Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Biomedical Imaging and Instrumentation

Biomedical Imaging and Instrumentation - Poster Session C

Poster F5 - Improving Quality of Automated Image Analysis for High-throughput Neuromodulator Screens

Friday, October 25, 2024

10:00 AM – 11:00 AM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

Ananya Sundararajan (she/her/hers)

Undergraduate Student
Case Western Reserve University
Irvine, California, United States

Co-Author(s)

FA

Fox Avery

Graduate Research Assistant
Worcester Polytechnic Institute, United States

Primary Investigator(s)

DA

Dirk Albrecht

Associate Professor
Worcester Polytechnic Institute, United States

Introduction: Chemical neuromodulators can be leveraged as potential therapies for neuropsychological disorders. We developed a high-throughput functional screening method that can analyze the effects of such neuromodulators on neural activity over time in living organisms [1]. This method yields large imaging data sets, which can increase the power and certainty of results. However, manual data processing is time consuming and often results in imprecise neuron detection [1]. A previously constructed automated data analysis system reduced manual processing time by 6-fold (22 hours of manual processing vs. 3.5 hours of automated processing), but resulted in ~10% increased variability, likely due to neuron selection and position tracking errors [2]. This work focuses on refining the automation of high-throughput data analysis to improve accuracy and validate neuron identification and quality. The increased accuracy and reduced manual processing time should allow for more sensitive and reliable analyses and identification of neuromodulators with therapeutic potential.

Materials and

Methods: Raw image data were obtained from prior experiments using Caenorhabditis elegans that co-expressed the fluorescent calcium sensor GCaMP and optogenetic excitatory ion channel Chrimson in AWA chemosensory neurons (Figure 1) [1]. Animals were immobilized through hydrogel fixation in 384-well plates and compounds from the Library of Pharmacologically Active Compounds (LOPAC, 1280 drugs) were added to the wells at 100 uM. 1% DMSO was used for negative controls and calcium channel blocker Nemadipine-A was used for positive controls (10 and 100 uM). Neurological responses to 5-second optogenetic activation (via 617 nm red light) were recorded every 6 hours for 18 hours [1].
Because active neurons increase fluorescence only during stimulation, individual candidate neurons were identified based on increased pixel brightness coincident with stimulation onset. A custom ImageJ script applied a Gaussian blur, minimum auto-threshold, and particle size filter. Candidate neurons were grouped by animal to accommodate minor movement between imaging intervals. A montage of the candidate neurons was created to validate chosen animals. Manual quality scoring was used to identify an automated algorithm that utilized max projections and thresholds to exclude candidates with excess movement, missing neurons, overlapping animals, and non-neurons. Pixel intensity data were recorded for each true neuron to generate a neural response trace. Mean peak neural response was averaged for each well and compared over time. Candidate hits were selected by a change in neural activity over 12 hours exceeding three standard deviations of negative vehicle controls (|Z-score| > 3).

Results, Conclusions, and Discussions: While manual processing of this data required 22 hours of active user interaction per 384-well plate, the previous automated analysis reduced user interaction time 40-fold at the expense of a small increase in uncertainty in neuron selections. The refined algorithm described here processed data for a 384-well plate with equal user interaction time as the previous algorithm. Although the computation time increased when compared to the previous algorithm (10 hours vs. 3.5 hours), the refined algorithm increased certainty of the neurons that are tracked through assessment of a montage visualization of candidate regions (Figure 2). Candidate neurons were effectively identified using functional responses that increased signal intensity only during stimulation, excluding autofluorescence, dust, gut signals, and non-targeted neurons. Further, animals were 100% correctly identified and matched across time points, enabling assessment of drug effects in each animal. However, up to 40-50% of candidate regions were found to contain errors, including excess movement during the imaging trial, autofluorescent gut flashes, overlapping animals, rotation of animals, flipping of animals between cycles, and excess background noise. From manual categorical binning of data quality, further post-processing algorithms were developed that automatically classified neuron quality based on minimal movement and appropriate size, shape and location. Although higher quality data selection was expected to reduce variability among negative controls, it remained roughly the same with a coefficient of variation of 10% (Figure 3). This suggests that the variability seen in the data set is reflective of true biological variability. Additionally, felodipine was identified as a hit, consistent with previous manual and automated analyses (Figure 3). The improved algorithm and data quality strengthen the certainty of candidate neuroactive compound hits. Further work may involve refining threshold criteria for neuron quality. The analyzed data set targeted intracellular disruption of neural response, since optogenetic activation and activity output were in the same neuron. Future work will include analyzing the effects of pharmacological modulators on intracellular communication like glutamatergic synapses in the AFD-AIY neural circuit. This system can accelerate other studies of functional neural imaging, such as investigating neural activity after traumatic brain injury or dose response for neuroactive compounds.

Acknowledgements (Optional): [1] Lagoy RC., Larsen E., Albrecht DR. (2021). Automated Functional Screening for Modulators of Optogenetically Activated Neural Responses in Living Organisms. Channelrhodopsin 2021 (pp. 221-233).
[2] Avery F., Albrecht DR. (2023). Automating Image Analysis for High-Throughput Screening of Neuromodulators in Living Organisms. BMES Annual Meeting; October, 2023; Seattle, WA.
This work was supported by NSF REU grant EEC2150076.