Midbrain local circuits shape sound intensity codes

Abstract

Hierarchical processing of sensory information requires interaction at multiple levels along the peripheral to central pathway. Recent evidence suggests that interaction between driving and modulating components can shape both top down and bottom up processing of sensory information. Here we show that a component inherited from extrinsic sources combines with local components to code sound intensity. By applying high concentrations of divalent cations to neurons in the nucleus of the inferior colliculus in the auditory midbrain, we show that as sound intensity increases, the source of synaptic efficacy changes from inherited inputs to local circuits. In neurons with a wide dynamic range response to intensity, inherited inputs increase firing rates at low sound intensities but saturate at mid-to-high intensities. Local circuits activate at high sound intensities and widen dynamic range by continuously increasing their output gain with intensity. Inherited inputs are necessary and sufficient to evoke tuned responses, however local circuits change peak output. Push-pull driving inhibition and excitation create net excitatory drive to intensity-variant neurons and tune neurons to intensity. Our results reveal that dynamic range and tuning re-emerge in the auditory midbrain through local circuits that are themselves variable or tuned.

Keywords: high divalents; inferior colliculus; local circuits; monosynaptic; sound intensity.

FIGURE 1

HiDi isolates inputs that form neuronal tuning curves. (A) Left: a head angle of 20° to the horizontal combined with a 90° electrode approach was optimal for accessing a wide range of characteristic frequencies. Middle: range of characteristic frequencies (CFs) in three mice. Depths were measured from the brain surface (the zero point on the abscissa). Acoustically driven responses were not observed in the first ~250 μm spanning the external and dorsal cortices. Right: sample of 40 neurons showing overlap of CF in control and HiDi (p = 1). The black and red circles overlap exactly. (B) Tuning curves are unaffected by 2.5 HiDi. Four cells are shown. Thresholds are indicated in each panel. From left to right: ANOVA: cell 1: 21–29 kHz; p = 0.41; cell 2: 12–22 kHz; p = 0.52; cell 3: 6–32 kHz; p = 0.37; cell 4: 20–34 kHz; p = 0.31.

FIGURE 2

External inputs and local circuits combine to widen dynamic range. (A) HiDi decreases firing rate at mid- to high sound intensities. Spike rasters in three monotonic neurons in control (top panels) and in 2.5 HiDi (bottom panels). CFs and recording depths are indicated. Horizontal bar at abscissa: tone duration. (B) Rate-intensity functions show a wide dynamic range, intensity-variant, response (RIF) for each of the cells in (A). HiDi decreases the dynamic range, and the RIF in HiDi, due to the monosynaptic input (RIF_M), is saturating or slightly non-monotonic at higher intensities. Firing rates are averaged over 12 presentations of the tone at each sound pressure level. (C) The gain of the local circuit, Gain_L, is the ratio of firing rates in control and HiDi. Gain_L = RIF/RIF_M, increases with sound intensity. Gain_L is plotted as a function of sound intensity for each of the neurons in (B). A gain <1 in the right panel is due to a slightly higher firing rate in HiDi at low intensities (~six spikes/s at 11 and 16 dB SPL). Sigmoidal fits. r² from left to right: 0.95956, 0.98407, and 0.94814. (D) Normalized RIFs in control and HiDi and normalized Gain_L for five neurons with control dynamic ranges ≥60 dB (gray lines). Black lines: population average of normalized RIFs and Gain_L. 33 cells. Mean and s.e.m. Sigmoidal fits: RIF, r² = 0.9942; RIF_M, r² = 0.9846; Gain_L, r² = 0.9933.

FIGURE 3

External and local influences on intensity-tuning. (A) Spike rasters of two intensity-tuned neurons (neuron 1, neuron 2) in control and HiDi. HiDi decreases peak firing rate in neuron 1 and increases it in neuron 2. (B) RIFs for each of the neurons in (A). Neuron 1: RIF_M and Gain_L are both excitatory. Top: RIF_M is tuned to the same range of intensities as the output RIF. Bottom: Gain_L, the ratio RIF/RIF_M, is also tuned to the same intensities as the output RIF. In this neuron, the local circuit supplies a gain of 1.9 at peak tuned intensities. Dotted line: gain of 1 implies no net effect of the local circuit. Neuron 2: RIF_M is excitatory, Gain_L is inhibitory. Both RIF_M and Gain_L are tuned to the same intensity range as the output RIF. The local circuit exerts a negative gain on firing rate. (C) Distribution of RIFs in intensity-tuned neurons (gray lines). Normalized data. Number of cells illustrated: RIF: 16; RIF_M: 19; Gain_L(E): 9; Gain_L(I): 8. Peaks for the output RIF, RIF_M, and Gain_L(E) are distributed over a 35 dB range and, for Gain_L(I), over 25 dB in the population. Black lines: population averages. Mean and SD. (D) Left: average normalized RIFs. Mean and SD. Number of cells: RIF: 16; RIF_M: 19; Gain_L(E): 9; Gain_L(I): 8. Gain_L(I) curves are normalized to the minima. Right: Gaussian fits. r² > 0.8793 for all curves.

FIGURE 4

Temporal activation of monosynaptic and local inputs. For both intensity-variant and tuned neurons, onset responses are averaged over the first 20 ms; sustained responses are averaged between 25 and 100 ms. Response onsets are measured from the mean first spike latency. RIFs are normalized. All data are from cells that exhibited a sustained response during a 100 ms tone. (A,B) Intensity-variant neurons. (A) Onset and sustained responses in control (top) and HiDi (bottom) for five cells with dynamic ranges >60 dB. (B) Population averages. Twelve cells. Mean and SD. The RIF due to the monosynaptic input has a short dynamic range during both the onset and sustained portions of the response to the tone. Gain_L was measured at 30, 60, 70, 80 dB SPL prior to normalization of the control and HiDi RIFs and averaged across cells. Local circuit gain does not increase during the onset portion (F_4,44 = 1.22; p = 0.31), but increases during the sustained response (F_4,44 = 9.72; p < 10^-5). (C,D) Intensity-tuned neurons. (C) Onset and sustained responses in control (top) and HiDi (bottom) for three cells with different tuning widths. (D) Population averages. 14 cells. Mean and SD. Since the HiDi and control functions were normalized, their peaks overlap. A slight increase in gain occurs during the onset portion of the tone (F_4,52 = 2.74; p = 0.038). Strong local circuit activation during the sustained portion of the tone occurs during the tuned region (vertical dotted lines). Gain_L was measured prior to normalization of the control and HiDi RIFs and averaged across cells.

FIGURE 5

Monosynaptic excitation and inhibition in wide dynamic range neurons. (A) Spike rasters. Firing rates decrease in HiDi (middle), but increase again in the inhibitory antagonists (bottom). Gabazine (Gz, 50 μM); strychnine (8 μM). The inhibition is a monosynaptic input. (B) RIFs for the cell in (A). The RIF in HiDi/Gz,/strychnine, due to the excitatory component of the monosynaptic input increases throughout the range of intensities, unlike the net monosynaptic input, RIF_M, which saturates. (C) RIF_M consists of an excitatory component, RIF_M(E), and an inhibitory monosynaptic component which exerts a gain, Gain_M(I) = RIF_M/RIF_M(E). Push–pull interaction between monosynaptic inhibition and excitation generates the net monosynaptic input. RIF_M(E)/RIF_M slope ratios: rising limb, 2.12 ± 0.085; falling limb, 2.24 ± 0.13; (D) average threshold and dynamic range. 16 cells. Mean and SEM.

FIGURE 6

Monosynaptic excitation and inhibition in intensity-tuned neurons. (A) Spike rasters in control (top); HiDi (middle); HiDi + Gz + strychnine (bottom). (B) RIFs for the three neurons. Left: the cell in (A); middle, right: two other cells. (C) Normalized RIF_M, RIF_M(E), and Gain_M(I) for the three cells in (B). Gain_M(I) changes the direction of its gain control with sound intensity. Hatched region: Gain_M(I) exhibits a “tuned” gain. (D) Left: linear fits (black lines) of the rising and falling limbs of RIF_M(E) and RIF_M for Neuron 1 in (B). Rising limb: RIF_M, r² = 0.90285; slope, 4.12821 spikes/s/dB SPL; RIF_M(E); r² = 0.92772; slope, 8.2253; falling limb: RIF_M, r² = 0.88935; slope, -3.30769; RIF_M(E), r² = 0.98401; slope, -6.3333. Right: population averages of tuned widths. Tuned widths were measured at half the peak height of normalized functions. 14 cells. Mean and SEM.