The new study, published in Frontiers in Synaptic Neuroscience, was led by John Carl Begley of Humboldt-Universität zu Berlin and the German Center for Neurodegenerative Diseases (DZNE), together with Harald Prüss of Charité – Universitätsmedizin Berlin, Paul Turko of the Shared Primary Neuron Facility, and Camin Dean of the DZNE and Bernstein Center for Computational Neuroscience Berlin. Their approach combines an established calcium imaging analysis tool called Suite2p with custom Python scripts to automate what has long been a painstaking, manual process of identifying and measuring synaptic activity in cultured neurons.

The technique at the core of the work is synaptic calcium imaging. Neurons in culture are engineered to express GCaMP6f, a genetically encoded fluorescent protein that lights up when calcium floods into the cell. By imaging neurons in a bathing solution that lacks magnesium and contains tetrodotoxin, a sodium channel blocker that silences action potentials, researchers can isolate a specific phenomenon: spontaneous glutamate release from presynaptic vesicles triggers calcium influx through NMDA receptors on the postsynaptic side. Each flash of fluorescence at a dendritic spine therefore reports both presynaptic vesicle fusion and postsynaptic receptor function simultaneously, making it a remarkably information-rich readout of synaptic health.

Until now, however, analyzing these recordings required researchers to manually draw regions of interest around candidate synapses, often using software such as FIJI, and then extract and analyze fluorescence traces with tools like MATLAB or MetaMorph. The process was slow, labor-intensive, vulnerable to experimenter bias, and capped datasets at roughly 1,000 to 2,000 synapses per condition. Because a single neuron typically forms thousands of synapses, most of the synaptic population went unmeasured. The Berlin team recognized that the bottleneck was not the imaging itself but the analysis, and they set out to eliminate it.

After evaluating existing calcium imaging tools, the researchers settled on Suite2p, an open-source Python package originally designed to detect thousands of neuronal cell bodies in two-photon recordings. They reconfigured its detection parameters for the tiny, punctate signals of individual synapses, which are roughly two micrometers in size, and added automated filtering steps. Regions of interest with a fluorescence skew greater than one, indicating contamination from surrounding neuropil, were excluded, as were elongated events along dendritic shafts rather than confined to spines. The pipeline then converts raw fluorescence into normalized signals, corrects for photobleaching using an adaptive baseline algorithm borrowed from spectroscopy, and detects calcium transients with a threshold set at 4.5 standard deviations above the baseline noise level.

The results are striking in scale. Across their experiments, the team analyzed 1,198,748 spontaneously active synapses and detected 10,228,773 unique synaptic calcium transients, all without a single manual region-of-interest selection. Individual fields of view contained between 1,700 and 3,300 synapses, and complete datasets ranged from 150,000 to 340,000 synapses. Validation experiments showed the automated approach achieved a false negative rate of just 3.7 percent and a false positive rate of 0.3 percent, figures the authors note can be tuned by adjusting filtering criteria. For statistics on such massive hierarchical datasets, the team employed generalized linear mixed-effects models, cross-checked against bootstrap resampling of 10,000 iterations.

To prove the pipeline could detect known biology, the researchers put it through a battery of pharmacological tests. Applying phorbol 12,13-dibutyrate, a compound that boosts presynaptic vesicle release through Munc13 and protein kinase C, increased synaptic event frequency more than twofold without changing the number of active synapses. Subsequent application of APV, a competitive NMDA receptor antagonist, collapsed the number of active synapses roughly sevenfold and cut event frequency to about two-thirds of control levels, confirming that the transients are genuinely NMDAR-dependent. Conversely, glycine, a co-agonist of the NMDA receptor, increased both event frequency by 55 percent and the number of active synapses, while the AMPA receptor antagonist NBQX left frequency unchanged but paradoxically increased transient amplitude, a finding the authors attribute to crosstalk between AMPA and NMDA receptors.

The pipeline also captured clinically relevant drug effects with unusual temporal resolution. Acute application of ketamine, an anesthetic and rapid-acting antidepressant under investigation for depression, PTSD, and pain, reduced the number of active synapses from about 2,179 to 1,769, while memantine, an extrasynaptic NMDA receptor antagonist used to treat Alzheimer’s disease symptoms, cut the count from roughly 2,376 to 1,427. Because the system images multiple regions in sequence, the researchers could track how these antagonists progressively silence synapses over 30 minutes, consistent with the fact that both drugs block NMDA receptors only after channels open. Overnight ketamine treatment reduced both event frequency and amplitude in a concentration-dependent manner, while overnight memantine reduced amplitude but left frequency largely intact.

Perhaps most compelling is what the pipeline revealed about autoimmune disease. The team tested three patient-derived NMDA receptor autoantibodies from people with anti-NMDAR encephalitis, a devastating condition estimated to affect one to five people per million each year and often misdiagnosed as primary psychiatric illness. Two of the antibodies, 003-102 and 008-218, significantly reduced synaptic event frequency, amplitude, and the number of active synapses, while the third, 007-124, which has the lowest binding affinity, had no measurable effect. Notably, antibody 008-218 suppressed synapse number more strongly than 003-102 despite similar binding affinities, likely because the two antibodies disable receptors through different mechanisms, one driving internalization and the other locking channels into inactive conformations. Such functional distinctions would be invisible in fixed-tissue immunostaining.

The authors calculated Z-prime scores, a standard metric of assay quality in drug screening, and found values between 0 and 0.5 for several parameters at the well level and above 0.5 for frequency and active synapse number at the replicate level, placing the assay firmly in the range usable for cell-based high-throughput screens. They envision the platform as a secondary medium-throughput screen for discovering compounds that protect synapse function, and suggest it could eventually be applied directly to patient cerebrospinal fluid samples to guide antibody-depleting therapies. With the analysis code openly available on GitHub, the pipeline opens the door to studying individual synapses at a scale never before realized, promising new insights into synaptic plasticity, disease mechanisms, and the search for drugs that keep these vital junctions working.

Subject of Research: An automated calcium imaging pipeline for high-throughput quantification of synaptic function in cultured neurons

Article Title: Automated ROI detection allows rapid quantification of synaptic activity across tens of thousands of synapses in cell culture

Article References: Automated ROI detection allows rapid quantification of synaptic activity across tens of thousands of synapses in cell culture. (n.d.). https://doi.org/10.3389/fnsyn.2026.1832103

Image Credits: AI Generated

DOI: 10.3389/fnsyn.2026.1832103

Keywords: calcium imaging, synapses, NMDA receptors, Suite2p, automated ROI detection, drug screening, autoantibodies, encephalitis, ketamine, memantine, presynaptic release, neurodegeneration