Pitch selectivity in ferret auditory cortex

8 Our perception of pitch – the tonal quality of a sound at a fundamental frequency (F0) – is essential for musical 9 melody, vocal communication, and attending to a voice in a crowded room. Pitch is distinct from simple 10 frequency tuning. Although pitch perception can arise from either harmonic spectral structure or temporal 11 regularity, it remains unclear how individual neurons represent these distinct cues. We recorded spiking activity 12 in hundreds of ferret auditory cortical neurons while systematically varying harmonic and temporal pitch cues. 13 A subset of neurons represented F0 based on harmonic content, another subset on temporal periodicity, and a 14 third population exhibited invariant pitch tuning across both cue classes. These findings provide the first 15 evidence for specialized pitch neurons in non-primates, and demonstrate that mammalian auditory cortex 16 employs dual spectral–temporal mechanisms for extracting F0. 17 Significance statement 18 Pitch allows us to enjoy music, understand speech, and follow voices in noisy settings, yet how the brain 19 represents pitch remains a long-standing question. Pitch selective neurons have been identified previously only 20 in primates, limiting our ability to investigate the neural of pitch. Here, we show that ferret auditory cortex also 21 contains such neurons, which extract pitch from both harmonic structure and temporal periodicity. This 22 discovery reveals that the brain uses dual, complementary strategies for encoding pitch. 23 Author contributions 24 VT analyzed the data and wrote the manuscript. QG collected the data, analyzed the data, and contributed to the 25 manuscript. KMMW conceptualized the experiment, collected the data, edited the manuscript, and secured the 26 funding. 27

28 This work was funded by Biotechnology and Biological Sciences Research Council grants to KMMW 29 (BB/M010929/1, BB/X013103/1) and a University of Oxford Clarendon Scholarship to VT. We are grateful to 30 Dr Ben Willmore and Dr Aleksandar Ivanov for assisting with data collection, to Prof Randy Bruno for helpful 31 comments on a draft manuscript, and to Prof Andrew King for sharing experimental equipment for this study. 32

33 Auditory cortex, Neural coding, sensory integration, temporal, spectral, pitch, perception 34 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 2

35 Sounds in nature are typically composed of many frequencies, yet we often perceive a natural sound to have a 36 single tonal quality along a low-to-high scale, known as its pitch. Pitch perception supports recognition of 37 musical melody, prosody in speech, and the segregation of sound sources in a complex acoustic environment, 38 such as following a conversation in a noisy restuarant1–4. Despite its essential role in hearing, it remains unclear 39 how pitch is extracted from acoustics and represented in the brain5. 40 Many sensory qualities are not represented by a labeled line code in the periphery. For example, color is a 41 percept synthesized by the visual system that is not directly dictated by a single wavelength of light. Similarly, 42 the perceived pitch is not simply the range of frequencies spanned by a sound. In the spectral domain, when all 43 frequency components are integer multiples (i.e. harmonics) of a common fundamental frequency (F0), a pitch 44 is perceived at the F0. This is true even for “missing fundamental” sounds, which contain no physical energy at 45 F0 itself6. The frequency-tuned auditory nerve fibers of the cochlea can represent this spacing of harmonics as a 46 place code using labeled lines, but only for lower “resolved” harmonics7. In the temporal domain, harmonic 47 sounds are periodic, with a waveform repeating at 1/F0. Auditory nerve fibers tuned to higher, unresolved 48 harmonics represent F0 in their spike timing by phase-locking to the sound envelope8. Psychophysical studies in 49 humans and animals have shown that pitch can be perceived when sounds contain exclusively harmonic or 50 temporal cues9–12. Therefore, the auditory system may use dual processing to identify F0. 51 Neurological studies13–15, lesion experiments in cats16, and neuroimaging in human listeners13,17 point to a key 52 role of auditory cortex in pitch processing. Microelectrode studies across a range of mammalian species have 53 described auditory cortical neurons sensitive to F018–20, but only in marmosets have specialized “pitch neurons” 54 been identified that extract F0 invariantly across sound types, including missing fundamentals21. These neurons 55 were clustered at the low-frequency border of primary and secondary auditory cortex, raising the possibility of a 56 specialized pitch area in the brain. 57 To investigate whether invariant representations of pitch exist in non-primate auditory cortex, we recorded the 58 F0 tuning of individual neurons in ferret primary auditory cortex while presenting sounds that varied in their 59 harmonic structure, phase, and temporal regularity. We found a population of pitch selective responses in 60 auditory cortex, with some neurons integrating the labeled line-coded harmonic cues (“harmonicity neurons”) 61 and others tuned to periodicity (“temporal neurons”). A subset of “pitch” neurons encoded F0 invariantly across 62 both cue types, reflecting the unified perception of the pitch evoked by these different sounds. Importantly, the 63 pure tone frequency tuning of these pitch selective neurons was not predictive of their F0-tuning, suggesting 64 that cortical pitch representations necessarily integrate over multiple frequency components of a sound. These 65 findings demonstrate that pitch selective neurons are not unique to primates. 66 67

68 We examined the hypothesis that pitch is represented in individual primary auditory cortical neurons of ferrets. 69 We presented anaesthetized ferrets with a collection of 13 stimulus types that varied in harmonic components, 70 phase, and the presence of a pink noise masker. Each stimulus type was presented across 17 different 71 fundamental frequencies (1/4 octave increments across 250 – 4000 Hz). We recorded responses from 1266 well-72 isolated single neurons across 4 ferrets. This included 1077 neurons in A1 and 189 in AAF. We found that 872 73 (81%) of the single neurons we recorded were sound responsive (paired t-test, p<0.05). Of these sound-74 responsive neurons, 165 (19%) were sensitive to the F0 of periodic click trains (1-way ANOVA, p<0.05). 75 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 3 Because click trains contain both resolved and unresolved harmonics, we reasoned that neurons encoding F0 76 through either the harmonic or temporal envelope cues should be sensitive to the F0 of these click trains. 77 Therefore, only sound-responsive neurons that were also sensitive to the F0 of click trains were included in 78 subsequent analysis. We refer to this population as “F0-sensitive neurons”. 79 Harmonicity neurons 80 We first aimed to test whether a subset of single neurons in the auditory cortex may extract pitch exclusively 81 from resolved harmonics. We reasoned that such "harmonicity neurons” would derive a sound’s F0 by 82 integrating the place code pattern of resolved harmonics across input neurons that are each tuned to individual 83 harmonics of the sound. Neurons using such a strategy would be unable to extract pitch information from 84 unresolved harmonics. They should therefore show similar F0-tuning across any sounds that contain resolved 85 harmonics, and may be either unresponsive or untuned to the F0 of sounds containing only unresolved 86 harmonics (Fig. 1a). 87 Within the F0-sensitive neurons described above, we identified putative harmonicity neurons as those that were 88 F0-sensitive to sound that contained only resolved harmonic complexes (1-way ANOVA, p<0.05), but were not 89 F0-sensitive to tone complexes containing only higher, unresolved harmonics (1-way ANOVA, p0.05). This 90 criterion identified 56 putative harmonicity neurons (34% of 165 F0-sensitive neurons). The F0 tuning curves of 91 3 putative harmonicity neurons are shown in Fig. 1b. 92 Note that this initial analysis identified neurons with spike rates that were modulated by the F0 of both click 93 trains and resolved harmonic tones, but it does not require that the neuron show similar F0-tuning curves for 94 these 2 stimuli. A harmonicity neuron should give a reliable read-out of a sound’s F0 across different types of 95 sounds that contain resolved harmonics in order to support pitch perception. Therefore, we next tested if these 96 putative harmonicity neurons showed consistent F0-tuning across sounds that contain resolved harmonics, but 97 have different timbres. These example neurons maintained their tuning when we deleting all the energy at the 98 frequency component corresponding to F0 (missing F0 sounds; Fig. 1c). Similarly, when we introduced 99 temporal and spectral noise to the original click trains with 5% jitter in the timing of their individual clicks, 100 these neurons were still tuned to the fundamental (CT 5% jitter; Fig. 1c). This invariant F0 tuning is visualized 101 as vertical bands of high spike rates at similar F0s across all 4 resolved harmonic stimuli in Fig. 1c, although 102 there is some variation in the width and peak of F0-tuning. Note that these neurons were not F0-sensitive when 103 only temporal F0 cues were available (red lines in Fig. 1b), highlighting their dependence on resolved 104 harmonics. 105 We used a bootstrapping approach to test if the timbre-invariant F0 tuning visualized in Fig. 1c was statistically 106 significant. For each putative harmonicity neuron, we calculated the correlations between F0-tuning curves for 107 all possible pairs of the 4 resolved harmonic sounds. We created a null distribution of these average correlations 108 from F0-randomized versions of the same neuron’s responses. If the true F0-tuning curve correlation was larger 109 than the 95th percentile of the null correlation distribution, we concluded that the neuron showed significant 110 generalization of pitch tuning across resolved harmonic stimuli (Fig. 1d). Of the 56 putative harmonicity 111 neurons tested, 59% (n=33) showed significantly correlated F0-tuning across the resolved harmonic stimuli 112 (p<0.05), and these were classified as harmonicity neurons. 113 114 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 4 115 Figure 1: Harmonicity neurons derive F0 from resolved harmonics. a) Spectra (top row) and waveforms (bottom row) of the 5 different 116 sound types used to classify harmonicity neurons. Each of these examples has an F0 of 2000 Hz. b) Each of the 3 panels shows the trial-117 averaged spike rate response of an example harmonicity neuron, as a function of F0. Shaded areas show the standard error of t he mean 118 (SEM). Harmonicity neurons derive F0 from resolved harmonics and therefore are sensitive to the F0 of click trains (bla ck line) and 119 resolved harmonics (blue line), but not to unresolved harmonics (red line). c) The color scale in each plot indicates the normalized trial-120 averaged spiking responses to 4 types of stimuli that contain resolved harmonics. d) In each plot, the s olid black line indicates the 121 average correlation between F0 tuning curves across all possible pairs of stimuli shown in c. The histogram (open bars) shows a null 122 distribution of the average F0-tuning correlation calculated when the tuning curves in c were shuffled randomly across F0. The dashed 123 black line shows the 95th percentile of this null distribution. Results for the same 3 example harmonicity neurons are shown in b, c, and 124 d. 125 126 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 5 Temporal neurons 127 We next examined whether neurons in auditory cortex can extract F0 exclusively from unresolved harmonics. 128 We hypothesized that such “temporal neurons” would derive a sound’s F0 by integrating the spike timings of 129 inputs that are phase-locked to the sound’s temporal periodicity. Unlike harmonic neurons, they would show 130 similar F0-tuning across any sounds that contain unresolved harmonics, regardless of whether resolved 131 harmonics were also present (Fig. 2a). In principle, temporal periodicity could also be derived from the phase-132 locked activity of neurons responding to resolved harmonics, so temporal neurons may be either unresponsive, 133 untuned, or tuned to the F0 of sounds containing only resolved harmonics. 134 We assessed periodicity tuning within our original set of 165 F0-sensitive neurons by identifying neurons that 135 were sensitive to the F0 (1-way ANOVA, p<0.05) of click trains that contained either: (1) a combination of 136 resolved and unresolved harmonics (Fig 2a; click train), or only high, unresolved harmonics (Fig 2a; high 137 harmonics) . The F0 tuning curves of 3 neurons reaching these two criteria are shown in Fig. 2b. In total, 44 138 neurons (27% of 165 F0-sensitive neurons) showed this temporal-based F0 sensitivity. 139 As with harmonicity neurons, these putative temporal neurons should provide a reliable read-out of a sound’s 140 F0 across stimuli with different frequency spectra in order to support pitch perception. We therefore tested if 141 these putative temporal neurons showed consistent F0-tuning across sounds that contain unresolved harmonics 142 but have different frequency spectra (Fig. 2a). These comprised the click trains and high harmonic stimuli 143 already described, as well as missing fundamental sounds and click trains with 5% jitter. A visual inspection of 144 the spike rate responses of the 3 example putative temporal neurons showed a striking similarity in F0-tuning 145 across the 4 sound types (Fig. 2c). We applied the same bootstrapping procedure described above for 146 harmonicity neurons (Fig. 1d), and found that over 84% of these putative temporal neurons showed statistically 147 similar F0-tuning across the 4 stimuli (p<0.05; Fig. 2d). Based on their ability to encode the F0 of unresolved 148 harmonics across a wide variety of sound spectra, we classified these as temporal neurons (n=37). 149 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 6 150 Figure 2: Temporal neurons derive F0 from unresolved harmonics. a) Spectra (top row) and waveforms (bottom row) of 4 different 151 sound types used to classify temporal neurons (F0=2000 Hz). b) Each panel shows the trial-averaged spike rate response for an example 152 temporal neuron as a function of F0. Shaded area shows the SEM. Temporal neurons show F0 sensitivity for click trains (black line) 153 and unresolved harmonics (red line). c) The normalized trial-averaged responses to 4 types of stimuli that contain unresolved harmonics, 154 plotted as in Fig. 1c. d) Results of bootstrapped correlation test of F0-tuning similarity across stimuli, plotted as in Fig. 1d. The same 3 155 example temporal neurons are shown in b, c and d. 156 157 158 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 7 Temporal neurons are sensitive to phase manipulations 159 If temporal neurons primarily derive F0 from the sound’s periodicity, they should be sensitive to manipulations 160 of the sound’s temporal fine structure. In the stimuli considered thus far, all harmonics were presented in phase. 161 Here, we compare the responses of temporal neurons to 2 versions of the high harmonic stimuli in which the 162 temporal cues were disrupted through phase manipulations. In the alternating phase (“alt. phase”) condition, the 163 starting phase of consecutive harmonics alternated between sine and cosine phase. This creates a temporal 164 periodicity at F0 and a second, comparatively weaker periodicity at 2F0 (Fig. 3a, middle). We therefore 165 predicted that the tuning curves of temporal neurons might be shifted to a lower F0 in response to the 166 alternating phase stimuli compared to the same sound presented in phase. In the randomized phase condition 167 (“rand. phase"), the phase of each harmonic was individually randomized, in order to diminish the temporal 168 periodicity at F0 and create a flatter temporal envelope (Fig. 3a, right). We predicted that temporal neurons 169 would show poor F0-tuning in response to the randomized phase stimulus compared to the same harmonics 170 presented in phase. It is important to note that both these manipulations left the harmonic content of the stimuli 171 as revealed in the sounds’ spectra completely unchanged (Fig. 3a), so any changes in neuronal response would 172 be due entirely to temporal sensitivity. 173 We found that a subset of temporal neurons was sensitive to phase manipulations. An example of a phase-174 sensitive temporal neuron is shown in Fig. 3b. This neuron was tuned to a lower F0 for the alternating phase 175 stimulus (orange line) compared to the in-phase version of this sound (black line), reflecting the increased 176 periodicity of the alternating phase version. Furthermore, this neuron showed weaker responses to sounds with 177 randomized phase, even when they were presented at the preferred F0 (green line). Therefore, this example 178 neuron showed the phase sensitivity predicted for temporal neurons. Not all temporal neurons showed this level 179 of phase sensitivity. A second example neuron shown (Fig. 3c) is one of the most phase-invariant temporal 180 neurons in our dataset, showing similar F0-tuning across all three high harmonic stimuli. 181 Phase sensitivity was found to be a common feature of temporal neurons overall. Their F0 tuning was 182 consistently and significantly lower in response to alternating phase stimuli than to the same harmonics 183 presented in phase (one-tailed paired t-test; t(36)=6.1, p=2.9x10-7; Fig. 3d). These phase manipulations also 184 elicited weaker spiking responses compared to same harmonics presented in phase (1-way ANOVA, 185 F(2)=12.86, p=9.79x10-6; Fig. 3e). Temporal neurons produced weaker spike rate responses at their preferred F0 186 in response to alternating phase stimuli (Tukey-Kramer HSD test; p=0.0015) and random phase stimuli (Tukey-187 Kramer HSD test; p=1.00x10-5), compared to harmonics in phase. Therefore, the population of temporal 188 neurons is sensitive to fine temporal manipulations, even across sounds with identical spectral content. 189 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 8 190 Figure 3: Temporal neurons are sensitive to phase manipulations. a) Spectra (top) and temporal waveforms (bottom) of stimuli in the 3 191 phase conditions are shown. The left panels show a tone complex containing high, unresolved harmonics, with each harmonic component 192 presented in cosine phase (black). This stimulus has a strong temporal periodicity at F0. The middle panels show the same high harmonic 193 complex presented with “alternating phase” (orange). The right panels show the high harmonic stimulus, with each harmonic component 194 starting in randomized phase (green). b) F0-tuning of a phase -sensitive temporal neuron. The firing rate during each of the 3 high 195 harmonic stimuli is shown as a function of F0. c) F0-tuning of a phase -invariant neuron, plotted as in b. d) The center of mass of a 196 neuron’s F0-tuning curve was defined as the weighted average of the trial -averaged responses elicited across all F0s presented. The 197 scatterplot shows that most temporal neurons had a lower center of mass for alternating phase stimuli compared to the same st imuli 198 presented in phase (paired t -test; p<0.05). e) Each temporal neuron’s response to a given phase manipulation was taken as the average 199 of its response at the best F0 (+/ - 0.25 octaves) of the in phase F0 -tuning curve. This response was then normalized across all phase 200 manipulations. Neural response magnitude was smaller for the alternating and random phase stimuli than the in phase stimuli (1 -way 201 ANOVA with Tukey-Kramer HSD test; thin line: p<0.01, thick line: p<0.001). 202 203 Pitch neurons encode F0 using both harmonic and temporal cues 204 Having identified harmonicity neurons that derive F0 from resolved harmonics and temporal neurons that derive 205 F0 from the periodicity of unresolved harmonics, we next investigated whether there may exist more general 206 “pitch neurons” that can derive F0 from both these acoustic cues. To identify pitch neurons, we applied a 207 protocol similar to the one used for temporal and harmonicity neurons. First, we defined putative pitch neurons 208 as those that were F0-sensitive for click trains, resolved harmonics, and unresolved harmonics (Fig. 4a), 209 combining the inclusion criteria of harmonicity and temporal neurons. We found 32 neurons that passed this 210 criterion, and three examples are shown in Fig. 4b. To determine if these putative pitch neurons could 211 generalize their F0-tuning across a wide range of sounds, we used the same bootstrapping approach described 212 above for harmonicity and temporal neurons to test if F0-tuning was similar across 5 stimulus classes: click 213 trains, low harmonics, high harmonics, missing F0 sounds, and click trains with modest (5%) jitter (Fig. 4a). 214 While these 5 stimuli have widely different frequency spectra and temporal fine structure, they should all evoke 215 a percept of pitch in both humans and ferrets12. 216 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 9 Our bootstrapped correlation analysis showed that 31 of the 32 putative pitch neurons had statistically similar 217 F0-tuning across all 5 stimuli (p<0.05; Fig. 4d), even when we include sounds with non-overlapping frequency 218 spectral (namely, low and high harmonic stimuli). This evidence extends the existence of acoustically invariant 219 “pitch neurons” beyond marmosets21, revealing a conserved cortical mechanism for pitch coding. 220 221 Figure 4: Pitch neurons can derive F0 from resolved and unresolved harmonics. a) Spectra (top ro w) and waveforms (bottom row) of 5 222 different sound types used to classify pitch neurons (F0 = 2000 Hz). b) Trial-averaged spike rate responses as a function of F0 for three 223 example pitch neurons. Pitch neurons can derive F0 from resolved or unresolved harm onics and therefore have similar F0 -tuning to 224 click trains (black line), resolved harmonics (blue line), and unresolved harmonics (red line). c) Normalized trial-averaged responses to 225 5 types of stimuli that contain resolved or unresolved harmonics, or bot h, plotted as in Fig. 1c. d) Results of bootstrapped correlation 226 test of F0-tuning similarity across stimuli, plotted as in Fig. 1d. The same 3 example temporal neurons are shown in b, c and d. 227 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 10 Sensitivity to pitch salience 228 If harmonicity and temporal neurons contribute to ferrets’ percept of pitch, we might expect their response 229 magnitude to decrease as pitch salience decreases. In human listeners22,23, pitch salience decreases as the 230 temporal regularity of a click train is degraded by randomly “jittering” the timing of individual clicks. Pitch 231 neurons in the marmoset have been shown to be sensitive to this jittering of click trains21. We investigated if 232 pitch selective neurons in ferrets are modulated by pitch salience, by presenting click trains with 5 levels of 233 temporal jitter (0%, 5%, 10%, 20%, and 40% jitter; Fig. 5a). 234 An example jitter-sensitive harmonicity neuron is shown in Fig. 5b. Its spike rate at best F0 decreases as jitter 235 increases. For comparison, Fig. 5c shows a temporal neuron which is not sensitive to jitter; its tuning curve and 236 response magnitude are consistent across jitter levels. Therefore, some neurons classified as harmonic and 237 temporal neurons were sensitive to this manipulation of temporal regularity. Next, we examined if the 238 populations of harmonicity and temporal neurons were jitter-sensitive. The spike rate response of each neuron 239 was calculated for click trains presented within a half octave of best F0, and for each of the 5 levels of jitter. 240 Responses were normalized by the average response across all jitters presented to remove variance in overall 241 firing levels across neurons. We found that harmonicity (1-way ANOVA; F(4)=6.24, p=1x10-4; Fig. 5d), 242 temporal (F(4)=16.5, p=1.6x10-11; Fig. 5e), and pitch (F(4)=13.6, p=1.8x10-9; Fig. 5d) neurons all showed 243 weaker spiking responses as the level of jitter increased. Therefore, all three classes of pitch selective neurons 244 encode temporal regularity in a manner consistent with a neural representation of pitch salience. 245 246 Figure 5: Temporal and harmonicity neurons are sensitive to temporal regularity. a) Spectra (top row) and waveform (bottom row) of 247 click trains with increasing temporal jitter between clicks (left to right). Jittering the temporal regularity of the clicks in the click train 248 introduces noise in the harmonic spectrum (top row) and periodicity (bottom row), reducing the salience of the perceived pitch in human 249 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 11 listeners22,23. b) F0-tuning of an example harmonicity neuron in response to click train stimuli with 0, 5, 10, 20, or 40% temporal jitter. 250 The tuning curve is consistent for low percentages of jitter, but becomes flat for jitter above 20%. c) F0-tuning for an example temporal 251 neuron, plotted as in b. The F0 -tuning of this neuron is not sensitive to the jitter manipulation. d) For each harmonicity neuron, the 252 normalized response at best F0 (+/- 0.25 octaves) was calculated for each jitter level. The responses of neurons decrease d as jitter level 253 increased (1 -way ANOVA with Tukey -Kramer HSD test; thin line: p<0.05, thick line: p<0.001). e) Jitter sensitivity at best F0 for 254 temporal neurons, plotted as in d. f) Jitter sensitivity at best F0 for pitch neurons, plotted as in d. 255 256 F0-tuning of pitch selective neurons is not predicted by pure tone frequency tuning 257 The above analyses have shown that a subpopulation of auditory cortical neurons can derive the fundamental 258 frequency that we hear as pitch from resolved harmonics (harmonicity neurons), temporal periodicity (temporal 259 neurons), or both these types of acoustical cues (pitch neurons). While pure tones elicit a percept of pitch in 260 humans and ferrets24, they do not contain harmonics from which to derive harmonic relations, and they only 261 provide temporal periodicity cues in one frequency channel of the labelled line. Thus, they would not provide 262 strong inputs for the type of harmonic and temporal neurons we have proposed here, which integrate across 263 frequency channels. Previous studies have shown a poor correspondence between pure tone frequency tuning 264 and preferences for the F0 of artificial vowel sounds in auditory cortex25, and pure tones are also poor predictors 265 of cortical responses to complex sounds more generally26. Therefore, we hypothesize that pure tone responses 266 may be a poor predictor of F0-tuning for complex sounds in the harmonic, temporal, and pitch neurons 267 described here. 268 We tested this hypothesis by comparing the F0-tuning of harmonicity, temporal, and pitch neurons measured in 269 response to click trains (containing resolved harmonic and temporal periodicity cues), low harmonic complexes 270 (resolved harmonics only), high harmonic complexes (temporal periodicity without resolved harmonics), and 271 pure tones (energy at F0 only). Fig. 6a shows the tuning curves for these 4 stimuli for 3 example pitch neurons. 272 The first example neuron (green border) shows considerable overlap in its pure tone frequency preference 273 (double-peaked tuning for 1414 and 2828 Hz) and best F0 for complex sounds (F0 preference for one of these 274 two frequencies for each complex sound). However, the other two example neurons display consistent F0 275 preferences for the 3 complex sounds, but non-overlapping frequency tuning derived from responses to pure 276 tones. 277 Overall, our harmonicity, temporal, and pitch neurons showed F0-tuning to complex sounds that did not readily 278 correspond to their pure tone frequency tuning. The scatterplot in Fig. 6b shows the correlation of F0-tuning in 279 pitch neurons between the click trains and resolved harmonic stimuli (x-axis) and between the click trains and 280 unresolved harmonic stimuli (y-axis). The clustering of neurons in the top right quadrant indicates that their F0-281 tuning measured with broadband click trains was similar that that measured with both resolved and unresolved 282 harmonics (Fig. 6b; resolved harmonic correlations: µ=0.46, σ=0.26; unresolved harmonic correlations: µ=0.35, 283 σ=0.32). This structure was not observed for pure tone stimuli (Fig. 6c; resolved harmonic correlations: µ=0.18, 284 σ=0.31; unresolved harmonic correlations: µ=0.06, σ=0.28). Similarly, the F0-tuning of harmonic (µ=0.02, 285 σ=0.39) and temporal neurons (µ=0.15, σ=0.30) were not positively correlated with their pure tone frequency 286 tuning (Fig. 6b, marginal histograms). These results demonstrate that complex pitch tuning in these neurons is 287 distinct from the mechanisms used to encode a single pure tone frequency. 288 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 12 289 Figure 5: Pitch neurons’ F0-tuning for complex sounds does not correspond to pure tone frequency tuning. a) Pure tone frequency tuning 290 and F0 tuning curves are shown for three 3 pitch neurons. Normalized trial -averaged spike rates (lines) are shown with SEM (shade d 291 regions). The green, purple, and orange borders around these examples indicate their identity in b and c. b) Axes indicate the correlation 292 between F0 -tuning measured with click trains and low harmonics (x -axis) or high harmonics (y -axis). F0 -tuning to cl ick trains is 293 positively correlated with F0-tuning to low harmonics in nearly all harmonicity neurons (blue histogram), and it is positively correlated 294 with F0-tuning to high harmonics in nearly all temporal neurons (red histogram). F0 -tuning to click trains is positively correlated with 295 F0-tuning for both low and high harmonics in most pitch neurons (scatter plot; each dot is one pitch neuron). c) Results plotted as in b, 296 but examining correlations with pure tone frequency tuning, rather than click train F 0-tuning. The distribution of correlations between 297 frequency tuning and F0-tuning to low harmonics is centered around zero in harmonicity neurons (blue histogram). Similarly, temporal 298 neurons’ frequency tuning is not correlated with F0 -tuning to high harmo nics (red histogram). In pitch neurons, pure tone frequency 299 tuning does not correlate systematically with F0-tuning to low or high harmonics (scatter plot). 300 301 Cochlear distortion products cannot account for the observed F0-tuning 302 When a missing fundamental sound is presented to the ear, the active mechanism of outer hair cell amplification 303 causes displacement of the basilar membrane at the position corresponding to F027,28. Perceptually, 304 psychophysical studies have shown that cochlear distortion products can improve pitch detection in human 305 subjects29,30. Therefore, it is important to determine if cochlear distortion products contribute to the F0-tuning 306 we observed in cortical neurons in response to our missing fundamental stimuli, particularly our high harmonic 307 stimuli which were designed to only contain unresolved harmonics. Distortion products at F0 are 10-15 dB 308 quieter than the harmonic components in a missing fundamental sound29, so we presented our low and high 309 harmonic stimuli in the presence of pink noise (5 dB below the harmonic components) to mask potential 310 distortion products in the ferret cochlea. If the observed F0-tuning persists in these neurons even in the presence 311 of the pink noise masker, this tuning is not entirely produced by distortion products. 312 We found that pitch selective neurons retained their F0-tuning even in the presence of a pink noise masker. Fig. 313 7a,c show similar F0-tuning with and without masker for example harmonicity and temporal neurons. This 314 finding held for the population of harmonicity and temporal neurons. F0-tuning curves for masked and 315 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 13 unmasked resolved harmonics were more similar for harmonicity neurons than other neurons (two sample t-test; 316 t(839)=8.7, p=2.5x10-17; Fig. 7b). Similarly, the F0-tuning for unresolved harmonics presented with and without 317 a pink noise masker were significantly similar in temporal neurons (t(843)=9.8, p=2.2x10-21; Fig. 7d). 318 Furthermore, the preferred F0 for masked and unmasked versions of the sounds were significantly correlated in 319 harmonicity and temporal neurons (Pearson correlation; r=0.55, p=8.7x10-4 for harmonicity neurons; r=0.57, 320 p=2.0x10-4 for temporal neurons; Fig. 7e,f). These results confirm that the pitch selectivity observed here is not 321 due to cochlear distortion products. 322 323 Figure 7: Distortion products do not explain observed F0-tuning. a) Each panel shows data from one example harmonicity neuron. Trial-324 averaged spike rates are plotted as a function of F0 for low harmonics presented with (dashed line) or without (solid line) a pink noise 325 masker. b) Histograms show th at the correlation betw een F0 -tuning for masked and unmasked low harmonics within harmonicity 326 neurons (blue bars) was higher than in other sound-responsive neurons (open bars). c) Each plot shows F0 -tuning for masked (dashed 327 line) and unmasked (solid line) high harmonics for two example temporal neurons, plotted as in a. d) Plotted as in b, histograms show 328 the correlation between F0 -tuning for masked and unmasked high harmonics within temporal neurons (red bars) and other sound -329 responsive neurons (open bars). e) For each harmonicity neuron, the scatter plot shows the best F0 measured with unmasked and masked 330 low harmonics. Larger dots indicate two harmonicity neurons. f) Best F0 of temporal neurons for unmasked and masked high harmonics, 331 plotted as in e. 332 333 Anatomical locations of harmonic, temporal and pitch neurons 334 Given the evidence for an anatomically segregated region of pitch neurons in the marmoset auditory cortex21, 335 we next explored where harmonicity, temporal, and pitch neurons were located across the surface of ferret 336 auditory cortex. The auditory cortex of each animal was mapped based on anatomical landmarks (e.g. sulci), 337 latency of neural responses to tones and noise bursts, and best frequencies (as described previously)18. Using 338 this mapping, each neuron was categorized to low- or high-frequency regions of A1 or AAF. The proportion of 339 F0-sensitive neurons classified as harmonicity, temporal, or pitch neurons is mapped in each region is 340 summarized in Supplementary Table 1 and visualized in Supplementary Fig. 1. Only 1 F0-sensitive neuron was 341 located in low-frequency AAF, so we omitted this region from further analysis. 342 To assess whether the three populations of neurons were homogeneously distributed across primary auditory 343 cortex, we applied a chi-square test for homogeneity, with the anatomical regions as the categorical variable. 344 We found that harmonicity neurons were distributed similarly across the three cortical regions (low frequency 345 A1, high frequency A1, and high frequency AAF; χ2(2)=1.78, p=0.41; Fig. 8a). Temporal neurons (Fig. 8b) and 346 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 14 pitch neurons (Fig. 8c) were significantly more prevalent in high-frequency A1 and AAF, particularly high-347 frequency AAF (χ2(2)=25.6, p=2.7x10-6 for temporal neurons; χ2(2)=16.6, p=2.5x10-4 for pitch neurons). In 348 summary, we did not find strong evidence for a single pitch centre in auditory cortex, although the proportion of 349 temporal and pitch neurons varied between primary fields and across the tonotopic gradient. 350 351

352 The fundamental frequency (F0) of harmonic sounds can be computed from either harmonic spacing or 353 temporal periodicity31,32. Computational models have shown that a single spectrotemporal algorithm for F0 354 extraction is also plausible33,34, but human psychophysical studies are not consistent with such an algorithm, and 355 instead suggest that pitch arises from the parallel processing of harmonic and temporal cues9,35. Behavioral 356 studies have demonstrated that ferrets, like humans, can discriminate pitch shifts in complex sounds24,36, 357 although ferrets base their pitch judgments more heavily on temporal periodicity than resolved harmonic cues12. 358 Previous studies have shown that neurons throughout ferret auditory cortex can represent the F0 of artificial 359 vowels, and almost always in parallel with other sound features18,25,37, and that their activity correlates with 360 pitch judgments during behaviour38. However, it is not clear if neurons in ferret auditory cortex represent F0 361 more generally across stimuli, process harmonic and temporal cues to extract F0, nor if they include specialized 362 pitch neurons. 363 Here, we used high-density microelectrode recordings and a diverse stimulus set to identify three new neural 364 subpopulations in ferret primary auditory cortex: harmonicity neurons that encode the F0 of resolved 365 harmonics; temporal neurons tuned to the periodicity of F0 in unresolved harmonics; and pitch neurons that can 366 extract F0 from both cue types. These populations showed properties previously described in primate pitch 367 neurons, including sensitivity to phase manipulations and temporal regularity, responses to missing 368 fundamentals at F0, and invariance to cochlear distortion products21,39,40. These F0 computations are well 369 supported by findings of complex pitch discrimination in behaving ferrets12,24, and could allow a listener to 370 generalize pitch across a wide array of sounds, as humans do when recognizing a melody across different 371 instruments. 372 A model of how harmonicity, temporal and pitch neurons may extract F0 from the representations of complex 373 sounds in the cochlea is presented in Fig. 8. Harmonicity neurons (H) could integrate inputs across labelled line 374 representations of the resolved harmonic components of a sound. This mechanism has been previously 375 described as a “harmonic sieve”, with excitatory inputs from neurons tuned to individual harmonics of the 376 preferred F0 and inhibitory inputs from intermediate frequencies41. Temporal neurons (T) may instead involve 377 circuits that compute the periodicity of spike times in neurons tuned to unresolved harmonics. These inputs may 378 also include neurons phase-locked to resolved harmonic components, as F0 can in theory be calculated as the 379 autocorrelation of spike times across neurons tuned to different harmonics42. Pitch neurons (P) can provide 380 spectrally-invariant representations of F0, perhaps by combining inputs from harmonic and temporal neurons 381 tuned to a common F0. The auditory nerve representations of resolved harmonic and temporal periodicity 382 described in Fig. 8 have been demonstrated in the microelectrode studies of Delgutte and colleagues8. The 383 evidence for the proposed dual processing of pitch cues at the level of auditory cortex is summarized below. 384 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 15 385 Figure 8: Schematic of proposed dual extraction of fundamental frequency (F0) from resolved and unresolved harmonics. Top: The 386 spectrum of a harmonic tone complex is shown, with each frequency component in a different color. Middle: The frequency tunin g 387 curves of individual cochlear filters are shown along the unrolled basilar membrane, with color indicating the best frequency of each 388 auditory nerve fiber. Because the tuning bandwidth of auditory nerve fibers is constant with respect to logarithmic frequency , tuning 389 curves become exponentially broader with increasing linear frequency. Low numbered harmonics are termed “resolved” if only on e 390 harmonic falls within the tuning curve of a single auditory nerve fiber, and therefore F0 is represented with a place code as the pattern 391 of harmonic activation across the population of resolved fibers. In contrast, auditory nerve fibers tuned to higher harmonic s are likely 392 to respond to multiple harmonics in the sound, so these are “unresolved” in the place code. Because multipl e unresolved harmonics 393 excite the same auditory nerve fiber, the spike times of auditory nerve fibers can phase lock to the sum of unresolved harmon ics at F0 394 (purple traces show spike output over time). This produces an explicit temporal representation of F0 for resolved harmonics. Bottom: 395 Our data show that there are distinct neural populations in auditory cortex to extract these F0 cues: harmonicity neurons (H) compute 396 F0 from the place code of exclusively resolved harmonics; and temporal neurons (T) extr act F0 from the temporal code of unresolved 397 harmonics. These two neural populations may then converge on a third neural population, pitch neurons (P), which encode F0 398 irrespective of the types of acoustic cues present in the sound. 399 400 Dual pitch processing in auditory cortex 401 Our findings add to the evidence in non-human primates that cortical neurons can represent F0 using resolved 402 harmonic or unresolved temporal envelope cues10,40,43. In macaques, populations of neurons in low-frequency 403 A1 can represent resolved harmonics as a place code, while higher-frequency neurons time-lock to the temporal 404 envelopes of unresolved harmonics20,43. In marmosets, Wang and colleagues identified a “pitch area” on the 405 low-frequency border of A1 and R, where individual neurons were tuned to the F0 of both pure tones and 406 missing fundamentals21,39. 407 Within the marmoset pitch centre, a small subpopulation of pitch neurons were sensitivity to alternating phase 408 manipulations40. A separate subset of neurons with higher best F0s (>450 Hz) was insensitive to phase 409 manipulations, but sensitive to resolved harmonic cues40. Ferret auditory cortex shows a similar division of 410 labour. Ferret temporal neurons show sensitivity to phase manipulations, including a reduced response to both 411 alternating phase and random phase harmonics (Fig. 3). They show similar tuning across complex periodic 412 stimuli, even when only unresolved harmonics are presented (Fig. 2). Harmonicity neurons in our study 413 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 16 generalized their F0-tuning across complex stimuli, but only for sounds containing resolved harmonics (Fig. 1). 414 As reported for marmoset pitch-selective neurons21,39, our temporal and harmonicity neurons were sensitive to 415 the temporal regularity of click trains (Fig. 5), demonstrating a neural parallel to pitch salience in human 416 listeners22,23. Our results suggest that dual cortical processes for extracting F0 may be widespread across 417 mammals. 418 Pitch-selective neurons in auditory cortex 419 Our results demonstrate that pitch-selective neurons exist in ferrets. As reported in marmosets, some pitch-420 selective neurons were sensitive to the temporal regularity of jittered click trains, were unaffected by the 421 presence of a masking noise to control for cochlear distortion products, and were sensitive to the phase of 422 unresolved harmonic components21,39. Their rarity was also comparable: about 3.5% of sound-responsive 423 neurons met our pitch neuron criteria, closely matching the ~3.6% reported in marmoset auditory cortex (Fig. 2 424 of Bendor and Wang, 200521), although about one-third of neurons are pitch selective in the marmoset pitch 425 center21,39,40. To our knowledge, this is the first demonstration of pitch-selective neurons outside primates. 426 Two differences between marmoset and ferret pitch neurons are noteworthy. First, ferret pitch neurons were not 427 typically well driven by pure tones, while marmoset pitch neurons were tuned to similar F0s for pure tones and 428 missing fundamentals21,40. This may reflect our differences in classification criteria, but it also raises the 429 possibility that pure-tone frequency and complex F0 extraction are mediated by distinct mechanisms. A single 430 tone has limited harmonic structure, providing a weak input to our model of temporal and harmonicity neurons, 431 which both pool information across harmonics (Fig. 8). 432 A second difference is that temporal, harmonicity, and pitch neurons were found distributed throughout low and 433 high frequency primary auditory fields in ferrets, albeit in variable proportions. This contrasts with the local 434 pitch area of marmosets21, but aligns with more distributed pitch representations reported in auditory cortex of 435 macaques20,43,44 and humans45–48. Even human fMRI studies showing evidence of a pitch center disagree on its 436 location within auditory cortex44,49,50. Overall, the bulk of evidence, suggests that a distributed network of pitch-437 selective neurons may be the more general organizational principle. Anatomical clustering in marmosets may 438 represent a species-specific specialization, or could reflect the criterion for pitch neurons to generalize their F0 439 tuning to pure tones. 440

and future directions 441 These findings establish that ferret auditory cortex implements dual pitch extraction strategies and contains 442 neurons specialized for invariant pitch representation. This supports the idea that pitch perception relies on 443 conserved cortical computations across mammals, including non-primates, and opens the door to studies of 444 pitch in a wider range of species. For example, if mice were shown to experience pitch perception, a toolbox of 445 power genetic manipulations would be available to further dissect pitch processing circuits. 446 Future work should test the causal role of these neurons in behavior. Despite the wealth of evidence for 447 anatomically localized pitch neurons in the marmoset, the behavioural consequences of manipulating this area 448 have not been explored. Comparative studies should also clarify how differences in cortical organization relate 449 to species’ ecological or trained demands. For example, do species with cochlea that provide fewer resolved 450 harmonics (e.g. mice) naturally have more neurons tuned to temporal than harmonic pitch cues? Would training 451 on resolved harmonics result in a higher proportion of temporal pitch neurons? Many key questions about the 452 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 17 neural mechanisms of pitch perception remain, and the identification of ferret pitch selective neurons suggests 453 that these questions can be answered in a wide range of animal models. 454 455

456 Animals 457 Two female and two male adult ferrets (age: µ = 29.3 weeks, σ = 18.4 weeks; Mustela putorius furo; Marshall 458 BioResources, UK) were used in this study. The animal procedures in this study were approved by the local 459 ethical review committee of the University of Oxford and were performed under UK Home Office license. 460 Surgical Procedure 461 We performed terminal electrophysiological recordings on each ferret. The animals were put under general 462 anesthesia via an intramuscular injection of ketamine (Vetalar; 5 mg/kg) and medetomidine (Domitor; 0.02 463 mg/kg). Anesthesia was maintained for the duration of the surgery and electrophysiology with a continuous 464 intravenous infusion of ketamine and medetomidine in Hartmann’s solution with 3.5% glucose and 465 dexamethasone (0.5 mg/ml/hr) via intravenous cannula. The ferrets were intubated and artificially ventilated 466 with medical O2. Blood oxygenation, electrocardiogram, respiratory rate, and end-tidal CO2 were continuously 467 monitored throughout the recording session, and a body temperature of 36-38°C was maintained with a heating 468 pad and warm air (3M Bair Hugger). To prevent corneal desiccation, eye ointment (Maxitrol; Alcon, UK) was 469 applied throughout the procedure. Every 6 hours, the animals received atropine (Atrocare; 0.06 mg/kg i.m.), or 470 when bradycardia or arrhythmia was observed. 471 Animals were placed in a custom stereotaxic frame and secured in place with a mouthpiece and ear bars. The 472 scalp was then shaved, cleaned (ChloraPrep; 2% chlorhexidine gluconate), and injected with bupivacaine 473 (Marcain, <1 mg/kg s.c.). The scalp was then incised, and the temporal muscles retracted. Using dental cement 474 (SuperBond; C&B, UK) and a stainless steel bone screw (Veterinary Instrumentation, UK), a steel holding bar 475 was attached to the skull. A craniotomy (10 mm diameter) was carried out over the right auditory cortex, and 476 the exposed dura was removed. Anatomical location was confirmed from stereotaxic coordinates (11 mm 477 ventral to the midline and 8 mm anterior to Lambda) and visual identification of the ectosylvian gyrus. The 478 surface of the brain was covered with a solution containing 1.25% agarose in 0.9% NaCl. Regularly throughout 479 recording, silicone oil was applied to the craniotomy. 480 Immediately prior to recording, the ear bars were removed, and the animal with the frame was transferred to an 481 electrically isolated anechoic chamber. An Ag/AgCl reference wire was inserted between the skull and the dura 482 at the edge of the craniotomy. A microelectrode probe (Neuropixels Phase 3)51 was inserted through the entire 483 depth of auditory cortex, orthogonally to the brain surface. The cortical area probed by each penetration was 484 identified based on its location relative to the ectosylvian gyrus, the local field potential response latencies, and 485 the frequency response area (FRA) shapes of cortical units recorded. Across the 18 cortical penetrations that 486 yielded sound-responsive units, 10 were in low frequency A1, 3 in high frequency A1, 2 in low frequency AAF, 487 and 3 in high frequency AAF. After a complete presentation of the stimulus set, the probe was removed and 488 reinserted in a new location within auditory cortex. Data were acquired using the SpikeGLX software 489 (https://github.com/billkarsh/SpikeGLX)52 and custom MATLAB scripts (Mathworks) at a 30 kHz sampling 490 rate. 491 492 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 18 Sound presentation 493 Stimuli were presented binaurally via earphones (Panasonic RP-HV094E-K) driven by a System 3 RP2.1 494 multiprocessor and headphone amplifier (Tucker-David Technologies). The earphones were closed-field 495 calibrated using a GRAS amplifier and 1/8 inch condenser microphone placed at the end of a model ferret ear 496 canal. Inverse filters were designed to ensure that the devices produced flat (less than ±3 dB) outputs across 200 497 Hz – 30 kHz. During the experiment, the earphones were coupled with otoscope speculae which were inserted 498 into each ear canal and sealed in place with Otoform (Dreve Otoplastik). Sounds were presented at 48,828 Hz 499 sampling rate. 500 Sound stimuli 501 The main stimulus set consisted of 13 sound types: pure tones and 12 different complex sounds, each presented 502 over a range of 17 F0s (200-4000 Hz in 0.5 octave increments). There were three broad categories of acoustic 503 manipulations in our complex stimuli: temporal regularity, harmonic components, and phase. In the temporal 504 regularity category, we presented click trains that were composed of biphasic pulses at a rate of F0. The 5 505 different types of “click train” stimuli were composed with different levels of temporal jitter between clicks (0, 506 5, 10, 20, and 40%). In the 0% jitter condition, the clicks were perfectly periodic, occurring with a spacing of 507 1/F0. In the jittered condition, each click from a regular click train was randomly shifted in time. The maximal 508 amount of time shift was defined as a percentage of the inter-click interval (5, 10, 20, or 40%), centered on the 509 timing of that click in a regular click train21,22. Therefore, stimuli with an increased jitter are less temporally 510 regular, and these are perceived by humans22 and ferrets (unpublished data from our lab) to have a weaker pitch 511 salience. 512 The second category, harmonic component manipulations, comprised harmonic sounds designed to include 513 either: 1) energy at all harmonics within our passband except at F0 (200 Hz – 30 kHz; “missing F0”); 2) only 514 harmonics that would be resolvable on the ferret cochlea (“low harm”); or 3) only harmonics that are 515 unresolved for ferrets (“high harm”). The resolved frequencies for each F0 were determined using a model of 516 ferret cochlear filters based on ferret auditory nerve recordings12,53. In addition, resolved and unresolved 517 harmonic stimuli were presented with and without a pink noise masker (“low mask” and “high mask”) to mask 518 cochlear distortion artefacts. The masker was presented at a sound level that was 5 dB below the harmonic 519 complex tone. In this condition, the noise masker would be 10 to 15 dB above the level of the expected 520 individual tonal components at F029. 521 For the phase manipulations, the third category, we presented unresolved harmonic stimuli with either 522 randomized phases across all harmonics (“rand. phase”), or with neighboring harmonics in alternating opposite 523 phases (sine and cosine phase, “alt. phase”). These stimuli were generated by adding sin waves at the frequency 524 of the desired harmonics for each combination of F0 and harmonic content. These stimuli did not include 525 energy at F0. All sounds were generated with custom Matlab scripts, and bandpass filtered from 200 Hz – 30 526 kHz. 527 Each sound was presented for either 200 ms (3 ferrets) or 300 ms (1 ferret) at 70 dB SPLA. Each unique 528 combination of sound type and F0 (e.g. resolved harmonics at 1000 Hz) was presented 13 times, in pseudo 529 random order across type and F0, with an interstimulus silent interval of 600 ms (3 ferrets) or 400 ms (1 ferret), 530 and linear onset and offset ramps of 25 ms. 531 In addition, a fuller set of pure tone stimuli was presented to assess frequency tuning in each penetration, so that 532 we could localize each of our microelectrode penetrations on the tonotopic map. Broadband gaussian noise 533 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 19 bursts of 100 ms were used as a search stimulus to identify sound-responsive neurons at the beginning of each 534 electrode penetration, followed by presentation of pure tones for frequency tuning classification, and then the 535 full set of pitch stimuli described above. Pure tones were presented at one of 15 different frequencies (500 Hz – 536 40 kHz in 0.45 octave increments) at 40, 50, 60, 70, or 80 dB. Tones lasted 100 ms followed by 500 ms of 537 silence. Each unique combination of tone frequency and intensity was presented 20 times to capture the 538 frequency response area of the electrode shank. FRAs were recorded for each penetration to identify the 539 tonotopic gradients across AAF and A1. 540 Spike sorting 541 The recorded signal was processed offline by first digitally high-pass filtering at 150 Hz. Common average 542 referencing was performed to remove noise across electrode channels54. Spiking activity was then detected and 543 clustered using Kilosort2 software (https://github.com/MouseLand/Kilosort2)55,56. Responses from single units 544 were manually curated using Phy (https://github.com/cortex-lab/phy)57. Units showing stereotypical spike 545 shapes with low variance, and a clear refractory period in their autocorrelation spike histogram were classified 546 as single units. Only well isolated single units were included in subsequent analyses. 547 Isolating F0-sensitive neurons 548 Each neuron was first assessed for sound responsivity using a paired-sample t-test (alpha = 0.05), which 549 compared spike rates during 150 ms of silence immediately prior to sound onset to rates during the 150 ms after 550 sound onset across all presentations of our sound stimuli. Only neurons found to be significantly responsive to 551 sound were included in subsequent analyses. We assessed neural sensitivity to F0s in response to each of the 13 552 sound types using a one-way ANOVA (alpha = 0.05), with spike rates during the first 100 ms after sound onset 553 as the dependent variable and F0 as the grouping variable. The best F0 for a given neuron was taken as the F0 554 eliciting the largest trial-averaged spike rate for a given stimulus type (e.g. click trains). 555 Assessing tuning similarity 556 We assessed the similarity of a neuron’s F0-tuning across pairs of sound types using a bootstrapped sampling 557 procedure. First, F0 tuning curves were constructed for a given neuron and for each stimulus type by averaging 558 the spike rate in the first 100 ms after sound onset across trials. Next, we calculated the correlation of F0-tuning 559 curves for every possible pair of sound types for that neuron. These pairwise correlations were averaged across 560 stimulus pairs to create a single metric of how consistent the neuron’s F0-tuning was across different sound 561 types. To test whether this average correlation of F0-tuning was greater than would be expected by chance, we 562 created a null correlation distribution from “shuffled” tuning curves, in which the neuron’s average responses 563 were randomly re-assigned across F0s. We repeated this procedure for 10,000 shuffled versions of each 564 neuron’s dataset. If the pairwise correlation of the unshuffled tuning curves was greater than the 95th percentile 565 of the null distribution, the neuron was considered to have consistent F0-tuning across the different sound types. 566 Declaration of generative AI and AI-assisted technologies in the manuscript preparation process 567 During the preparation of this work the authors used ChatGPT in order to make suggestions for reducing the 568

word count. After using this tool, the authors reviewed and edited the content as needed and take full 569 responsibility for the content of the published article. 570 571 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 20

572 1. Assmann, P. F. & Summerfield, Q. Perceptual segregation of concurrent vowels. The Journal of the 573 Acoustical Society of America 82, S120 (1987). 574 2. Cheveigné, A. de. Concurrent vowel identification. III. A neural model of harmonic interference 575 cancellation. The Journal of the Acoustical Society of America 101, 2857–2865 (1997). 576 3. Bird, J. & Darwin, C. J. Effects of a difference in fundamental frequency in separating two sentences. in 577 Psychophysical and physiological advances in hearing (eds Palmer, A., Rees, A., Summerfield, A. Q. & 578 Meddis, R.) 263–269 (Whurr, London, 1998). 579 4. McPherson, M. J., Grace, R. C. & McDermott, J. H. Harmonicity aids hearing in noise. Attention, 580 Perception, & Psychophysics 84, 1016–1042 (2022). 581 5. Wang, X. & Walker, K. M. M. Neural Mechanisms for the Abstraction and Use of Pitch Information in 582 Auditory Cortex. Journal of Neuroscience 32, 13339–13342 (2012). 583 6. Houtsma, A. J. M. & Goldstein, J. L. The Central Origin of the Pitch of Complex Tones: Evidence from 584 Musical Interval Recognition. The Journal of the Acoustical Society of America 51, 520–529 (1972). 585 7. Moore, B. C. J. & Gockel, H. E. Resolvability of components in complex tones and implications for 586 theories of pitch perception. Hearing Research 276, 88–97 (2011). 587 8. Cariani, P. A. & Delgutte, B. Neural correlates of the pitch of complex tones. I. Pitch and pitch salience. 588 Journal of Neurophysiology 76, 1698–1716 (1996). 589 9. Shackleton, T. M. & Carlyon, R. P. The role of resolved and unresolved harmonics in pitch perception and 590 frequency modulation discrimination. The Journal of the Acoustical Society of America 95, 3529–3540 591 (1994). 592 10. Song, X., Osmanski, M. S., Guo, Y. & Wang, X. Complex pitch perception mechanisms are shared by 593 humans and a New World monkey. Proceedings of the National Academy of Sciences 113, 781–786 (2016). 594 11. Sumner, C. J. et al. Mammalian behavior and physiology converge to confirm sharper cochlear tuning in 595 humans. Proceedings of the National Academy of Sciences 115, 11322–11326 (2018). 596 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 21 12. Walker, K. M., Gonzalez, R., Kang, J. Z., McDermott, J. H. & King, A. J. Across-species differences in 597 pitch perception are consistent with differences in cochlear filtering. eLife 8, e41626 (2019). 598 13. Zatorre, R. J. Pitch perception of complex tones and human temporal‐lobe function. The Journal of the 599 Acoustical Society of America 84, 566–572 (1988). 600 14. Tramo, M. J., Shah, G. D. & Braida, L. D. Functional Role of Auditory Cortex in Frequency Processing and 601 Pitch Perception. Journal of Neurophysiology 87, 122–139 (2002). 602 15. Stewart, L., Von Kriegstein, K., Warren, J. D. & Griffiths, T. D. Music and the brain: disorders of musical 603 listening. Brain 129, 2533–2553 (2006). 604 16. Whitfield, I. C. Auditory cortex and the pitch of complex tones. The Journal of the Acoustical Society of 605 America 67, 644–647 (1980). 606 17. Warrier, C. M. & Zatorre, R. J. Right temporal cortex is critical for utilization of melodic contextual cues in 607 a pitch constancy task. Brain 127, 1616–1625 (2004). 608 18. Bizley, J. K., Walker, K. M. M., Silverman, B. W., King, A. J. & Schnupp, J. W. H. Interdependent 609 Encoding of Pitch, Timbre, and Spatial Location in Auditory Cortex. Journal of Neuroscience 29, 2064–610 2075 (2009). 611 19. Griffiths, T. D. & Hall, D. A. Mapping Pitch Representation in Neural Ensembles with fMRI. Journal of 612 Neuroscience 32, 13343–13347 (2012). 613 20. Fishman, Y. I., Micheyl, C. & Steinschneider, M. Neural Representation of Harmonic Complex Tones in 614 Primary Auditory Cortex of the Awake Monkey. Journal of Neuroscience 33, 10312–10323 (2013). 615 21. Bendor, D. & Wang, X. The neuronal representation of pitch in primate auditory cortex. Nature 436, 1161–616 1165 (2005). 617 22. Pollack, I. Detection and Relative Discrimination of Auditory “Jitter”. The Journal of the Acoustical Society 618 of America 43, 308–315 (1968). 619 23. Yost, W. A., Mapes-Riordan, D., Shofner, W., Dye, R. & Sheft, S. Pitch strength of regular-interval click 620 trains with different length ‘runs’ of regular intervals. The Journal of the Acoustical Society of America 117, 621 3054–3068 (2005). 622 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 22 24. Walker, K. M. M., Schnupp, J. W. H., Hart-Schnupp, S. M. B., King, A. J. & Bizley, J. K. Pitch 623 discrimination by ferrets for simple and complex sounds. The Journal of the Acoustical Society of America 624 126, 1321–1335 (2009). 625 25. Bizley, J. K., Walker, K. M. M., King, A. J. & Schnupp, J. W. H. Neural Ensemble Codes for Stimulus 626 Periodicity in Auditory Cortex. Journal of Neuroscience 30, 5078–5091 (2010). 627 26. Laudanski, J., Edeline, J.-M. & Huetz, C. Differences between spectro-temporal receptive fields derived 628 from artificial and natural stimuli in the auditory cortex. PLoS One 7, e50539 (2012). 629 27. Verpy, E. et al. Stereocilin-deficient mice reveal the origin of cochlear waveform distortions. Nature 456, 630 255–258 (2008). 631 28. Bowling, T., Wen, H., Meenderink, S. W. F., Dong, W. & Meaud, J. Intracochlear distortion products are 632 broadly generated by outer hair cells but their contributions to otoacoustic emissions are spatially restricted. 633 Scientific Reports 11, 13651 (2021). 634 29. Pressnitzer, D. & Patterson, R. D. Distortion products and the perceived pitch of harmonic complex tones. 635 in Physiological and Psychophysical Bases of Auditory Function (eds Breebart, D. J., Houtsma, A. J. M., 636 Kohlrausch, A., Prijs, V. F. & Schoonoven, R.) 97–104 (Shaker, Maastricht, 2001). 637 30. Smalt, C. J., Krishnan, A., Bidelman, G. M., Ananthakrishnan, S. & Gandour, J. T. Distortion products and 638 their influence on representation of pitch-relevant information in the human brainstem for unresolved 639 harmonic complex tones. Hearing Research 292, 26–34 (2012). 640 31. Oxenham, A. J. Pitch Perception. Journal of Neuroscience 32, 13335–13338 (2012). 641 32. Oxenham, A. J. Questions and controversies surrounding the perception and neural coding of pitch. 642 Frontiers in Neuroscience 16, 1074752 (2023). 643 33. Meddis, R. & O’Mard, L. A unitary model of pitch perception. The Journal of the Acoustical Society of 644 America 102, 1811–1820 (1997). 645 34. Pitch: Neural Coding and Perception. vol. 24 (Springer New York, New York, NY, 2005). 646 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 23 35. Oxenham, A. J., Micheyl, C., Keebler, M. V., Loper, A. & Santurette, S. Pitch perception beyond the 647 traditional existence region of pitch. Proceedings of the National Academy of Sciences 108, 7629–7634 648 (2011). 649 36. Yin, P., Fritz, J. B. & Shamma, S. A. Do ferrets perceive relative pitch? The Journal of the Acoustical 650 Society of America 127, 1673–1680 (2010). 651 37. Walker, K. M. M., Bizley, J. K., King, A. J. & Schnupp, J. W. H. Multiplexed and robust representations of 652 sound features in auditory cortex. J Neurosci 31, 14565–14576 (2011). 653 38. Bizley, J. K., Walker, K. M. M., Nodal, F. R., King, A. J. & Schnupp, J. W. H. Auditory cortex represents 654 both pitch judgments and the corresponding acoustic cues. Current Biology 23, 620–625 (2013). 655 39. Bendor, D. & Wang, X. Neural Coding of Periodicity in Marmoset Auditory Cortex. Journal of 656 Neurophysiology 103, 1809–1822 (2010). 657 40. Bendor, D., Osmanski, M. S. & Wang, X. Dual-Pitch Processing Mechanisms in Primate Auditory Cortex. 658 Journal of Neuroscience 32, 16149–16161 (2012). 659 41. Goldstein, J. L. An optimum processor theory for the central formation of the pitch of complex tones. J 660 Acoust Soc Am 55, 1061–1069 (1974). 661 42. Licklider, J. C. R. A duplex thory of pitch perception. Experientia 7, 128–134 (1951). 662 43. Steinschneider, M., Reser, D. H., Fishman, Y. I., Schroeder, C. E. & Arezzo, J. C. Click train encoding in 663 primary auditory cortex of the awake monkey: Evidence for two mechanisms subserving pitch perception. 664 The Journal of the Acoustical Society of America 104, 2935–2955 (1998). 665 44. Norman-Haignere, S. V., Kanwisher, N., McDermott, J. H. & Conway, B. R. Divergence in the functional 666 organization of human and macaque auditory cortex revealed by fMRI responses to harmonic tones. Nature 667 Neuroscience 22, 1057–1060 (2019). 668 45. Griffiths, T. D. et al. Direct recordings of pitch responses from human auditory cortex. Current Biology 20, 669 1128–1132 (2010). 670 46. Allen, E. J., Burton, P. C., Olman, C. A. & Oxenham, A. J. Representations of Pitch and Timbre Variation 671 in Human Auditory Cortex. The Journal of Neuroscience 37, 1284–1293 (2017). 672 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint 24 47. Gander, P. E. et al. Direct electrophysiological mapping of human pitch-related processing in auditory 673 cortex. Neuroimage 202, 116076 (2019). 674 48. Berger, J. I. et al. Distribution of multiunit pitch responses recorded intracranially from human auditory 675 cortex. Cerebral Cortex 33, 9105–9116 (2023). 676 49. Patterson, R. D., Uppenkamp, S., Johnsrude, I. S. & Griffiths, T. D. The Processing of Temporal Pitch and 677 Melody Information in Auditory Cortex. Neuron 36, 767–776 (2002). 678 50. Plack, C. J., Barker, D. & Hall, D. A. Pitch coding and pitch processing in the human brain. Hearing 679 Research 307, 53–64 (2014). 680 51. Jun, J. J. et al. Fully integrated silicon probes for high-density recording of neural activity. Nature 551, 681 232–236 (2017). 682 52. Karsh, B. SpikeGLX. Janelia Research Campus (2017). 683 53. Sumner, C. J. & Palmer, A. R. Auditory nerve fibre responses in the ferret. European Journal of 684 Neuroscience 36, 2428–2439 (2012). 685 54. Ludwig, K. A. et al. Using a common average reference to improve cortical neuron recordings from 686 microelectrode arrays. Journal of Neurophysiology 101, 1679–1689 (2009). 687 55. Pachitariu, M., Steinmetz, N., Kadir, S., Carandini, M. & Kenneth D., H. Kilosort: realtime spike-sorting 688 for extracellular electrophysiology with hundreds of channels. bioRxiv (2016) doi:10.1101/061481. 689 56. Pachitariu, M., Sridhar, S., Pennington, J. & Stringer, C. Spike sorting with Kilosort4. Nature Methods 21, 690 914–921 (2024). 691 57. Rossant, C. et al. Spike sorting for large, dense electrode arrays. Nature Neuroscience 19, 634–641 (2016). 692 693 694 695 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 4, 2025. ; https://doi.org/10.1101/2025.11.03.686220doi: bioRxiv preprint