Survey of hippocampal responses to sound in naïve mice reveals widespread activation by broadband noise onsets

Abstract

Recent studies suggest some hippocampal (HC) neurons respond to passively presented sounds in naïve subjects, but the specificity and prevalence of these responses remain unclear. We used Neuropixels probes to record unit activity across layers in mid-ventral HC and auditory cortex (ACtx) of awake, untrained mice (male and female) while presenting diverse sounds at typical environmental levels (65–70 dB SPL). A subset of HC neurons exhibited reliable, short-latency responses to passive sounds, including tones and broadband noise. HC units showed evidence of tuning for tone frequency but not spectrotemporal features in continuous dynamic moving ripples. Across sound types, HC responses overwhelmingly occurred at stimulus onset; they quickly adapted to continuous sounds and did not respond at sound offset. Among all sounds tested, broadband noise was most effective at driving HC activity. Spectral manipulations indicated response prevalence scaled with increasing spectral bandwidth and density. Similar responses were also observed for visual flash and contrast modulated noise movies, although these were less common than for broadband noise. Sound-evoked face movements, quantified by total face motion energy (FME), correlated with population-level HC activity. However, many individual units responded regardless of FME strength, suggesting both auditory and motor-correlated inputs. Together, our results show that abrupt sound onsets are sufficient to activate many HC neurons in the absence of learning or behavioral engagement. This supports a role for HC in detecting salient environmental changes and supports the idea that auditory inputs contribute directly to HC function.

Introduction

Hippocampus (HC) has a well-established role in processing behaviorally important sounds. For instance, trace conditioning experiments demonstrate HC is critical for bridging the temporal gap between a tone and subsequent shock (Clark & Squire, 1998; McEchron et al., 1998; Moyer et al., 1990; Solomon et al., 1986). In tone discrimination learning, HC responses begin to distinguish frequencies predictive of shock or safety in parallel with behavioral discrimination (Freeman et al., 1996). Many other studies have observed sound-related activity in HC during auditory tasks, which is often absent or sharply diminished during passive playback (Aronov et al., 2017; Kumar et al., 2016). These studies suggest the primary role of HC in processing sound may be in learning contingencies and orchestrating behavioral responses.

Nevertheless, several recent findings indicate some HC neurons respond to passively presented sounds even in untrained subjects (Billig et al., 2022; Bimbard et al., 2023; Martorell et al., 2019; Xiao et al., 2018). Such responses are observed at sound levels well below startle and pain thresholds and typically have short latencies (~15–25 ms). They have been observed using a variety of sounds including broadband noise, tone pips, and natural sounds. One study noted units responsive to passively presented sounds were largely distinct from units that responded to the same sounds during task engagement (Aronov et al., 2017). Thus, auditory inputs are capable of evoking HC responses outside of learning and goal-directed behavioral contexts.

Few studies have attempted to identify specific acoustic features that drive HC responses in passive settings. Sound level is clearly important, considering HC noise thresholds are roughly 30 dB higher than auditory cortex (ACtx). One study found substantially more HC units were activated by broadband noise than pure tones (Xiao et al., 2018), consistent with a similar dissociation in upstream entorhinal cortex, medial septum, and pontine nuclei (G.-W. Zhang et al., 2018). Little else is known about the passive environmental sound features that evoke HC responses. It is also unknown whether and how these responses might complement processing in canonical auditory pathway stations. Thus, their prevalence and functional significance are unclear. The current study was motivated by two primary questions. First, which aspects of the auditory environment evoke HC responses in naïve subjects, and how prevalent are these responses? Second, how do responses in HC compare with canonical auditory pathway stations? We addressed these questions in a series of experiments examining a wide range of sound features using Neuropixels probes to simultaneously record from large unit samples in HC and ACtx.

Materials and Methods

Subjects and surgery

All procedures were approved by the Institutional Animal Care and Use Committee at the University of California, San Francisco. Subjects were adult male and female C57BL/6 mice aged ~1.5–3 months old (sample sizes and specific age ranges are provided for each experiment below). Most expressed optogenetic effectors targeting interneuron subpopulations, and some were wildtype littermates of 5XFAD transgenics (genotyping performed by Transnetyx). All mice were housed socially under a 12H-12H light-dark cycle. Surgeries were performed under isoflurane anesthesia. Once anesthetized, lidocaine was injected subcutaneously near the incision site for local analgesia, and meloxicam and buprenorphine were administered subcutaneously for systemic analgesia. A headbar was implanted above the right temporal lobe with dental cement. Additional postoperative doses of meloxicam and buprenorphine were administered intraperitoneally, and the subjects were allowed to recover at least two days. In a second procedure, a small craniotomy (~1–2 mm diameter) was made above ACtx (~2.5–4 mm posterior to bregma) and sealed with silicone elastomer (Kwik-Cast, World Precision Instruments). Meloxicam was provided for perioperative analgesia as before. Electrophysiological recordings were typically conducted within 1–3 days following craniotomy and occasionally the day of the craniotomy following 3+ hours recovery.

In vivo electrophysiology

All experiments were conducted inside a dark sound attenuation chamber (Industrial Acoustics Company). Subjects were headfixed on a spherical treadmill permitting free movement or rest (Figure 1A; Bigelow et al., 2022; Niell & Stryker, 2008). Extracellular recordings were conducted with Neuropixels probes lowered ~3500 μm below the brain surface (1 μm/s) in a trajectory spanning ACtx, HC, and thalamus (Figure 1B, left). Targeting these areas required probe entry at ~45° within a limited anterior-posterior range spanning approximately 2.8–3.6 mm caudal to bregma. The probe trajectory thus sampled units primarily within the CA1 and CA3 regions of mid-ventral HC. Some penetrations may have intersected CA2/3 instead of CA1/3. All Neuropixels channels within the recording span were active, permitting dense sampling of units across all layers of HC and ACtx. Prior to recording, the craniotomy was filled with 2% agarose and allowed to settle for 20+ minutes. Neuropixels data were recorded using OpenEphys or SpikeGLX. Spike sorting was performed offline using KiloSort 2.0 (Pachitariu et al., 2024). We used physiological features to estimate the boundaries of ACtx and HC offline, including depth-dependent transitions in local field potential power, multiunit and single firing rates, unit isolation density, and sound-evoked responses (Figure 1C). In some recordings, we applied Di-I to the probe for histological visualization of the probe trajectory (Figure 1B, right). Although we often recorded from putative auditory thalamus (units below HC with short latency, tuned responses to tones), we focused analysis on ACtx as recordings more consistently targeted this area and its boundaries were less ambiguous (e.g., unit isolation gaps reflecting pia and white matter). A minority of recordings included only ACtx or HC (e.g., anterior sites hitting ACtx but missing ventral HC). These were included in group data analyses since our primary aim was to define global differences between regions. Spike waveforms for example units are shown throughout as median ±MAD across all recorded spikes with 0.5 ms scale bars.

Figure 1.

Passive sounds evoke short latency responses in some hippocampal units of naive mice. (A) Sounds were presented passively to naive mice headfixed on a spherical treadmill. (B) Neuropixels probes enabled simultaneous recording in HC and ACtx. (C) Multiple physiological features were used for offline estimation of HC and ACtx boundaries including depth-dependent transitions in (i) tone-evoked multiunit activity, (ii) LFP power, (iii) multiunit event rates, (iv) unit isolation density, and (v) prevalence of tuned units. (D) Sound-evoked response latencies in HC and Actx units. (i) Onset latency calculations for example HC units with excited (top) and suppressed (bottom) responses to sound. (ii) Percentages of HC units with significant onset responses (left), with response latencies for tones (middle) and noise (right). (iii) Percentages of onset-responsive units and latencies for ACtx units as in (iii).

Auditory and visual stimuli

All stimuli were generated and presented with MATLAB (Mathworks). Sounds were sampled at 192 kHz and presented with an OctaCapture soundcard (Roland) through an open-field speaker (MF1 or ES1, Tucker-Davis Technologies) positioned ~20 cm from the left (contralateral) ear. Except where otherwise indicated, sounds were gated with 5 ms cosine² ramps and presented at 65–70 dB SPL. Sound levels were calibrated using either a Brüel & Kjær Model 4939 microphone (Model 2209 meter), or a combination of Avisoft-Bioacoustics CM16/CMPA and Fifine K669B microphones for frequencies above and below ~5 kHz, respectively.

Sound types tested in the experiments below included the following: Tones: sinusoid functions of constant frequency (range: 4–64 kHz); Broadband noise: white noise filtered with a passband of either 2–64 or 4–64 kHz; Dynamic moving ripple (DMR): a noise signal continuously modulated by an envelope defined by a smooth, random walk through spectrotemporal modulation space (spectral modulation range: 0–2 cycles/octave; temporal modulation range: −20–20 cycles/s; modulation depth: 40 dB; Escabí et al., 2003). The noise signal contains uniform octave band energy, created by summing multiple tone carriers (64/octave), each with randomized phase and frequencies spaced evenly on an octave scale between 2–64 or 4–64 kHz. Random double sweep (RDS): two uncorrelated frequency-modulated sweeps, i.e., tones of continuously variable frequency across time (frequency range: 4–64 kHz; sweep rate range: −50–50 octaves/s; Gourévitch et al., 2015); Clicks: 0.1-ms positive square wave pulses; Vocalization: a vocalization produced by a male mouse in an isolated cage after exposure to female urine, recorded with an Avisoft-Bioacoustics CM16/CMPA ultrasonic microphone (acquired at 250 kHz then down-sampled to 192 kHz). The vocalization contained energy broadly distributed between ~5–96 kHz including frequency-modulated components (Figure S1); Bandpass noise: to avoid filter artifacts at narrow bandwidths, these sounds were created by summing many tones (2¹⁵) with randomized phases and frequencies equally spaced on an octave scale within a range specified by a given center frequency and bandwidth (center frequency range: 4–32 kHz; bandwidth range: 1/8–2 octaves); Chords: harmonic tone complexes created by summing tones (1–256 carriers/octave) with randomized phases and frequencies equally spaced on an octave scale between 2–64 kHz.

In one experiment (Figure 4), we presented a battery of diverse sound types with the following parameters: Tones: frequencies 4–64 kHz (1/3 octave steps), levels 20–80 dB (15 dB steps), 100 ms duration; DMR: frequency range 4–64 kHz, spectral modulations 0–2 cycles/octave, temporal modulations −20–20 cycles/s, 1 s duration (repeating segments) and 5 s duration (non-repeating segments); RDS: frequency range 4–64 kHz, sweep rates −50–50 octaves/s, 1 s duration (repeating segments) and 5 s duration (non-repeating segments); RDS + noise: the same 1 s repeating segment presented with 1.6 s broadband noise at 55, 65, 75 dB (white noise filtered with a passband spanning 1/2 octave above and below the 4–64 kHz RDS frequency range); Noise train: four-pulse broadband noise train presented at levels spanning 20–80 dB (15 dB steps; white noise filtered with a 4–64 kHz passband; 150 ms pulse and inter-pulse interval); Click trains: 0.1 ms square wave pulses presented at rates 2–32 Hz (octave spacing); Vocalization: natural mouse vocalization presented as it was recorded (natural) and after randomizing the samples in time (scrambled), effectively creating a noisy comparison sound. The same vocalization was presented in all recordings. All sounds were interleaved in random order with each sound presented 50 times (300 ms inter-trial intervals) except tone pips, which were repeated 12 times per frequency-level combination. All sounds were presented at 65 dB SPL other than the specified tone and noise level manipulations.

Figure 4.

Hippocampal units respond best to broadband noise among diverse sound types. (A) Diverse sound types interleaved in pseudorandom order within the same experiment. (B) Example HC unit showing responses to RDS + noise, noise train, repeated DMR, and 80 dB tones. (C) Example ACtx unit responding to all tested sound types. (D) Percentages of units with responses to one or more of the tested sounds. (E) HC units primarily responded to sounds that included broadband noise, especially the noise train and RDS + noise stimuli. (F) ACtx units responded to all sound types.

Two experiments presented visual stimuli on a monitor positioned ~25 cm in front of the mouse. In one experiment, we presented full field visual flash trains (Morrill & Hasenstaub, 2018). In a second, we presented a contrast modulated noise (CMN) movie, which is designed to maximally activate visual pathway neurons by smoothly transitioning among variable spatiotemporal features over time (Niell & Stryker, 2008; Piscopo et al., 2013). The monitor was black during inter-trial intervals, such that transitions to flash or CMN reflected a highly salient and abrupt visual onset event. In these experiments, auditory and visual stimuli were presented alone or together using Psychophysics Toolbox Version 3 (Kleiner et al., 2007). Monitor luminance was calibrated to 25 cd/m² for 50% gray at eye position (SpectraScan PR-670). Auditory-visual onset synchrony was calibrated within ~1 ms using a diode to record monitor luminance values.

Treadmill and face movement recordings

We used custom software to log treadmill movements detected by optical USB mice positioned near the surface of the treadmill as in our previous work (Bigelow et al., 2022). In most experiments, we also recorded face movements using a Mako U-130B camera (Allied Vision) and varifocal lens (B&H Photo Video) illuminated by infrared LED and positioned to include nose, mouth, whisker, and eye regions (Figure 7A). Frame acquisitions were triggered by an Arduino UNO at either ~15 or ~30 frames/s. We then used FaceMap (Syeda et al., 2024) offline to calculate face motion energy (FME) across time. This includes converting videos into difference frames, reducing dimensionality across time with PCA, and projecting individual difference frames onto principal components for a time series reflecting relative FME in arbitrary units. Outlier values were common in both treadmill and face movement recordings; we thus limited each time series between the 99 and 0.1 percentiles for the recording. We then normalized these values as percentages of the maximum within each recording.

Figure 7.

Hippocampal responses to sound correlate with face movement. (A) Sounds evoke face motion responses on some trials. (i) Raw video frames are converted into face motion energy (FME) signals using FaceMap. (ii) Example single trial FME responses to a broadband noise train illustrating high trial-to-trial variability. (B) Similar stimulus features evoke HC activity and FME. Similar patterns of HC activity and FME responses were evoked by (i) bandpass noise, (ii) chords, and (iii) noise/flash train stimuli. (C) Sound responses in many HC units occur regardless of face movement. (i) Trials with high baseline FME and treadmill movement were excluded to isolate stimulus-evoked FME. (ii) Trials were classified as High or Low FME (cutoff: 2 SD change from baseline) then equated via subsampling prior to (iii) averaging spiking responses. Example unit responses showing diverse dependence on FME in both (iv) HC and (v) ACtx. (vi) Percentages of units with reliable sound responses on trials with High FME only, Low FME only, or both.

Analysis

In most cases, we used peristimulus-time histograms (PSTHs) to assess spiking responses to sound, with exceptions indicated below. Spiking responses in sensory areas and elsewhere have diverse temporal dynamics, featuring a wide range of latencies and durations in response to stimulus onset and/or offset, reflecting excitation and/or suppression relative to pre-stimulus firing. To accommodate this wide range of potential response patterns, we used a response reliability metric to determine whether stimulus-aligned firing patterns deviated significantly from chance for each unit (Bigelow et al., 2022). First, we calculated the correlation coefficient between two PSTHs constructed from random trial halves (without replacement). We then repeated the process 100–1000 times and defined response reliability as the mean across iterations. We further defined chance reliability in the same way after circularly shifting individual trials randomly in time, thus breaking the temporal relationship between stimulus and response without altering other statistics such as spike counts and inter-spike intervals. Defined in this way, units with trial-averaged responses that deviated in any way from random stimulus-aligned firing patterns – including onset transients, sustained excitation/suppression, and/or offset responses – produced higher data-half correlation values than the null condition. A p-value was then calculated by z-score transforming the observed reliability value using the mean and standard deviation of the chance (shifted) distribution and identifying its associated probability in the cumulative normal distribution function. We then adjusted the raw p-values for multiple comparisons using Benjamini–Hochberg False Discovery Rate (FDR) correction applied across all recorded units for a given experiment (Benjamini & Hochberg, 1995). Except where otherwise indicated, we used a conservative α = 0.001 to identify responses that clearly deviated from chance regardless of their specific response patterns and effect sizes. In many experiments, stimuli varied along one or more dimensions (e.g., tone frequency). In such cases, we calculated reliability separately for each stimulus value, as well as a single composite reliability value reflecting concatenated responses across all stimulus values.

We analyzed responses to tone frequency-level combinations (Figure 4) by constructing frequency response area (FRA) functions reflecting trial-averaged firing within 50 ms of tone onset. We then calculated the response reliability statistics above by subsampling random trial halves as above. We analyzed responses to continuous DMR sounds (Figures 2 and 4) using reverse correlation (Bigelow et al., 2022). For each unit, we calculated the spike-triggered average (STA) by summing windowed stimulus segments (time × frequency) around with each spike and dividing by spike count (Figure 2Bi). The STA is thus a linear estimate of the unit’s spectrotemporal receptive field (STRF). We used a temporal window spanning 200 ms before and 50 ms after each spike discretized in 5-ms bins, and a frequency window spanning the full stimulus range (2–64 kHz), discretized in 1/32 octave bands. Spikes from the first 200 ms of the stimulus were not included in STA calculations to reduce bias introduced by onset transients. We also calculated a null STA, reflecting time-frequency bins expected by chance, using the same procedure after reversing the stimulus in time. For display purposes, we transformed STA time-frequency bin values into z-scores relative to the null STA distribution, smoothed with a gaussian window (σ = 3 bins), and set values below |2.33| to zero to display only values less likely to occur by chance (p < 0.01). We then assessed the statistical significance of the STA using the response reliability metric described above. For experiments using continuous 30-min DMR, this was implemented by discretizing the full stimulus into contiguous 1-min trials.

Figure 2.

Hippocampal units show tuned responses to tone pips but not continuous dynamic moving ripples. (A) Some HC units show frequency tuning for tones. Example unit responses to tones in (i) HC and (ii) ACtx. (iii) Percentages of units with reliable responses to tones (left), and their corresponding best frequencies (top right) and bandwidths (bottom right). (B) HC units are not tuned for features in continuous DMR. (i) DMR stimulus and illustration of STRF calculated by averaging the stimulus segments preceding each spike for an example ACtx unit. (ii) Example STRFs from HCs unit lacking obvious structure (iii) Example ACtx STRFs. (iv) Percentages of units with reliable STRFs indicating HC units did not encode features in continuous DMR. (C) Some HC units responded to repeated a DMR segment. (i) Repeated DMR segment and example ACtx unit response. (ii) Example HC units showing obvious onset responses. (iii) Example ACtx units showing responses throughout the stimulus. (iv) Percentages of with reliable responses to the DMR segment.

Results

We used Neuropixels probes to simultaneously record HC and ACtx responses to a wide variety of sounds presented passively to awake, naïve mice. We cumulatively recorded from thousands of units in each region across nine separate experiments (Table 1), thus enabling detailed insights into the stimulus preferences and response characteristics of units in each area.

Table 1.

Unit sample sizes included in each experiment

Experiment	Figures	HC units	ACtx units
Tones	1,2, 8	2,128	2,263
Noise train	1,7	2,648	3,152
Continuous DMR	2	3,829	4,716
Continuous noise	3	883	1,347
Diverse sounds	2, 4	1,051	1,657
Bandpass noise	5, 8	1,463	1,940
Chords	5	1,463	1,940
Noise & flash train	6, 8	3,663	5,474
RDS & CMN	6	3,291	5,021

Passive sounds evoke short latency responses in some hippocampal units of naïve mice

We first replicated earlier findings that some HC units show short-latency responses to passive sounds, that HC responses are more prevalent for noise than tones, and that fewer units are responsive overall in HC than ACtx (Figure 1D). To do this, we constructed trial-averaged PSTHs for each unit (2-ms bins, 3rd-order Savitzky-Golay smoothing filter, 10 ms window) then examined for deviations from pre-stimulus firing (Figure 1Di). Units with activity above or below pre-stimulus firing (mean ±3 SD) for at least three consecutive bins within the first 150 ms were included in subsequent latency calculations (offset responses are considered in Figure 3). For tone experiments, PSTHs were averaged across 120 repetitions each of frequencies spanning 4–64 kHz in 0.2 octave steps (150 ms duration, 70 dB SPL). The experiments included 2,128 HC units and 2,263 ACtx units across 37 recordings from 17 mice (12 female; median age: 69 days, range: 40–97 days). Units with measurable onset latencies were much more prevalent in ACtx (1,493 of 2,263; 66.0%) than HC (124 of 2,128; 5.8%). Consistent with prior studies (Xiao et al., 2018), HC latencies were 23 ±13.4 ms (median ±MAD; Figure 1Dii, middle) across excited and suppressed responses, compared to 21 ±16.7 ms for ACtx units (Figure 1Diii, middle). Excited responses typically occurred earlier than suppressed responses in both areas (HC: 23 ±13.1 vs. 34 ±11.5 ms; ACtx: 19 ±15.4 vs. 29 ±22.9 ms).

Figure 3.

Hippocampal units respond primarily to sound onset. (A) HC responses adapt to 2 s broadband noise. (i) Individual HC unit responses (top) and means (bottom) separated into groups reflecting excitation and suppression relative to baseline firing. Population rates averaged within Early and Late windows (right) indicated activity returned near baseline levels by the end of the stimulus. (ii) ACtx unit responses organized as in (i). (B) Onset responses persist in pulse noise. (i) HC unit responses (top) and means (bottom) as in (Ai). Window averaged rates (right) indicated onset responses remained different from baseline through the final pulse. (ii) ACtx responses organized as in (i). (C) HC units do not respond to noise offset. (i) Example HC units showing responses at noise onset only. (ii) Example ACtx units with responses at noise onset and/or offset. (iii) Onsets and offsets were measured within 50-ms windows surrounding the peak and compared to an equivalent pre-stimulus window. Offset responses were also compared to a pre-offset window to ensure did not reflect sustained responses during the sound period (iv) Percentages of units with responses at noise onset only, offset only, or both, indicating a lack of offset responses in HC units.

We similarly calculated response latencies for broadband noise in separate recordings presenting 500 repetitions of a four-pulse noise train (Olsen & Hasenstaub, 2022; white noise filtered with a 4–64 kHz passband; 150 ms pulse and inter-pulse intervals; 65 dB SPL; 3 s inter-trial interval). The dataset included 2,648 HC units and 3,152 ACtx units from 31 recordings in 16 mice (6 female; median age: 74 days, range: 53–87 days). Units with measurable response latencies within the first 150 ms pulse were again more common in ACtx (1,666 of 3,152; 52.8%) than HC (938 of 2,648; 35.4%). Similar to tone experiments, median response latencies for noise were 25 ±15.2 ms in HC compared to 17 ±20.2 ms in ACtx, with excited responses typically occurring earlier than suppressed responses (HC: 21 ±18.1 vs. 29 ±10.9 ms; ACtx: 17 ±18.9 vs. 29 ±22.6 ms). Together, these experiments confirmed earlier findings that some HC units in naïve mice respond to passive tones and noise with short latencies (trailing ACtx by just a few milliseconds). We explored the finding that many more HC units responded to the noise train than tones in additional experiments below (Figures 4–5).

Figure 5.

Hippocampal preference for noise reflects spectral bandwidth and density. (A) HC units are highly sensitive to spectral bandwidth. (i) Stimuli comprised tones and bandpass noise with bandwidths ranging from 1/8 to 2 octaves. (ii) Example HC units with strong responses to 2 octave noise. (iii) Example ACtx units selective for tones (top) and narrowband noise (bottom). (iv) Percentages of units with reliable responses (left), and the bandwidths eliciting the best response (top right) and significant responses (bottom right). (B) HC units are sensitive to spectral density. (i) Chord stimuli comprising 1 to 256 tones/octave. (ii) Example HC units responsive to chords but not tones. (iii) Example ACtx units with strong responses to tones (left) and all tested sounds (right). (iv) Percentages of units with reliable responses (left), and the sound types eliciting the best response (top right) and significant responses (bottom right).

Hippocampus shows frequency tuning for tone pips but not continuous dynamic moving ripples

An obvious question regarding tone responses in HC is whether they are frequency tuned. If so, how do their tuning characteristics compare with ACtx? Might they be tuned for additional sound features that drive selective firing in ACtx units? We addressed these questions by first examining tone frequencies that evoked responses in each area. Of 2,128 HC units, 124 had significant composite response reliability (5.8%) compared to 1,695 of 2,263 in ACtx (74.9%). Across responsive units, reliability values were significantly lower in HC (median r = 0.25) than ACtx (median r = 0.52; p < 10⁻²², Wilcoxon rank sum test). Example units in Figure 2A show responses with clearly limited frequency ranges in each area. Best frequencies in HC units (the frequency with maximum response reliability) clustered almost exclusively between 8 and 32 kHz (Figure 2Aiii, top), corresponding to the most sensitive hearing range in mice. Best frequencies for ACtx units were similarly most common between 8 and 32 kHz but also covered the octaves above and below. Total responsive bandwidth (all tone frequencies with reliability p < 0.05) was 1 ±0.3 octaves in HC (median ±MAD) compared to 2 ±0.9 octaves in ACtx, a significant difference (p < 10⁻³³, Wilcoxon rank sum test). Thus, some HC units are tuned for tone frequency with more stereotypical and narrow frequency preferences than ACtx units.

Would auditory tuning observed for tone frequency generalize to other sound features? Auditory neurons show selectivity for many sound features including modulation frequencies in the spectral and temporal domains (Atencio & Schreiner, 2008). A standard approach for measuring these preferences is presenting continuous DMR and averaging across the stimulus segments preceding each spike (Atencio & Schreiner, 2013), i.e., the STA (Figure 2Bi). We calculated STAs for each HC and ACtx unit in a dataset of 73 recordings from 29 mice (17 female; median age: 79 days, range: 54–97 days). Consistent with previous studies, the majority of ACtx units had reliable STAs (3,159 of 4,716; 67.0%). In contrast, not a single HC unit out of 3,829 had a reliable STA (Figure 2Biv). Thus, HC tuning was sparse, stereotypical, and narrow for tone pips but non-existent for continuous DMR.

By design, continuous DMR lacked recurring silent interstimulus intervals and sound onsets present in the tone and noise train experiments. We speculated HC responses in those experiments may have depended on transitions from silence to sound. To test this idea, we analyzed responses to 50 repetitions of a 1-s DMR separated by 300 ms of silence, which were interleaved with other sound types in a separate experiment described further below (Figure 4). Across 26 recordings in 11 mice (9 female; median age: 71 days, range: 58–96 days), we found 71 of 1,051 HC units (6.8%) responded to this version of DMR (Figure 2Civ) – comparable to the percentage responsive to tones (Figure 2Aiii). As with tone responses, response reliability values in units that responded to the DMR segment were significantly lower in HC (median r = 0.46) than ACtx (median r = 0.54; p = 0.0012, Wilcoxon rank sum test). Together, these experiments suggested sound onsets may be critical for HC responses.

Hippocampus primarily responds to sound onset

We conducted three additional analyses aimed at clarifying whether the predominance of onset responses in HC might reflect adaptation to sounds with stationary stimulus statistics, and whether transitions from sound to silence (offsets) might be similarly effective in activating HC units as transitions from silence to sound (onsets). We first examined activity evoked by 2 s continuous broadband noise segments separated by 1.4–1.5 s silent intervals in an experiment presenting 0.5, 1, and 2 s segments 50 times each in random order (4–64 kHz bandpass filtered white noise; 65 dB SPL). This dataset included 883 HC and 1,347 ACtx units from 22 recordings in 5 mice (all male; median age: 83 days, range: 76–97 days). We further analyzed units with significant increases or decreases in mean firing during the stimulus window relative to pre-stimulus firing (p < 0.001, Wilcoxon signed rank tests), reflecting 240 units from HC (27.2%) and 842 from ACtx (62.5%). As seen in Figure 3Ai, HC units with both excited and suppressed responses showed obvious onset responses but typically returned to baseline firing before stimulus end. Wilcoxon signed rank tests confirmed activity in excited HC units differed significantly from baseline within the first 150 ms of the stimulus (“Early” window; p < 10⁻²⁰) but not the last 150 ms (“Late” window; p = 0.47). Similarly, suppressed HC units had large changes from baseline in the Early window (p < 10⁻²⁰), and only subtle but significant changes in the late window (p < 10⁻⁴). By contrast, responses in many ACtx units were sustained throughout the entire stimulus window, giving rise to highly significant group differences from baseline across response types and time windows (all p-values < 10⁻¹⁸). Thus, HC units respond robustly at noise onset but adapt quickly to continued noise.

Would HC units similarly adapt to a repetitive pattern of alternating noise and silence? We reanalyzed responses to the noise train stimuli used for latency calculations above described above using the same Early and Late firing windows described in Figure 3A. Onset transients diminished following the initial pulse but remained clear in HC population-averaged firing (Figure 3B). This was confirmed by significant Wilcoxon signed rank tests for each response type, firing window, and area (all p-values < 10⁻²⁹). Thus, even brief silence (150 ms) is sufficient for onset responses in HC.

Sound offset responses are common throughout the auditory pathway and believed important for processing temporal features in sound (Malone et al., 2015; Olsen & Hasenstaub, 2022). As expected, responses at noise offset were apparent in many of ACtx units in Figures 3A and 3B. We quantified such responses using continuous noise of three durations (0.5, 1, and 2 s). The dataset was the same used to analyze responses to 2-s continuous noise in Figure 3A. We analyzed only excitatory responses since offset suppression is difficult to disambiguate from post-stimulus recovery effects (Kopp-Scheinpflug et al., 2018). To accommodate potentially different onset and offset latencies, we first searched 100 ms following noise onset and offset in 1 ms steps for the 50 ms window that maximized firing. We then considered onset responses significant that differed from and equivalent baseline window immediately preceding the noise (p < 0.001; Wilcoxon signed rank tests). Offset responses were analyzed similarly, except were required to differ from two baselines, one before noise onset and the other before noise offset, to ensure elevated firing did not merely reflect continuation of sustained activity during the stimulus period (Figure 3Ciii, Bigelow et al., 2022; Olsen & Hasenstaub, 2022). Using these criteria, we observed offset responses in roughly one of five ACtx neurons, but no HC units (Figure 3Civ). Thus, the onset of a sound – but neither its continuation nor offset – is relevant to HC.

Hippocampus responds best to broadband noise among diverse sound types

Percentages of responsive HC units ranged widely among separate experiments testing different sound types above; 5.8% responded to tones (Figure 2A), 6.8% to repeated DMR segments (Figure 2C), and 27.2% to broadband noise segments (Figure 3A). We also measured onset responses in 35.4% of units using a broadband noise train repeated 500 times (Figure 1C). These differences may be partly influenced by experiment parameters such as stimulus repetitions and inter-trial intervals but also suggest differences due to acoustic features. Diverse search stimuli are often used when probing auditory pathway stations because individual neurons often respond robustly to one sound type or feature but not others. We thus examined whether HC units might show similar preferences among a battery of diverse sound types interleaved within the same experiment. The battery was designed to elicit responses in as many ACtx neurons as possible, and included the sounds depicted in Figure 4A. Responses to most sound types were analyzed using PSTH response reliability. Non-repeating DMR and RDS segments were analyzed using STA reliability. Tone responses were analyzed using FRA reliability as well as PSTH reliability using frequency-averaged responses at each level. The dataset included 1,051 HC units and 1,657 ACtx units from 26 recordings in 11 mice (9 female; median age: 71 days, range: 58–96 days).

As expected, the vast majority of ACtx units (1,494 of 1,657; 90.2%) had reliable responses to at least one sound in this battery (Figure 4D). Many HC units also responded to at least one sound (196 of 1,051; 18.7%). Parsing these responses by sound type revealed HC units primarily responded to sounds with broadband noise characteristics (Figure 4E); few to none responded to sounds other than noise train, RDS + noise, and repeated DMR. This pattern was not observed for ACtx units, which responded to all sounds and especially well to RDS, DMR, and loud noise trains (Figure 4F). Surprisingly, no HC units responded to the brief broadband signal generated by 0.1 ms click pulses, even when presented at a comparable repetition rate to the pulsed noise train (4 Hz). Click responses were robust in many ACtx units, suggesting differences between HC and auditory pathway stations in the way acoustic energy is integrated across time. Because loudness percepts correlate with sound duration (Clayton et al., 2024), this could be related to the higher sound level thresholds in reticular-limbic stations (Xiao et al., 2018).

Hippocampal responses to broadband noise reflect spectral bandwidth and density

Broadband noise was clearly the most effective stimulus for activating HC units in the experiments above. But which acoustic features explain its effectiveness? In other words, what makes noise “noisy”? We contrasted responses to noise and tones by manipulating two parameters: spectral bandwidth and density.

To assess spectral bandwidth, we presented tones and bandpass noise with bandwidths ranging from 1/8 to 2 octaves (octave spacing) and center frequencies spanning 4–32 kHz (0.2 octave spacing; 150 ms duration). Sound levels were equated for all stimuli at 65 dB SPL and each frequency-bandwidth combination was repeated 50 times with 300 ms inter-trial intervals. The dataset included 1,463 HC units and 1,940 ACtx units from 22 recordings in 5 mice (all male; median age: 83 days, range: 76–97 days). Many units in HC (279 of 1,463; 19.1%) and most in ACtx (1,687 of 1,940; 87.0%) had significant composite response reliability for this stimulus set (Figure 5Aiv). As with tone responses (Figure 2), response reliability was significantly lower in responsive HC units (median r = 0.10) than ACtx units (median r = 0.35; p < 10⁻⁶³, Wilcoxon rank sum test). Separating these responses by condition revealed HC responses were indeed strongly dependent on bandwidth, with nearly all units responding best to 2 octave stimuli (Figure 5Aiv, top; measured by composite reliability calculated across frequencies within each bandwidth condition). This distribution was significantly different from ACtx (p < 10⁻⁶⁶, Wilcoxon rank sum test), where many units preferred either tones or 2 octave noise, and some preferred intermediate bandwidths.

We assessed spectral density using chord analogs of the four-pulse noise train used for experiments above, comprising harmonic tone complexes constructed from 1 to 256 carriers/octave (Figure 5Bi). The chords thus had equivalent bandwidth but variable density. We also included broadband noise and tone trains (8 kHz) for comparison. We equated all sound levels at 65 dB SPL and repeated each stimulus 50 times with ~2 s inter-trial intervals. The dataset was collected together with bandpass noise experiments above. Similar to bandpass noise, this stimulus set evoked responses in many HC units (359 of 1,463; 24.5%) and most ACtx units (1,493 of 1,940; 77.0%), measured by composite response reliability (Figure 5Biv). Reliability values were again lower in responsive HC units (median r = 0.20) than ACtx units (median r = 0.39; p < 10⁻⁴⁷, Wilcoxon rank sum test). Population analysis indicated HC units responded best with 4 or more carriers/octave (Figure 5Biv, top), again reflecting preference for “noisier” sounds relative to ACtx units (p < 10⁻¹², Wilcoxon rank sum test).

Some hippocampal units respond to passive visual flash and contrast-modulated noise

The experiments above indicated most responses to passive sound in HC reflect the sudden occurrence of noise in an otherwise quiet environment. These outcomes suggested units with such responses may reflect a more general environmental change detection process. If so, they might respond to other environmental changes including visual events. We tested this possibility in two experiments.

We first examined whether HC units respond to full-field visual flashes, which, in the otherwise dim recording chamber, produced a sudden and highly salient environmental change. Flash trains were presented with the same temporal dynamics as the noise train stimuli described above (150-ms flash and inter-flash intervals, repeated four times). Noise and flash trains were presented alone or together with 50–200 repetitions of each condition randomly interleaved with 2–3 s inter-trial intervals. Noise trains were presented at 65 dB SPL as above. The dataset included 3,663 HC units and 5,474 ACtx units from 64 recordings in 35 mice (24 female; median age: 79 days, range: 58–96 days). Over one third of HC units in this sample responded to the noise train (Figure 6Aii, left; 1,469 of 3,663; 40.1%). Responses to flash train alone were far less common (117 of 3,663; 3.2%). Two observations suggested audiovisual convergence in many of the responsive units. First, more units responded to noiseflash trains presented together than either alone (1,785 of 3,663; 48.7%). Second, most flashresponsive HC units also responded to noise (Figure 6Aiii, right). Most ACtx units responded to noise (4,946 of 5,474; 90.4%), and the percentage of flash responsive units was slightly higher than HC (369 of 5,474; 6.7%). Reliability values for both noise and flash train were significantly lower for responsive units in HC than ACtx (noise train: HC median r = 0.49, ACtx median r = 0.73, p < 10⁻¹⁶⁵, Wilcoxon rank sum test; flash train: HC median r = 0.40, ACtx median r = 0.52, p < 10⁻⁹, Wilcoxon rank sum test). Thus, both noise and flash presented passively to naïve mice evoked HC responses, sometimes in the same units, supporting the notion that HC is sensitive to abrupt, salient environmental changes.

Figure 6.

Some hippocampal units respond to passive visual flash and contrast-modulated noise. (A) Responses to white noise and/or flash trains. (i) Example HC units responsive to noise only (top two units) or both noise and flash (bottom units). (ii) Example ACtx units as in (ii). (iii) Percentages of units with reliable responses to noise and/or flash trains indicating responses to noise are more prevalent in HC units (left) and most flash responsive units also respond to noise (right). (B) Responses to RDS or CMN. Example segments of the RDS (i) and CMN (ii) stimuli. (iii) Example HC units responsive to RDS only (top two units) or both RDS and CMN (bottom unit). (iv) Example ACtx units as in (iii). (v) Percentages of units with reliable responses to RDS or CMN indicating responses to RDS were more prevalent than CMN (left) and that only a few units responded to both (right).

Although flashes reflect a salient environmental change, they lack specific stimulus features such as edges and motion for which most neurons in visual cortex and other stations are tuned. Thus, it is possible fewer HC units responded to flash because it is less effective in driving upstream visual neurons than noise is for driving auditory neurons. To examine this possibility, we conducted a second experiment using auditory and visual stimuli each containing diverse features designed to activate as many neurons in their respective sensory pathways as possible. The visual stimulus was a CMN movie, with each frame defined by a smooth transition through a pseudorandom spatiotemporal feature subspace. Previous studies report this stimulus is highly effective in driving heterogeneously tuned neurons in mouse visual cortex and thalamus (Niell & Stryker, 2008; Piscopo et al., 2013). The auditory stimulus was RDS (presented at 65 dB SPL), which is conceptually related to CMN in that its frequency modulation vectors vary within a pseudorandom subspace across time, thus producing diverse spectrotemporal features that are highly effective in driving most auditory pathway neurons (Bigelow et al., 2022; Gourévitch et al., 2015). The RDS and CMN stimuli lasted 15 s (example segments shown in Figure 6Bi–ii) and were repeated 80 times in pseudorandom order with 2–4 s inter-trial intervals. This dataset included 3,291 HC units and 5,021 ACtx units from 60 recordings in 29 mice (25 female; median age: 79 days, range: 55–96 days). Representative HC units responded at RDS and/or CMN onset (Figure 6Biii). Most ACtx units showed activity peaks and troughs throughout the RDS, indicating sensitivity to specific spectotremporal features, with some units also showing CMN responses (Figure 6Biv). Population analyses summarized in Figure 6Bv indicated few HC units responded to CMN (39 of 3,291; 1.2%). Thus, the finding that fewer HC units responded to flash than noise was not well explained by the lack of spatiotemporal features preferred by visual pathway neurons. HC responses to RDS (207 of 3,291; 6.3%) were less prevalent than noise in the prior experiment, providing additional support for sensitivity for broadband noise in HC. In ACtx, nearly all units responded to RDS (4,907 of 5,021; 97.7%). The percentage of CMN responsive units was slightly higher than HC (205 of 5,021; 4.1%), and nearly all coincided with RDS responses. As with noise/flash train experiments, reliability values for RDS and CMN were significantly lower for responsive units in HC than ACtx (RDS: HC median r = 0.13, ACtx median r = 0.62, p < 10⁻¹⁰⁸, Wilcoxon rank sum test; CMN: HC median r = 0.13, ACtx median r = 0.17, p < 10⁻⁴, Wilcoxon rank sum test).

Hippocampal responses to sound correlate with face movement

Recent studies from independent groups have reported that sounds can evoke face movement responses, which may correlate with certain aspects of the sound. Two findings are especially relevant to our study. First, face movement evoked by natural sounds correlated strongly with population activity in HC (Bimbard et al., 2023). Second, face movement responses were stronger for broadband noise than tones or visual stimuli (Clayton et al., 2024; Olsen & Hasenstaub, 2025). We thus tested whether the prevalence of HC responses to noise vs. tones or visual flash in our study was related to stimulus evoked face movement.

We used FaceMap (Syeda et al., 2024) to transform our face video recordings into FME signals (Figure 7A). Consistent with previous studies, we observed clear increases in FME aligned to sound onsets in some trials (e.g., example trace with repeated noise train in Figure 7Aii). Trials with evoked face movements were observed to be interspersed among others without obvious FME responses to the same sounds. We also observed transient FME increases during some silent inter-trial intervals. These findings replicate recent findings suggesting sounds can elicit FME responses with a high degree of trial variability (Olsen & Hasenstaub, 2025).

To determine whether FME responses were related to acoustic parameters predicting HC activation, we calculated average peri-stimulus FME for each of the stimuli presented in the bandpass noise, chord, and visual flash experiments above. We then defined a global population activity index in HC reflecting the absolute value of z-scored activity relative to pre-stimulus baseline, averaged across all responsive units. As seen in Figure 7B, FME responses were very tightly correlated with population HC activity across experiments and stimulus features.

Considering identical sounds evoked highly variable FME responses across trials (Figure 7Aii), we wondered whether responses of individual HC units might be predicted by these outcomes. We therefore reanalyzed responses in the experiment described above comprising 500 noise train repetitions. We were primarily interested in the relationship between stimulus-evoked FME and HC activity and thus excluded trials with high spontaneous FME during inter-trial intervals. Treadmill recordings obtained in most experiments (27/31) indicated movements were typically associated with, but not necessary for elevated FME. We therefore excluded any trial for which mean treadmill activity or FME within a 1 s window prior to stimulus onset exceeded 5% of the maximum for the recording (Figure 7Ci). We then transformed FME values for each trial into z-scores relative to baseline (500 ms pre-stimulus window; Figure 7Ci). Trials for which mean z-scored FME was greater than two within the first 500 ms of stimulus onset were classified as high FME (Olsen & Hasenstaub, 2025). We then constructed PSTHs from high and low FME trials and calculated response reliability as in previous analyses (Figure 7Ciii). Because reliability and other spiking measures are affected by trial counts, we first equated trial counts by random subsampling from the condition with more trials. Experiments with at least 20 trials per condition were included in the final analysis, reducing the original sample to 2,041 HC units and 2,293 ACtx units from 22 recordings in 11 mice (5 female).

Example units in Figure 7Civ illustrate that FME correlated responses were highly variable among HC units. In some cases, firing was substantially elevated on high FME trials, especially following the initial pulse in the noise train (top row). Responses in other units showed no obvious relation to FME (bottom row). This variability is reprised by group data Figure 7Cvi, indicating some units responded reliably on high FME trials only (9.0%), others responded on high or low FME trials (7.8%), and only a few responded on low FME trials only (0.8%). Similar observations held for ACtx units, although a much larger proportion responded regardless of FME (>40%). These results confirm previous findings that population HC responses correlate with FME. Critically, however, they reveal substantial variability at the level of individual neurons and trials, and clarify that many individual units in HC respond to sounds regardless of FME.

Sound responsive hippocampal units are found in all layers

The previous sections showed that HC includes units that respond to passive sounds. Our Neuropixels probe trajectory sampled neurons across layers of the mid-ventral CA1/3 (possibly CA2/3 in some recordings; Figure 1B, Figure 8Ai), each of which features distinct cellular compositions, connectivity patterns, and functional properties. We thus examined whether sound responsive units were distributed throughout the recorded range in HC, or alternatively, whether they clustered within specific layers or regions. To do this, we first constructed bootstrap confidence intervals (99%, 1000 iterations) for the fractional depth distributions of all recorded units, regardless of sound responsivity (Figure 8Aii, left). The unit depth distributions were sharply bimodal, with peaks in unit isolation density indicative of CA1/3 pyramidal layers. We then compared the depth distributions of sound-responsive units against the confidence intervals to determine whether they followed the unit isolation density patterns that would be expected by chance if they were randomly distributed among recorded units (Figure 8Aii, middle). We finally calculated mean pre-stimulus firing rates for units within each depth bin to assess whether sound responsivity was related to differences in baseline firing among layers (Figure 8Aii, right).

Figure 8.

Sound responsive units are distributed throughout hippocampal layers. (A) HC unit depth calculations. (i) The Neuropixels probe trajectory sampled across HC layers. Orange and green arrows indicate pyramidal layers. (ii) Fractional depths of all recorded units (left; solid line: bootstrap median; dotted lines: 99% bootstrap confidence intervals), the subset of noise-responsive units (middle), and overlay (right). The magenta (cyan) arrows indicate depth bins in which responsive units were more (less) prevalent than expected by chance. (iii) Mean firing rates (±SEM) for units within each fractional depth bin. Data in ii-iii reflect noise train responses summarized in Figure 6A. (B) Responsive unit distributions by stimulus type: (i) Tone responses reflect the dataset summarized in Figure 2A. (ii) Bandpass noise responses reflect the dataset summarized in Figure 5A. (iii-iv) Noise and flash train responses reflect the dataset summarized in Figure 6A.

We included three datasets in this analysis: tones (Figure 2A), bandpass noise (Figure 5A), and noise/flash train (Figure 6A). As seen in Figure 8Bi–iii, sound-responsive units from all three experiments were found throughout HC layers. Across experiments, the proportion of sound responsive units fell below confidence interval bins corresponding to the pyramidal layers, and above confidence interval bins between these layers, reflecting stratum radiatum (SR) and lacunosum-moleculare (SLM) neurons. Units responsive to visual flash train followed a partially overlapping distribution (Figure 8Biv), although with fewer responsive units in the depth bins just below the putative CA1 pyramidal layer. The distributions of responsive units roughly paralleled the laminar differences in pre-stimulus firing rates; smaller percentages of putative pyramidal layer neurons were classified as sound-responsive, and neurons in these depth bins tended to have lower spontaneous firing rates. It follows that the non-random distributions of responsive units could at least partially reflect their higher firing rates and thus higher likelihood of reaching statistical significance (Bigelow et al., 2022; Olsen & Hasenstaub, 2022). Consistent with this possibility, we found that pre-stimulus firing rates were significantly correlated with response reliability values in all datasets (all correlation coefficients > 0.25, all p-values < 10⁻²²). An exception to this general trend was that bins outside pyramidal layers (putative stratum oriens; SO) had higher firing rates and fewer sound-responsive units (but more visual-responsive units). Thus, units responsive to sensory events are distributed throughout ventral HC, with only subtle deviations from the patterns expected by chance given the unit density and firing rate distributions.

Discussion

Summary of results.

Previous studies established that some HC units respond to sound outside learning and behavioral contexts (Billig et al., 2022). However, these studies left open questions about specific sound features capable of driving HC responses, the prevalence of such responses, and how they compare to responses of neurons in the canonical auditory pathway. Here we made significant progress toward answering these questions by testing a wide range of sound properties while recording from large unit samples in HC and ACtx. We found that roughly one in twenty HC units responded to tone pips (Figure 2). We replicated previous findings (Xiao et al., 2018) by showing that many more units responded to broadband noise (~30–40%; Figures 1, 3, 4, 6). Our results further clarified that preference for noise over tones depends on spectral bandwidth and density (Figure 5). We also noted that responses overwhelmingly occurred at stimulus onset (Figure 3), trailed responses in ACtx by just a few milliseconds (Figure 1), and could be either excitatory or suppressive (Figures 1, 3). Unlike ACtx, HC units did not reliably encode the spectrotemporal features of continuous DMR or RDS sounds (Figures 2, 4). They also did not respond continuously to ongoing broadband noise, or to noise offsets (Figure 3). Despite their preference for broadband noise onsets, HC units did not respond to the broadband but brief signal generated by clicks (Figure 4). In summary, the most effective acoustic event for evoking HC responses in naïve mice was a transition from silence to broadband noise.

Sources of auditory inputs to HC.

Given its position as a major hub at the interface of sensory, limbic, and modulatory circuits, acoustic information is accessible to the HC via many convergent channels. The canonical auditory pathway, which routes auditory signals to HC via projections from ACtx to the perirhinal and entorhinal cortices (PC, EC) may contribute; however, we note that the clear preference of HC units for broadband noise is consistent with the previously reported selectivity for noise over tones in reticular-limbic pathway stations including pontine nuclei, medial septum, and EC (Xiao et al., 2018; G.-W. Zhang et al., 2018). An additional pathway is suggested by the presence of frequency-tuned responses to passive sounds in the cholinergic basal forebrain (CBF; Zhu et al., 2023), which receives inputs from auditory thalamus and in turn sends dense projections to ventral HC (Gergues et al., 2020). Notably, CBF neurons have a narrow range of best frequencies, which partially overlap with the restricted range of BFs we observed in tone-responsive HC units (Figure 2).

Organization of sound responses within HC.

Given our probe trajectory and narrow anterior-posterior range, units in our study were predominantly sampled within the mid-ventral CA1/3 regions, across layers. Unit depth distributions were sharply bimodal, with peaks in unit isolation reflecting the densely-packed pyramidal layer neurons (Figure 8). Sound responsive units were observed across layers but were more prevalent than expected by chance between the pyramidal layers. However, the greater prevalence of sound-evoked responses in nonpyramidal layers may reflect these neurons’ characteristically higher spontaneous firing rates, which increase the likelihood of detecting significant stimulus-locked activity within the constraints of typical recording durations. Alternatively, it is possible that SR/SLM neurons may be differentially responsive to environmental sounds; indeed, neurons in these areas receive direct inputs from EC and are capable of strongly modulating activity in CA1 pyramidal cells (Dvorak-Carbone & Schuman, 1999), which comprise the primary outputs of HC. Thus, salient environmental sounds could modulate HC output via activation of a ventral EC-SR/SLM-CA1 pathway. This is consistent with previous work implicating ventral HC as a hub for integrating environmental events with internal state to guide behavioral responses (Turner et al., 2022). Considering the role of ventral (but not dorsal) HC in anxiety-related behaviors and approach/avoidance dynamics (Gergues et al., 2020; Turner et al., 2022), additional studies comparing sound responses in dorsal and ventral HC could clarify the behavioral relevance of abrupt environmental sounds in untrained animals.

Role of motor or modulatory inputs.

Recent studies have found that sounds – especially broadband noise – sometimes evoke face movement responses, and that either the movements themselves or the associated changes in neuromodulatory state can produce changes in firing rate that are easily mistaken for genuine sensory responses (Bimbard et al., 2023; Clayton et al., 2024; Olsen & Hasenstaub, 2025), including changes in HC population responses (Bimbard et al 2023). Might HC responses to sound also reflect motor-correlated inputs? Here we found that many HC units responded to sound even on trials without strong face movement (Figure 7), consistent with the possibility that such neurons are driven by auditory inputs. This possibility is also consistent with the short response latencies of many HC units (Figure 1). However, the presence of a significant fraction of units responding only on high face movement trials implies that sound-aligned responses across HC may reflect a complex volley of inputs from both auditory and motor-correlated sources.

Sound and visual responses in HC.

Two experiments showed that some HC units were also activated by passive visual events, suggesting responses to bottom-up sensory drive are not strictly limited to audition. Responses to auditory and visual events were not equally common in our study; while ~40% of HC units responded to broadband noise trains, only ~3% responded to visual flash trains with the same temporal dynamics (Figure 6). To test whether stimuli containing richer spatiotemporal features would drive more visual responses in the HC, we compared HC responses to contrast modulated noise (visual stimuli) and random double sweep (sound stimuli). Only ~1% of units responded to the CMN, while ~6% of units responded to the RDS. One interpretation of these results would be that sound features have preferential access to HC circuits and thus may have a disproportionate impact on the behaviors they support. Alternatively, it may be that fewer HC units responded to the visual stimuli tested in our experiments simply because these visual stimuli lacked the (unknown) features for which most HC units are receptive, in the same way that HC responses to otherwise identical sounds depended on specific spectral features (Figure 5). We also did not equate auditory and visual stimuli in terms of psychophysical detection threshold or other criteria. Thus, although flash and noise trains both reflected abrupt, highly salient environmental events, the sounds may nonetheless have been more perceptually salient to the mice. Finally, although we recorded from large HC unit samples in each experiment, they were obtained within a limited range of mid-ventral CA1/3. It is possible that other HC regions may not feature the same asymmetric distribution of auditory and visual responses. Distinguishing among these possibilities will require investigating a wider range of stimuli and HC regions. Additional experiments presenting vibrotactile or olfactory stimuli could further clarify the multisensory nature of passive sensory processing in HC.

Relevance for hearing health and disease.

The finding that sound responsive neurons are common in HC could be relevant to several aspects of hearing health and disease. Occupational and environmental noise hazards have well known adverse effects on auditory pathway structures, which are believed to contribute to hearing deficits such as presbycusis and hidden hearing loss (Gourévitch et al., 2014). This raises the question of whether HC neurons might similarly be affected by environmental noise exposure, and if so, whether these changes could affect general cognitive health. Indeed, both environmental noise exposure and hearing loss induction in rodents produce spatial memory impairments along with structural and functional changes in HC (Liu et al., 2016, 2018; Qian & Ricci, 2020, Zhang et al 2021), and hearing loss exacerbates the pathology associated with Alzheimer’s disease (Paciello et al., 2021). Hippocampal responsiveness to sound may thus represent an underappreciated pathway through which auditory dysfunction contributes to cognitive decline and dementia risk, but more work is needed to determine whether this pathway is biologically or clinically significant.

Significance statement.

Hippocampus is critical for learning and memory, but its role in sensory processing is less understood. Here, we show many hippocampal neurons in awake, untrained mice respond to passive sounds, especially broadband noise. Sound onsets – transitions from silence to sound – are critical for these responses, suggesting a role in detecting abrupt, salient environmental changes. Consistent with this possibility, some units also responded to visual events, though fewer than responded to noise. In contrast to auditory cortex, hippocampal units were not reliably tuned for spectrotemporal modulation features, suggesting independent functional roles. The prevalence of passive auditory processing in hippocampus builds on previous work suggesting hearing may interact with general cognitive health.