Activation of a Visual Cortical Column by a Directionally Selective Thalamocortical Neuron

SUMMARY

The retinas of rabbits and rodents have directionally selective (DS) retinal ganglion cells that convey directional signals through the lateral geniculate nucleus (LGN) of the thalamus to the primary visual cortex (V1). Notably, the function and synaptic impact in V1 of these directional LGN signals are unknown. Here we measured, in awake rabbits, the synaptic impact generated in V1 by individual LGN DS neurons. We show that these neurons make fast and strong connections in layers 4 and 6, with postsynaptic effects that are similar to those made by LGN concentric neurons, the main thalamic drivers of V1. By contrast, the synaptic impact of LGN DS neurons on superficial cortical layers was not detectable. These results suggest that LGN DS neurons activate a cortical column by targeting the main cortical input layers and that the role of DS input to superficial cortical layers is likely to be weak and/or modulatory.

INTRODUCTION

Considerable efforts have gone into understanding how neurons signal the direction of movement in the mammalian visual system. In cats and primates, retinal and thalamic neurons respond equally to different directions of movement, and therefore, direction selectivity is thought to be generated de novo within the primary visual cortex (V1). In rabbits (Barlow et al., 1964; Fried et al., 2002, 2005) and in mice (Huberman et al., 2009; Weng et al., 2005; Yoshida et al., 2001), however, directionally selective (DS) neurons are found in the retina and in the lateral geniculate nucleus (LGN) (Levick et al., 1969; Piscopo et al., 2013; Stewart et al., 1971; Swadlow and Weyand, 1985), and in both species, DS LGN neurons have been shown to project to V1 (Cruz-Martin et al., 2014; Kondo and Ohki, 2016; Sun et al., 2016; Swadlow and Weyand, 1985). If this information is conveyed to layer 4 (L4) and layer 6 (L6) of V1, it could contribute to other mechanisms that generate DS responses in V1 (Lien and Scanziani, 2018). Recent studies in the mouse have examined the laminar profile of axonal terminals and boutons of LGN DS neurons in V1 and have reported such inputs to both superficial layers of cortex, as well as deeper layers (Cruz-Martin et al., 2014; Kondo and Ohki, 2016; Sun et al., 2016). However, the postsynaptic effects of LGN DS axons have not been examined, and understanding these synaptic effects is essential for understanding the function of LGN DS neurons and their possible role in generating cortical direction selectivity. Here, we investigate this question by examining the laminar distribution of presynaptic and postsynaptic effects generated by single LGN DS neurons within the visual cortex of awake rabbits. To do this we recorded the extracellular spikes of single LGN DS neurons and examined, in the retinotopically aligned region of V1, the laminar profile of presynaptic (axonal) and postsynaptic local field potentials (LFPs) and current sinks that each of these cells generated within the cortex. Using this method (single-axon spike-triggered LFP and current source density analysis), we have previously studied the pre-and postsynaptic LFPs and currents generated within V1 by individual LGN concentric cells in both rabbits and cats (Jin et al., 2008, 2011; Stoelzel et al., 2008, 2009). Here we show that single LGN DS neurons, like concentric LGN cells, provide their primary synaptic drive to L4, and some also provide input to L6, but none has a detectable synaptic input in the superficial layers. We also show that the strength, rise time, and synaptic dynamics of LGN DS synaptic inputs to V1 are similar to those of LGN concentric cells. Finally, we show that the postsynaptic responses generated in V1 by LGN DS neurons, like those of LGN concentric neurons, are strong enough to drive spiking activity in single L4 and L6 cortical neurons.

RESULTS

In brief, our experimental strategy was as follows: (1) Identify and characterize the receptive field (RF) properties of an LGN DS neuron. (2) Locate the retinotopically aligned region of V1 using a fine-diameter “mapping” microelectrode. (3) Replace the mapping electrode with a 16-channel laminar recording probe positioned perpendicular to the cortical surface and extending through the layers of the cortex. (4) Record thousands of spontaneous spikes from the LGN DS neuron while recording field potentials through the cortical probe sites. (5) Compute the spike-triggered laminar profile of the LFPs and current sinks and sources generated (where current sinks reflect inward, de-polarizing currents, and sources reflect outward currents; e.g., Freeman and Nicholson, 1975; Hagen et al., 2017), thereby gaining a measure of the laminar profile of presynaptic and postsynaptic responses generated in V1 by the single LGN DS neuron under study.

We recorded from 30 DS neurons in the rabbit LGN, which had RFs that showed some overlap with cortical recording sites. For 20 of these cells, the retinotopic alignment was very good, with RF centers of LGN and cortex being separated by less than half the diameter of the LGN RF (mean separation for these 20 cases 0.24 ± 0.03 LGN RF diameters; Figure S1). We used these very well aligned cases for the bulk of our analyses. LGN DS neurons were very directional, with a mean response in the preferred direction ~10 times greater than in the opposite (null) direction (mean DS index [DSI] 0.82 ± 0.02, range 0.62–0.95). For comparative purposes, we also present data from 43 LGN concentric neurons, some studied simultaneously with LGN DS neurons and some drawn from our previous measurements (Stoelzel et al., 2008). As is the case for the LGN DS neurons, we limit analyses of concentric cells to those cases in which the RF centers of LGN and cortex were separated by less than half LGN RF diameters (n = 38, mean separation for these 38 cases 0.22 ± 0.03 LGN RF diameters).

The recording situation is shown in Figure 1A. The LGN DS neuron had an RF that generated spatially overlapping ON and OFF responses (Figure 1B) and showed very strong directional selectivity (Figure 1C; F0 response in preferred and null direction = 48.4 and 4.2 spikes/s, respectively; DSI = 0.84). The preferred directions for each of the 30 DS neurons studied clustered around one of the four cardinal directions (Figure 1D). The depth profile of the spike-triggered LFPs, and the derived current source density (CSD) analysis are given in Figures 1E and 1F, and a colorized version of the spike-triggered CSD profile is given in Figure 1G. Figure 1H shows an expanded version of the spike-triggered LFP response shown in channel 5 (of Figure 1E), in the upper portion of L4. Note that this LGN DS neuron generated strong presynaptic (axonal) and postsynaptic response in L4 and no detectable postsynaptic responses in the superficial or deep layers. We have previously shown, in rabbit somatosensory thalamocortical neurons that project to S1 (Swadlow and Gusev, 2000), that whereas the postsynaptic component of such responses (Figures 1G and 1H, right oblique arrows) is blocked by AMPA/kainate receptor antagonists, the short-latency pre-synaptic (axonal) component is not. Very similar spike-triggered LFP and CSD profiles are generated by LGN neurons with concentric RFs in V1 of both rabbits (Stoelzel et al., 2008, 2009) and cats (Jin et al., 2008, 2011) and, as noted above, by ventrobasal thalamocortical neurons projecting to S1 barrel cortex of the rabbit (Hagen et al., 2017; Swadlow and Gusev, 2000; Swadlow et al., 2002).

Figure 1. Postsynaptic Impact of LGN DS Neurons.

(A) We recorded from one or more LGN neurons in awake rabbits using an array of independently movable microelectrodes. Spike-triggered field potentials, generated from spontaneous spikes of the LGN neuron, were recorded from the retinotopically aligned region of V1 using a 16-channel silicone probe (vertical spacing, 100 mm) or single electrodes.

(B) RFs of an LGN DS neuron and the multiunit cortical RFs from the aligned region of V1 were mapped with sparse noise and plotted using reverse correlation. Retinotopic alignment of the LGN DS cell’s ON responses (red) and OFF responses (blue) with a L4 cortical site (black) was very good, with the RF centers of the LGN and cortical fields separated by less than half the diameter of the LGN RF.

(C) Polar plot of the directional selectivity of this LGN DS neuron showing a strong preference for motion from posterior to anterior visual field.

(D) Distribution of preferred directions of all LGN DS neurons studied. Numbers of cells that fall into directions of cardinal axes are indicated by numbers.

(E and F) Spike-triggered average field potentials (for this cell, generated by 9,947 LGN spikes) were plotted on all cortical probe sites (E), and the spike-triggered CSD depth profile of these averages was derived (F). Downward and upward responses in (F) represent current sinks (inward currents) and sources, respectively.

(G) Color map of the spike-triggered CSD depth profile shown in (F). Here and in Figures 2 and 3, the solid horizontal arrow in colorized CSD profiles (on the left) indicates the cortical channel that corresponded to the reversal point of the field potential generated by a diffuse flash stimulus, brackets indicate estimate of the upper and lower boundaries of L4 (STAR Methods), and red and blue represent current sinks and sources, respectively.

(H) A spike-triggered average LFP elicited from the cortical site with the peak postsynaptic response (channel 5 in E), showing a strong presynaptic (left oblique arrow) and postsynaptic (right oblique arrow) response. Vertical dashed line indicates time of the LGN spike.

Depth profiles of the pre-and postsynaptic impacts generated by LGN DS neurons and LGN concentric neurons were very similar. In five cases, we studied the spike-triggered impacts generated by an LGN DS neuron and one or more LGN concentric neurons. These two LGN cell types were either simultaneously or sequentially recorded, and the impact in V1 was studied using the same cortical recording probe that was unchanged in depth. A total of eight LGN DS/LGN concentric cell pairs were studied in this manner. Notably, the depths within V1 of the pre-synaptic and postsynaptic impacts of all of these LGN DS/LGN concentric cell pairs were very similar. Each of these cells had a peak postsynaptic impact in L4, and in all but one case, the depths of the peak-amplitude postsynaptic impacts were separated by 200 mm or less (in the remaining pair, the peaks were separated by 300 mm). One such pair is shown in Figures 2A– 2C, in which both the LGN DS and the LGN concentric neuron generated their strongest synaptic impact at an identical depth, near the superficial border of L4. Note that the RF of the LGN DS neuron (Figure 2A) generated spatially overlapping ON and OFF responses (as did all of the LGN DS neurons studied here; e.g., Figure 1B). Moreover, all LGN DS neurons responded in a highly transient manner to both dark and light stationary spots presented for 1–2 s (as is characteristic of ON/OFF LGN DS neurons). Moreover, all LGN DS neurons studied showed pronounced non-linear spatial summation in their responses to sinusoidal drifting gratings (F1/F0 ratios < 1, mean 0.40 ± 0.03, range 0.09–0.67). By contrast, LGN concentric cells (which generate either ON or OFF responses in the RF center) responded in either a sustained or transient manner to a stationary stimulus, and most showed linear spatial summation in their responses to sinusoidal drifting gratings (F1/F0 ratios > 1; Figure 3D in Hei et al., 2014). The LGN concentric cell shown here (in Figures 2A and 2C) had an on-center RF and responded in a sustained manner to a light spot, and responses to drifting grating stimuli yielded an F1/F0 ratio of 1.52.

Figure 2. Comparing the Impact Generated in V1 by LGN DS Neurons and Concentric Neurons: Depth and Strength of Impact, Thalamocortical Conduction Times, and Spontaneous Impulse Activity.

(A) RFs of a pair of simultaneously recorded LGN DS and LGN concentric neurons and the RF of the retinotopically aligned recording site in L4 (mapped and plotted as in Figure 1B). (B and C) Depth profiles, recorded simultaneously and on the same probe in V1, of the spike-triggered presynaptic (axonal, left oblique arrows) and postsynaptic (right oblique arrows) currents generated by this LGN DS neuron (B) and concentric neuron (C). Although these two LGN neurons have a similar depth profile, the DS neuron has a somewhat longer latency.

(D) The frequency distribution of depths for the maximum amplitude postsynaptic responses generated by well-aligned LGN DS neurons (n = 20 cells, n = 2 rabbits) and concentric neurons (n = 38 cells, n = 3 rabbits) (the same cells were used for analysis in F–I), normalized to the flash reversal point (depth of “0”; STAR Methods). Four of the DS neurons and 7 of the concentric neurons also generated a postsynaptic response near the L5/L6 border. Asterisks show the depth of V1 SINs (n = 10 cells, n = 1 rabbit) that, on the basis of cross-correlation analysis, received a synaptic drive from an LGN DS neuron (Figure 4).

(E) For comparative purposes, the depth distributions of the cell bodies of corticotectal neurons, antidromically activated from the superior colliculus (green bars, taken from Bereshpolova et al., 2007), and the depth distribution of the most superficial L6 corticogeniculate neurons (antidromically activated via LGN stimulation) found in similarly studied penetrations (Stoelzel et al., 2017) are shown.

(F) The distribution of thalamocortical presynaptic (axonal) latencies for LGN DS neurons and LGN concentric neurons. The inset shows the mean latencies for LGN DS and LGN concentric neurons (p < 0.001, Mann-Whitney U test). (G–I) Comparison of average amplitude (G; p = 0.606, Mann-Whitney U test) and rise time (H; p = 0.2415, t test) of the peak postsynaptic responses generated by well-aligned LGN DS and concentric neurons. The spontaneous firing rates of these LGN populations were also compared (I; p = 0.729, t test). Error bars indicate SEM.

Figure 3. Thalamocortical Synapses of LGN DS Neurons Display Short-Term Depression.

LGN DS spikes with long preceding interspike intervals generate more powerful postsynaptic potentials than spikes with short preceding intervals. (A1 and B1) Individual action potentials were selected from spontaneous spike trains of LGN DS neurons, on the basis of their preceding interspike interval. Arrows indicate selected spikes. New spike trains were generated using only those spikes with preceding interspike intervals within a specific range. Short-interval spikes (A1) were those with preceding interspike intervals between 5 and 20 ms, while long-interval spikes (B1) were those with intervals of 500–3,000 ms. (A2 and B2) Spike-triggered LFP and CSD profiles were generated using only these selected spikes (short, n = 3,348 spikes; long, n = 405 spikes). Gain settings and color intensities for both spike-triggered CSD (stCSD) profiles are identical.

(C) Spike-triggered field potentials generated in L4 by this LGN DS neuron. The spike-triggered average responses based on differing preceding interspike intervals are shown. As in Figure 1H, the dashed vertical line indicates the onset of a thalamic spike.

(D) Postsynaptic amplitude reductions generated by thalamic spikes with short versus long preceding intervals for 16 LGN DS neurons (left) compared with those seen in LGN concentric transient (middle) and concentric sustained (right) neurons (data from concentric sustained versus transient neurons from Stoelzel et al., 2008). LGN DS neurons generated a postsynaptic response that was reduced at the short preceding interspike intervals to a similar extent to what was seen in concentric transient neurons (p = 0.817, t test) but significantly less reduced than the reduction seen in concentric sustained cells (p < 0.001, t test).

We compared the relative depth within V1 of the peak-amplitude postsynaptic current sinks generated by the 20 well-aligned LGN DS neurons and for 38 similarly well-aligned LGN concentric cells (Figure 2D). These depths are normalized to the depth of the reversal point in the flash-evoked potential, which occurs ~100 mm above the superficial border of L4 (Stoelzel et al., 2008). Note that the synaptic impacts generated by both the LGN DS neurons and the concentric neurons show strong peaks at depths that correspond to the estimated position of L4. Four of the 20 LGN DS neurons and seven of the 38 LGN concentric cells showed, in addition to the L4 impact, a clear postsynaptic impact deeper in the cortex, near the estimated position of the L5/L6 border. For reference purposes, Figure 2E shows the position of somatic recordings from corticotectal neurons (data from Bereshpolova et al., 2007) and L6 corticogeniculate neurons (data from Stoelzel et al., 2017). Note the strong degree of overlap in the depths of the corticogeniculate neurons with the deeper synaptic impacts generated by LGN DS and LGN concentric cells.

Although presynaptic (axonal) and postsynaptic depth profiles generated in V1 by LGN DS and LGN concentric neurons are quite similar, the thalamocortical axonal conduction times of the LGN DS neurons were somewhat longer. Thalamocortical axonal conduction time is given by the latency of the presynaptic component of the spike-triggered response (left oblique arrow in Figures 1G, 1H, 2B, and 2C). Although the latency distributions of LGN DS and LGN concentric cells overlapped considerably (Figure 2F), the mean axonal latencies of the LGN concentric cells were 0.4–0.5 ms faster than those of the LGN DS neurons (Figure 2F, inset; 0.82 ± 0.06 versus 1.28 ± 0.10 ms, respectively; p < 0.001, Mann-Whitney U test).

Of course, the amplitudes of all spike-triggered pre-and postsynaptic effects were strongly dependent on retinotopic alignment of LGN and cortical RFs. For this reason, our analyses of these effects were limited to cases in which the centers of the LGN and cortical RFs were separated by less than half LGN RF diameters (represented by cells included within the orange shading in Figure S1). Note that 20 of 22 such well-aligned LGN DS cells generated a clear postsynaptic (and presynaptic) impact at the aligned cortical sites.

Although peak postsynaptic responses generated in V1 by LGN DS and LGN concentric cells occur at very similar cortical depths, the synaptic impact they generate and their spontaneous firing rates could be very different. However, this is not the case. We compared the amplitude (Figure 2G) and rise times (Figure 2H) of the postsynaptic impact generated in V1 by the 20 well-aligned LGN DS neurons with the impacts generated by similarly well-aligned LGN concentric neurons and found them to be very similar (DS versus Con: postsynaptic potential [PSP] amplitude, 9.78 ± 1.32 versus 10.73 ± 1.06 mV, p = 0.606, Mann-Whitney U test; PSP rise time, 0.59 ± 0.04 versus 0.50 ± 0.05 ms, p = 0.2415, t test).

The spontaneous firing rates were also very similar for LGN DS and concentric neurons (Figure 2I; 7.83 ± 0.43 versus 7.60 ± 0.43 spikes/s; p = 0.729, t test), as are the effects of electroencephalographic arousal on the spontaneous firing and burst rates (with spontaneous rates going up when alert and burst rates going down; Figure S2).

We also examined the synaptic dynamics of LGN DS thalamocortical input to L4 and found that they were very similar to that of LGN thalamocortical concentric cells, with both populations exhibiting significant activity-dependent depression. As was found for LGN concentric cells that synapse in V1 (Stoelzel et al., 2008), and for ventrobasal thalamocortical neurons synapsing in somatosensory cortex (Swadlow and Gusev, 2001; Swadlow et al., 2002), the spikes of LGN DS neurons with short preceding interspike intervals (Figures 3A1 and 3A2) generated synaptic responses in V1 that are much weaker than those generated by spikes with long preceding intervals (Figures 3B1 and 3B2). Serving as a control, the presynaptic (axonal) response (left oblique arrow in Figures 3A2, 3B2, and 3C) is unaffected by preceding interspike interval. This is to be expected for synapses formed by axons with high rates of spontaneous activity and depressing synaptic dynamics. In such neurons, the long preceding interspike intervals allow recovery from the chronic synaptic depression generated by the high rates of spontaneous activity seen in these thalamocortical neurons (~8 Hz; Figure 2I; also see Gil et al., 1997; Castro-Alamancos and Oldford, 2002; Stoelzel et al., 2008; Swadlow et al., 2002). This in turn results in weaker postsynaptic responses following short preceding intervals. Figure 3D (left) shows the relative amplitude (percentage reduction) of this reduced response generated by LGN DS neurons with short versus long preceding interspike intervals. For comparative purposes, we also show the reduced responses seen in previously studied LGN concentric sustained and transient neurons (concentric cell data from Stoelzel et al., 2008). The LGN DS neurons show activity-dependent depression that is very similar to what was seen in LGN concentric transient cells and somewhat (but significantly, p < 0.001) less than that seen in concentric sustained cells.

The above results show that LGN DS neurons have strong (and depressing) synapses and therefore may be able to potently drive targeted cells in V1. To examine this, we tested a population of cortical neurons that are known to be strongly connected to topographically aligned thalamocortical neurons: putative fast-spike inhibitory interneurons (suspected inhibitory interneurons [SINs]). These cells mediate a fast, sensitive feedforward inhibition in sensory neocortex, and we previously showed, using cross-correlation methods, that LGN concentric neurons are strongly connected to L4 SINs of V1 (Zhuang et al., 2013) and that somatosensory ventrobasal thalamocortical neurons are similarly connected to L4 SINs of S1 (Swadlow and Gusev, 2001, 2002). Therefore, we asked whether LGN DS neurons also made strong functional connections with SINs of V1. We found that they did. Figure 4A shows a case in which retinotopic alignment of the LGN DS neuron (black contours) and the V1 SIN in L4 (red and blue contours) is nearly perfect. The orientation-directional tuning curves show that the LGN DS neuron (left) has a very strong preference for downward movement, but the SIN (right) responds similarly to all directions (which is typical for these cells; Zhuang et al., 2013). The sharp, short-latency peak in the cross-correlogram (onset at ~1.8 ms) shows a brief increase in SIN spike probability following the LGN spike, indicative of monosynaptic connectivity (Alonso et al., 2001; Swadlow and Gusev, 2001, 2002; Zhuang et al., 2013). Nine of the ten SINs activated by LGN DS neurons were found in L4, and one was located near the L5/L6 border (the depth of these SINs within V1 is shown by the asterisks in Figure 2D). The efficacy of these ten functional connections between LGN DS neurons and V1 SINs had a median value of 2.4%, slightly greater than that seen for LGN concentric cells synapsing onto SINs of V1 (Zhuang et al., 2013) and ventrobasal thalamocortical neurons synapsing onto SINs of S1 (Swadlow and Gusev, 2002). Figure 4B shows another such well-aligned, “connected” cell pairs, in which the presynaptic LGN DS neuron preferred nasal-to-temporal (visual field) stimulus movements. In this case, the V1 SIN was located near the L5/L6 border.

Figure 4. LGN DS Neurons Drive V1 Neurons.

(A) Cross-correlation of spontaneous spike trains of an LGN DS neuron and a retinotopically aligned SIN in L4 of V1. The LGN DS neuron (red and blue contours) shows overlapping ON/OFF responses) and the V1 SIN (black contours) also shows overlapping ON/OFF responses. The orientation-directional tuning curves show that the LGN DS neuron (left) has a very strong preference for downward movement, but the SIN (right) responds similarly to all directions of movement. The black contour plot shows the F0 responses, and the red contour shows F1 responses (which are minimal in both neurons). The sharp, short-latency peak in the cross-correlogram (onset at ~1.8 ms) shows a brief increase in L4 SIN’s spike probability following the LGN DS spike, indicative of monosynaptic connectivity (Swadlow and Gusev, 2001, 2002; Zhuang et al., 2013).

(B) Another LGN DS neuron making a functional contact with a V1 SIN. In this case the SIN was located near the L5/L6 border.

DISCUSSION

Our results show that LGN DS neurons powerfully activate the retinotopically aligned region of rabbit V1. Moreover, like LGN concentric neurons (Stoelzel et al., 2008), LGN DS neurons make the strongest connections within L4 and L6. Importantly, these synapses have all the characteristics of powerful thalamic “drivers” of cortical activity (e.g., fast, strong connections that show synaptic depression; Sherman and Guillery, 1998). These results are consistent with the finding that LGN DS axons make multiple synaptic contacts in V1 (Kondo and Ohki, 2016; Sun et al., 2016). Our results also suggest that the connections that LGN DS axons make in the superficial layers of V1 (Cruz-Martin et al., 2014; Kondo and Ohki, 2016; Sun et al., 2016) may not be sufficiently powerful to serve as drivers of cortical activity in these layers and may, instead, be modulatory, and perhaps affected by brain state and/or behavior.

Our cross-correlation results show that input from LGN DS neurons does indeed drive cortical neurons in L4/L6. Like LGN concentric cells, LGN DS neurons provide a strong input to retinotopically aligned V1 SINs, an input that should contribute to the response properties of these cells. The notable lack of directional selectivity in the responses of the SINs strongly suggests that these SINs do not receive a selective input from LGN DS neurons with a single directional preference. Instead, they probably pool inputs from LGN DS neurons with multiple directional preferences and, possibly, also from concentric LGN cells. If so, this would mirror the situation seen in S1 barrel cortex, where L4 SINs lack a preference for the direction of vibrissae displacement but receive inputs from thalamocortical neurons with multiple directional preferences (Swadlow and Gusev, 2002).

It should be emphasized that both of our experimental approaches, single-axon spike-triggered LFP and CSD analysis and extracellular cross-correlation analysis, are well suited for examining the pre-and postsynaptic effects of strong, focal synaptic inputs to a cortical region (Hagen et al., 2017). Neither method, however, is well suited for detecting weak or diffuse synaptic inputs. Thus, our results do not contradict previous results showing LGN DS axons and boutons in superficial cortical layers (Cruz-Martin et al., 2014; Kondo and Ohki, 2016; Sun et al., 2016). However, they do imply that any such input is significantly weaker and/or less focal than the input to L4 and L6.

Our results indicate that synapses of LGN DS neurons and LGN concentric neurons provide a remarkably similar input to L4/L6. Thus: (1) they both generate extracellular monosynaptic PSPs with similar rise time and amplitude. (2) They both show synaptic depression typical of strong thalamocortical synapses (Gil et al., 1997). (3) They both have similar spontaneous firing rates in the awake alert state, lower rates in the non-alert state, and high rates of “bursting” when non-alert (Figure S2). (4) They both show a similar cortical spread estimated by the decay of postsynaptic responses as the LGN-V1 retinotopic alignment is reduced. (5) They both have fast axonal conduction times, with LGN DS axons being ~0.5 ms slower than LGN concentric axons. (6) Finally, our cross-correlation results show that synaptic input from LGN DS neurons can be strong enough to drive spiking activity in single cortical neurons within both L4 and L6. Thus, just as is seen for LGN concentric afferents, LGN DS afferents provide a rapid, focal, and potently driving input to L4 and L6, the primary input layers of V1.

Our work leaves open the question of whether LGN DS neurons provide and/or contribute to directional and/or orientation selectivity of L4 and L6 simple cells. However, it shows that, like LGN concentric cells, LGN DS cells are capable of providing a strong, “driving” input to the cortex. Such input could contribute to the directionality of simple cells if the DS input were sufficiently selective (i.e., if the directional responses of the LGN DS cells were matched to the directional responses of the simple cells). Moreover, the temporal delay in the LGN DS input to V1 (Figure2F) could contribute to the responses seen in simple cells either directly (through excitatory mechanisms; Stanley et al., 2012; Lien and Scanziani, 2018) or via delayed asymmetric feedforward inhibition. It is notable, however, that aside from their DS responses, most response properties of LGN DS neurons are very different than those of L4 simple cells. Thus: (1) whereas LGN DS neurons have spatially overlapping ON and OFF subfields, L4 simple cells have spatially non-overlapping ON and/or OFF subfields. (2) Whereas LGN DS afferents show pronounced non-linear spatial summation (F1/F0 ratios < 1) in response to drifting gratings (Skottun et al., 1991), L4 simple cells show pronounced linear spatial summation. (3) Whereas LGN DS cells have preferred directions lying on the four cardinal axes, L4 simple cells have broadly distributed preferred directions (Hei et al., 2014; Zhuang et al., 2013). These differences suggest that LGN DS neurons are not likely to selectively “drive” L4 simple cells with the same directional preferences and convey their directional RF properties upon them. This would be consistent with recent results in the mouse indicating a synthesis of DS in V1 by highly selective spatiotemporal convergence of LGN concentric neurons (Lien and Scanziani, 2018). However, input to V1 simple cells from LGN DS neurons could provide a directional bias to their response preferences and, through their powerful drive onto V1 SINs, provide a mechanism for fast and strong feedforward inhibition that sharpens cortical sensory tuning around the four cardinal directions of motion.

METHODS

Simultaneous recordings were obtained from LGN neurons and from the retinotopically aligned region of V1 in awake adult female Dutch-Belted rabbits. All experiments were conducted with the approval of the University of Connecticut Animal Care and Use Committee in accordance with National Institutes of Health guidelines.

Electrophysiological recordings.

Initial surgery was performed under ketamine-acepromazyne anesthesia using aseptic procedures. After removal of the skin and fascia above the skull, the bones of the dorsal surface of the skull were fused together using stainless steel screws and acrylic cement. A stainless steel rod (6 mm in diameter, thinned to 2 mm in places to conserve space on the skull) was oriented in a rostrocaudal direction and cemented to the acrylic mass. The rabbit was held rigidly by this rod during later surgery and recording sessions. A layer of acrylic cement or silicone rubber always covered the exposed skull between recording sessions, and the silicone rubber was also used to buffer the wound margins from the acrylic cement on the skull. Subsequent recordings were obtained in the awake state through a small entrance in the skull (< 1 mm). All electrophysiological activity was acquired using a Plexon data acquisition system (Plexon, Dallas, TX).

Thalamic activity was recorded using a concentric array of seven independently controlled electrodes placed within the LGN (electrode separation ~200 um, Swadlow et al., 2005). The microelectrodes were built from 40 mm quartz-insulated platinum/tungsten filaments that were pulled and sharpened to a fine tip (impedance, 1.5–3 MOhm). Extracellular single-unit recordings and cortical local field potentials were recorded from the retinotopically aligned region of V1 using either 16-channel silicon probes with 100 microns vertical spacing (NeuroNexus Technologies) or the same fine-diameter single electrodes, that were moved through the depth of the cortex. Multiunit recordings from superficial layers of superior colliculus (SC) were obtained using low impedance (< 1 MOhm) electrodes. Two stimulating electrodes (platinum/iridium microwire, 75 mm in diameter and insulated to ~½ mm of the tip) were also implanted in LGN. These were attached to the LGN microelectrode guide tubes and were used to activate cortical neurons following LGN stimulation (Zhuang et al., 2013). Hippocampal EEG was recorded using two electrodes implanted above and below the CA1 layer and used for monitoring brain states. The location of L4 was determined by the reversal point of the field potential generated by a repeated brief diffuse flash stimulus (Stoelzel et al., 2008). L6 probe sites were identified by the presence of antidromically activated corticogeniculate neurons following electrical stimulation of the LGN. Putative fast-spike interneurons (SINs) of L4 and L6 were identified by a high-frequency (> 600 Hz) burst of three or more spikes elicited by electrical stimulation via the LGN stimulating electrodes. All interneurons studied here responded to LGN stimulation at very short latencies (< 2.2 ms mean = 1.89 ± 0.07 ms) with spikes of short duration (negative plus positive components = 0.565 ± 0.015 ms, peak to peak intervals = 0.188 ± 0.009 ms). The receptive fields of these cells all showed a strong spatial overlap of ON and OFF subregions. These characteristics have all been linked to putative fast-spike inhibitory interneurons of sensory neocortex (Swadlow, 2003; Zhuang et al.; 2013)

Receptive field and visual stimulation.

All visual stimuli were presented on a CRT monitor (Nec MultiSync 40 3 30 cm, mean luminance 48 cd/m2, refresh rate 160 Hz). First, the experiment began by searching for one or more LGN DS cells, which are relatively rare in rabbit LGN (< 7%, Hei et al., 2014; Stoelzel et al., 2008; Swadlow and Weyand, 1985). Our strategy, therefore, was to focus on the DS cells, and not spend too much time on other cells. To achieve this, all cells were initially tested for directional responding, and cells that showed roughly equivalent responses were usually not studied further. Then, for cells that showed a directional preference during this initial testing, RFs to stationary flashing stimuli were mapped using sparse noise stimulation made of white and black squares (0.5–2° in a grid of 30 3 22 degrees) and the raw ON and OFF RF matrices were generated by reverse correlation method. ON-OFF subfields of RF were calculated in spikes per second, normalized by the response peak, and transformed into contour plots (each line represents a 10% decrement), that were smoothed using bicubic interpolation. Next, responses to moving stimuli were tested using sine-wave drifting gratings that were optimized for size, temporal frequency, spatial frequency, and contrast. The direction/orientation was pseudorandomly changed to test 12 or 24 directions of motion (in 30° or 15° steps). Each directional grating lasted for 3–8 s followed by 2 s uniform mid-luminance background. Directional tuning for each cell was measured as the average of F0 responses taken for each direction. After an LGN DS cell was identified, a direction-selective index (DSI) was calculated, and only cells with DSI > 0.4 were considered as DS cells (Hei et al., 2014). Third, the cell’s sustained/transient property was measured with flashing stationary stimuli (dark or light), which were optimized by size, orientation, and contrast polarity to elicit the strongest response possible, and were presented on the cell’s RF center for 2 s with 2 s gaps between stimulation. This test was performed when subjects were in the alert state (below). We generally studied only DS neurons, but in some cases we compared DS neurons with nearby concentric LGN cells, studied simultaneously or sequentially on the same microelectrode. Concentric cells showed strong surround inhibition with no or very poor orientation selectivity. They also were classified as sustained or transient.

Achieving retinotopic alignment.

These studies required precise retinotopic alignment between thalamic and cortical recording sites. To achieve this, as a first approximation, cortical RF maps were obtained from the multiunit activity recorded at multiple depths within the cortex using a single movable microelectrode. Based on the known topography of V1, the axis of the mapping electrode was adjusted by retracting and reinserting it at a slightly different angle until the recording sites at different depths were very well-aligned, based on their highly overlapping receptive field locations (see Figure 1A of Bereshpolova et al., 2007). Then, the mapping electrode was replaced with the 16-channel probe. After mapping, the cell’s RF center was constantly tracked (below), and all the visual stimuli were presented to the cell’s RF center. Once RFs were tested and mapped, a large number of spontaneous spikes (usually ~10000 LGN DS spikes) was recorded.

During the experiments, RF maps were used to assess the degree of spatial overlap between the simultaneously recorded LGN DS and cortical RFs. For each LGN DS neuron, the strength and time course of response were measured by summing the 3 by 3 grid elements (3 3 3 degrees) around the element with maximal value for every time interval of the sliding window. These measures were fitted with a polynomial function. The maximum amplitude and half rise time were used to define the strongest subregion of LGN DS cell for assessing the degree of spatial overlap with cortex. Then RF maps of both sites were generated using 20 ms windows, smoothed with a Gaussian filter and fitted with output ellipse. The distance between the spatial location of the RF center of the LGN DS cell and the center of the cortical RF was calculated based on ellipse parameters (width and height defined by 30% of the peak value). This distance was normalized to the RF “diameter” of the thalamic neuron along the appropriate axis.

Cross-correlation analyses.

Cross-correlation analysis was used for assessing functional connectivity between LGN DS neurons and putative cortical fast-spike inhibitory interneurons. Several thousand spontaneous spikes were usually recorded from each neuron. When this was achieved, cross-correlograms, based on the spontaneous firing of LGN neurons and FS interneurons, were computed. Monosynaptic connectivity was inferred from the presence of significant peaks in the probability of SIN firing at intervals of 1.1–2.5 ms following thalamic spikes. A peak in a cross-correlogram was defined as significant when at least two of three successive bins (0.1 ms bin width) in the peak exceeded the 0.01 confidence level. Baseline firing was calculated between −4 and +1 ms of the time of thalamic firing (Swadlow and Gusev, 2001, 2002).

Monitoring eye position.

The eye position of the awake rabbit is usually very stable (Collewijn, 1971; Swadlow and Weyand, 1985; Bezdudnaya et al., 2006). During recording sessions, the eye position was continuously monitored by mapping, using a second LCD monitor, the SC multiunit RF position with sparse noise. If an eye movement occurred during testing, the relation between the RF center of the LGN cell and the superior colliculus multiunit RF center was used to dynamically place the stimulus on the LGN RF. Data within 15 s of eye movements was discarded in offline analysis (Zhuang et al., 2013). The relationship between the RFs in the SC and LGN was determined when the rabbit’s eye was stable. At the same time, for most cells, the pupil position and size were monitored by an infrared high-speed camera system (ViewPoint EyeTracker System; Arrington Research), which was mounted 40 cm away from the rabbit eye.

Monitoring brain state.

In the rabbit LGN and cortex, the response of neurons to stationary stimuli can be classified as either sustained or transient. However, the sustained response is severely decreased in the absence of arousal in both LGN (Bezdudnaya et al., 2006; Swadlow and Weyand, 1985) and V1 (Zhuang et al., 2013). Therefore, to classify LGN and cortical neurons as sustained or transient this test was done in the alert state. For that reason we monitored hippocampal and cortical EEG. The alert state is indicated by hippocampal “theta” activity (5–7 Hz) and cortical desynchrony, and the nonalert state is indicated by hippocampal high-voltage irregular activity, and more slow-wave activity in the neocortex (Bereshpolova et al., 2011; Bezdudnaya et al., 2006; Swadlow and Gusev, 2001). To compare effect of arousal state on spontaneous firing rate and burst rate, the recordings were segmented by visual inspection into alert versus nonalert states based on the hippocampal and cortical EEG (as above). Bursts were identified as clusters of two or more spikes with ISIs of % 4 ms, where the initial spike of the burst had a preceding interval of at least 100 ms (Lu et al., 1992; Bezdudnaya et al., 2006).

Quantification and Statistical Analysis.

Spike waveforms were identified online and verified offline by Plexon cluster analysis software. All data were then analyzed using Plexon NeuroExplorer (Nex Technologies, Inc.), MATLAB (The MathWorks, Inc.), Anaconda (Python distribution, Anaconda, Inc.).

Directional tuning for each cell was measured as the average of F0 responses measured with gratings pseudorandomly drifting in 12 or 24 directions. For each cell, the preferred direction, direction selective index (DSI), orientation selective index (OSI) and circular variances were calculated. The preferred direction was computed as the vector sum of the responses in all the directions. The DSI, were calculated as follows:

$$DSI=\left(R\_pref-R\_null\right)/\left(R\_pref+R\_null\right)$$

where Rpref is the F0 response in the measured preferred direction, which was defined as the stimulus direction closest to the vector sum of responses across all directions. Rnull is the F0 response in the stimulus direction 180 ° opposite of the preferred direction.

The tuning curve of each cell was fitted by the von Mises distribution, modified from Elstrott et al. (Elstrott et al., 2008):

$$R=R0+R_{\max }{e}^{k\cos \left(x-\mu \right)}/e^k$$

where R is the F0 response in any given direction x; Rmax is the maximum F0 response; μ is the preferred direction in radians and k is the concentration parameter for tuning width.

Spike-triggered LFP/CSD analysis.

Methods and rationale for localizing the pre-and postsynaptic responses of single thalamocortical cells through the depth of the cortex have been described (Hagen et al., 2017; Jin et al., 2008; 2011; Stoelzel et al., 2008; 2009; Swadlow and Gusev, 2000; Swadlow et al., 2002). Spontaneous impulses of the LGN DS cells were recorded along with the field potentials generated through the depths of the retinotopically aligned region of V1. Current-source density (CSD) profiles were generated from the field profiles according to the method described by Freeman and Nicholson (1975). First, we duplicated the uppermost and lowermost field trace (Vaknin et al., 1988), which converted our 16 recording channels to a total of 18 channels. Next, we smoothed the traces to reduce high spatial-frequency noise components. Next, we calculated the second derivative, and this yielded a total of 14 traces. In the CSD profiles, current sinks are indicated by downward deflections and sources are indicated by upward deflections. To facilitate visualization of CSD profiles, we generated color image plots. These were plotted by linear interpolation along the depth axis, with red and blue representing current sinks and sources, respectively. Green is approximately zero, normalized to the 1 ms period before the thalamic spike.

For most analyses, we used all of the spikes in the dataset except those that occurred within 5 ms of another spike. We eliminated LGN spikes with short (< 5 ms) interspike intervals to avoid generating compound averages from any high-frequency spikes, such as those that occur during thalamic bursts. When examining the effects of preceding interspike interval on the spike-triggered potentials, we analyzed only those LGNd spikes that were (1) preceded by interspike intervals of a given value and (2) were not followed by another spike within 5 ms.

Two conditions had to be met before an LGN neuron was considered to have generated a postsynaptic impact in either L4 or L6: (1) the response had to be preceded by a clear axon terminal response R 0.75 mV in amplitude; and (2) the postsynaptic response had to follow this axonal component by < 1 ms and consist of a sharp negativity in the LFP and spike-triggered record. The latency, peak amplitude and rise time of the axonal and postsynaptic responses were measured from the spike-triggered field potentials generated on the peak channel. To study synaptic dynamics, the preceding interspike interval of each spike was measured, and the amplitude of the synaptic impact generated by the spikes with various categories of intervals was then compared (Stoelzel et al., 2008; Swadlow and Gusev, 2001; Swadlow et al., 2002).

Statistical analysis.

Parametric tests were used when groups passed normality (using Shapiro–Wilk test). When a dataset failed normality, comparisons were made using the Wilcoxon signed rank test for paired data and the Mann–Whitney U test for unpaired data. Mean ± SEM is reported for all confidence intervals. All tests were two-tailed. The significance level for all statistical tests was set at p < 0.05.