Widespread Decoding of Tactile Input Patterns Among Thalamic Neurons

Abstract

Whereas, there is data to support that cuneothalamic projections predominantly reach a topographically confined volume of the rat thalamus, the ventroposterior lateral (VPL) nucleus, recent findings show that cortical neurons that process tactile inputs are widely distributed across the neocortex. Since cortical neurons project back to the thalamus, the latter observation would suggest that thalamic neurons could contain information about tactile inputs, in principle regardless of where in the thalamus they are located. Here we use a previously introduced electrotactile interface for producing sets of highly reproducible tactile afferent spatiotemporal activation patterns from the tip of digit 2 and record neurons throughout widespread parts of the thalamus of the anesthetized rat. We find that a majority of thalamic neurons, regardless of location, respond to single pulse tactile inputs and generate spike responses to such tactile stimulation patterns that can be used to identify which of the inputs that was provided, at above-chance decoding performance levels. Thalamic neurons with short response latency times, compatible with a direct tactile afferent input via the cuneate nucleus, were typically among the best decoders. Thalamic neurons with longer response latency times as a rule were also found to be able to decode the digit 2 inputs, though typically at a lower decoding performance than the thalamic neurons with presumed direct cuneate inputs. These findings provide support for that tactile information arising from any specific skin area is widely available in the thalamocortical circuitry.

Keywords: information processing; integrative neurophysiology; neurophysiology; tactile; thalamus.

FIGURE 1

Spike firing metrics suggested that thalamic neurons formed a continuum, thus not offering a basis for subdividing our thalamic neurons into classes. (A) The spike shapes of two different thalamic neurons. Each example consists of 10 superimposed traces. (B) Relationship between spike shape parameters across all thalamic neurons recorded. The 3D plot at the top contains a multilinear regression overlaid on the data points. Below, each pairwise relationship is plotted separately. (C) Relationships between three different measures of spike firing regularity. The 3D plot on top contains a multilinear regression overlaid on the data points. Below, each pairwise relationship is shown separately.

FIGURE 2

Location of neuronal recording sites in stereotaxic coordinates with the outline of the thalamic nuclei indicated. All thalamic neuron recording sites are shown in horizontal, sagittal and coronal view. For each viewing plane, the plane is split into three ranges and all recording sites within each range is shown. The ranges of each outline level are here presented as lower limit-upper limit the plane of the outline section illustrated; Paxinos and Watson, 2006). The left column shows the horizontal plane with top row: 4.9–5.3 mm (5.3 mm), middle row: 5.6–6.3 mm (6.1 mm) and bottom row: 6.3–7 mm (6.6 mm). In the same manner the middle column shows the sagittal plane with top row: 1.7–2.4 mm (2.1 mm), middle row 2.4–3.1 mm (2.9 mm) and bottom row: 3.1–3.8 mm (3.4 mm). Right column show the coronal range with top row_ –1.8 to –2.5 mm (–2.1 mm), middle row: –2.5 to –3.2 mm (–2.8 mm) and bottom row: –3.2 to –3.9 mm (–3.6 mm). Rt, Reticular thalamic nucleus; VPL, ventroposterior lateral nucleus; VPM, ventroposterior medial nucleus; VL, ventrolateral nucleus; PO, posterior complex; AVVL, anteroventral nucleus of thalamus ventrolateral part; ANG, anterior nuclear group.

FIGURE 3

A majority of our recorded thalamic neurons had above chance decoding of tactile input. (A) Sample raw responses of two different neurons to two different stimulation patterns. Below each sample, the presented stimulation pattern is shown with black markers with light gray lines extending upwards. (B) Schematic of the rat forepaw showing the location of the four pairs of electrotactile stimulation electrodes (Supplementary Figure 1 illustrates the full set of stimulation patterns). (C) Peristimulus Time Histograms (PSTHs) and Kernel Density Estimations (KDEs) of all of the responses to one of the stimulation patterns for the two sample cells. The KDEs of the two cells are shown superimposed and normalized in the diagram to the right. Below each PSTH/KDE the presented stimulation pattern is shown with black markers with light gray lines extending upwards. (D) Confusion matrices of the decoding performance for the two sample cells across all eight stimulation patterns. (E) Decoding performance for the entire population of recorded neurons (blue bars). Orange bars show the corresponding distribution of the decoding following shuffling of the stimulation pattern labels. Green bar below the x-axis corresponds to the mean plus 2 SD of the shuffled data, which was the decision boundary for counting a thalamic neuron as being an above-chance decoder.

FIGURE 4

Response latency times and their relationship to decoding performance. (A) PSTHs and KDEs of the responses to one stimulation pattern in four sample cells. The blue dashed line indicates the estimated response onset latency time in each case (note that the response latency time for each neuron was calculated using the responses from all eight stimulation patterns). No response latency time could be identified for the neuron to the far right. (B) All detected response latency times for neurons during stimulation patterns. (C) All detected response latency times based on responses evoked by single pulse stimulation. (D) Relationship between decoding performance (F1-score) and the logarithm of the response latency time for pattern stimulation. Dashed red line indicates the decoding level threshold defined in Figure 3E. (E) Relationship between the decoding performance and the average firing frequency.

FIGURE 5

Relationships between location and decoding level and response latency time. (A) Schematic of the recording area from a dorsal view. (B) Sagittal section of the brain (with the thalamic nuclei outlined according to Paxinos and Watson, 2006) illustrating one electrode track (white dotted box). (C) 3D location of thalamic neurons visualized in three viewing directions, mapped onto one plane each (horizontal 6.3 mm, sagittal 2.9 mm and coronal –2.8 mm; more precise locations displayed in Figure 2) with color coded decoding performance (top row of plots) and response latency times (bottom row of plots). Response latency times of 10 ms or lower have been colored red. Dotted red lines in each plot indicate the stereotactic planes in the other two plots, for reference. (D) Similar display as in (C) but as 3D projections, with decoding performance shown to the left and response latency times to the right. Note that for visualization, multiple neurons recorded during the same recording session and electrode have been shifted 40 μm in the mediolateral axis, and neurons recorded very closely to each other according to stereotaxic coordinates, but in different sessions, have been shifted 40 μm in the rostrocaudal axis. V1, primary visual cortex; V2, secondary visual cortex; HPF, hippocampal formation; PtA, parietal area; S1, primary somatosensory cortex; CPu, caudate putamen (striatum); LPLR, lateroposero laterorostral nucleus; LDVL, laterodorsal ventrolateral nucleus.