Spectrotemporal receptive fields in the inferior colliculus revealing selectivity for spectral motion in conspecific vocalizations

Abstract

Frequency modulations are a prominent feature of animal vocalizations and human speech. Here we investigated how neurons in the inferior colliculus (IC) of Mexican free-tailed bats respond to the frequency-modulated (FM) direction and velocity of complex signals by extracting their spectrotemporal receptive fields (STRFs) using a family of upward- and downward-moving ripple stimuli. STRFs were obtained in more than half of the cells that were sampled. To verify the validity of each STRF, we compared their features both with tone-evoked responses and by convolving the STRF with several conspecific calls. We show that responses to tones are in close agreement with the STRF and that the responses predicted by convolutions compare favorably with responses evoked by those calls. The high predictability showed that the STRF captured most of the excitatory and inhibitory properties of IC cells. Most neurons were selective for the direction and velocity of spectral motion with a majority favoring the downward FM direction, and most had spectrum-time inseparability that correlated with their direction selectivity. Furthermore, blocking inhibition significantly reduced the directional selectivity of these neurons, suggesting that inhibition shapes FM direction selectivity in the IC. Finally, we decomposed the natural calls into their ripple components and show that most species-specific calls have downward-sweeping FM components with sweep velocities that correspond with the preferred sweep velocities of IC neurons. This close quantitative correspondence among features of signals and responses suggests that IC cells are tuned by inhibition to respond optimally to spectral motion cues present in their conspecific vocalizations.

Figure 1.

Constructing an STRF from neural responses to moving ripple stimuli. A, A subset of moving ripple stimuli with different temporal and spectral modulation rates. B, A single moving ripple stimulus with a temporal modulation rate (ω) of 8 Hz and a spectral modulation rate (Ω) of 0.9 cycles per octave. C, Neural response of an IC cell to the ripple stimulus shown in B. Magnitude and phase of the response are extracted by calculating the amplitude and unwrapped phase of the fundamental present in the recorded period histogram with its period equivalent to one ripple cycle (1/ω). D, Magnitude matrix showing response magnitudes for all ripple stimuli presented. E, Phase matrix showing response phases to the same ripple set. F, By using a range of temporal and spectral modulation rates (ω, Ω), a ripple transfer function is obtained with its magnitude (D) and phase (E) derived from the neural responses to each ripple stimulus. An STRF for the neuron is constructed by linearly summing each ripple stimulus in our set, scaled by its response magnitude and shifted by its response phase, which is analogous to taking the inverse Fourier transform of the ripple transfer function.

Figure 2.

Features of STRFs and their agreement with tone-evoked responses. A, B, STRF generated by a family of moving ripples is shown as a two-dimensional plot, and the same STRF plotted in three dimensions is shown below. Excitation is indicated in red and inhibition in blue, with black contour lines depicting significant regions (see Materials and Methods). The BF_STRF, the frequency on the spectral axis of the STRF that had the highest peak on the temporal axis, was ∼25 kHz. The STRF also had both onset and offset inhibitions, whose frequencies corresponded to but were temporally separated from the excitatory region, and an inhibitory region that flanked the low frequency side of the excitatory region. C, The tuning curve of the same neuron generated by tone-evoked responses. The BF, the frequency to which the neuron was most sensitive, of the tone-evoked tuning curve was also 25 kHz. D, E, Plots showing agreement between tone evoked responses and STRFs in 43 IC neurons.

Figure 3.

STRFs provide accurate predictions of responses to species-specific calls. Spectrograms of each species-specific vocalization are shown in the top, with the evoked response of each IC neuron (red) and the predicted response of its STRF (blue) displayed below each call. Each row shows the predicted and evoked responses of an IC neuron with its STRF on the left. Predicted responses were generated by convolving the STRF of the neuron with the spectrogram of each call. Correlation between predicted and actual response is shown in top right of each panel. A, B, Two neurons in which there were high correlations between their STRF predictions and their actual responses. Convolutions predicted the call selectivity of these neurons because they predicted very low response magnitudes for the calls that evoked little or virtually no responses but predicted high response magnitudes for those calls that evoked strong responses. Those calls that evoked little or no responses were used to demonstrate that convolving STRFs with those calls also predicted little or no activity, but no correlation coefficient was computed for those calls.

Figure 4.

Effects of blocking inhibition on the STRF and tuning curve of an IC neuron. The reduction in surrounding inhibitory regions in the STRF of a neuron before blocking inhibition (control) and while inhibitory receptors were blocked. Blocking inhibition also caused a small expansion of the excitatory region of the neuron. Both bicuculline (60 nA injection current) and strychnine (60 nA injection current) were applied to the cell.

Figure 5.

Inseparability and direction selectivity of STRFs. A, Distribution of the inseparability indices across all IC neurons with a valid STRF that yielded a predictability of 0.3 or higher. B, Distribution of direction selectivity indices across the same neurons. C, Correlation between the two indices indicating a strong contribution of inseparability of the STRFs to direction selectivity. D, Blocking inhibitory receptors of 12 IC neurons reduced their direction selectivity and to a lesser extent their spectrum–time inseparability, indicating that both properties are shaped by inhibition.

Figure 6.

A, B, Contour plots of the magnitude matrices of one neuron containing the ripple responses obtained before (A) and while (B) inhibition was blocked. Before inhibition was blocked, the neuron responded strongly to downward-sweeping ripples in quadrant 1 (Q1) and hardly at all to upward-sweeping ripples in quadrant 2 (Q2), and thus was directionally selective. Its directional selectivity index was −0.91. When inhibition was blocked, the range of SMRs as well as TMRs to which the cell responded increased in both quadrants of the magnitude matrix. Thus, the directional selectivity index was reduced to −0.35 attributable to the smaller difference in overall power between the two quadrants. The resulting STRFs before and while inhibition was blocked are shown in C and D, respectively.

Figure 7.

Decomposing a synthetic FM sweep into its ripple components. A, A spectrogram of a synthetic, downward FM sweep moving at a constant velocity indicated by its slope. B, Magnitude matrix of the FM sweep shown in A obtained by the two-dimensional Fourier transform of its spectrogram. Because FM velocity is equal to the ratio of temporal to spectral modulation rates (TMR/SMR), the ripple composition of the sweep clusters around a line passing through the origin (0,0) with a slope that indicates the FM velocity of the sweep.

Figure 8.

Velocity tuning of IC neurons. A, Contour plot of the first quadrant in the magnitude matrix of an IC neuron showing responses to downward-moving ripples. B, Responses were fit with a two-dimensional Gaussian, which estimated the orientation angle as well as the spread of responses across the quadrant. The orientation of responses was compared with the velocity line (blue) passing through the peak (black dot). The slope of the line represents the BV of the neuron. The deviation of the orientation angle from the angle of the BV line (orientation error) indicates the degree to which an IC neuron is tuned for velocity, in which the larger the error the poorer the tuning. The neuron shown had a BV of 105 octaves/s, an s_x/s_y ratio of 4.5, and an orientation error of 5°. C, Response peaks and orientation lines are shown for 30 IC neurons in reference to different velocity lines. Most IC neurons had a BV between 0 and 100 octaves/s, with a mean of ∼60 octaves/s. D, Distribution of orientation errors obtained from the 30 neurons showing a mean of 0° and an SD of 7°. This shows that most IC neurons in our sample had a strong degree of velocity tuning.

Figure 9.

Decomposing conspecific calls into their ripple components. The top row shows examples of four syllables from four bat calls. The bottom row displays their magnitude matrices that show the ripple composition of each syllable. Similar to the decomposition of the synthetic FM sweep, the tilt in the ripple composition of each matrix indicates the averaged FM velocity present in each syllable. The FM velocity of each syllable is indicated by the dotted blue line in its matrix.

Figure 10.

FM velocities in communication calls and IC neurons. A, Distribution of FM velocities found in 21 calls containing 32 syllables. B, Distribution of the best velocities to which IC neurons are tuned. The two distributions are well correlated (r = 0.7), suggesting that the FM velocities to which IC neurons are tuned and the velocities present in their conspecific calls correspond closely.