Neurons are complex cells with dynamic morphogenesis involving the elongation of a developing process, called a neurite. Neurite outgrowth occurs in response to many stimuli and illnesses and is studied in various disease contexts with the hope of aiding development of nervous system therapeutics. However, researchers who study neuron morphogenesis face the difficult task of visualizing and measuring neuronal outgrowth. Normally, images of the neurite structures are captured with microscopy and fed into automated image analysis programs that quantify neurite elongation and branching; however, these images often have low signal-to-noise-ratios (SNR) and are of insufficient clarity and resolution to reliably quantify neurite growth. Researcher Alvason Li from the Corey and Zhu labs (Vaccine and Infectious Disease Division) recently developed a novel system for automated identification of neurite outgrowth and published their method in Scientific Reports.
To capture images of neurites, neurons are placed in two-chamber microfluidic devices which allow for their directional growth. Neurites are then stained with a fluorescent antibody and imaged with an inverted microscope. It is this step of the process that causes low SNR images, as dyes can cause high intensity fluorescent background signal and cellular debris can introduce noise. After images are captured, a computer software program uses statistical tracing to scan the image to detect neurite branching and growth in an unbiased and automated manner. In recent years, several groups have developed stochastic algorithms to aid in quantification of neurite elongation, including the hidden-Markov-model (HHM), a type of Dynamic Bayesian Network statistical model that predicts outcomes based on the previous state. Despite improving the field of automated measuring models, quantification of elongation in neurite images with low SNR remains unreliable with current methods. Li, who initially “was inspired by the Markov Chain method in phylogenetic tree analysis,” began to apply random-reaction-seed (RRS), a model based in HHM, to neurite detection. RRS is a dynamic tracing algorithm that infers an outcome based on learned probabilities: in this case RRS predicts the maximum-likelihood path of the neurite based on the dynamic search process raised by a random starting point.
In order to test the reliability of RRS in the context of neurite outgrowth quantification, the authors first compared RRS-mediated detection of neurite elongation to deterministic algorithms, simpler models that always produce the same output with any given input. However, many of these methods require time-consuming interventions of human visual intelligence during the analysis process—such as manual election of parameters such as contrast—which introduce bias, inconsistency, and inaccuracy. RRS-mediated measurements, when compared to deterministic-mediated neurite analysis, performed better with low SNR images and did not require manual parameter adjustment. The authors then compared RRS to other standard HHM-based algorithms. Some of the current HHM models rely on the presence of the soma body, the core of the neuron cell from which the dendritic neurites branch, as a reference point for measuring branching length. However, microfluidics chambers in which neurons are imaged maintain the soma bodies and neurites in discrete compartments, creating a need for models that do not rely on soma bodies for neurite tracing. RRS, a random-seeding algorithm, does not require a specific seeding point such as the soma body to be able to identify the net-like structure of branching neurites. RRS method is more accurate and effective in quantifying neurite growth than both deterministic and existing statistical HHM-based models.
Li and colleagues, who study peripheral nervous system in human skin in the context of herpes simplex virus 2 (HSV-2) infection, are beginning to implement the RRS method to measure neurite changes in tissues. HSV-2 infects genital skin and affects axons and local innervating nerve networks, however, the sensory neurons are bundled distantly in the ganglia away from the site and cannot be imaged. Therefore, the RRS methods has a clear advantage over others methods of quantification. Accordingly, the authors have begun to expand the uses for RRS and have successfully imaged and quantified nerve fibers in tissues, suggesting that it is “highly possible” that the “RRS method could be applied to neurite measurement in skin biopsies,” Li explains. Furthermore, Li and colleagues believe that the RRS-mediated robust detection of neurites in planar analysis could be extended to other uses such as three-dimensional live-cell and retinal vessel imaging. Li elaborates, saying that “a hybrid system combining this stochastic RRS method and deep learning (machine intelligence) method will be promising for biomedical feature extractions from tissue images. For instance, this hybrid intelligent system has provided our lab a super way to precisely quantify the spatial distribution of chimeric antigen receptor T cells in tissue. By applying this enhanced RRS method for cell membrane detection and multiplex RNA FISH, we can phenotype nearly a quarter million cells in a complex tissue section at the single cell level.”
Li AZ, Corey L, Zhu J. 2019. Random-reaction-seed method for automated identification of neurite elongation and branching. Scientific Reports. 2019; 9: 2908. doi: 10.108/s41598-019-39962-0
This work was supported by the National Institutes of Health.
Fred Hutch/UW Cancer Consortium members Jia Zhu and Larry Corey contributed to this work.