Tonotopic and multisensory organization of the mouse dorsal inferior colliculus revealed by two-photon imaging

The dorsal (DCIC) and lateral cortices (LCIC) of the inferior colliculus are major targets of the auditory and non-auditory cortical areas, suggesting a role in complex multimodal information processing. However, relatively little is known about their functional organization. We utilized in vivo two-photon Ca2+ imaging in awake mice expressing GCaMP6s in GABAergic or non-GABAergic neurons in the IC to investigate their spatial organization. We found different classes of temporal response, which we confirmed with simultaneous juxtacellular electrophysiology. Both GABAergic and non-GABAergic neurons showed spatial microheterogeneity in their temporal responses. In contrast, a robust reversed rostromedial-caudolateral gradient of frequency tuning was conserved between the two groups, and even among the subclasses. This, together with the existence of a subset of neurons sensitive to spontaneous movements, provides functional evidence for redefining the border between DCIC and LCIC.


13
The inferior colliculus (IC) is a major auditory processing center, which receives input from most 14 brainstem auditory nuclei. The IC is usually divided into three main regions: the central nucleus (CNIC), 15 the lateral cortex (LCIC), and the dorsal cortex (DCIC; Oliver 2005). The latter two, together also known 16 as the "shell region" of the IC, are part of the non-lemniscal pathway, and are heavily innervated by 17 feedback projections from the cerebral cortex. The DCIC is often defined as the part that covers the CNIC 18 dorsally (e.g. Zhou and Shore 2006). The LCIC is characterized by distinct neurochemical modules, which 19 can be visualized with various histochemical methods such as acetylcholinesterase or NADPH diaphorase 20 staining as well as dense labelling of GABAergic terminals and sparse calretinin immunoreactivity 21 Here, we describe the use of two transgenic mouse lines to characterize GABAergic and glutamatergic 43 neuronal subpopulations in the dorsal IC in awake animals using in vivo two-photon Ca 2+ imaging. We 44 studied GABAergic neurons with a Gad2-IRES-Cre mouse (Taniguchi et al. 2011) that was crossed with 45 the GCaMP6s reporter line Ai96 (Madisen et al. 2015) and a sub-population of non-GABAergic neurons 46 using the Thy1-driven GCaMP6s transgenic line GP4.3 (Dana et al. 2014). We show a rich diversity of 47 sound-evoked responses in both GABAergic and non-GABAergic neurons in the awake mouse, confirmed 48 with simultaneous juxtacellular electrophysiology in awake animals. Remarkably, we observed a reversal 49 of the characteristic frequency gradient in the rostromedial-caudolateral direction, which was conserved 50 between GABAergic and non-GABAergic cells, as well as among cells with different classes of sound-51 evoked response. Moreover, we found a subset of neurons that were responsive to spontaneous 52 movement to the animal, and were potentially associated with multisensory neurochemical modules 53 (Lesicko et al. 2016). These findings suggest that the large majority of the dorsal IC surface belongs to 54 the LCIC. 55

Expression of GCaMP6s in IC subpopulation with transgenic mice 57
To better understand the response heterogeneity observed in IC neurons, we attempted to divide the 58 neuronal population through selective expression of the GCaMP reporter. Two transgenic mouse lines 59 were used: GP4.3, a Thy1-driven GCaMP6s expression line (Dana et al. 2014), and a cross between 60 Gad2-IRES-Cre line (Taniguchi et al. 2011) and Ai96, a Cre-dependent GCaMP6s reporter line (Madisen et 61 al. 2015). From here on we will refer to the latter as Gad2;Ai96. 62 To confirm the neurochemical identity of GCaMP6s-positive neurons in the transgenic lines, we 63 performed immunolabelling of GAD67.  Figure 1C. In particular, we observed that 69 GCaMP6s-positive neurons are enriched at the border of neurochemical modules characterized by 70 dense GAD67 terminals ( Figure 1D). 71 In contrast, we detected positive GAD67 immunoreactivity in >93% (1702/1828) of GCaMP-positive 72 neurons in the Gad2;Ai96 mice. Again, we observed that a fraction of the GAD67-positive cells in this 73 mouse line were not labelled by GCaMP6s. 74 In addition to GAD67, the calcium buffer parvalbumin (PV) is another marker for neurochemical modules 75 in the rat (Chernock et al. 2004), while the calcium buffer calretinin (CR) shows a complementary 76 expression (i.e. in the extramodular region) in juvenile and early postnatal mice (Dillingham et al. 2017). 77 We therefore investigated the co-expression of PV and CR in the two transgenic lines. We found that the 78

Sound-evoked changes in FGCaMP of IC neurons 84
Unanaesthetized, head-fixed mice were subjected to 1-s pure tones of various frequencies and 85 intensities ( Figure 2). The use of longer stimuli allowed a detailed characterization of the kinetics of the 86 fluorescence change (ΔFGCaMP). Importantly, we observed both increases, which were either transient or 87 sustained, and decreases in FGCaMP during the presentation of stimuli, and some cells showed a sharp 88 increase in FGCaMP directly after the cessation of the sound ( Figure 2E). The different fluorescence 89 response classes were classified as onset/sustained, inhibitory and offset, respectively ( Figure 3A-D). The 90 decrease in fluorescence upon sound stimulation was likely due to sound-evoked inhibition of 91 spontaneous firing in these cells. Some cells showed a mixture of response kinetics, some even to the 92 same stimulus. Particularly common was the combination of inhibitory and offset responses (e.g. Figure  93 3E,M). Another common combination were onset-offset cells in which a lower frequency elicited an 94 onset/sustained response, while a slightly higher frequency elicited an offset response (e.g. Figure 3I).

Relationship between cell type and soma size 114
GABAergic cells had a larger soma size ( Figure 4C), defined as the area of the ROI in imaging experiments, 115 as had been reported previously (cat: Oliver et al. 1994; rat: Merchán et al. 2005). In our dataset there 116 was a tendency for larger cells to have a more sustained response ( Figure 4D). 117 Detailed kinetics of the different response classes 118 The temporal kinetics of the FGCaMP response of each cell was assessed by averaging the fluorescence 119 change across all stimuli that showed a significant response. Single exponential functions were fit on the 120 onset (0-1 s re stimulus onset) and the offset/decay (0.5-4 s re stimulus offset) periods ( electrophysiological recordings showed that it can represent a cell that quickly adapted during our one-132 second long stimuli (example in Figure 5A-D). We observed sustained fluorescence responses that 133 corresponded to sustained firing ( Figure 5D) with different amount of adaptation. Inhibitory responses 134 corresponded to cells that reduced their spontaneous firing upon sound stimulation ( Figure 5D). 135 We further characterized the relationship between fluorescence responses and firing patterns by fitting 136 a model to our data set ( Figure 5E,F; see Materials and Methods for equations). We found that on 137 average, an action potential led to a median increase of 0.35 (GP4.3; n = 5 cells) or 0.14 (Gad2;Ai96; n = 138 12 cells) times the minimal fluorescence (F0), with a median zero-to-peak rise time of 480 ms (GP4.3) or 139 560 ms (Gad2;Ai96) and a median decay time constant of 1.04 s (GP4.3) or 1.36 s (Gad2;Ai96). The 140 model explains between 50% and 95% of the variance in the fluorescence (median: 79%; average ± s.d.: 141 77 ± 13%). 142

Spatial distribution of frequency tuning 143
The widespread and homogeneous expression of GCaMP6s in IC of the transgenic mice allowed a good 144 overview of its functional organization. We aligned the position of the cells from multiple animals (GP4.3: 145 n = 7 mice; Gad2;Ai96: n = 6 mice) by anatomical landmarks (midline, anterior and posterior extent of 146 the exposed IC, lateral extent of the exposed IC), and plotted the cells on this common anatomical 147 coordinate system ( Figure 6A; Video 1). 148 Interestingly, we observed a central strip running in the caudomedial-rostrolateral orientation of cells 149 responding to lower frequency, while CF progressively increased in both caudolateral and rostromedial 150 directions. We tried to estimate the orientation of the spatial organization by projecting the x,y-151 coordinates of all neurons onto an axis with a parametrized angle θ, while simultaneously fitting the log-152 transformed CF using a polynomial model (see Materials and Methods). We used a polynomial model as 153 a generic fit as a means to extract any general direction along which CFs seem to diverge, without 154 imposing any presumption about the spatial dependence of CF. We found that a fourth order 155 polynomial captured the variance maximally; further increasing its order did not lead to better fits. The 156 results of the fit are presented as contour lines on Figure 6A, with the best orientation (θ) 50° from the 157 medial-lateral (x) axis. 158 Barnstedt et al. (2015) suggested from clustering analysis that these low frequency neurons were from 159 the most dorsal end of the central nucleus. While we do not exclude the possibility of imaging into the 160 central nucleus of the IC, we believe the observed tonotopy is a good representation of the IC shell 161 region because all neurons reported in this data set did lie within 160 μm from the pia surface. In 162 addition, all CFs were represented among the most superficial neurons (<40 μm deep, Figure 6B). There 163 was, however, an over-representation of low CF neurons among the deepest imaged (121 -160 μm 164 deep, Figure 6B). We attribute this to a sampling bias where the low frequency region coincided with 165 the center of the cranial window, thus providing the best optical access for deeper imaging. 166 To better visualize the reversal of tonotopy, we projected the neurons onto our best orientation ( Figure  167 6C-E). There was a high count of low CF (~4 kHz) neurons near the 600 μm position ( Figure 6C), with the 168 mode going towards ~20 kHz at the 1300 μm end. The increase in CF on the rostromedial side was not as 169 pronounced, likely due to the low number of neurons sampled in this region. The solid line in Figure 6C  170 shows our best polynomial fit, which explained around 13% of the variance in CFs. A similar spatial 171 distribution of CF was observed for GABAergic and glutamatergic neurons ( Figure 6D) and for cells 172 showing different response classes ( Figure 6E). 173

Spatial distribution for response classes 174
We next asked whether there was any obvious spatial organization of the different response classes. 175 Figure 7A shows the proportion of different response classes along our presumed tonotopic axis, which 176 looked rather homogenous, without any obvious difference between cells rostromedial and caudolateral 177 to the CF minimum. The same pattern holds for the orthogonal direction ( Figure 7B). cochlear nucleus, which also receive somatosensory inputs (Wu et al. 2014). This prompted us to 184 investigate whether these "spontaneous" activities could be attributed to non-auditory inputs. Our 185 recordings were performed while the animal was passively awake, which allowed us to observe and 186 correlate the voluntary movement of the animal to simultaneously recorded calcium transients. 187 We found that the onset of many of these spontaneous transients coincided with movement events of 188 the animal (paw and facial movements in Figure 8A). By correlating movement and fluorescence, we 189 detected 165 (out of 1359) cells that showed a positive correlation (r > 0.25) with facial movements 190 during the spontaneous recording period. To exclude potential false-positive detection due to motion 191 artefacts (i.e. cells moving into and out of focus), we excluded cells with fluorescence transients that did 192 not show the typical ~1 s exponential decay kinetics, which last much longer than the brief image shifts 193 that could accompany movement events (image shifts in Figure  Since animal movement inevitably produces sound that may activate the IC neurons, leading to an 195 apparent movement sensitivity, we also excluded the cells that showed excitatory or offset responses in 196 their FRA. In the end, 41 cells showed spontaneous calcium transients that correlated with animal 197 movement and that could not be explained by their FRA, which was either inhibited or showed no clear 198 pure tone evoked responses (example FRAs in Figure 8F). These cells seemed to be enriched at the 199 caudolateral side of the dorsal IC ( Figure 8C Comparison of tonotopic organization with histological data and literature 204 Figure 9A and G show two relatively superficial brain sections (within 80 μm and 120 μm from dorsal 205 surface, respectively) from one Gad2;Ai96 and one GP4.3 animal stained for GAD67 after two photon 206 imaging. We overlaid the line representing the neurons with the lowest CF and compared its location 207 with the border of LCIC and DCIC traced from the latest Allen Reference Atlas (CCFv3) and from the 208 classical reference atlas by Paxinos and Franklin (2001). The minimum frequency line aligned with the 209 demarcation from the Allen Reference Atlas, while that by Paxinos Figure 10E). Aligning this to post hoc histological staining ( Figure 10D), we can demonstrate 223 that cell 4 has dendritic branches extending both inside and outside of GAD67 modules ( Figure 10F). We 224 suggest that this may be a functional connection scheme for extramodular neurons to enable them to 225 integrate somatosensory and auditory inputs to the IC ( Figure 10G). 226

227
We investigated the functional organization of the dorsal IC in awake mice at the single neuron level. 228 We found that the dorsal IC is tonotopically organized, with isofrequency bands running in a 229 rostrolateral-caudomedial orientation with frequency tuning varying along the caudolateral-230 rostromedial direction. The lowest CF band was found in the middle, dividing the dorsal IC into two 231 reversed tonotopic gradients. In the caudolateral part, but not in the rostromedial part, neurons that 232 were sensitive to whisker or other movements were found, which was in agreement with the view that 233 the caudolateral and rostromedial part corresponded to the LCIC and DCIC, respectively. We observed 234 four different types of firing patterns in response to tones, but other than the tonotopical organization, Another striking finding was therefore that we saw clear evidence for offset responses in 3-11% of 280 sound-responsive cells. A whole-cell study in the mouse IC showed that in most cases offset responses 281 are inherited from upstream, but that they may also be generated de novo as a rebound from inhibition 282 (Kasai et al. 2012). Similar to the inhibition class, we therefore suggest that the prominence of offset 283 responses may be related to the increased impact of synaptic inhibition in the IC of awake animals. A 284 study using communication calls suggests that offset responses may be even more prominent following 285 more complex sound stimuli, such as communication calls (Akimov et al. 2017), possibly by a summation 286 of inhibition at different frequency bands (Sanchez et al. 2008). 287

Tonotopical organization of the dorsal IC 288
We observed a tonotopic gradient that ran along a gradient from caudolateral to rostromedial. The 289 gradient reversed at the site where the CFs were the lowest. This reversal is in line with the banding 290 patterns observed with epifluorescent calcium imaging in unanesthetized mice before hearing onset; 291 neurons in these bands fire together, and this activity depended on the cochlea, suggesting that they 292 represented groups of cells with similar CF (Babola et al. 2018). Interestingly, our data are compatible 293 with two earlier calcium imaging studies, even though in the first, which focused on medial regions of 294 the dorsal IC, the gradients were reported to run from lateral (low-frequency) to medial (high-frequency) 295 (Ito et al. 2014), whereas in the second they were reported in a roughly (medio)rostral (low-frequency) 296 to (latero)caudal (high-frequency) direction (Barnstedt et al. 2015). Following pooling of the data from 297 several experiments, we observed a gradient reversal, which, interestingly, was also observed in one 298 animal in the latter study (Barnstedt et al. 2015). 299 We found considerable variability in CFs within a band. Imprecisions in the alignment of imaging areas 300 and variability between animals may have contributed to the variability, but substantial variability was 301 also observed within a single animal, in line with previous results (Ito et al. 2014; Barnstedt et al. 2015). The four response types showed a similar tonotopic organization, but considerable microheterogeneity. 307 This microheterogeneity was larger than in an earlier study in which bulk loading of an organic Ca

Presence of somatosensory inputs in LCIC 321
We monitored spontaneous whisker and general body movements, and used strict criteria to exclude 322 movement artefacts. As we restricted the analysis to neurons that were not excited by tones, we 323 consider it unlikely that the observed somatosensory responses were instead caused by self-generated 324 sounds. Despite these severe restrictions, we did find that a few percent of the cells were excited during 325 whisker or body movements, suggesting a substantial role for somatosensory (or motor) inputs, 326 especially considering that in previous research only a minority of cells have been shown to respond to 327 unimodal tactile stimuli, and the dominant effect of somatosensory inputs appears to be inhibitory 328 Through an opening in the head plate, a craniotomy of 3 mm diameter centered at one of the ICs was 415 made by thinning and removing the skull bone. A cranial window, made by gluing a 3 mm cover slip (CS-416 3R-0; Warner Instrument Inc, Hamden, CT, USA) on a custom built, 500 µm thick steel ring with UV-417 cured optical adhesive (NOA68; Norland Products), was installed over the exposed brain surface and 418 secured with superglue. Each animal was allowed to recover for at least two days before the first 419 measurements. For in vivo electrophysiology, the cranial window construct was gently removed, and the 420 dura mater covering the IC and part of the cerebellum was carefully punctured and removed with a pair 421 of fine forceps. 422 After all recordings had been done, animals received an intraperitoneal injection of pentobarbital (300 423 mg/kg) and were perfused transcardially, first with physiological saline (Baxter Healthcare, Zurich, 424 AlexaFluor-, Cy3-or Cy5-conjugated secondary antibodies (Invitrogen or Jackson ImmunoResearch). To 496 ensure a more homogeneous GAD67 fluorescence for the post-hoc immunostaining of imaged brains, 497 the brain slices were incubated twice in the primary antibody solution, each for 1 week at 4 °C. The 498 secondary antibody was also applied twice, but overnight at room temperature. Cell counting was 499 performed manually in FIJI using the Cell Counter plug-in on confocal z-stacks. 500

General Analysis 501
Data analysis was performed with Igor Pro (WaveMetrics, Inc., Lake Oswego, OR, USA) using custom 502 written procedures. Two-photon images and behavior video were aligned to ClampEx data using 503 stimulus timing. Whisking behavior was assessed by calculating the root-mean-square of the frame to 504 frame intensity difference in an area at the whisker pad. 505 Movement artefacts in two-photon images were corrected based on the built-in ImageRegistration 506 operation in Igor Pro, which is based on a published algorithm (Thevenaz et al. 1998). Neuronal cell 507 bodies were identified visually based on the average image of the motion-corrected image series, and a 508 higher sampling Z-stack of the area (0.5×0.5 µm pixels; 1 µm z steps) taken directly after each 509 experiment. Regions-of-interests (ROIs) were drawn around cell bodies, and an average fluorescence 510 was extracted for each ROI after subtracting the average fluorescence fluctuation of the surrounding 511 background for further analysis. The background fluorescence was defined as the average fluorescence 512 in a 2 µm wide contour surrounding the ROI, excluding any pixel directly belonging to another ROI. 513

Analysis of Frequency Response Areas 514
The frequency response area (FRA) of each ROI was analyzed largely as described previously (Geis et al., 515 2011). The fluorescence trace within 1 s of each stimulus was taken as the stimulus related waveform. 516 To extract the stimulus-related response, we extracted the signal autocorrelation (also named 517 "consistency index" by Barnstedt et al., 2015) by calculating the average Pearson correlation coefficient 518 among fluorescence waveforms to the same stimuli. 519 Kinetics of sound-evoked responses were fitted by single exponential functions. 520

Orientation of CF Gradient 521
To find the most prominent direction of CF gradient, we parametrized the location of each neuron as a 522

Ground-truth and modelling of GCaMP6s fluorescence 540
Traces from juxtacellular recording were first subjected to a digital DC remove filter which subtracts at 541 each point in time the average potential within ±1 ms to remove DC drift or offset introduced by 542 nanostimulation (Houweling et al. 2010). Traces were blanked around the start and end of each current 543 injection (2 ms before and 3 ms after) to remove stimulus artefacts. Spikes were then detected by a 544 simple thresholding procedure, with spike times defined as the peak time of the spikes, which 545 presumably corresponds to the maximum rate of rise of the action potential.         Epifluorescence image of the IC in a horizontal brain section stained for GAD67. This brain slice was from 915 a Gad2;Ai96 mouse after two-photon imaging. Black straight line indicates the minimum frequency 916 location derived from fitting two-photon imaging data. Dashed curve represents the demarcation 917 between the dorsal and the lateral (external) cortices traced from version 3 of the Allen Reference Atlas. 918 Circles connected with solid and broken lines mark the demarcation at the dorsal brain surface traced 919 from the atlas by Paxinos and Franklin (2001). The rostral and caudal end of this demarcation are 920 marked by a dashed line because at the indicated positions on the anterior-posterior axis the whole 921 structure was labeled as LCIC or DCIC, respectively. Modules with dense GAD67 staining, considered to 922 be a hallmark for the LCIC, were observed both medially and laterally from each of the three 923 demarcations (arrowheads). We observed a region in the center of the IC whose GAD67 staining density 924 was at least as strong as in the neurochemical modules (square labelled B).