Fine-tuning TrailMap: The utility of transfer learning to improve the performance of deep learning in axon segmentation of light-sheet microscopy images.

Light-sheet microscopy has made possible the 3D imaging of both fixed and live biological tissue, with samples as large as the entire mouse brain. However, segmentation and quantification of that data remains a time-consuming manual undertaking. Machine learning methods promise the possibility of automating this process. This study seeks to advance the performance of prior models through optimizing transfer learning. We fine-tuned the existing TrailMap model using expert-labeled data from noradrenergic axonal structures in the mouse brain. By changing the cross-entropy weights and using augmentation, we demonstrate a generally improved adjusted F1-score over using the originally trained TrailMap model within our test datasets.

Fine-tuning TrailMap: The utility of transfer learning to improve the performance of deep learning in axon segmentation of light-sheet microscopy images.

Light-sheet microscopy has made possible the 3D imaging of both fixed and live biological tissue, with samples as large as the entire mouse brain. However, segmentation and quantification of that data remains a time-consuming manual undertaking. Machine learning methods promise the possibility of automating this process. This study seeks to advance the performance of prior models through optimizing transfer learning. We fine-tuned the existing TrailMap model using expert-labeled data from noradrenergic axonal structures in the mouse brain. By fine-tuning the final two layers of the neural network at a lower learning rate of the TrailMap model, we demonstrate an improved recall and an occasionally improved adjusted F1-score within our test dataset over using the originally trained TrailMap model.

Locomotion activates PKA through dopamine and adenosine in striatal neurons.

The canonical model of striatal function predicts that animal locomotion is associated with the opposing regulation of protein kinase A (PKA) in direct and indirect pathway striatal spiny projection neurons (SPNs) by dopamine1-7. However, the precise dynamics of PKA in dorsolateral SPNs during locomotion remain to be determined. It is also unclear whether other neuromodulators are involved. Here we show that PKA activity in both types of SPNs is essential for normal locomotion. Using two-photon fluorescence lifetime imaging8-10 of a PKA sensor10 through gradient index lenses, we measured PKA activity within individual SPNs of the mouse dorsolateral striatum during locomotion. Consistent with the canonical view, dopamine activated PKA activity in direct pathway SPNs during locomotion through the dopamine D1 receptor. However, indirect pathway SPNs exhibited a greater increase in PKA activity, which was largely abolished through the blockade of adenosine A2A receptors. In agreement with these results, fibre photometry measurements of an adenosine sensor11 revealed an acute increase in extracellular adenosine during locomotion. Functionally, antagonism of dopamine or adenosine receptors resulted in distinct changes in SPN PKA activity, neuronal activity and locomotion. Together, our results suggest that acute adenosine accumulation interplays with dopamine release to orchestrate PKA activity in SPNs and proper striatal function during animal locomotion.

High-fidelity, efficient, and reversible labeling of endogenous proteins using CRISPR-based designer exon insertion.

Precise and efficient insertion of large DNA fragments into somatic cells using gene editing technologies to label or modify endogenous proteins remains challenging. Non-specific insertions/deletions (INDELs) resulting from the non-homologous end joining pathway make the process error-prone. Further, the insert is not readily removable. Here, we describe a method called CRISPR-mediated insertion of exon (CRISPIE) that can precisely and reversibly label endogenous proteins using CRISPR/Cas9-based editing. CRISPIE inserts a designer donor module, which consists of an exon encoding the protein sequence flanked by intron sequences, into an intronic location in the target gene. INDELs at the insertion junction will be spliced out, leaving mRNAs nearly error-free. We used CRISPIE to fluorescently label endogenous proteins in mammalian neurons in vivo with previously unachieved efficiency. We demonstrate that this method is broadly applicable, and that the insert can be readily removed later. CRISPIE permits protein sequence insertion with high fidelity, efficiency, and flexibility.

Microglia Regulate Pruning of Specialized Synapses in the Auditory Brainstem.

The assembly of uniquely organized sound localization circuits in the brainstem requires precise developmental mechanisms. Glial cells have been shown to shape synaptic connections in the retinogeniculate system during development, but their contributions to specialized auditory synapses have not been identified. Here we investigated the role of microglia in auditory brainstem circuit assembly, focusing on the formation and pruning of the calyx of Held in the medial nucleus of the trapezoid body (MNTB). Microglia were pharmacologically depleted in mice early in development using subcutaneous injections of an inhibitor of colony stimulating factor 1 receptor, which is essential for microglia survival. Brainstems were examined prior to and just after hearing onset, at postnatal days (P) 8 and P13, respectively. We found that at P13 there were significantly more polyinnervated MNTB neurons when microglia were depleted, consistent with a defect in pruning. Expression of glial fibrillary acidic protein (GFAP), a mature astrocyte marker that normally appears in the MNTB late in development, was significantly decreased in microglia-depleted mice at P13, suggesting a delay in astrocyte maturation. Our results demonstrate that monoinnervation of MNTB neurons by the calyx of Held is significantly disrupted or delayed in the absence of microglia. This finding may reflect a direct role for microglia in synaptic pruning. A secondary role for microglia may be in the maturation of astrocytes in MNTB. These findings highlight the significant function of glia in pruning during calyx of Held development.

Central Compensation in Auditory Brainstem after Damaging Noise Exposure.

Noise exposure is one of the most common causes of hearing loss and peripheral damage to the auditory system. A growing literature suggests that the auditory system can compensate for peripheral loss through increased central neural activity. The current study sought to investigate the link between noise exposure, increases in central gain, synaptic reorganization, and auditory function. All axons of the auditory nerve project to the cochlear nucleus, making it a requisite nucleus for sound detection. As the first synapse in the central auditory system, the cochlear nucleus is well positioned to respond plastically to loss of peripheral input. To investigate noise-induced compensation in the central auditory system, we measured auditory brainstem responses (ABRs) and auditory perception and collected tissue from mice exposed to broadband noise. Noise-exposed mice showed elevated ABR thresholds, reduced ABR wave 1 amplitudes, and spiral ganglion neuron loss. Despite peripheral damage, noise-exposed mice were hyperreactive to loud sounds and showed nearly normal behavioral sound detection thresholds. Ratios of late ABR peaks (2-4) relative to the first ABR peak indicated that brainstem pathways were hyperactive in noise-exposed mice, while anatomical analysis indicated there was an imbalance between expression of excitatory and inhibitory proteins in the ventral cochlear nucleus. The results of the current study suggest that a reorganization of excitation and inhibition in the ventral cochlear nucleus may drive hyperactivity in the central auditory system. This increase in central gain can compensate for peripheral loss to restore some aspects of auditory function.

Hidden hearing loss and endbulbs of Held: Evidence for central pathology before detection of ABR threshold increases.

Reductions in sound-evoked activity in the auditory nerve due to hearing loss have been shown to cause pathological changes in central auditory structures. Hearing loss due strictly to the aging process are less well documented. In this study of CBA/CaH mice, we provide evidence for age-related pathology in the endbulb of Held, a large axosomatic ending arising from myelinated auditory nerve fibers. Endbulbs are known to be involved in the processing of temporal cues used for sound localization and speech comprehension. Hearing thresholds as measured by auditory brainstem response (ABR) thresholds remained stable up to one year, whereas suprathreshold amplitudes of early ABR waves decreased by up to 50% in older mice, similar to that reported for age-related cochlear synaptopathy (Sergeyenko et al., 2013). The reduction of ABR response magnitude with age correlated closely in time with the gradual atrophy of endbulbs of Held, and is consistent with the hypothesis that endbulb integrity is dependent upon normal levels of spike activity in the auditory nerve. These results indicate that central auditory pathologies emerge as consequence of so-called "hidden" hearing loss and suggest that such brain changes require consideration when devising therapeutic interventions.

The effect of progressive hearing loss on the morphology of endbulbs of Held and bushy cells.

Studies of congenital and early-onset deafness have demonstrated that an absence of peripheral sound-evoked activity in the auditory nerve causes pathological changes in central auditory structures. The aim of this study was to establish whether progressive acquired hearing loss could lead to similar brain changes that would degrade the precision of signal transmission. We used complementary physiologic hearing tests and microscopic techniques to study the combined effect of both magnitude and duration of hearing loss on one of the first auditory synapses in the brain, the endbulb of Held (EB), along with its bushy cell (BC) target in the anteroventral cochlear nucleus. We compared two hearing mouse strains (CBA/Ca and heterozygous shaker-2+/-) against a model of early-onset progressive hearing loss (DBA/2) and a model of congenital deafness (homozygous shaker-2-/-), examining each strain at 1, 3, and 6 months of age. Furthermore, we employed a frequency model of the mouse cochlear nucleus to constrain our analyses to regions most likely to exhibit graded changes in hearing function with time. No significant differences in the gross morphology of EB or BC structure were observed in 1-month-old animals, indicating uninterrupted development. However, in animals with hearing loss, both EBs and BCs exhibited a graded reduction in size that paralleled the hearing loss, with the most severe pathology seen in deaf 6-month-old shaker-2-/- mice. Ultrastructural pathologies associated with hearing loss were less dramatic: minor changes were observed in terminal size but mitochondrial fraction and postsynaptic densities remained relatively stable. These results indicate that acquired progressive hearing loss can have consequences on auditory brain structure, with prolonged loss leading to greater pathologies. Our findings suggest a role for early intervention with assistive devices in order to mitigate long-term pathology and loss of function.

Descending projections from the inferior colliculus to the dorsal cochlear nucleus are excitatory.

Ascending projections of the dorsal cochlear nucleus (DCN) target primarily the contralateral inferior colliculus (IC). In turn, the IC sends bilateral descending projections back to the DCN. We sought to determine the nature of these descending axons in order to infer circuit mechanisms of signal processing at one of the earliest stages of the central auditory pathway. An anterograde tracer was injected in the IC of CBA/Ca mice to reveal terminal characteristics of the descending axons. Retrograde tracer deposits were made in the DCN of CBA/Ca and transgenic GAD67-EGFP mice to investigate the cells giving rise to these projections. A multiunit best frequency was determined for each injection site. Brains were processed by using standard histologic methods for visualization and examined by fluorescent, brightfield, and electron microscopy. Descending projections from the IC were inferred to be excitatory because the cell bodies of retrogradely labeled neurons did not colabel with EGFP expression in neurons of GAD67-EGFP mice. Furthermore, additional experiments yielded no glycinergic or cholinergic positive cells in the IC, and descending projections to the DCN were colabeled with antibodies against VGluT2, a glutamate transporter. Anterogradely labeled endings in the DCN formed asymmetric postsynaptic densities, a feature of excitatory neurotransmission. These descending projections to the DCN from the IC were topographic and suggest a feedback pathway that could underlie a frequency-specific enhancement of some acoustic signals and suppression of others. The involvement of this IC-DCN circuit is especially noteworthy when considering the gating of ascending signal streams for auditory processing. J. Comp. Neurol. 525:773-793, 2017. © 2016 Wiley Periodicals, Inc.

Tonotopic organization of vertical cells in the dorsal cochlear nucleus of the CBA/J mouse.

The systematic and topographic representation of frequency is a first principle of organization throughout the auditory system. The dorsal cochlear nucleus (DCN) receives direct tonotopic projections from the auditory nerve (AN) as well as secondary and descending projections from other sources. Among the recipients of AN input in the DCN are vertical cells (also called tuberculoventral cells), glycinergic interneurons thought to provide on- or near-best-frequency feed-forward inhibition to principal cells in the DCN and various cells in the anteroventral cochlear nucleus (AVCN). Differing lines of physiological and anatomical evidence suggest that vertical cells and their projections are organized with respect to frequency, but this has not been conclusively demonstrated in the intact mammalian brain. To address this issue, we retrogradely labeled vertical cells via physiologically targeted injections in the AVCN of the CBA/J mouse. Results from multiple cases were merged with a normalized 3D template of the cochlear nucleus (Muniak et al. [2013] J. Comp. Neurol. 521:1510-1532) to demonstrate quantitatively that the arrangement of vertical cells is tonotopic and aligned to the innervation pattern of the AN. These results suggest that vertical cells are well positioned for providing immediate, frequency-specific inhibition onto cells of the DCN and AVCN to facilitate spectral processing.

3D model of frequency representation in the cochlear nucleus of the CBA/J mouse.

The relationship between structure and function is an invaluable context with which to explore biological mechanisms of normal and dysfunctional hearing. The systematic and topographic representation of frequency originates at the cochlea, and is retained throughout much of the central auditory system. The cochlear nucleus (CN), which initiates all ascending auditory pathways, represents an essential link for understanding frequency organization. A model of the CN that maps frequency representation in 3D would facilitate investigations of possible frequency specializations and pathologic changes that disturb frequency organization. Toward this goal, we reconstructed in 3D the trajectories of labeled auditory nerve (AN) fibers following multiunit recordings and dye injections in the anteroventral CN of the CBA/J mouse. We observed that each injection produced a continuous sheet of labeled AN fibers. Individual cases were normalized to a template using 3D alignment procedures that revealed a systematic and tonotopic arrangement of AN fibers in each subdivision with a clear indication of isofrequency laminae. The combined dataset was used to mathematically derive a 3D quantitative map of frequency organization throughout the entire volume of the CN. This model, available online (http://3D.ryugolab.com/), can serve as a tool for quantitatively testing hypotheses concerning frequency and location in the CN.

Preparation of an awake mouse for recording neural responses and injecting tracers.

It is well known that anesthesia alters neural response properties in various regions of the brain. In the auditory system, fundamental response properties of brainstem neurons including threshold, frequency specificity, and inhibitory sidebands are altered in significant ways under anesthesia. These observations prompted physiologists to seek ways to record from single neurons without the contaminating effects of anesthesia. One result was a decerebrate preparation, where the brainstem was completely transected at the level of the midbrain. The drawbacks of this preparation are a formidable surgery, the elimination of descending projections from the forebrain, and an inability to use sensory stimulation to examine structures above the midbrain. A different strategy has been to implant electrode arrays chronically to record from single neurons and multiunit clusters while the animal is awake and/or behaving. These techniques however are not compatible with injecting tracer dyes after first electrophysiologically characterizing a brain structure. To avoid altering neural response properties with anesthetics while recording electrophysiological response properties from single neurons, we have adapted a head restraint technique long used in bats to mouse. Using this method, we are able to conduct electrophysiological recordings over several days in the unanesthetized mouse. At the end of the recording sessions, we can then inject a dye to reconstruct electrode positions and recording sites or inject a tracer so that pathways to and from the recording loci can be determined. This method allows for well isolated single neuron recordings over multiple days without the use anesthetics.

The spiral ganglion: connecting the peripheral and central auditory systems.

In mammals, the initial bridge between the physical world of sound and perception of that sound is established by neurons of the spiral ganglion. The cell bodies of these neurons give rise to peripheral processes that contact acoustic receptors in the organ of Corti, and the central processes collect together to form the auditory nerve that projects into the brain. In order to better understand hearing at this initial stage, we need to know the following about spiral ganglion neurons: (1) their cell biology including cytoplasmic, cytoskeletal, and membrane properties, (2) their peripheral and central connections including synaptic structure; (3) the nature of their neural signaling; and (4) their capacity for plasticity and rehabilitation. In this report, we will update the progress on these topics and indicate important issues still awaiting resolution.

The neural coding of stimulus intensity: linking the population response of mechanoreceptive afferents with psychophysical behavior.

How specific aspects of a stimulus are encoded at different stages of neural processing is a critical question in sensory neuroscience. In the present study, we investigated the neural code underlying the perception of stimulus intensity in the somatosensory system. We first characterized the responses of SA1 (slowly adapting type 1), RA (rapidly adapting), and PC (Pacinian) afferents of macaque monkeys to sinusoidal, diharmonic, and bandpass noise stimuli. We then had human subjects rate the perceived intensity of a subset of these stimuli. On the basis of these neurophysiological and psychophysical measurements, we evaluated a series of hypotheses about which aspect(s) of the neural activity evoked at the somatosensory periphery account for perception. We evaluated three types of neural codes. The first consisted of population codes based on the firing rate of neurons located directly under the probe. The second included population codes based on the firing rate of the entire population of active neurons. The third included codes based on the number of active afferents. We found that the response evoked in the localized population is logarithmic with stimulus amplitude (given a constant frequency composition), whereas the population response across all neurons is linear with stimulus amplitude. We conclude that stimulus intensity is best accounted for by the firing rate evoked in afferents located under or near the locus of stimulation, weighted by afferent type.