Michael D. Kim

Live imaging of multicolor-labeled cells in Drosophila

We describe LOLLIbow, a Brainbow-based live imaging system with applications in developmental biology and neurobiology. The development of an animal, including the environmentally sensitive adaptation of its brain, is thought to proceed through continual orchestration among diverse cell types as they divide, migrate, transform and interact with one another within the body. To facilitate direct visualization of such dynamic morphogenesis by individual cells in vivo, we have modified the original Brainbow for Drosophila in which live imaging is practical during much of its development. Our system offers permanent fluorescent labels that reveal fine morphological details of individual cells without requiring dissection or fixation of the samples. It also features a non-invasive means to control the timing of stochastic tricolor cell labeling with a light pulse. We demonstrate applicability of the new system in a variety of settings that could benefit from direct imaging of the developing multicellular organism with single-cell resolution.

Three-dimensional neurite tracing under globally varying contrast

We study the 3D neurite tracing problem in different imaging modalities. We consider that the examined images do not provide sufficient contrast between neurite and background, and the signal-to-noise ratio varies spatially. We first split the stack into box sub-volumes, and inside each box we evolve simultaneously a number of different open-curve snakes. The curves deform based on three criteria: local image statistics, local shape smoothness, and a term that enforces pairwise attraction between snakes, given their spatial proximity and shapes. We validate our method using larva Drosophila sensory neurons imaged with confocal laser scanning microscopy, as well as publicly available datasets.

Specification of Dendritogenesis Site in Drosophila aCC Motoneuron by Membrane Enrichment of Pak1 through Dscam1.

Precise positioning of dendritic branches is a critical step in the establishment of neuronal circuitry. However, there is limited knowledge on how environmental cues translate into dendrite initiation or branching at a specific position. Here, through a combination of mutation, RNAi, and imaging experiments, we found that a Dscam-Dock-Pak1 hierarchical interaction defines the stereotypical dendrite growth site in the Drosophila aCC motoneuron. This interaction localizes the Cdc42 effector Pak1 to the plasma membrane at the dendrite initiation site before the activation of Cdc42. Ectopic expression of membrane-anchored Pak1 overrides this spatial specification of dendritogenesis, confirming its function in guiding Cdc42 signaling. We further discovered that Dscam1 localization in aCC occurs through an inter-neuronal contact that involves Dscam1 in the partner MP1 neuron. These findings elucidate a mechanism by which Dscam1 controls neuronal morphogenesis through spatial regulation of Cdc42 signaling and, subsequently, cytoskeletal remodeling.

Three-Dimensional Motor Neuron Morphology Estimation in the Drosophila Ventral Nerve Cord

Type-specific dendritic arborization patterns dictate synaptic connectivity and are fundamental determinants of neuronal function. We exploit the morphological stereotypy and relative simplicity of the Drosophila nervous system to model the diverse neuronal morphologies of individual motor neurons (MNs) and understand underlying principles of synaptic connectivity in a motor circuit. Our computational approach aims at the reconstruction of the neuron morphology, namely the robust segmentation of the neuron volumes from their surroundings with the simultaneous partitioning into their compartments, namely the soma, axon, and dendrites. We use the idea of cosegmentation, where every image along the z -axis (depth) is segmented using information from “neighboring” depths. We use 3-D Haar-like features to model appearance. Because soma and axon are determined by their distinctive shapes, we define an implicit shape representation of the 2-D segmentation sets to drive cosegmentation and achieve the desired partitioning. We validate our method using image stacks depicting single neurons labeled with green fluorescent protein (GFP) and serially imaged with laser scanning confocal microscopy.

CRF formulation of active contour population for efficient three-dimensional neurite tracing

We present a conditional random field for three-dimensional neurite tracing, using a population of open curve active contours. We aim at increased robustness under spatially varying neurite-background contrast, and at the same time reducing the computational complexity compared to the state-of-the-art. While most existing active contour based methods perform tracing by evolving multiple snakes along the neurite centerline in a sequential manner, our approach implements a simultaneous evolution, reducing the complexity as we show theoretically in our algorithm analysis and experimentally. Our approach provides increased accuracy in ambiguous regions (e.g., low contrast, neurite bifurcations and crossovers, etc.), by exploiting interactions among spatially neighboring snakes. We illustrate the performance of our method and compare it with existing frameworks using sample volumes of wild-type sensory neurons in the larval Drosophila.

Patterning motor neurons in the Drosophila ventral nerve cord using latent state Conditional Random Fields

Type-specific dendritic arborization patterns dictate synaptic connectivity and are fundamental determinants of neuronal function. We exploit the morphological stereotypy and relative simplicity of the Drosophila nervous system to model the diverse neuronal morphologies of individual motor neurons (MNs) and understand underlying principles of synaptic connectivity in a motor circuit. In our analysis, we use images depicting single neurons labeled with green fluorescent protein (GFP) and serially imaged with laser scanning confocal microscopy. We model morphology with a novel formulation of Conditional Random Fields, a latent state CRF, to capture the highly varying compartment-based structure of the neurons (soma-axon-dendrites). We integrate a multi-class logistic model as the local potential function for combining compartment features. All parameters are learned in a single procedure, while L1-norm logistic model parameters are added in the maximum pseudo-likelihood model for learning with better scalability. The regularization hyper-parameters are chosen with a minimum cross-validation generalization error model.

Motor neuron morphology estimation for its classification in the Drosophila brain

Type-specific dendritic arborization patterns dictate synaptic connectivity and are fundamental determinants of neuronal function. We exploit the morphological stereotypy and relative simplicity of the Drosophila nervous system to model the diverse dendritic morphologies of individual motor neurons (MNs) to understand underlying principles of synaptic connectivity in a motor circuit. The genetic tractability of Drosophila allows us to label single MNs with green fluorescent protein (GFP) and serially reconstruct identifiable MNs in 3D with confocal microscopy. Our computational approach aims at the robust segmentation of the MN volumes and the simultaneous partitioning into their compartments, namely the soma, axon and dendrites. We use the idea of co-segmentation, where every image along the z-axis (depth) is clustered using information from ‘neighboring’ depths. As appearance we use a 3D extension of Haar features and for the shape we define an implicit representation of the segmentation domain.

Part-based motor neuron recognition in the Drosophila ventral nerve cord.

We exploit the morphological stereotypy and relative simplicity of the Drosophila nervous system to model the diverse neuronal morphologies of individual motor neurons and understand underlying principles of synaptic connectivity in a motor circuit. In our analysis, we use images depicting single neurons labeled with green fluorescent protein (GFP) and serially imaged with laser scanning confocal microscopy. We model morphology with a novel formulation of Conditional Random Fields, a hierarchical latent-state CRF, to capture the highly varying compartment-based structure of the neurons (soma-axon-dendrites). In the training phase, we follow two approaches: (i) hierarchical learning, where compartment labels are given, and (ii) latent-state learning, where compartment labels are not given in the samples. We demonstrate the accuracy of our approach using wild-type motor neurons in the larval ventral nerve cord. However, our method can also be used for the identification of motor neuron mutations, as well as the automated annotation of the motor circuitry in wild type and mutant animals. Our method is directly applicable to the recognition of compartment-defined structures.

Motor neuron recognition in the Drosophila ventral nerve cord

We exploit the morphological stereotypy and relative simplicity of the Drosophila nervous system to model the diverse neuronal morphologies of individual motor neurons and understand underlying principles of synaptic connectivity in a motor circuit. In our analysis, we use images depicting single neurons labeled with green fluorescent protein (GFP) and serially imaged with laser scanning confocal microscopy. We model morphology with a novel formulation of Conditional Random Fields, a hierarchical latent-state CRF, to capture the highly varying compartment-based structure of the neurons (soma-axon-dendrites). In the training phase, we follow two approaches: (i) hierarchical learning, were compartment labels are given, and (ii) latent-state learning, where compartment labels are not given in the training samples. We demonstrate the accuracy of our approach using wild-type MNs in the larval ventral nerve cord. However, our method can also be used for the identification of MN mutations, as well as the automated annotation of the motor circuitry in wild type and mutant animals.

Neurite reconstruction from time-lapse sequences using co-segmentation

We introduce a novel segmentation method for time-lapse image stacks of neurites based on the co-segmentation principle. Our method aggregates information from multiple stacks to improve the segmentation task, using a neurite model and a tree similarity term. The neurite model takes into account branching characteristics, such as local shape smoothness and continuity, while the tree similarity term exploits the local branch dynamics across image stacks. Our approach improves accuracy in ambiguous regions, handling successfully out-of-focus effects and branching bifurcations. We validated our method using Drosophila sensory neuron datasets and made comparisons with existing methods.