Overview

A central question in neuroscience is how sensory stimuli from the environment are detected and processed by neural circuits to generate perception and behavior. Our laboratory has developed new molecular viral-genetic approaches to characterize, target and manipulate defined neural cell types in the olfactory cortex of mice. We have identified, using in vivo two-photon microscopy and electrophysiological recordings, fundamental principles of odor information coding in cortical neural networks. These recent advances open up new opportunities to explore how diverse neural cell types contribute to odor information coding, and how this information is transmitted to downstream target areas involved in multimodal sensory integration, perception, and motor control. We are also interested in how brain and behavioral state and learning and experience shape olfactory neural network functions.

The olfactory system of mice provides a simple, tractable model system of outstanding ethological importance. Furthermore, olfactory neural circuits are particularly vulnerable to aging and neurodegenerative disease, thus representing a highly relevant experimental model system for clinical and translational neuroscience research.

Odor coding in cortical neural circuits

Keywords: in vivo imaging, behavioral monitoring, computational modeling

Olfactory perception and behaviors critically depend on our ability to detect and discriminate odorant molecules, identify changes in the composition of odorant mixtures, and accurately track dynamic fluctuations in odor concentration. Using in vivo 2-photon calcium imaging and population coding analyses, we have shown that odor identity can accurately be decoded from ensembles of neurons in the mouse piriform cortex (Roland et al., 2017). We have proposed a model in which distinct perceptual features of odors, such as odor identity, intensity, or valence, are encoded in independent subnetworks of neurons in the olfactory cortex.

An important question is whether odor information encoded in piriform ensembles is selectively transmitted to different downstream brain targets, and how information routing is shaped by brain and behavioral state and learning and experience. We address these questions by chronically recording neuronal activity in the piriform cortex of awake, behaving mice. We use gradient-index (GRIN) lens technology and 2-photon and mini-endoscope microscopy, combined with high-resolution video tracking of behavior. We monitor the activity of specific subsets of neurons by labelling neurons based on their molecular identity or projection targets, using intersectional viral/genetic gene targeting. Computational approaches including supervised and unsupervised machine learning allow us to generate quantitative mechanistic models of neural circuit functions.

Molecularly distinct mitral cell types selectively project to different cortical centers for olfaction. Schematic representation of experimental design and results: injection of a retro-propagated virus expressing nuclear GFP into posterior piriform cortex (purple) and anterior olfactory nucleus (blue) for single-nucleus RNA sequencing. We found distinct molecularly-defined mitral and tufted cell types (UMAP, first square), with distinct gene regulatory network topology (second square) and preferential connectivity patterns along the anterior-posterior axis of the olfactory cortex (third square and drawing on the right). 

Olfactory learning and behavior

Keywords: associative learning, neural circuit plasticity, Fos-tagging

Odor memories are exceptionally robust and essential for animal survival. The olfactory cortex has long been hypothesized to encode odor memories, yet the cellular substrates for olfactory learning and memory remained unknown. Using intersectional, cFos-based genetic manipulations (‘‘Fos tagging’’), we have shown that olfactory fear conditioning activates sparse and distributed ensembles of neurons in the mouse piriform cortex (Meissner-Bernard et al., 2019). Chemogenetic silencing of these Fos-tagged piriform ensembles selectively interferes with odor fear memory retrieval but does not compromise basic odor detection and discrimination. Furthermore, chemogenetic reactivation of piriform neurons that were Fos tagged during olfactory fear conditioning causes a decrease in exploratory behavior and an increase in stress hormone levels in the blood, mimicking odor-evoked fear memory recall. 

Ongoing experiments in the lab explore the neural mechanisms for olfactory-spatial associations in a variety of behavioral tasks, combined with mini-endoscope imaging in freely moving mice. Together, our experiments identify specific ensembles of piriform neurons as critical components of an olfactory fear memory trace and provide new experimental avenues for investigating memory formation, retrieval, and memory dysfunction in the olfactory system.

Data engineering and computational modeling

A ubiquitous modern challenge for neuroscience researchers is working with large multimodal data sets, which often include combinations of calcium imaging, electrophysiology, behavioral tracking through video and other sensors, and electrical or optogenetic stimulation. Due to the large scale of such data, data acquisition, processing, storage and sharing poses major challenges. 

To tackle this issue we have developed a custom data processing pipeline, built around Suite2P and the Neurodata Without Borders (NWB) file format, to standardize our data processing workflow. All code is open source and available on GitLab – Fleischmann Lab. We use Suite2p for cell segmentation and extraction of calcium activity time series, and, in parallel, we  parse behavioral data coming from an angular wheel sensor and a pressure sensor recording the animal’s sniffing. We then merge together behavioral data and calcium imaging data into one single NWB file. This custom data processing pipeline allows researchers to incorporate robust, shareable, standard file formats into their workflow and enables efficient, reproducible analyses. 

We leverage our workflow to reverse engineer computational models of neural circuit function and behavior. Through multi-regression and generalized linear models, we extract the behavioral data contained in each NWB file and model each cell independently in the context of its circuit activity. These techniques allow us to generate cell profiles and identify overlapping ensembles through which we can explore the influence of odor cues and behavioral state on neuronal activity. We also use unsupervised learning algorithms to visualize, quantify, and integrate information contained within these large multidimensional data sets. For example, we are currently testing how Laplacian Eigenmaps and Tensor Component Analysis can capture latent circuit structure and identify functionally distinct neuronal subpopulations in an unsupervised manner. 

Data engineering and computational modelling work is done in close collaboration with Dr. Jason Ritt, Scientific Director of Quantitative Neuroscience at the Robert J. and Nancy D. Carney Institute for Brain Science.

Biomarkers for neurodegenerative disease

Many devastating neurodegenerative diseases, including Alzheimer’s and Parkinson’s Disease present similar symptoms, including progressive loss of cognitive and motor functions that eventually lead to dementia. Normal Pressure Hydrocephalus (NPH) is a neurodegenerative disease characterized by the accumulation of cerebrospinal fluid (CSF) in the ventricles of the brain and presents with cognitive, motor, and urinary symptoms. NPH is thought to account for up to 5-10% of all dementia cases but, unlike most other neurodegenerative diseases, can be treated.

The standard treatment for NPH is a surgical procedure to implant a shunt in the brain ventricle, which drains excessive CSF. With early diagnosis, shunt implantation can revert symptoms and dramatically ameliorate patients’ quality of life. However, because of overlapping symptoms with other neurodegenerative diseases, correct diagnosis of NPH is difficult. Consequently, an estimated 80% of NPH cases remain unidentified. Moreover, no current test can predict shunt surgery benefit.

Using next-generation RNA sequencing (RNA-seq), liquid-chromatography mass spectroscopy (LC-MS), and immunoassays, we are developing a biomarker test for the diagnosis of NPH and prediction of surgery success. Combining this multi-omic approach with novel machine learning and data science tools, our NPH diagnostic test will allow physicians to accurately identify patients with NPH and recommend shunt surgeries to those who will benefit.

This interdisciplinary project, led by Dr. Maria Grazia Ruocco, is a collaboration across Brown University, including Dr. Petra Klinge in the Department of Neurosurgery at Rhode Island Hospital, Dr. Thomas Serre in the Department of Cognitive, Linguistic, and Psychological Sciences, and our lab. The commercialization of this diagnostic tool is being developed by the Brown-affiliated biotechnology startup Adelle Diagnostics, Inc.