Abstract
Selectivity for direction of motion is a key feature of primary visual cortical neurons. Visual experience is required for direction selectivity in carnivore and primate visual cortex, but the circuit mechanisms of its formation remain incompletely understood. Here, we examined how developing lateral geniculate nucleus (LGN) neurons may contribute to cortical direction selectivity. Using in vivo electrophysiology techniques, we examined LGN receptive field properties of visually naive female ferrets before and after exposure to 6 h of motion stimuli to assess the effect of acute visual experience on LGN cell development. We found that acute experience with motion stimuli did not significantly affect the weak orientation or direction selectivity of LGN neurons. In addition, we found that neither latency nor sustainedness or transience of LGN neurons significantly changed with acute experience. These results suggest that the direction selectivity that emerges in cortex after acute experience is computed in cortex and cannot be explained by changes in LGN cells.
SIGNIFICANCE STATEMENT The development of typical neural circuitry requires experience-independent and experience-dependent factors. In the visual cortex of carnivores and primates, selectivity for motion arises as a result of experience, but we do not understand whether the major brain area that sits between the retina and the visual cortex—the lateral geniculate nucleus of the thalamus—also participates. Here, we found that lateral geniculate neurons do not exhibit changes as a result of several hours of visual experience with moving stimuli at a time when visual cortical neurons undergo a rapid change. We conclude that lateral geniculate neurons do not participate in this plasticity and that changes in cortex are likely responsible for the development of direction selectivity in carnivores and primates.
Introduction
Detection of direction of motion is a key component of mammalian visual processing that emerges early in the visual pathway. In primates and carnivores, direction selectivity is thought to first emerge in the primary visual cortex (V1; Hubel and Wiesel, 1962; Livingstone and Hubel, 1988) and requires visual experience for its development (Chapman and Stryker, 1993; White et al., 2001; Li et al., 2006). Early deprivation of visual experience in kittens (Zhou et al., 1995) and ferrets (Li et al., 2006) impairs development of direction selectivity, and this deficit cannot be recovered despite subsequent introduction to normal visual experience (Li et al., 2006). However, direction selectivity can be rapidly and artificially induced in visually naive ferrets with only 3–6 h of visual exposure to moving stimuli (Li et al., 2008; Van Hooser et al., 2012; Ritter et al., 2017) allowing us to probe its development in a laboratory setting to study its mechanisms.
What circuitry underlies this rapid developmental emergence? The major feedforward input to the visual cortex arrives via the lateral geniculate nucleus (LGN), and here we sought to understand what, if any, contribution to emerging direction selectivity was made by changes in LGN neurons. Although the developing visual cortex exhibits rapid changes in direction selectivity with a few hours of visual experience, it is unknown if developing LGN cells also exhibit substantial receptive field plasticity. One could imagine two major hypotheses for how potential developmental plasticity in LGN neurons might influence direction selectivity in V1.
First, LGN neurons could acquire some substantial direction selectivity themselves as a result of experience so that the input that each cell provides to cortex exhibits increased direction selectivity as a result of experience. Although this hypothesis would be at odds with the spot detector model for a majority of LGN receptive fields (Kuffler, 1953) recent studies have shown that some species have substantial direction-selective or orientation-selective channels that run through the LGN (Marshel et al., 2012; Scholl et al., 2013; Zhao et al., 2013; Cruz-Martín et al., 2014; Sun et al., 2016; Hillier et al., 2017) making this an important possibility to assess.
Second, LGN cells themselves may be unselective for orientation or direction, but there could be experience-dependent changes in their receptive fields that are necessary for the expression of direction selectivity in visual cortex. For example, direction selectivity requires some spatiotemporal offset in inputs (Reichardt, 1961; Barlow and Levick, 1965; Adelson and Bergen, 1985; Watson and Ahumada, 1985; Suarez et al., 1995; Maex and Orban, 1996) in that inputs to one side of the receptive field of the cell have different latencies than the inputs to the other side (Movshon et al., 1978b; McLean and Palmer, 1989; Albrecht and Geisler, 1991; Reid et al., 1991; Saul and Humphrey, 1992b; DeAngelis et al., 1993; Jagadeesh et al., 1997; Livingstone, 1998; Priebe and Ferster, 2005). Perhaps visual experience is required to establish some property, such as a spread in LGN cell latencies (Saul and Humphrey, 1990, 1992a; Ferster et al., 1996; Alonso et al., 2001; Stanley et al., 2012) or a change in the sustained/transient nature of responses (Marr and Ullman, 1981; Lien and Scanziani, 2018) that is necessary for direction selectivity to be computed in cortex, even if the computation is conducted in cortex.
Using in vivo electrophysiology, we examined the receptive field properties of individual thalamic neurons before and after acute visual stimulation with drifting sinusoidal gratings. We found that orientation selectivity, direction selectivity, response latency, and sustainedness/transience of individual LGN neurons did not significantly change with acute visual experience. That is, we found no evidence of rapid changes in LGN receptive fields when cortex is undergoing a substantial increase in direction selectivity. These results are consistent with the idea that the changes driving an increase in cortical direction selectivity are first occurring in V1, either at the convergence of LGN inputs or in the V1 circuitry itself.
Materials and Methods
Experimental design
All experimental procedures were approved by the Brandeis University Animal Care and Use Committee and performed in compliance with National Institutes of Health Guide for the Care and Use of Laboratory Animals. Ferrets (Mustela putorius furo; n = 29), age postnatal day (P) 30–35 were used for electrophysiological experiments. Females were used because young animals required cohousing with sexually mature females, and cohousing with males causes undue stress in mature females.
Surgical procedures
Ferrets were initially sedated with ketamine (20 mg/kg, i.m.). Atropine (0.16–0.8 mg/kg, i.m.) and dexamethasone (0.5 mg/kg, i.m.) were used to reduce salivary and bronchial secretion and to reduce inflammation and swelling, respectively. The animal was anesthetized with a mixture of isoflurane, oxygen, and nitrous oxide through a mask while a tracheostomy was performed. After completion of the tracheostomy, animals were ventilated with 1–2% isoflurane in a 2:1 mixture of nitrous oxide and oxygen. Next, a cannula was inserted into the intraperitoneal cavity for delivery of 5% dextrose in lactated Ringer's solution (3 ml/kg/h). Body temperature was maintained at 37°C using a thermostatically controlled heating pad. End-tidal CO2 levels and respiration rate were monitored and kept within the appropriate physiological range (3.5–4%). The animal was held in place by a custom stereotaxic frame that did not obstruct vision. Silicone oil was placed on the eyes to prevent corneal damage. An O-ring (10 mm diameter, 2 mm thickness) was cemented to the skull using Zap Gel and Zip Kicker (ZAP) over LGN in the right hemisphere, and a 6 × 6 mm craniotomy was made in the interior of the O-ring. The dura was removed with a 31 gauge needle. All wound margins were infused with bupivacaine. Before beginning electrophysiological recordings, the ferret was paralyzed with the neuromuscular blocker gallamine triethiodide (10–30 mg/kg/h) through the intraperitoneal cavity to suppress spontaneous eye movements, and the nitrous oxide-oxygen mixture was adjusted to 1:1. The electrocardiogram (ECG) of the animal was continuously monitored to ensure adequate anesthesia, and the percentage of isoflurane was increased if the ECG indicated any distress.
Electrophysiology
Following removal of the dura, warm agarose (2–4% in PBS) was applied to the craniotomy to prevent brain pulsation. Then a custom (https://github.com/VH-Lab/vhlab-parts) 3D-printed chamber-and-grid (Realize) was cemented to the O-ring using Zap Gel and Zip Kicker (ZAP) so that an electrode could be driven through the grid and inserted perpendicularly to the surface of the brain. A low-impedance tungsten electrode (0.1 MΩ; catalog #TM33B01, World Precision Instruments) was inserted through different grid holes using an MP-285 manipulator (Sutter Instruments) until the LGN was initially located by monitoring modulation to handheld stimuli on a loudspeaker. The LGN was identified by identifying an area that exhibited reliable responses to repeated stimulation with an ophthalmoscope or manually controlled computer stimulus. The LGN was then mapped by driving the electrode through adjacent holes, and a recording location was selected where receptive fields were in front of the animal (within 30°). Then either a linear microelectrode array, microwire brush array (Microprobes), or a tetrode array (Plexon S-probe) was inserted in the same location for recording. The signal was amplified using an RHD2000 amplifying/digitizing chip and USB interface board (Intan Technologies), and stimulus timing information was acquired using a Micro1401 acquisition board and Spike2 software (Cambridge Electronic Design). Units were initially automatically sorted off-line by a mixture of Gaussians model using the KlustaKwik algorithm (http://klustakwik.sourceforge.net; Harris et al., 2000) and manually adjusted using custom software in MATLAB (http://github.com/VH-Lab/vhlab-toolbox-matlab, function math/cluster_points_gui.m) to isolate single units. Recordings were made at the beginning of the experiment and after 6 h of training with visual stimuli.
To compare results in LGN with prior results in visual cortex, we re-examined data from Ritter et al. (2017), which used the same setup and stimuli but used different multichannel electrodes that resulted in multiunit data rather than the single units described here. Nevertheless, the columnar organization of orientation and direction selectivity in ferret visual cortex (Weliky et al., 1996; Rao et al., 1997) means that multiunit activity is a good readout of changes in orientation and direction selectivity as assessed with spiking activity.
Identifying the LGN
The primary method we used to identify the LGN was the presence of distinct and repeatable responses to manual visual stimulation on the loudspeaker, its depth as measured by the micromanipulator (∼10 mm below the cortical surface), and its structure when mapped with low impedance electrodes. Using these criteria in past studies of LGN in squirrel (Van Hooser et al., 2003) and tree shrew (Van Hooser et al., 2013), histologic reconstructions always showed electrodes to be in LGN and never outside LGN. Here, in two cases we performed histology to verify that the recordings occurred in the LGN. Because we were concerned about the reusability of our expensive multichannel electrodes, we did not apply a label or produce electrolytic lesions using these multichannel electrodes. Instead, on completion of experiments, a traditional single-channel tungsten electrode coated in the fluorescent dye DiI (DiCarlo et al., 1996) was inserted at the same grid hole location and depth and left in place for 20 min. Animals were then transcardially perfused, and the brain was placed in 4% paraformaldehyde in 0.1 m PBS at 4°C for 24 h and then moved to 10% sucrose in PBS for 24–48 h. This was followed by placement in 30% sucrose in PBS at 4°C until sectioning. The recording hemisphere of the brain was sectioned coronally or sagittally into 50–150 μm sections using a sliding microtome (Leica SM2010R). All staining procedures were performed on a shaker. We washed sections in 0.1 m PBS 3 × 5 min and permeabilized in 0.3% Triton X-100 diluted in PBS for 2 h at room temperature. Some slices were stained with 5% 500/525 NeuroTrace (fluorescent Nissl; Thermo Fisher Scientific) for 20 min, and others were incubated in fluorophore-conjugated anti-NeuN antibody (Alexa Fluor 488 Rabbit anti-NeuN, catalog #ABN78A4, Millipore) at a 1:300 dilution overnight (>12 h) at 4°C. Sections were then washed 3 × 5 min in PBS and mounted on slides until air dried. Slides were then coverslipped with Fluoromount-G medium (Electron Microscopy Sciences), and edges were sealed using nail polish (Electron Microscopy Sciences). Histologic sections were viewed using a fluorescent microscope (Keyence BX-Z 710), and electrode tracks were reconstructed using DiI dye traces. In all reconstructed cases, the electrode was verified to be in the LGN.
Visual stimulation
Visual stimuli were created in MATLAB (MathWorks) software using the Psychophysics Toolbox (Brainard, 1997; Pelli, 1997; Kleiner et al., 2007) and displayed on a 21 inch flat face CRT monitor (GDM-520, Sony) with a resolution of 800 × 600 and a refresh rate of 100 Hz. The monitor was placed at a such a distance that it subtended 51° × 42° of visual space. We manually mapped receptive fields by displaying circular patches of drifting sinusoidal gratings at different positions and moving the monitor to accommodate different eccentricities while listening to the responses on a loudspeaker.
Drifting grating stimuli were full-field, high-contrast sinusoidal gratings (2 s duration; 3.5 s interstimulus interval) presented pseudorandomly, with direction of motion (in steps of 45°) in either of the two directions orthogonal to the axis of orientation. For spatial and temporal frequency tuning, direction was covaried with stimuli consisting of drifting sinusoidal gratings at six different spatial frequencies [0.01–0.32 cycles per degree (cpd)] and six different temporal frequencies (0.5–32 Hz), respectively, at 100% contrast. When spatial frequency was varied, temporal frequency was held fixed at either 1 or 4 Hz, and when temporal frequency was varied, spatial frequency was held fixed at 0.08 cpd. Spatial phase stimuli consisted of counterphase static sinusoidal gratings (spatial frequency 0.08 cycles per degree and temporal frequency 1 Hz) at six different spatial phases.
Data analysis
We recorded from a total of 563 LGN neurons as identified by spike sorting. Spikes were initially detected by identifying signals that were negative and had an absolute amplitude of >4 SDs of the noise. For drifting gratings that examined orientation/direction, we examined mean responses (F0) and modulation at the stimulus frequency (F1). If a the F1 response of a cell was greater than the mean response (F0), F1 was used to calculate index values. If F0 was greater, it was used for calculations (Movshon et al., 1978a, b; Heimel et al., 2005). Neurons were included in analysis if they exhibited significant variation across all conditions including blank by an ANOVA test, p < 0.05; there was no exclusion criterion based on neuron firing rates, which could be quite low (spontaneous firing rate before training: 0.51 spikes/s, SEM, 0.069 spikes/s). We calculated responses for 207 neurons. The fitless vector measures, circular variance (CV), and directional circular variance (DirCV), were calculated as previously described (Ringach et al., 2002; Mazurek et al., 2014) and 1-CV and 1-DirCV were used to quantify selectivity for orientation and direction, respectively. Direction tuning curves were examined for stimuli where direction and spatial frequency were covaried, and the spatial frequency with the highest response (F0 or F1 response, whichever was greater) was chosen for computing 1-CV and 1-DirCV. For display purposes only, we also showed double Gaussian fits, prepared as in Mazurek et al. (2014).
For spatial phase stimuli, the spatial phase that elicited the greatest response (F1) was chosen for further analysis. Raster responses from 200 trials were averaged for peristimulus time histograms (PSTHs; 10 ms time bins) of cell responses, and PSTHs were used to calculate peak rate, maintained rate, response latency, and transience/sustainedness. Values were calculated as previously described (Van Hooser et al., 2003). In brief, responses were binned into 10 ms increments. The bin with the greatest number of spikes was used for peak latency. Peak firing rate (PR) was defined as the mean rate during the 10 ms bin centered on the peak latency. Initial latency was taken as the first bin with a spike count greater than or equal to one-half the maximum. There was one cell that responded with such a delayed latency that its peak response occurred in the opposite phase. We assigned the latency of this cell to be 500 ms + peak rate. Maintained firing rate (MR) was defined as the mean firing rate during a 100–500 ms window after the peak latency time. Transience was defined as the transient time constant, τtrans, where MR = PR * exp(−300 ms/τtrans).
We performed Monte Carlo simulations to estimate the number of highly direction-selective cells needed to identify a cardinal bias. For N cells starting with six cells, we drew angle preferences randomly and manually set 50% of these cells to be cardinal angles. We binned these angles into 60° slices. We performed a Rayleigh test (MATLAB, function circ_rtest) to determine whether the data differed from uniform with p < 0.05. We then performed 100 simulations for each increasing value of N, and chose the N that provided significant detection in 95% (1.00–0.05) of these simulations.
Results
At the time of eye opening in ferrets, corresponding to approximately P 30–33, neurons in the V1 are selective for orientation of stimuli but are not selective for direction of motion of stimuli (Li et al., 2006, 2008). Direction selectivity develops with a few days of visual experience and increases to mature levels within the first 2–3 weeks of visual experience (Li et al., 2006). Furthermore, experience is required for the increase in cortical direction selectivity. Animals that have been dark reared at the time of eye opening until P45 never develop cortical selectivity for stimulus direction even if they are typically reared beginning at P45 (Li et al., 2006). However, direction selectivity can be rapidly induced in a laboratory environment by providing young ferrets with 3–6 h of exposure to drifting gratings (Li et al., 2008; Van Hooser et al., 2012).
To understand how the receptive field properties of LGN cells may be changing during this time of great plasticity of receptive fields in visual cortex, we made in vivo electrophysiological recordings in the LGN of visually naive ferrets (P30–P35; n = 29) that had between 0–2 d of visual experience after eye opening. To measure the responses of cells in the naive state, we needed to use methods that would allow us to record the responses of many LGN neurons simultaneously. The traditional single-cell recording approach of carefully introducing stimuli to an individual neuron for an hour or more would have meant that the third or fourth neuron recorded would have undergone extensive visual experience. The ferret LGN is small and is located ∼1 cm below the brain surface, so we developed a grid-and-chamber system (Fig. 1a) that allowed us to map the LGN with low impedance electrodes before precisely introducing a multichannel recording electrode to the brain at the same insertion position and direction as our preferred mapping penetration (Fig. 1b).
Methods for single-channel mapping and multichannel recording from ferret LGN. a, Top-down and side view of grid-and-chamber design used for locating and mapping LGN. b, Schematic for initial mapping and placement of multichannel recording electrode for experiment duration. Grid-and-chamber was secured to the skull of the animal, a single-channel mapping electrode was lowered into different grid holes, and responses to a light stimulus were monitored. When visual responses characteristic of LGN were located, consecutive grid holes were mapped with the single-channel electrode until a location with a central receptive field with high signal-to-noise ratio was identified. Then the multichannel electrode was lowered into the same location for the experiment. c, Experimental setup with timeline of experimental paradigm. Experiment began with a test phase to record initial responses to moving sinusoidal gratings of varying orientations, directions, spatial and temporal frequencies, and spatial phases. Then the animal was exposed to 6 h of visual stimulation with bidirectional sinusoidal gratings moving at 0° or 90°. Following this, the same set of test stimuli was used to measure responses again. d, Stably isolated spike cluster and waveform recorded from a neuron across test phases. e, Histologic confirmation of electrode track in ferret LGN. Electrode was coated in fluorescent dye DiI for visualization. Green shows neurons stained with anti-NeuN. Red is DiI. Right, Earlier mapping penetrations. LGN is outlined in gray. Scale bar, 100 µm.
After making initial receptive field measurements, we provided the animal with 6 h of visual stimulus exposure to a training motion stimulus that consisted of drifting sinusoidal gratings (0.08 cpd; temporal frequency 1–4 Hz) moving back and forth along the axis of motion orthogonal to the grating orientation (Li et al., 2008; Van Hooser et al., 2012; Ritter et al., 2017; Roy et al., 2020a; Fig. 1c). In this way, we obtained measures of LGN receptive fields before and after visual experience. In this work, although we were able to isolate single units (Fig. 1d), we did not find that we could commonly hold the same cells before and after visual experience, so we analyzed these groups of neurons as separate populations and intended no claims about following receptive fields of individual neurons as the lab has previously done in two-photon imaging work (Li et al., 2008; Van Hooser et al., 2012; Roy et al., 2020a). An example of a histologic section of LGN showing the fluorescent DiI that was on our labeling electrode is shown in Figure 1e.
Orientation and direction selectivity
We wanted to test the hypothesis that acute visual experience might cause an increase in orientation or direction selectivity in individual LGN cells. If this were true, then the rapid increase in direction selectivity that is observed in visual cortical neurons might reflect a direct contribution of individual LGN inputs.
The strengths of orientation and direction tuning were quantified by calculation of an orientation selectivity index (1-CV; see above, Materials and Methods) and a direction selectivity index (1-DirCV; see above, Materials and Methods).
We first compared the orientation selectivity values of naive LGN cells with a population of naive and trained ferret visual cortical cells that were recorded with the same stimuli, experimental setup, and analysis code and reported in Ritter et al. (2017). The only difference between the studies in addition to the area of recording is that Ritter et al. (2017) used different electrodes and analyzed multiunit data rather than single-unit data. Initial orientation indexes for LGN cells were weak, exhibiting a median value of 0.12. Representative tuning curves for naive LGN cells are shown in Figure 2a (top row), and tuning curves for naive and experienced cortical cells from layers II to VI are shown in Figure 2a,b (bottom rows). Additional raw data, in the form of raster plots and peristimulus time histograms for the LGN cells in Figure 2, a and b, are shown in Figure 3. We found that orientation selectivity index values in naive LGN neurons (median, 0.12) were much lower than in naive cortical neurons [median, 0.435; Kruskal–Wallis test H(1) = 135.8034, p < 1e-308] and trained cortical neurons [median, 0.53; Kruskal–Wallis test H(1) = 135.4603, p < 1e-308; Fig. 2c]. For comparison, the median (50th percentile) orientation selectivity index value in the naive cortex corresponded to the 90th percentile orientation selectivity index value in the naive and experienced LGN.
Lack of increase in orientation and direction selectivity in LGN neurons with 6 h of experience with moving stimuli. a, Top row, Tuning curves of three example LGN neurons before visual stimulation. Bottom row, Three naive primary visual cortex neurons (cortex; Ritter et al., 2017). A double Gaussian fit (Mazurek et al., 2014) is shown and in bold if orientation tuning was significant (Hotelling-T2, p < 0.05), but later quantification is done using circular variance. Dashed line indicates response to control gray screen. Raster plots are shown for LGN cells in Figure 3. b, Top row, Tuning curves of three example LGN neurons. Bottom row, Three example cortical neurons in ferrets after visual stimulation. Conventions are the same as in a. c, Cumulative histogram of orientation selectivity in LGN before (LGN Naive, n = 167) and after (LGN Trained, n = 60) 6 h of visual stimulation compared with data previously collected from ferret V1 (Ritter et al., 2017) before (Cortex Naive, n = 174) and after 6 h of visual stimulation (Cortex Trained, n = 131). Degree of orientation selectivity is expressed as 1-CV in orientation space, n indicates cell number, *indicates reference group, and lines point to groups that are significantly different from the reference group. Orientation selectivity index values in naive LGN neurons (median, 0.12) were much lower than in naive [median, 0.435; Kruskal–Wallis test H(1) = 135.8034, p < 1e-308] and experienced [median, 0.53; Kruskal–Wallis test H(1) = 135.4603, p < 1e-308] cortical neurons. Visually naive LGN neurons did not significantly increase in their selectivity for orientation following 6 h of training [median, 0.11; Kruskal–Wallis test, H(1) = 0.1037, p = 0.7474]. d, Cumulative histogram of direction selectivity in LGN before (LGN Naive) and after (LGN Trained) 6 h of visual stimulation compared with data previously collected from cortical neurons before (Cortex Naive) and after (Cortex Trained) 6 h of visual stimulation (Ritter et al., 2017). Degree of direction selectivity is expressed as 1-DirCV in direction space. Conventions are the same as in c. Direction selectivity index values in naive LGN neurons (median, 0.14) were not significantly lower than in naive cortical neurons [median, 0.12; Kruskal–Wallis test H(1) = 1.8852, p = 0.1697] but were significantly lower than trained [median, 0.20; Kruskal–Wallis test H(1) = 11.6684, p = 6.3570e-04] cortical neurons. Visually naive LGN neurons did not significantly increase in their selectivity for direction following 6 h of training [median, 0.155; Kruskal–Wallis test, H(1) = 0.0012, p = 0.9726].
Cycle-by-cycle raster plots and peristimulus time histograms for LGN cells in Figure 2. Animals were shown several drifting sinusoidal gratings at different directions (indicated by arrow) at the preferred spatial frequency and the temporal frequency indicated by the time axis. Each cycle of the grating constituted a line in the raster plot. ai–aiii, For example, for each direction, cell ai observed five repetitions of gratings that consisted of eight cycles shown at four spikes/s (the time axis is 250 ms) for a total of 40 cycles. Cells that observed gratings at 1 spikes/s have a 1 s time axis. Peristimulus time histograms are binned in 10 ms bins. Cells ai, aii, and aiii are the same LGN cells shown in Figure 2a, top row, with data taken before training. bi–biii, Cells bi, bii, and biii are the same LGN cells shown in Figure 2b, top row, with data taken after training.
We next asked whether 6 h of visual experience, which causes rapid and reliable induction of direction selectivity in visual cortex (Li et al., 2008; Van Hooser et al., 2012; Roy et al., 2016; Ritter et al., 2017; Roy et al., 2020a), caused changes in the amount of orientation selectivity measured in LGN cells. We found that visually naive LGN neurons did not significantly increase their selectivity for orientation following 6 h of training [trained LGN cell median, 0.12, Kruskal–Wallis test, H(1) = 0.1037, p = 0.7474; Fig. 2c]. We then compared the LGN orientation tuning values for the after (trained) condition with the naive and trained cortical cells. Representative tuning curves for trained LGN cells and experienced cortical cells are shown in Figure 2, a and b, bottom rows, respectively. We found that LGN orientation tuning values for the after (trained) condition were significantly lower than the values of the visually naive cortical neurons [Kruskal–Wallis test, H(1) = 74.3566, p < 1e-308] and the trained cortical neurons [Kruskal–Wallis test, H(1) = 79.3016, p < 1e-308]. When we compared orientation selectivity values for visually naive LGN neurons with LGN neurons following 6 h of training, we found no significant difference [Kruskal–Wallis test, H(1) = 0.1037, p = 0.7474].
After examining orientation index values, we compared direction selectivity index values of naive LGN neurons (median, 0.14) with cortical values and found that they were not significantly lower than naive cortical neurons [median, 0.12: Kruskal–Wallis test H(1) = 1.8852, p = 0.1697] but were significantly lower than trained cortical neurons [median, 0.20; Kruskal–Wallis test H(1) = 11.6684, p < 6.3570e-04; Fig. 2d]. We then compared direction selectivity values for trained LGN neurons with naive and trained cortical cells and found no significant difference between trained LGN and naive cortical neurons [Kruskal–Wallis test, H(1) = 1.1816 p = 0.2770] but found a significant difference between trained LGN neurons and trained cortical neurons [Kruskal–Wallis test, H(1) = 6.7742, p = 0.0092]. When we compared direction selectivity values for visually naive LGN neurons with LGN neurons following training, we found no significant difference [median, 0.155; Kruskal–Wallis test, H(1) = 0.0012, p = 0.9726; Fig. 2d].
Comparisons of orientation and direction selectivity to other mammals
The presence of a few LGN cells with very high empirical orientation and direction selectivity index values begged the question of whether these cells resembled the populations of robust orientation- or direction-selective cells identified in the mouse (Marshel et al., 2012; Piscopo et al., 2013) or rabbit (Swadlow and Weyand, 1985; Hei et al., 2014). In these species, direction-selective cells exhibit direction preference values that are highly aligned to the cardinal directions (up, down, left, right) and essentially do not exist with other preferences. Further, in the superficial mouse LGN, direction-selective cells exhibit direction preferences for left/right (anterior/posterior; Marshel et al., 2012). In all these species, orientation- and direction-selective cells exhibit high firing rates and robust tuning.
To better compare the angle preferences of our cells with those of mouse and rabbit, we plotted orientation (Fig. 4a,b) and direction (Fig. 4c,d) preferences in polar coordinates (with the length representing the selectivity index) for LGN and cortical cells from Ritter et al. (2017). Unlike in mouse (Marshel et al., 2012) and rabbit (Hei et al., 2014), where biases for cardinal orientations and directions are strong, the orientation or direction biases in young ferret LGN neurons were more subtle, although many of these biases significantly differed from a pure uniform distribution of angles. To restrict ourselves to neurons with some selectivity, we only analyzed angle preferences from neurons that exhibited selectivity index values of 0.2 or higher. The distribution of naive LGN neuron orientation preferences differed significantly from uniform (Rayleigh's test, z = 44.7, p = 1.36e-33), as did the distribution of naive LGN neuron direction preferences (Rayleigh's test, z = 5.8; p < 0.0028). The trained LGN neurons also exhibited significant deviations from uniform in orientation preference (Rayleigh's test, z = 15.9, p < 1.9e-11), although trained LGN neurons did not exhibit biased direction preferences (Rayleigh's test, z = 0.16, p < 0.85).
LGN neurons exhibit slight biases in their distribution of orientation and direction preferences, and many LGN neurons with high empirical orientation or direction selectivity index values exhibit very weak firing rates. a, Vector plots indicating orientation selectivity and orientation preference for all cells from LGN (top) and V1 (bottom; Ritter et al., 2017) before visual stimulation (Naive). Degree of orientation selectivity is expressed as 1-CV in orientation space, where length of vector indicates orientation selectivity index, ranging from zero to one. Preferred orientation angle ranges from 0° (horizontal) to 90° (vertical) and back to 0°/180° (horizontal). b, Vector plots indicating orientation selectivity for all cells from LGN (top) and V1 (bottom; Ritter et al., 2017) after visual stimulation (Trained). Conventions are the same as in Figure 3a. c, Vector plots indicating direction selectivity for all cells from LGN (top) and V1 (bottom; Ritter et al., 2017) before visual stimulation (Naive). Degree of direction selectivity is expressed as 1-DirCV in direction space, where length of vector indicates direction selectivity index, ranging from zero to one. Arrows indicate stimulus direction. d, Vector plots indicating direction selectivity for all cells from LGN (top) and V1 (bottom; Ritter et al., 2017) after visual stimulation (Trained). Conventions are the same as in Figure 3c. e, Firing rates (spikes/s) to orientation that elicited the greatest response for all LGN cells. Degree of orientation selectivity is expressed as 1-CV in orientation space. f, Firing rates (spikes/s) to direction that elicited the greatest response for all LGN cells. Degree of direction selectivity is expressed as 1-DirCV in direction space. g, Firing rates to orientation that elicited the greatest response for all V1 cells (Ritter et al., 2017). Conventions are the same as in e. h, Firing rates to directions that elicited the greatest response for all V1 cells (Ritter et al., 2017). Conventions are the same as in f.
Cortical cells also exhibited a biased distribution of orientation angle preferences, consistent with previous intrinsic signal imaging studies of ferret that found small orientation preference biases toward the cardinal directions, particularly toward the horizontal, in immature ferret cortex (Coppola et al., 1998; Coppola and White, 2004). The orientation preferences of naive cortical neurons differed from uniform (Rayleigh's test, z = 71.5, p < 1.23e-36), as did trained cortical neurons (Rayleigh's test, z = 53.7, p < 1.68e-27). By contrast, direction angle preferences in cortex did not exhibit any deviations from uniform in either naive cortex (Rayleigh's test, z = 0.99, p = 0.37) or trained cortex (Rayleigh's test, z = 0.158, p = 0.85).
Overall, the tuning preferences were different from uniform for orientation for both LGN and cortex, but direction preference biases were only seen in naive LGN. We did not observe any prominent cardinal peaks as seen in adult mouse or rabbit LGN (Marshel et al., 2012; Hei et al., 2014). Given the small number of highly directional cells that we observed here, we cannot say rigorously whether there is any cardinal bias (A Monte Carlo simulation suggests 75–85 directional cells would be needed to have a 95% chance of detecting a bias, if 50% of cells were part of a cardinal population.). Further, we cannot say whether a higher percentage of cells in adult animals would exhibit substantial orientation or direction selectivity.
As a next step, we sought to determine whether LGN cells with empirically high orientation and direction selectivity also exhibited robust firing rates. We plotted the response to the preferred stimulus against orientation and direction selectivity index values. In LGN, there was a strong negative correlation between maximum firing rate and orientation (Fig. 4e; correlation coefficient test, naive, ρ = −0.2782, p = 0.0003; trained, ρ = −0.4901, p = 0.0012) and direction (Fig. 4f; correlation coefficient test, naive, ρ = −0.2716, p = 0.0004; trained, ρ = −0.3386, p = 0.0081) selectivity index values. Although all these cells exhibited significant firing by passing an ANOVA between all stimulus conditions and a control stimulus, many (though not all) of the LGN cells with high orientation or direction selectivity index values were those with the weakest responses. All of these weakly firing cells exhibited significant variation in responses across stimuli in an ANOVA test (see above, Materials and Methods), but absolute firing rate was not used as a criteria for inclusion or exclusion. Therefore, we don't find much evidence of robust orientation- and direction-selective cells among our population of LGN cells, which had receptive fields centrally located (in front of the animal).
In cortex, there was a weaker relationship between maximum firing rate and selectivity index for orientation (Fig. 4g; naive, ρ = −0.0987, p = 0.1950; trained, ρ = −0.1330, p = 0.1298), but a strong negative correlation for direction (Fig. 4h; naive, ρ = −0.3399, p < 1e-4; trained ρ = −0.1869, p = 0.0326). Further, there were many examples of cells with high firing rates and high orientation or direction preference index values. This evidence suggests that there are robust orientation- and direction-selective neurons in cortex, whereas there are few such cells in our LGN sample.
Sustainedness/transience
Our first analyses showed that LGN cells did not exhibit increases in orientation or direction selectivity themselves, but it remained possible that other short-term changes in receptive field properties could aid a cortical calculation of direction selectivity. The circuitry that underlies direction selectivity in carnivore visual cortex is still a matter of research, but we can gain some clues about what receptive field properties might be important for the input to such a circuit from the literature.
Marr and Ullman (1981) hypothesized that direction selectivity might arise by combining a low-latency, transient input at one location (such as a Y cell) with a longer-latency, sustained input (such as an X cell) at a neighboring location to create a direction-selective cell. This neuron would be selective for motion that arose on the side of the location of the sustained input and moved in the direction of the transient input (Fig. 5a). Evidence for this arrangement was found in the projections of the mouse LGN to direction-selective cells in visual cortex (Lien and Scanziani, 2018). If there was a major experience-dependent change in the sustainedness or transience of LGN cells, such as the appearance of sustained responses or transient responses, then the availability of these signals could have an impact on the calculation of cortical direction selectivity.
Schematic illustrating how direction selectivity could arise in V1 from combined LGN inputs. a, A direction-selective V1 cell is created by combining an input at one location (such as a Y cell) with a sustained input (such as an X cell) at a neighboring location to create a direction-selective cell. Right, Hypothesized space-time receptive fields of the LGN cells are shown before and after visual experience. In this hypothesis, the X cell becomes more sustained with experience. The cortical neuron would be selective for motion that arose on the side of the input location of the X cell and moved in the direction of the transient input. If LGN cells underwent experience-dependent changes in their sustainedness or transience, direction selectivity in V1 could be amplified. b, A direction-selective V1 cell is created by combining inputs selective for specific spatial locations in the receptive field of the cell and are stimulated by a moving stimulus in a precise temporal order. In this case, a direction-selective V1 cell could be created if the latency of the Y cell was reduced by experience, resulting in a short-latency (SL) LGN cell input at one location and a longer latency (LL) LGN cell input at a neighboring location. If LGN cells exhibited changes in latencies with experience, direction selectivity in V1 could be amplified.
We examined the sustainedness and transience of LGN neurons using counterphase sinusoidal gratings (0.08 cycles/degree; temporal frequency 1 Hz) presented at six different spatial phases. These stimuli, when shown at a phase that lined up with the receptive field center, flipped the stimulus provided to the center from ON to OFF each 0.5 s. We chose the spatial phase that elicited the strongest response (F1) for analysis. From the PSTH we found that the peak firing rate of the LGN neurons did not significantly change after exposure to the training stimuli (t(54) = −0.0893, p = 0.9291; Fig. 6a–e). The maintained rate also did not significantly change after exposure to the training stimuli (t(54) = −1.5392, p = 0.1296; Fig. 6f). The transience values for the naive LGN neurons (median, 0.9008) changed very slightly with exposure to training stimuli (median, 0.87455; t(54) = 1.9932, p = 0.0513), but not significantly (Fig. 6g). In summary, 6 h of visual experience had no significant effect on orientation or direction selectivity, or sustainedness or transience of LGN neurons. This suggests that there was no change in the availability of sustained or transient signals to cortex as a result of short-term experience.
Spatial phase analysis, sustainedness and transience, and latency of LGN cells. a, Three example naive LGN neuron responses (F1) to six different spatial phases. b, Raster plots from a naive LGN neurons responses to the spatial phase that elicited the strongest response. Responses from 200 trials were averaged for PSTHs (10 ms time bins) of cell responses. White background indicates one phase, and gray background indicates new phase. From the PSTH of the response we computed peak response (red dashed line), maintained response, and transience. c, Three example trained LGN neuron responses (F1) to six different spatial phases. d, Raster plots from c trained LGN neurons responses to the spatial phase that elicited the strongest response. Conventions are the same as in b. e, Box plot of peak rate values for naive (before; n = 22) and trained (after; n = 34) LGN cells. Central line indicates median (top), and bottom of box indicate 25th and 75th percentiles, whiskers indicate minimum and maximum values not considered outliers, and outliers are indicated by plus sign. There was no significant change in peak rate with training (t(54) = −0.0893, p = 0.9291). f, Box plot of maintained rate values for naive and trained LGN cells. Conventions are the same as in e. There was no significant change in maintained rate with training (t(54) = −1.5392, p = 0.1296). g, Box plot describing transience of cells before and after training. Conventions are the same as in e. There was no significant difference in transience with training (t(54) = 1.9932, p = 0.0513). h, Box plot of response latencies of cells before and after training. There was no significant change in response latency after training (t(54) = −0.5325, p = 0.5966). i, Box plot of coefficient of variation cells of peak response across trials before and after training. There was no significant difference in coefficient of variation after training (t(54) = 0.2519, p = 0.8021).
Latency and reliability
In our next set of analyses, we examined changes in response latency. According to the Reichardt (1961) model, maximum activation of a direction-selective cell occurs when its inputs are selective for specific spatial locations in the receptive field of the cell and are stimulated by a moving stimulus in a precise temporal order. In this case, when a stimulus moves in the preferred direction of the cell it activates the inputs with longer latencies first and progresses toward activation of shorter latencies, allowing for the subthreshold inputs to simultaneously arrive at the soma and summate of the cell, causing an action potential (Fig. 5b). A stimulus moving in the opposite direction, therefore, would activate cells with the shortest latencies first and progress toward longer latencies and a subthreshold response by the postsynaptic cell. If we saw a change in response latencies of thalamic inputs with short-term experience, such as a sharpening of latency tuning, this might suggest that a precise arrangement of thalamic inputs onto cortical cells contributes to direction selectivity development (Reichardt and Poggio, 1976; Reichardt, 1987).
To test this model, we examined initial response latencies of naive LGN cells and compared these with response latencies following exposure to training stimuli. We found that initial latencies did not significantly change with 6 h of visual experience (t(54) = −0.5325, p = 0.5966; Fig. 6h). This suggests that short-term experience does not cause a change in response latencies of thalamocortical inputs.
We also quantified the reliability of responses by measuring the coefficient of variation of the peak response across trials and found no significant difference between initial and trained responses (t(54) = 0.2519, p < 0.8021; Fig. 6i).
Spatial and temporal frequency
Although we found no change in response latencies, we were interested in assessing whether there were changes in spatial frequency tuning and temporal frequency tuning with short-term visual experience. The primary reason we covaried spatial and temporal frequency with direction was to ensure that we assessed orientation and direction at the preferred spatial and temporal frequency of each cell, but these data allows us to examine spatial and temporal frequency tuning for stimulation in the preferred direction. We examined the responses of cells to six different spatial frequencies and six different temporal frequencies using moving sinusoidal gratings at the preferred direction of each cell (Fig. 7a,b). We found no significant change in spatial frequency tuning with visual stimulation training [Kruskal–Wallis test, H(1) = 0.0044, p = 0.9470; Fig. 7c].
Spatial and temporal frequency analysis of LGN cells. a, Spatial frequency tuning curves of three representative LGN cells before 6 h of visual stimulation training (top row) and after training (bottom row). Top and bottom tuning curves do not come from the same cell. Error bars indicate SE. b, Temporal frequency tuning curves of three representative LGN cells before 6 h of visual stimulation training (top row) and after training (bottom row). Conventions are the same as in a. c, Cumulative histogram of LGN cells preferred spatial frequency (cpd) before and after training. Preferred spatial frequencies did not significantly increase with training [Kruskal–Wallis test, H(1) = 0.0044, p = 0.9470]. d, Cumulative histogram of LGN cells preferred temporal frequency (Hz) before and after training. Preferred temporal frequencies significantly increased with training [Kruskal–Wallis test, H(1) = 32.7565, p = 1.0446e−8].
We similarly examined responses of cells to differing temporal frequencies. In previous recordings in visual cortex (Ritter et al., 2017), temporal frequency preferences increased over the duration of experiments for animals that were provided with 6–9 h of experience with either drifting gratings or control animals that merely watched a gray screen for 6–9 h. These increases in temporal frequency preferences occurred in all groups, including those that exhibited increases in direction selectivity (the animals that watched moving gratings) and those that did not exhibit increases in direction selectivity (the animals that watched a gray screen). Therefore, we expected to see an increase in temporal frequency preferences with the duration of the experiment. Indeed, when we examined responses of cells to different temporal frequencies, we found a significant increase in the preferred temporal frequency of LGN neurons with training [Kruskal–Wallis test, H(1) = 32.7565, p = 1.0446e-8; Fig. 7d]. The mechanism of this process is unknown, but these increases in temporal frequency did not correlate with increases in direction selectivity in previous studies and rather were only correlated with the duration of the experiment (Ritter et al., 2017).
Discussion
We assessed the impact of early visual experience on developing LGN neurons in visually naive ferrets by characterizing the response properties of immature LGN neurons and the impact of moving visual stimuli. We found that naive ferret LGN cells did not exhibit strong orientation or direction selectivity. Further, the low orientation and direction selectivity index values of ferret LGN neurons were not increased by 6 h of exposure to a moving stimulus. Other receptive field properties that might influence inputs to a circuit that computes direction selectivity, such as the degree of sustainedness and transience or response latency, were also not altered by short-term experience. These data suggest that orientation and direction selectivity, and early changes in visual response properties, develop downstream from LGN. These results are consistent with the idea that V1 is the earliest stage in the ferret visual system that exhibits substantial orientation and direction selectivity in both developing animals and adults.
Orientation and direction selectivity in LGN
Previous studies in mouse and rabbit (but not squirrel; Zaltsman et al., 2015) have shown that there is a high percentage of orientation- and direction-selective cells in LGN (Swadlow and Weyand, 1985; Marshel et al., 2012; Piscopo et al., 2013; Scholl et al., 2013; Zhao et al., 2013; Suresh et al., 2016) and that, in mice, cortex receives orientation-selective inputs from LGN (Cruz-Martín et al., 2014; Kondo and Ohki, 2016; Sun et al., 2016). In contrast, we found that LGN cells in ferret are not strongly tuned for orientation and direction at the onset of visual experience or with short-term visual experience. Although our surveys were limited to the part of the visual field that was directly in front of the animal, cortical neurons that serve the same visual region exhibit substantial direction selectivity, and indeed, it is a property of a typical cell (Gilbert, 1977; Weliky et al., 1996; Clemens et al., 2012). Our findings imply that individual thalamocortical cells do not significantly contribute to direction selectivity in cortex at this developmental stage. Had it been the case that LGN presynaptic inputs to V1 were orientation- and direction-selective, it could have been possible that LGN was performing the computation that contributes to cortical direction selectivity and that V1 was inheriting direction selectivity.
It should be noted that a limitation of our study is that the layer of origin was not identified for the cells recorded here, and it is possible that orientation- or direction-selective cells could be enriched in layers or portions of the LGN that we did not sample.
It is unlikely that LGN cells contribute to the rapid emergence of direction selectivity
Before this study, what roles could LGN cells have conceivably played in the rapid development of direction selectivity in ferret that occurs following 3–9 h of visual stimulation with moving stimuli? Li et al. (2008) showed that direction selectivity develops only for the orientation columns that are stimulated; for example, if the animal observed vertical stimuli moving left and right, then direction selectivity developed in the vertical orientation columns but not in the horizontal orientation columns. Therefore, if changes in LGN receptive fields could have contributed exclusively to this increase in direction selectivity, one would have to imagine that it would be through increases in direction selectivity in individual LGN cells that somehow projected to the vertical orientation columns. This outcome was unlikely, as it is at odds with the prevailing spot detector view of most LGN receptive fields. Nevertheless, the young brain is highly plastic, and it was prudent to check for this possible contribution. Our results here do not provide any evidence in favor of an increase in orientation or direction selectivity for LGN cells, so it is highly unlikely that such an increase is responsible for the rapid increases in cortical direction selectivity.
A more likely possibility was that changes in both cortex and LGN could have contributed to the rapid increases in cortical direction selectivity with experience. An optogenetic study that provided direct stimulation of the cortical surface for 9 h showed that activity provided to cortex alone, without direct stimulation of the LGN, was sufficient to increase direction selectivity in the naive ferret cortex (Roy et al., 2016). These experiments showed that cortical stimulation and resulting changes were enough, in principle, to cause increased direction selectivity. However, these experiments did not tell us that plasticity in cortex was the only process that contributes to the increases in direction selectivity with acute visual experience. It was possible that concomitant experience-dependent changes in LGN receptive fields also contributed to experience-dependent changes that were observed in cortical cells.
In this article, we outlined how changes in the available LGN cell latencies or increases in sustainedness or transience could have contributed to increased direction selectivity by providing cortex with more reliable information about stimulus direction. Changes of this type would not have been enough by themselves to explain the increases in direction selectivity in ferret; presumably, all LGN cells that are activated by moving gratings would have undergone these changes, yet Li et al. (2008) only observed changes for particular orientation columns. Clearly, changes in cortex that were specific to the orientation columns stimulated would also have been needed to take advantage of the increase in information that might have been provided by LGN cells, but it would have allowed for a dual role of LGN and cortex in the development of direction selectivity.
Instead, we found no evidence for short-term changes in LGN receptive field latencies or sustainedness/transience and no overall evidence of changes in LGN that might have contributed, even in part, to the rapid emergence of direction selectivity in cortex. Given (1) the orientation specificity of the emergence of direction selectivity (Li et al., 2008) and that orientation selectivity is first found in cortex (Hubel and Wiesel, 1959), (2) the fact that direct optogenetic stimulation of the cortex causes emergence of direction selectivity (Roy et al., 2016), and (3) we find no evidence of changes in LGN receptive fields with short-term experience (this article), we conclude that the rapid emergence of direction selectivity in ferret visual cortex involves only processes that are taking place in the cortex. These changes could include modifications of LGN to V1 cell connections in cortex, modification of intracortical connections, and also include, in part, changes to cortical neuronal excitability (Roy et al., 2020b).
Rapid development versus normal development
Under normal conditions, the experience-dependent development of direction selectivity in carnivore visual cortex takes place over ∼2 weeks after eye opening (Li et al., 2006). During this time, LGN receptive fields do become smaller and exhibit shorter latencies (Tavazoie and Reid, 2000), temporal frequency preferences increase (Cai et al., 1997), and the latencies of the fastest inputs to visual cortical neurons also decrease (Roy et al., 2020b). The results here show that these changes in LGN receptive fields are not necessary for increases in cortical direction selectivity.
This result has important implications for the circuit mechanisms that underlie direction selectivity in V1. Many models that are influenced by the Reichardt (1961) model rely on precise delays of feedforward inputs at different locations (Saul and Humphrey, 1990; Feidler et al., 1997; Van Hooser et al., 2014), but direction selectivity can be formed acutely in cortex well before the final latencies of LGN cells are established. Because direction selectivity persists despite changes in the latencies of the input cells, these results provide indirect support for models of direction selectivity that do not require precisely timed excitatory input.
One class of such models rely on feedforward inhibition that modifies feedforward excitation. This feedforward inhibition might arise in an asymmetric manner via lateral cortical connections, so that more inhibition is triggered by stimulation on the null side, as in the rodent and lagomorph retina (Barlow and Levick, 1965; Euler et al., 2014). A study that used optogenetic stimulation uncovered asymmetric null-side inhibition onto layer 2/3 neurons in ferret visual cortex (Scholl et al., 2019), and it's possible that this configuration is found in layer 4 as well. However, intracellular studies by Priebe and Ferster (2005) found no evidence for strong null-side inhibition in layer 4 neurons in cat. Instead, disynaptic feedforward inhibition that arrives at particular phases of stimulation may confer direction selectivity (Priebe and Ferster, 2005; Freeman, 2021). A mechanism like this might be robust to refinements in LGN cell latency as the disynaptic delay from feedforward interneurons to cortical excitatory neurons would be relatively constant even as the latencies of the feedforward inputs changed.
Another feedforward model that does not require such exquisite timing of LGN inputs is the sustained/transient convergence model (Marr and Ullman, 1981; Lien and Scanziani, 2018). If inputs at one side of the receptive field are transient, and inputs at the other side of the receptive field are sustained, then the temporal summation of these two inputs could still produce a direction-selective signal, even if the absolute latencies of these inputs changed over development.
Footnotes
This work was supported by National Institutes of Health–National Eye Institute Grant EY022122. We thank David Landesman for initial contributions to the chamber prototype and members of the Van Hooser lab for comments.
The authors declare no competing financial interests.
- Correspondence should be addressed to Stephen D. Van Hooser at vanhoosr{at}brandeis.edu
