1. Neuroscience
Download icon

Three rules govern thalamocortical connectivity of fast-spike inhibitory interneurons in the visual cortex

  1. Yulia Bereshpolova
  2. Xiaojuan Hei
  3. Jose-Manuel Alonso
  4. Harvey A Swadlow  Is a corresponding author
  1. Department of Psychological Sciences, University of Connecticut, United States
  2. Department of Biological and Vision Sciences, State University of New York College of Optometry, United States
Research Article
  • Cited 1
  • Views 796
  • Annotations
Cite this article as: eLife 2020;9:e60102 doi: 10.7554/eLife.60102

Abstract

Some cortical neurons receive highly selective thalamocortical (TC) input, but others do not. Here, we examine connectivity of single thalamic neurons (lateral geniculate nucleus, LGN) onto putative fast-spike inhibitory interneurons in layer 4 of rabbit visual cortex. We show that three ‘rules’ regulate this connectivity. These rules concern: (1) the precision of retinotopic alignment, (2) the amplitude of the postsynaptic local field potential elicited near the interneuron by spikes of the LGN neuron, and (3) the interneuron’s response latency to strong, synchronous LGN input. We found that virtually all first-order fast-spike interneurons receive input from nearly all LGN axons that synapse nearby, regardless of their visual response properties. This was not the case for neighboring regular-spiking neurons. We conclude that profuse and highly promiscuous TC inputs to layer-4 fast-spike inhibitory interneurons generate response properties that are well-suited to mediate a fast, sensitive, and broadly tuned feed-forward inhibition of visual cortical excitatory neurons.

Introduction

Neurons within the visual pathway form highly specific connections to preserve the precise retinotopic organization needed for visual acuity. Such connection specificity requires an active developmental process of synaptic pruning that limits the number of neurons receiving input from the same axon. In thalamus, each retinal afferent makes profuse retinogeniculate connections early during development but the connections are dramatically pruned at later developmental stages leaving only a small subset of neurons connected to the same retinal afferent (Hamos et al., 1987). Similarly, thalamic axons are dramatically pruned during cortical development leaving only a subset of neurons connected to the same thalamic axon in the mature visual cortex (Alonso et al., 2001; Taylor et al., 2018). However, the specificity of thalamocortical (TC) connectivity differs for different classes of neurons in layer 4 (L4) of sensory neocortex. For all cell types, a high degree of topographic precision is a necessary condition for TC connectivity, but in some systems other requirements are also stringently imposed. For example, synaptic connectivity between cells of the lateral geniculate nucleus (LGN) and simple cells of the feline visual cortex (V1) is highly dependent on precise retinotopic alignment, but also requires a similarity of a number of receptive field (RF) properties of the thalamic inputs and cortical target neurons (Alonso et al., 2001; Hirsch and Martinez, 2006; Sedigh-Sarvestani et al., 2017). This specificity results in a relatively low connection probability between retinotopically-aligned LGN neurons and L4 simple cells, and accounts, in part, for the orientation and direction selectivity of visual cortical simple cells (Alonso et al., 2001; Lien and Scanziani, 2018; Sedigh-Sarvestani et al., 2017). By contrast, the probability of a synaptic connection between a TC neuron and a topographically aligned cortical fast-spike inhibitory neuron is much higher. For example, a connection probability of ~2/3 is seen between ventrobasal thalamic neurons and putative fast-spike interneurons (suspected inhibitory interneurons, SINs) in the aligned somatosensory cortex of rabbits (Swadlow and Gusev, 2002) and rats (Bruno and Simons, 2002). In cat visual cortex, these inhibitory cells also respond to thalamic inputs with less selectivity than do the (presumptive) spiny simple cells (Sedigh-Sarvestani et al., 2017).

Here, we examined the synaptic connectivity between LGN concentric neurons (the most populous cell-type in the LGN) and L4 SINs in rabbit V1 (Zhuang et al., 2013). We found that three necessary conditions, or ‘rules’, regulate the connectivity between these populations. One rule concerns the precision of TC retinotopic alignment, another concerns the peak amplitude of the postsynaptic local field potential (LFP) elicited near the interneuron by the specific LGN neuron under study, and a third concerns the latency of the specific SIN’s response to visual stimulation and electrical stimulation of the thalamus. Together, these rules generate a highly accurate prediction (in 41 of 42 cases) of whether an LGN-SIN pair will be synaptically connected. Moreover, the dense, highly divergent/convergent TC network that results from the application of these ‘rules’ is consistent with the broadly tuned RF properties of many fast-spike inhibitory cortical neurons (Bruno and Simons, 2002; Hirsch et al., 2003; Nowak et al., 2008; Swadlow and Gusev, 2002; Zhuang et al., 2013). It is also consistent with the fast, potent, and broadly tuned feed-forward inhibition onto L4 spiny cells (Bagnall et al., 2011; Cruikshank et al., 2007; Gabernet et al., 2005; Miller et al., 2001a; Miller et al., 2001b; Swadlow, 2003; Taylor et al., 2018).

Results

We studied the functional connectivity between 50 concentric LGN neurons and 47 L4 SINs (64 LGN-SIN pairs) that had RFs aligned, to various extents, with those of the LGN RFs. Figure 1A shows the recording situation. Once an LGN neuron was isolated and the RF was plotted, a recording electrode was placed in the retinotopically-aligned region of V1 (after appropriate mapping procedures). Cortical recording electrodes consisted of either a 16-channel laminar probe, aligned perpendicular to the cortical surface, or a single microelectrode, similarly aligned, that was slowly advanced into L4. A stimulating electrode was also located in the LGN for gross electrical stimulation of TC inputs. This was used to identify V1 SINs. As described in Methods, SINs were identified by the high-frequency spike discharge elicited by electrical stimulation of the thalamus (3 + spikes elicited at frequencies of >600 Hz). SINs also had spikes of very short duration. Figure 1 (see also supplementary figure 1) illustrates how we measured the properties of L4 SINs, and the frequency distributions of these measures, compared with those of L4 simple cells and SINs, previously studied in awake rabbit V1 using the same methods (from Zhuang et al., 2013).

Figure 1 with 1 supplement see all
Experimental approach.

(A) Schematic of the experimental preparation. Extracellular spikes were recorded simultaneously from one or more LGN cells and from one or more L4 SINs. Cortical recordings were obtained from either a 16-channel laminar probe (aligned perpendicular to the cortical surface) or from a single movable microelectrode. Electrical stimulation of the LGN was used to identify V1 SINs. (B) Mapping with the sparse noise stimulation allowed to obtain spatiotemporal RFs and ON or OFF subfields of LGN neurons and V1 SINs. (C) An example of the spatiotemporal RF maps of an LGN concentric cell and SIN of L4 shown as a color maps in a series of time delays after stimulus onset. (D) Spatial RF maps for the same LGN-SIN pair shown in C. The spatial maps were obtained by reverse-correlating neuronal responses with the white or black stimuli in a time window (±15 ms) around the response peak. Colors correspond to the response sign (red, ON; blue, OFF). (E) Temporal profiles of the spatial RF for the same ON-center LGN neuron (upper trace) and ON-OFF balanced SIN (lower traces) shown in C-D. Spatial response is normalized to maximum within each cell. Red and blue points are the average response at each time step and lines are a fitted polynomial function of ON and OFF responses respectively. The vertical dashed lines mark the latencies of RFs. Peak latency of the response is taken as the latency to the peak of the interpolating function (vertical lines), duration of the response is taken as the full width at half maximum value (horizontal lines). (F) Cross-correlograms for the same LGN-SIN pair indicating synaptic connectivity between LGN neuron and retinotopically-aligned SINs in L4. ‘0’ on the x-axis represents the time of the LGN spikes. The correlograms are based on spontaneous activity.

Once the LGN neurons and V1 SINs were identified, we used sparse noise stimulation (Figure 1B) to further assess their retinotopic alignment, spatiotemporal RFs and other RF properties. Figure 1C shows an example of the spatiotemporal RF maps of an LGN concentric cell and an L4 SIN, shown as a color maps in a series of time delays after stimulus onset. By definition, LGN concentric neurons yield only ON-center or OFF-center responses at short latencies. By contrast, the RFs of SINs consists of highly overlapping ON and OFF response zones, which usually occur at similar amplitudes and latencies (Zhuang et al., 2013). In this case, the LGN cell was ON-center (Figure 1C, top two rows) and the SIN responded at similar amplitude and time course for light and dark stimuli (Figure 1C, bottom two rows). Figure 1D shows the spatial RF maps for this LGN/SIN pair, taken at the temporal window that yielded the peak response for each cell. Figure 1E shows, for the same LGN neuron (above) and L4 SIN (below), how the ON and OFF responses (and their spatial distributions) evolve over time. The LGN cell clearly responded earlier than the SIN, and the ON and OFF responses of the SIN occurred with roughly similar amplitude and time course.

To assess synaptic connectivity between LGN cells and L4 SINs, we employed cross-correlation analysis. In order to avoid stimulus-induced correlations, spontaneous spike trains were used in these analyses (e.g. Bereshpolova et al., 2011; Bereshpolova et al., 2019; Swadlow and Gusev, 2001; Swadlow and Gusev, 2002; Zhuang et al., 2013). Figure 1F shows, for this LGN/SIN pair, a strong, brief (~1 ms) increase in SIN spike probability beginning ~1.3 ms (with the peak at 1.7 ms) following the LGN spike. Such short-latency, brief increases in spike probability are a hallmark of TC synaptic connectivity as measured by extracellular cross-correlation of spike trains (e.g. Reid and Alonso, 1995; Tanaka, 1983; Swadlow and Gusev, 2001). The ‘efficacy’ of this connection (an index of the probability that an LGN spike will elicit a SIN spike) was 2.6%.

Retinotopic alignment was a dominant factor governing synaptic connectivity between LGN concentric cells and L4 SINs. Figure 2A shows, for all concentric LGN cells studied, how the likelihood of a synaptic connection with L4 SINs depends on retinotopic alignment. The proportion of connected pairs is high (31/42 cell pairs, ~73%) when alignment is within ½ of the diameter of the LGN RF center and drops off precipitously when misaligned by > ½ of a RF diameter. This implies a very high degree of divergence from single LGN neurons to multiple aligned SINs, and convergence from multiple LGN neurons to individual retinotopically-aligned L4 SINs. Importantly, however, although precise retinotopic alignment was a necessary correlate of synaptic connectivity, it was not sufficient. Thus, 11 of 42 (~26%) highly-aligned LGN/SIN pairs (where RF centers separated by <½ of an LGN RF diameter) showed no signs of synaptic connectivity. We initially thought that these retinotopically aligned but non-synaptically connected cases might be due to a dissimilarity in some LGN and cortical RF properties (as is the case for simple cells in cat visual cortex [Alonso et al., 2001]). For example, the center responses of LGN concentric cells are either ON or OFF and, although most SINs show highly overlapping ON and OFF subfields, some are more dominated by ON or OFF responses.

Rule 1: Retinotopic alignment.

(A) Relationship between retinotopic alignment of LGN and L4 SIN RF centers (x-axis) and (1) the efficacy of the connection between concentric LGN neurons and SINs (filled circles, y-axis, left side) and (2) the percent connected cell pairs (solid line, y-axis right side). The proportion of connected pairs is high (~73%) when alignment is good (i.e. when the distance between thalamic and cortical RF centers is less than 1/2 the diameter of the LGN RF), and drops off precipitously as misalignment increases. Non-connected cell pairs were ascribed an efficacy of ‘0’ and are shown by open circles. Insets show cross-correlograms for four LGN-SIN pairs with different degrees of RF alignment. Red arrow denotes the data point for cell pair shown in B-E. (B-E) Importantly, LGN-SIN synaptic connectivity does not depend on similarity of RF properties other than alignment. An example of connected cell pair with dissimilar RF sign (LGN cell was OFF-center, SIN was ON-dominated) and dissimilar sustained/transient cell class. (B) Spatial RF maps from a LGN neuron and a retinotopically-aligned L4 SIN. (C) LGN and SIN RFs are dissimilar in their RF sign. The LGN neuron yields only an OFF response at short latencies, and SIN yields earlier and stronger ON response. (D) These LGN and SIN RFs are also dissimilar in their sustained/transient responses. PSTHs of the responses of the LGN cell (upper) and the SIN (lower) to an optimal stationary stimulus placed over the RF center, presented for two seconds (arrows mark stimulus onset). The LGN cell responded in a purely transient manner, but the SIN’s response had both transient and sustained components and was therefore classified as a ‘sustained’ cell (Bezdudnaya et al., 2006; Stoelzel et al., 2008; Zhuang et al., 2013). (E) Cross-correlation of the spike trains (spontaneous spikes only) of the LGN cell and the SIN, indicating synaptic connectivity.

Figure 2B shows such a case, where the RF maps from a concentric LGN neuron (OFF-center, yielding only an OFF response at short latencies), and from a very well-aligned L4 SIN (which yielded both ON and OFF responses at similar latencies). Post-stimulus time histograms (PSTHs) of the responses to light/dark stimuli are shown in Figure 2C, and show that the SIN response to the light stimulus was stronger than the response to the dark (OFF) stimulus. Thus, in this case, there was a clear dissimilarity between the RF properties of the LGN neuron (OFF-center) and SIN (ON-dominated). There was also a dissimilarity in the temporal dynamics of their response to a stationary stimulus (Figure 2D). LGN neurons can be classified as ‘sustained’ or ‘transient’ (Cleland et al., 1971; Swadlow and Weyand, 1985; Bezdudnaya et al., 2006; Stoelzel et al., 2008), and V1 SINs can also be classified in this manner (Swadlow and Weyand, 1987; Zhuang et al., 2013). In this case (Figure 2D), the LGN cell (upper histogram) responded in a transient manner, but the SIN’s response (lower histogram) was sustained. Thus, this LGN/SIN pair, although very well-aligned retinotopically, was dissimilar with respect to both of these RF properties (ON/OFF and Sustained/Transient). Because of this, we thought that this cell pair would be one of those that were not synaptically connected, but we were wrong. Cross-correlation of their spike trains revealed that this LGN-SIN cell pair was very well connected (Figure 1E), with LGN spikes generating a strong, brief increase in SIN spike probability beginning ~1.8 ms following the thalamic spike, and lasting about 1 ms. We found that neither similarity of LGN and SIN ON/OFF responses, nor similarity of sustained/transient responses was a significant factor in predicting the synaptic connectivity of retinotopically-aligned LGN/SIN pairs. Moreover, the probability of observing the connection in pairs where both parameters were similar did not significantly differ from pairs where both parameters were dissimilar (X2 test, p=0.769).

Two factors explain nearly all cases in which retinotopically-aligned, LGN/SIN cell pairs are not synaptically connected

In addition to retinotopic alignment, a critical factor controlling TC synaptic connectivity concerns the strength of the synaptic drive from the LGN cell to the local network within which the aligned SIN is imbedded. TC neurons can differ greatly in the overall synaptic drive that they produce at different depths within L4 (Humphrey et al., 1985; Jin et al., 2008; Stoelzel et al., 2008). Even neighboring LGN neurons may selectively target deep vs. superficial portions of L4 and some may not even project to V1 (e.g. interneurons). When recording in vivo, such differences are reflected by the amplitude of the spike-triggered LFPs and/or current source density (CSD) profiles generated by the spikes of single thalamic neurons at different cortical depths (Jin et al., 2008; Jin et al., 2011; Stoelzel et al., 2008; Swadlow and Gusev, 2002). To examine the effect of such differences, our cortical recording electrodes were filtered appropriately to enable recording both spikes (from the SINs and other cortical neurons) as well as low-amplitude LFPs that were triggered by the spikes of the retinotopically-aligned LGN neurons (which required methods of spike-triggered averaging to be revealed). We then determined how the amplitude of the monosynaptic component of the spike-triggered LFP generated near the SIN by spikes of the LGN neuron (Swadlow and Gusev, 2001; Swadlow et al., 2002; Stoelzel et al., 2008) was related to the probability of a synaptic connection of that LGN neuron with the SIN under study. Figure 3A–C shows an instructive case in which the RFs of two simultaneously recorded LGN neurons were both retinotopically aligned with the RF of a single SIN (Figure 3A) that was located superficially within L4. This SIN was recorded on the 8th channel of the laminar probe (shown by asterisk in Figure 3B1and B2). One of these LGN neurons generated a maximum response more superficially in L4 (Figure 3B1, peak response at the same depth as the SIN), and this LGN neuron was synaptically connected to the SIN (Figure 3C1). The other LGN neuron generated a postsynaptic response that was deeper in L4 (Figure 3B2), with almost no response in superficial L4 (where the SIN was located). This LGN neuron showed no evidence of synaptic connectivity with the SIN (Figure 3C2), despite the good retinotopic alignment of their RFs. Figure 3D shows, for all 42 of our well-aligned LGN-SIN cell pairs, the significant relationship between the peak amplitude of the spike-triggered postsynaptic LFP response generated by the LGN neuron in the near vicinity of the aligned SIN, and the efficacy of the synaptic connection with the SIN. Non-connected cell pairs were ascribed an efficacy of ‘0’ and are shown by open circles. Synaptic efficacies of ‘0’ (no connection) were found for each of the 6 LGN neurons that failed to generate a postsynaptic LFP at the site of the SIN under study (represented by the large open circle, lower left). Thus, 6 of the 11 cases in which no synaptic connectivity was observed, despite excellent retinotopic alignment, can be accounted for by a very weak (or absent) LGN input to the vicinity of the SIN under study.

Rule 2: The ‘strength’ with which the LGN neuron provides a synaptic drive to the local network within which the aligned SIN is imbedded.

(A-C) An example of the RF maps of two simultaneously recorded LGN neurons and a retinotopically-aligned SIN recorded on a linear probe. (A) The RFs of two LGN neurons (red - ON-center, blue - OFF-center) along with the RF of the SIN located within superficial L4 of V1. (B1, B2) Spike-triggered LFPs and the colorized CSD profiles generated in V1 by the spikes of these two LGN cells. The SIN was recorded on the 8th channel from the top of the probe (asterisk). The spike-triggered LFP/CSD generated by the OFF-center LGN cell had a maximum response more superficially in L4, at the same depth as the SIN, and this LGN neuron was synaptically connected to the SIN (C1). The ON-center LGN neuron generated a postsynaptic response that was deeper in L4 (B2), with almost no response in superficial L4 (where the SIN was located, and this LGN neuron was not connected to the SIN (C2)), despite the good retinotopic alignment of their RFs. (D) For all 42 well-aligned LGN-SIN cell pairs, the relationship between the amplitude of the spike-triggered postsynaptic response generated by the LGN neuron recorded near the SIN and the efficacy of the synaptic connection with the retinotopically-aligned SIN (non-connected cell pairs were ascribed an efficacy of ‘0’ and are shown by open circles). The largest open circle illustrates LGN cells with no spike-triggered postsynaptic LFP and no connection with a SIN revealed by cross-correlation analysis. Red arrows denote data points derived from the LGN-SIN pairs shown in B1, C1 (right arrow) and B2, C2 (left arrow).

Next, we show that all-but-one of the remaining ‘unexplained’ cases of retinotopically aligned, but unconnected LGN-SIN pairs can be accounted for by a long-latency synaptic response of these ‘unconnected’ SINs to strong electrical stimulation of the LGN, and to visual stimulation. Thus, they appear to receive minimal direct input from LGN neurons. Figure 4A shows the distribution of synaptic latencies for SINs that were connected or were not connected to their aligned LGN neuron. None of the SINs responding at synaptic latencies of >3 ms to electrical stimulation of the thalamus were synaptically connected to their paired and well-aligned LGN neuron (31/37 of the SINs showing synaptic latencies < 3 ms were connected). Notably, SINs with long synaptic latencies to electrical stimulation of the LGN (>3 ms) also responded at longer latencies to visual stimulation than did those that responded at shorter synaptic latencies (<3 ms) (Figure 4B, 35.69 ± 1.78 ms vs. 25.77 ± 0.38 ms, respectively, p<0.001), suggesting that the long synaptic latencies of these SINs to electrical stimulation of the LGN reflect a fundamental difference in synaptic connectivity of these cells with the visual thalamus. Importantly, the spike duration of these long-latency SINs did not differ from those of the other SINs (0.467 ± 0.022 ms vs. 0.449 ± 0.022 ms, p=0.337). Moreover, the long-latency SINs had similar RFs to those SINs that displayed short synaptic latencies (<3 ms) to electrical stimulation of the thalamus and were synaptically connected to one (or more) LGN cells. Thus, they all displayed spatially overlapping ON and OFF subfields, showed weak or no orientation selectivity, and responded in a non-linear manner to drifting grating stimulation (F1/F0 rations of <1). The depth within L4 of these SINs was also similar to the depths of the short-latency SINs. (199 ± 90 μm vs. 184 ± 22 μm, beneath the estimated superficial border of L4, p=0.822).

Rule 3: A third factor explains why some retinotopically-aligned LGN/SIN cell pairs were not synaptically connected despite strong retinotopic alignment and a strong postsynaptic LFP response generated by the LGN neuron in the vicinity of the SIN.

This factor concerns the global synaptic connectivity of the specific L4 SIN with the LGN. Some L4 SINs appear to receive no (or very weak) direct synaptic input from the LGN, as indicated by a failure to respond at a short latency to either strong electrical stimulation of the LGN or visual stimulation. This accounts for the lack of connectivity of some well-aligned LGN-SIN pairs. (A) The distribution of synaptic latencies to electrical stimulation of the LGN, for SINs that were, or were not connected to their aligned LGN neuron. Note that none of the SINs responding at synaptic latencies of >3 ms were synaptically connected to its paired, well-aligned LGN neurons. (B) The relationship between synaptic latency to electrical stimulation of the LGN and the synaptic latency to visual stimulation. (C) For well-aligned connected and unconnected LGN-SIN pairs, the relationship between the amplitude of the postsynaptic LFP response generated by the LGN cell in the network surrounding SIN and the latency of the SIN to electrical stimulation of the LGN. These two variables account for all-but-one (indicated by the red arrow) of the cases in which the LGN/SIN pairs were not synaptically connected, despite precise retinotopic alignment.

Figure 4C summarizes the extent to which (a) the amplitude of the postsynaptic LFP generated by the LGN neuron at the site of the SIN under study, and (b) latency of the SIN response to electrical stimulation of the thalamus account for all-but-one (small red arrow) of the cases in which the LGN-SIN cell pairs were not synaptically connected, despite precise retinotopic alignment. Thus, 31/32 well-aligned LGN-SIN cases were synaptically connected provided that (1) they received a measurable postsynaptic input from the LGN input (spike-triggered post-synaptic LFP amplitude >1 μV), and (2) the synaptic latency to electrical stimulation of the LGN was <3 ms. The only exception to these rules was a single LGN cell with a weak spike-triggered LFP amplitude that did not connect to a retinotopically-aligned SIN (1/31 well-aligned LGN-SIN pairs).

Neighboring ‘regular-spiking’ neurons are more selectively connected to aligned LGN neurons than are SINs

Finally, the very high connection probability between retinotopically-aligned LGN/SINs pairs contrasts with the functional connectivity between LGN neurons and neighboring cortical neurons that were not SINs. We identified 28 neurons that were not SINs, but were located very near (in the same penetration and within 150 μm, either above or below) to a SIN that was synaptically connected to the same LGN neuron. Only three of these neurons (Figure 5A–B) received a functional input from the same LGN neuron (~11%) despite the fact that the spike-triggered LFP generated by the LGN neuron (recorded at the site of this non-SIN) was similar, albeit somewhat lower in amplitude than that measured at the site of the connected SIN (Figure 5B, 7.32 ± 0.59 μV and 8.72 ± 0.69 μV, respectively, p=0.023, paired t test). These non-SINs were ‘regular-spiking’, having considerably longer-duration spikes (0.921 ± 0.028 ms) than the neighboring connected SINs (0.455 ± 0.022 ms, p<0.001, Figure 5C).

Functional connectivity between LGN neurons and L4 non-SINs.

(A)The relationship between the amplitude of the spike-triggered postsynaptic response generated by the LGN neuron, recorded at the site of the non-SIN, and the efficacy of any synaptic connection with the non-SIN. Only three of these non-SINs (red) showed synaptic connectivity. Each of these non-SINs was located within 150 μm (either above or below) of a SIN that did receive a synaptic input from the same LGN neuron. (B) The relationship between the peak amplitude of the spike-triggered postsynaptic responses generated by a single LGN neuron, at the site of a synaptically connected SIN (y-axis) and at the site of a non-SIN (x-axis, located at a vertical distance of <150 μm from the connected SIN). For all data points, the SIN is connected to the LGN neuron, for red points, both the SIN and the non-sin are connected. (C) The frequency distribution of spike waveform durations for the L4 non-SINs shown in A and for the neighboring connected SINs. The L4 non-SINs had considerably longer-duration spikes than did the neighboring connected SINs.

Discussion

The three rules governing TC connectivity onto L4 SINs

We show that three basic conditions must be met to ensure that an L4 SIN will receive a synaptic drive from an LGN concentric cell. The first, and most salient of these is retinotopic alignment. When this sole condition is met, 73% of LGN-SIN pairs display functional connectivity. The second necessary condition to ensure synaptic connectivity of LGN-SIN pairs concerns the strength of the postsynaptic LFP generated near the SIN by the specific LGN neuron under study. We looked only at the peak amplitude of the initial 1 ms of this postsynaptic response to ensure that we were examining a monosynaptic effect. The synaptic impact of some well-aligned LGN neurons may be stratified within L4 and be quite strong at some depths but weak at others (e.g. the case shown in Figure 3). It is not surprising, therefore, that an LGN neuron that generates little or no monosynaptic drive at the cortical depth of the L4 SIN would fail to make a synaptic connection with that SIN, even when retinotopic alignment is precise. In hindsight, this factor may appear obvious, for how could an LGN cell provide a synaptic drive to a SIN if it does not generate synapses near it? However, previous single-cell studies of TC connectivity (i.e. cross-correlation studies) have not taken a measure of the synaptic profile provided by the TC neuron to the region around the cortical cells under study. In previous cross-correlation studies, this factor has simply been unknown. Here, we gained a measure of this by examining the spike-triggered LFP generated near the SIN by spikes of the single LGN cell under study. This is a fairly easy measure to take in cross-correlation studies using extracellular microelectrodes, and we suggest that its general employment would reduce the variability seen in many cross-correlation studies of synaptic connectivity.

The third factor that must be met to ensure a connectivity between LGN neurons and retinotopically-aligned L4 SINs concerns the fact that some L4 SINs do not appear to receive a significant monosynaptic input from any LGN cells. We infer this from our finding that (1) these L4 SINs respond at very long synaptic latencies (>3 ms) to strong electrical stimulation of the LGN, (2) that these cells also respond at long latencies to visual stimulation, and (3) that cross-correlation analysis reveals no synaptically connectivity between these SINs and retinotopically-aligned LGN cells. Notably, these cells responded, like other L4 SINs, with a burst of high-frequency spikes (at >600 Hz) to electrical stimulation of the LGN (albeit, at longer latencies), and their spike durations were as short as those of other SINs. They are, therefore, indistinguishable from other putative fast-spike interneurons, based on traditional extracellular measures (e.g. Bruno and Simons, 2002; Zhuang et al., 2013). Their estimated depth within L4 was also similar to other SINs. Together, these data support the notion that these ‘long-latency’ SINs represent a different sub-class of fast-spike inhibitory interneuron in L4 that is connected to the thalamus in a different manner than are those that respond at short synaptic latencies to LGN stimulation. Similarly, recent studies in several cortical systems indicate that fast-spiking parvalbumin-expressing interneurons may be divisible into multiple functional subclasses (Dávid et al., 2007; Garcia-Junco-Clemente et al., 2019; Shin and Moore, 2019). Our results are supportive of this notion. Surprisingly, these ‘long-latency’ SINs, which seem to lack significant monosynaptic input from the LGN have RF that are indistinguishable from those of SINs which do receive strong monosynaptic LGN input. This, of course, raises the question of how these RFs are synthesized.

The synthesis of RFs in L4 SINs

L4 SINs in rabbit V1 can be divided into two classes: those that respond at short latencies to strong electrical stimulation of the LGN (<2 ms, short-latency SINs), and those that respond at long latencies (>3 ms, long-latency SINs, Figure 4). Aside from the latency of the visual response, the RFs of these two cell classes are virtually identical, consisting of overlapping ON and OFF zones, and poor or no orientation/direction selectivity, and non-linear responses to drifting visual gratings. By contrast, most regular-spiking neurons (non-SINs) in L4 have ‘simple’ RFs consisting of one or more spatially separate ON and/or OFF zones, marked orientation/direction selectivity, and linear responses to drifting visual gratings (Zhuang et al., 2013). Our results show that, virtually all of the short-latency (first-order) L4 SINs receives a highly convergent input from virtually all of the LGN axons that synapse in their vicinity (Figure 4). Thus, the short-latency SINs receive monosynaptic input from ON- and OFF-center LGN axons, as well as from LGN cells with both sustained and transient response properties. The overlapping ON and OFF responding, and the broadly tuned response properties of these SINs are what might be expected to result from such a promiscuous TC input.

A more puzzling question concerns how RFs are synthesized in the ‘long-latency’ L4 SINs. Of course, these SINs might receive some monosynaptic thalamic input (either subthreshold, or very long-latency) despite the fact that strong and synchronous electrical stimulation of the LGN failed to generate short-latency spikes in these cells. Notably, there is little evidence for LGN axons with long conduction times in rabbit, as the vast majority of LGN neurons have TC conduction times of <2 ms (~98%, Stoelzel et al., 2008; Swadlow and Weyand, 1985). Of course, one possibility is that these SINs receive their input from a yet-to-be-discovered slowly conducting LGN pathway. Alternatively, RFs of the long-latency SINs may be synthesized from a highly convergent input from neighboring spiny (simple) cells. Thus, in L4 of somatosensory barrel cortex, fast-spike interneurons receive a powerful and highly convergent input from neighboring spiny (regular-spiking) neurons. This input (spiny to fast-spike) is twice as strong and seven times more probable than is the input to other neighboring spiny neurons (Beierlein et al., 2003). In V1, the L4 spiny neurons are largely simple cells. So, we could speculate that the ‘long-latency’ SINs synthesize their ON/OFF and broadly tuned RFs from such strong and highly convergent input from neighboring simple cells. The result would be a SIN RF with spatially overlapping ON/OFF zones (because the simple cells have both ON and OFF subfields) and minimal orientation tuning (because all orientations are represented by the neighboring simple cells and there are no orientation columns in Rabbit V1). Such disynaptic excitatory input would also be expected to shape the RFs of the short-latency SINs, which would have RFs that are synthesized both from the highly convergent monosynaptic LGN inputs and from the (disynaptic) input from neighboring spiny neurons (Beierlein et al., 2003), resulting in the same RF structure and lack of orientation tuning, but a shorter latency to visual stimulation and electrical stimulation of the thalamus.

The sparseness of TC connections to L4 regular-spiking (presumptive spiny neurons)

By contrast to the profuse TC connectivity seen in L4 SINs, regular-spiking neurons (presumptive spiny neurons), found just above or below an L4 SIN that is connected to an LGN neuron (within 150 microns in the same vertical penetration) have a much lower connection probability (~11%, compare Figures 5A and 3D), despite the proven efficacy of the tested LGN neuron in driving the neighboring SIN. We have previously shown that > 80% of such regular-spiking non-SINs in L4 of rabbit V1 are simple cells (Zhuang et al., 2013). Similarly, in L4 of cat V1, simple cells receive a highly selective input from LGN TC neurons that depends on retinotopic alignment and spatiotemporal RF match, while the inputs to fast-spiking neurons are less selective (Alonso et al., 2001; Reid and Alonso, 1995; Sedigh-Sarvestani et al., 2017). Findings in somatosensory barrel cortex of rats are similar, where regular-spiking L4 neurons receive input from ventrobasal TC neurons at a much lower probability than do L4 fast-spike neurons (Bruno and Simons, 2002).

Comparing the RFs of rabbit L4 SINs with those of other species

These results, as well as previous results in rabbit V1, indicate that the RFs of putative fast-spike GABAergic inhibitory interneurons form a uniform class of ‘complex’ cells, with spatially overlapping ON and OFF subfields and little or no orientation or directional selectivity. In the cat, however, the RF properties of fast-spike interneurons are more heterogeneous. Thus, Hirsch et al., 2003 recorded intracellularly from 10 neurons with smooth dendrites and presumptive fast-spiking properties. They found that four of these cells had RFs that closely resembled the SINs of this study (spatially overlapping ON/OFF subfields, little orientation/direction selectivity), and the remaining 6 of these cells had classic ‘simple’ RFs. Similarly, Nowak et al., 2008 recorded intracellularly and used cluster analysis to segregate cell types based on electrophysiological and RF properties. They found that some fast-spike cells were very broadly tuned to orientation, and they concluded that this class of fast-spike cell was "... likely to correspond to the non-orientation selective, complex inhibitory neurons of layer four described by Hirsch et al., 2003”. By contrast, Cardin et al., 2007 found only small differences between fast-spike and regular-spike cell classes in cat L4. Thus, it seems that the SINs found in rabbit V1 (Swadlow and Weyand, 1987Swadlow, 1988; Zhuang et al., 2013; this study) may have their counterparts in feline V1, but that the well-tuned fast-spike interneurons with ’simple’ RFs found in cat L4 are not found in rabbit V1. These cells have not yet been studied in L4 of mouse (or monkey) V1. In layer 2/3 of mouse, however, inhibitory interneurons of multiple classes were found to be very broadly tuned to orientation and a number of other stimulus features (Kerlin et al., 2010; Sohya et al., 2007). We would speculate that in the mouse (and other rodents), the RF properties of L4 fast-spike neurons will prove to be very similar to those found in rabbit V1.

Conclusions

The promiscuous connectivity seen here between LGN concentric neurons and retinotopically-aligned L4 SINs is reminiscent of the ‘complete transmission line’ between nodes of a network, described by Griffith, 1963, and elaborated by Abeles, 1991 (who adapted such a network as the basis of his ‘synfire chain’). In such networks, each member of one node excites each member of the successive node via richly convergent/divergent synaptic connectivity. Neurons in the output node (SINs, in this case) are expected to show high sensitivity (because of their profuse inputs), but they suffer a decreased ability to discriminate among any selective properties of the input elements. Therefore, neurons in the output node will be broadly tuned to stimulus features represented by cells at the input node. These network characteristics nicely describe many of the response properties of L4 SINs in rabbit V1 (Zhuang et al., 2013). Thus, these cells are more sensitive to stimulus contrast than are L4 simple cells and have a spatial RF that consists of spatially overlapping ON and OFF subfields (despite the fact that the concentric LGN input cells responded to either ON or OFF in the RF center). Moreover, they are more broadly tuned to a number of stimulus parameters (spatial and temporal frequency, stimulus direction and orientation) than are simple cells of V1.

Similarly, SINs of somatosensory barrel cortex are synaptically connected to neurons in ventrobasal thalamus in a richly divergent/convergent manner. In that system, cross-correlation studies reveal a connection probability of ~2/3 between SINs in an L4 barrel, and TC neurons of the topographically aligned thalamic barreloid (Bruno and Simons, 2002; Swadlow et al., 2002). Notably, SINs of S1 have RF properties that are highly analogous to those of V1, and their RF response properties are highly consistent with the expected properties of the ‘complete transmission line’ of Griffith, 1963. Thus, they display ‘high sensitivity’, showing lowest thresholds to whisker stimulation of any neurons found in the S1 barrel, and they show very broad temporal and directional tuning (Swadlow, 1989; Swadlow, 1995). The lack of directional tuning in SINs of L4 barrel cortex is thought to result from the highly convergent input to these from multiple neurons in their aligned thalamic barreloid that display different directional preferences (Swadlow and Gusev, 2002).

These results have implications for our current understanding of TC development and function. They suggest that TC pruning during development is likely to be more pronounced in excitatory than in fast-spike inhibitory cortical neurons leading to differences in their RF selectivity (Alonso and Swadlow, 2005; Cardin et al., 2007; Bruno and Simons, 2002). This connectivity difference is likely to generalize across sensory systems and species. For example, in both visual and somatosensory cortex, each thalamic afferent is known to make connection with a small percentage of cortical excitatory neurons that are retinotopically or somatotopically aligned, a percentage that is estimated to be ~15 to 33% in cat visual cortex (Alonso et al., 2001; Sedigh-Sarvestani et al., 2017), 11% in rabbit visual cortex (as shown here, Figure 5A), and 37% in rat barrel cortex (Bruno and Simons, 2002). Also, in both visual and somatosensory cortex, each thalamic afferent makes a higher percentage of connections with fast-spike inhibitory than excitatory neurons (Swadlow and Gusev, 2002, Alonso and Swadlow, 2005; Bruno and Simons, 2002; Sedigh-Sarvestani et al., 2017). While these similarities in TC connectivity are well known, all estimates of connection probability reported in the past failed to take into account the laminar specificity of the axonal arbor from the thalamic afferent and the presence of L4 cortical neurons that may receive weak or no direct thalamic input.

Remarkably, our results indicate that, when the laminar specificity of the axonal arbor is taken into account, each thalamic afferent makes connection with nearly all of the first-order (short-latency) fast-spike inhibitory neurons that are within reach of the axonal arbor regardless of the response properties of the LGN and cortical neurons. This is clearly not the case with neighboring L4 regular-spiking neurons. The ‘three rules’ of synaptic connectivity that we describe are both necessary and sufficient for predicting connectivity between LGN neurons and L4 SINs (40/41 cases, Figure 4C). Together, they predict connection probability with a surprising high level of precision that is not yet possible for any other TC circuit in any other species. This nonselective pooling of thalamic afferents can be amplified by nonselective pooling of inhibitory neurons connecting to the same cortical excitatory neuron (Taylor et al., 2018).

In conclusion, we have shown that concentric LGN TC neurons form functional synaptic connections with retinotopically-aligned L4 SINs at a very high probability (~73%). This probability increases to nearly 100% when considering two additional factors: whether the aligned LGN TC neuron synapses at the depth in L4 where the SIN is located, and whether the aligned L4 SIN receives short-latency input from any LGN cells. Thus, virtually every first-order fast-spike interneuron in layer four receives input from nearly all of the LGN axons that synapse nearby (40/41 cases). We propose that the observed promiscuous, highly convergent TC connectivity onto retinotopically/topographically aligned L4 fast-spike inhibitory interneurons may be a general feature of sensory neocortex, and that this feature plays a causal role in generating the fast, sensitive, broadly tuned, and powerful feed-forward TC inhibition of excitatory neurons that is seen within L4 (Bagnall et al., 2011; Cruikshank et al., 2007Cruikshank et al., 2010; Gabernet et al., 2005; Hull et al., 2009; Miller et al., 2001b; Swadlow, 1995; Swadlow, 2003; Taylor et al., 2018).

Materials and methods

Experimental procedures and data analysis

Request a detailed protocol

Extracellular single-unit recordings were obtained from LGN neurons and from the retinotopically-aligned region of V1 in four awake adult female Dutch-Belted rabbits. The general surgical procedures for chronic recordings and anesthesia have been described in detail previously (Swadlow et al., 2002; Bereshpolova et al., 2007; Stoelzel et al., 2008; Zhuang et al., 2013) and are reported only briefly here. All animal procedures were conducted with the approval of the University of Connecticut Animal Care and Use Committee in accordance with the National Institutes of Health guidelines.

Initial surgery was performed under ketamine-acepromazine 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. The silicone rubber was used to buffer the wound margins and was covered with the acrylic cement to create base for an attachment of chronically implanted equipment. A layer of acrylic cement also always covered the exposed skull between recording sessions.

A concentric array of seven independently movable electrodes was placed within the LGN (electrode separation ~200 um, Swadlow et al., 2005). Recordings were made using fine-diameter (40 micron) quartz-insulated platinum/tungsten electrodes tapered and sharpened to a fine tip. Extracellular single-unit recordings 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. All electrophysiological activity was recorded in the awake state (e.g. Bereshpolova et al., 2007; Bereshpolova et al., 2019; Bezdudnaya et al., 2006; Zhuang et al., 2013) and acquired using a Plexon data acquisition system (Plexon, Dallas, TX). Signals were amplified, bandpass filtered and sorted to identify single units. LFPs were also recorded, filtered at 2 Hz to 1.9 kHz (half-amplitude), and sampled continuously at 10 kHz.

Visual stimulation and RFs analysis

Request a detailed protocol

All stimuli were generated using custom-made program (Visual C++, DirectX 7), and presented on a CRT monitor (Nec MultiSync 40 × 30 cm, mean luminance 48 cd/m2, refresh rate 160 Hz). RFs were mapped using sparse noise (Jones and Palmer, 1987), made of white and black squares (0.5–2° in a grid of 30 × 22 degrees) in a pseudorandom sequence, and matrices of the raw ON and OFF RF data were generated by reverse correlation method. A Gaussian filter and bicubic interpolation were applied to facilitate visualization. The luminance of gray background was adjusted so the contrasts from the white and black squares were equal.

Spatial RF maps were calculated for a series of time delays (using sliding time window with 1 ms step and 10 ms duration) between stimulus and response. For each time delay after stimulus onset, the spatial RF was computed by averaging spikes that follow the presentation of the stimulus at each position of the grid. The average spatial responses at each time delay were fit with a polynomial function. From these fits, we extracted temporal RF parameters such as latency of visual response (defined as the time at which the variance first crosses the three times SD above the mean variance of noise at delay 0–10 ms after stimulus onset), peak latency of the response (measured as the time at which the interpolating function reached maximum value), duration of the response (defined as the full width of the interpolating function at half maximum value).

To assess whether the spatial RF in the restrict peak area of SINs was dominated by ON or by OFF subfields, a ‘sign index’ was calculated (Van Hooser et al., 2013):

Sign index=| RON( i,j) ROFF(i,j) |RON( i,j)+ ROFF(i,j),

Where RON(i,j) is the response to all bright squares at position i,j and ROFF(i,j) is the response to all dark squares at position i,j. The sign index ranged from 0 to 1, with 0 representing a balanced spatial RF, and one an ON-dominated or OFF-dominated spatial RF.

The sign index and response latency of RF were used to define the strongest subregion of cortical cell for assessing the degree of spatial overlap with LGN RF.

Assessing retinotopic alignment

Request a detailed protocol

To achieve precise retinotopic alignment between thalamic and cortical recording sites, 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 with each other, and with the LGN neuron, based on their highly overlapping RF locations. In some cases, the mapping electrode was replaced with the 16-channel probe. Once RFs were tested and mapped, a large number of spontaneous spikes (usually several thousand from each neuron) was recorded.

To quantitatively evaluate a spatial relationship between thalamic and cortical RFs, spatial RF structures at the peak response latency were reconstructed by calculating the center of mass of the response in visual space using the absolute value of all significant pixels in each RF. Then RF maps of both sites were fitted with an ellipse. Ellipse parameters such as width, height and elongation angle were extracted using the covariance matrix, eigenvalues and eigenvectors of a complex Hermitian. This approach provided the center position and aspect ratio of the spatial RFs. The distance between the spatial location of the RF center of the LGN cell and the center of the SIN 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.

Cells and layers identification

Request a detailed protocol

Identification of cortical layers was based on the reversal point of the field potential generated by a diffuse flash stimulus (Stoelzel et al., 2008). SINs were identified by firing a burst of 3 or more high-frequency spikes (>600 Hz) to electrical stimulation of the LGN (Swadlow, 1989; Swadlow, 2003; Zhuang et al., 2013). SINs also had spikes of very short duration (Figure 1—figure supplement 1).

Sustained/transient identification: A flashing stimulus was presented over each cell’s RF (at least 1 s on, 1 s off). Baseline activity was calculated as the average firing rate of the neuron for the period preceding the start of the stimulus for a half a second. Maintained discharge was calculated from the average firing rate from 0.5 to 1 s after stimulus onset. Concentric cells of LGN, which had a maintained response above baseline of more than 10 spikes per second, were classified as sustained (Bezdudnaya et al., 2006; Cano et al., 2006). The cortical sustained SINs were required to have a maintained response at least two times higher than baseline. Cells that had a ratio lower than two were classified as transient (Zhuang et al., 2013). Note that whereas ‘transient’ neurons have only a transient response component, ‘sustained’ neurons may have both.

LFP and CSD analysis

Request a detailed protocol

Methods and rationale for localizing the pre-and postsynaptic responses generated by single TC neurons through the depth of the cortex have been described previously (Swadlow et al., 2002; Stoelzel et al., 2008; Jin et al., 2011). Spontaneous extracellular single-unit activity of the LGN cells was recorded along with the LFPs generated through the depths of the retinotopically-aligned region of V1. Spike-triggered averages of LFPs were generated from spontaneous LGN spikes. One-dimensional CSD profiles were computed from the voltage traces of the field profiles according to the method described by Freeman and Nicholson, 1975. Estimates for the CSD at the top and bottom recording sites were provided by the method of Vaknin et al., 1988. In the CSD profiles and their color maps, current sinks (red) are indicated by downward deflections and sources (blue) are indicated by upward deflections.

In order to assess the extent of the postsynaptic response generated by the LGN cell near the SIN under study (‘Rule 2’ in Figure 3), we measured the spike-triggered average responses generated on the probe channel (or single electrode) on which the SIN was recorded. An LGN neuron was considered to have generated a postsynaptic impact at the depth of recorded SIN after fulfilling the following two conditions (1) presence of a clear axon terminal potential ≥0.75 μV in amplitude; and (2) the postsynaptic response had to follow this axonal component by <1 ms and consist of a sharp negativity in the spike-triggered LFP of >1 μV in amplitude. To be sure that we were measuring the amplitude of the monosynaptic spike-triggered LFP, we measured the peak amplitude of this response during the initial 1 ms following the onset of the postsynaptic response.

Assessing TC connectivity

Request a detailed protocol

The connectivity within simultaneously recorded geniculocortical pairs of neurons was assessed by using the cross-correlation analysis. Monosynaptic connection was inferred from the presence of a significant peak in the probability of SIN firing at intervals of 1–4 ms (reflecting the presynaptic axonal conduction time, the synaptic delay, and the rise-time of postsynaptic firing). 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.

After detecting a significant peak in the cross-correlogram, we determined an efficacy value (Levick et al., 1972) based on a brief window (+0.6 ms) centered on the peak. Efficacy was calculated by counting the number of SIN spikes occurring during this window, subtracting the baseline number of expected spikes, and dividing this value by the number of triggering TC spikes. The number of SIN spikes expected by chance within each bin was based on the mean number of spikes per bin that occurred during a 5 ms window, from 4 ms before to 1 ms after the TC spike.

All the p values provided in Results represent the results of independent sample t test, if not specified. Data are provided as mean ± SEM.

Data availability

All data generated or analysed during this study are included in the manuscript and supporting files.

References

    1. Levick WR
    2. Cleland BG
    3. Dubin MW
    (1972)
    Lateral geniculate neurons of cat: retinal inputs and physiology
    Investigative Ophthalmology 11:302–311.

Decision letter

  1. John R Huguenard
    Senior Editor; Stanford University School of Medicine, United States
  2. Diego Contreras
    Reviewing Editor; University of Pennsylvania School of Medicine, Department of Neuroscience, United States
  3. Barry Connors
    Reviewer; Brown University, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Thank you for submitting your article "Three rules govern thalamocortical connectivity of fast-spike interneurons in the visual cortex" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a Guest Reviewing Editor, and the evaluation has been overseen by John Huguenard as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Barry Connors (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, we are asking editors to accept without delay manuscripts, like yours, that they judge can stand as eLife papers without additional data, even if they feel that they would make the manuscript stronger. Thus the revisions requested below only address clarity and presentation.

Summary:

This manuscript shows, in the visual system of the adult rabbit, that LGN neurons and V1 suspected interneurons (SINs) in L4 show positive cross-correlograms when (1) RFs overlap spatially more than 50%, (2) LGN spikes trigger a measurable LFP response near the recorded SIN and (3) SIN have short latencies to electrical (< 3ms) and visual (< 30 ms) stimulation. SINs with longer latencies (n=5) supposedly do not receive LGN input and do not show positive cross-correlograms even with 100% spatial overlap. Thus, the results show that direct synaptic connections between LGN cells and SINs are predicted by three properties of the cells involved, and are insensitive to other response features. The study represents another important step forward in the understanding of thalamocortical transmission of sensory signals to neocortical layer 4.

Revisions:

1) The three reviewers agreed that the article is clearly written, well-documented and illustrated and very pleasant to read, the methodology is powerful and flawless, the results are convincing. The study represents another important step forward in the understanding of thalamocortical transmission of sensory signals to neocortical layer 4. However, the three reviewers agree on the confusing nature of the terminology pre and postsynaptic and the need for additional analysis of non-SIN neurons.

2) Issue regarding terminology:

One reviewer is concerned that the term mismatch applied to the sign of the RF and the time course of the response seems misleading. For example, I don't see a mismatch in time course in Figure 2 since the SIN response has both a transient and a sustained response and the transient response is similar in time to the LGN response. Other LGN or cortical neurons, which you are not recording from, are responsible for the sustained portion of the SIN response. One could say that the matching is incomplete but not that there is a mismatch as the time course of the LGN input is clearly expressed in the SIN response. Similarly with the RF sign overlap, the ON response of the SIN in Figure 2C is obviously caused by an ON LGN cell which you are not recording from, but there is a clear response to the OFF LGN cell. Under the authors definition of "mismatch" a SIN and an LGN neuron will never "match" since SINs have overlapping ON and OFF subregions.

3) Main issue concerning pre and postsynaptic components of the response:

The summary of the comments posted below concerns the use of the terms pre and postsynaptic. Essentially, the LFP amplitude is generated by postsynaptic processes, thus, describing it as presynaptic process is confusing. For example, changes in input resistance caused by changes in brain state would greatly change the amplitude of the LFP. An indirect measure of the presynaptic input would be provided by the sink caused by the invasion of synaptic terminals by incoming action potentials, which the authors showed in the past but did not provide here. Furthermore, response delay as a presumption of lack of LGN input is just as much pre as it is postsynaptic, in fact it refers to the same phenomenon as rule 2: size of LGN input, which in rule 3 is zero. Ultimately, both measures are indirect measures of connectivity and synaptic activation and the otherwise unambiguous terms pre and postsynaptic make the issue very confusing. All three reviewers view this terminology and the presentation of the results as oversimplifying and overselling conclusions since extracellular recordings of spikes and fields provide only indirect evidence of pre and postsynaptic measures, they do not provide any specific information on actual pre and postsynaptic processes. Thus, the authors should clarify the terminology throughout and do not utilize those terms. More importantly, the authors should pursue the analysis of the non-SIN neurons (Supplementary figure 2) to determine whether they also follow the three rules (see "e" below).

4) Issue of context:

The manuscript would be more appealing if it had a more general introduction, and a discussion that touches on the implications of their connectivity rules for sensory processing, cortical microcircuitry, and even the development of thalamocortical synapses onto SINs. Their data do provide some important clues about microcircuitry. There are, for example, SINs in their sample that do not receive monosynaptic thalamic input, yet the cells seem to reside in the termination zones of TC terminals. Something must determine whether TC axons connect or not to particular SINs, although the paper does not hint at anything mechanistic.

While the essential revisions are listed above, for the benefit of the authors we are pasting the detailed comments of the reviewers below in their entirety in case they find them useful in revising their resubmission:

a) Figure 3D shows a nice correlation between local spike-triggered-LFP amplitude and LGN-SIN efficacy. This is a beautiful result, but it raises a couple of questions of data analysis and semantics. If I understand the senior author's previous work on st-CSDs, the method identifies current sinks generated by both LGN intracortical axon spikes and sinks generated by LGN synapses upon cortical neurons (with the sinks perhaps generated mainly by vertically oriented pyramidal cells). If that's correct, then the only purely "presynaptic" component is the short-latency, brief, upwardly propagating axonal sink component. The longer latency, more localized component is the "postsynaptic" response. While the later sink obviously also depends on presynaptic components (synaptic functions require pre and post contributions, by definition), the sink is directly generated by postsynaptic mechanisms.

b) My questions: Does LGN-SIN efficacy correlate with any CSD measures? Figure 3D plots the peak postsynaptic LFP amplitude only, whereas the CSDs provide (in principle) additional spatial information (at least along the axis of the electrode array). CSDs also provide a true "presynaptic" measure, the early sinks, whereas the later sinks combine measures of pre- and postsynaptic functions. In the two example pairs shown in Figure 3, there is not only a big difference in the laminar locus of the LGN-triggered responses, but the axonal CSDs are very different in amplitude. This seems like a possible confound. Are the axonal sinks/spikes too small and unreliable to analyze across the cell sample? My first semantic quibble is about calling Rule 2 a "presynaptic factor", considering that the measure used to deduce it (the late L4 LFP) is largely a postsynaptic process.

c) To continue my pedantic semantic questions, why is Rule 3 a "postsynaptic factor", any more than Rule 2 is a "presynaptic factor"? Both designations seem to confuse the mechanistic issues. Rules 2 and 3 both involve different features of the same basic mechanism: either LGN axons form synapses on the SIN in question, or they don't. While the number and strength of those synapses are important factors, the methods used here cannot disambiguate those variables or infer anything exclusively pre- or postsynaptic about the connections. Rule 2 is about local connectivity (a SIN may make strong synapses with some LGN cells, but not with the cells being recorded in a particular experiment); rule 3 is about global connectivity (some SINs apparently don't get monosynaptic input from any LGN cells). This is an enormously important result, but I worry that the authors' semantic choices may obscure their conclusions for some readers. Labeling the rules "presynaptic" or "postsynaptic" is misleading, in my opinion. The problem is particularly acute in the Abstract and Introduction, where the rules are stated but only identified by their presynaptic/postsynaptic monikers and not further explained. The Discussion, by contrast, provides a very clear and compelling explanation of the authors' conclusions.

d) Although the results are very interesting they may be actually trivial, as the authors honestly mention in their Discussion. Indeed, rules 2 and 3 are probably simply spatial and temporal functional evidences of direct anatomical connection. Rule 2 is spatial and is a way to know whether the input arrives close to the recorded cell. Rule 3 is temporal and can be interpreted as expected delays between first-order and second-order cells (see Bullier et al., 1980 for instance). So if there is not the proper spatial conditions – the main input not in the right location – and/or not the proper temporal conditions – too long latency to be a first-order cell -, then it is entirely expected and logical that there is the connection is indirect. The results are really expected. Would rules 2 and 3 be linked to functional properties such as the RF structure or selectivity, as the authors also unsuccessfully investigated, the results would have had stronger impact because it would help understanding the relationship between the structure of the thalamo-cortical connectivity and the emergence of function. Currently, the benefit of these 3 rules is, as discussed, a methodological improvement for more unequivocal estimation of a direct connection: "This is a fairly easy measure to take in cross- correlation studies using extracellular microelectrodes, and we suggest that its general employment would reduce the variability seen in many cross-correlation studies of synaptic connectivity. " I agree it is an important methodological improvement for the field, but I feel not suited for eLife.

e) However, there may be one possibility, with the current dataset, to potentially increase the impact of this result. I suggest that the authors make a complete parallel analysis of the non-SIN neurons analysed in Supplementary figure 2. Adding the information on the retinotopic alignment (rule 1) and latency of the response (rule 3) will allow to state whether these neurons obey as well to the 3 rules or not. If they do, as is expected, then the 3 rules are really 3 independent functional measure of direct connection that can be applied to any neurons. If they don't, then it seems that the result is less trivial than expected, that will make it probably more interesting. The discussion of the potential circuitry and impact of such results on the thalamo-cortical connectivity would make it more suitable for eLife.

f) I don't really agree with the Abstract sentence "These rules concern (1) retinotopic alignment, (2) a property of the presynaptic axon, and (3) a property of the postsynaptic cortical interneuron". Rule 2 is not really a property of the presynaptic axon, which could be interpreted as conduction velocity for instance, but simply the laminar localization of the presynaptic terminals. The authors have not shown that rule 3 is a property of the cortical interneuron (integration time or Magno vs. Parvo cells for instance), since it can just be first vs. second-order cells. There are three major evidences in favor of the second hypothesis : 1) it is predictive of direct connection, 2) the authors haven't been able to relate the difference of connectivity to "sustained" or "transient" cells categories and 3) actually short and long latency SIN are undistinguishable functional properties. I think rule 2 would be better described as spatial connectivity profile and rule 3 has temporal properties that reflects 1st vs. 2nd order cells.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your article "Three rules govern thalamocortical connectivity of fast-spike inhibitory interneurons in the visual cortex" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Guest Reviewing Editor and John Huguenard as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Barry Connors (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, we are asking editors to accept without delay manuscripts, like yours, that they judge can stand as eLife papers without additional data, even if they feel that they would make the manuscript stronger. Thus the revisions requested below only address clarity and presentation.

Summary:

The authors largely addressed all concerns raised in the first round of review. Overall, the impression remains positive both about the importance of the results and suitability for eLife. However, important concerns were raised that need to be addressed concerning language and presentation.

The main points are:

1) Better definition of the term "postsynaptic". See detailed explanation in reviewer 2 comment.

2) Clarification of SINs in comparison to layer 4 interneurons in other species. See comments by reviewer 1.

3) Enhancement of the main point, which is written in the response letter, that "virtually every first-order fast-spike interneuron in layer 4 receives input from nearly all of the LGN axons that synapse nearby (40/41 cases)". See comments by reviewer 3.

In addition to the written reviews, the three reviewers had a lengthy discussion regarding the revised manuscript. The reviewers praised the difficult and elegant technique used. Three key points of the discussion were:

1) The non-trivial nature of the high % of LGN inputs to SINs. It was noted that the main point that needs to come across to the reader is the sparseness of connections to RS cells, which are neighbors to the SINs and yet receive much less input (11% in this manuscript, but other examples in the literature are key to make the point). For those that have spent long hours (and years) chasing monosynaptic connections between thalamus and cortex, the numbers provided here are indeed surprising. However, without a clear notion of the difficulty and probability of finding connected pairs the three rules may appear rather trivial. The three reviewers agreed that in presenting the results, the authors should make an effort to convey this notion. When recording pairs, in which the cortical cell is an RS, a connection can be explained by the respective spatial and temporal properties of the two cells in the pair. But a lack of connection can almost never be explained. Why are thalamocortical axons so picky? The surprising finding here is that the authors are able to explain all cases in which there are NO connections based on the three rules. Which obviously means: 1) they can explain connectivity to all SINs and 2) SINs get a lot more LGN input than RS cells. These two points are novel and surprising and have convinced the reviewers of the suitability of the manuscript to eLife. However, it should be made clear also that a lot remains to be done to understand why RS cells are sparsely connected.

2) The perhaps unique nature of SINs. As indicated by reviewer 1, fast-spiking (FS) interneurons in cats and monkeys are simple cells with spatially separate RF subregions (the jury is still out on mice it seems). SINs in rabbits are, instead, complex cells. Perhaps they fulfill a unique functional role in rabbits, and for that receive many more connections. Or perhaps SINS and the FS cells typically identified in cat and monkey neurophysiological studies are distinctly different subtype of neurons. Can the authors provide more insight into the comparison of rabbit SINS and cat/monkey FS cells?

3) Emphasis and clarification must be placed in the existence of a group of SINs that do not receive any connection from LGN, which suggests the existence of two types of SINs with perhaps very different functional roles. The two main potential sources of excitatory input to SINS are LGN synapses and synapses from local spiny cells. Data suggest two groups of SINS: Most are driven by strong, convergent LGN input, and this makes them sensitive and visually nonspecific. Another small group receives no LGN input, yet they are still visually sensitive and nonspecific, suggesting they are largely driven by strong, convergent input from local spiny cells, which is entirely consistent with a variety of anatomical and paired-cell in vitro recording studies that the authors do not cite, local FS interneurons in general have high probabilities of connectivity (~0.5) with spiny neurons in their local neighborhood.

This point was also viewed by the reviewers as a key and novel result that needs emphasis and clarification.

Reviewer #1:

An important notion that needs to be made very clear in the manuscript is that there are no interneurons in layer 4 of cat or monkey V1 that look like SINs. The vast majority of neurons in thalamorecipient layer 4 are simple cells, i.e. they have segregated ON and OFF subregions (apart from center surround neurons in monkey which are not relevant to this Discussion). In contrast, SINs are complex cells with overlapping subregions of opposing contrast. That makes SIN a unique category of cells and perhaps that is the reason why they have such large percentage of input compared to cat L4 neurons. Based on that, comparisons of LGN % connectivity from other species are only mildly informative and the best comparison is with neurons in rabbit cortex itself. (which the authors don't fully provide).

Comparisons with barrel cortex are tricky because the functional properties that may determine convergence are not well understood, although sparse connectivity of VB to barrel neurons does enhance the uniqueness of SINs notoriously promiscuous connectivity.

So the authors show that, if there is RF overlap (rule1) and thalamic input to the vicinity of the SIN (st-LFP, rule2), the SIN will receive a connection, unless the SIN does not receive thalamic input at all (e-stim,rule3). Thus, SINs receive a lot more input than RS cells, which, under the same circumstances, only receive a small percentage of the available thalamic input. Sure, this is interesting and important, but by itself it does not create a universal new knowledge about L4 gabaergic interneurons because of the uniqueness of SINs. Comparing their promiscuous input to cat or monkey L4 interneurons is incorrect as in those two species FS neurons in layer 4 are simple cells with segregated subregions and are selective to orientation, direction and spatial frequency (SINs are very broadly tuned as shown by the authors in Zhuang et al). The broader tuning properties of layer 4 FS cells in cat and monkey is no match for the SINs.

Reviewer #2:

The authors have done an excellent job of responding to the reviewers' comments. Their revised manuscript provides clearer explanations of the "three rules", and new additions to the Introduction and Discussion broaden the context and appeal of the results.

I have one small suggestion. While the authors were very responsive to the reviewers' concerns about the meanings of "presynaptic" and "postsynaptic" in the original manuscript, the revision still has a phrase that will (in my opinion) confuse many readers. It first appears in the Abstract as a summary of rule #2: "strength of the postsynaptic response elicited near the interneuron by the LGN neuron", and again in a slightly different form in the Introduction.

Confusion will arise because is unclear In the Abstract and Introduction what "postsynaptic response" is referred to. Most readers' default assumption will be that it refers to the response of the postsynaptic neuron under study (i.e. a SIN), and the readers' own response might be to think, "well, of course you need a strong postsynaptic response from a SIN to consider it connected to the LGN cell". That's obvious and uninteresting.

But actually, the authors are referring to the "spike-triggered postsynaptic LFP response generated by the LGN neuron" (spelled out clearly much later in the paper). In other words, the authors' "postsynaptic measure, i.e. an st-LFP (or CSD), is derived not from the SIN under study but from other neighboring unidentified neurons; another reasonable interpretation of this measure is that the st-LFP/CSD is an estimate of the strength and location of the presynaptic terminals of the single LGN cell under study.

Bottom line: I think the authors need to find a better way to state their rule #2 early in the paper if they want their readers to appreciate the point.

Reviewer #3:

In their revised version the authors addressed correctly and convincingly most of the concerns of the first review and the papers gains globally clarity. There remains a substantive concern regarding the analysis of non-SIN neurons

The authors have explained lacking experimental data to fully address the question and this is granted. However, can the authors be less conservative ?

It is understood that the authors do not have enough data to check for the 1st and 3rd rule. However, the result for rule 2 is quite different: only 11% of them are connected compared to 73% for the SIN, comparing Figure 3D with 5A is quite convincing that nonSIN and SIN behave differently.

This would be important to help the author convince the reader that the behavior of non-SIN are different and hereby be more convincing that their result is not trivial.

What they answered to this concern, and their discussion is not fully convincing in the current state. In their answer, the authors stated that : “these (3 rules) "necessary conditions" are the ONLY requirements (other than retinotopic alignment) that must be fulfilled to generate a functional synaptic connection. That these three requirements are both necessary and sufficient is new, and that is surprising.”

It is new indeed, but at first glance do not seem surprising (see below).

The authors further argue that:

“All previous measures of thalamocortical connectivity indicated that only a subset of cortical neurons located within the axonal field of a thalamic afferent receives connection from that afferent. This is true for regular spiking neurons in the cat visual cortex (15%-33%, Sedigh-Sarvestani et al., 2017; Alonso et al., 2001), regular spiking neurons in rat barrel cortex (37%, Bruno and Simons, 2002) and fast spiking neurons in rat barrel cortex (63%, Bruno and Simons, 2002). This specific wiring is also true for other stages of sensory processing (e.g. 4/20, 20% for retinogeniculate connections in cats, Hamos et al., 1987). Given the strong evidence from the past that geniculate afferents make connection with only a subset of cortical neurons, our results are clearly surprising”

This argument concerns rule 2 only and demonstrate that, in itself, it is not sufficient. So further constrains must exist, as a consequence it does not make their result surprising, but rather expected at first glance. And alignment of RF and first-order cell are not the most surprising parameters to look at for further constrains.

Their last sentence “This finding is clearly unexpected and has major consequences for understanding how thalamocortical inhibitory networks develop and are functionally organized.” seems like an overstatement.

Note from editor: this comment was made prior to the discussion among the reviewers. The reviewers have agreed on the suitability for eLife upon discussion. To make the current manuscript more convincing and suitable for eLife, I believe the authors have more work to do. First, maybe by exploiting more the large difference between Figure 3D and 5A. Second, by improving their discussion on that point that is not clear and not as convincing as their reply to our first review (“That these three requirements are both necessary and sufficient is new, and that is surprising”):

It is not clear what is the logical flow that leads to some sentences, for instance “Under this scheme, excitatory cortical neurons become selective to sensory stimuli through two complementary mechanisms: a selected pool of thalamic inputs that build the basic receptive field structure and a nonselective feed-forward inhibitory network that sharpens receptive field selectivity.”. They speak about “different type of fast spiking cells” which is misleading (what “types” are they referring to ?). “Each thalamic afferent makes connection with nearly all of the first-order fast-spike inhibitory neurons that are within reach” is a circular argument, if they make connection they are first-order cells, this argument could be better written.

https://doi.org/10.7554/eLife.60102.sa1

Author response

Revisions:

1) The three reviewers agreed that the article is clearly written, well-documented and illustrated and very pleasant to read, the methodology is powerful and flawless, the results are convincing. The study represents another important step forward in the understanding of thalamocortical transmission of sensory signals to neocortical layer 4. However, the three reviewers agree on the confusing nature of the terminology pre and postsynaptic and the need for additional analysis of non-SIN neurons.

We thank the reviewers for their careful and generally positive review of our work. We believe that we have complied with all reviewers’ requests, save one, the request for a “complete parallel analysis of the non-SIN neurons”. Although we have done some further analyses of these data (below, new Figure. 5B) We feel that we cannot do the “complete parallel analysis” requested without extensive further experimentation This is because we have limited receptive-field measures on these cells, little data on the latency of their responses to thalamic electrical stimulation and visual stimulation, and a very limited number that were connected (n = 3, Figure. 5). Please see our further discussion of this below.

2) Issue regarding terminology:

One reviewer is concerned that the term mismatch applied to the sign of the RF and the time course of the response seems misleading. For example, I don't see a mismatch in time course in Figure 2 since the SIN response has both a transient and a sustained response and the transient response is similar in time to the LGN response. Other LGN or cortical neurons, which you are not recording from, are responsible for the sustained portion of the SIN response. One could say that the matching is incomplete but not that there is a mismatch as the time course of the LGN input is clearly expressed in the SIN response. Similarly with the RF sign overlap, the ON response of the SIN in Figure 2C is obviously caused by an ON LGN cell which you are not recording from, but there is a clear response to the OFF LGN cell. Under the authors definition of "mismatch" a SIN and an LGN neuron will never "match" since SINs have overlapping ON and OFF subregions.

We agree with the reviewer’s argument that the terms “match” and “mismatch” are too strong for describing the ON/OFF agreement between LGN neurons and SINs, because LGN cells are either ON or OFF centered, but SINs have both ON and OFF responses. However, SINs may be ON-dominated or OFF-dominated, and we do maintain that it is appropriate to say that the response properties of an ON center LGN response and an ON-Dominated SIN are “more similar” than are the response properties of an ON-center LGN cell and an OFF-dominated SIN. Therefore, we have eliminated the match/mismatch terms in favor of the terms similarity/dissimilarity (a less dichotomous distinction).

We have done the same for sustained vs. transient responding. Please note, however, that there is a 40+ year history in the literature classifying visual neurons (Retina, LGN and cortex) as sustained or transient based on the presence or absence of a sustained response component (e.g. Cleland, Levick and Sanderson, J. Physiol,, 1973; Cleland et al., 1976, Bezdudnaya et al., 2006; Cano et al., 2006; Stoelzel et al., 2008; Zhuang et al., 2013). In these schemes, “transient” cells have only a transient response component, while “sustained” cells may have both.

3) Main issue concerning pre and postsynaptic components of the response:

The summary of the comments posted below concerns the use of the terms pre and postsynaptic. Essentially, the LFP amplitude is generated by postsynaptic processes, thus, describing it as presynaptic process is confusing. For example, changes in input resistance caused by changes in brain state would greatly change the amplitude of the LFP. An indirect measure of the presynaptic input would be provided by the sink caused by the invasion of synaptic terminals by incoming action potentials, which the authors showed in the past but did not provide here. Furthermore, response delay as a presumption of lack of LGN input is just as much pre as it is postsynaptic, in fact it refers to the same phenomenon as rule 2: size of LGN input, which in rule 3 is zero. Ultimately, both measures are indirect measures of connectivity and synaptic activation and the otherwise unambiguous terms pre and postsynaptic make the issue very confusing. All three reviewers view this terminology and the presentation of the results as oversimplifying and overselling conclusions since extracellular recordings of spikes and fields provide only indirect evidence of pre and postsynaptic measures, they do not provide any specific information on actual pre and postsynaptic processes. Thus, the authors should clarify the terminology throughout and do not utilize those terms.

We agree completely that this is unduly confusing and have made the appropriate changes throughout the manuscript.

More importantly, the authors should pursue the analysis of the non-SIN neurons (Supplementary figure 2) to determine whether they also follow the three rules (see "e" below).

We would like very much to fully comply with this this important suggestion. However, we do not have enough data on the non-SINs. Importantly, only three non-SINs (3/28, ~11%) showed connectivity (Supplementary figure 2), despite the fact that they were very near (< 150 μm vertically) to SINs that were connected to the same LGN axon, and the spike-triggered postsynaptic LFPs elicited by the same LGN neuron were nearly as strong near the non-SIN as they were near the SINs (see new Figure. 5B and associated text). Moreover, we did not record the response latencies to electrical or visual stimulation for many of the non-SINs, or study their receptive fields in sufficient detail. Therefore, we would not, without extensive further data collection, be able to determine whether our “rules” apply to these cells as they do for the SINs. We are hopeful that the reviewers would agree with our conservative view: That there is, at present, simply not enough relevant data on the non-SINs to make convincing arguments beyond the very limited statements that we made (i.e. that regular-spiking non-SINs were less frequently connected than were SINs that were found in the same terminal LGN terminal field). We emphasize, however, that the main point of our study is not simply that layer 4 fast-spike interneurons are more connected to the thalamus than are the spiny regular spiking neurons. We believe that our most important finding is that virtually every first-order fast-spike interneuron in layer 4 receives input from nearly all of the LGN axons that synapse nearby (40/41 cases).

4) Issue of context:

The manuscript would be more appealing if it had a more general introduction, and a discussion that touches on the implications of their connectivity rules for sensory processing, cortical microcircuitry, and even the development of thalamocortical synapses onto SINs. Their data do provide some important clues about microcircuitry. There are, for example, SINs in their sample that do not receive monosynaptic thalamic input, yet the cells seem to reside in the termination zones of TC terminals. Something must determine whether TC axons connect or not to particular SINs, although the paper does not hint at anything mechanistic.

We have modified the Introduction and Discussion in line with this suggestion.

While the essential revisions are listed above, for the benefit of the authors we are pasting the detailed comments of the reviewers below in their entirety in case they find them useful in revising their resubmission:

a) Figure 3D shows a nice correlation between local spike-triggered-LFP amplitude and LGN-SIN efficacy. This is a beautiful result, but it raises a couple of questions of data analysis and semantics. If I understand the senior author's previous work on st-CSDs, the method identifies current sinks generated by both LGN intracortical axon spikes and sinks generated by LGN synapses upon cortical neurons (with the sinks perhaps generated mainly by vertically oriented pyramidal cells). If that's correct, then the only purely "presynaptic" component is the short-latency, brief, upwardly propagating axonal sink component. The longer latency, more localized component is the "postsynaptic" response. While the later sink obviously also depends on presynaptic components (synaptic functions require pre and post contributions, by definition), the sink is directly generated by postsynaptic mechanisms.

We agree that the observed later sink is generated purely by postsynaptic mechanisms. However, presynaptic terminals are necessary for this postsynaptic sink and our thinking was that this fast postsynaptic sink can be used to infer the density of presynaptic terminals in the area. Nevertheless, we agree that this is unnecessarily confusing, so we have changed this terminology throughout the manuscript, as described above (and suggested by the reviewers)

b) My questions: Does LGN-SIN efficacy correlate with any CSD measures? Figure 3D plots the peak postsynaptic LFP amplitude only, whereas the CSDs provide (in principle) additional spatial information (at least along the axis of the electrode array). CSDs also provide a true "presynaptic" measure, the early sinks, whereas the later sinks combine measures of pre- and postsynaptic functions. In the two example pairs shown in Figure 3, there is not only a big difference in the laminar locus of the LGN-triggered responses, but the axonal CSDs are very different in amplitude. This seems like a possible confound. Are the axonal sinks/spikes too small and unreliable to analyze across the cell sample? My first semantic quibble is about calling Rule 2 a "presynaptic factor", considering that the measure used to deduce it (the late L4 LFP) is largely a postsynaptic process.

As described in Figure 1, our cortical recordings were made either on single electrodes or on a laminar probe. Since the axonal and postsynaptic LFPs can be measured on single electrodes (Swadlow and Gusev, 2000) and on the laminar probes, they provided a more complete data set than did the CSD measures. Notably, the axonal responses (at the same depth as the SIN under study) is poorly correlated with synaptic efficacy (R2 = 0.0011). This is because the axonal and the postsynaptic responses often reach a peak at different depths. Sometimes the axonal sinks can be seen deep in the cortex at earliest latency (where there is no clear postsynaptic sink, as in our Figure 3B1). Indeed, sometimes there is a strong axonal response seen ascending through the cortex, but no postsynaptic response (e.g. Jin et al., 2011, Figure 1C case on right). Presumably, this case represents recordings from an ascending axon near our probe that turns laterally in the cortex and then arborizes and forms synapses (away from the recording probe).

We agree with the “first semantic quibble” and have changed the description of our “rules” throughout the manuscript.

c) To continue my pedantic semantic questions, why is Rule 3 a "postsynaptic factor", any more than Rule 2 is a "presynaptic factor"? Both designations seem to confuse the mechanistic issues. Rules 2 and 3 both involve different features of the same basic mechanism: either LGN axons form synapses on the SIN in question, or they don't. While the number and strength of those synapses are important factors, the methods used here cannot disambiguate those variables or infer anything exclusively pre- or postsynaptic about the connections. Rule 2 is about local connectivity (a SIN may make strong synapses with some LGN cells, but not with the cells being recorded in a particular experiment); rule 3 is about global connectivity (some SINs apparently don't get monosynaptic input from any LGN cells). This is an enormously important result, but I worry that the authors' semantic choices may obscure their conclusions for some readers. Labeling the rules "presynaptic" or "postsynaptic" is misleading, in my opinion. The problem is particularly acute in the Abstract and Introduction, where the rules are stated but only identified by their presynaptic/postsynaptic monikers and not further explained. The Discussion, by contrast, provides a very clear and compelling explanation of the authors' conclusions.

We agree (as described above) and have changed the text and figures appropriately.

d) Although the results are very interesting they may be actually trivial, as the authors honestly mention in their discussion. Indeed, rules 2 and 3 are probably simply spatial and temporal functional evidences of direct anatomical connection. Rule 2 is spatial and is a way to know whether the input arrives close to the recorded cell. Rule 3 is temporal and can be interpreted as expected delays between first-order and second-order cells (see Bullier et al., 1980 for instance). So if there is not the proper spatial conditions – the main input not in the right location – and/or not the proper temporal conditions – too long latency to be a first-order cell -, then it is entirely expected and logical that there is the connection is indirect. The results are really expected. Would rules 2 and 3 be linked to functional properties such as the RF structure or selectivity, as the authors also unsuccessfully investigated, the results would have had stronger impact because it would help understanding the relationship between the structure of the thalamo-cortical connectivity and the emergence of function. Currently, the benefit of these 3 rules is, as discussed, a methodological improvement for more unequivocal estimation of a direct connection: "This is a fairly easy measure to take in cross- correlation studies using extracellular microelectrodes, and we suggest that its general employment would reduce the variability seen in many cross-correlation studies of synaptic connectivity. " I agree it is an important methodological improvement for the field, but I feel not suited for eLife.

We respectfully disagree. Of course, for connectivity to occur, presynaptic terminals must be nearby (rule 2), and the postsynaptic neuron must be receptive to some LGN input (rule 3). Our ability to measure these variables while recording in vivo is new, and we agree it is a nice methodological contribution. However, what is new and surprising (as a conceptual contribution) is that these “necessary conditions” are the ONLY requirements (other than retinotopic alignment) that must be fulfilled to generate a functional synaptic connection. That these three requirements are both necessary and sufficient is new, and that is surprising.

All previous measures of thalamocortical connectivity indicated that only a subset of cortical neurons located within the axonal field of a thalamic afferent receives connection from that afferent. This is true for regular spiking neurons in the cat visual cortex (15%-33%, Sedigh-Sarvestani et al., 2017; Alonso et al., 2001), regular spiking neurons in rat barrel cortex (37%, Bruno and Simons, 2002) and fast spiking neurons in rat barrel cortex (63%, Bruno and Simons, 2002). This specific wiring is also true for other stages of sensory processing (e.g. 4/20, 20% for retinogeniculate connections in cats, Hamos et al., 1987). Given the strong evidence from the past that geniculate afferents make connection with only a subset of cortical neurons, our results are clearly surprising. They indicate that EVERY geniculate axon makes connection with ALL first-order fast spiking cortical neurons that are within reach of the axonal arbor (first-order defined as inhibitory neurons that receive thalamic input). This finding is clearly unexpected and has major consequences for understanding how thalamocortical inhibitory networks develop and are functionally organized.

e) However, there may be one possibility, with the current dataset, to potentially increase the impact of this result. I suggest that the authors make a complete parallel analysis of the non-SIN neurons analysed in Supplementary figure 2. Adding the information on the retinotopic alignment (rule 1) and latency of the response (rule 3) will allow to state whether these neurons obey as well to the 3 rules or not. If they do, as is expected, then the 3 rules are really 3 independent functional measure of direct connection that can be applied to any neurons. If they don't, then it seems that the result is less trivial than expected, that will make it probably more interesting. The discussion of the potential circuitry and impact of such results on the thalamo-cortical connectivity would make it more suitable for eLife.

As noted above, we would like to comply more fully with this request, but we don’t have enough data on non-SINs to make meaningful conclusions other than the single conclusion that we reached: that they were less frequently connected than were the SINs, despite being in the synaptic field of the same LGN neuron that was driving a neighboring SIN.

f) I don't really agree with the Abstract sentence "These rules concern (1) retinotopic alignment, (2) a property of the presynaptic axon, and (3) a property of the postsynaptic cortical interneuron". Rule 2 is not really a property of the presynaptic axon, which could be interpreted as conduction velocity for instance, but simply the laminar localization of the presynaptic terminals. The authors have not shown that rule 3 is a property of the cortical interneuron (integration time or Magno vs. Parvo cells for instance), since it can just be first vs. second-order cells. There are three major evidences in favor of the second hypothesis : 1) it is predictive of direct connection, 2) the authors haven't been able to relate the difference of connectivity to "sustained" or "transient" cells categories and 3) actually short and long latency SIN are undistinguishable functional properties. I think rule 2 would be better described as spatial connectivity profile and rule 3 has temporal properties that reflects 1st vs. 2nd order cells.

We agree (see above) and have changed the wording of the “rules”.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The authors largely addressed all concerns raised in the first round of review. Overall, the impression remains positive both about the importance of the results and suitability for eLife. However, important concerns were raised that need to be addressed concerning language and presentation.

The main points are:

1) Better definition of the term "postsynaptic". See detailed explanation in reviewer 2 comment.

We have changed our definition in the Abstract and throughout the manuscript, in line with the reviewers’ suggestion.

2) Clarification of SINs in comparison to layer 4 interneurons in other species. See comments by reviewer 1.

We have added a section on this in the Discussion entitled “Comparing the receptive fields of rabbit L4 SINs with those of other species”.

3) Enhancement of the main point, which is written in the response letter, that "virtually every first-order fast-spike interneuron in layer 4 receives input from nearly all of the LGN axons that synapse nearby (40/41 cases)". See comments by reviewer 3.

We have highlighted this idea in the Abstract (“We conclude that virtually every first-order fast-spike interneuron in layer 4 receives input from nearly all of the LGN axons that synapse nearby.”) and elsewhere in the text. In addition, we added a new section in Discussion where this is discussed further relative to the sparseness of connections to regular spiking cells: “Comparisons with L4 regular spiking (presumptive spiny) neurons”.

In addition to the written reviews, the three reviewers had a lengthy discussion regarding the revised manuscript. The reviewers praised the difficult and elegant technique used. Three key points of the discussion were:

1) The non-trivial nature of the high % of LGN inputs to SINs. It was noted that the main point that needs to come across to the reader is the sparseness of connections to RS cells, which are neighbors to the SINs and yet receive much less input (11% in this manuscript, but other examples in the literature are key to make the point). For those that have spent long hours (and years) chasing monosynaptic connections between thalamus and cortex, the numbers provided here are indeed surprising. However, without a clear notion of the difficulty and probability of finding connected pairs the three rules may appear rather trivial. The three reviewers agreed that in presenting the results, the authors should make an effort to convey this notion. When recording pairs, in which the cortical cell is an RS, a connection can be explained by the respective spatial and temporal properties of the two cells in the pair. But a lack of connection can almost never be explained. Why are thalamocortical axons so picky? The surprising finding here is that the authors are able to explain all cases in which there are NO connections based on the three rules. Which obviously means: 1) they can explain connectivity to all SINs and 2) SINs get a lot more LGN input than RS cells. These two points are novel and surprising and have convinced the reviewers of the suitability of the manuscript to eLife. However, it should be made clear also that a lot remains to be done to understand why RS cells are sparsely connected.

To better emphasize this, we have added a section to the Discussion “Comparisons with L4 regular spiking (presumptive spiny) neurons”.

2) The perhaps unique nature of SINs. As indicated by reviewer 1, fast-spiking (FS) interneurons in cats and monkeys are simple cells with spatially separate RF subregions (the jury is still out on mice it seems). SINs in rabbits are, instead, complex cells. Perhaps they fulfill a unique functional role in rabbits, and for that receive many more connections. Or perhaps SINS and the FS cells typically identified in cat and monkey neurophysiological studies are distinctly different subtype of neurons. Can the authors provide more insight into the comparison of rabbit SINS and cat/monkey FS cells?

We respectfully disagree with the reviewer on this issue and have added a section to the Discussion concerning such comparisons: “Comparing the receptive fields of rabbit L4 SINs with those of other species”.

3) Emphasis and clarification must be placed in the existence of a group of SINs that do not receive any connection from LGN, which suggests the existence of two types of SINs with perhaps very different functional roles. The two main potential sources of excitatory input to SINS are LGN synapses and synapses from local spiny cells. Data suggest two groups of SINS: Most are driven by strong, convergent LGN input, and this makes them sensitive and visually nonspecific. Another small group receives no LGN input, yet they are still visually sensitive and nonspecific, suggesting they are largely driven by strong, convergent input from local spiny cells which is entirely consistent with a variety of anatomical and paired-cell in vitro recording studies that the authors do not cite, local FS interneurons in general have high probabilities of connectivity (~0.5) with spiny neurons in their local neighborhood.

This point was also viewed by the reviewers as a key and novel result that needs emphasis and clarification.

We have added a new section to the Discussion: “The synthesis of receptive fields in L4 SINs”) which contains a paragraph dealing these “long-latency” SINs.

Reviewer #1:

An important notion that needs to be made very clear in the manuscript is that there are no interneurons in layer 4 of cat or monkey V1 that look like SINs. The vast majority of neurons in thalamorecipient layer 4 are simple cells, i.e. they have segregated ON and OFF subregions (apart from center surround neurons in monkey which are not relevant to this discussion). In contrast, SINs are complex cells with overlapping subregions of opposing contrast. That makes SIN a unique category of cells and perhaps that is the reason why they have such large percentage of input compared to cat L4 neurons. Based on that, comparisons of LGN % connectivity from other species are only mildly informative and the best comparison is with neurons in rabbit cortex itself. (which the authors don't fully provide).

Comparisons with barrel cortex are tricky because the functional properties that may determine convergence are not well understood, although sparse connectivity of VB to barrel neurons does enhance the uniqueness of SINs notoriously promiscuous connectivity.

So the authors show that, if there is RF overlap (rule1) and thalamic input to the vicinity of the SIN (st-LFP, rule2), the SIN will receive a connection, unless the SIN does not receive thalamic input at all (e-stim,rule3). Thus, SINs receive a lot more input than RS cells, which, under the same circumstances, only receive a small percentage of the available thalamic input. Sure, this is interesting and important, but by itself it does not create a universal new knowledge about L4 gabaergic interneurons because of the uniqueness of SINs. Comparing their promiscuous input to cat or monkey L4 interneurons is incorrect as in those two species FS neurons in layer 4 are simple cells with segregated subregions and are selective to orientation, direction and spatial frequency (SINs are very broadly tuned as shown by the authors in Zhuang et al). The broader tuning properties of layer 4 FS cells in cat and monkey is no match for the SINs.

We disagree strongly (but very respectfully!) with that statement, and explain our thinking in the new Discussion section entitled “ Comparing the receptive field of rabbit L4 SINs with those of other species ”.

Reviewer #2:

The authors have done an excellent job of responding to the reviewers' comments. Their revised manuscript provides clearer explanations of the "three rules", and new additions to the Introduction and Discussion broaden the context and appeal of the results.

I have one small suggestion. While the authors were very responsive to the reviewers' concerns about the meanings of "presynaptic" and "postsynaptic" in the original manuscript, the revision still has a phrase that will (in my opinion) confuse many readers. It first appears in the Abstract as a summary of rule #2: "strength of the postsynaptic response elicited near the interneuron by the LGN neuron", and again in a slightly different form in the Introduction.

Confusion will arise because is unclear In the Abstract and Introduction what "postsynaptic response" is referred to. Most readers' default assumption will be that it refers to the response of the postsynaptic neuron under study (i.e. a SIN), and the readers' own response might be to think, "well, of course you need a strong postsynaptic response from a SIN to consider it connected to the LGN cell". That's obvious and uninteresting.

But actually, the authors are referring to the "spike-triggered postsynaptic LFP response generated by the LGN neuron" (spelled out clearly much later in the paper). In other words, the authors' "postsynaptic measure, i.e. an st-LFP (or CSD), is derived not from the SIN under study but from other neighboring unidentified neurons; another reasonable interpretation of this measure is that the st-LFP/CSD is an estimate of the strength and location of the presynaptic terminals of the single LGN cell under study.

Bottom line: I think the authors need to find a better way to state their rule #2 early in the paper if they want their readers to appreciate the point.

Thank you, we agree and have restated this “rule” in the Abstract, and everywhere it comes up in the paper.

Reviewer #3:

In their revised version the authors addressed correctly and convincingly most of the concerns of the first review and the papers gains globally clarity. There remains a substantive concern regarding the analysis of non-SIN neurons

The authors have explained lacking experimental data to fully address the question and this is granted. However, can the authors be less conservative ?

It is understood that the authors do not have enough data to check for the 1st and 3rd rule. However, the result for rule 2 is quite different: only 11% of them are connected compared to 73% for the SIN, comparing Figure 3D with 5A is quite convincing that nonSIN and SIN behave differently.

This would be important to help the author convince the reader that the behavior of non-SIN are different and hereby be more convincing that their result is not trivial.

We have tried to be “less conservative” in this revision and have included a new section in the Discussion that concerns the “The sparseness of TC connections to L4 regular spiking (presumptive spiny) neurons”. In that section, as the reviewer suggests, we direct the reader to “compare Figures 5A and 3D”. We also now note the RS results in the Abstract.

What they answered to this concern, and their discussion is not fully convincing in the current state. In their answer, the authors stated that : “these (3 rules) "necessary conditions" are the ONLY requirements (other than retinotopic alignment) that must be fulfilled to generate a functional synaptic connection. That these three requirements are both necessary and sufficient is new, and that is surprising.”

It is new indeed, but at first glance do not seem surprising (see below).

The authors further argue that:

“All previous measures of thalamocortical connectivity indicated that only a subset of cortical neurons located within the axonal field of a thalamic afferent receives connection from that afferent. This is true for regular spiking neurons in the cat visual cortex (15%-33%, Sedigh-Sarvestani et al., 2017; Alonso et al., 2001), regular spiking neurons in rat barrel cortex (37%, Bruno and Simons, 2002) and fast spiking neurons in rat barrel cortex (63%, Bruno and Simons, 2002). This specific wiring is also true for other stages of sensory processing (e.g. 4/20, 20% for retinogeniculate connections in cats, Hamos et al., 1987). Given the strong evidence from the past that geniculate afferents make connection with only a subset of cortical neurons, our results are clearly surprising”

Regarding our “surprise” to find that the three rules were necessary and sufficient to predict TC connectivity, we now state that: “The “three rules” of synaptic connectivity that we describe are both necessary and sufficient for predicting connectivity between LGN neurons and L4 SINs (40/41 cases, Figure 4C). Together, they predict connection probability with a surprising high level of precision that is not yet possible for any other thalamocortical circuit in any other species.”

This argument concerns rule 2 only and demonstrate that, in itself, it is not sufficient. So further constrains must exist, as a consequence it does not make their result surprising, but rather expected at first glance. And alignment of RF and first-order cell are not the most surprising parameters to look at for further constrains.

Their last sentence “This finding is clearly unexpected and has major consequences for understanding how thalamocortical inhibitory networks develop and are functionally organized.” seems like an overstatement.

We have removed that statement.

https://doi.org/10.7554/eLife.60102.sa2

Article and author information

Author details

  1. Yulia Bereshpolova

    Department of Psychological Sciences, University of Connecticut, Storrs, United States
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Supervision, Validation, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7117-7255
  2. Xiaojuan Hei

    Department of Psychological Sciences, University of Connecticut, Storrs, United States
    Contribution
    Software, Formal analysis, Validation, Investigation
    Competing interests
    No competing interests declared
  3. Jose-Manuel Alonso

    1. Department of Psychological Sciences, University of Connecticut, Storrs, United States
    2. Department of Biological and Vision Sciences, State University of New York College of Optometry, New York, United States
    Contribution
    Conceptualization, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  4. Harvey A Swadlow

    1. Department of Psychological Sciences, University of Connecticut, Storrs, United States
    2. Department of Biological and Vision Sciences, State University of New York College of Optometry, New York, United States
    Contribution
    Conceptualization, Resources, Data curation, Supervision, Funding acquisition, Validation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    harvey.swadlow@uconn.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1477-3250

Funding

National Institutes of Health (R01EY028905)

  • Harvey Swadlow

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Ethics

Animal experimentation: All animal procedures were conducted with the approval of the University of Connecticut Animal Care and Use Committee i(Protocol No. A19-040) n accordance with the National Institutes of Health guidelines.

Senior Editor

  1. John R Huguenard, Stanford University School of Medicine, United States

Reviewing Editor

  1. Diego Contreras, University of Pennsylvania School of Medicine, Department of Neuroscience, United States

Reviewer

  1. Barry Connors, Brown University, United States

Publication history

  1. Received: June 16, 2020
  2. Accepted: November 30, 2020
  3. Version of Record published: December 8, 2020 (version 1)

Copyright

© 2020, Bereshpolova et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 796
    Page views
  • 109
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

  1. Further reading

Further reading

    1. Neuroscience
    Darjan Salaj et al.
    Research Article

    For solving tasks such as recognizing a song, answering a question, or inverting a sequence of symbols, cortical microcircuits need to integrate and manipulate information that was dispersed over time during the preceding seconds. Creating biologically realistic models for the underlying computations, especially with spiking neurons and for behaviorally relevant integration time spans, is notoriously difficult. We examine the role of spike frequency adaptation in such computations and find that it has a surprisingly large impact. The inclusion of this well-known property of a substantial fraction of neurons in the neocortex – especially in higher areas of the human neocortex – moves the performance of spiking neural network models for computations on network inputs that are temporally dispersed from a fairly low level up to the performance level of the human brain.

    1. Neuroscience
    Attila Ozsvár et al.
    Research Article

    Summation of ionotropic receptor-mediated responses is critical in neuronal computation by shaping input-output characteristics of neurons. However, arithmetics of summation for metabotropic signals are not known. We characterized the combined ionotropic and metabotropic output of neocortical neurogliaform cells (NGFCs) using electrophysiological and anatomical methods in the rat cerebral cortex. These experiments revealed that GABA receptors are activated outside release sites and confirmed coactivation of putative NGFCs in superficial cortical layers in vivo. Triple recordings from presynaptic NGFCs converging to a postsynaptic neuron revealed sublinear summation of ionotropic GABAA responses and linear summation of metabotropic GABAB responses. Based on a model combining properties of volume transmission and distributions of all NGFC axon terminals, we predict that in 83% of cases one or two NGFCs can provide input to a point in the neuropil. We suggest that interactions of metabotropic GABAergic responses remain linear even if most superficial layer interneurons specialized to recruit GABAB receptors are simultaneously active.