigma	G4625
TEA-Cl	Sigma	T2265
CsCl	Sigma	289329
HEPES	Sigma	SLBX2493
EGTA	Sigma	324626
ATP	Sigma	A2383
GTP	Sigma	G8877
Phosphocreatine	Sigma	P7936
K-Gluconate	Sigma	G4500
MgATP	Sigma	A9187
Glucose	Sigma	SLBC6575V
Sucrose	Sigma	D00168514
Agarose	OmniPur	3332C511
Muscimol	Fisher Scientific	28910
Clozapine-N-oxide	Tocris	34233–69–7
Alexa Fluor® 488	Invitrogen	C-22841
4-Aminopyridine	Alfa Aesar	A12405
6-Cyano-7-nitroquinoxaline-2,3-dione	Sigma	C239
Tetrodotoxin	Sigma	T8024
Bupivacaine	Sigma	1078507
Critical Commercial Assays
N.A.
Deposited Data
N.A.
Experimental Models: Cell Lines
N.A.
Experimental Models: Organisms/Strains
Mouse: C57BL/6J	The Jackson Laboratory	RRID: IMSR_JAX:000664
Mouse: Ai14	The Jackson Laboratory	RRID: IMSR_JAX:007914
Mouse: VGluT2-ires-Cre	The Jackson Laboratory	RRID: IMSR_JAX: 016963
Mouse: GAD2-ires-Cre	The Jackson Laboratory	RRID: IMSR_JAX: 010802
Mouse: PV-Cre	The Jackson Laboratory	RRID:IMSR_JAX: 017320
Mouse: SOM-ires-Cre	The Jackson Laboratory	RRID:IMSR_JAX: 013044
Oligonucleotides
N.A.
Recombinant DNA
N.A.
Software and Algorithms
LabVIEW	LabVIEW	http://www.ni.com; RRID: SCR_014325
Matlab	Matlab	http://www.mathworks.com/; RRID: SCR_001622
Allen Reference Atlas	Allen Institute	http://www.brain-map.org; RRID: SCR_006491
Offline sorter	Plexon	https://plexon.com
Prism	GraphPad	https://www.graphpad.com/scientific-software/prism/; RRID: SCR_00279
Fiji	NIH	https://fiji.sc/; RRID: SCR_002285
Other
NI board for data acquisition	National Instrument
Multi-channel silicone probe	Neuronexus Technologies	A1x64-Poly2–6mm-23s-160
Monitor	ViewSonic

Highlights.

• LP enhances feature selectivity in V1 by providing subtractive surround suppression.

• This suppression is mediated by LP innervation of layer 1 inhibitory neurons in V1.

• The bottom-up retina-SC-LP-Vl pathway constructs a differential visual circuit.

• LP helps V1 to maintain its orientation selectivity under varying noise background.