Physiological characterization of a rare subpopulation of doublet-spiking neurons in the ferret lateral geniculate nucleus

Abstract

Interest in exploring homologies in the early visual pathways of rodents, carnivores, and primates has recently grown. Retinas of these species contain morphologically and physiologically heterogeneous retinal ganglion cells that form the basis for parallel visual information processing streams. Whether rare retinal ganglion cells with unusual visual response properties in carnivores and primates project to the visual thalamus and drive unusual visual responses among thalamic relay neurons is poorly understood. We surveyed neurophysiological responses among hundreds of lateral geniculate nucleus (LGN) neurons in ferrets and observed a novel subpopulation of LGN neurons displaying doublet-spiking waveforms. Some visual response properties of doublet-spiking LGN neurons, like contrast and temporal frequency tuning, were intermediate to those of X and Y LGN neurons. Interestingly, most doublet-spiking LGN neurons were tuned for orientation and displayed direction selectivity for horizontal motion. Spatiotemporal receptive fields of doublet-spiking neurons were diverse and included center/surround organization, On/Off responses, and elongated separate On and Off subregions. Optogenetic activation of corticogeniculate feedback did not alter the tuning or spatiotemporal receptive fields of doublet-spiking neurons, suggesting that their unusual tuning properties were inherited from retinal inputs. The doublet-spiking LGN neurons were found throughout the depth of LGN recording penetrations. Together these findings suggest that while extremely rare (<2% of recorded LGN neurons), unique subpopulations of LGN neurons in carnivores receive retinal inputs that confer them with nonstandard visual response properties like direction selectivity. These results suggest that neuronal circuits for nonstandard visual computations are common across a variety of species, even though their proportions vary.

NEW & NOTEWORTHY Interest in visual system homologies across species has recently increased. Across species, retinas contain diverse retinal ganglion cells including cells with unusual visual response properties. It is unclear whether rare retinal ganglion cells in carnivores project to and drive similarly unique visual responses in the visual thalamus. We discovered a rare subpopulation of thalamic neurons defined by unique spike shape and visual response properties, suggesting that nonstandard visual computations are common to many species.

INTRODUCTION

Recently, there is renewed interest in homologies in the structure, organization, and function of the early visual pathways across mammalian species. As the mouse has become a popular model of visual system structure and function, many have sought evidence for homologies with more traditional animal models of vision such as carnivores and primates. While there are obvious differences in visual capabilities, central visual specializations, and cortical area organization across these species, it is nonetheless interesting to consider homologies, especially in the early processing stages of the visual hierarchy. For example, in rodents, carnivores, and primates, there are numerous morphologically and physiologically distinct retinal ganglion cells (RGCs), the output projection neurons of the retina (Boycott and Wässle 1999; Masland 2001; Thoreson and Dacey 2019). These distinct RGC classes form the basis for parallel visual information processing streams, a hallmark of early visual system organization. The vast majority of RGCs in carnivores and primates display standard receptive field properties, namely center/surround spatial organization (Field and Chichilnisky 2007; Troy and Shou 2002). In mice, other RGC types, like direction-selective RGCs (Weng et al. 2005), are as abundant or even more prevalent than center/surround RGCs (Baden et al. 2016). Many of these non-center/surround RGCs in mice, including direction-selective RGCs, project to the superior colliculus in lieu of or in addition to projecting to the visual thalamus, the dorsal lateral geniculate nucleus (LGN; Seabrook et al. 2017). Although rare, direction-selective RGCs have also been observed in carnivore and primate retinas (Cleland and Levick 1974; Schiller and Malpeli 1977). Furthermore, some koniocellular LGN neurons in marmosets display direction-selective responses (Eiber et al. 2018). Whether and how nonstandard visual response properties among carnivore and primate LGN neurons are attributed to inputs from rare nonstandard RGCs, like direction-selective RGCs, remain open questions. In this study, we describe a novel subpopulation of LGN neurons in the ferret that displayed nonstandard physiological properties including tuning for stimulus orientation and direction. Optogenetic manipulation of corticogeniculate feedback did not systematically alter the unusual tuning properties of these LGN neurons, suggesting that their tuning for orientation and direction was inherited from retinal inputs.

The physiological response properties of LGN neurons have been well described in carnivores (Bullier and Norton 1979; Cleland et al. 1971; Derrington and Fuchs 1979; Hoffmann et al. 1972; Hubel 1960; Saul and Humphrey 1990), including ferrets (Price and Morgan 1987; Stryker and Zahs 1983; Zahs and Stryker 1985). The majority of LGN neurons receive inputs from alpha or beta RGCs and relay X or Y parallel stream signals to the visual cortex (Sherman and Guillery 2006). X and Y LGN neurons have center/surround receptive fields but are differentiated by sustained versus transient responses to stationary gratings and preferences for higher or lower luminance contrasts and lower or higher temporal frequencies, respectively (Derrington and Fuchs 1979; Price and Morgan 1987). Some nonstandard responses such as color selectivity have been observed among cat LGN neurons (Buzás et al. 2013); however, it is challenging to characterize nonstandard LGN responses given that such a small minority (<10%) of RGCs display visual responses that differ from the standard center/surround responses of the majority alpha and beta RGCs (Cleland and Levick 1974). Furthermore, whether and in what capacity these rare RGCs project to the LGN in carnivores and primates are not known.

We conducted a large survey of neurophysiological responses among LGN neurons in ferrets, and we discovered a novel subpopulation of LGN neurons that were defined by a unique and consistent spike waveform feature: “doublet”-spike waveforms. Doublet-shaped spike waveforms consisted of two distinct downward deflections within less than 1 ms of each other but not separated into multiple distinct action potentials. Tuning measurements for doublet-spiking LGN neurons revealed contrast and temporal frequency preferences intermediate to those of X and Y neurons and tuning for stimulus orientation and direction in many cases. These unusual tuning properties were consistent with nonstandard spatiotemporal receptive fields including On/Off and elongated subregions. Optogenetic activation of corticogeniculate feedback did not alter doublet-spiking neuronal tuning, suggesting these properties were inherited from retinal inputs. Together these findings support the existence of rare and unusual subpopulations of LGN neurons in carnivores that likely receive direct retinal input from rare and nonstandard RGCs. Although doublet-spiking LGN neurons made up a tiny proportion of recorded LGN neurons in ferrets, their presence suggests that neuronal circuits conveying nonstandard visual computations from the retina to the LGN are common to multiple species.

MATERIALS AND METHODS

This study involved new analyses of data collected as a part of a previous study of the functional role of corticogeniculate (CG) circuits in vision (Hasse and Briggs 2017) as well as new data collected following the same procedures. Data were surveyed from ~600 LGN neurons recorded in 30 adult female ferrets (Mustela putorius furo). Data reported here were collected from 14 ferrets. All of the procedures performed conformed to the guidelines set forth by the National Institutes of Health and the U.S. Department of Agriculture and were approved by the Institutional Animal Care and Use Committees at the University of Rochester and the Geisel School of Medicine at Dartmouth.

Summary of methods for which details have been published previously.

All of the ferrets received an injection of a genetically modified rabies virus (SAD∆G-ChR2-mCherry) targeting the lateral geniculate nucleus of the thalamus (LGN) to express channelrhodopsin2 (ChR2) and mCherry selectively in CG neurons in the visual cortex, as described previously (Hasse et al. 2019; Hasse and Briggs 2017). Surgical preparation and virus injection, neurophysiological recording of LGN and V1 neurons, visual and optogenetic stimulation, spike sorting and data analyses, and histological processing of brain tissue have all been described in detail previously (Hasse and Briggs 2017). Briefly, in a sterile surgical procedure and under full anesthesia, 5 μl of rabies virus were injected, targeting the LGN (Hasse et al. 2019; Hasse and Briggs 2017). Ferrets recovered for 7 to 11 days after which a terminal neurophysiological recording experiment was conducted under anesthesia and with paralytic to prevent eye movements. Multielectrode arrays (7-channel Eckhorn Matrix from Thomas Recording Gmbh, Giessen, Germany and 24-contact V-probe from Plexon Inc., Dallas, TX) placed in the LGN and in area 17 recorded LGN and V1 neurons in response to visual stimuli displayed on a CRT monitor placed ~50 cm in front of ferrets’ eyes and optogenetic stimulation (LED emitting ~464-nm light) delivered via fiberoptic probe (Doric Lenses Inc., Quebec, CAN or PlexBright by Plexon) placed at the surface of area 17 near the V-probe (Hasse and Briggs 2017). Visual stimuli included drifting sinusoidal gratings and m-sequence white noise stimuli. Gratings were presented for 2 s with 2 s of mean gray luminance in between presentations; each grating type was displayed two to three times per condition (no LED, with LED). Half of the trials included visual stimulation alone (no LED) and the other half were visual stimulation paired with optogenetic stimulation (with LED), at the drift rate of displayed gratings or continuously on for m-sequence stimuli (Hasse and Briggs 2017). Following the end of the neurophysiological recording session, ferrets were euthanized via overdose and perfused, and then brain tissue was histologically processed to verify virus expression at the injection site and in CG neurons (Hasse et al. 2019; Hasse and Briggs 2017).

Spike waveform and tuning analyses.

LGN single units were spike sorted offline following standard clustering procedures. Waveform data and interspike interval (ISI) distributions were computed for all LGN single units. The percentage of total spikes in the ISI bin corresponding to 1 ms, akin to short ISI violations, was computed per neuron. Additionally, bin times corresponding to the ISI distribution maximum were determined per single unit. LGN neuronal tuning in response to gratings varying in contrast, temporal frequency, spatial frequency, orientation, and size was measured via curve fits to stimulus-evoked firing rates. Contrast curves were fit with power functions, orientation curves were fit with Gaussians, and spatial and temporal frequency curves were fit with smoothing spline functions. The following tuning metrics were analyzed for this study: contrast to evoke a half-maximal response (c50), preferred temporal frequency (TF), preferred orientation, orientation half-width at half-height (HWHH) for the preferred orientation, 1 minus the circular variance (1 − CV) for orientation tuning, and direction selectivity (DS) computed as the difference divided by the sum of the response to the preferred and anti-preferred (+180°) orientations. LGN neurons were considered tuned for orientation if one or two neighboring data points were above 2 SD of the mean evoked responses across orientations. LGN neuronal responses to m-sequence stimuli were analyzed in two ways for all LGN neurons. Spatiotemporal receptive field maps were generated by computing the spike-triggered average (STA) of reverse-correlated sequence frames. Additionally, principal components analysis (PCA) was used to compute second, third, and fourth principal component (PC) responses from the peak temporal window (usually the frames corresponding to 40–20 ms preceding the spike). The first PC was redundant as it was equivalent to the STA in all cases. Average spontaneous and visually evoked firing rates of LGN neurons were computed from responses measured before and during grating presentation, respectively.

Summary of animal and cell numbers.

In 8 of 14 ferrets, virus was successfully injected into the LGN revealing expression of ChR2 and mCherry in CG neurons (example LGN injection site and labeled CG neurons shown in Fig. 2, A and B). Doublet-spiking LGN neurons recorded in these animals are labeled in blue throughout the figures. In the remaining six ferrets, there was no virus in the LGN and no label in the visual cortex so LGN neurons recorded in these animals served as control for optogenetic stimulation and are labeled in black throughout the figures. Seven doublet-spiking LGN neurons were recorded in five ferrets in which virus injected into the LGN induced ChR2 and mCherry expression in CG neurons. Four doublet-spiking LGN neurons were recorded in four ferrets in which no virus was detected in the LGN. Thirty-seven LGN neurons (LNs) with regular shaped spike waveforms were recorded in 5 ferrets in which virus injected into the LGN-induced ChR2 and mCherry expression in CG neurons, and 13 LNs were recorded in 2 ferrets in which no virus was detected in the LGN. Doublet-spiking neurons and LNs were qualitatively and quantitatively similar within each subpopulation across recording sessions/animals (Figs. 1–5) and were therefore combined into their respective categories for all population-level analyses.

Fig. 2.

Examples of virus expression in lateral geniculate nucleus (LGN) and V1 and relative depths of doublet-spiking LGN neurons. A: site of rabies virus injection (black stain) in LGN, shown in coronal section stained against cytochrome oxidase activity. Scale bar = 375 μm; dorsal is up, and medial is left. B: virus-infected and stained corticogeniculate (CG) neurons in layer 6 of area 17 shown in coronal section and stained against cytochrome oxidase activity. Layers are labeled at left. Scale bar = 250 μm. C: depth of 10 doublet-spiking LGN neuron relative to the most dorsal and ventral LGN recording sites per session. Black and blue dots indicate no virus and virus expression, respectively, as in Fig. 1.

Fig. 1.

Waveforms of doublet-spiking lateral geniculate nucleus (LGN) neurons. A, rows 1–5: Individual spike waveforms (50–100 randomly selected waveforms each, colored) overlaid with average ± 1 SD of all spike waveforms (solid black lines, dashed black lines) for 10 of 11 doublet-spiking LGN neurons. Average ± SD waveforms were computed from all waveforms per neuron, regardless of shape. Neurons are identified by cell numbers at the top of rows 1–5; black text indicates neurons recorded from animals in which no virus was detected in the LGN, and blue text indicates neurons recorded from animals in which virus injected into the LGN induced channelrhodopsin2 (ChR2) and mCherry expression in corticogeniculate (CG) neurons. Interspike interval (ISI) distributions illustrated for each neuron at right. ISI x-axis is from 0 to 100 ms for all. Row 6: 2 example LGN neurons (LNs) with regular shaped spike waveforms and their ISI distributions. B: percentage of total spikes in the 1st bin, corresponding to 1 ms (or short ISI violations) for doublet-spiking neurons and LNs. Black and blue dots indicate no virus and virus expression, respectively, as described in A. Red dots and lines indicate average values and SE, respectively. C: percentage of waveforms with 2 negative peaks for each doublet-spiking neuron, LN, and combination of 2 LNs into 1 unit (2xLNs). *Significant differences between doublet-spiking neurons, LNs, and 2xLNs (P = 1.2 × 10⁻⁶, Kruskal-Wallis test). Conventions as in B. D: SD in peak-to-peak time for doublet-spiking neurons, LNs, and 2xLNs. *Significant difference between doublet-spiking neurons, LNs, and 2xLNs (P = 0.002; Kruskal-Wallis test). Conventions as in B. E: average peak-to-peak time for doublet-spiking neurons, LNs, and 2xLNs. Conventions as in B.

Fig. 5.

Spatial receptive field components for doublet-spiking lateral geniculate nucleus (LGN) neurons. A: spatiotemporal receptive field generated from the spike-triggered average (STA) of m-sequence frames for cell 3, recorded from an animal with no virus expression in LGN or V1. Top and bottom rows: STAs are without and with LED stimulation, respectively. Temporal windows indicate time preceding spikes (at 0 ms). Scale bar = 1°. B–D: principal component analysis (PCA) of m-sequence frames at the peak temporal window (40–20 ms) for cell 1 (from a no-virus animal; B), cell 9 (C), and cell 6 (D), the latter 2 from animals in which virus was expressed in corticogeniculate (CG) neurons (blue cell titles). Top and bottom rows: principal components (PCs 2, 3, and/or 4) are without and with LED stimulation, respectively. Scale bars = 1°.

Classification of doublet-spiking LGN neurons including statistical comparisons.

Preliminary qualitative identification of doublet-spiking waveforms among LGN single-unit recordings involved examining the shape of individual and average waveforms for distinct features of doublet spikes, namely two distinct peaks within the same downward deflection that were close together in time but were not separated into multiple action potentials. Spike sorting of putative doublet-spiking LGN neurons was reexamined to ensure that doublet spikes were not the result of two separate spike waveforms. Of the 11 doublet-spiking neurons identified, 8 were the only well-isolated unit recorded on the electrode. The remaining three doublet-spiking neurons were recorded on an electrode along with another well-isolated single-unit. To ensure that the doublet-spiking units were separate from the other isolated units, a cross-correlation between the spike trains of the simultaneously recorded units was computed. There were no significant cross-correlation peaks in the spike trains of the three groupings of units recorded on the same electrodes, indicating that doublet-spiking neurons were unique units and not contaminated by spikes from nearby units. The percentage of spikes in the first bin (e.g., short ISI violations) and the bin times corresponding to the maximum of the ISI distributions were compared across doublet-spiking LGN neurons and LNs using Mann-Whitney U tests. Additionally, the presence of the same doublet-shaped waveform throughout the entire duration of the ~2-h long recording session was confirmed to consider a putative unit to be a doublet-spiking neuron.

Three tests, described below, were employed to determine whether LGN neurons designated as doublet-spiking had waveforms that were distinct from other LNs. In these three tests, putative doublet-spiking LGN neurons were compared with two control groups of LNs. Fifty LNs were randomly selected and used as the first control group. To test the possibility that doublet waveforms were due to coincident spiking among two separate single units recorded on the same channel but mistakenly sorted into a single unit, we manually resorted spikes on an additional 11 randomly selected channels containing two LNs by combining them into a single unit; these were categorized as “2xLN” units.

The three tests to determine whether doublet-spiking LGN neuronal waveforms differed from LN or 2xLN waveforms were as follows. In the first test, the percentage of total waveforms with doublet shape for each unit was calculated. Doublet-shaped waveforms were defined by two minima within the same downward deflection and within a 0.6-ms window. The percentage of doublet-shaped waveforms was computed relative to the total number of spike waveforms recorded from the same unit over the entire ~2-h long recording session. In the second test, the average peak-to-peak time (in ms) was computed between the first and second minima for all waveforms for each unit. In the third test, the standard deviation (SD) of the peak-to-peak time (in ms) between the first and second minima for all waveforms per unit was calculated. Population-level comparisons of the percentage of doublet-shaped waveforms, the average peak-to-peak time, and the SD of the peak-to-peak time were made between doublet-spiking LGN neurons, LNs, and 2xLNs using nonparametric ANOVA tests (Kruskal-Wallis) that include corrections for multiple comparisons.

Analysis of relative depth and statistical comparisons of tuning metrics across neuronal types.

To determine the relative depth of doublet-spiking LGN neurons relative to other LGN neurons, we computed the recording position of doublet-spiking LGN neurons relative to the most dorsal (given a value of 1) and the most ventral LGN (given a value of 0) neuronal recordings made with the same electrode penetration. Importantly, the most dorsal and ventral recording positions do not necessarily indicate the top and bottom of the LGN but the upper and lower depths of each recording penetration.

Tuning metrics (c50 and preferred TF) were compared across doublet-spiking LGN neurons and LNs using Mann-Whitney U tests. Preferred orientation and 1 − CV were compared across doublet-spiking LGN neurons and LNs that displayed orientation tuning using Mann-Whitney U tests. A minority of LNs (5 of 50, 4 of which were recorded in the same ferret) displayed some orientation tuning, consistent with prior observations that the majority of ferret LGN neurons have center/surround classical receptive fields and do not display orientation tuning or direction selectivity (Price and Morgan 1987; Stryker and Zahs 1983) similar to cats (Derrington and Fuchs 1979; Sherman and Spear 1982).

RESULTS

In carnivores, including ferrets, the vast majority of LGN neurons fall into the physiologically defined X or Y classes, with a smaller proportion of LGN neurons assigned to the “catch-all” W class (Bullier and Norton 1979; Cleland et al. 1971; Derrington and Fuchs 1979; Hoffmann et al. 1972; Price and Morgan 1987; Saul and Humphrey 1990; Sherman and Spear 1982; Stryker and Zahs 1983). A minority (<10%) of RGCs display nonstandard visual physiological responses, i.e., that differ from the standard center/surround responses of the majority alpha and beta RGCs that contribute to the X and Y streams, respectively (Cleland and Levick 1974). While less common visual responses are attributed to some cat LGN neurons (Buzás et al. 2013), it remains an open question whether rare RGCs provide sufficient input to generate nonstandard visual response properties in the carnivore LGN. This question is especially timely given renewed interest in homologies in visual circuits across a variety of animal models.

In surveying over 500 neurons recorded in the LGN of ferrets, we observed 11 LGN neurons that all displayed a unique spiking waveform shape consisting of two minima at the trough of the action potential within 0.6 ms of each other (Fig. 1A). We termed these “doublet”-spiking waveforms because of their clear double downward deflections without separation of multiple distinct action potentials. Doublet-spiking waveforms were qualitatively consistent across all 11 doublet-spiking LGN neurons but distinct from the waveforms of LGN neurons (LNs) with regular spike waveforms (Fig. 1A, colored lines are individual waveforms, and row 6 shows representative LN waveforms). There were no differences in doublet waveform shape across doublet-spiking LGN neurons recorded in ferrets in which virus injected into the LGN (Fig. 2A) produced ChR2 and mCherry expression in corticogeniculate neurons (Fig. 2B) compared with those recorded in ferrets in which virus was not injected into the LGN (Fig. 1A, blue and black cell labels, respectively). Spike sorting errors were not the source of doublet-spiking waveforms, as ISI distributions (Fig. 1A, right distributions per cell) revealed few short ISI violations, or the percentage of total spikes in the first time bin, corresponding to 1 ms (Fig. 1B). Percentages of spikes in the first bin ranged from 0.02 to 0.66% for doublet-spiking LGN neurons and were not different from the those of LNs (P = 0.72; average percent spikes in 1st bin for doublet-spiking LGN neurons = 0.24 ± 0.08%, for LNs = 0.47 ± 0.1%; Fig. 1B). Additionally, ISI distribution peaks did not differ between doublet-spiking LGN neurons and LNs (P = 0.13; average ISI peak for doublet-spiking LGN neurons = 3.6 ± 0.2 ms, for LNs = 3.8 ± 0.4 ms). Furthermore, doublet-spiking waveforms were not an artifact of multiple units sorted together on the same recording electrode. The majority of doublet-spiking LGN neurons were the only well-isolated unit recorded per electrode. Three doublet-spiking LGN neurons recorded along with other well-isolated units had no significant overlapping spike times according to cross-correlation analyses. These results suggest that doublet-spiking LGN neurons were unique units, not contaminated by nearby unit spiking.

Having confirmed that doublet-spiking waveforms were not an artifact of multiunit activity or spike sorting, we next determined whether doublet-spiking LGN neuronal waveforms were significantly different from waveforms of LNs. Waveforms of doublet-spiking LGN neurons were compared with waveforms of 50 randomly selected LNs. As a secondary confirmation that doublet-shaped waveforms could not be generated from multi-unit activity, we also compared doublet-spiking LGN neuronal waveforms to 11 additional pairs of LNs recorded on the same electrode and sorted together into a new “unit” (termed 2xLN units). Doublet-spiking LGN neurons were significantly more likely to have waveforms with two negative peaks compared with LNs and 2xLNs (P = 1.2 × 10⁻⁶; average percent 2-peak waveforms for doublet-spiking LGN neurons = 61 ± 5.5%, for LNs = 22 ± 1.5%, for 2xLNs = 18 ± 4%; Fig. 1C). Additionally, the time delay between the two negative peaks was significantly more consistent for doublet-spiking LGN neurons compared with LNs and 2xLNs (P = 0.002; average SD of peak-to-peak time for doublet-spiking LGN neurons = 0.06 ± 0.006 ms, for LNs = 0.1 ± 0.004ms, for 2xLNs = 0.1 ± 0.01 ms; Fig. 1D). Although not significant, the average peak-to-peak width (in ms) was less for doublet-spiking neuron waveforms compared with those of LNs and 2xLNs (P = 0.6; average peak-to-peak time for doublet-spiking LGN neurons = 0.16 ± 0.01 ms, for LNs = 0.19 ± 0.01 ms, for 2xLNs = 0.18 ± 0.02; Fig. 1E) suggesting that the time delay between negative peaks in doublet-shaped waveforms was smaller than delays due to rare variations in LN action potentials. Additionally, doublet-spiking LGN neuron waveforms were not appreciably different in terms of short ISI violations, percentages of two-peak waveforms, peak-to-peak widths, or the standard deviation of peak-to-peak widths for neurons recorded in ferrets with or without virus in the LGN (blue and black dots in Fig. 1, B–E). Together these results suggested that doublet-shaped waveforms were a consistent and unique feature of the extracellular action potentials of doublet-spiking LGN neurons and were not due to spike sorting artifacts, interactions between neighboring units, or effects of corticogeniculate feedback on action potential shape or timing. Based on the unique and consistent doublet-shaped waveform feature we observed, we next explored the visual response properties of doublet-spiking LGN neurons to determine whether they represented a distinct, nonstandard subpopulation of LGN neurons.

Although ferret LGN is a laminar structure with some separation of X/Y and W cell classes in the A/A1 and C layers, respectively (Zahs and Stryker 1985), there was no obvious pattern in the position of doublet-spiking LGN neurons relative to the most dorsally or ventrally recorded LGN neurons per penetration (Fig. 2C). Doublet-spiking LGN neurons in ferrets with or without virus in the LGN were similarly distributed per recording penetration (Fig. 2C, blue and black dots). From these patterns, it seemed unlikely that doublet-spiking LGN neurons were restricted to a single LGN layer. Doublet-spiking LGN neurons displayed a range of spontaneous and visually-evoked firing rates (2–37 Hz) that were within the distribution of those observed for X and Y LGN neurons (average spontaneous firing rate = 9.8 ± 2 Hz; average visually evoked firing rate = 13.4 ± 2.5 Hz).

Doublet-spiking LGN neurons also displayed tuning for stimulus contrast, temporal frequency, and spatial frequency that was within the range of X and Y LGN neurons (Fig. 3A). Interestingly, doublet-spiking LGN neurons tended to have lower contrast to evoke half maximal response (c50) values than LNs as a whole (average c50 for doublet-spiking neurons = 29 ± 5.4%, for LNs = 41.1 ± 3.3%; Fig. 3B, compare filled to open dots), more similar to Y LGN neurons. However, doublet-spiking LGN neurons also tended to prefer lower temporal frequencies than LNs as a whole (average preferred TF for doublet-spiking neurons = 7.4 ± 1.2 Hz, for LNs = 9.2 ± 1 Hz; Fig. 3C, compare filled to open dots), more similar to X LGN neurons. Based on contrast and temporal frequency tuning, doublet-spiking LGN neurons had intermediate preferences relative to X and Y LGN neurons. As we observed previously for LGN X and Y neurons, there was no effect of LED activation of corticogeniculate feedback on the c50 or preferred temporal frequencies of doublet-spiking LGN neurons (Fig. 3, B and C).

Fig. 3.

Tuning and LED modulation of tuning metrics for doublet-spiking lateral geniculate nucleus (LGN) neurons. A: tuning curves for 2 representative doublet-spiking LGN neurons (cells 4 and 6) for contrast (top row), spatial frequency (middle row), and temporal frequency (bottom row). Data points are dots, error bars represent SE, and lines are curve fits. Black data points and curves (left columns per cell) are data from no LED trials and cyan data points and curves (right columns per cell) are data from trials with LED stimulation of corticogeniculate (CG) neurons. Cell 4 (black label, green outline) was recorded in an animal with no virus in the LGN and cell 6 (blue label, orange outline) was in an animal in which virus injected into the LGN yielded channelrhodopsin2 (ChR2) and mCherry expression in CG neurons. B and C: contrast to evoke a half-maximal response (c50; B) and preferred temporal frequency (TF; C) with and without LED stimulation of CG neurons for doublet-spiking LGN neurons. Black and blue filled dots indicating no virus or virus in LGN, as in Fig. 1, and LGN neurons (LNs) are open black circles. Green, orange, and yellow outlines indicate cells 4, 6, and 10 respectively, matching boxes around cell labels in A and in Fig. 4A.

Somewhat surprisingly, 8 of 10 doublet-spiking LGN neurons displayed orientation tuning (Fig. 4, A–D) and 6 of 10 had some direction selectivity (Fig. 4E). Notably, most doublet-spiking LGN neurons preferred vertical orientations and horizontal directions of grating motion. All but one of the tuned doublet-spiking LGN neurons preferred grating orientations within 30° degrees of vertical, as illustrated by the two example sets of tuning curves (Fig. 4A) and the histogram of preferred orientations among doublet-spiking LGN neurons (Fig. 4B). The majority of LNs in our sample did not display orientation tuning, consistent with tuning data from X and Y LGN neurons that do not display selectivity for stimulus orientation or direction due to their center/surround receptive fields (Price and Morgan 1987; Stryker and Zahs 1983). The preferred orientations for the minority of LNs showing some orientation bias (5 of 50, 4 of which were recorded from a single anatomical location in one ferret) were similarly distributed compared with those of doublet-spiking LGN neurons (Fig. 4B, gray outlines). Doublet-spiking LGN neurons also preferred a narrower range of orientations relative to V1 neurons in ferret V1, which are ~7–30% more likely to be tuned to cardinal than oblique orientations but otherwise demonstrate distributed tuning across orientations (Coppola et al. 1998; Usrey et al. 2003). Orientation tuning quantified by 1 − CV and half-width at half-height revealed relatively sharp orientation tuning among tuned doublet-spiking LGN neurons and a mixture of tuned and weakly biased or untuned responses among LNs (Fig. 4, C and D). Most of the doublet-spiking LGN neurons tuned for orientation were also direction selective (Fig. 4E). Interestingly, the lone doublet-spiking neuron that was tuned to a nonvertical orientation did not show direction selectivity. Accordingly, all of the doublet-spiking LGN neurons that were direction selective preferred horizontal motion direction, consistent with direction-selective LGN neurons encountered in mice (Marshel et al. 2012; Scholl et al. 2013). As with tuning for contrast and temporal frequency, LED activation of corticogeniculate neurons did not systematically adjust orientation or direction tuning preferences among doublet-spiking LGN neurons (Fig. 4, D and E).

Fig. 4.

Orientation tuning and direction selectivity for doublet-spiking lateral geniculate nucleus (LGN) neurons. A: example orientation tuning curves for cells 4 and 10. Data points are dots, error bars represent SE, and lines are curve fits. Black data points and curves are data from no LED trials. Dashed gray lines illustrate spontaneous activity levels. Insets are polar plots of preferred grating orientation and direction with cardinal orientations labeled. The outermost ring corresponds to a firing rate of 15 Hz for both examples (inset: spike rates are baseline-subtracted). Cell 4 (black label, green outline) was recorded in an animal with no virus in the LGN, and cell 10 (blue label, yellow outline) was in an animal in which virus injected into the LGN yielded channelrhodopsin2 (ChR2) and mCherry expression in corticogeniculate (CG) neurons. B: distribution of preferred orientations for doublet-spiking neurons and LGN neurons (LNs) with some orientation tuning bias (gray outlines). Black and blue bars indicate no virus or virus in LGN. Colored arrows indicate cells in Figs. 3–5 with matching outlines. C: distribution of 1 minus circular variance (1 − CV) values for doublet-spiking neurons and LNs. Conventions as in B. Note different left and right y-axes. Doublet-spiking and LN neurons with 1 − CV = 0 were not tuned for orientation. D: orientation tuning as half-width at half-height (HWHH) with and without LED stimulation of CG neurons for doublet-spiking LGN neurons. Black and blue filled dots indicating no virus or virus in LGN, as in Fig. 3. Dots with colored outlines match cells in Figs. 3–5. Neurons with HWHH = 180° are untuned for orientation. E: direction selectivity (DS) with and without LED stimulation of CG neurons for doublet-spiking LGN neurons, conventions as in D. Neurons with DS = 0 are untuned for direction.

When available, spatiotemporal receptive field maps corroborated orientation and direction tuning preferences among doublet-spiking LGN neurons. Cell 3 was an Off cell with a standard, spatially linear, center/surround receptive field organization (Fig. 5A). Consistent with this receptive field structure, cell 3 was untuned for grating orientation and was not direction selective. Cell 3 was unusual in that its response to an Off pixel in its receptive field was quite variable, including both long and short latency responses (Fig. 5A). Unsurprisingly given cell 3 was recorded in an animal with no virus in the LGN, there was no effect of LED activation of visual cortex on its spatiotemporal receptive field. The three other doublet-spiking LGN neurons for which we obtained spatiotemporal receptive field maps were nonstandard in that they had a combination of On and Off responses in their receptive field and/or their receptive fields included vertically elongated and separate On and Off subregions (Fig. 5, B–D). For these three neurons, spike-triggered averages of white noise stimulus frames did not yield any spatiotemporal structure due to nonlinear spatial summation, but PCA analysis of stimulus frames revealed both On and Off subregions. In particular, cells 1 and 6 had vertically elongated On and Off subregions that were spatially offset, indicating orientation preference for vertical gratings (Fig. 5, B and D). Consistent with this receptive field structure, both cells 1 and 6 were tuned for vertical orientations and displayed some horizontal motion direction selectivity (red and orange arrows and outlined dots in Fig. 4, B–E). Neither of these cells showed marked changes in receptive field structure during LED illumination of visual cortex (Fig. 5, B and D). Cell 9 also had a center/surround receptive field organization, but it responded to both On and Off pixels in its receptive field (Fig. 5C). Interestingly, the On response of cell 9 was absent with LED activation of corticogeniculate feedback. Cell 9 also displayed some orientation tuning, perhaps due to its mixture of On and Off responsivity, but it was not direction selective.

Finally, we measured doublet-spiking LGN neuronal response statistics with and without LED illumination of the visual cortex, as we have done for LGN X and Y neurons previously (Hasse and Briggs 2017). There were no changes in response latency, response precision, mean spike count, or spike count variability across LED conditions for four doublet-spiking LGN neurons recorded in ferrets in which virus infected corticogeniculate neurons (P > 0.25 for all) nor for one doublet-spiking LGN neuron recorded in a ferret with no virus in the LGN.

DISCUSSION

A survey of neurophysiological responses among over 500 LGN neurons in ferrets revealed a novel and rare (<2% of LGN neurons) subpopulation of LGN neurons that were defined by a unique and consistent spike waveform feature. These neurons all displayed waveforms with two distinct spike minima within less than 1 ms, and the majority of spikes fired over recording sessions lasting hours had the same “doublet” waveform shape (Fig. 1). We verified that doublet-spiking waveforms were not an artifact of spike sorting errors and differed in shape compared with waveforms of LGN neurons with regular spike waveforms. Interestingly, doublet-spiking LGN neurons tended to prefer visual stimuli at lower contrast and with lower temporal frequencies, suggesting visual response preferences intermediate to those of X and Y LGN neurons (Hasse and Briggs 2017; Price and Morgan 1987; Stryker and Zahs 1983). Surprisingly, many doublet-spiking LGN neurons displayed orientation tuning and direction selectivity reminiscent of direction-selective LGN neurons recorded in mice (Marshel et al. 2012; Scholl et al. 2013). Receptive fields of doublet-spiking neurons could be center/surround, On/Off, or vertically elongated with spatially separated On and Off subfields. Importantly, optogenetic activation of corticogeniculate feedback did not produce changes in the visual response properties or spatiotemporal receptive fields of doublet-spiking LGN neurons, suggesting their unusual responses were inherited from their retinal inputs. Additionally, doublet-spiking LGN neurons were found throughout the depth of LGN recording penetrations. These findings suggest that rare and unique LGN subpopulations exist in the carnivore and are likely driven by input from rare and nonstandard RGCs. Finally, although doublet-spiking LGN neurons had a number of common characteristics including doublet-spike waveforms and preferences for low contrast and high temporal frequencies, they varied in spatial receptive field structure, which generated orientation and direction tuning for some doublet-spiking LGN neurons. Thus, even among this rare subpopulation of LGN neurons, there was additional diversity.

It is possible that doublet-spiking LGN neurons are a subset of the catch-all category of W LGN neurons. Most doublet-spiking LGN neurons were recorded more ventral to X or Y neuronal recordings along electrode tract penetrations, with two recorded more dorsal to X or Y neurons. It is possible to cross from C to A1/A layers or vice versa depending on the rostral/caudal and medial/lateral position of the recording electrode. Without lesions or dye-coated electrode tracts, it is not possible to be certain about the laminar location of doublet-spiking LGN neurons. The only direction-selective LGN neurons reported in primates have been located in the koniocellular layers (Eiber et al. 2018), which, if homologous to the direction-selective doublet-spiking LGN neurons encountered here, could suggest membership in the W class. Another possibility is that doublet-spiking LGN neurons are local LGN inhibitory interneurons. However, this seems unlikely for a number of reasons. First, LGN inhibitory interneurons make up more than 2% of LGN neurons (Montero and Singer 1985) and even though their cell somas are small, many have been recorded previously in cats (Wang et al. 2011). Doublet-spiking waveforms have not been reported for these inhibitory interneurons (Wang et al. 2011). Furthermore, inhibitory interneurons display receptive field properties that are more similar to X or Y neurons (Montero and Singer 1985), consistent with the fact that they receive direct retinal input from alpha and beta RGCs (Dubin and Cleland 1977). Doublet-spiking LGN neurons could also arise due to a combination of inputs from multiple RGCs with spatially offset or diverse tuning properties, similar to the convergence of diverse RGC inputs onto LGN neurons observed in mice (Liang et al. 2018; Litvina and Chen 2017). Compared with mice, carnivore RGC inputs onto LGN neurons may be less divergent and more likely to share common spatial and tuning properties (Usrey et al. 1999). However, convergent input from diverse RGCs would be consistent with the notion that doublet-spiking LGN neurons differ from X and Y LGN neurons and may show more similarities to LGN neurons observed in mice.

Another possibility is that doublet-spiking waveforms were simply burst spikes recorded from LGN neurons. This possibility also seems unlikely for several reasons. Burst spikes occur less often than tonic spikes; ~10–15% of total LGN neuronal spikes are burst spikes in anesthetized cats (Denning and Reinagel 2005; Lesica and Stanley 2004; Rivadulla et al. 2003). A hallmark of doublet-spiking LGN neurons was that more than half of the spikes, greater than 90% of spikes in two cases, had the same characteristic two-peaked doublet-shaped waveform. By comparison, LNs displayed spikes with two peaks less than 40% of the time, in line with the percentage of burst spikes observed previously in LGN neurons. Furthermore, the standard deviation in the width between the doublet peaks per waveform was quite low (Fig. 1D), suggesting that doublet-shaped spikes were generated by the biophysical properties of ion channels expressed within this subpopulation of LGN neurons, not coincident arrival of excitatory postsynaptic potentials (EPSPs). Additionally, tuning properties and spatiotemporal receptive field maps were computed from all spikes, regardless of their shape, per trial for doublet-spiking LGN neurons. As a rule, doublet-spiking LGN neurons displayed unusual visual responses like tuning for low contrast and low temporal frequencies (Fig. 3A), tuning for orientation and direction (Fig. 4), and On/Off receptive fields (Fig. 5, B–D). These properties are not consistent with LGN X or Y neurons, suggesting that doublet-spiking LGN neurons were not simply LGN neurons firing burst spikes.

It is important to point out that it is unlikely the doublet-spiking waveforms we measured here were an artifact of dendritic or axonal currents. LGN spikes with complex waveforms have been observed when the recording electrode is very close to the cell body, enabling recordings of retinogeniculate EPSPs or S potentials, and/or axon-generated spikes in addition to somatic spikes (Bishop et al. 1962; Kaplan and Shapley 1984; Sincich et al. 2007). We did not reliably detect these types of currents in our recordings. However, the doublet-spiking waveforms we observed differed from the complex waveforms described previously. LGN spiking waveforms mixed with EPSPs have large amplitudes (greater than 0.5 mV; Sincich et al. 2007), and the timing of currents relative to spikes is variable within individual LGN neurons. The doublet-shaped waveforms we measured had smaller amplitudes and consistent peak-to-peak widths (Fig. 1), inconsistent with EPSP origins. The doublet-spiking waveforms we observed bore some resemblance to doublet-shaped spiking waveforms described previously during the transition between extracellular and intracellular glass pipette recordings of motor neurons (Nelson and Frank 1964). However, as expected for intracellularly recorded action potentials, the amplitudes of the doublet-shaped waveforms described by Nelson and Frank (1964) were an order of magnitude greater than the spike amplitudes we observed, suggesting that our recordings were consistently extracellular. Furthermore, we were able to record from doublet-spiking LGN neurons for hours, while intracellular recordings in vivo typically do not last long.

Artifacts that manifest as double spikes also can appear when recordings made further away (~100 μm) from the cell body pick up back-propagating signals along dendrites. Prior studies have noted “W-shaped,” low-amplitude waveforms when recordings are made near distal dendrites, possibly due to mixture of dendritic signals and action potentials (Gold et al. 2006; Loópez-Jury et al. 2018). In some cases, these complex waveforms include two peaks separated by milliseconds. Additionally, these types of artifacts were recorded with multielectrode arrays of silicone electrode contacts that have low impedance and thus pick up a variety of signals over longer distances. Here we utilized single-wire electrodes made of tungsten-in-glass that had high impedance (average impedance = 1.6 ± 0.05 MΩ; range = 1–2 MΩ) and were optimal for isolating single unit spikes. Furthermore, the doublet-spiking waveforms described here all had two distinct peaks within less than 0.6 ms of each other, representing an order of magnitude shorter time between peaks than many back-propagation artifacts reported previously. Also, our simulation of spike sorting errors using single LNs recorded on the same contact and sorted together to form 2xLN units did not reveal consistent percentages of doublet-shaped waveforms, nor did doublet-like waveforms from 2xLN units have consistent peak-to-peak widths or standard deviations (Fig. 1, C–E). While we cannot entirely rule out the possibility that extracellular recording conditions contributed to doublet-shaped waveforms, we believe our observations are most consistent with unique expression of ion channels and/or unique cellular structure in doublet-spiking LGN neurons. A unique combination of ion channels that are less subject to inactivation could generate doublet-shaped spike waveforms, akin to the expression of T-type calcium channels that generate bursts in thalamic neurons (Lu et al. 1992). Alternatively, doublet-spiking LGN neurons could contain multiple electrical compartments within the soma or the axon initial segment, such as multiple spike initiation zones or proximal axonal branches, that generate doublet-shaped waveforms. Further study of the ion channel expression and biophysical properties of these unique neurons is required to elucidate the mechanism underlying doublet-shaped spike waveforms.

Although, as laid out above, it seems unlikely that doublet-spiking waveforms were a result of burst firing or recordings from LGN neuronal dendrites or retinal afferents, it is nonetheless possible that doublet-spiking LGN neurons do not constitute a novel subpopulation of LGN neurons. Regardless, it is noteworthy that novel visual response properties were independently observed among this unusual group of doublet-spiking LGN neurons. It is interesting to consider the possible source for these unusual visual response properties like orientation and direction selectivity. Given that the vast majority of RGCs have center/surround receptive fields, one possible source for orientation tuning and direction selectivity is from the visual cortex via corticogeniculate feedback. We used optogenetic methods to activate corticogeniculate neurons while recording from doublet-spiking LGN neurons, and activation of cortical feedback did not systematically alter the receptive field properties of doublet-spiking neurons. Two doublet-spiking LGN neurons recorded in ferrets in which virus injected into the LGN generated expression of ChR2 in corticogeniculate neurons showed some changes in direction selectivity with optogenetic activation of feedback. However, these effects were in the opposite direction and similar magnitude changes in direction selectivity were observed for two doublet-spiking LGN neurons recorded in ferrets in which no virus was in the LGN (Fig. 4E). These data suggest that while all four of these example doublet-spiking LGN neurons displayed robust direction selectivity in at least one experimental condition, direction selectivity was somewhat variable across measurements. Only one doublet-spiking neuron showed a change (a sharpening) in orientation tuning with optogenetic activation of corticogeniculate feedback (Fig. 4D). Furthermore, optogenetic activation of corticogeniculate feedback did not alter any other tuning properties, response latency or precision, or spatiotemporal receptive field properties of doublet-spiking neurons, in marked contrast to results of optogenetic activation of corticogeniculate feedback on the response timing and precision of LGN X and Y neurons (Hasse and Briggs 2017). Together these results suggest that corticogeniculate feedback from V1 minimally impacts the receptive field properties or temporal dynamics of doublet-spiking LGN neurons. If doublet-spiking LGN neurons receive feedback from the visual cortex, it must be contributing to other, as yet unmeasured aspect of their responses. Accordingly, the most likely source for the unusual response properties of doublet-spiking LGN neurons is the retina. Further evidence supporting a retinal, rather than cortical, origin for tuning among doublet-spiking LGN neurons is found in their spatial receptive field structure. Direction-selective doublet-spiking LGN neurons with spatially separated On/Off subfields likely inherited their tuning from direction-selective RGCs, as in the mouse. In mouse, direction selectivity in RGCs arises through asymmetric excitation and inhibition from distributed amacrine cell inputs (Hanson et al. 2019) and direction selectivity is then conferred onto direction-selective On/Off LGN neurons via RGC inputs (Suresh et al. 2016). In V1 simple cells of carnivores, direction selectivity emerges through space/time inseparable receptive field structure (McLean and Palmer 1989), in contrast to the separable receptive field structure observed for direction-selective doublet-spiking LGN neurons.

Although unusual visual response properties have been observed among a small subset of LGN neurons in carnivores and primates (Buzás et al. 2013; Eiber et al. 2018), it was unclear whether inputs from rare RGCs were the primary source for these signals. Our results provide compelling evidence that nonstandard visual computations are relayed directly from the retina to the LGN by unique and rare neuronal circuits in a variety of species including carnivores.

GRANTS

This work was funded by National Eye Institute Grants EY-018683 and EY-025219 (to F.B.) and T32-EY-007125 (to A.J.M.) and the Whitehall Foundation (2013-05-06). J.M.H. was supported by a Graduate Fellowship from the Albert J. Ryan Foundation.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.J.M. and J.M.H. performed experiments; A.J.M. and J.M.H. analyzed data; A.J.M., J.M.H., and F.B. interpreted results of experiments; A.J.M. and F.B. prepared figures; A.J.M. and F.B. drafted manuscript; A.J.M., J.M.H., and F.B. edited and revised manuscript; A.J.M., J.M.H., and F.B. approved final version of manuscript; J.M.H. and F.B. conceived and designed research.