Microseconds-level coding of echo delay in the auditory brainstem of an FM-echolocating bat

Abstract

Echolocating big brown bats (Eptesicus fuscus) detect changes in ultrasonic echo delay with an acuity as sharp as 1 µs or less. How this perceptual feat is accomplished in the nervous system remains unresolved. Here, we examined the precision of latency registration (latency jitter) in neural population responses as a possible mechanism underlying the bat’s hyperacuity. We recorded local field potentials in the cochlear nucleus and inferior colliculus of anesthetized big brown bats to sequences of sounds consisting of a simulated frequency-modulated broadcast followed, at various echo delays, by a four-echo cascade. Latencies of the first negative response peak to the broadcast and to the first echo in the cascade were shorter in the cochlear nucleus than in the inferior colliculus, but latency jitter of this peak was comparable in both brainstem nuclei. Mean latency jitter, averaged over all stimulus conditions, was 51 µs in the cochlear nucleus and 56 µs in the inferior colliculus. Latency jitter to the successive echoes in the echo cascades was larger, with means of 125 µs and 111 µs, respectively. These values are lower than values commonly reported for single-neuron latency variability in bats and other mammals, and they approach within an order of magnitude the big brown bat’s psychophysical performance. Latency jitter for synchronized population responses on a scale of microseconds reduces the gap between neurophysiological and behavioral measures of acuity. Further systems-level analysis is necessary for understanding neural mechanisms of perception.

NEW & NOTEWORTHY Echolocating big brown bats resolve time delays with a sharp precision of 1 µs or less. How this hyperacuity is accomplished in the auditory system is unknown. We now report that the precision of latency registration (latency jitter) in population activity from two brainstem nuclei in response to simulated echolocation sounds is in the range of tens of microseconds. These values are smaller than observed in single neuron responses and approach the bat’s psychophysical acuity.

INTRODUCTION

Aerial-hawking insectivorous bats use ultrasonic echolocation for navigation and prey catching (1). Bats broadcasting frequency modulated (FM) echolocation calls (FM bats) determine the distance to a target, such as a small insect, by measuring the time delay between the emitted broadcast and the returning echo, and they classify the size and shape of the target by the spectral structure of echoes (2). Psychophysical experiments show that the FM big brown bat, Eptesicus fuscus, can detect changes in echo delay with acuity as sharp as 1 µs or less (3–8). Even in the most basic range or delay discrimination tests, bats of all species tested have thresholds as small as 50 µs (9). The neurobiological specializations underlying this hyperacute temporal processing remain the subject of considerable experimentation, computational modeling, and debate (10–14).

Experiments on the neural basis of echo delay acuity have searched for neurons within the bat’s central auditory system that could code target distance by firing selectively and sharply to specific time delays between simulated broadcast-echo pairs (the echolocation broadcast and one returning echo). Such delay-tuned neurons have been identified in the auditory midbrain [inferior colliculus (IC)] of several species of FM bats (15–19). In the big brown bat, the range of time delays (best delays: 2–30 ms) and response latencies to which these neurons selectively fire spans the behavioral operating range (up to 5 m) for detecting echoes from insect-sized targets (12, 20). However, delay-tuned neurons have very broad tuning curves, with a sharpness (tuning width) on the order of milliseconds, about equal to the value of the tuned delay itself (15, 16). This broad tuning indicates that delay tuning based solely on a firing rate code cannot account for the bat’s perceptual acuity for echo delay (5, 7, 8, 12).

The lack of correspondence between the sharpness of neural delay tuning and the bat’s perceptual acuity has motivated the search for alternative neural codes based on spike timing and latency (12, 13, 21–24) rather than on firing rate. Neurons in the IC typically respond to individual simulated FM broadcasts and echoes with single spikes (i.e., they are phasic on-responders). The time that elapses between each broadcast’s response and each echo’s response then registers echo delay, with accuracy related to latency variability across the different tuned frequencies of the responding neurons (12, 14). Accurate tuning for echo delay requires highly precise time coding [“nearly constant” (25); “low variability” (26)] of the neural on-response latency. Several studies in the big brown bat’s IC quantified latency jitter (variability) of on-responses of individual neurons by calculating the standard deviations (SDs) of latencies across repeated stimulus presentations (i.e., the trial-to-trial variability). In the first demonstration of delay tuning, Feng et al. (15) observed that some delay-tuned neurons in the intertectal nucleus of the IC coded time delays between broadcast-echo pairs with latency variability as low as 150 μs. Further experiments (16, 19, 26) reported latency jitters in the range of 0.1–6 ms, with a small number of neurons having values as low as 50 µs. In intracellular recordings from the IC, Voytenko and Galayzuk (27) found latency jitters of <100 µs in the initial hyperpolarizing on-response. The smallest values obtained in these studies begin to approach the bat’s psychophysical performance, but the numbers of IC neurons showing this precise timing are few; more typically, values of single neuron latency jitter are in the range of hundreds of microseconds.

Population responses might better reflect behavioral performance than do single neurons (13, 21, 26). Luo et al. (13) recorded extracellular field potentials (i.e., population responses filtered between 200 and 600 Hz) in the IC of awake big brown bats in response to FM broadcasts and broadcast-echo pairs. Latencies to these stimuli were coded by these population responses with high precision, with a median jitter of 104 µs and a smallest value of 53 µs. In contrast, the median SD from the corresponding multiple unit response (filtered between 600 and 3000 Hz) was considerably larger, 425 µs. These data deliver supporting evidence that summed population responses could underlie, at least partially, the big brown bat’s temporal delay hyperacuity.

In the experiment reported here, we examined the precision of latency variability in the IC and then in the cochlear nucleus (CN), the first synapse in the ascending auditory pathway, using population responses instead of single neuron responses. We recorded local field potentials (LFPs), which reflect the synchronized activity of groups of neurons (28). We hypothesized that the latency jitter of LFPs would be tighter than that reported in studies of single-neuron responses. Our results complement previous population studies of the IC (13, 21, 26) by using a different naturalistic stimulus set and by extending the analysis to the CN, a brain area understudied in FM echolocation. Two previous studies of the CN analyzed single-neuron responses to tone bursts (29, 30), but only one of those studies (30) calculated the latency variability of the neural response. That study reported some jitter values under 100 μs, but only in 3% of the sample of recorded neurons. A more recent experiment (31) recorded multiple unit activity from the CN in response to isolated FM broadcasts and found mean on-response latency jitter of 45 μs across a range of stimulus levels, shorter than estimated from single neuron data (30). No study to date has examined latency variability in the CN to naturalistic sequences of broadcasts and echoes and how this compares to that in the IC.

Here, we asked if the precision of latency coding reported in earlier single neuron and population analyses of the IC (13, 26) arises de novo from intrinsic local IC circuitry or whether it is as precise in the CN and transmitted upwards without modification. To do this, we recorded LFPs from the CN, and we reanalyzed LFP data reported in a previous study of the IC (32) to calculate and compare values of latency jitter. The stimulus we used was a multiple echo (echo flow) cascade, consisting of an FM downsweep mimicking the big brown bat’s natural echolocation broadcast followed at varying delays by a sequence of four returning echoes. We quantified response latency and latency jitter to the broadcast and to all four echoes in the cascade. We hypothesized that the precision of latency registration, on the order of tens of microseconds, observed in CN multiple unit activity in response to FM broadcasts presented alone (31) would also be manifested in LFPs to broadcast-echo cascades. We compared our data with those previously reported in the CN and IC of other mammals to ask if precise latency jitter in the bat’s auditory brainstem is a specialization unique to echolocation.

MATERIALS AND METHODS

Animals

Data presented here are based on recordings from five adult big brown bats (2 males, 3 females; ages unknown beyond 2 years), wild-captured from local barns, as authorized by scientific collecting permits from the state of Rhode Island (No. 2011 and 2012). The bats were vaccinated against rabies and were socially housed in a dedicated biohazard level 2 colony room maintained at temperatures of 20–24°C and 55%–65% relative humidity. They had unlimited access to vitamin-enriched water and were fed daily with live mealworms (Tenebrio larvae) in amounts to keep their body weights within the healthy range of 15–20 g. All personnel working with the bats were vaccinated against rabies. Husbandry and experimental procedures were approved by the Brown University Institutional Animal Care and Use Committee (Protocol No. 1111037) and are consistent with US federal guidelines.

Acoustic Stimuli

The experimental stimuli, termed echo flow series (32), consisted of one FM broadcast followed by seven different group time delays by a cascade of four echoes. These series of broadcast-echo cascades resemble what the bat would receive from four reflecting objects, spaced over several meters, while it flew past them. This is a simplified model for flying parallel to vegetation, such as a row of trees. Because of the width of the big brown bat’s biosonar beam (33), a single broadcast produces echoes from all objects within the beam, so that each individual broadcast triggers the return of several echoes at varying time delays. Previous research on FM bats analyzed IC responses to pairs of tones (34, 35) and to simulated echolocation sounds, presented as isolated broadcast-echo pairs or in naturalistic broadcast-echo sequences (13, 15–19, 21, 22, 26). Only two studies to date, both in the IC, examined the neural representation of an FM broadcast followed by cascades of echoes (32, 36), but neither of those experiments quantified latency jitter.

Stimuli were created in MATLAB (Mathworks, Natick, MA) at a sampling rate of 500 kHz and saved as .wav files. Two-harmonic hyperbolic FM downsweeps, resembling the bat’s natural broadcast, consisted of a first harmonic decreasing from 50 to 20 kHz and a second harmonic decreasing from 100 to 40 kHz. Stimulus duration was 2.5 ms with 0.4 ms cosine rise/fall profile. The echo flow stimulus (Fig. 1) consisted of one FM downsweep (the broadcast; termed the echo flow broadcast in Ref. 32) followed by four echoes (the cascade) with the same spectral content but different amplitudes. Stimuli were designed as a set of seven consecutive decreasing broadcast-echo delays, ranging from 20.4 to 14.4 ms (1-ms steps). These echo delays mimic the presence of successive reflecting objects at decreasing distances from and parallel to a flying bat. The intervals between the echoes in the echo cascade were 2.1, 3.5, and 6.8 ms. The first echo was lower in amplitude by 1.5 dB than the broadcast; the three successive echoes were further attenuated by 1.5, 1.7, 2.8, and 5.9 dB.

Figure 1.

Digitized stimulus waveforms of the echo flow stimuli. In each individual plot, waveform 1 (beginning at time 0) is the broadcast, a frequency-modulated (FM) downsweep resembling the bat’s natural echolocation broadcast. The 4 subsequent waveforms are the echoes, each of which is an FM downsweep with the same spectral content of the broadcast but at different amplitudes. The echo delay between the broadcast and the first echo in the echo cascade decreases in seven 1-ms steps from 20.4 to 14.4 ms (top to bottom; equivalent target ranges: 3.52 to 2.48 m). Waveform peak-to-decay envelope compensates for loudspeaker frequency response. During experiments, echo flow stimuli were presented as a set (complete series of 7 decreasing echo delays).

During experiments, stimulus .wav files were accessed in Labview (National Instruments, Austin, TX) and loaded to one D/A channel of a National Instruments PCI-6111e board. An on-board counter triggered the D/A conversion process (500-kHz clock rate). The analog files were low pass filtered at 200 kHz (Wavetek Rockland Model 442, San Diego, CA), attenuated (Hewlett Packard 350 D, Loveland, CO), amplified (Harmon/Kardon PM465, Stamford, CT), and presented through an ultrasonic loudspeaker (EAS-10TH1000, Panasonic, Osaka, Japan) located 50 cm away from the midline of the anesthetized bat’s head. Stimulus levels were calibrated with a Brüel & Kjaer (Naerum, Denmark) Model 4135 one-fourth-inch microphone placed at this same distance away from the loudspeaker. At 0 dB of attenuation, the speaker output at the bat’s ears was ∼100 dB SPL re 20 µPa.

Surgical Procedures

Bats were anesthetized for surgery and during recordings by intramuscular injections of a mixture of medetomidine/midazolam/fentanyl (0.4/4.0/0.4 μg/g body weight). Anesthetized bats’ bodies were lightly wrapped in a cotton surgical towel with a small heating pad placed under the ventrum, and the animals were positioned dorsal-side-up onto a custom-built platform for surgery. Head fur was shaved and muscles retracted. A small craniotomy was made through the skull overlying either the left or the right IC. The bats’ body temperatures were maintained within the range of 30–34°C by the application of fresh heating pads onto the back. Supplemental doses of the anesthetic mixture were administered as needed to maintain anesthesia during physiological recordings.

Electrophysiological Recordings

Bats were placed on a platform in a sound-attenuating chamber (Industrial Acoustics, North Aurora, IL). A low impedance (<1,000 kΩ) tungsten electrode was inserted into the IC through the craniotomy, at an angle of ∼12° ipsilaterally from the midline, and advanced contralaterally through the (opposite) IC by a hydraulic microdrive (PC-5N, Narishige, Tokyo Japan). This angle of insertion was chosen based on a stereotaxic atlas of the big brown bat’s brain (37) to target, first, central and medial regions of the IC (from the surface down to depths of ∼2,000 µm) and, second, the CN (depths of 3,000–4,000 µm). An indifferent needle electrode was inserted into the muscle at the base of the skull. The physiological signal was amplified (PAR Model 113, Princeton Applied Research Corporation, Princeton, NJ), bandpass-filtered (10 Hz to 10 kHz, Wavetek Rockland Model 442), and digitized at 20 kHz (National Instruments PCI-6111e board).

The echo flow stimuli were presented as a set of seven consecutive decreasing echo delays, from longer to shorter delays (Fig. 1), with each echo delay presented for 40 repetitions at an interstimulus interval of 250 ms. Stimuli were set to an attenuation range of −30 to −50 dB and presented typically from higher to lower levels. We made no effort to quantify the response threshold, but at these attenuation levels (∼50–70 dB SPL re 20 μPa based on loudspeaker calibration), responses were above the threshold as estimated visually.

At the completion of CN recordings, electrolytic lesions were made using a current generator (100 μA, 1.4 min, Model 51595, Stoelting, Wood Dale, IL). Bats were then deeply anesthetized by intraperitoneal injection of 0.3 mL Beuthanasia and transcardially perfused with 4% paraformaldehyde solution. Brains were removed, embedded in agarose (5% in 0.9% saline), sliced on a vibratome (50-μm coronal sections) onto gelatin-subbed slides, and processed with cresyl violet acetate solution to identify CN lesion sites and nuclear boundaries.

Data Analysis

Digitized neural responses were imported into MATLAB and bandpass filtered using a second order Butterworth filter between 143 and 500 Hz to display the LFP, as used previously (32). A custom-written MATLAB code based on the findpeaks function detected all peaks in the LFP, up to a time interval of 50 ms poststimulus onset, which exceeded baseline criterion (3 times the amplitude of the root-mean-square activity in the first ms following stimulus onset, i.e., during the acoustic propagation time from the loudspeaker to the bat’s ears). For each detected (above criterion threshold) peak in the LFP, the program calculated the peak amplitude, the standard deviation of peak amplitude, peak latency, and the standard deviation of peak latency over the 40 repetitions of each echo delay. We first identified the prominent peaks in the response to the longest echo delay and followed those same peaks in the responses to the other, shorter echo delays. Response latencies to the broadcast and to the first echo in the echo cascade were calculated from the time delay between stimulus onset and the first detected negative peak in the LFP (which was usually, but not always, the first detected peak in the response). We also extracted the latency jitter (or SD) of the largest positive peak to the broadcast and of the four largest positive peaks to the echo cascade that exceeded threshold criterion.

Statistical analyses were performed using SPSS Statistics v28 (IBM, Armonk, NY). Because of the small sample size and the presence of outliers, we used nonparametric tests (Kendall’s τ, one-tailed; Friedman’s Q, two-tailed) to evaluate how response latency and latency jitter changed with stimulus attenuation and echo delay.

RESULTS

In a previous study (32), we recorded LFPs to the echo flow stimuli from 12 recording sites in the IC of eight bats. In five of these eight bats, we then lowered the electrode to depths of 3,000–4,000 μm below the dorsal surface of the IC to target the CN. CN recordings were not possible in the other three bats because those animals had recovered from anesthesia by the end of the IC recordings. Only one CN site was sampled in each bat, so data are based on recordings from five different sites/animals. Animals were euthanized at the end of CN recordings. Electrolytic lesions were identified around the ventral and medial borders of the CN at levels near the insertion point of the eighth cranial nerve (Fig. 2).

Figure 2.

Approximate position of recording sites in the cochlear nucleus (CN), superimposed on 2 images of the bat’s auditory brainstem. Images are based on cresyl-violet stained sections (thickness 50 µm, sliced in the coronal plane) modified from the atlas of the big brown bat brain (37); x: approximate position of recording sites, displaced by ±100 µm in the anterior/posterior dimension of the 2 planes of section illustrated here. A: level of the anteroventral CN at the entry point of the eighth nerve. Atlas level: 49; bregma: −4.3 mm; lambda: −0.45 mm. Lesion sites for bats BI, OL, and RO. B: level of the anterior region of the posteroventral CN, close to the trapezoid body. Bregma: −4.9 mm; lambda: −1.05 mm. Lesion sites for bats MK and TH.

Latency and Latency Jitter in the Cochlear Nucleus

Figure 3 shows LFP waveforms from one CN recording site to the echo flow stimulus at each of the seven echo delays at stimulus attenuations of −30, −40, and −50 dB. To show the variability in the response, in each plot, the mean waveform at each echo delay is shown in red, and the waveforms in response to each of the 40 repetitions are shown in black. The LFP to the broadcast and to the first echo in the cascade consists of a short-latency negative peak followed by a positive peak. There is little variability in these peaks across the 40 stimulus repetitions; that is, the waveforms are all superimposed. In contrast, there is more variability in the waveforms in the echo delay interval and in the responses to the echo cascade. The latency to the first negative peak in the broadcast remains consistent across echo delays, while that to the first echo decreases with decreasing echo delay. These trends occur at the other recording sites, although the relative amplitudes and latencies of the detected peaks vary, related to differences in the locations of the recording electrode or in absolute threshold of response at each site.

Figure 3.

Local field potentials (LFPs) from 1 recording site in the cochlear nucleus at 3 different attenuation levels (columns: −30 dB, −40 dB, and −50 dB). Labels on the left show the echo flow stimuli (EF1 to EF7) in decreasing echo delays from 20.4 to 14.4 ms. The x-axis shows time in ms, up to 45 ms, and the y-axis shows amplitude. The red waveform in each plot is the mean LFP across 40 repetitions of the stimulus. The black lines show the individual responses to each of these 40 repetitions. There is little variability around the mean waveform in the first negative and first positive peaks to the broadcast and in the first negative peak to the first echo. Variability increases in the echo delay interval and in the responses to the echo cascade. Responses are not corrected for acoustic transmission time. Waveforms were plotted using MATLAB graphics and then imported into Adobe Photoshop 2024 for size adjustment.

Mean latency and latency jitter of the first negative peak in the LFP in response to the broadcast at each of the seven echo delays are plotted in Fig. 4, A and B. Mean latencies to the broadcast, across all bats, are 3.7, 3.9, and 4.2 ms, with individual values varying from 3 to 6.4 ms (Fig. 4A). These latencies increase significantly with increasing stimulus attenuation (τ = 0.28, P < 0.001). Mean amplitude-latency trading, a metric of how changing echo amplitude results in changes in neural response latencies (23), for the broadcast is 20 µs/dB.

Figure 4.

Latency and latency jitter of the first negative local field potential (LFP) peak to the broadcast (A and B) and to the first echo in the echo cascade (C and D) at 3 stimulus attenuation levels (colored symbols) and at each echo delay (x-axis), averaged across the 5 bats. A: mean latency to the broadcast at each of the 7 echo delays (from longer to shorter, x-axis). Error bars are 1 SD around the mean latency. Latency increases with increasing stimulus attenuation. B: latency jitter is the SD around each latency value, averaged across 40 stimulus repetitions for each individual bat and then averaged across bats. Error bars are 1 SD around this mean. Latency jitter increases with increasing stimulus attenuation. C: latencies of the first negative LFP peak to the first echo decrease with decreasing echo delay. D: latency jitter to the first echo is stable across echo delays but increases with increasing stimulus attenuation.

Mean latency jitter to the broadcast (Fig. 4B) is relatively stable across echo delays (when combined across stimulus levels; τ = 0.08, P = 0.10) but increases significantly with increasing stimulus attenuation (when combined across echo delays; τ = 0.38, P < 0.001). Mean latency jitter to the broadcast across echo delays and across bats is 32 µs at −30 dB, 44 µs at −40 dB, and 67 µs at −50 dB (range: 22–153 µs; overall mean: 48 μs).

Latency to the first negative LFP peak in response to the first echo, averaged across all bats, decreases significantly with decreasing echo delay (at −30 dB: τ = 0.78; at −40 dB: τ = 0.71; at −50 dB: τ = 0.7; all values P < 0.001; Fig. 4C). The mean amplitude-latency trading ratio for these data is 18 µs/dB. Latency jitter of this peak increases with increasing stimulus attenuation (means of 44, 50, and 63 µs at the 3 stimulus levels; Q = 14.7, df = 2, P < 0.001; Fig. 4D). Individual values range from 27–101 µs across bats, stimulus attenuations, and echo delays; the overall mean is 54 μs. The latency jitter remains stable across echo delays (Q = 10.3, df = 6, P < 0.11).

We also calculated the latency jitter of the first positive peak following the negative peak (Fig. 3) in response to the broadcast and of the four largest positive LFP peaks to the echo cascades that exceeded the threshold. These values were averaged across the seven echo delays and the five bats. The mean latency jitter for the broadcast calculated in this way is 45 µs at −30 dB, 50 µs at −40 dB, and 64 µs at −50 dB, with an overall mean of 53 μs. Mean latency jitter for the four echoes in the echo cascade is 127 µs at −30 dB, 117 µs at −40 dB, and 133 µs at −50 dB, with an overall mean of 124 μs. Mean jitter values for all four echoes separately at each attenuation level are shown in Table 1. Latency variability increases from the first to the second echo and then remains relatively steady over the rest of the echo cascade.

Table 1.

Latency variability of the positive peaks in the local field potential in response to the four echoes in the echo cascades

Attenuation Level	Cochlear Nucleus				Inferior Colliculus
	E1	E2	E3	E4	E1	E2	E3	E4
−30 dB	68	120	174	148	69	77	147	136
−40 dB	64	94	128	161	67	54	135	164
−50 dB	71	149	137	177	96	82	137	169

Data are means (jitter, µs) over all echo delays and all recording sites. E1, E2, E3, E4: echoes 1, 2, 3, 4.

Latency Jitter in the Inferior Colliculus

To obtain latency jitter in IC responses to the echo flow stimuli, we reanalyzed previous data (32) but only using responses from the same five bats (eight recording sites at 300–2,000 μm below the dorsal surface of the IC) from which CN data were available. Examples of the variability in LFPs from one recording site are displayed in Fig. 5. As in the cochlear nucleus, LFPs from the inferior colliculus show little variability in responses to the broadcast and to the first echo across the 40 stimulus repetitions.

Figure 5.

Local field potentials (LFPs) from one recording site in the inferior colliculus at three different attenuation levels (columns: −30 dB, −40 dB, and −50 dB). Labels on the left show the echo flow stimuli (EF1 to EF7) in decreasing echo delays from 20.4 to 14.4 ms. The x-axis shows time in ms, up to 50 ms, and the y-axis shows amplitude. The red waveform in each plot is the mean LFP across 40 repetitions of the stimulus. The black lines show the individual responses to each of these 40 repetitions. There is little variability around the mean waveform in the first negative and first positive peaks to the broadcast and in the first negative peak to the first echo. Variability increases in the echo delay interval and in the responses to the echo cascade. Responses are not corrected for acoustic transmission time. Waveforms were plotted using MATLAB graphics and then imported into Adobe Photoshop 2024 for size adjustment.

Mean IC latencies to the broadcast are 6.7, 7, and 7.5 ms, respectively, at stimulus attenuation levels of −30, −40, and −50 dB (overall latency range: 5.2–9.9 ms). Mean latency jitter of the first negative LFP peak (Fig. 5A) is 44 µs at −30 dB (8 sites), 67 µs at −40 dB (8 sites), and 46 µs at −50 dB (6 sites), with an overall mean of 52 μs. These values do not change significantly with stimulus attenuation (Q = 3.6, df = 2, P = 0.08). Latency jitter to the broadcast does not vary with echo delay (combined across stimulus level; τ = 0.07, P = 0.11). Mean latencies to the first negative peak in the LFP to the first echo, corrected for and averaged across echo delay, are 7.4, 7.6, and 8.2 ms at these same attenuation level. Latency jitter is 66 µs at −30 dB, 51 µs at −40 dB, and 61 µs at −50 dB, with an overall mean of 59 μs (Fig. 5B). These values remain stable with changes in stimulus attenuation (Q = 0.9, df = 2, P = 0.63). There is no significant relationship between latency jitter and echo delay (Q = 10.7, df = 6, P = 0.099).

We also analyzed the latency jitter of the largest positive peaks in the IC LFPs in response to the broadcast and to the echo cascade (Fig. 6). The mean latency jitter of the positive peak to the broadcast is 42 µs at −30 dB, 55 µs at −40 dB, and 78 µs at −50 dB, with an overall mean of 58 μs. The mean latency jitter of the four positive peaks in the echo cascade is longer, 107 µs at −30 dB, 105 µs at −40 dB, and 121 µs at −50 dB (overall mean: 111 μs). Values for the four individual echoes are shown in Table 1. Latency variability is similar for the first two echoes and then increases in response to the remaining echoes in the cascades.

Figure 6.

Latency jitter of the first negative peak in local field potentials from the inferior colliculus for the broadcast (A) and the first echo (B). Plotted are the mean latency jitter (µs) and 1 SD around this mean at 3 stimulus attenuation levels at each echo delay averaged across recording sites. Data are based on 8 recording sites at −30 and −40 dB and 6 recording sites at −50 dB.

DISCUSSION

In this experiment, we quantified the latency and latency jitter of peaks in LFPs from the CN and from the IC of five anesthetized big brown bats in response to a digitally synthesized mimic of an FM echolocation broadcast followed by a cascade of returning echoes, stimuli similar to that a flying bat might encounter in an environment containing several different ensonified objects. Independent variables were stimulus attenuation level and the echo delay (the time delay between the broadcast and the first echo in each echo cascade). We report finding very small latency jitters (small SDs) to the broadcast and to the echoes in both CN and IC population responses. In the CN, the mean latency jitter of the first negative peak in the LFP, averaged across stimulus attenuations, is 48 μs for the broadcast and 54 μs for the first echo in the cascade, with an overall mean of 51 μs. In the IC, the mean jitter of this response peak is remarkably similar, 52 µs for the broadcast and 59 μs for the first echo, with an overall mean of 56 μs. These mean values are near the limit of the 20-kHz sampling rate used to digitize the data, and so may underestimate actual values. Latency jitter to the four echoes in the cascade, quantified from the four largest positive peaks in the LFPs, is larger than that for the first negative peak but remains in the microsecond range (means of 125 µs in the CN and 111 µs in the IC). The precision in the tens of microseconds range for the averaged synchronous neural firing in LFPs across stimulus presentations found here is sharper than typical values reported in previous single neuron studies (15, 26, 27, 30) and approaches within an order of magnitude the big brown bat’s psychophysical acuity (3–8). Moreover, the precision of latency registration seen in our data is lower than what has been typically seen in the CN and IC in other mammals (38–42). These comparisons are consistent with the hypothesis that precise latency registration in the auditory brainstem is a specialization for the fast temporal processing that underlies echolocation.

Latency Registration in the Bat’s Cochlear Nucleus

Our data contribute to the limited database on sound processing in the CN of FM bats. Two earlier studies [recording sites not identified (29) and recordings in anteroventral and posteroventral CN (30)] recorded single neuron responses to tones in anesthetized little brown bats and awake big brown bats, respectively. On-response latencies in these two studies were 0.6–5.6 ms [mean: 1.8 ms (29)] and 1.3–5.7 ms [10 dB above thresholds (30)]. A study of multiple unit responses from the ventral CN of anesthetized big brown bats (31) to an isolated FM broadcast quantified on-response latencies of 1.3–4 ms at different stimulus levels, varying consistent with amplitude-latency trading (23). This comparison is important because changes in single-neuron response latencies with echo amplitude correspond to the bat’s perceived delay (6).

We now show that the latencies of the first negative peak in the LFP in response to the broadcast and to the first echo in an echo cascade range from 2.9 to 6.5 ms across stimulus levels and also undergo amplitude-latency trading. Response latencies are generally comparable between these experiments, in spite of differences in type of neural response (single neuron, multiple unit, and LFPs), stimulus type (tones compared to FM downsweeps), stimulus levels, and anesthesia [sodium pentobarbital (29); awake animals (30); and medetomidine/midazolam/fentanyl (31; the present study)].

Simmons et al. (31) recorded multiple unit responses to FM sweeps varying in sweep direction at the same five CN recording sites sampled in this experiment. They quantified mean latency jitter, across stimulus levels, of the on-response to FM downsweeps as 45 μs. Here, we show that the comparable mean latency jitter of the first negative peak in the LFP to echo flow stimuli is similar to that of the multiple unit response −47 µs to the broadcast and 54 µs to the first echo in the cascade. For both isolated broadcasts and to broadcasts presented before an echo cascade, latency jitters increase with increasing stimulus attenuation and undergo amplitude-latency trading. On the other hand, latency jitter does not change with echo delay. We also show that the latency jitter of the predominant positive peak to the four echoes in the cascade is larger, with a mean of 125 µs across echo delays and stimulus levels. These larger jitter values are expected from the recruitment of additional neurons contributing to the LFP during the presentation of the entire sequence of echoes.

A single neuron study of the CN (30) reported a mean latency jitter to characteristic frequency tones (10 dB above threshold) of 910 μs; only 3% of neurons had jitters <100 µs (lowest values were not reported, and values were not separated by recording site in the CN). In a smaller sample of “level-tolerant” neurons, the reported latency jitter was 310 μs. These values are considerably larger than those we observed in LFP responses to both the broadcast and to the echo cascades. We suggest that in our experiment, the responses of those 3% of single neurons with jitters <100 µs (30) were most reflected in the LFP.

Single neuron studies in the CN of other mammals (cat: Refs. 38, 39; guinea pig: Ref. 40) calculated latency jitter to characteristic frequency tones. In barbiturate anesthetized cats, Rhode and Smith (Table 1 in Ref. 38) reported mean jitter values of 90 μs in onset units in the ventral CN; other units with different firing patterns had larger values. In awake cats, onset units from the posteroventral CN had mean jitter values of 1.4 ms, although some individual values were as low as 20–50 μs (Table 2 in Ref. 39). It is interesting to note that the latency jitter was smaller in anesthetized compared to awake cats. Winter and Palmer (40) found latency jitters of 100–400 μs in onset responses from the CN of anesthetized guinea pigs. Overall, these comparisons show that mean latency jitter in LFPs in the big brown bat ventral CN is lower than those reported for onset neurons in the CN of these other mammals.

Cochlear Nucleus Organization

Data in this study were collected as part of a larger experiment in which LFPs were also recorded from the IC (32), with the electrode then lowered to the area of the CN and the experiment repeated. The angle of electrode insertion at the surface of the IC limited the region of the CN we sampled. This, along with the few CN sites we sampled (1 site in each of 5 bats), is a limitation of our work. All recovered lesion sites were found in the ventral CN (posterior portion of the anteroventral CN, anterior portion of the posteroventral CN, as labeled by Ref. 37), close to the level of the entry point of the eighth cranial nerve and medially near the area of the trapezoid body. It is not known if the area we sampled is somewhat special in terms of latency registration of FM signals, or whether other areas of the CN would exhibit similarly precise latency jitters. Because of the structural complexity of the CN, functional differences between its subdivisions are expected and have indeed been observed in both bats and other mammals (30, 38–40, 43, 44). Yet, Haplea et al. (30) reported no major differences in response properties, including latency or latency jitter, between single neurons in the anteroventral and posteroventral CN, even though neurons in those areas exhibited a variety of firing patterns. More research examining population latency registration in a wider area of the CN is required to determine the extent to which CN is homogeneous for latency registration.

Aside from the relatively larger size of the anteroventral CN, the structural and immunohistochemical organization of CN in FM bats is similar to that in other mammals (29, 43, 45–48). The CN area from which we recorded contains multipolar cells and shows high expression of the protein connexin-36, a marker for gap junctions and possible electrical transmission (45, 48). Archival data from our laboratory indicate that neurobiotin injections into the ventral CN label multipolar cells and fibers projecting towards the trapezoid body, and thus comprise part of the ascending auditory pathway (consistent with the findings of Ref. 49 based on transport of horseradish peroxidase).

Latency Jitter in the Inferior Colliculus

Latency jitter of LFPs from the IC in this experiment is remarkably similar to that we observed in the CN. Overall, the mean IC latency jitter in response to the broadcast is 52 μs and that in response to the first echo is 59 μs. One difference between the IC and the CN data is that latency jitter in the IC does not vary statistically with stimulus level and does not undergo amplitude-latency trading. This may reflect the relatively larger extent of the IC we sampled (recording depths from 300 to 2,000 um) compared to the more restricted area of the CN, and the heterogeneous response properties of individual neurons in different IC lamina that were sampled. It may also be a limitation of our sample size in both experiments.

The latency jitters we calculated from IC LFPs are smaller than those derived from responses of single neurons to either tones or single FM downsweeps (broadcasts). One study in awake big brown bats (30) reported a mean value of latency jitter (SD) to tones of 4.98 ms; 34% of neurons had SDs <1 ms, but only 2% had SDs <100 µs. For 15% of the neurons labeled as level tolerant, the average latency jitter was 450 µs. The results of that study suggested that latency jitter increased from the CN to the IC, a finding in opposition to ours. An experiment quantifying single neuron responses in awake Mexican free-tailed bats (24) reported latency SDs to tones ranging from 440 μs to 3.5 ms, but smaller values in response to FM downsweeps (250 μs or less). In awake big brown bats (26), latency jitters to FM downsweeps in the range of 50 μs to 6 ms were found, with 18% of neurons having SDs below 500 μs. In that study, latency jitters were not related to absolute latency, and small jitters were found even in neurons with on-response latencies as long as 25–30 ms. Another study in the IC of awake big brown bats (21) found that LFPs to FM downsweeps had greater latency stability than single neuron spikes, but that study did not provide explicit values of latency jitter. Sanderson and Simmons (22) recorded responses of single IC neurons to broadcast-echo pairs and found latency jitters of the first spike to the broadcast in the range of 0.122–5.51 ms, with a median value of 490 μs. Experiments based on intracellular recordings from the IC in awake bats (27) reported latency jitters of the onset hyperpolarizing potential in response to FM downsweeps ranging from 0.03 to 1.9 ms, with a mean value of 117 µs. The smallest values reported in all of these studies begin to approach, but are larger than, what we quantified in LFP responses from anesthetized bats.

Ferragamo et al. (26) acknowledged that the values of latency jitter from single-neuron responses are too high to explain the big brown bat’s psychophysical acuity and so argued for the examination of population-level rather than single-neuron responses as an explanation. Luo et al. (13) tested this hypothesis by recording population responses (extracellular field potentials) using silicon probes from 528 sites in the IC of awake big brown bats. Stimuli were single broadcast-echo pairs presented at a stimulus level of 70 dB SPL, close to our stimulus attenuation level of −30 dB. The mean latency jitter calculated in that experiment was 104 µs, and the lowest was 53 µs. These values are within the range of but somewhat larger than we found in our data. This difference likely reflects the larger sample size in (13), suggesting that the IC is not homogeneous in its temporal precision. We note that the filter cutoffs used in these two studies to isolate field potentials (200–600 Hz in Ref. 13; 143–500 Hz here) are very similar, but the type of filter employed differed (elliptic filter from Wave_Crus algorithms in Ref. 13; second-order Butterworth filter here). It is not clear how the type of filter may have affected the calculations of latency variability. Overall, findings from these two studies confirm the low latency jitter in the big brown bat’s IC as obtained from a summed population rather than single neuron responses.

Similar to the comparisons of CN data across species, latency jitter of IC neurons in other mammals is overall larger, indicating less temporal precision, than seen in LFPs from the echolocating big brown bat. In anesthetized guinea pigs, latency jitters of multiple unit activity were 1 ms and higher, depending on stimulus type (short vs. long tones), stimulus level (threshold vs. suprathreshold), and location across IC lamina (Figs. 7 and 8 in Ref. 41). A study of single neuron responses to tones in the anesthetized mouse IC (42) reported variations in latency jitter across neurons with different firing patterns, with phasic neurons showing the smallest jitter values (130–170 µs; mean of 367 µs). The comparison with mice is particularly interesting, given that these animals hear ultrasound but do not echolocate.

Future Directions

LFP recordings pick up synaptic and action potentials aggregated from the synchronous activity of groups of neurons close to the recording electrode (28). However, it is not known how many neurons contribute to each LFP nor is it understood the extent to which LFPs reflect action potentials, soma spikes, or dendritic depolarization. Nonetheless, our results show that response latencies are more stable and less variable (i.e., less jitter from presentation to presentation) when they are compounded from synchronized neural activity across groups of neurons than from analog averaging of responses from individual neurons obtained one at a time. The importance of pooling of neural responses is not surprising because some numbers of individual neurons are likely involved in the ability of echolocating bats to perceive echo delay with such fine acuity, although we do not know the identity of the neurons.

On the other hand, the complexity of echo processing must involve more than just pooling responses together by averaging: Although echo delay is the same for each frequency in the FM downsweep of a broadcast, successive sweep frequencies occur at different times. This means that the process of combining responses must take into account their dispersion in time by compressing (i.e., “dechirping”) the FM downsweep to remove its time duration (50). We propose that the aggregation of responses to form perception of delay is based on precise latency registration, not simply on delay tuning. The larger question remains as to whether the neural code for delay involves processing operations that go beyond just combining responses to achieve behavior-like acuity; perhaps the time scale of response latency has been altered by interactions analogous to heterodyning, for example (11).

Our data are limited by the small number of recording sites we sampled, but they are important because they extend earlier experiments (13, 18, 19) based on the use of stimuli the bat experiences while echolocating, rather than on responses to single tones or isolated broadcast-echo pairs, and they provide the first evidence of acute time processing in the first synapse of the ascending auditory pathway. Replication of our study with more extensive sampling is critical for understanding the role of the CN in the internal sculpting of response properties and in influencing sound processing in other auditory areas. Moreover, to fully understand the mechanisms of precise latency registration in the central auditory system and its relation to perception, it is important to repeat this experiment in other brain areas receiving input from the CN. Previous studies of single neuron responses to tones in the ventral nucleus of the lateral lemniscus (30, 51) calculated a mean latency jitter of 140 μs, smaller than reported by the same experimenters in the CN and IC. In particular, 75% of neurons in the ventral lateral lemniscus had SDs <0.1 ms, compared to 3% in the CN and 2% in the IC. It is possible that the low values of latency jitter we recorded in IC LFPs represent the contribution of these low variability neurons in the ventral lateral lemniscus. Little is known about latency registration in the FM bat’s auditory cortex. Maciás et al. (52) found that latency jitter in the auditory cortex of anesthetized Mexican free-tailed bats varied with the repetition rate of echoes. They plotted SDs of <3 ms (Fig. 3 in Ref. 52), but the exact numbers were not reported.

The presence of comparable latency jitters in the CN and IC suggests that echo delay is not registered by neural response latencies at the first synapse in the lower auditory pathway and then sharpened (perhaps by excitatory/inhibitory interactions) at the level of the midbrain to approximate behaviorally measured accuracy. Our data show that latency registration is substantially unchanged at the CN and the IC, suggesting that timing accuracy is not enhanced at these successive stages of processing. The data do not imply, however, that sharpening does not occur at other levels of the auditory pathway. The tens-of-microsecond values of latency jitter that we calculated in LFPs from the ventral CN and the IC approach within an order of magnitude, but do not reach, the big brown bat’s psychophysical acuity for echo delay. Some other neural process, as yet undiscovered, must intervene to mediate perceptual acuity. A full understanding of the big brown bat’s temporal hyperacuity requires additional exploration into population response properties at all levels of the auditory system.

DATA AVAILABILITY

Source data are available through the Brown digital repository at https://repository.library.brown.edu/studio/item/bdr:uk2vd5zh/.

GRANTS

This study was supported by US Office of Naval Research Grants N00014-14-1-05880 (to J.A.S.), N00014-17-1-2736 (to J.A.S. and A.M.S.), and N00014-24-1-2665 (to J.A.S. and A.M.S.).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.M.S. and J.A.S. conceived and designed research; A.M.S. and M.W. performed experiments; A.M.S. and M.W. analyzed data; A.M.S. and J.A.S. interpreted results of experiments; A.M.S. and J.A.S. prepared figures; A.M.S. drafted manuscript; A.M.S. and J.A.S. edited and revised manuscript; A.M.S., M.W., and J.A.S. approved final version of manuscript.