Neural representation of object approach in a decision-making motor circuit

Abstract

Although behavior is ultimately guided by decision-making neurons and their associated networks, the mechanisms underlying neural decision-making in a behaviorally relevant context remain mostly elusive. To address this question, we analyzed goldfish escapes in response to distinct visual looming stimuli with high-speed video and compared them with electrophysiological responses of the Mauthner cell (M-cell), the threshold detector that initiates such behaviors. These looming stimuli evoke powerful and fast body-bend (C-start) escapes with response probabilities between 0.7 and 0.91 and mean latencies ranging from 142 to 716 ms. Chronic recordings showed that these C-starts are correlated with M-cell activity. Analysis of response latency as a function of the different optical parameters characterizing the stimuli suggests response threshold is closely correlated to a dynamically scaled function of angular retinal image size, (t), specifically kappa(t) = (t-delta x e(-beta(t-delta)), where the exponential term progressively reduces the weight of (t). Intracellular recordings show that looming stimuli typically evoked bursts of graded EPSPs with peak amplitudes up to 9 mV in the M-cell. The proposed scaling function kappa(t) predicts the slope of the depolarizing envelope of these EPSPs and the timing of the largest peak. An analysis of the firing rate of presynaptic inhibitory interneurons suggests the timing of the EPSP peak is shaped by an interaction of excitatory and inhibitory inputs to the M-cell and corresponds to the temporal window in which the probabilistic decision of whether or not to escape is reached.

Figure 1.

Visual-evoked escapes in goldfish. A, Behavioral setup and stimulus presentation. Left and top right, Projecting the image of an expanding disk onto a screen above the animal triggers a short-latency body bend (top right). Bottom right, Chronic recordings of brainstem activity show a single Mauthner axon spike (*) preceding a visual evoked C-start. Later impulses are attributed to firing of other reticulospinal neurons. The bottom trace indicates onset and offset of the looming stimulus (Stim.). The apparent collision time (c.t.) is indicated by an arrowhead. B, Stimulus characterization. Plotted is the time course of the projected disk size (diameter) for eight different stimuli used in the behavior and physiology experiments. Characteristics of three additional control stimuli are described in Results. C, Mean response latencies (open bars) and their relationship to TTC (dark bars) for seven different stimuli that differ in size and/or approach velocity (± SEM; n = number of responses in N = number of animals). D, Stimulus-specific cumulative response probabilities plotted against response latencies normalized with respect to stimulus duration. The numbers distinguish individual stimuli. Stimulus notations and sample sizes apply to Figures 23–4.

Figure 2.

Elementary optical parameters of looming stimuli at response onset. Left, Time course of θ(t) (A), θ′(t) (B), and τ(t) (C) for seven different looming stimuli with corresponding mean escape latencies for stimuli 1–3 (black dots) and stimuli 4–7 (open circles). The apparent collision time (c.t.) for each stimulus is indicated by an arrowhead. To account for 25 ms central processing time (Canfield, 2003) and 10 ms motor output delay (Preuss and Faber, 2003), latencies were adjusted by 35 ms. Note that τ(t) is independent of object size and thus similar for stimuli 1–3. The right panels are plots of the mean escape latencies (±SEM, horizontal error bars) versus the mean threshold values at response onset (±SEM, vertical error bars) for stimuli 1–7 (±SEM). The solid black lines indicate the best linear fits of threshold values, whereas dotted lines are the best horizontal linear fits of the data. Statistical comparison (GEE method and ANOVA) of the threshold values showed significant differences for θ(t), θ′(t), and τ(t) (see Results).

Figure 3.

Evaluation of candidate scaling functions at response onset. Left, Time courses of the composite functions η(t) (A), κ(t) (B), and ω(t) (C) for seven different looming stimuli with indicated mean escape latencies for stimuli 1–3 (black dots) and stimuli 4–7 (open circles). The apparent collision time (c.t.) for each stimulus is indicated by an arrowhead. Right, Plots of the mean escape latencies (±SEM, horizontal error bars) versus the mean threshold values at response onset (±SEM, vertical error bars) for stimuli 1–7 (±SEM). The solid black lines indicate the best linear fits of threshold values, whereas dotted lines represent the best horizontal linear fits of the data. Statistical comparison (GEE method and ANOVA) of the threshold values showed significant differences for η(t) and ω but not for κ(t).

Figure 4.

Quantitative relationships between escape latency and stimulus properties. A, B, Plots of mean escape latency ± SEM, expressed in terms of the TTC versus d/V (A) and (d/V)^1/2 (B), with linear fits. In both, TTC is adjusted by 35 ms (see Materials and Methods). Corresponding slopes, used to calculate values of candidate functions at response onset, and the correlation coefficients are S₁ = 2.873 s·rad⁻¹ (r = 0.921) and S₂ = 1.270 s·rad⁻¹ (r = 0.8840).

Figure 5.

Visual-evoked responses recorded in the M-cell ventral dendrite. As with the behavior, PSPs were evoked by six looming stimuli that either had the same (stimuli 1–3) or different (stimuli 2, 4, and 6) approach speeds. The superimposed black traces represent the corresponding time courses of κ(t), shifted by 25 ms to account for sensory processing time in the visual pathway (Canfield, 2003). The bottom traces indicate the onset and offset of the looming stimuli. The apparent collision time for each stimulus is indicated by an arrowhead. Note difference in time scale for top and bottom traces. The bottom right panel shows the mean PSP peak time (± SEM, vertical error bars) plotted versus the predicted peak times of κ(t) for stimuli 1–6 (14, 24, 13, 14, 13, 30 responses in 5 animals for stimuli 1–6, respectively). Stim., Stimulus.

Figure 6.

M-cell visual-evoked responses are looming specific. A1, A2, Consecutive traces recorded from an M-cell soma of PSPs evoked by a receding disk (stimulus 2 reversed) and a looming stimulus (stimulus 1), respectively. B1–B3, Successive traces, from another M-cell, of responses to the looming stimulus 2, presented normally, reversed (i.e., receding), and with a checkerboard pattern. In both sets of data, the bottom trace denotes stimulus onset and offset. B/W, Black and white.

Figure 7.

Comparison of rates of rise of evoked PSPs and the corresponding κ(t) functions. A, Sample PSP (top trace) evoked in the M-cell soma by the looming stimulus 6 (black) with corresponding polynomial fit (red). The apparent collision time is indicated by an arrowhead. The individual slopes from PSP onset to PSP peak were calculated using an algorithm (see Material and Methods for details) that detected the onset and peak of the response (black arrows) with respect to a 50 ms baseline before stimulus onset, assuming a linear rise (blue line). Bottom trace, Negative step indicates timing of the stimulus onset and offset. B, Mean values of normalized (Norm.) slopes for the PSPs (± SEM) and for κ(t), η(t), and ω(t) for the six stimuli (9, 12, 7, 14, 5, and 19 responses in 5 animals for stimuli 1–6, respectively).

Figure 8.

Visual-evoked postsynaptic inhibition. A, Schematic of identified visual system projections to the M-cell including feedforward inhibition mediated by midbrain (PHP) interneurons (Zottoli et al., 1987). B, Simultaneous, intracellular recordings from the M-cell soma and extracellular recordings of visual-evoked presynaptic action potentials from a population of inhibitory PHP cells. The top trace shows an evoked PSP, and the middle trace shows the corresponding train of EHPs in response to a visual looming stimulus. The bottom trace indicates onset and offset of the looming stimulus. The inset (above) is a segment of the same recording on an expanded time scale. C, Voltage-clamp (SEVC) recording from a M-cell depolarized 20 mV from resting potential (300 ms voltage step). Visual evoked IPSCs and EPSCs are indicated by outward and inward currents, respectively. D, Transformation of the experimentally derived scaling function e^−βθ(t)(solid line; β = 0.05) into the inhibitory function f(t) = 1 − e^−βθ(t) (dotted line) and its time derivative f′(t) (dashed line). E, Frequency–time distribution of EHP events evoked by the indicated looming stimuli (total of 9, 11, and 14 responses in 4 animals for stimuli 1, 4, and 6, respectively) with the superimposed time course of f′(t) (dashed lines). * indicates the mean PSP peak time for the three stimuli (data from Fig. 5). The apparent collision time for each stimulus is indicated by an arrowhead. Stim., Stimulus.