Machine Learning Pipeline for Pathological Analysis of Synaptic Loss in Human iPSC-Derived Neuronal Model of Human Prion Disease

1. Overview

Scientific Rationale and Importance: A defining feature shared by both prion and Alzheimer’s disease is early synaptic loss and neuronal hyperactivity that closely parallels cognitive decline; however, the molecular mechanisms driving these processes remain unclear. Currently, there is a lack of human-relevant models to study synapse loss in prion disease (PrD). Previous investigations have primarily relied on mouse models and primary neurons. While these are powerful tools, they lack a human genetic background, are difficult to scale for therapeutic testing, and often depend on manual, subjective quantification. Here, we propose two key innovations:

Determine how prion-mediated signaling affects synapse loss in a novel human induced Pluripotent Stem Cell (iPSC)-derived neuronal model.
Develop a machine-learning–based pipeline to analyze the morphology, number, and distribution of synapses, amenable to high-throughput image analysis.

Funding, Facilities, and Ethics Approvals: The proposed image analysis pipeline operates exclusively on de-identified iPSC-derived excitatory neuron (iN) images and does not involve human subjects or personally identifiable information, thereby minimizing ethical risk. Computing will be performed at the San Diego Supercomputer Center (SDSC), and high-throughput imaging will be conducted using the BioPipeline High Content Imaging System at the Nikon Imaging Center, University of California (UC) San Diego. The PI has extensive expertise in modeling neurodegenerative processes using transgenic mice, primary neurons, human postmortem tissue, and human iPSC-derived neuronal systems. This expertise, combined with established collaborations and access to advanced imaging platforms at UC San Diego, uniquely positions the team to carry out this project successfully.

Timeline: 6 months.

Months 1–2: Training of machine-learning–based pixel and object classification models with preliminary datasets.
Months 2–3: Culture and differentiation of human iPSCs into excitatory neurons (iNs).
Months 4–5: Treatment with a PrP^Sc-mimetic antibody and confocal imaging.
Months 5–6: Final data analysis and preprint preparation.

Data Sharing Commitment: All analysis code and processing pipelines will be released publicly on GitHub under an open-source license. Preprints will be uploaded to ResearchHub and bioRxiv to ensure immediate community access.

2. Introduction

Synaptic dysfunction and loss are the strongest correlates of cognitive decline in Alzheimer's Disease (AD)^1-4, underscoring the need to identify the mechanisms that drive synaptic loss. Similarly, synaptic loss is an early neuropathological feature of experimental prion disease models⁵ and is observed in human brains with sporadic Creutzfeldt-Jakob disease (sCJD)⁶. Yet, the signaling mechanisms driving synaptic injury in human neurons remain poorly understood. One potential mechanism involves persistent neuronal hyperactivity driven by aberrant glutamatergic receptor signaling. This sustained activity elevates intracellular Ca²⁺, leading to excitotoxicity⁷, which may contribute to early synaptic atrophy in prion disease (PrD).

Patients with sCJD exhibit signs of heightened neuronal activity, including enhanced functional connectivity in the default mode networks on fMRI⁸, periodic sharp wave complexes on EEG⁹, and myoclonus¹⁰. Similarly, prion-infected mice exhibit early neuronal hyperactivity characterized by elevated Arc and dysregulation of glutamatergic receptors, specifically reduced mGlu5 and increased phosphorylated GluA1, before pronounced neuronal loss occurs¹⁷. Consistent with a state of chronic excitation, prion-infected brains display elongated and concave post-synaptic structures¹⁸. These findings suggest that neuronal hyperactivity and glutamatergic dysregulation are central, early features of prion disease and may represent the earliest steps leading to synaptic degeneration.

The proposed work addresses two main limitations in the field:

Aim 1: While mouse models are powerful, they may not fully recapitulate human-specific neuronal responses due to differences in receptor composition and intracellular signaling cascades. Recently, human iPSC-derived neurons carrying PRNP mutations revealed aberrant synaptic structure¹⁹, but these genetic models do not elucidate the acute signaling events triggered by PrP^C stimulation, nor do they recapitulate the synaptic loss observed in human prion disease. We have developed a novel iN model that recapitulates increased Arc expression upon exposure to a PrP^Sc-mimetic antibody. We will now test how this treatment contributes to longitudinal synapse loss in human neurons.

Aim 2: Despite advances in optical microscopy, which have enabled increasingly detailed visualization of neuronal structure and activity across spatial and temporal scales, the field still primarily relies on manual quantification. Current computational image analysis methods for neuronal morphology or injury^20,21 are not amicable for large datasets as they face significant limitations in segmentation, morphological reconstruction, and feature quantification. We will apply machine-learning–based pixel and object classification to develop a novel, publicly available analysis pipeline to overcome these barriers.

Impact Statement:

These aims will create the first human high-throughput pipeline to study PrD by pairing a biologically relevant human model with an advanced machine learning–based image analysis pipeline. We will enable high-throughput, quantitative, and unbiased assessment of disease phenotypes at scale. Because PrP^C interacts with other pathogenic aggregates, including Aβο, α-synuclein, and tau ^11-15, the knowledge gained will impact future studies on AD and related tauopathies and synucleinopathies.

3. Methods

Aim 1. Determine early synaptic changes in human iPSC-derived neurons following PrP^C stimulation.

Our goal is to address the temporal progression of synaptic injury following PrP^C stimulation in human neurons.

We have established a platform using human excitatory iNs that reproduces the neuronal hyperactivity phenotype observed in human PrD brains. RNA-seq and immunofluorescence confirmed robust expression of neuronal glutamatergic markers, and PRNP, but not pluripotent, glial, or inhibitory neuronal transcripts¹⁶. We stimulated iNs with the PrP^Sc-mimetic antibody, POM1, from two hours to five days. By two hours, iNs showed a 2.4-fold increase in Arc and phosphorylation of ERK1/2, recapitulating the neuronal hyperactivity phenotype observed in prion-infected brains¹⁷. We hypothesize that PrP^Cstimulation will activate glutamatergic receptors, leading to subsequent synapse loss in human neurons. To test this hypothesis, we will treat iNs with PrP^Sc-mimetic antibody (150 nM) for 1, 3, 5, 7, and 14 days and quantify dendritic area as well as pre- and post-synaptic numbers using immunofluorescent staining for MAP-2, synapsin-1, and PSD-95, respectively. Nonspecific IgG treatment will be used as a control. High-content imaging will be conducted in the UCSD Nikon Imaging Center using a fully automated robotic system equipped with a spinning-disk confocal microscope. We will analyze ≥150 neurons across 3 independent biological replicates (≥50 neurons per replicate) for soma and dendrite metrics. ≥50 neurons with high-resolution synaptic labeling for perisomatic synapse quantification, yielding >1,000 synapses for analysis. Power analyses indicate that this sample size provides >80% power (α = 0.05) to detect ≥10–15% differences between automated and manual measurements. The analysis in Aim 1 will be conducted using manual measurements by trained researchers. As such, guaranteeing Aim 1 is not dependent on Aim 2.

Statistical analyses: For all aims, differences among the means will be analyzed using a t-test with Welch’s correction, or ANOVA, followed by a Tukey post-hoc test for pair-wise comparison.

Aim 2: Develop and validate an open, reproducible image analysis pipeline for quantitative neuronal morphometry.

Our goal is to develop and validate a robust, semi-automated image analysis pipeline integrating ImageJ and ilastik to enable reliable, scalable quantification of neuronal structure from our human iNs. This pipeline will specifically improve segmentation and detection, morphological reconstruction and tracing, and quantification of structural features in neuronal imaging data.

To develop this pipeline, we will use available images of iNs treated with the PrP^Sc-mimetic antibody (POM1) for 3 or 5 days. We imaged MAP2 and SYP to evaluate injury to dendrites and pre-synapses, respectively. Next, we will develop a modular workflow that combines ilastik’s machine-learning–based pixel and object classification with ImageJ’s established tools for image processing, skeletonization, and morphometric analysis. The pipeline will be optimized for fluorescently labeled neuronal preparations and will support batch processing to ensure scalability and reproducibility. The pipeline will quantify the following key metrics:

1. Neuronal density – number of neurons per unit area, derived from automated soma detection and segmentation.

2. Dendritic complexity – number of primary and secondary dendrites per neuron, extracted from skeletonized reconstructions.

3. Dendritic length – total dendrite length per neuron, computed from traced dendritic arbors.

4. Perisomatic synapse density – number of synapses localized within the first 100 nm from the neuronal cell body, using co-localized synaptic markers and spatial filtering relative to the segmented soma.

By leveraging ilastik’s interactive learning capabilities, the pipeline will reduce dependence on hard-coded thresholds and improve performance in images with variable signal-to-noise ratios, uneven illumination, and dense neuronal labeling.

Validation Strategy: Pipeline performance will be validated against expert manual annotations and existing analysis methods (as described in Aim 1). Validation will focus on three key aspects. Segmentation and classification accuracy: assessed using the following equations (1) precision, (2) recall, and (3) F1 scores for soma and dendrite detection with true positive (TP), false positive (FP), and false negative (FN).

Morphological (4) accuracy, evaluated by comparing automated neuronal density, dendritic counts and lengths, and perisomatic synapse density to manually traced ground-truth reconstructions.

Reproducibility: We will assess reproducibility across independent users and imaging sessions. Manual ground-truth datasets will be generated by blinded expert annotators using standardized annotation protocols.

Ethics and Data Management: The project uses de-identified images (no human subjects). Data management will follow FAIR (Findable, Accessible, Interoperable, Reusable) principles. Raw and processed image datasets, along with associated metadata (imaging parameters, preprocessing steps, and analysis settings), will be systematically organized and version-controlled. All scripts, ilastik project files, and macros will be archived on GitHub under an open-source license.

4. Pilot Data (Optional)

We established a platform using human excitatory iNs that reproduces the neuronal hyperactivity phenotype, including increased Arc observed in human prion disease-affected brains¹⁶. Immunofluorescence in our iNs showed expression of mature neuronal markers (MAP2, Tuj1, and NeuN), excitatory vesicular glutamate transporters (vGlut), and post-synaptic protein (PSD95) (Fig. 1). PrP^C was expressed in the soma and in distinct puncta around the Tuj1+ neurites, congruent with synaptic localization of PrP^C (Fig. 1).

We treated our iNs with a PrP^Sc mimetic antibody and an unspecific IgG for 1 - 3 days to determine the effect of prion signaling in synapse loss. High-content imaging was conducted using a fully automated robotic system equipped with a spinning-disk confocal microscope to image dendrites and pre-synapses (MAP2 and SYP, respectively) from treated iNs (Fig 2A). Next, to determine the effect of PrP signaling in dendritic injury, we used ImageJ to semi-automatically detect MAP2+ structures. Preliminary tests showed that standard ImageJ tools failed to correctly segment MAP2+ dendrites due to signal intensity variability (Fig 2B). However, by implementing machine-learning–based pixel classification, we successfully segmented MAP2+ neurites (Fig 2C).

Further training to detect SYP+ structures allowed us to batch process images to determine pre-synapses (Fig. 3). Yet, additional training and optimization are necessary to validate our analysis. Next, it is essential to integrate the machine-learning–based classification into a unified platform to perform segmentation and detection, morphological reconstruction and tracing, and quantification of structural features in neuronal imaging data.

5. Budget

Category	Description	Amount (USD)
Tissue Culture	Reagents for iPSC and iN maintenance	$3,000
Imaging Reagents	Antibodies and optical imaging dishes	$2,500
Microscopy Fees	Confocal microscopy hourly rates	$2,000
Computing/Storage	SDSC fees and external HDD storage	$500
Dissemination	Poster printing and conference registration	$2,000
Total		$10,000

6. References

DeKosky, S. T. & Scheff, S. W. Synapse loss in frontal cortex biopsies in Alzheimer's disease. Ann. Neurol. 27, 457–464 (1990).
DeKosky, S. T., Scheff, S. W. & Styren, S. D. Structural correlates of cognition in dementia. Neurodegeneration 5, 417–421 (1996).
Selkoe, D. J. Alzheimer's disease is a synaptic failure. Science 298, 789–791 (2002).
Scheff, S. W. et al. Synaptic alterations in CA1 in mild Alzheimer disease and mild cognitive impairment. Neurology 68, 1501–1508 (2007).
Cunningham, C. et al. Synaptic changes characterize early behavioural signs in the ME7 model of murine prion disease. Eur. J. Neurosci. 17, 2147–2155 (2003).
Ferrer, I. Synaptic pathology and cell death in the cerebellum in Creutzfeldt-Jakob disease. Cerebellum 1, 213–222 (2002).
Verma, M., Lizama, B. N. & Chu, C. T. Excitotoxicity, calcium and mitochondria: a triad in synaptic neurodegeneration. Transl. Neurodegener. 11, 3 (2022).
Paoletti, M. et al. Default Mode Network quantitative diffusion and resting-state functional magnetic resonance imaging correlates in sporadic Creutzfeldt-Jakob disease. Hum. Brain Mapp. 43, 4158–4173 (2022).
Wieser, H. G., Schindler, K. & Zumsteg, D. EEG in Creutzfeldt-Jakob disease. Clin. Neurophysiol. 117, 935–951 (2006).
Sequeira, D. et al. Prevalence and Treatments of Movement Disorders in Prion Diseases. Mov. Disord. 37, 1893–1903 (2022).
Prusiner, S. B. et al. Evidence for alpha-synuclein prions causing multiple system atrophy in humans with parkinsonism. Proc. Natl. Acad. Sci. U. S. A. 112, E5308–5317 (2015).
Prusiner, S. B. et al. Scrapie prions aggregate to form amyloid-like birefringent rods. Cell 35, 349–358 (1983).
Prusiner, S. B. et al. Further purification and characterization of scrapie prions. Biochemistry 21, 6942–6950 (1982).
Ferreira, D. G. et al. $\alpha$-synuclein interacts with $PrP^C$ to induce cognitive impairment through mGluR5 and NMDAR2B. Nat. Neurosci. 20, 1569–1579 (2017).
Balducci, C. et al. Tau oligomers impair memory and synaptic plasticity through the cellular prion protein. Acta Neuropathol. Commun. 13, 17 (2025).
Ojeda-Juarez, D. et al. $PrP^C$-induced signaling in human neurons activates phospholipase $C\gamma1$ and an Arc/Arg3.1 response. bioRxiv (2025).
Ojeda-Juarez, D. et al. Prions induce an early Arc response and a subsequent reduction in mGluR5 in the hippocampus. Neurobiol. Dis. 172, 105834 (2022).
Lawrence, J. A. et al. Diminished Neuronal ESCRT-0 Function Exacerbates AMPA Receptor Derangement and Accelerates Prion-Induced Neurodegeneration. J. Neurosci. 43, 3970–3984 (2023).
Gojanovich, A. D. et al. Abnormal synaptic architecture in iPSC-derived neurons from a multi-generational family with genetic Creutzfeldt-Jakob disease. Stem Cell Reports 19, 1474–1488 (2024).
Choquet, D. et al. Advanced imaging and labelling methods to decipher brain cell organization and function. Nat. Rev. Neurosci. 22, 237–255 (2021).
Benisty, H. et al. Review of data processing of functional optical microscopy for neuroscience. Neurophotonics9, 041402 (2022)

Machine Learning Pipeline for Pathological Analysis of Synaptic Loss in Human iPSC-Derived Neuronal Model of Human Prion Disease

Machine Learning Pipeline for Pathological Analysis of Synaptic Loss in Human iPSC-Derived Neuronal Model of Human Prion Disease

Funding Progress

Funders

Related Funding

Request for Proposals - Neuroscience

Nonprofit Recipient

DOI