Emergence of Direction Selectivity at the Convergence of ... · 1 Emergence of Direction...

Emergence of Direction Selectivity at the Convergence of Thalamo-Cortical Synapses in Visual 1

Cortex 2

3

Anthony D Lien 1-3 and Massimo Scanziani 1-4 41 Neurosciences Graduate Program, University of California San Diego, La Jolla, California, USA. 2 5

Center for Neural Circuits and Behavior, Neurobiology Section and Department of Neuroscience, 6

University of California San Diego, La Jolla, California, USA. 3 Department of Physiology, University of 7

California San Francisco, San Francisco, California, USA. 4Howard Hughes Medical Institute, University 8

of California San Francisco, San Francisco, California, USA. 9

10

Abstract 11

12

Detecting the direction of an object’s motion is essential for our representation of the visual 13

environment. Visual cortex is one of the main stages in the mammalian nervous system where 14

motion direction may be computed de novo. Experiments and theories indicate that cortical neurons 15

respond selectively to motion direction by combining inputs that provide information about distinct 16

spatial locations with distinct time-delays. Despite the importance of this spatiotemporal offset for 17

direction selectivity its origin and cellular mechanisms are not fully understood. We show that 18

~80+/-10 thalamic neurons responding with distinct time-courses to stimuli in distinct locations 19

contribute to the excitation of mouse visual cortical neurons during visual stimulation. Integration 20

of thalamic inputs with the appropriate spatiotemporal offset provides cortical neurons with the 21

primordial bias for direction selectivity. These data show how cortical neurons selectively combine 22

the spatiotemporal response diversity of thalamic neurons to extract fundamental features of the 23

visual world. 24

25

26

The ability to detect the direction of motion of objects in the visual world is an essential property of 27

sensory processing. In the primary visual cortex of mammals many neurons preferentially respond to 28

stimuli moving in a specific direction within their receptive field 1,2. In primates and carnivores, this 29

sensitivity to motion direction is believed not to be inherited from earlier stages of visual processing but 30

computed de novo in visual cortex. Here, neurons likely extract directional motion from the visual scene 31

peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/244293doi: bioRxiv preprint first posted online Jan. 7, 2018;

http://dx.doi.org/10.1101/244293

by combining inputs that respond with different temporal delays to stimuli presented in different locations 32

of the neuron’s receptive field, that is by combining spatially and temporally offset inputs 3-11. All 33

theoretical models of direction selectivity, from the most general 12,13 to those that more closely mimic the 34

receptive field structure of visual cortical neurons 14-17, are based on this spatiotemporal offset. However, 35

the cellular mechanisms and synaptic connectivity patterns that generate the spatiotemporal offset of 36

visual responses in cortical neurons remain speculative. Some models propose that direction selectivity 37

emerges through intracortical interactions while others posit that direction selectivity is first generated 38

through thalamo-cortical interactions. Other models, based on the rodents’ visual system, suggest that 39

direction selectivity may actually not be computed in cortex but is inherited from earlier stages of visual 40

processing. Models of direction selectivity generated through intracortical interactions (Fig. 1a) propose 41

that anisotropic connectivity patterns between cortical neurons result in a spatial offset between excitation 42

and inhibition or in a spatial offset between excitatory inputs with distinct time-courses 10,15,16,18-22. Models 43

proposing that direction selectivity results from thalamo-cortical interactions suggest that spatially offset 44

thalamic inputs with different time-courses in their responses to visual stimuli converge onto individual 45

cortical neurons (Fig. 1a) 7,23-26. Models proposing that direction selectivity is not computed in the cortex, 46

suggest that direction selectivity already existing in the retina may be transmitted to the cortex via the 47

thalamus (Fig. 1a) 27-30. Here we take advantage of our ability to isolate individual thalamic inputs onto 48

layer 4 (L4) cortical neurons of mice to identify the mechanism for the de novo generation of direction 49

selectivity in primary visual cortex (V1). 50

51

Thalamic excitation reports direction of motion 52

To investigate the synaptic mechanisms underlying direction selectivity, we performed whole-cell patch-53

clamp recordings from neurons located between 300-550 um from the pial surface in V1 of anesthetized 54

mice while displaying visual stimuli to the contralateral eye 31. This range of depth corresponds largely to 55

the radial extent of L4 but may include some deep L2/3 and superficial L5 neurons. For simplicity we will 56

refer to the neurons recorded at this depth as L4 neurons. We recorded the spiking and the membrane 57

potential (Vm) in response to 12 different stimuli presented on a monitor and consisting of gratings (arrays 58

of dark and light bars) of 6 orientations drifting in either one of two opposite directions along the axis 59

perpendicular to the gratings’ bars. When presented with gratings at their preferred orientation some L4 60

neurons (Fig. 1c) fired many more action potentials in response to movement in one direction as compared 61

to the opposite, consistent with previous studies 32. Vm fluctuated at the fundamental temporal frequency 62


http://dx.doi.org/10.1101/244293

of the drifting grating (F1 modulation), that is at the frequency at which any pixel on the monitor 63

transitions from dark to light and back to dark (Fig. 1c and d). The direction selectivity index (DSI, see 64

methods) of the amplitude of the F1 modulation of Vm (VmF1) correlated with the DSI of the neuron’s 65

spiking response (r=0.52 p=0.000103, Fig. 1g, left) and in most cells the preferred direction was the same 66

for both parameters (46/51 matching preferred direction p<0.0001 binomial test; Fig. 1g left). 67

68

To determine whether the direction preference of L4 neurons is already apparent in the thalamic excitatory 69

synaptic input we isolated thalamic excitation by silencing cortical excitatory neurons via photo-70

activation of cortical inhibitory interneurons expressing Channelrhodopsin 2 (ChR2) 33, while recording 71

from L4 neurons in the voltage clamp configuration at the reversal potential of inhibition, as described 72

previously 31 and presenting gratings drifting in either of the two direction perpendicular to the previously 73

determined preferred orientation. Cortical silencing completely abolished firing in cortical excitatory 74

neurons, including in the deepest cortical layers, and reduced visually evoked excitation by ~65% 75

(63±16%, drifting grating; n=66 cells; 68±16%, static grating; n=53 cells; Extended Data Fig. 1), 76

consistent with the fact that during visual stimulation the majority of excitation received by L4 neurons is 77

of cortical origin 26,31,34. The residual excitation represents the isolated thalamic input 31,34. 78

79

To determine whether thalamic excitation shows direction selectivity we used two parameters: First, the 80

amplitude of the F1 modulation of the thalamic excitatory current. Second, the thalamic excitatory current 81

integrated over the duration of the visual stimulus (thalamic charge), i.e. the total amount of thalamic 82

excitation received by a L4 neuron in response to the stimulus. The amplitude of the F1 modulation of 83

thalamic excitation (ThalF1) showed direction selectivity in 38% of the neurons (we defined ThalF1 as 84

being direction selective when the DSI exceeded 0.3; 25/66 cells; Fig. 1h left). Importantly, the preferred 85

direction of ThalF1 predicted the preferred direction of VmF1 in the majority of cells (37/52 cells 86

p=0.0032; binomial test; Fig. 1g middle; if the test is restricted to cells with DSI VmF1>0.3: 22/28 had the 87

same preferred direction; p=0.0037, binomial test; the DSI of ThalF1 also correlated with the DSI of 88

VmF1 [r=0.33; p=0.0168; n=52 cells; Fig. 1g middle]). 89

90

In contrast, the thalamic charge showed no direction selectivity (in 64/66 cells DSI did not exceed 0.3) 91

and was on average the same for the preferred and non-preferred directions of ThalF1 (for cells with DSI 92

ThalF1 > 0.3 the normalized thalamic charge was 0.98 ± 0.03 in the preferred and 0.93 ± 0.09 in the non-93


http://dx.doi.org/10.1101/244293

preferred direction, p=0.13 n=25 cells; for DSI ThalF1 < 0.3: 0.97 ± 0.06 preferred, 0.95 ± 0.06 non-94

preferred, p = 0.22 n=41 cells, Wilcoxon rank-sum test, Fig. 1h right). The DSI of thalamic charge did not 95

correlate with the DSI of VmF1 (r=0.2 p=0.16 n=52 cells, Fig. 1g right) and did not predict the preferred 96

direction of VmF1 (32/52 had the same preferred direction; p: 0.126, binomial test; if the test is restricted 97

to cells with DSI VmF1>0.3: 18/28 had the same preferred direction; p=0.1849, binomial test; Fig. 1g 98

right). Finally, to quantify the possible contribution of a thalamic charge bias to the DSI of ThalF1 we 99

equalized the thalamic charge evoked by gratings drifting in both directions (see methods). This procedure 100

led to a very small, yet significant reduction in DSI of ThalF1 (subset of neurons where DSI of ThalF1 > 101

0.3: DSI before equalization: 0.49 +/- 0.15; DSI after equalization: 0.46 +/- 0.17; p=0.022, paired t-test; 102

n=25; Extended Data Figure 2). These data demonstrate that direction preference in L4 neurons is already 103

prominent in the amplitude modulation of thalamic excitation 26 but much less so in the charge. Thus, an 104

essentially equal quantity of thalamic excitation is differently distributed in time depending on the 105

direction of the stimulus. 106

107

The spatiotemporal offset of thalamic excitation 108

What accounts for the differential temporal distribution of thalamic excitation for stimuli moving in 109

opposite directions? The amplitude of thalamic excitation in response to a moving stimulus is the 110

summation of the momentary thalamic excitation driven by the current position of the stimulus with the 111

residual excitation generated by the previous stimulus position within the receptive field. Heterogeneity in 112

the time-course of thalamic excitation elicited by stimuli at different receptive field locations could cause 113

the observed differences in the amplitude modulation for motion in opposite directions 7,23-26. 114

115

To determine whether the time-course of thalamic excitation depends on the position of the stimulus, we 116

presented, while silencing cortex, static gratings of the preferred orientation at 16 different spatial phases, 117

separated by 22.5º and presented randomly (250 ms each; Fig. 2a-b). The static gratings triggered thalamic 118

excitatory postsynaptic currents (EPSCs) with a latency of 39.8 ± 7.3 ms (time to 20% of peak amplitude, 119

n=53 neurons), an amplitude of -105 ± 61 pA (range -28 to -280 pA, measured for the spatial phase 120

eliciting the largest amplitude, n=53 neurons) and a duration of 145± 33 ms (range: 46-180 ms duration; 121

n=53 neurons; time from 10% to 90% of integral). 122

123


http://dx.doi.org/10.1101/244293

To validate that static gratings allow us to recapitulate the dynamics of thalamic excitation in response to 124

drifting gratings we computed the algebraic sum of thalamic EPSCs evoked by each of the 16 phases of 125

the static gratings (Fig. 2c,d). The currents were staggered in time (by 31.25 ms, i.e. by 1/16 of the period 126

of the drifting grating) to match the temporal sequence at which each of the 16 individual spatial phases 127

occur during a drifting grating (Fig. 2c,d); we also staggered the gratings by shorter intervals to mimic 128

higher temporal frequencies; Extended Data Fig. 3). We compared the algebraic sum in which EPSCs 129

were ordered according to the spatial phase sequence simulating the motion of the grating in one direction 130

against the sum simulating motion in the opposite direction. In the large majority of recorded L4 neurons 131

the F1 amplitude modulation of the summed thalamic EPSCs evoked by static gratings had a direction 132

preference matching that of ThalF1 evoked by drifting gratings (Fig. 2g; 41/53 matching preferred 133

direction p<0.0001, binomial test; for neurons where DSI ThalF1 to drifting gratings >0.3: 18/23 matching 134

preferred direction p=0.01; binomial test; among the 12 neurons where direction preference was not 135

matched 7 were not direction selective (DSI ThalF1 to drifting gratings <0.3)). While we have no 136

explanation for the unmatched directional preference of the remaining 5 cells, the thalamic charge of these 137

neurons in response to drifting gratings was on average not direction selective (DSI of thalamic charge: 138

0.09 +/- 0.07; n = 5) consistent with the data above). These results validate static gratings as an approach 139

to determine the dynamics of thalamic excitation underlying direction preference. Below we restrict our 140

analysis to those 46 neurons where the DSI to static gratings is > -0.1 (Fig. 2g). 141

142

The amplitude of thalamic EPSCs evoked by static gratings depended on the spatial phase of the grating, 143

consistent with the spatial separation in ON and OFF sub-regions of thalamic excitation 31 (Fig. 2a). 144

Strikingly, however, the time-course of the decay of the thalamic EPSCs also depended on the spatial 145

phase (Fig. 2a, asterisks). We quantified the impact of the spatial phase of the static grating on the time-146

course of the thalamic EPSC by measuring the integral of the early (30-110 ms from stimulus onset, Fig. 147

2a-b shaded pink regions) and late (110-230 ms, Fig. 2a-b shaded gray regions) portion of the EPSC, to 148

capture the peak and the decay, respectively (Fig. 2e). Both the magnitude of the early and the late EPSC 149

fluctuated sinusoidally with a period spanning the 16 phases of the static gratings (i.e. one whole cycle). 150

Crucially, the phase relationship between the modulation of the magnitude of the early and late EPSC was 151

not fixed but varied from cell to cell (Fig. 2h top), ranging from 0º difference (i.e. the spatial phase that 152

triggers the largest early EPSC also triggers the largest late EPSC) to 87º difference (i.e. the spatial phase 153

that triggers the largest early EPSC is shifted relative to the phase that triggers the largest late EPSC). We 154


http://dx.doi.org/10.1101/244293

discovered that the phase relationship between the early and late thalamic EPSC predicted the direction 155

preference of the cell to drifting gratings. The predicted preferred direction was the direction of movement 156

in which the spatial phase eliciting the largest late EPSC preceded the spatial phase eliciting the largest 157

early EPSC. Intuitively, two consecutive EPSCs with distinct decays will summate to produce a larger 158

peak current if the slow one precedes the fast one than if the fast one precedes the slow one. In 89% of the 159

neurons (41 out of 46), the predicted preferred direction matched that observed with drifting gratings (p < 160

0.0001; Binomial test). In those few cells where the phase relationship did not predict the preferred 161

direction of drifting gratings, the phase difference was close to 0º and the cells were poorly direction 162

selective (DSI ThalF1 < 0.3; the phase difference between the early and late EPSCs correlated with the 163

DSI in response to drifting gratings [r=0.405 p=0.0053; n=46 cells; phase difference for DSI ThalF1 < 164

0.3: 10.8 ± 11.5º; n=28 cells; phase difference for DSI ThalF1 > 0.3: 33.8 ± 26.9º; n=18 cells; p = 165

0.000603, Wilcoxon rank-sum test] Fig. 2h, top). To visualize the phase dependent shift in the early and 166

late EPSC we computed the spatiotemporal receptive field that is, a heat map in which the time-course of 167

the thalamic EPSC is plotted for each spatial phase of the static grating (Fig. 2f). For each time bin, we 168

identified the phase with the largest excitation and fitted a linear function through the data. Steeper slopes 169

indicate larger phase differences between the early and late EPSC. Negative slopes predict that the 170

preferred direction is for gratings drifting in the direction of increasing spatial phase. Again, in 89% of the 171

neurons (41 out of 46), the predicted preferred direction matched that observed with drifting gratings (p < 172

0.0001; Binomial test; the slope correlated with the DSI in response to drifting gratings [r = -0.38; p= 173

0.009, n=46 cells; slope for DSI ThalF1 < 0.3: -104 ± 129º/s; n=28 cells phase difference for DSI ThalF1 174

> 0.3: -311 ± 232º/s; n=18 cells p = 0.00204 Wilcoxon rank-sum test] Fig. 2h, bottom). These results 175

demonstrate that different positions of the stimulus trigger thalamic excitation with different early and late 176

components. Importantly, the difference between the spatial positions triggering the largest early and the 177

largest late component provides the initial bias for a preferred direction of motion in L4 neurons. 178

179

Combining individual thalamic inputs with distinct spatiotemporal response properties 180

By what mechanism is the time-course of thalamic excitation modulated by the spatial phase of the 181

grating? We tested the possibility that the time-course of thalamic excitation depends on the time-course 182

of the firing of the thalamic neurons they receive input from. For this we needed to record from 183

synaptically connected thalamo-cortical pairs 35,36 and compare the time-course of firing of presynaptic 184

thalamic neurons with the time-course of decay of thalamic excitation. We performed extracellular 185


http://dx.doi.org/10.1101/244293

recordings from thalamic neurons using four shank linear probes inserted in the dorsolateral geniculate 186

nucleus (dLGN), the primary thalamic relay from the retina to V1, while simultaneously recording 187

thalamic excitation from L4 neurons during cortical silencing. The response of thalamic neurons to static 188

gratings varied from transient to sustained, consistent with previous reports37 (Extended Data Fig. 4). 189

While the firing rate of thalamic neurons was modulated by the phase of the grating, the type of response, 190

i.e. transient or sustained, remained the same irrespective of phase (Extended Data Fig. 4). 191

192

Monosynaptic connections between thalamic neurons and L4 cortical neurons were identified based on 193

latency, time course, and probability of events in the cortical neuron following spikes in individual 194

thalamic neurons during the presentation of drifting gratings under cortical silencing conditions (see 195

methods and Extended Data Fig. 5). Out of 40 L4 whole cell recordings in 24 mice and a total of 739 196

isolated thalamic units, 23 thalamo-cortical pairs, recorded in 15 mice satisfied those criteria and were 197

considered monosynaptically connected. Those 23 connected pairs consisted of 17 L4 neurons and 23 198

thalamic units, because while for most L4 neurons we found only one presynaptic thalamic unit, for two 199

L4 neurons we isolated two presynaptic units and in two L4 neuron we isolated three presynaptic units 200

(Extended Data Fig. 6). Unitary EPSCs (uEPSCs) had an average latency of 2.09 ± 0.51 ms (peak time of 201

the peri-spike time histogram (PSpTH, see methods and Extended Data Fig. 5)), an average and median 202

peak amplitude of -10.8 ± 11.6 pA and 6.4 pA, respectively (range -1.8 to -48.4 pA), and a jitter of 211 ± 203

47µs (half-width at half-max of PSpTH; n=23 connected pairs; uEPSPs had a mean and median amplitude 204

of 0.81+/-0.79 mV and 0.57 mV, respectively and ranged from 0.14 - 3.4 mV; see methods and Extended 205

Data Fig. 5). These thalamo-cortical connected pairs allowed us to compare the time course of thalamic 206

excitation recorded in the postsynaptic L4 neurons with the time course of firing of their presynaptic 207

thalamic input neurons. 208

209

The time course of firing in presynaptic thalamic units matched the duration of thalamic excitation in 210

postsynaptic L4 neurons. The example in Figure 3 shows two units, 1 and 2, isolated in the thalamus, both 211

converging on the same cortical neuron recorded in L4 (Fig. 3a). These two units responded maximally to 212

distinct phases of the static grating (Fig. 3c, bottom). Furthermore, these two units also had distinct 213

response types, unit 1 showing transient and unit 2 sustained responses to the stimulus. Spatial phases of 214

the static grating eliciting transient activity in unit 1 and little or no activity in unit 2 also elicited fast 215

decaying thalamic excitation in the L4 neuron (Fig. 3d, left). In striking contrast, spatial phases of the 216


http://dx.doi.org/10.1101/244293

grating that triggered sustained activity in unit 2 elicited slow decaying thalamic excitation in the L4 217

neuron (Fig. 3d, right). Thus, the time-course of firing of these two thalamic neurons matched the time-218

course of thalamic excitation in their postsynaptic L4 neuron. 219

220

To compare the time-course of thalamic excitation recorded in L4 neurons with the time-course of the 221

firing of their presynaptic thalamic neurons across experiments we created a heat-map of the thalamic 222

EPSC amplitudes ranked according to their duration (Fig. 3e, left; see Methods). Every row in the heat-223

map reports the amplitude of the thalamic EPSC in one L4 neuron in response to one of the 16 spatial 224

phases of the grating. Every row in the adjacent heat-map (Fig. 3e, right) shows the amplitude of the 225

simultaneously recorded peri-stimulus time histogram (PSTH) of the corresponding presynaptic thalamic 226

neuron. There was a marked and highly significant correlation between the time-course of the thalamic 227

EPSCs in L4 neurons and the time-course of the firing of their presynaptic thalamic neurons (average 228

pairwise Pearsons correlation: 0.40; n=208 paired EPSC/PSTH spatial phase responses; significantly 229

different than correlation of shuffled PSTHs; p<0.0001; see methods). Accordingly, the time-course of the 230

average PSTH of thalamic neurons firing in response to static gratings triggering slow thalamic EPSCs 231

was significantly more sustained than the time-course of the average PSTH in response to gratings 232

eliciting fast EPSCs (Fig. 3f). Thus, the time-course of thalamic excitation onto a L4 neuron depends on 233

the spatial phase and follows the time-course of the activity of its presynaptic thalamic neurons. 234

235

These results suggest that L4 neurons extract motion direction by combining thalamic inputs with distinct 236

spatial and temporal response properties. In other words, the spatiotemporal receptive field of thalamic 237

excitation of a L4 cortical neuron (e.g. Fig. 2h) should reflect the spatiotemporal activity pattern of the 238

combined population of its presynaptic thalamic neurons. Consistent with this possibility even the 239

combined spatiotemporal activity pattern of units 1 and 2 (from the example above) approximates the 240

spatiotemporal receptive field of their common L4 target neuron (Fig. 4a), with the more sustained unit 2 241

being active at lower and the transient unit 1 being active at higher spatial phases. Because recordings of 242

multiple thalamic units that are converging on a simultaneously recorded cortical neuron are rare, we 243

verified the above hypothesis by generating a “compound”, direction selective cortical neuron and its set 244

of thalamic inputs. We combined all eight thalamo-cortical connected pairs in which the cortical neuron 245

had a DSI of thalamic excitation larger than 0.3 (total of six L4 neurons: four neurons with one 246

presynaptic thalamic unit, two neurons with two presynaptic thalamic units; Fig. 4b). We aligned and 247


http://dx.doi.org/10.1101/244293

averaged the spatiotemporal receptive fields of thalamic excitation of the six cortical neurons to obtain the 248

spatiotemporal receptive field of our compound cortical neuron (Fig. 4b). The spatiotemporal receptive 249

field of the compound neuron had a negative slope, consistent with the fact that the preferred directions of 250

its six constituent neurons was for gratings drifting in the direction of increasing spatial phase. Strikingly, 251

the resulting combined spatiotemporal activity pattern of the eight presynaptic thalamic inputs also had a 252

negative slope: while the more sustained units were active at the lower spatial phases, the higher spatial 253

phases were dominated by transient thalamic units. Thus the combined spatiotemporal activity pattern of 254

the eight presynaptic thalamic units predicted the direction preference of the compound postsynaptic 255

neuron (Fig. 4b). 256

257

These data imply that for a direction selective L4 cortical neuron, the temporal distribution of activity of 258

the population of its thalamic inputs depends on the order in which the spatial phases of a grating are 259

presented. For a grating drifting in the preferred direction, they should fire together in phase to produce 260

the observed large amplitude fluctuations of thalamic excitation in the postsynaptic neuron. In the 261

opposite direction their firing should be more homogeneously distributed in time. We used our direction 262

selective compound neuron to determine the behavior of the presynaptic population of thalamic neurons to 263

gratings drifting in the preferred and opposite direction (Fig. 4c). The responses of the six L4 neurons 264

contributing to the compound neuron, and of their eight presynaptic thalamic input neurons, were 265

temporally aligned using the phase of the F1 modulation of thalamic excitation (see Methods). 266

Importantly, there was no correlation between the DSI of the firing of individual thalamic neurons and the 267

DSI of their postsynaptic L4 cortical target neuron (Extended Data Fig. 7). In the preferred direction the 268

eight thalamic neurons fired in phase to produce a strongly F1 modulated population activity (Fig. 4c). In 269

the opposite direction the same thalamic neurons fired in a more distributed manner resulting in a much 270

less pronounced F1 modulation of the population activity. Thus, our data demonstrate that the 271

convergence of transient and sustained thalamic neurons responding to distinct spatial phases produces a 272

spatiotemporal offset that confers direction preference to their target L4 cortical neuron. 273

274

A simple model in which two thalamic neurons, a transient and a sustained, preferring distinct spatial 275

phases of the stimulus, converge on the same cortical neuron, captures the essence of the above 276

observations (Fig. 5). Because of their transient and the sustained firing, the two thalamic neurons 277

generate a fast and a slow decaying EPSC onto the cortical neuron, respectively, whose amplitude 278


http://dx.doi.org/10.1101/244293

depends on the spatial phase of the stimulus. Because the sustained and the transient neuron prefer distinct 279

stimulus phases the decay of the compound EPSC in the cortical neuron changes with the phase of the 280

stimulus. The direction preference and the DSI of the cortical neuron depend on the relative offset of the 281

spatial phases of the two converging thalamic neurons (Fig. 5). 282

283

Finally, our 23 paired recordings allow us to estimate the number of thalamic neurons that contribute to 284

the response of a L4 neuron to drifting gratings (Fig. 6). We convolved the spike train of each thalamic 285

unit in response to drifting gratings with its uEPSC (Fig. 6a), integrated the resulting current and 286

compared the obtained charge with the total thalamic charge received by its postsynaptic L4 neuron in 287

response to drifting gratings (Fig. 6b). This metric hence combines the size of the uEPSC relative to the 288

total thalamic excitation and the firing rate of the presynaptic thalamic unit. A large contribution of an 289

individual thalamic unit (unitary contribution) may be due to a large uEPSC and/or a high firing rate. The 290

unitary contribution averaged 1.25 ± 1.51% (n = 23 pairs). Randomly sampling the pool of 23 unitary 291

contributions indicate that on average 80±10 thalamic neurons contribute to the excitation of a L4 neuron 292

(Fig. 6b; see Extended Data Fig. 8 for an estimate of the range). The unitary contribution was the same no 293

matter whether the stimulus drifted in the preferred or opposite direction (1.25 ± 1.54% preferred; 1.25 ± 294

1.51% opposite direction; p=0.52 Wilcoxon signed rank test), consistent with the lack of direction 295

selectivity of the thalamic charge. 296

297

Discussion 298

These results show that the initial seed for the generation of direction preference in L4 neurons of V1 299

originates from the combination of thalamic inputs with distinct spatiotemporal response properties. Thus, 300

the direction of motion is extracted from the visual scene at the earliest stage of cortical processing by 301

integrating thalamic inputs that convey information about distinct spatial locations with distinct time 302

courses. While these results reveal the mechanism for the primordial bias in direction selectivity in visual 303

cortex other intracortical mechanisms may further shape and amplify this bias. These mechanisms may 304

include a sharpening by anisotropic intracortical inhibition or excitation 10,15,16,18-21, as well as a 305

sharpening and amplification by cortical recurrent excitatory circuits 15,31,34, by the intrinsic excitable 306

properties of the neuronal membrane 11 and the active properties of dendrites 38. In contrast, the fact that 307

the isolated thalamic charge is, on average, the same irrespective of the direction of the stimulus (Fig. 308

1g,h), that equalizing the thalamic charge for the two directions of the stimulus has a minor impact on the 309


http://dx.doi.org/10.1101/244293

DSI of F1Thal (Extended Data Fig. 2) and that the direction preference of a cortical neuron does not 310

correlate with the direction bias of its thalamic inputs (Extended Data Fig. 7) shows that direction 311

selectivity in earlier, pre-cortical stages of visual processing, e.g. the retina, play a minor role in the 312

emergence of direction selectivity in L4 neurons. However, we cannot exclude the possibility that 313

anesthesia may have reduced the impact of other mechanisms in the emergence of directional selectivity. 314

We optogenetically excite inhibitory neurons to silence visual cortex and this perturbation may even 315

extend beyond the borders of V1. Most known excitatory input to L4 in V1 originate from either within 316

V1 or from the dLGN and thus we believe that most visually evoked excitation isolated during cortical 317

silencing originates from the dLGN. Other sources of visually evoked excitation from yet uncharacterized 318

cortical or sub-cortical sources, whose response to visual stimuli does not depend on V1 may have, 319

however, contributed to the recorded excitation. 320

321

The two classical models for the detection of directional motion by the nervous system, the Reichardt 12 322

and the Barlow and Levick 13 models are based on spatiotemporal offsets in excitation or in inhibition 323

relative to excitation, respectively. Both types of offsets have been proposed as a mechanism for the 324

generation of direction selectivity in visual cortex 10,15,16,18-20. Our results indicate that the spatiotemporal 325

offset of excitatory inputs relative to each other, originally proposed in the Reichardt model, more closely 326

captures the mechanism for the emergence of direction preference in mouse L4 cortical neurons. 327

328

Differences in the time-course of the visual response of thalamic neurons, categorized as transient and 329

sustained39 as well as lagged and non-lagged 23,40, have been reported in several species including cats39 33023,40, monkeys 41 and mice 37, where they were originally proposed to contribute to direction 331

selectivity23,24. Thus, the mechanism identified in the present study may be generalizable to other species. 332

Some of the dLGN neurons categorized here as sustained had a late onset reminiscent of lagged cells 23,40 333

(Extended Data Fig. 4). While our model (Fig. 5) only considers transient and sustained dLGN neurons 334

with the same onset, late onset dLGN neurons could further enhance the direction selectivity of the 335

cortical neuron. 336

337

The distribution of uEPSC amplitudes in our dataset is skewed towards smaller values and large amplitude 338

EPSCs were likely undersampled. Furthermore, submillisecond synchrony between thalamic units may 339

have led us to misclassify some thalamic units as presynaptic to the recorded L4 neuron when they in fact 340


http://dx.doi.org/10.1101/244293

only fired some of their action potential in synchrony with an actual presynaptic unit. These two factors 341

may have led to an underestimate of the average uEPSC amplitude and hence an overestimate of the 342

number of units contributing to the visual response of a L4 neuron (Extended Data Fig. 8). Nevertheless, 343

compared to previous thalamo-cortical paired recordings our estimate of uEPSP amplitude is very similar 344

to that reported in rodent somatosensory cortex in vivo 35 and in vitro 42, and somewhat larger than in cat 345

visual cortex in vivo 36. Furthermore the convergence of thalamic neurons onto L4 cortical neurons 346

estimated here (~80) resembles estimates in somatosensory cortex based on paired recordings in vivo 35 347

suggesting a common thalamo-cortical convergence across distinct sensory systems. 348

349

In conclusion, in the mammalian nervous system direction selectivity is generated de novo in at least two 350

stages of visual processing, the retina and the cortex. While, based on studies in primates and carnivores, 351

it was originally believed that retinal direction selectivity was not transmitted to the geniculo-cortical 352

pathway but instead exclusively conveyed to other subcortical structures, recent evidence in rodents 353

indicates that dLGN neurons can inherit direction selectivity from the retina 27,37,43 and that they project to 354

the cortex 28,29. However, abolishing direction selectivity in the retina does not eliminate direction 355

selectivity in cortex 30 consistent with its de novo emergence in cortex described here. Whether direction 356

selectivity computed in the cortex through the mechanisms illustrated here is combined, at some stage of 357

cortical processing, with direction selectivity inherited from the retina or whether these two channels stay 358

separated remains to be established. 359


http://dx.doi.org/10.1101/244293

Methods 360

Experiments were performed in accordance with the regulations of the Institutional Animal Care and Use 361

Committee of the University of California, San Diego and the Administrative Panel on Laboratory Animal 362

Care at Stanford University. 363

364

Mice 365

Mice were heterozygous male or female offspring of PV-Cre 44, JAX #008069) or VGat-ChR2 45, 366

JAX#014548) transgenic mice crossed with ICR white wildtype animals. Data were obtained from 41 367

VGat-ChR2 mice, and 7 PV-Cre mice. The number of cells and mice for each experiment are detailed 368

below and in the main text. 369

370

Drifting grating current clamp and voltage clamp recordings (Fig. 1a-g): 52 cells, 31 VGat-ChR2 mice. 371

Drifting grating voltage clamp recordings (Fig. 1h, Extended Data Fig. 1c-d): 61 cells in 36 VGat-ChR2 372

mice; 5 cells in 5 PV-Cre mice. 373

Drifting grating/static grating voltage clamp recordings (Fig. 2) 48 cells in 29 Vgat-ChR2 mice; 5 cells in 374

5 PV-Cre mice. 375

Drifting grating/static grating V1 voltage clamp/LGN silicon probe recordings (Fig. 3-5, Extended Data 376

Fig. 3-4): 40 dual recordings in 24 Vgat-ChR2 mice. 377

Cortical extracellular recording (Extended Data Fig. 1a): 2 PV-Cre mice; 5 Vgat-ChR2 mice. 378

Static grating LGN silicon probe recording (Extended Data Fig. 4): 24 recordings in 24 Vgat-ChR2 mice. 379

380

Solutions 381

Artificial Cerebrospinal Fluid (ACSF): 140 mM NaCl, 5 mM KCl, 10 mM D-glucose, 10 mM HEPES, 2 382

mM CaCl2, 2 mM MgSO4, pH 7.4. 383

Potassium-based intracellular solution: 135 mM potassium gluconate, 8 mM NaCl, 10 mM HEPES, 4 mM 384

Mg-ATP, 0.3 mM Na-GTP, 0.3 mM EGTA, pH 7.4. 385

Cesium-based intracellular solution: 125 mM cesium methanesulphonate, 8 mM NaCl, 10 mM HEPES, 4 386

mM Mg-ATP, 0.3 mM Na-GTP, 0.3 mM EGTA, 2 mM QX-314, pH 7.4 387

388

389

390


http://dx.doi.org/10.1101/244293

ChR2 expression in V1 inhibitory interneurons 391

AAV1-Flex-ChR2-tdTom (University of Pennsylvania Vector Core) was injected in the visual cortex of 392

neonatal (P0-1) PV-Cre mice to achieve ChR2 expression in PV+ inhibitory interneurons (PV-ChR2 393

mice) as previously described 31 VGAT-ChR2 mice express ChR2 in cortical inhibitory interneuron 394

populations therefore no viral injections were required. 395

396

Surgery for recording experiments 397

Adult mice (5-12 weeks) were anesthetized with a combination of urethane (1.5 g/kg, IP), chlorprothixene 398

(2-4 mg/kg, IP), and light isoflurane (0.5% in O2). A drop of silicon oil was applied to the eyes. The scalp 399

was removed, the skull cleaned and a metal head-fixation bar was affixed to the skull using dental acrylic. 400

A 1-2 mm diameter craniotomy was performed over V1 in one hemisphere (1 mm anterior of the 401

lambdoid suture, 2.5 mm lateral to the midline). In simultaneous V1 and LGN recording experiments, a 402

second narrow elongated craniotomy was performed (spanning 2-3 mm posterior of bregma, 3.4 mm 403

lateral to the midline) for insertion of the silicon probe array into the LGN. In both craniotomies the dura 404

was removed. Recording began shortly after completion of surgery. 405

406

V1 whole cell recording 407

Whole cell recordings were made using the blind patch technique 46 at a depth of 300-550 µm 408

corresponding to layer 4. Patch pipettes with 4-6 MΩ resistance were pulled from borosilicate glass and 409

filled with intracellular solution. Potassium-based intracellular solution was used in 54 cells (52 recorded 410

in both current clamp and voltage clamp, 2 recorded in voltage clamp only). 12 cells were recorded with 411

cesium-based intracellular solution (voltage clamp only). Prior to insertion of patch pipettes, a drop of 412

low-melting point agarose (1.5% in ACSF) was applied to the brain surface to reduce movement. Pipettes 413

were rapidly inserted into the visual cortex while applying high positive pressure (2.5 psi). Pressure was 414

reduced to 0.5 PSI upon reaching a depth of 200-300 µm and the pipette was advanced in 2 µm steps 415

while monitoring the pipette resistance. When a sudden increase in resistance was encountered, positive 416

pressure was released and light suction was applied to achieve a gigaohm seal. Brief pulses of suction 417

were applied to break the seal and achieve whole-cell configuration. Series resistance was 20-50 MΩ. 418

Signals were amplified (Multiclamp 700B, Molecular Devices) and digitized at 10 kHz (DigiData, 419

Molecular Devices) or 31.25 kHz (PCIe-6259, National Instruments). Membrane potential and spiking 420


http://dx.doi.org/10.1101/244293

responses were recorded in current clamp configuration. Excitatory currents were recorded in voltage 421

clamp at -70 mV near the reversal potential of inhibition. 422

423

LGN unit recording 424

Prior to initiating V1 whole-cell recordings, a 4-shank 32 channel silicon probe (Buzsaki32, NeuroNexus) 425

was inserted into the LGN craniotomy (coordinates listed above) with the shanks distributed along the 426

anterior/posterior axis. The probe was inserted at a 55 degree angle above horizontal in the coronal plane 427

such that it advanced along the lateral to medial and dorsal to ventral directions. The probe was advanced 428

in 5 µm steps until robust visual responses were observed in the multiunit activity (2.5-2.8 mm distance). 429

The shanks typically spanned the anterior/posterior extent of the dLGN from 2-3 mm posterior of bregma. 430

Retinotopy as assessed with coarse receptive field mapping of multiunit activity was consistent with 431

Piscopo et al, 2013. Signals were amplified (Model 4000, AM Systems) and digitized at 31.25 kHz (PCIe-432

6259, National Instruments). For technical reasons, the top-most site on each shank was not recorded. The 433

probe was allowed to settle for 30-60 minutes before collecting data. At the end of the experiment, the 434

mouse was sacrificed under deep anesthesia and the brain was fixed in 4% paraformaldehyde. The tissue 435

was sectioned for post-hoc verification of the recording sites. 436

437

Retinotopic alignment of V1 and LGN recordings 438

To improve the chances of recording from synaptically connected LGN and V1 neurons, coarse receptive 439

field maps of the LGN multiunit activity were generated and the corresponding retinotopic region of V1 440

was identified by mapping the receptive field of the LFP signal at various locations in the V1 craniotomy. 441

The number of V1 locations that were sampled before finding the corresponding cortical location was 442

approximately 1-5 although this was not precisely documented during the experiments. Subsequent V1 443

whole-cell recordings were targeted to this region of V1. V1 LFP was recorded using a patch pipette filled 444

with ACSF inserted to a depth of 200-400 µm. 445

446

Cortical silencing 447

Visual cortex was silenced by illuminating the V1 craniotomy with a 1mm fiber optic coupled to a blue 448

LED (470 nm; 20 mW total output, Doric) positioned several mm above the craniotomy or through the 449

objective (20x) of a fluorescence microscope with a blue LED (470 nm, 2.3 mW total output, Thorlabs) 450

coupled to the excitation port. The LED turned on 650 ms prior to the onset of a visual stimulus trial and 451


http://dx.doi.org/10.1101/244293

lasted throughout the duration of the visual stimulus. Trials with cortical silencing were interleaved with 452

trials without illumination in which cortical activity was intact. To validate the effectiveness of cortical 453

silencing, spiking responses of V1 neurons in response to drifting gratings were recorded using loose-454

patch or silicon probes (Buzsaki32 or A1x32-Edge-5mm-20-177, NeuroNexus) during the silencing 455

protocol. Cortical silencing in both PV-ChR2 and VGat-ChR2 mice suppressed nearly all spiking of non-456

narrow spiking neurons in V1 (~99% suppression, Extended Data Fig 1b). Some of the experiments 457

illustrated in Extended Data Fig 1b appear in previous publications and were also performed by one of the 458

authors (PV-ChR2: all loose patch recordings are from Fig.1b in 31;VGat-ChR2: 64 out of 138 units are 459

from Supplemental Fig. 7g-h in 47). 460

461

Visual stimuli 462

Visual stimuli were presented on an LCD monitor (75 cd/m2 mean luminance, gamma corrected) to the 463

eye contralateral to the hemisphere in which recordings were performed. 464

465

Drifting gratings 466

Drifting gratings stimuli were full-field, full contrast drifting bar gratings (0.04 cyc/deg spatial frequency, 467

2Hz temporal frequency). For determining the preferred orientation, 12 different drifting grating stimuli 468

were presented consisting of 6 evenly-spaced orientations (30 degree increment) drifting in one of two 469

opposite directions along the axis perpendicular to the grating bars. Drifting gratings were presented for 470

2.3 s and were preceded and followed by a mean luminance gray screen. Presentation of a single grating 471

was considered one visual stimulus trial (see cortical silencing methods). In 52/66 recordings, the 472

preferred orientation and direction of the cell were determined in current clamp using the full set of 12 473

orientations/directions followed by recording in the voltage clamp configuration using drifting grating 474

stimuli restricted to the 2 opposite directions at the preferred orientation. In these cells the orientation of 475

the stimulus may not exactly match the preferred orientation of thalamic excitation. In the remaining 14 476

cells the recording was started in the voltage clamp configuration and the preferred orientation was 477

determined in voltage clamp during cortical silencing by presenting the full set of 12 478

orientations/directions. The orientation and/or direction of the grating stimuli were presented in random 479

order. 480

481

Static gratings 482


http://dx.doi.org/10.1101/244293

Static grating stimuli were full-field, full contrast bar gratings (0.04 cyc/deg spatial frequency) of the 483

preferred orientation of the cortical neuron. 16 evenly-spaced spatial phases were presented (22.5 deg 484

increment equal to 1/16 of a cycle). A series of 5 static gratings of randomly chosen spatial phase were 485

presented sequentially. Each static grating was presented for 0.25 s and followed by a mean luminance 486

gray screen for 0.25 s. Each 5-grating sequence was considered one visual stimulation trial (see cortical 487

silencing methods) and was preceded and followed by a mean luminance gray screen. 488

489

Direction selectivity index 490

Direction selectivity index (DSI) for responses to two drifting grating stimuli moving in opposite 491

directions was calculated as: 492

(RespPref - RespNull)/RespPrefDir 493

Where RespPref and RespNull are the size of the responses to the directions that gave the larger and 494

smaller responses, respectively. This limits the range of DSI from 0-1. The responses used to calculate the 495

DSI were the spike rate, the F1 amplitude modulation of Vm, the F1 amplitude modulation of the thalamic 496

excitation and the thalamic excitatory charge (see Drifting grating analysis below). When the DSI was 497

compared for two different response parameters within the same cell e.g. DSI of the F1 modulation of 498

thalamic excitation versus the DSI of the F1 modulation of Vm (Fig. 1g middle), the DSI was defined 499

relative to the preferred direction of one of the parameters, the reference parameter. If the preferred 500

direction of the two parameters were different, the DSI of the non-reference parameter was multiplied by -501

1. In the example above the reference parameter was Vm. Hence, the DSI of the reference parameter is 502

always positive (range 0 to +1) but the DSI of the non-reference parameter can be negative or positive 503

(range -1 to +1) with negative and positive values indicating that the two parameters prefer opposite or the 504

same direction, respectively. The absolute value of the DSI indicates the degree of selectivity. 505

506

Drifting grating analysis 507

Drifting grating responses were evaluated in a time window from 0.3 s after stimulus onset to the end of 508

the stimulus (2 s total or 4 complete cycles). Spikes in current clamp recordings were detected by 509

identifying time points where the membrane potential exceeded -15 mV. The time of a spike was defined 510

as the time of the peak depolarization in a 1.5 ms window following each threshold crossing. Prior to 511

calculating the DSI of F1 amplitude for subthreshold membrane potential responses (Fig. 1g), all spikes 512

were removed from current clamp recordings by replacing the time window from 1 ms before to 2 ms 513


http://dx.doi.org/10.1101/244293

after the spike time with linear interpolation. In voltage clamp recordings, the holding current in a 0.4 s 514

window prior to the onset of visual stimulus was subtracted for each trial, computed from the bottom 5th 515

percentile of the distribution of current values, which should include the periods with the least amount of 516

spontaneous excitatory activity. The DSI based on spikes was calculated from the average spike rate 517

during the response window. F1 amplitude, for both Vm and thalamic excitation, was derived from the 518

amplitude of a sinusoidal fit to the cycle average of drifting grating responses. The period of the cycle 519

average was 0.5 s, matching the temporal frequency of the drifting grating. The excitatory charge was 520

calculated from the integral of drifting grating excitatory current responses. 521

522

Static grating analysis 523

Analysis was restricted to responses collected under cortical silencing conditions (thalamic excitation). 524

The thalamic excitation in response to each spatial phase of the static grating were averaged together. The 525

average value of the current from 0-24ms after stimulus onset was subtracted. This time window was prior 526

to any observable visual response. 527

528

Static grating summation 529

The algebraic sum of thalamic EPSCs evoked by each of the 16 phases of the static gratings was 530

computed for each cortical neuron. Before computing the sum, the 16 EPSCs were staggered in time 531

relative to each other in proportion to the phase offset between the phase of the static grating used to 532

evoke each of the EPSCs and a fixed reference phase. The absolute time separation was 31.25 ms for 533

every 22.5 degrees of phase offset such that a full cycle (16*31.25 ms) corresponds to 500 ms, i.e. a full 534

cycle of a 2Hz drifting grating. We computed the sum for positive and negative time separations (+ or - 535

31.25ms) corresponding to the temporal sequence of phases of a grating drifting in one or the opposite 536

direction. The DSI and F1 amplitudes of the resulting algebraic summations were analyzed as described 537

above. 538

539

Static grating spatiotemporal analysis 540

Spatiotemporal analysis was restricted to cortical neurons whose summed static grating responses 541

predicted the preferred direction in response to drifting gratings (DSI summed static relative to drifting > -542

0.1). The spatial phases were ordered such that the relationship between the sequence of spatial phases and 543


http://dx.doi.org/10.1101/244293

the preferred direction in response to drifting gratings was the same for each neuron. Specifically, the 544

preferred direction is in the upward direction for static grating heatmaps (Fig. 2f and 4a,b). 545

546

Early/Late spatiotemporal analysis 547

Early and late thalamic excitation was defined as the excitatory charge from 30-110 ms and 110-230 ms 548

after the onset of the static grating stimulus, respectively. The preferred spatial phase of early and late 549

thalamic excitation was calculated from the vector average across all spatial phases of early or late 550

thalamic excitation, respectively. For population average of early and late thalamic excitation (Fig. 2e 551

right), the early and late thalamic excitation were upsampled in the spatial phase dimension by 4x and 552

shifted along the spatial phase axis so that the preferred spatial phase of early thalamic excitation occurred 553

at a spatial phase of 180 degrees. 554

555

Spatiotemporal slope analysis 556

The preferred spatial phase was calculated for 5 time bins (40ms width starting from 30ms after stimulus 557

onset). A linear fit to these 5 preferred spatial phase values was performed to calculate the spatiotemporal 558

slope. For population averages of static grating heatmaps (Fig. 2f right, Fig. 4b), each cortical neuron’s 559

static grating responses were normalized by the peak response across all spatial phases, upsampled in the 560

spatial phase dimension by 4x and shifted along the spatial phase axis so that the preferred spatial phase of 561

the earliest time bin (30-70ms) occurred at a spatial phase of 180 degrees. 562

563

Thalamic unit spike sorting 564

The UltraMegaSort 48,49 spike sorting software was used to detect, cluster, and assign spike waveforms 565

into single units as previously described 50. Spike waveforms on the 7 recorded sites of an individual 566

shank were clustered using a k-means algorithm followed by manual assignment of clusters with distinct 567

waveform profiles into single units. 568

569

Identification of monosynaptically-connected thalamocortical pairs 570

For each simultaneously recorded thalamic unit and cortical neuron, a brief segment of the excitatory 571

current recorded in the cortical neuron during drifting grating visual stimulation with cortical silencing 572

around the time of each thalamic spike was extracted (25 ms before and after the spike). We looked for the 573

presence of a fast inward (negative) deflection in thalamic excitation occurring 1-4 ms after the thalamic 574


http://dx.doi.org/10.1101/244293

spike. We first calculated the first derivative of thalamic excitation of the spike-triggered sweeps 575

(dtThExc). The spike-triggered average of dtThExc was computed and z-scored. Thalamic units 576

containing a z-score peak exceeding -5 in the time window of 1-4 ms after the thalamic spike were 577

considered candidate presynaptic units. To ensure that the negative peak in the spike-triggered average of 578

dtThExc was due to an actual increase in the probability of fast inward events occurring in the 1-4 window 579

following the spike, we detected the occurrence of such events in the spike-triggered dtThExc sweeps. 580

Events were detected as decreasing threshold crossings of dtThExc where the threshold was set as -581

1*(standard deviation of all spike-triggered dtThExc sweeps for each thalamic unit). A peri-spike time 582

histogram (PSpTH) was assembled from these event detections (0.2 ms bin size) and z-scored. Candidate 583

presynaptic units with a z-score peak exceeding 3.5 in the time window 1-4 ms after the thalamic spike 584

were considered monosynaptically connected to the cortical neuron. Differentiation was performed on 585

data that was resampled at 10 kHz using consecutive samples. Following differentiation, sweeps were 586

smoothed (0.3 ms running average) to generate dtThExc. Z-scores of the spike-triggered average of 587

dtThExc and PSpTH were computed by first subtracting a smoothed version (3 ms running median to 588

remove fluctuations slower than several ms) and then z-scoring using the average and standard deviation 589

of time points 0-5 ms prior to spike. 590

591

Latency of monosynaptic response was defined as the time of the peak in the z-scored PSpTH relative to 592

the spike. Jitter was defined as the half-width at half-max of this peak. The unitary EPSC (uEPSC) was 593

derived from the spike-triggered thalamic excitation of each monosynaptically-connected thalamic unit. 594

While the identified monosynaptic connections had sub-millisecond jitter, the uEPSC may ride on top of a 595

slower envelope of thalamic excitation driven by the response of other synaptically connected inputs to 596

the visual stimulus. This component was estimated by shifting the trial number of thalamic responses by 1 597

trial relative to that of thalamic excitation for each direction of the drifting grating and computing a spike-598

triggered average based on the trial-shifted data35. The trial-shifted spike-triggered average was subtracted 599

from the uEPSC (shift-subtracted uEPSC) for amplitude and contribution (see “Thalamic unit 600

contribution” below) analyses. The baseline from 0-1 ms after the thalamic spike was also subtracted. 601

Unsubtracted uEPSCs are shown in the main figures. Shift-subtracted uEPSCs are shown in Extended 602

Data Fig. 3. The amplitude of monosynaptic connections was defined as the most negative (inward) value 603

of the shift-subtracted uEPSC from 0-3 ms after the onset of monosynaptic response (i.e., the time of 604

thalamic spike + latency for each pair) minus the average value in a 0.2 ms window just prior to the onset. 605


http://dx.doi.org/10.1101/244293

Amplitude of uEPSP and spontaneous uEPSC were determined in the same manner using the shift-606

subtracted spike-triggered average of Vm during drifting gratings under control conditions or the spike-607

triggered average of thalamic excitation during the 500ms prior to visual stimulation during cortical 608

silencing, respectively. Spontaneous uEPSCs were only characterized for pairs with at least 30 609

spontaneous spikes. One thalamic unit that passed the criteria for monosynaptic connection did not exhibit 610

a clear spontaneous uEPSC in the simultaneously recorded cortical neuron despite firing a sufficient 611

number of spontaneous spikes and hence was not considered to be monosynaptically connected. In the 4 612

experiments in which multiple presynaptic thalamic neurons were recorded, the spike sorting of 613

monosynaptically connected thalamic units was verified using KiloSort software 51. For one of these 614

experiments, this yielded an additional thalamic unit which passed the criteria for monosynaptic 615

connection. 616

617

In all subsequent connected pair analyses, each pair was treated independently without regard for whether 618

or not the postsynaptic neuron was common to other connected pairs. Thus the postsynaptic responses of a 619

cortical neuron may be represented multiple times if more than one pre-synaptic thalamic unit to that 620

cortical neuron was identified (e.g. the pairs in Fig. 3a). 621

622

623

Connected pair static grating analysis 624

Out of 23 pairs, 20 of the pre-synaptic thalamic units had spiking responses to static gratings. For each 625

pre-synaptic thalamic unit, peri-stimulus time histograms (PSTH) of the spiking were constructed for 626

responses (10 ms binning, upsampled to 10 kHz, smoothed by 20 ms running average) to each spatial 627

phase of the static grating under cortical silencing conditions. 628

629

Duration of thalamic excitation response to static gratings 630

The duration was defined as the time point after stimulus onset at which the excitatory charge reached 631

90% of its maximum value. In Figure 3, analyses were restricted to those static grating responses in 632

connected pairs in which both the pre-synaptic LGN firing and the postsynaptic thalamic excitation were 633

at least 10% of that elicited by the spatial phase that gave the largest response. Each response was peak 634

normalized. 635

636


http://dx.doi.org/10.1101/244293

To test statistical significance of the average pairwise Pearsons correlation between thalamic excitation 637

(EPSC) and thalamic spiking (PSTH) spatial phase responses, the PSTH responses across all pairs and 638

spatial phases were shuffled relative to their corresponding EPSC response so that each EPSC response 639

was reassigned to a random PSTH response. For each shuffle, the average pairwise correlation was 640

calculated, and this was procedure was repeated for 10,000 shuffles. The average pairwise correlation of 641

the real data was compared to the distribution of shuffled average pairwise correlations. None of the 642

shuffled average pairwise correlations exceeded that of the real data. Use of the Spearman correlation 643

produced similar results. 644

645

Compound cortical neuron: static grating 646

For each pair, the static grating heatmaps of the presynaptic PSTH and the postsynaptic thalamic 647

excitation were shifted along the spatial phase axis so that the preferred spatial phase of the earliest time 648

bin (30-70ms) of thalamic excitation occurred at a spatial phase of 180 degrees and the sequence of spatial 649

phases was ordered so that the preferred direction of thalamic excitation in response to drifting gratings 650

was in the upward direction. Heatmaps were normalized by the peak response across all spatial phases, 651

upsampled in the spatial phase dimension by 4x, and averaged across all pairs. 652

653

Compound cortical neuron: drifting grating 654

Cycle-average responses to drifting gratings of the preferred and non-preferred direction of thalamic 655

excitation of all pairs were averaged together. Before averaging, the EPSC and PSTH cycle-average for 656

each pair was shifted in time by the same amount so that the F1 peak of thalamic excitation occurred at 657

250 ms and peak normalized. 658

659

Thalamic unit contribution 660

For each pair, the excitatory current contributed by the thalamic unit was calculated by convolving its 661

shift-subtracted uEPSC from 0-15 ms after the thalamic spike with its spike train during drifting gratings 662

and averaging across trials. The shift-subtracted uEPSC was truncated to the time point at which it 663

returned to baseline if this occurred before 15 ms. The contributed charge was the integral of the trial-664

averaged convolution across the stimulus duration. Dividing the contributed charge by the thalamic 665

excitatory charge evoked by the same drifting grating stimulus resulted in the fractional contribution of 666

the thalamic unit. 667


http://dx.doi.org/10.1101/244293

668

Statistical analysis 669

Statistical analyses were done in IgorPro. No statistical methods were used to pre-determine sample sizes, 670

but our sample sizes were similar to those reported in previous publications in the field. All data are 671

presented as mean ± s.d. Normality of the data were not tested and nonparametric two-sided Wilcoxon 672

rank-sum or Wilcoxon signed-rank tests were used for unpaired or paired tests, respectively. Fraction of 673

cells with matching direction preference was compared to a chance value of 0.5 using two-tailed binomial 674

test. Experiments and analysis were not blinded. 675

676

Code availability 677

Custom code used in this study are available from the corresponding author upon reasonable request. 678

679

Data availability 680

The datasets generated during and/or analyzed during the current study are available from the 681

corresponding author upon reasonable request. 682


http://dx.doi.org/10.1101/244293

683References 6841 Hubel, D. H. & Wiesel, T. N. Receptive fields, binocular interaction and functional architecture in 685

the cat's visual cortex. J Physiol 160, 106-154 (1962). 686

2 Hubel, D. H. & Wiesel, T. N. Receptive fields of single neurones in the cat's striate cortex. J 687Physiol 148, 574-591 (1959). 688

3 Movshon, J. A., Thompson, I. D. & Tolhurst, D. J. Spatial summation in the receptive fields of 689simple cells in the cat's striate cortex. J Physiol 283, 53-77 (1978). 690

4 McLean, J. & Palmer, L. A. Contribution of linear spatiotemporal receptive field structure to 691velocity selectivity of simple cells in area 17 of cat. Vision Res 29, 675-679 (1989). 692

5 Albrecht, D. G. & Geisler, W. S. Motion selectivity and the contrast-response function of simple 693cells in the visual cortex. Visual neuroscience 7, 531-546 (1991). 694

6 Reid, R. C., Soodak, R. E. & Shapley, R. M. Directional selectivity and spatiotemporal structure of 695receptive fields of simple cells in cat striate cortex. J Neurophysiol 66, 505-529 (1991). 696

7 Saul, A. B. & Humphrey, A. L. Evidence of input from lagged cells in the lateral geniculate 697nucleus to simple cells in cortical area 17 of the cat. J Neurophysiol 68, 1190-1208 (1992). 698

8 DeAngelis, G. C., Ohzawa, I. & Freeman, R. D. Spatiotemporal organization of simple-cell 699receptive fields in the cat's striate cortex. II. Linearity of temporal and spatial summation. J 700Neurophysiol 69, 1118-1135 (1993). 701

9 Jagadeesh, B., Wheat, H. S., Kontsevich, L. L., Tyler, C. W. & Ferster, D. Direction selectivity of 702synaptic potentials in simple cells of the cat visual cortex. J Neurophysiol 78, 2772-2789 (1997). 703

10 Livingstone, M. S. Mechanisms of direction selectivity in macaque V1. Neuron 20, 509-526 704(1998). 705

11 Priebe, N. J. & Ferster, D. Direction selectivity of excitation and inhibition in simple cells of the 706cat primary visual cortex. Neuron 45, 133-145, doi:10.1016/j.neuron.2004.12.024 (2005). 707

12 Reichardt, W. Autocorrelation, a principle for the evaluation of sensory information by the central 708nervous system. 303-317 (M.I.T. Press; John Wiley, 1961). 709

13 Barlow, H. B. & Levick, W. R. The mechanism of directionally selective units in rabbit's retina. J 710Physiol 178, 477-504 (1965). 711

14 Adelson, E. H. & Bergen, J. R. Spatiotemporal energy models for the perception of motion. 712Journal of the Optical Society of America. A, Optics and image science 2, 284-299 (1985). 713

15 Suarez, H., Koch, C. & Douglas, R. Modeling direction selectivity of simple cells in striate visual 714cortex within the framework of the canonical microcircuit. J Neurosci 15, 6700-6719 (1995). 715

16 Maex, R. & Orban, G. A. Model circuit of spiking neurons generating directional selectivity in 716simple cells. J Neurophysiol 75, 1515-1545 (1996). 717

17 Watson, A. B. & Ahumada, A. J. Model of human visual-motion sensing. JOSA A 2, 322-342 718(1985). 719

18 Goodwin, A., Henry, G. & Bishop, P. Direction selectivity of simple striate cells: properties and 720mechanism. Journal of Neurophysiology 38, 1500-1523 (1975). 721


http://dx.doi.org/10.1101/244293

19 Emerson, R. C. & Gerstein, G. L. Simple striate neurons in the cat. II. Mechanisms underlying 722directional asymmetry and directional selectivity. Journal of Neurophysiology 40, 136-155 (1977). 723

20 Ganz, L. & Felder, R. Mechanism of directional selectivity in simple neurons of the cat's visual 724cortex analyzed with stationary flash sequences. Journal of Neurophysiology 51, 294-324 (1984). 725

21 Li, Y.-t., Liu, B.-h., Chou, X.-l., Zhang, L. I. & Tao, H. W. Strengthening of direction selectivity 726by broadly tuned and spatiotemporally slightly offset inhibition in mouse visual cortex. Cerebral 727Cortex 25, 2466-2477 (2014). 728

22 Sillito, A. Inhibitory processes underlying the directional specificity of simple, complex and 729hypercomplex cells in the cat's visual cortex. The Journal of Physiology 271, 699-720 (1977). 730

23 Saul, A. B. & Humphrey, A. L. Spatial and temporal response properties of lagged and nonlagged 731cells in cat lateral geniculate nucleus. J Neurophysiol 64, 206-224 (1990). 732

24 Alonso, J.-M., Usrey, W. M. & Reid, R. C. Rules of connectivity between geniculate cells and 733simple cells in cat primary visual cortex. Journal of Neuroscience 21, 4002-4015 (2001). 734

25 Stanley, G. B. et al. Visual orientation and directional selectivity through thalamic synchrony. 735Journal of Neuroscience 32, 9073-9088 (2012). 736

26 Ferster, D., Chung, S. & Wheat, H. Orientation selectivity of thalamic input to simple cells of cat 737visual cortex. Nature 380, 249 (1996). 738

27 Marshel, J. H., Kaye, A. P., Nauhaus, I. & Callaway, E. M. Anterior-posterior direction opponency 739in the superficial mouse lateral geniculate nucleus. Neuron 76, 713-720 (2012). 740

28 Cruz-Martín, A. et al. A dedicated circuit links direction-selective retinal ganglion cells to the 741primary visual cortex. Nature 507, 358-361 (2014). 742

29 Sun, W., Tan, Z., Mensh, B. D. & Ji, N. Thalamus provides layer 4 of primary visual cortex with 743orientation-and direction-tuned inputs. Nature neuroscience 19, 308-315 (2016). 744

30 Hillier, D. et al. Causal evidence for retina-dependent and-independent visual motion 745computations in mouse cortex. Nature Neuroscience (2017). 746

31 Lien, A. D. & Scanziani, M. Tuned thalamic excitation is amplified by visual cortical circuits. Nat 747Neurosci 16, 1315-1323, doi:10.1038/nn.3488 (2013). 748

32 Niell, C. M. & Stryker, M. P. Highly selective receptive fields in mouse visual cortex. J Neurosci 74928, 7520-7536 (2008). 750

33 Boyden, E. S., Zhang, F., Bamberg, E., Nagel, G. & Deisseroth, K. Millisecond-timescale, 751genetically targeted optical control of neural activity. Nat Neurosci 8, 1263-1268 (2005). 752

34 Li, Y.-t., Ibrahim, L. A., Liu, B.-h., Zhang, L. I. & Tao, H. W. Linear transformation of 753thalamocortical input by intracortical excitation. Nature neuroscience 16, 1324-1330 (2013). 754

35 Bruno, R. M. & Sakmann, B. Cortex is driven by weak but synchronously active thalamocortical 755synapses. Science 312, 1622-1627 (2006). 756

36 Sedigh-Sarvestani, M. et al. Intracellular, In Vivo, Dynamics of Thalamocortical Synapses in 757Visual Cortex. Journal of Neuroscience 37, 5250-5262 (2017). 758

37 Piscopo, D. M., El-Danaf, R. N., Huberman, A. D. & Niell, C. M. Diverse visual features encoded 759in mouse lateral geniculate nucleus. Journal of Neuroscience 33, 4642-4656 (2013). 760


http://dx.doi.org/10.1101/244293

38 Smith, S. L., Smith, I. T., Branco, T. & Hausser, M. Dendritic spikes enhance stimulus selectivity 761in cortical neurons in vivo. Nature 503, 115-120, doi:10.1038/nature12600 (2013). 762

39 Cleland, B., Dubin, M. & Levick, W. Sustained and transient neurones in the cat's retina and 763lateral geniculate nucleus. The Journal of Physiology 217, 473 (1971). 764

40 Mastronarde, D. N. Two classes of single-input X-cells in cat lateral geniculate nucleus. I. 765Receptive-field properties and classification of cells. J Neurophysiol 57, 357-380 (1987). 766

41 Marrocco, R. T. Sustained and transient cells in monkey lateral geniculate nucleus: conduction 767velocites and response properties. Journal of Neurophysiology 39, 340-353 (1976). 768

42 Hu, H. & Agmon, A. Differential Excitation of Distally versus Proximally Targeting Cortical 769Interneurons by Unitary Thalamocortical Bursts. J Neurosci 36, 6906-6916, 770doi:10.1523/JNEUROSCI.0739-16.2016 (2016). 771

43 Zhao, X., Chen, H., Liu, X. & Cang, J. Orientation-selective responses in the mouse lateral 772geniculate nucleus. Journal of Neuroscience 33, 12751-12763 (2013). 773

774Method References 775 776

44 Hippenmeyer, S. et al. A developmental switch in the response of DRG neurons to ETS 777transcription factor signaling. PLoS biology 3, e159, doi:10.1371/journal.pbio.0030159 (2005). 778

45 Zhao, S. et al. Cell type-specific channelrhodopsin-2 transgenic mice for optogenetic dissection of 779neural circuitry function. Nat Methods 8, 745-752 (2011). 780

46 Margrie, T. W., Brecht, M. & Sakmann, B. In vivo, low-resistance, whole-cell recordings from 781neurons in the anaesthetized and awake mammalian brain. Pflugers Archiv : European journal of 782physiology 444, 491-498, doi:10.1007/s00424-002-0831-z (2002). 783

47 Saleem, A. B. et al. Subcortical Source and Modulation of the Narrowband Gamma Oscillation in 784Mouse Visual Cortex. Neuron 93, 315-322, doi:10.1016/j.neuron.2016.12.028 (2017). 785

48 Fee, M. S., Mitra, P. P. & Kleinfeld, D. Automatic sorting of multiple unit neuronal signals in the 786presence of anisotropic and non-Gaussian variability. J Neurosci Methods 69, 175-188 (1996). 787

49 Hill, D. N., Mehta, S. B. & Kleinfeld, D. Quality metrics to accompany spike sorting of 788extracellular signals. J Neurosci 31, 8699-8705. 789

50 Olsen, S. R., Bortone, D. S., Adesnik, H. & Scanziani, M. Gain control by layer six in cortical 790circuits of vision. Nature 483, 47-52 (2012). 791

51 Pachitariu, M., Steinmetz, N. A., Kadir, S. N., Carandini, M. & Harris, K. D. Fast and accurate 792spike sorting of high-channel count probes with KiloSort. Adv. Neural Inf. Process. Syst. 4448–7934456 (2016). 794

795

796

797

798


http://dx.doi.org/10.1101/244293

799

800

801

802Acknowledgments 803

We thank J. Evora for help with genotyping and mouse husbandry, S. Hestrin for allowing us to perform 804

some of the experiments in his lab, R. Beltramo for helping with extracellular recordings, J.S. Isaacson 805

and B.L. Bloodgood for comments on the manuscript and the members of the Scanziani and Isaacson 806

laboratories for helpful discussions of this project. This project was supported by the Gatsby Charitable 807

Foundation and the Howard Hughes Medical Institute. 808

809

Author contributions 810

A.D.L. and M.S. designed the study. A.D.L. conducted all experiments and analysis. A.D.L. and M.S. 811

wrote the paper. 812


http://dx.doi.org/10.1101/244293

Time (s) Time (s) Time (s) Time (s)

50

0

-200

-100

0

0.80.40.0DSI Vm F1

PrefDir

Non-PrefDir

0.80.40.0DSI Vm F1

-1.0 -1.0 -1.0

-0.5 -0.5 -0.5

0.0 0.0 0.0

0.5 0.5 0.5

1.0 1.0 1.0

0.80.40.0DSI Spike

DS

I Vm

F1

DS

I Tha

l. F1

DS

I Tha

l. F1

DS

I Tha

l. C

harg

e

50

0

-100

0

F1Control

ChargeThalamic EPSC

2.01.00.0 2.01.00.0 2.01.00.0 2.01.00.0

-60 -60-40 -40

-20 -200 0

270° 90° 270° 90°

Th C

harg

e (N

orm

)

1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0

a b

c d

e f

g h

2/34

56

2/34

56

Cortex

Thalamus

LED Whole Cell rec.

Whole Cell rec.

ChR

ChR

ChR

Figure 1

30 pA250 ms

40 pA250 ms

spik

es p

er s

curr

ent (

pA)

Vm

(mV

)

CortexCortex

Model 1 Model 2

Thalamus

DSDS

Thalamus

Δt

ΔtCortex

Model 3

DS

Thalamus

Retina

DS

DS


http://dx.doi.org/10.1101/244293

Figure 1 813Amplitude modulation of thalamic excitation is direction selective 814a. Two alternative models for the emergence of direction selectivity. Left: Model 1: The spatiotemporal 815

offset of synaptic inputs onto a direction selective cortical cell (DS) is generated by intra-cortical 816circuits. Right: Model 2: The spatiotemporal offset is generated by the convergence of thalamic inputs 817with distinct spatial and temporal properties. 818

b. Experimental configuration to isolate thalamic excitation: whole cell recordings in layer 4 of 819anesthetized mice during visual stimulation while photo-activating Channelrhodopsin expressing 820inhibitory neurons to silence cortex. 821

c. Top: Current clamp recordings of a direction selective neuron in response to gratings drifting in the 822preferred (left) and non-preferred (right) direction (4 superimposed sweeps). Vm: membrane potential. 823Bottom: Peri-stimulus time histogram. 824

d. As in (c) but for a non-direction selective neuron. 825e. Voltage clamp recording (V holding: -70 mV) of the neuron illustrated in (c) under control conditions 826

(gray traces) and during cortical silencing (black traces) to isolate thalamic excitation. Inset is a cycle 827average and the red trace a sinusoidal fit. The arrow (F1) shows the amplitude of the modulation. The 828blue shaded area is the current integral used to estimate the thalamic excitatory charge. 829

f. As (e) but for neuron illustrated in (d). 830g. Summary plots for 52 recordings (31 mice). The scatter plots show the relationship between the 831

direction selectivity index (DSI) of the spikes (DSI spike), the F1 modulation of the membrane potential 832(DSI Vm F1), the F1 modulation of thalamic excitation (DSI Thal. F1), and the thalamic charge (DSI 833Thal. Charge). Negative values on the y-axis illustrate opposite direction preference as compared to 834index of x-axis. Note the absence of correlation between the DSI of the thalamic charge and the F1 835modulation of Vm. Green and blue filled data points refer to the neurons in (c,e) and (d,f), respectively. 836

h. Left: Distribution of DSIs of the F1 modulation of thalamic excitation for all recorded neurons. 66 837recordings (41 mice). Above a DSI of 0.3 (horizontal dotted line) the thalamic excitation is considered 838direction selective. Right: The charge of thalamic excitation in the preferred and non-preferred 839directions for each recorded neuron. For each cell the charge is normalized by the largest charge. Filled 840data points are neurons from (c,e) and (d,f). 841


http://dx.doi.org/10.1101/244293

100 pA200 ms

100 pA200 ms

*

*

**

**

100 pA200 ms

Static Grating

Direction Selective Neuron

Non-Direction Selective Neuron

F1

Drifting Grating

60 pA200 ms

F1

a

b

c d

e

f

g h

Direction Selective Neuron

time (ms) time (ms) time (ms)time (ms)

Non-Direction Selective Neuron

Pre

ferr

ed D

irect

ion

of D

riftin

g G

ratin

g

Preferred Direction of Drifting Grating

Pre

ferr

ed D

irect

ion

of D

riftin

g G

ratin

g

Pre

ferr

ed D

irect

ion

of D

riftin

g G

ratin

g

Preferred Direction of Drifting Grating

Pre

ferr

ed D

irect

ion

of D

riftin

g G

ratin

g

Non

-Pre

ferr

ed D

irect

ion

of D

riftin

g G

ratin

g

Non

-Pre

ferr

ed D

irect

ion

of D

riftin

g G

ratin

g

0 0 00400 400 400400600 600 6006001000 1000 10001000200 200 200200800 800 800800

* *

** **

100 ms***

***

100 ms

-1.0

-0.5

0.0

0.5

1.0

1.00.80.60.40.20.0DSI (drifting gratings)

DS

I (st

atic

gra

tings

)

250250 200200 150150 100100 5050 00

80

60

40

20

0

-600

-400

-200

0

0.40.20.0DSI (drifting gratings)

Δ p

hase

(deg

)sl

ope

(deg

/s)

1.00.80.60.40.20.0-0.2

720

720

540

540

360

360

180

180

0

0

spatial phase (deg)

spatial phase (deg)

1.00.80.60.40.20.0-0.2

720

720

540

540

360

360

180

180

0

0

spatial phase (deg)

spatial phase (deg)

inte

gral

(nor

m.)

inte

gral

(nor

m.)

315270225180135

04590

spat

ial p

hase

(deg

)

Direction Selective (example)

Static Grating Drifting Grating Direction Selective (example) Direction Selective (all)

Direction Selective (all)

Non-Dir. Selective (example)

Non-Dir. Selective (example)

Non-Dir. Selective (all)

Non-Dir. Selective (all)

Early EPSC

250 250200 200150 150100 10050 500 0

315270225180135

04590

spat

ial p

hase

(deg

)

time (ms)

time (ms)

time (ms)

time (ms)

1.00.80.6

-1000

-300-200

-500-400

pA pA

0

1

Late EPSC

0 45 90 135 180 225 270 315

0 45 90 135 180 225 270 315

Figure 2 peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/244293doi: bioRxiv preprint first posted online Jan. 7, 2018;

http://dx.doi.org/10.1101/244293

Figure 2 842Time-course of thalamic excitation to static stimuli explains direction selectivity to moving stimuli. 843a. Thalamic EPSCs recorded in a L4 direction selective neuron (DSI F1 thalamic excitation: 0.53) in 844

response to static gratings presented at 16 different phases (0-337.5º, 22.5º spacing) of the preferred 845orientation (250 ms duration). The early (30-100 ms after stimulus onset) and late (110-230 ms after 846stimulus onset) portion of the EPSC are shaded in pink and gray, respectively. The right inset is a cycle 847average of thalamic excitation recorded in the same neuron in response to a drifting grating in the 848preferred (left) and non-preferred (right) direction and the black trace a sinusoidal fit. The arrow (F1) 849shows the amplitude of the modulation. The left inset shows, on an expanded time-scale the 850superimposition of two example thalamic EPSCs (asterisks; peak scaled) elicited by static gratings of 851two different phases. Note the different decay time-course. 852

b. As in (a) but for a non-direction selective neuron (DSI F1 thalamic excitation: 0.16). 853c. Each green trace is the thalamic EPSC to one of the 16 phases of the static grating of the neuron in (a). 854

Each trace is the concatenation of identical EPSCs separated by 500 ms, i.e. a full cycle of the drifting 855grating. Traces are staggered in time relative to one another by 31.25 ms, i.e. a 16th of a full cycle of the 856drifting grating. Asterisks mark EPSC shown in (a). Traces on the left are ordered according to the 857sequence of phases of a grating drifting in the preferred direction. Traces on the right are ordered 858according to the non- preferred direction. The bottom trace is the algebraic sum of the green traces 859above and the black trace is a sinusoidal fit. Note the pronounced F1 modulation of the left as compared 860to the right trace. 861

d. As in (c) but for the non-direction selective neuron in shown in (b). 862e. Phase dependent modulation of the integral of the early (pink) and late (gray) portion of the thalamic 863

EPSC. Top left: example neuron in (a). Bottom left example neuron in (b). The cycle is repeated twice 864for clarity. Right: summary for all neurons with DSI> 0.3 (top; n = 18) and DSI<0.3 (bottom; n = 28). 865Plots were aligned so that the preferred spatial phase of the early EPSC (estimated by vector averaging, 866see methods) occurred at 180º and ordered such that increasing spatial phase corresponded to the 867drifting grating preferred direction. Example neurons from (a) and (b) were shifted by +7.2º and -18.75º 868degrees, respectively. Note the phase shift between the early and late portion of thalamic EPSCs for 869neurons with DSI>0.3 (vertical bars aligned at the peak of the cycle). 870

f. Spatiotemporal receptive field of thalamic EPSCs to static gratings. Hotter colors signify larger 871amplitudes (see scale to the right). Red circles mark the phase with the largest excitation at a given time. 872The red line is a linear fit. Top left: example neuron in (a). Bottom left example neuron in (b). Right: 873summary for all neurons with DSI> 0.3 (top; n = 18) and DSI<0.3 (bottom; n = 28). Population heat-874maps were aligned so that the preferred spatial phase of the earliest time bin occurred at 180º and 875ordered such that increasing spatial phase corresponded to the drifting grating preferred direction. Note 876the different slopes of the linear fit. 877

g. The DSI in response to drifting gratings is plotted against the DSI computed by the summed responses 878to static gratings. 53 recordings (34 mice). Note that for most experiments the direction preference to 879drifting gratings matches the direction preference of the response computed by summing the response to 880static gratings. Filled circles corresponds to cells in (c) and (d). 881

h. Top: The phase difference between the early and late portion of the thalamic EPSC (see (e)) is plotted 882against the DSI in response to drifting gratings. Bottom: The slope of the linear fits (see (g)) is plotted 883against the DSI in response to drifting gratings. 46 recordings (32 mice). The vertical dotted line marks 884DSI=0.3. 885


http://dx.doi.org/10.1101/244293

-60

-40

-20

0

60

0

250 ms

250 250

250

200 200

200

150 150

150

100 100

100

50 50

50

0 0

0

0

1

0 0

-10

-20-20

-30

-40

-40

-60-50

25

0

a

b

c

d

e f

time (s)

time (ms)

curr

ent (p

A)

curr

ent (p

A)

spik

es p

er

sspik

es p

er

s

curr

ent (p

A)

time (ms)

time (ms)time (ms)time (ms)

time (s)

Unit 1

uEPSC

uEPSC

PSpTH

PSpTH

time of spike (Unit 1)

time of spike (Unit 2)

Unit 1

Unit 2

Unit 2

-80

-40

0

2.01.51.00.50.0 2.01.51.00.50.0

-60

-40

-20

0

mem

b. pot. (

mV

)

Static grating

Drifting grating

*

*

**

**20 20

10 10

0 0

60 60

40 40

20 20

0 0

.05

p v

alu

e -2

10

-3

10

-4

10

Thalamic EPSC

Thalamic EPSC

conne

cte

d p

air r

esponse #

PSTH PSTH

0

200 200100 1000 0

s

}

}

V1

dLGN

Unit 1Unit 2

6

5

4

Extracell.

Rec.

Whole cell

Rec.

2/3

8 pA

1 kHz

5 ms

15 pA

1 kHz

5 ms

200

150

100

50

Figure 3peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission.

The copyright holder for this preprint (which was not. http://dx.doi.org/10.1101/244293doi: bioRxiv preprint first posted online Jan. 7, 2018;

http://dx.doi.org/10.1101/244293

Figure 3 886The time-course of firing of thalamic neurons explains the time-course of thalamic excitation. 887a. Left: Schematic of experimental configuration. Right: Two unitary excitatory postsynaptic currents 888

(uEPSC; blue and red traces) from two presynaptic thalamic units (Unit 1 and 2) recorded in the same 889L4 cortical neuron while silencing cortex. The arrows indicate the time of the spike of the respective 890unit. The gray trace is the peri spike time histogram (PSpTH) of events recorded in the L4 neuron (see 891methods and Extended Data Fig. 5 ). The increase in the frequency of events shortly after the spike of 892the unit corresponds to the onset of the uEPSC. The dotted line marks an event frequency of 0.25 kHz. 893

b. The direction selective response of the L4 neuron in (a) to gratings drifting in the preferred (left) and 894non-preferred (right) direction. Top: current clamp recordings (action potentials are truncated). Bottom: 895Voltage clamp recordings during cortical silencing to isolate thalamic excitation. 896

c. Responses to static gratings presented at 16 different phases of the preferred orientation of the neuron in 897(c). Top row: Thalamic excitation recorded in the L4 neuron (same as (b)). Bottom row: Peri-stimulus 898time histograms (PSTHs) of the same two thalamic units as (a). 899

d. Responses to two distinct phases of the static grating (from (c)) on expanded axes. Note that the phase 900of the static grating eliciting thalamic excitation with a slow decay (left) triggers a transient and a 901sustained response in units 1 and 2, respectively. The phase eliciting a fast decaying thalamic excitation 902(right) triggers a response in unit 1 only. 903

e. Heat-maps of responses to static gratings for 20 thalamo-cortical connected pairs (15 postsynaptic L4 904cortical neurons; 20 presynaptic thalamic units; 13 mice; only pairs where the thalamic unit responded to 905static gratings were included; see methods). Each row on the left panel is the amplitude of thalamic 906excitation recorded in one of the 15 L4 cortical neurons in response to one of the 16 phases of a static 907grating presented at the preferred orientation for that neuron. The rows are ordered according to the 908duration of thalamic excitation. The red line is the time at which the integral (charge) of the thalamic 909excitatory response reaches 90% of total. The amplitude of the PSTHs of the simultaneously recorded 910presynaptic thalamic units are shown on the corresponding rows on the right panel. Note that slower 911thalamic excitation (left panel, top) is accompanied by longer lasting PSTHs (right panel, top). The red 912line on the right panel is a copy of the red line on the left panel to facilitate comparison. Heatmap values 913are smoothed along the y-axis. 914

f. Top: Average PSTHs of the top and the bottom rows of the right heat-map in (f) are shown in purple 915and black respectively (brackets in (e) illustrate the range of rows used for the average PSTHs). Bottom: 916average thalamic excitation for the same rows. P-values are for Wilcoxon rank-sum test comparing 917black and purple traces in 20ms bins. 918


http://dx.doi.org/10.1101/244293

-0.7-0.6-0.5-0.4-0.3-0.2

0.30

12345 6

7 8

0.250.200.150.100.050.00

Postsynaptic thalamic excitation

Combined presynaptic thalamic units

Indvidual presynaptic thalamic units

Preferred direction Non-preferred direction

Uni

tsS

umm

ed u

nit a

ctiv

ity

s

Unit 1

Thal. EPSC

Thal. EPSC (all) Units (all)

Units 1 + 2

Unit 2

spat

ial p

hase

(deg

)

1.00.80.60.40.20.0time (s)

1.00.80.60.40.20.0time (s)

a

b

c

200 200100 1000 0

270

180

0

90

spat

ial p

hase

(deg

)

270

180

0

90

spat

ial p

hase

(deg

)

270

180

0

90

time (ms) time (ms)

0

1

Pre

ferr

ed D

irect

ion

of D

riftin

g G

ratin

gP

refe

rred

Dire

ctio

n of

Drif

ting

Gra

ting


http://dx.doi.org/10.1101/244293

Figure 4 919The temporal distribution of thalamic activity underlies cortical direction selectivity 920a. Top: Spatiotemporal receptive field of the firing of thalamic unit 1 (left) and 2 (right; same units as in 921

Fig. 3d) in response to static gratings. Note the phase difference and the different time-course in the 922responses between the two units. Bottom, left: Spatiotemporal receptive field of thalamic excitation 923recorded in a L4 cortical neuron (same neuron as in Fig. 3d) that was postsynaptic to units 1 and 2. 924Right: Combined spatiotemporal receptive field of units 1 and 2. Note the similarity between the 925summed spatiotemporal receptive field of the firing of units 1 and 2 and the spatiotemporal receptive 926field of thalamic excitation. 927

b. Left: Average spatiotemporal receptive field of thalamic excitation to static gratings from 8 pairs where 928the cortical L4 neuron had a DSI > 0.3. Spatiotemporal receptive fields were aligned so that the 929preferred spatial phase of the earliest time bin occurred at 180º and ordered such that increasing spatial 930phase corresponded to the drifting grating preferred direction. Right: Combined spatiotemporal 931receptive field of the firing of the 8 thalamic units presynaptic to the 6 cortical neurons. 932

c. Top: Summed activity of the same 8 presynaptic thalamic units to gratings drifting in the preferred (left, 933red traces) and non-preferred (right, black traces) direction (two identical cycles are shown for clarity; 934pink and gray lines are sinusoidal fits). The PSTHs of each of these units is shown below (units 1 and 2 935correspond to units 1 and 2 in (a)). The units were temporally aligned relative to each other using the 936phase of the F1 modulation of thalamic excitation recorded in their postsynaptic L4 target neurons in 937response to gratings drifting in the preferred and non-preferred direction. Bottom: The average of the 938phase aligned thalamic excitation recorded in the postsynaptic L4 cortical neurons from the same 8 939pairs. 940


http://dx.doi.org/10.1101/244293

Cortex

DS

Thalamus

decay: 3.5τ

Decay of sustained EPSC (τ):

decay: 1τ

o0 o45 o90 o135 o180 o225 o270 o315

S

T

T

S

S

T+S

T+S

T

2τ

0

200

time (τ) time (τ)

Preferred direction Non-Pref. direction

phas

e (d

eg)

Am

plitu

de

0time (τ)

2 4 0time (τ)

2 4 0time (τ)

2 4

a b

c d e

54321

2010 10 10 0

*

*

-0.4-0.20.00.20.4

F1 D

SI

900phase shift (deg)

270180

1.5 2 3.5 5 8


http://dx.doi.org/10.1101/244293

Figure 5 941A simple model of direction selectivity 942

a. Schematic of the model: A cortical neuron (DS) receives two excitatory inputs, one that generates 943currents with a fast decay (1 tau; dark gray) and the other that generates currents with a slower decay 944(3.5 tau; light gray). The slow and the fast decays of the currents reflect the sustained (S) and the 945transient (T) firing of the thalamic input neurons in response to a visual stimulus, respectively. The 946receptive fields of the two thalamic neurons are offset in space relative to each other. 947

b. The excitatory currents mediated by the T and S neurons and their sum (T+S; green) in response to 948static gratings presented at distinct spatial phases. Because the receptive fields of the two thalamic 949neurons are offset (by 90 degrees in phase space for this particular example) the currents that each 950generate in the cortical neurons peak (red asterisks) in response to distinct phases of the static grating. 951As a consequence of this offset, the decay of the compound EPSC (T+S) depends on the spatial phase of 952the stimulus. 953

c. The spatiotemporal receptive field of the currents generated by the T and the S neuron and by their sum 954(T+S). The spatiotemporal receptive field of the sum but not of the individual components is tilted. 955

d. Response of the cortical neurons to gratings drifting in either direction. The cycle period is 10 tau. 956e. The DSI of the cortical neuron is plotted against the spatial phase difference (phase shift) of the T and S 957

neurons for the example above (black) and for different decays of the EPSC generated by the sustained 958thalamic input neuron (colored lines). The vertical green line marks the 90 degrees shift of the example 959above. 960

961


http://dx.doi.org/10.1101/244293

80

40

0Spik

es p

er

s

8

6

4

2

0

contr

ibution p

ref. (

%)

86420

contribution non-pref. (%)

uEPSC

PSTH

Drifting Grating

Thalamic excitation

50 pA

500 ms

Extracell.

Rec.

Whole Cell

Rec.

a

b

PSpTH

time of spike

8 pA

750 Hz

5 ms

Figure 6peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission.


http://dx.doi.org/10.1101/244293

962Figure 6 963Contribution of individual thalamic neurons to thalamic excitation in visual cortex 964a. Left: Schematic of experimental configuration. Right: A unitary excitatory postsynaptic current 965

(uEPSC) from a presynaptic thalamic unit recorded in a L4 cortical neuron while silencing cortex. The 966arrows indicate the time of the spike. The gray trace is the peri spike time histogram (PSpTH) of events 967recorded in the L4 neuron (see methods and Extended Data Fig. 5). 968

b. Top: Peristimulus time histogram (PSTH) of this thalamic unit in response to a drifting grating. Bottom: 969Black trace: Thalamic excitatory current recorded in the L4 neuron in response to drifting grating. Blue 970trace: Unitary thalamic excitation in response to drifting grating computed by convolving the uEPSC 971with the PSTH. Shaded areas are the time integral of thalamic excitation. This thalamic neuron 972contributed ~4% (unitary contribution) of total thalamic excitation. Inset: comparison of unitary 973contributions computed for the preferred and the non-preferred direction for all thalamo-cortical pairs 974(n=23). Green circles: from cortical neurons with DSI of thalamic excitation >0.3 (n = 8). Blue circles 975from cortical neurons with DSI<0.3 (n = 15). Filled blue circle is the example above. 976

977 978 979


http://dx.doi.org/10.1101/244293

time (s)

20

15

10

5

0CtrlCtrl LEDLED

VGAT-ChR2 PV-ChR2

spik

e ra

te (H

z)

2/34

56

2/34

56

CortexChR

ChR

ChR

Extracell.Rec.

a b c

12

8

4

0

spi

ke ra

te (H

z)

3210time (s)

3210

Depth: 615 μm

450-650 μm650-950 μm

Depth: 890 μm

1.00.90.8

0.70.60.5

corti

cal d

epth

(mm

)

20100resid. spikes (%)

Visual stim. Visual stim.LED LED

16

12

8

4

0

-3S

pike

s/10

ms

bin

(x10

)

-3S

pike

s/10

ms

bin

(x10

)4

2

0

6

1.00.80.60.40.2remaining fraction

-300

-200

-100

0

8004000 0time (ms)

Drifting grating

Drifting grating

Static grating

Static grating

ampl

itude

(pA

)

Control

LED

2/34

56

2/34

56

Cortex

Thalamus

Whole Cell Rec.

ChR

ChR

ChR

-200

-100

0

200100

remaining fraction

8

6

4

2

01.00.80.60.40.20.0

time (ms)

num

ber o

f neu

rons

8

6

4

2

00.0

num

ber o

f neu

rons

d

e

f g

h

Extended Data Figure 1peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission.


http://dx.doi.org/10.1101/244293

Extended Data Figure 1. Cortical Silencing a. Experimental configuration: Channelrhodopsin (ChR2) is expressed in cortical inhibitory neurons to

suppress neuronal activity upon illumination with a blue LED while performing extracellular recordings. b. Visually evoked activity (full field drifting gratings) from units isolated throughout the cortical depth is

suppressed upon LED illumination. Black lines: ChR2 was expressed into all GABAergic neurons (vGat-ChR2 mouse line; 138 units recorded with silicon probes; 25 units recorded above 500 μm from the cortical surface (98.9 ± 2.7% silencing) and 113 recorded below 500 μm (99.4 ± 2% silencing); 5 mice). Red lines: ChR2 was conditionally expressed in parvalbumin expressing neurons through viral injection into the visual cortex of the PV-Cre mouse line (13 loose patch recordings in layer 4; 2 mice; 100% silencing).

c. As in (b) but specifically for units recorded between 450-650 μm depth (black; 99.8+/- 0.6% silencing; n = 26; 3 mice) and 650-950 μm (green; putative layer 6; 99.8+/- 2.4% silencing; n = 48; 3 mice) from the cortical surface. These units are a subset of the units from vGat-ChR2 mice illustrated in (b) where the exact recording depth could be estimated. All units from vGat-ChR2 mouse line.

d. Percent visually evoked spikes remaining during LED illumination across cortical depths deeper than 450 um. Same units as in (c)

e. Peristimulus time histogram of two units located at 615 μm (left) and 890 μm depth (right) in response to drifting gratings under control conditions (black) and during cortical silencing (blue). The duration of the visual stimulus and of the LED illumination is illustrated by the horizontal bars.

f. Experimental configuration: as in (a) but whole cell recordings from layer 4 neurons instead of extracellular recordings.

g. Whole cell voltage clamp recording (V holding: -70 mV) of a layer 4 neuron (same neuron as in Fig. 2a). Cycle average in response to drifting gratings (left; two identical cycles are shown for clarity) and to static gratings (right; average of 10 traces). Gray: Control conditions. Black: During LED illumination to isolate the thalamic component of excitation.

h. Distribution of residual excitatory charge upon LED illumination for drifting gratings (66 recordings similar to (g) left) and static gratings presented at the preferred spatial phase (53 recordings similar to (g) right).


http://dx.doi.org/10.1101/244293

3.0

3.0

2.5

2.5

2.0

2.0

1.5

1.5

1.0

1.0

0.5

0.5

Time (s)

Time (s)

3.0

3.0

2.5

2.5

2.0

2.0

1.5

1.5

1.0

1.0

0.5

0.5

Time (s)

Time (s)

-100

-50

0

-50

0

pApA

Time (ms)4002000

Time (ms)4002000

-30-20

0

20

pA

Time (ms)4002000

Time (ms)4002000

-30-20

0

20

pA

1.0

0.8

0.6

0.4

0.2

0.0

DS

I Tha

l. F

1 (C

harg

e eq

ualiz

ed)

1.00.80.60.40.20.0DSI Thal F1

Before Charge Equalization

After Charge Equalization

Preferred dir. Non-Preferred dir.a b

c



http://dx.doi.org/10.1101/244293

Extended Data Figure 2. The contribution of the directional preference of the thalamic charge to the direction selectivity index of ThalF1 In those neurons in which the preferred direction of the thalamic charge matched that of VmF1 (positive DSI Thal. Charge; Fig. 1g right) the absolute value of the DSI of the thalamic charge (0.087+/-0.053; n=32) was significantly different than the absolute value of the DSI of the thalamic charge in those neurons in which the preferred direction of the thalamic charge did not match that of VmF1 (negative DSI Thal. Charge; Fig. 1g right; 0.048+/-0.037 (n=20); p= 0.003; t-test). To determine the impact of this slight bias in thalamic charge on DSIThalF1 we have equalized the thalamic charge evoked by gratings drifting in both directions. a. Example recording from a cortical neuron where the charge of thalamic excitation is larger in the

preferred as compared to the non-preferred direction. Top: thalamic excitation as recorded (non equalized) in response to a grating drifting in the preferred (left) and non-preferred (right) direction. Bottom: same as top but after scaling the response to the non-preferred direction such that the charge is the same in either direction (charge equalization).

b. Cycle average of thalamic excitation with superimposed sinusoidal fit (red). Top: as recorded; Bottom: after charge equalization. After charge equalization the direction preference is maintained but, for this particular example, the DSI is reduced (see filled datapoint in (c)).

c. Scatter plot for all recordings (Green: DSI> 0.3; Blue DSI<0.3). The filled data point is the example above. The equalization leads to only a very small change in the DSI of the thalamic F1 amplitude modulation (All neurons: DSI before equalization: 0.28 +/- 0.20; DSI after equalization: 0.26 +/- 0.21 (p=0.034; paired t-test; n=66); Subset of neurons where DSIThalF1 > 0.3 (green): DSI before equalization: 0.49 +/- 0.15; DSI after equalization: 0.46 +/- 0.17 (p=0.022; paired t-test; n=25); Subset of neurons where DSIThalF1 < 0.3 (blue): DSI before equalization: 0.16 +/- 0.09; DSI after equalization: 0.14 +/- 0.11(p=0.300; paired t-test; n=41).


http://dx.doi.org/10.1101/244293

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

DS

I (f

rom

sta

tic g

ratings)

DS

I (from

sta

tic g

ratin

gs; n

orm

.)

DS

I (from

sta

tic g

ratin

gs; n

orm

.)

8765432

temporal frequency (Hz)

-1.0

-0.5

0.0

0.5

1.0

DSI to drifting gratings >0.3

DSI to drifting gratings <0.3

DSI to drifting gratings >0.3 (peak norm.)

DSI to drifting gratings <0.3 (peak norm.)

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

DS

I (f

rom

sta

tic g

ratings)

8765432

temporal frequency (Hz)

-1.0

-0.5

0.0

0.5

1.0

Full response Cut at half cycle

Extended Data Figure 3


http://dx.doi.org/10.1101/244293

Extended Data Figure 3: Predicting the DSI for Various Temporal Frequencies Direction selectivity index (DSI) predicted from the response to static gratings. The amplitude of the F1 modulation was determined from the algebraic sum of the thalamic EPSCs evoked by each of the 16 phases of the static grating. The thalamic EPSCs were staggered in time to mimic different temporal frequencies of a drifting grating (e.g. at 4Hz a cycle lasts 250 ms and hence the response to each one of the phases is staggered by 15.6 ms (250/16 ms) relative to the preceding one). The DSI was computed by comparing the F1 modulation of the sum in which EPSCs were ordered according to the spatial phase sequence simulating the motion of the grating in one direction against the sum simulating motion in the opposite direction. Green and blue traces, average of all cells whose DSI ThalF1 to drifting gratings was larger or smaller than 0.3, respectively. Dotted traces: the computed DSI was normalized to the peak for each cell; right ordinate. Left panel: the full 250 ms response to static grating was used to compute the DSI at each temporal frequency. Note the reversal of direction preference at higher temporal frequencies. Right panel: Only the initial x milliseconds of the response to static gratings were used to compute the DSI, x being the half period of the temporal frequency to be computed (e.g. for 4 Hz, x = 125 ms). The rationale for this approach is that the interactions between excitatory inputs that are relevant for the emergence of DS likely occur within a half cycle.


http://dx.doi.org/10.1101/244293

-1.0

-1.0

-1.0

-0.5

-0.5

-0.5

0

0

0

0.5

0.5

0.5

1.0

1.0

1.0

-1.0

-1.0

-1.0

-0.5

-0.5

-0.5

0

0

0

0.5

0.5

0.5

1.0

1.0

1.0

early/late Index

150

100

50

0)esahp laitaps .ferp no xedni etal/ylrae yb detros( tinu N

GL

2500 time (ms)

preferred spatial phase

preferred spatial phaseΔ+45 deg

Δ-45 deg

Δ-45 deg spatial phase

Δ-45 deg sp. phase

Δ+45 deg sp. phase

Δ+45 deg sp. phase

Δ+45 deg spatial phase

175 120

0 0

firin

g ra

te

0

1

2001000time (ms)

early late

2001000time (ms)

100

spik

es/s

360

270

270

180

180

90

90

0

0

spat

ial p

hase

(deg

)

pref

. spa

tial p

hase

pref

. spa

tial p

hase

Δ-4

5 de

g sp

. pha

se

spat

ial p

hase

(deg

)

360

270

270

180

180

90

90

0

0

spat

ial p

hase

(deg

)sp

atia

l pha

se (d

eg)

spik

es/s

spik

es/s

} }

a b

c d

e

f

40

30

20

10

0# th

alam

ic u

nits

1.00.80.60.40.20.0DSI F0

-1.0

-1.0

-1.0

-0.5

-0.5

-0.5

0

0

0

0.5

0.5

0.5

1.0

1.0

1.0

-1.0

-1.0

-1.0

-0.5

-0.5

-0.5

0

0

0

0.5

0.5

0.5

1.0

1.0

1.0

early/late Index

Δ-45 deg sp. phase

Δ+45 deg sp. phase

Δ+45 deg sp. phase

DS

I F0

0.3

1.0

pref

. spa

tial p

hase

pref

. spa

tial p

hase

Δ-4

5 de

g sp

. pha

se



http://dx.doi.org/10.1101/244293

Extended Data Figure 4. The response dynamics of dLGN units to static gratings is similar across phases. a. Example dLGN unit with transient response to static gratings. Top: Spatiotemporal receptive field.

Bottom: Peri-stimulus time histograms (PSTHs) in response to each phase of the static grating used to construct the spatiotemporal receptive field illustrated above. The PSTH at the preferred phase is highlighted by a thicker trace. The preferred phase is defined as the phase closest to the vector average of the response at each phase. The brackets show the time windows over which the early and late firing rates were averaged (Re and Rl, respectively) in order to compute the early/late index [(Re-Rl)/(Re+Rl)]. This unit has an early/late index of 1 for static gratings presented at the preferred phase and of 0.88 and 1 for gratings presented at phases of ± 45 degrees from the preferred phase.

b. As in (a) but for an example dLGN unit with sustained response to static grating. The arrows illustrate the preferred spatial phase and the phases separated by ±45 degrees. This unit has an early/late index of -0.3 for static gratings presented at the preferred phase and of -0.08 and 0.22 for gratings presented at phases of ± 45 degrees from the preferred phase.

c. Heat-maps of responses to static gratings for 177 thalamic units (24 mice). Left: Each row is the amplitude of the PSTH of one of the dLGN units in response to the preferred phase of the static grating. The units are ordered according to their early/late index in response to the preferred phase. Middle: Same as left but in response to a static grating whose phase is 45 degrees below the preferred phase. The order of the units has not been changed; i.e. it is the same as on the left. Right: Same as left but in response to a static grating whose phase is 45 degrees above the preferred phase. The order is the same as on the left. Note that transient and sustained units maintain their characteristic firing dynamics even in response to static gratings presented at phases of ±45 degree from the preferred phase.

d. Scatter plots of the early/late index computed in response to static gratings presented at the preferred phase and at phases of ±45 degrees from the preferred. Note that in all plots the data are close to the unity line.

e. Distribution of direction selectivity indexes of the firing rates (DSI F0) of dLGN units in panel d. The vertical dotted line is DSI F0 = 0.3.

f. As in (d) but specifically for those dLGN units with a DSI F0 larger than 0.3 (n=22 units).


http://dx.doi.org/10.1101/244293

0 5 10 15 20 25−8

−4

0

4

8

12

PC 1

PC

2

-50

-40-30

-20

-10

0

10

pA

/ms

-50

-40-30

-20

-10

0

10

pA

/ms

-50

-40-30

-20

-10

0

10

pA

/ms

250

0

-20 -10 0 10 20time (ms)

2 ms

-20 -10 0 10 20time (ms)

threshold

threshold

threshold

jitter (half width

at half max) latency

time window

-20 -10 0 10 20time (ms)

-20 -10 0 10 20time (ms)

-20 -10 0 10 20time (ms)

-20 -10 0 10 20time (ms)time (ms)

-400-200

0

pA

7006005004003002001000

-200

0

pA

/ms

pA

/ms

trace #

-60 -60

-40 -40

-20 -20

0 0

pA

pA

1086420-2-4time (ms)

1086420-2-4time (ms)

1086420-2-4time (ms)

1086420-2-4time (ms)

1086420-2-4time (ms)

1086420-2-4time (ms)

-40

-30

-20

-10

0

z-s

core

-40

-30

-20

-10

0

z-s

core

-40

-30

-20

-10

0

z-s

core

0.25

0.20

0.15

0.10

0.05

0.00

events

per

0.2

ms 20

15

10

5

0

z-s

core

uEPSC (unshuffled - shuffled)

Current (I)

Unit x

Unit x

Unit y

Unit y

Unit z

Unit z

Unit z Unit z

Derivative of Current (dI/dt)

Spike times

Spike triggered average of dI/dt

Spike triggered average of ISpike triggered dI/dt (example traces) Spike triggered dI/dt (all traces)

Peri Spike Time Histogram (PSpTH) z-scored PSpTH

Unfiltered average High pass, z-scored average

0

750

shuffled

unshuffled

a

b

c

d e



http://dx.doi.org/10.1101/244293

Extended Data Figure 5. Criteria for identifying thalamocortical pairs Thalamocortical pairs were identified based on two criteria: Criterion 1 (illustrated in panel c) sets a threshold for the dLGN unit spike triggered average of the time derivative of the current recorded in L4 cortical neurons. Criterion 2 (illustrated in panel d) sets a threshold and a time window for the distribution of events detected in the time derivative of the L4 current around the time of the spike in the dLGN unit. Both criteria have to be satisfied for the dLGN unit and the L4 cortical neuron to be considered a pair. a. Isolation of units in the dLGN. Right: First two principal components illustrating three separable clusters

attributed to three independent dLGN units (units x, y and z in red, gray and blue, respectively). Left: Electrophysiological recording illustrating the average spike shape recorded from 7 electrodes for the three dLGN units.

b. Differentiation of the current recorded in L4 neurons. Same experiment as in (a). Top trace: The current recorded in the whole cell configuration from a L4 cortical neuron (holding voltage:-70 mV) in response to the presentation of a drifting grating (single trial). Middle trace: The temporal derivative of the above current (dI/dt). Lower panel: the times at which each one of the three dLGN units from (a) (x: red; y: gray; z: blue) fired during the same trial.

c. Criterion 1. Left panels: spike triggered average of dI/dt of the current recorded in the L4 neuron for the three dLGN units illustrated in (a). Time 0 denotes the time of the spike. Right panels, same spike triggered averages shown on the left after low pass filtering and z-scoring (see methods). Note that only unit z (blue) crosses the 5z threshold.

d. Criterion 2. Top left: Seven individual time derivatives of currents (dI/dt) recorded in the L4 neuron (same as in (b)) aligned relative to seven spikes recorded in unit z (time 0 denotes the time of the spike). Each asterisk shows an event crossing the threshold of -36pA/ms. Top right: same as left but represented as a heat-map of the amplitude of dI/dt for 761 traces (the heatmap color scale ranges from +50 pA/ms to -200 pA/ms). This heat-map clearly illustrates an increase in event probability around 2 ms following the spike in unit z. Bottom left: the Peri Spike Time Histogram (PSpTH) for the events detected in the 761 traces illustrated above. The peak of the PSpTH is used to determine the latency (i.e. the time interval between the spike recorded in the dLGN unit and the occurrence of a postsynaptic response detected in the L4 cortical neuron). The half width at half max is used to determine the jitter of that response (in this example the latency is: 2 ms and the jitter is 188 microseconds). Bottom right: Same as left but z-scored. The PSpTH must cross a threshold of 3.5z within 1-4 ms after the spike in the dLGN unit for the dLGN unit to be considered synaptically connected to the L4 neuron.

e. Left: Unit z spike triggered average of the response recorded in the same L4 neuron as in (a). Continuous blue line: unshuffled trials. Dotted line: shuffled trials (see methods). Right the difference between the shuffled and unshuffled trials is used to isolate the unitary EPSC (uEPSC) between unit z and the recorded L4 neuron.


http://dx.doi.org/10.1101/244293

-20

-10

0

Am

pl. (

pA)

Am

pl. (

mV

)

Time (ms)

Time (ms)Spa

tial p

hase

(deg

)

0.80.60.40.20.0

27018090

0200 2001000

Pair 0 e/l index:

-20-10

010

3210

Pair 1

-20-10

0

3210

Pair 2

-40

-20

0

1.51.00.50.0

Pair 3*

-40-20

0

2.01.51.00.50.0

Pair 4

-6-4-20

0.60.40.20.0

Pair 5

1050

-5

0.40.20.0

Pair 6

-4-202

0.20.10.0

-0.1

Pair 7*

-10-50

1.51.00.50.0

Pair 8

-6-4-20

-0.2

0.0

0.2

Pair 9*

-4-20

1.0

0.5

0.0

Pair 10

-4-20

0.30.20.10.0

Pair 11

-5

0

0.5

0.0

Pair 12

-15-10

-50

0.80.60.40.20.0

Pair 13

-10-50

1.0

0.5

0.0

Pair 14

100

-10

-1

0

1

Pair 15

-10-50

1050-5

-0.50.00.5

Pair 16

-202

0.40.20.0

-0.2-0.4

Pair 17

-2-101

0.20.0

-0.2

Pair 18

-4-2024

0.60.40.20.0

Pair 19

-4-2

0

-0.10.00.1

Pair 20

-5

0

5

1.00.50.0

Pair 21

-4

-2

0

0.4

0.2

0.0

Pair 22

0.67 0.79 0.99 0.89 -0.04

-0.51 0.97 -0.53 0.52

0.86 -0.19 0.53 0.10 1.00 1.00

-0.72 0.57 -0.22 0.77 0.66

uEPSP

uEPSC uEPSC spont

3

2

1

0

uEP

SP

ampl

itude

(mV

)

50403020100uEPSC amplitude (pA)

uEPSC amplitude (pA)Spo

nt. u

EP

SC

am

pli.

(pA

)

40

30

20

10

0403020100

a

b



http://dx.doi.org/10.1101/244293

Extended Data Figure 6. Unitary EPSCs, EPSPs, and spatiotemporal receptive fields of presynaptic thalamic units. a. Each panel shows the shift-subtracted unitary EPSP (uEPSP; top red, see methods), the

shift-subtracted unitary EPSC (uEPSC; middle blue, see methods) recorded during visual stimulation and the spatiotemporal receptive field (heat-map; bottom) of one of the 23 thalamic units connected to a layer 4 neuron. For some pairs the uEPSC recorded during spontaneous activity (gray, see methods) is also shown. uEPSCs are recorded during cortical silencing; uEPSPs are recorded under control conditions. The vertical line at time 0 marks the time of the peak of the extracellularly recorded action potential in the presynaptic thalamic unit. Pair numbers of the same color were recorded in the same postsynaptic layer 4 cortical neuron. The heat-map shows the spatiotemporal receptive field of the thalamic unit in response to static gratings. Each spatiotemporal receptive field is centered (157.5 degrees) on the preferred spatial phase (defined as the phase that produced the most spikes) of its unit except for converging pairs which are aligned to the average preferred phase of the converging units. The response of pairs marked by an asterisks were not included in the analysis of static gratings because of their poor response to those stimuli. The early/late (e/l) index (see Extended Data Fig. 4) for the preferred phase of the presynaptic thalamic unit is given for each pair on the top right except for pairs with an asterisk. uEPSCs (blue) are the average of 49 - 970 spike triggered traces. uEPSPs (red) are the average of 101 - 1496 spike triggered traces. Spontaneous uEPSCs (gray) are the average of 30 - 412 spike triggered traces. Units of pairs 10, 11, 12, 13, 14, 18, 21, and 22 are the eight presynaptic units to the compound neuron in Figure 4 and correspond to the units numbered in Figure 4 as 6, 7, 1, 2, 8, 5, 4, and 3, respectively.

b. Top panel: Correlation between the amplitude of the visually evoked uEPSC (blue in (a)) and the spontaneously occurring uEPSC (gray in (a); r=0.95 p=5.9e-9 n=17) for those pairs in which both could be recorded. The gray line is unity. Bottom panel: Correlation between the amplitude of the visually evoked uEPSC (blue in (a)) and the uEPSP (red in (a); r=0.59 p=0.0028) for all pairs. The gray line is linear fit to the data with a slope of 0.056 mV/pA.


http://dx.doi.org/10.1101/244293

-1.0

-0.5

0.0

0.5

1.0

thala

mic

unit

DS

I F

0

-1.0

-0.5

0.0

0.5

1.0

0.50.40.30.20.10.0

thalamic excitation

DSI F1

thala

mic

unit

DS

I F

1

1.00.80.60.40.20.0

time (s)

1.00.80.60.40.20.0

time (s)

Preferred direction Non-preferred direction

1

2

3

4

5

6

7

8

PSTHs of indvidual presynaptic thalamic units U

nits

Firing rate equalized

0.30

0.25

0.20

0.15

0.10

0.05

0.00

Combined PSTHs of presynaptic thalamic units

Sum

med u

nit a

ctivity

a b



http://dx.doi.org/10.1101/244293

Extended Data Figure 7. The direction selectivity of presynaptic thalamic units does not contribute to the direction selectivity of layer 4 cortical neurons. a. Top: The DSI of the average firing rate of each presynaptic thalamic unit (thalamic unit DSI F0) is

plotted against the DSI of the F1 modulation of thalamic excitation recorded in the postsynaptic L4 cortical neuron (thalamic excitation DSI F1; r=0.09, p=0.68; 12/23 matching preferred direction, p=1.00, binomial test; n = 23 pairs). Bottom: The DSI of the F1 modulation of the firing rate of each presynaptic thalamic unit (thalamic unit DSI F1) is plotted against the DSI of the F1 modulation of thalamic excitation recorded in the postsynaptic L4 cortical neuron (r=0.16, p=0.46; 11/23 matching preferred direction, p=1.00, binomial test; same 23 pairs as left). Note that the DSI of thalamic units does not predict the DSI of the F1 of thalamic excitation.

b. Top: The PSTHs of each of the 8 units contributing to the compound neuron (from Fig. 4). The units were temporally aligned relative to each other using the phase of the F1 modulation of thalamic excitation recorded in their postsynaptic L4 target neurons in response to gratings drifting in the preferred (red) and non-preferred direction (black). Two identical cycles are shown for clarity. The equalized PSTHs (i.e. the PSTHs that were scaled such that the firing rate of the thalamic unit is the same in either direction) are shown in green. Only the first cycle is equalized to facilitate comparison. Bottom: Summed PSTHs of the 8 presynaptic thalamic units (pink and gray lines are sinusoidal fits; from Fig. 4). The green traces are the summed activity of the equalized PSTHs. Note the similarity between the control and the equalized combined activity.


http://dx.doi.org/10.1101/244293

200

180

160

140

120

100

80

60

40

20

0# of

uni

tary

con

tribu

tions

nee

ded

to e

qual

the

visu

ally

evo

ked

thal

amic

cha

rge

# of unitary contributions needed to equal the visually evoked thalamic charge

1.00.90.80.70.60.50.40.30.20.10.0fraction of big contributors

300

200

100

0120110100908070605040

-4fre

quen

cy (x

10)



http://dx.doi.org/10.1101/244293

Extended Data Figure 8. Number of thalamic units contributing to the visually evoked response of

a L4 cortical neuron.

Top panel: Distribution of the number of unitary contributions of dLGN neurons necessary to equal the

total thalamic charge recorded in a L4 cortical neuron in response to a grating drifting in the preferred

direction during cortical silencing. The distribution is obtained by randomly sampling with replacement the

individual contributions from each of the 23 pairs, 10,000 times. In each iteration, unitary contributions

were sampled until their sum reached 100%. To compute the unitary contribution of a dLGN unit we first

convolved the spike train of the unit in response to a drifting grating with the uEPSC that that unit evoked

in the postsynaptic L4 cortical neuron, integrated the resulting current in time and normalized the obtained

charge by the total charge recorded in the postsynaptic cortical neuron in response to the drifting grating

during cortical silencing. Unitary contributions are expressed in percent of the total charge. On average

80.9 +/- 10.7 dLGN units (average +/- std) contribute to the visually evoked thalamic current in a L4

cortical neuron. Bottom panel: Number of unitary contributions of dLGN neurons necessary to equal the total thalamic

charge as a function of the fraction of “big contributors”. Because the units that contribute a large fraction

of the total charge (big contributors) may have been under-sampled (as a consequence of a skewed

distribution) we have arbitrarily increased their fraction in the pool of unitary contributions and determined

the average number of unitary contributions necessary to equal the total charge, as above. Big

contributors are those dLGN units that contribute to more than 2% of the total charge. They represent

26% of all unitary contributions in our data set of 23 pairs (6 pairs; arrow). Increasing the fraction of big

contributors (x-axis) progressively reduces the average number of dLGN units necessary to equal the

total thalamic charge evoked in response to visual stimulation (y-axis). Each data point is the average +/-

sem.


http://dx.doi.org/10.1101/244293

Date post:	24-Jul-2020
Category:	Documents
Upload:	others
View:	1 times
Download:	0 times

Emergence of Direction Selectivity at the Convergence of ... · 1 Emergence of Direction...

Documents