Bilateral widefield calcium imaging reveals circuit asymmetries and lateralized functional activation of the mouse auditory cortex

Many studies comparing ACx topography across animals show some degree of stereotypic organization; however, individual differences are also prominently observed (12, 13). To quantify individual differences in topographic organization, we created a quantitative method to compare topography and then utilized this method to investigate the similarity of topographic organization within an individual animal and across mice.

We captured 10 min of spontaneous activity as well as responses to tone stimuli on both cameras simultaneously (Fig. 1 A–C). We then calculated the correlation between each pixel in the left hemisphere and each pixel in the right hemisphere during spontaneous activity (Fig. 2A). We found high interhemispheric correlations (R > 0.95) in areas corresponding to secondary somatosensory cortices (S2) (24–26) and somatosensory parietal–ventral areas (PV) (27–30). We used these two landmarks to create a cartesian coordinate system and mapped the locations of peak responses to sound stimulation from both hemispheres onto this coordinate system. For each cortical area (n = 5 areas), we then measured the spatial distance (in μm) between the locations of peak response in the left and right hemispheres within mice (Fig. 2 A and B). Using this analysis, we found that the average distance between locations of peak response in the left and right hemispheres within an individual mouse across all frequencies and cortical areas, denoted as LvR, was 169 μm (SD: 104.9 μm) (Fig. 2 C and D). Thus, peak areas of activation in each hemisphere were highly similar.

The distance between locations of peak response across the hemispheres did not depend on sound frequency (Fig. 2E). Within ACx, most areas show similar hemispheric differences and variability [primary auditory cortex (A1): 149.5 ± 73.2 μm; A2: 189.6 ± 84.9 μm; anterior AAF: 158.4 ± 66.0 μm; DM: 128.8 ± 159.8 μm] (Fig. 2F). The ultrasonic field (UF) showed the highest difference across hemispheres (247.0 ± 159.8 μm) (Fig. 2F), possibly due to weaker fluorescence levels emitted from this area than all other cortical areas (Fig. 2F).

Next, we compared the location of peak responses across mice. We grouped the data points representing difference in location of peak response between the left hemispheres across mice (LvL*), between the right hemispheres across mice (RvR*), and between the left and right hemispheres across mice (LvR*) (Materials and Methods). The average difference in the location of peak responses across all mice was 306 ± 166.7 μm for the left and 295.5 ± 168.1 μm for the right hemisphere (Fig. 2D). Comparing left versus right hemispheres across mice results in a difference of 309.9 ± 179.2 μm. Our findings show that across mice, the locations of peak response only differ by ~300 μm and that relative positioning of vectors is similar across all frequency levels (Fig. 2 C and D). Thus, there is a stereotypical geometry of peak response across mice (Fig. 2 B and C). For all cases, the difference in peak location across mice is significantly higher than differences within individuals (L Across vs. LR Within: P = 6.26e-16, R Across vs. LR Within: P = 1.12e-13, LR Across vs. LR Within: P = 2.10e-13, two-sample t test, Bonferroni corrected). This finding highlights that while ACx organization is stereotyped, the topographic organization between hemispheres is more similar within mice than across mice.

Our results indicate that the mouse auditory cortex demonstrates stereotypical subfield organization across mice and that topography between the hemispheres within an individual mouse is symmetrical regardless of stimulus frequency or cortical region (Fig. 2G). Higher similarity within an individual suggests that individual sensory experience might impact cortical organization, while stereotypy across animals also suggests a hard-wired, stereotypical mechanism underlying organization.

PET and fMRI imaging studies in humans demonstrate that the left hemisphere has a larger area of activation in response to speech stimuli during both passive listening experiments (3–5, 31) and behavioral tasks (2). In mice, electrophysiological mapping also shows a larger overall active area in the left hemisphere than the right; however, these experiments did not compare hemispheres within individual mice and solely measured the area of responses to pure tone stimuli (20). C-Fos labeling studies found a higher number of labeled neurons in the left hemisphere in response to vocalizations, but the use of c-Fos labeling prevents the comparative analysis of pure tone and vocalization stimuli and precludes collection of multiple trials within an individual animal (19).

We thus set out to confirm whether the left hemisphere has a larger total area of activation by comparing the activity in both hemispheres in response to pure tone and adult vocalization stimuli within individual mice. We calculated the total area of activation—using 35% and 50% of the maximum image value as thresholds (Materials and Methods)—as well as the fractional size for each ACx area (SI Appendix, Figs. S3 and S5 C and F). While tones activated A1, AAF, DM, A2, and UF, vocalization did not strongly activate A1 (Fig. 3A).

For males and females grouped together, the activation area was similar between the left hemisphere for all three stimulus categories for both 50% and 35% thresholds (SI Appendix, Fig. S5 A and D). Separated by sex, males and females also had nearly identical responsive areas for ultrasonic tones and vocalizations (SI Appendix, Fig. S5 B and E).

Comparing population averages of the fractional size of each ACx showed that relative sizes were similar across hemispheres for low-frequency tones, high-frequency tones, and vocalizations (SI Appendix, Fig. S5 C and F). However, high variability in measurements was observed between mice. To minimize the effect of individual animal differences, we computed a Hemispheric Area Bias Index (HAB): (% total area_Left − % total area_Right)/(% total area_Left + % total area_Right) for A1, A2, and DM for each mouse (Fig. 3 B–D and SI Appendix, Fig. S4 B–D). When comparing the HAB, we found a left hemisphere bias in A2 to 4 to 64 kHz tones at both thresholds and 70 to 95 kHz tones at 35% threshold and a right hemisphere bias in DM in response to 4 to 64 kHz tones at both thresholds (Fig. 3D and SI Appendix, Fig. S4D). Together, these findings suggest that the hemispheres show differences in the way in which the activity is divided across ACx areas but that across-animal variability is high.

Our results indicate that the activated areas in ACx of each hemisphere differ. In addition to differences in the area of activation, the amplitude of neural responses could show hemispheric differences. Indeed, in the left hemisphere human ACx, e.g., the left planum temporale (PT), activation is higher than that of the right hemisphere (4). While behavioral studies suggest that there is a functional specialization of the left hemisphere for mouse vocalizations and pup calls (18, 21, 32), no direct comparison of the amplitude of activation between the hemispheres exists in mice. Given that each cortical area within the mouse ACx has a different peak response to tones and vocalization stimuli, with strongest responses to vocalizations outside of A1 (Fig. 3A), conflicting reports exist on which cortical area underlies vocalization processing, e.g., A2 for pup calls (18), DM of the left hemisphere of females for courtship calls (11), or DP or VPAF for general higher-order processing (13).

To resolve these issues, we calculated the peak response for each ACx area and stimulus by finding the maximum normalized fluorescence in each cortical area in response to pure tone and adult vocalization stimuli and compared this measurement across areas and hemispheres. We analyzed adult pain calls separately because these vocalizations contain a higher power of low frequencies. Adult pain calls are similar to pup wriggling calls but span a larger overall frequency spectrum, allowing us to not only identify hemispheric bias more independent of frequency bias but also uncover potential sex-specific biases (18, 23, 33, 34). First, we calculated the Hemispheric Response Bias (HRB: Response_Left − Response_Right)/(Response_Left + Response_Right) for each mouse individually (Fig. 4A). In A2, the HRB was positive for 9/10 mice, indicating that the left hemisphere was more strongly activated than the right hemisphere (Fig. 4A). Of note, 1/10 mice had a starkly opposite trend in response to 70 to 95 kHz tones and vocalizations at 50 and 65 dB (outlined in red) (Fig. 4A). Human studies show that in rare cases, language lateralization can be mirrored, especially in individuals who are left handed (5, 35, 36); therefore, perhaps a small percentage of mice also demonstrate a form of “handedness” in functional lateralization. We removed this mouse from the group averages for subsequent analysis.

We plotted the average peak responses across trials for each hemisphere (SI Appendix, Fig. S6 A and B). Each line connects the left and right hemisphere response within one mouse. We found that left A2 showed larger responses in response to 70 to 95 kHz tones (P = 0.040, n = 9, Wilcoxon rank-sum) and adult vocalizations (P = 0.045, n = 9, Wilcoxon rank-sum) at 50 dB. Computing the HRB based on sex showed that for adult vocalizations, the HRB is similar across both males (HRB: 0.21, SEM: 0.045) and females (HRB: 0.15, SEM: 0.050) at 50 dB and 65 dB (male ratio: 0.10, SEM: 0.050; female ratio: 0.12, SEM: 0.34) (Fig. 4A).

However, only females show significant left hemisphere response bias to adult pain calls at 65 dB (P = 0.007, n = 4, one-sample t test) with an HRB of 0.14 ± 0.021, which is similar to that of adult vocalizations. Female responses to vocalizations and adult pain calls only show a left hemisphere bias in A2 at 65 dB, whereas male responses show bias to vocalizations at 50 dB (Fig. 4A). Thus, A2 shows a sex-specific HRB for vocalizations.

Bias to vocalization could be due to a bias in frequency responses. We thus investigated how the response bias depends on tone frequency. We find that A2 shows a left hemisphere bias only for high-frequency tones (32 kHz to 82 kHz) (Fig. 4B). When we averaged peak activation in A2 across all mice for each vocalization, we found a left hemisphere response bias (P < 0.05, n = 9, paired-sample t test) in 8/9 vocalization stimuli (Fig. 4C). In addition, the left hemisphere A2 showed higher activation than all other cortical areas regardless of the hemisphere (SI Appendix, Fig. S6 A and B). None of the other cortical areas analyzed (A1, AAF, DM, and UF) showed a hemispheric difference to tones (at any frequency) nor vocalizations (SI Appendix, Fig. S6 A and B). Given that we observe a response bias in left A2 to pain calls containing low frequencies, this indicates that left A2 is more selective to vocalization stimuli not only due to higher responsiveness to frequencies within the vocalization range. Together, these results show that ACx and, in particular, A2 show functional lateralization to different stimuli in a sex-dependent manner.

So far, we have shown that while the functional areas are similar across hemispheres, there exists lateralization of the strength of functional activation. We next sought to understand the functional connectivity giving rise to this lateralization. Bilateral imaging allows us to calculate the cross-hemisphere correlations to derive functional connectivity maps. In particular, correlations in spontaneous activity (37) and noise correlations (38) are reflective of connectivity between neurons (8, 39, 40) both within or across hemispheres. Noise correlations are a measure of shared trial-to-trial response fluctuations between two neurons or two cortical areas (38, 41). Recorded at resting state, analyzing correlated fluctuations in spontaneous activity is an alternative way of inferring functional connectivity (39, 42). There are strengths to each method, so we used both to investigate the functional connectivity of ACx within and across hemispheres.

We first set out to investigate the pattern of interhemispheric correlations in spontaneous activity. We used 9-pure-tone stimuli to identify sound-responsive regions and identified ROIs (n = 10 mice). Subsequently, we recorded 10 min of spontaneous activity (12,000 frames) in silence without adjusting the head position and any imaging setting. We then collected all the ROIs identified from each tone for each animal and calculated the interhemispheric correlations in the spontaneous activity of the identified regions. In general, matching auditory fields (Fig. 5A) showed higher mean correlation than “nonmatching fields” (Fig. 5B). Moreover, in matched fields, left DM paired with right DM (DM-DM) showed a high correlation coefficient of 0.93, second to the UF-UF (norm. correlation coefficient = 0.99), and the A2-A2 had the lowest correlation coefficient [normalized (for each animal) correlation coefficient = 0.89] among the five pairings (all P < 0.05, n = 10, bootstrapping hypothesis test, boots = 100).

To investigate whether the correlations between nonmatching fields were similar across interhemispheric pairings, e.g., is the correlation between left DM and right AAF the same as between left AAF and right DM, each nonmatching field pair was placed side by side. This comparison showed that the correlation of left DM to right AAF was not significantly weaker than that of right DM and left AAF (Fig. 5B); however, when we compare the correlation difference within individual animals (Fig. 5C), e.g., $(CorrA{1}_{\mathit{left}}-CorrD{M}_{\mathit{right}})/(Corr{A1}_{\mathit{left}}+CorrD{M}_{\mathit{right}})$ , it shows a bias to the right in DM-AAF (P = 0.012, n = 9, Wilcoxon sign-rank) and a marginal bias in A2-AAF (P = 0.055, n = 9, Wilcoxon sign-rank). This indicates that while asymmetries exist, the individual variability is high. Individual mice showed varying degrees of asymmetry, but asymmetries in DM-AAF were consistently present (Fig. 5C). These results indicate that a number of auditory regions show asymmetries in their functional correlations and that most asymmetries include DM and AAF. Lumping ROIs from each subregion (e.g., A1) for each animal ignores the fact that there are tonotopic differences across subareas and that lumping might obscure more specific effects. We thus analyzed ROIs defined by responsiveness to different frequencies separately. Given the tuning properties of ACx neurons, some pixels could be activated by tones with nearby frequencies and thus contribute to multiple ROIs. To minimize this potential confound, we only used ROIs corresponding to 4 kHz, 16 kHz, and 64 kHz, which reduces the coactivation to <30% (SI Appendix, Fig. S7). By separating subareas, we still find that A2 shows the lowest interhemispheric correlations and that the interhemispheric bias exists in DM-AAF, A2-AAF, and A1-DM (Fig. 5 D and E).

We next used all the ROIs identified and paired across the brain. As a consequence, each animal has multiple ROIs for each field to pair with the other hemisphere (e.g., the low- and high-frequency areas of A1), and ROIs determined by close-by frequencies can have overlap. In this analysis, spontaneous activity correlations between nonmatched fields show significant asymmetries between A1-A2, A1-AAF, DM-A2, DM-AAF, and A2-AAF (SI Appendix, Fig. S8).

Sound stimuli increase overall activity levels in the cortex; therefore, noise correlation analysis might allow the detection of interactions between cells having low spontaneous activity rates and thus reveal additional asymmetries. We measured noise correlations during 1 s tone presentation. In general, noise correlations matched results from spontaneous activity with matching auditory fields showing higher correlations than “nonmatching fields” (SI Appendix, Fig. S9). Moreover, as with correlations determined from spontaneous activity, DM-DM, had the highest correlation, whereas A2-A2 had the lowest (SI Appendix, Fig. S9). The interhemispheric functional connectivity between nonmatched fields revealed by noise correlation was similar (SI Appendix, Fig. S9B), possibly due to contamination of the stimulus responses into noise correlations (43) or the higher sensitivity of our spontaneous activity recordings due to longer duration recording (10 min vs. 1 s, thus 12,000 vs. 20 frames). The noise correlation between A1 and DM (SI Appendix, Fig. S9B, DM-A1: 0.73, A1-DM: 0.72) and between DM and UF (SI Appendix, Fig. S9B, UF-DM: 0.77, DM-UF: 0.78) was equivalent or higher than the lowest value found in matched fields A2-A2 (SI Appendix, Fig. S9A, r = 0.72), which suggests that DM and UF might be parts of A1 instead of separate fields (11).

Together, our results show that the matched-field functional connectivity was higher than that of nonmatched fields and that A2-A2 shows the lowest interhemispheric functional connectivity. Our results also show that asymmetrical interhemispheric functional connectivity exists between nonmatching fields of the auditory cortex.

We showed that ACx fields are functionally connected across the hemispheres and that circuit asymmetries exist. Sound experience can shape functional circuits in ACx and tonotopic maps (44–47). We thus investigated whether the development of interhemispheric functional connectivity is influenced by sensory experience. To answer this question, we imaged mice with congenital sensorineural hearing loss. Otoferlin is required for synaptic release from cochlear inner hair cells, and mutation in the otoferlin-encoded gene (OTOF) causes sensorineural hearing loss (48–51). OTOF knockout (OTOF^−/−) mice show reduced auditory cortical responses in early development (47), diminished auditory brainstem response (ABR) 3 wk after birth, and are profoundly deaf as adult (52). We compared the interhemispheric functional connectivity in normally hearing F1 mice (female n = 5, male n = 5) with otoferlin knockout mice (OTOF^−/−, female n = 3, male n = 2) by recording spontaneous activity from both hemispheres for 10 min. However, since in the OTOF^−/−, sound-evoked responses cannot be utilized to locate ACx, we developed two different approaches to recognize ROIs falling within the putative ACxs. The first method relies on fitting an average ACx geometry model derived from normal hearing control (OTOF^+/+) mice (Fig. 2), and the second method relies on clustering neighboring pixels that were intrahemispherically correlated. Compared with normal hearing mice (Fig. 6 A, Left), interhemispheric functional connectivity was lower in OTOF^−/− mice (Fig. 6 A, Right and Fig. 6B and SI Appendix, Fig. S12). Despite the overall reduction in correlation, the matched fields remained relatively higher in correlation than the nonmatching fields in OTOF^−/−. These results suggest that sensory experience is required to establish interhemispheric functional correlations.

To visualize how functional connectivity differs between the OTOF^-/- and OTOF^+/+ groups, we thresholded (r = 0.7) (Fig. 6 C and D) and found that the only ROI pair that stayed above this threshold was between projected left UF and right UF. UF is located near S2, which is dominated by head representation, and we speculate that sensory deprivation of auditory inputs may have induced cross-modal plasticity and that S2 may have taken over putative UF territory (53–56). Studies on congenital deaf mice (57) and in other deaf animals (58, 59) show that regions of ACx, particularly A1 and AAF, can respond to somatosensory input primarily from the head region. Therefore, hearing is required to establish interhemispheric functional correlations between ACxs, and the presence of strong functional interhemispheric correlations between putative UF might be evidence of cross-modal reorganization.

We next investigated the effects of sound experience on intrahemispheric functional connectivity. Fig. 6 E and F shows intrahemispherically correlated pixels above a threshold at 0.95 for control and OTOF^−/− mice with one example, respectively. To visualize the functional connectivity, each pair of correlated pixels was linked with yellow lines with red terminal endpoints. Therefore, the regions covered with appreciable yellow shading indicate the clustering of functionally connected neuronal populations. As was observed from the decreased total area (comparing Upper panels of Fig. 6 E and F) of correlated neighboring pixels and the total number of ROIs identified by clustering the correlated pixels (Lower panels of Fig. 6 E and F), the intra-hemispheric functional connectivity of OTOF^−/− within the projected ACx area appeared to reduce. To compare the observations between animals, the number of ROIs and correlated areas were normalized against the effective areas before comparing the two mouse groups. We defined the effective area for each hemisphere as the total number of pixels that were not blocked by the spatial mask, which is used to exclude the somatosensory cortex. Because the effective area may affect the number of ROIs and the correlated neighboring pixels identified, we divided the two values by the effective areas for each animal resulting in a measure for effective ROI number and correlated area ratio, respectively. The effective ROI number between control and OTOF^−/− mice was similar for both hemispheres (Fig. 6 G, Lower), but the range was larger for the left hemisphere of the OTOF^−/− mice. Conversely, the mean correlated area ratios (Fig. 6 G, Upper) for OTOF^−/− mice were lower compared to control mice (P = 0.03 left hemisphere; P = 0.03 right hemisphere, n = 10 for OTOF^+/+, n = 5 for OTOF^−/−, bootstrapping hypothesis test, boots = 100), which implies that the intrahemispheric functional connectivity within the projected ACx region was weaker and that the origin of this connection was affected by the absence of hearing experience.

All procedures were approved by the Johns Hopkins University Animal Care and Use Committee. For hearing mice, a total of 10 mice were used (5 female and 5 male imaged between 6 and 34 wk of age). Thy1-GCaMP6s (Jackson Laboratory Stock #024275) were crossed with CBA/CAJ (Jackson Laboratory Stock #000654) to prevent early-onset hearing loss (73). C57BL/6 mice are homozygous for the recessive Cdh23 allele, which causes this hearing loss. The resulting offspring are heterozygous Ahl+/Cdh23, ensuring that they have healthy hearing and express GCaMP6s under the Thy1 promoter, which is present in excitatory neurons. For congenital deaf mice, a total of five mice were used (three female and two male imaged between 8 and 14 wk of age). OTOF^−/− mice expressing GCaMP6s were created by crossing offspring of OTOF^−/− and Thy1-GCaMP6s breedings. Founders for our OTOF^−/− breeding colony were generously provided by Tobias Moser (Göttingen). All mice were housed in a 12-h reverse light/dark cycle room in the institutional animal colony before and after experiments.

The bilateral widefield microscope (Fig. 1A) consisted of a pair of identical widefield calcium imaging units synchronized by a shared trigger, each collecting images from one side of the temporal lobe. Blue LEDs (M470L4, Thorlabs)—filtered between 450 and 490 nm (ET470/40x, Chroma)—illuminated each side of the brain through the cranial windows. Emission filters (AT535/40m, Chroma) with bandpass 515-55 nm collected the green fluorescence with peak 490 nm emitted from calcium indicator, GCaMP6s. A dichroic mirror 1 long pass at 499 nm (MD499-FITC, Thorlabs) was used to reflect excitation light while allowing fluorescent emission to propagate to the CMOS camera (CS235MU, Thorlabs). To ensure identical illumination power of two units, we modulated the power between 2.9 and 3.1 mW with a DC voltage. Identical dry objectives (UplanSApo 4×/0.16, Olympus) and tube lens (AC508-150A, Thorlabs) were used on both units to ensure equivalent lateral amplification of 3.3.

Mice were injected with 0.1cc dexamethasone (2 mg/mL, VetOne) 2 to 3 h before surgery to reduce swelling during surgery. Mice were initially anesthetized with 4% isoflurane (Fluriso, VetOne) with continued exposure at 1.5 to 2.5% throughout surgery. Rectal body temperature of the mice was maintained at 36 °C using a heating pad. Hair was removed with scissors and a hair removal agent (Nair). The skin was sterilized by applying Betadine and then 70% ethanol three times. Skin on top of the head was then removed, exposing the skull. Connective tissue was scraped away from the skull using a scalpel in order to prepare the surface for application of the headpost, and the temporal muscle on both sides was resected to create space for two cranial window implants. Cyanoacrylate (Vetbond) was applied to open incisions to prevent bleeding and infection. The headpost was then secured to the top of the head using both super glue (Loctite) and dental acrylic (C&B Metabond, Parkell). A craniotomy with a diameter of approximately 3 mm was performed over the left auditory cortex first. The cranial window consisted of two circular 3-mm glass coverslips and one circular 4-mm glass coverslip secured together using optic glue (NOA71-Norland Products). Silicone elastomer (Kwik-Sil, World Precision Instruments) was carefully placed around the edge of the 4-mm coverslip, and the window was implanted with 3-mm coverslips facing toward the cortex. Dental acrylic was carefully applied around the 4-mm coverslip to further secure the window in place. The same process was then repeated on the right side of the skull: Anatomical landmarks such as the temporal ridge, lambdoid suture, and the medial cerebral artery were used to ensure window symmetry across hemispheres. Immediately following surgery, mice were injected with 0.1cc dexamethasone, 0.05cc cefazolin (1g/vial, West Ward Pharmaceuticals), and 5 mg/kg of carprofen (0.5 mg/mL, Zoetis US) and placed under a heat lamp for recovery. For 3 d following surgery, mice were injected with 0.1cc dexamethasone and carprofen once a day to ensure proper healing from surgery. In addition, antibiotic water (Sulfamethoxazole and Trimethoprim Oral Suspension, USP 200 mg/40 mg per 5 mL, Aurobindo Pharms USA; 6 mL solution diluted in 100 mL water) replaced normal drinking water for 7 d following the procedure.

All data collection was completed at least 1 wk after the craniotomy surgery. Throughout the experiment, the mice were awake with the headplate fixed to a 3-axis translational stage, allowing position adjustments. A dark chamber padded with sound-absorbing foam enclosed the mice, the bilateral widefield microscope, and the speaker to minimize outside noise and light. The imaging system and the sound delivery system were controlled by a custom-made MATLAB graphical user interface (GUI). We set the imaging parameters of both imaging subsystems identically: frame rate 20 Hz, 4-by-4 binning with additional 10.9 dB gain. Binning was required to enhance the signal-to-noise ratio (SNR), and the additional gain was used to remove readout noise under low light.

Stimuli used in sound-evoked response experiments include two sets of pure tones and two sets of vocalizations. The first set of tones fall within the typical mouse hearing ranging between 4 kHz and 64 kHz with logarithmic spacing and two tones per octave, meaning a total of nine frequency levels. The other set of tones consists of five ultrasonic tones, ranging between 70 kHz and 94 kHz with an even spacing of 6 kHz, which coincides with the mouse vocalization spectrum. As for vocalization stimuli, the first set consists of nine different vocalization stimuli created using the “virtual mouse organ” (23), in which syllables recorded from mice were combined into bouts using a third-order Markov model. Each bout fell between 64 kHz and 94 kHz and contained 10 syllables. Vocalizations of p100 mice and older were used. The tenth vocalization stimulus consists of a bout of 10 adult pain calls which contain abundant sound energy at low frequencies (Fig. 1C and SI Appendix, Fig. S1A). Both tones and vocalization stimuli were played for 1 s with a 3 to 3.5-s pause of silence in between. Each tone and vocalization stimulus was played at 35 dB, 50 dB, and 65 dB. Stimuli were played in a random order to prevent the prediction of stimulus. For both tones and vocalizations, each stimulus was repeated 10 times. Sound waveforms were loaded into an RX6 multifunction processor (Tucker-Davis Technologies) and fed through a PA5 attenuator (Tucker-Davis Technologies) to output the intended sound level. Output from the PA5 was fed to an ED1 speaker driver and, subsequently, an ES1 electrostatic speaker (Tucker-Davis Technologies). Because auditory stimuli presented to one ear preferentially activates the contralateral hemisphere and since we intended to image neural activity in both hemispheres, the speaker was placed directly in front of the mouse at a distance of 10 cm to prevent bias in the activity of one hemisphere. Pure tone stimuli were generated and calibrated by custom MATLAB script; for details, see ref. 12. The peak amplitude of the tones was calibrated to 70 dB SPL using a Brüel & Kjaer 49440A microphone (SI Appendix, Fig. S1B). Vocalization stimuli were fed through a whitening filter from 4 to 95 kHz and calibrated so that the average of the peak sound level was equivalent to 70 dB (SI Appendix, Fig. S1C). There were no systematic differences in sound levels (<0.05 dB between 4 and 64 kHz) measured at the position of each ear.

Spontaneous activity was recorded for 10 min in silence without presenting any sensory stimulus to the mice. For all hearing mice, sound-evoked response experiments were completed first followed by the spontaneous activity experiment without releasing the mouse to ensure identical field of view throughout all experiments. In total, all experiments together lasted ~67 min. For congenitally deaf mice, only the spontaneous activity experiment was conducted.

All the data analysis was done by custom MATLAB scripts. Because the response nature of spontaneous activity is different from that of sound-evoked activity, we have two protocols to calculate the fractional difference in fluorescence signals, ΔF/F₀, with respect to the baseline F₀. In sound-evoked response experiments, the baseline was defined as the average fluorescence intensity four frames preceding the stimulus onset. In spontaneous response experiments, the baseline (F₀) of ΔF/F₀ was defined as the average pixel values of the first 550 image frames (~30 s) with linear adjustment according to the drifting trend. The overall fluorescence drifting level for each time point was calculated by taking the moving average, with a window size of 550 frames, of the original fluctuation trace.

In the present study, a variety of approaches are used to define ROIs, and these can be classified into three categories: sound-stimulation method, cofluctuation correlation method, and map projection method. When presenting pure tones and vocalizations, the sound-stimulation method is used to reveal the topology of ACx, map out the tonotopic gradient, and contrast the response strengths in different cortical areas in both hemispheres. In addition, the sound stimulation method is also used in functional connectivity analysis in OTOF^+/+ mice, where the noise correlation and the synchronicity of the spontaneous activity within the sound-sensitive areas are calculated. Alternatively, the cofluctuation method can be used to identify ROIs by analyzing the cofluctuating regions within or across the hemispheres. Because this method does not require neurons to respond to the sound stimulus, it can be applied to both OTOF^+/+ and OTOF^−/− mice. The map projection method is specifically used on OTOF^−/− mice. We determined the ACx topology map from sound stimulation on multiple OTOF^+/+ mice and projected this map onto the cortical surface of the OTOF^−/− mice. Both the cofluctuation and map projection methods are used for functional connectivity analysis.

For the sound-stimulation method, peaks of activation were calculated by averaging across the 20 playout frames, and then averaging across the 10 trials in order to obtain a single “activation image” that characterizes the response to each stimulus. Then, a Gaussian filter with sigma 1.5 was applied to the activation images corresponding to each stimulus. ROI masks were overlaid over the original activation image to isolate the active regions. Each area of activation was assigned to a particular cortical area A1, A2, AAF, DM, or UF using the tonotopic gradients as reference (SI Appendix, Fig. S3).

Functional connectivity measurements evaluate the degree of correlation between spontaneous activity of different areas. Hence, by definition, “functionally connected neurons” fluctuate together (74), which means that they either share the same source of fluctuation or transmits fluctuation to each other by different or indirect connections. To find intrahemispheric connected areas, the pixel-wise correlation was calculated. To perform pixel-wise correlation calculation, each image frame from the stitched ΔF/F image series was first reshaped into a column vector. Each column vector was then arranged in time order to form a 2D matrix, where each row represents the fluctuation of ΔF/F over the time course for the corresponding pixel. Pearson’s correlation coefficient was next calculated between all the possible combinations of rows. By applying a given correlation coefficient threshold and resuming the pixel locations back to their 2D form, the functional connectivity between pixels within and across hemispheres was visualized. The red-shaded regions shown in SI Appendix, Fig. S10B demonstrate the characteristic pattern when applying a relatively high correlation threshold across hemispheres, 0.85 in this case, which correspond to the secondary (S2) somatosensory cortex and somatosensory parietal–ventral area (PV) in both hemispheres, respectively (75). We used S2 and PV as the dorsorostral boundary of the ACx and applied a correlation threshold of 0.95 within each hemisphere to identify potential ACx (yellow regions in SI Appendix, Fig. S10B). After clustering the neighboring correlated pixels, ROIs can be defined, shown in SI Appendix, Fig. S10C, and demonstrated a similar pattern as by the sound-stimulation method, shown in SI Appendix, Fig. S10A.

The map projection method was developed because the OTOF^−/− mice do not respond to sound stimulation. The method is based on the assumption that the relative location between ACx and the axis formed between S2 and PV is fixed. The projection of the ACx topography model derived from OTOF^+/+ mice onto OTOF^−/− mice was done to define ROIs. An ACx topography model is an averaged map of sound-responsive ROIs identified from 10 OTOF^+/+ mice. For the convenience of averaging the model across animals, we transferred the ROIs from local coordinates to global coordinates by taking the centroid of S2 as a reference for shifting and rotating the ROIs with respect to S2 centroid until the line formed by centroids of S2 and PV was vertically oriented, as shown in SI Appendix, Fig. S11B. The projection of the ROIs was made by aligning the orientation and position of S2 and PV in OTOF^−/− mice (SI Appendix, Fig. S11A) with that of the model (SI Appendix, Fig. S11 B and C).

The peak fluorescence response within each area was calculated by finding the maximum pixel value within the ROI and then taking the average of a 3 × 3 pixel square around the maximum pixel. The accuracy of each peak location was confirmed by manually ensuring that the location fell on the cortex rather than blood vessels or window edges. Only activity falling within the top 90th percentile was considered a peak for geometric comparisons. For comparison of activation strength of each cortical area, if activity within a particular cortical area did not fall within the top 90th percentile, a 3 × 3 square around maximum pixel value of the total area of that region based on tonotopic gradients was used (SI Appendix, Fig. S3).

The area of activation to each stimulus was calculated by averaging the activation images across all attenuation levels (35 dB, 50 dB, and 65 dB) and then applying a Gaussian filter using “imgaussfilt” with sigma 1.5 to the averaged image to eliminate noise (SI Appendix, Fig. S2). Singular value decomposition was performed on the Gaussian filtered images, and the first 10 principal components were used. The images were binarized using “imbinarize” at two different thresholds: one threshold at 50% of the maximum value in the image and one threshold at 35% of the maximum value. The binarized images for each stimulus were summed together, and any pixel >1 was considered active. The number of pixels within the binarized images was calculated using the “bwarea” function. The number of pixels was multiplied by a conversion factor in order to find the area in $\mathrm{\mu m}$ . We used a 0.01-mm stage calibration slide (MUHWA) to measure the effective pixel resolution of our setup to be 1.86 $\mathrm{\mu m}$ . The conversion factor was then calculated by:

where $d$ corresponds to the downsampling factor (2), and $b$ corresponds to the binning factor (4). The conversion factor was calculated to be 2.21 × 10⁻⁴ mm²/pixel. Division of the auditory cortex into subregions was achieved by overlaying tonotopic gradients to manually trace which regions of the binarized image correspond to what cortical areas (SI Appendix, Fig. S2). Percent of total area is calculated as follows:

Hemispheric Area Bias was calculated using the following equation: (% total area_Left − % total area_Right)/(% total area_Left + % total area_Right). The HAB was compared across hemispheres and across mice, and either rank-sum or paired-sample t tests were used depending on the distribution in order to determine differences in distribution of activation across hemispheres.

The geometry between cortical areas was calculated using a global coordinate system by setting hotspots of correlated activity (R > 0.95) that are located in the somatosensory cortex as origin. The centroids of each hotspot were calculated for both hemispheres by taking the average pixel locations within a particular hotspot. Then, the dorsal centroid and rostral centroid were used to create the y axis of a global coordinate system, and the existing coordinate system of each image was rotated accordingly (Fig. 2A). The location(s) of peak activation in (x, y) coordinates in response to each stimulus presented (4 to 64 kHz, played at 35 dB, 50 dB, and 65 dB) were determined. For each stimulus, the location of peak activation was determined for each ACx area (five areas), in each hemisphere (2), across all mice (10 mice). The (x, y) coordinates of peak activation in response to each stimulus were averaged over all sound levels. Coordinates for 4 kHz and 5.7 kHz were averaged together and represent the “low-frequency” group. Coordinates for 16 kHz and 22.6 kHz were averaged together and represent the “midfrequency” group. Coordinates for 45.3 kHz and 64 kHz were averaged together and represent the “high-frequency” group. Vectors between the new origin and each peak location (with coordinates rotated to the new axis) were determined (Fig. 2A). Right hemisphere axes were rotated around the y axis, aligned, and overlayed onto the left hemisphere axes to visualize differences between left and right hemisphere locations (Fig. 2B). The following comparisons were made. LvR: The difference between the (x, y) coordinates of the peak location in the left hemisphere and the peak location in the right hemisphere (in a unit of distance) within each mouse was calculated. For each frequency group (low, medium, and high), there were one to five locations of peak activation depending on whether the cortical area (five areas) had a peak activation greater than the 90th percentile. The locations of peak activation were then compared within all mice (10 mice). LvL*: The difference between the (x, y) coordinates of the peak location in the left hemisphere of one mouse and the corresponding (x, y) coordinates of the peak location in the left hemisphere in all other mice was calculated. For instance, the difference between peak location in left A1 to low frequencies in mouse 1 and the peak location in left A1 to low frequencies in mouse 2 to 10 was compared. This was repeated for all cortical areas and frequencies. RvR*: The same method was used as in LvL* except using the right hemisphere locations. LvR*: The difference between the (x, y) coordinates of the peak location in the left hemisphere of one mouse and the corresponding (x, y) coordinates of the peak location in the right hemisphere of all other mice was calculated. For instance, the difference between peak location in left A1 to low frequencies in mouse 1 and the peak location in right A1 to low frequencies in mouse 2 to 10 was compared.

Given an ROI (see the ROI Definitions section), a ΔF/F trace (nearly 10 min long for spontaneous experiments and 1 s for sound-evoked experiments) can be calculated by taking the average value within the ROI. These ROI-specific traces would be paired together to calculate Pearson’s correlation coefficient indicating the functional connectivity between ROIs.

The responses evoked by nine pure tones were used to calculate noise correlation. The ROIs used here were defined by sound stimulations. For the purpose of determining functional connectivity between ROIs, comparisons were only made between traces that were collected from the same sound presentation trial. The following equation is used to calculate the noise correlation (76):

where p and q stand for two distinct ROIs, and v is the ΔF/F trace representing the corresponding ROI during sound presentation trial i. In the present study, we calculate noise correlation between ROIs across hemispheres. Each trace was calculated within an observation window of 1 s at a frame rate of 20 Hz.

G.C., C.-T.C., and P.O.K. designed research; G.C. and C.-T.C. performed research; G.C. and C.-T.C. analyzed data; and G.C., C.-T.C., and P.O.K. wrote the paper.

The authors declare no competing interest.