A deep learning-based approach to model anomalous diffusion of membrane proteins: the case of the nicotinic acetylcholine receptor.

We present a concatenated deep-learning multiple neural network system for the analysis of single-molecule trajectories. We apply this machine learning-based analysis to characterize the translational diffusion of the nicotinic acetylcholine receptor at the plasma membrane, experimentally interrogated using superresolution optical microscopy. The receptor protein displays a heterogeneous diffusion behavior that goes beyond the ensemble level, with individual trajectories exhibiting more than one diffusive state, requiring the optimization of the neural networks through a hyperparameter analysis for different numbers of steps and durations, especially for short trajectories (<50 steps) where the accuracy of the models is most sensitive to localization errors. We next use the statistical models to test for Brownian, continuous-time random walk and fractional Brownian motion, and introduce and implement an additional, two-state model combining Brownian walks and obstructed diffusion mechanisms, enabling us to partition the two-state trajectories into segments, each of which is independently subjected to multiple analysis. The concatenated multi-network system evaluates and selects those physical models that most accurately describe the receptor's translational diffusion. We show that the two-state Brownian-obstructed diffusion model can account for the experimentally observed anomalous diffusion (mostly subdiffusive) of the population and the heterogeneous single-molecule behavior, accurately describing the majority (72.5 to 88.7% for α-bungarotoxin-labeled receptor and between 73.5 and 90.3% for antibody-labeled molecules) of the experimentally observed trajectories, with only ~15% of the trajectories fitting to the fractional Brownian motion model.

Carlos Gutiérrez-Merino: Synergy of Theory and Experimentation in Biological Membrane Research.

Professor Carlos Gutiérrez-Merino, a prominent scientist working in the complex realm of biological membranes, has made significant theoretical and experimental contributions to the field. Contemporaneous with the development of the fluid-mosaic model of Singer and Nicolson, the Förster resonance energy transfer (FRET) approach has become an invaluable tool for studying molecular interactions in membranes, providing structural insights on a scale of 1-10 nm and remaining important alongside evolving perspectives on membrane structures. In the last few decades, Gutiérrez-Merino's work has covered multiple facets in the field of FRET, with his contributions producing significant advances in quantitative membrane biology. His more recent experimental work expanded the ground concepts of FRET to high-resolution cell imaging. Commencing in the late 1980s, a series of collaborations between Gutiérrez-Merino and the authors involved research visits and joint investigations focused on the nicotinic acetylcholine receptor and its relation to membrane lipids, fostering a lasting friendship.

Insulin resistance induced by long-term hyperinsulinemia abolishes the effects of acute insulin exposure on cell-surface nicotinic acetylcholine receptor levels and actin cytoskeleton morphology.

Using CHO-K1/A5 cells, a clonal cell line that robustly expresses adult muscle-type nicotinic acetylcholine receptor (nAChR), we explored whether insulin resistance in these mammalian cells affects cell-surface expression of the nAChR, its endocytic internalization, and actin cytoskeleton integrity. Acute nanomolar insulin stimulation resulted in a slow increase in nAChR cell-surface levels, reaching maximum levels at ∼1 h. Long periods of insulin incubation caused CHO-K1/A5 cells to become insulin resistant, as previously observed with several other cell types. Furthermore, long-term insulin treatment abolished the effects of short-term insulin exposure on cell-surface nAChR levels, suggestive of a desensitization phenomenon. It also affected the kinetics of ligand-induced nAChR internalization. Since the integrity of the cortical actin cytoskeleton affects nAChR endocytosis, we also studied the effects of long-term insulin treatment on this meshwork. We found that it significantly affected the cortical actin morphology of CHO-K1/A5 cells and the response of the actin cytoskeleton to a subsequent short-term insulin stimulus. Overall, the present results show for the first time the effects of insulin signaling on cell-surface nAChR expression and actin cytoskeleton-associated internalization.

Apolipoprotein E4 heterologous expression, purification under non-denaturing conditions, and effects on neuronal clonal cell lines.

The ε4 allele of the apolipoprotein E gene (APOE4) constitutes the main genetic risk factor for late-onset Alzheimer disease (AD). High amounts of pure apolipoprotein E4 (ApoE4), in a rapid and reproducible fashion, could be of value for studying its pathophysiological roles in AD. The aim of the present work was to optimize a preparative method to obtain highly purified recombinant ApoE4 (rApoE4) with full biological activity. rApoE4 was expressed in the E. Coli BL21(D3) strain and a soluble form of the protein was purified by a combination of affinity and size-exclusion chromatography that precluded a denaturation step. The structural integrity and the biochemical activity of the purified rApoE4 were confirmed by circular dichroism and a lipid-binding assay. Several biological parameters affected by rApoE4, such as mitochondrial morphology, mitochondrial membrane potential and reactive oxygen species production were studied in CNh cells, a neuronal cell line, and neurodifferentiation and dendritogenesis were analyzed in the SH-SY5Y neuroblastoma cell line. The improved rApoE4 purification technique reported here enables the production of highly purified protein that retain the structural properties and functional activity of the native protein, as confirmed by tests in two different neuronal cell lines in culture.

Structure and function meet at the nicotinic acetylcholine receptor-lipid interface.

The nicotinic acetylcholine receptor (nAChR) is a transmembrane protein that mediates fast intercellular communication in response to the endogenous neurotransmitter acetylcholine. It is the best characterized and archetypal molecule in the superfamily of pentameric ligand-gated ion channels (pLGICs). As a typical transmembrane macromolecule, it interacts extensively with its vicinal lipid microenvironment. Experimental evidence provides a wealth of information on receptor-lipid crosstalk: the nAChR exerts influence on its immediate membrane environment and conversely, the lipid moiety modulates ligand binding, affinity state transitions and gating of ion translocation functions of the receptor protein. Recent cryogenic electron microscopy (cryo-EM) studies have unveiled the occurrence of sites for phospholipids and cholesterol on the lipid-exposed regions of neuronal and electroplax nAChRs, confirming early spectroscopic and affinity labeling studies demonstrating the close contact of lipid molecules with the receptor transmembrane segments. This new data provides structural support to the postulated "lipid sensor" ability displayed by the outer ring of M4 transmembrane domains and their modulatory role on nAChR function, as we postulated a decade ago. Borrowing from the best characterized nAChR, the electroplax (muscle-type) receptor, and exploiting new structural information on the neuronal nAChR, it is now possible to achieve an improved depiction of these sites. In combination with site-directed mutagenesis, single-channel electrophysiology, and molecular dynamics studies, the new structural information delivers a more comprehensive portrayal of these lipid-sensitive loci, providing mechanistic explanations for their ability to modulate nAChR properties and raising the possibility of targetting them in disease.

Fluorescence Studies of Nicotinic Acetylcholine Receptor and Its Associated Lipid Milieu: The Influence of Erwin London's Methodological Approaches.

Erwin London dedicated considerable effort to understanding lipid interactions with membrane-resident proteins and how these interactions shaped the formation and maintenance of lipid phases and domains. In this endeavor, he developed ad hoc techniques that greatly contributed to advancements in the field. We have employed and/or modified/extended some of his methodological approaches and applied them to investigate lipid interaction with the nicotinic acetylcholine receptor (nAChR) protein, the paradigm member of the superfamily of rapid pentameric ligand-gated ion channels (pLGIC). Our experimental systems ranged from purified receptor protein reconstituted into synthetic lipid membranes having known effects on receptor function, to cellular systems subjected to modification of their lipid content, e.g., varying cholesterol levels. We have often employed fluorescence techniques, including fluorescence quenching of diphenylhexatriene (DPH) extrinsic fluorescence and of nAChR intrinsic fluorescence by nitroxide spin-labeled phospholipids, DPH anisotropy, excimer formation of pyrene-phosphatidylcholine, and Förster resonance energy transfer (FRET) from the protein moiety to the extrinsic probes Laurdan, DPH, or pyrene-phospholipid to characterize various biophysical properties of lipid-receptor interactions. Some of these strategies are revisited in this review. Special attention is devoted to the anionic phospholipid phosphatidic acid (PA), which stabilizes the functional resting form of the nAChR. The receptor protein was shown to organize its PA-containing immediate microenvironment into microdomains with high lateral packing density and rigidity. PA and cholesterol appear to compete for the same binding sites on the nAChR protein.

Mapping the Nicotinic Acetylcholine Receptor Nanocluster Topography at the Cell Membrane with STED and STORM Nanoscopies.

The cell-surface topography and density of nicotinic acetylcholine receptors (nAChRs) play a key functional role in the synapse. Here we employ in parallel two labeling and two super-resolution microscopy strategies to characterize the distribution of this receptor at the plasma membrane of the mammalian clonal cell line CHO-K1/A5. Cells were interrogated with two targeted techniques (confocal microscopy and stimulated emission depletion (STED) nanoscopy) and single-molecule nanoscopy (stochastic optical reconstruction microscopy, STORM) using the same fluorophore, Alexa Fluor 647, tagged onto either α-bungarotoxin (BTX) or the monoclonal antibody mAb35. Analysis of the topography of nanometer-sized aggregates ("nanoclusters") was carried out using STORMGraph, a quantitative clustering analysis for single-molecule localization microscopy based on graph theory and community detection, and ASTRICS, an inter-cluster similarity algorithm based on computational geometry. Antibody-induced crosslinking of receptors resulted in nanoclusters with a larger number of receptor molecules and higher densities than those observed in BTX-labeled samples. STORM and STED provided complementary information, STED rendering a direct map of the mesoscale nAChR distribution at distances ~10-times larger than the nanocluster centroid distances measured in STORM samples. By applying photon threshold filtering analysis, we show that it is also possible to detect the mesoscale organization in STORM images.

Cholesterol in myasthenia gravis.

The cholinergic neuromuscular junction is the paradigm peripheral synapse between a motor neuron nerve ending and a skeletal muscle fiber. In vertebrates, acetylcholine is released from the presynaptic site and binds to the nicotinic acetylcholine receptor at the postsynaptic membrane. A variety of pathologies among which myasthenia gravis stands out can impact on this rapid and efficient signaling mechanism, including autoimmune diseases affecting the nicotinic receptor or other synaptic proteins. Cholesterol is an essential component of biomembranes and is particularly rich at the postsynaptic membrane, where it interacts with and modulates many properties of the nicotinic receptor. The profound changes inflicted by myasthenia gravis on the postsynaptic membrane necessarily involve cholesterol. This review analyzes some aspects of myasthenia gravis pathophysiology and associated postsynaptic membrane dysfunction, including dysregulation of cholesterol metabolism in the myocyte brought about by antibody-receptor interactions. In addition, given the extensive therapeutic use of statins as the typical cholesterol-lowering drugs, we discuss their effects on skeletal muscle and the possible implications for MG patients under chronic treatment with this type of compound.

Fluorescence sensors for imaging membrane lipid domains and cholesterol.

Lipid membrane domains are supramolecular lateral heterogeneities of biological membranes. Of nanoscopic dimensions, they constitute specialized hubs used by the cell as transient signaling platforms for a great variety of biologically important mechanisms. Their property to form and dissolve in the bulk lipid bilayer endow them with the ability to engage in highly dynamic processes, and temporarily recruit subpopulations of membrane proteins in reduced nanometric compartments that can coalesce to form larger mesoscale assemblies. Cholesterol is an essential component of these lipid domains; its unique molecular structure is suitable for interacting intricately with crevices and cavities of transmembrane protein surfaces through its rough ß face while "talking" to fatty acid acyl chains of glycerophospholipids and sphingolipids via its smooth α face. Progress in the field of membrane domains has been closely associated with innovative improvements in fluorescence microscopy and new fluorescence sensors. These advances enabled the exploration of the biophysical properties of lipids and their supramolecular platforms. Here I review the rationale behind the use of biosensors over the last few decades and their contributions towards elucidation of the in-plane and transbilayer topography of cholesterol-enriched lipid domains and their molecular constituents. The challenges introduced by super-resolution optical microscopy are discussed, as well as possible scenarios for future developments in the field, including virtual ("no staining") staining.

Antibody-induced crosslinking and cholesterol-sensitive, anomalous diffusion of nicotinic acetylcholine receptors.

Synaptic strength depends on the number of cell-surface neurotransmitter receptors in dynamic equilibrium with intracellular pools. Dysregulation of this homeostatic balance occurs, for example in myasthenia gravis, an autoimmune disease characterized by a decrease in the number of postsynaptic nicotinic acetylcholine receptors (nAChRs). Monoclonal antibody mAb35 mimics this effect. Here we use STORM nanoscopy to characterize the individual and ensemble dynamics of monoclonal antibody-crosslinked receptors in the clonal cell line CHO-K1/A5, which robustly expresses adult muscle-type nAChRs. Antibody labeling of live cells results in 80% receptor immobilization. The remaining mobile fraction exhibits a heterogeneous combination of Brownian and anomalous diffusion. Single-molecule trajectories exhibit a two-state switching behavior between free Brownian walks and anticorrelated walks within confinement areas. The latter act as permeable fences (~34 nm radius, ~400 ms lifetime). Dynamic clustering, trapping, and immobilization also occur in larger nanocluster zones (120-180 nm radius) with longer lifetimes (11 ± 1 s), in a strongly cholesterol-sensitive manner. Cholesterol depletion increases the size of the clustering phenomenon; cholesterol enrichment has the opposite effect. The disclosed high proportion of monoclonal antibody-crosslinked immobile receptors, together with their anomalous, cholesterol-sensitive diffusion and clustering, provides new insights into the antibody-enhanced antigenic modulation that leads to physiopathological internalization and degradation of receptors in myasthenia.

Lovastatin Differentially Regulates α7 and α4 Neuronal Nicotinic Acetylcholine Receptor Levels in Rat Hippocampal Neurons.

Neuronal α7 and α4ß2 are the predominant nicotinic acetylcholine receptor (nAChR) subtypes found in the brain, particularly in the hippocampus. The effects of lovastatin, an inhibitor of cholesterol biosynthesis, on these two nAChRs endogenously expressed in rat hippocampal neuronal cells were evaluated in the 0.01-1 µM range. Chronic (14 days) lovastatin treatment augmented cell-surface levels of α7 and α4 nAChRs, as measured by fluorescence microscopy and radioactive ligand binding assays. This was accompanied in both cases by an increase in total protein receptor levels as determined by Western blots. At low lovastatin concentrations (10-100 nM), the increase in α4 nAChR in neurites was higher than in neuronal cell somata; the opposite occurred at higher (0.5-1 µM) lovastatin concentrations. In contrast, neurite α7 nAChRs raised more than somatic α7 nAChRs at all lovastatin concentrations tested. These results indicate that cholesterol levels homeostatically regulate α7 and α4 nAChR levels in a differential manner through mechanisms that depend on statin concentration and receptor localization. The neuroprotective pleomorphic effects of statins may act by reestablishing the homeostatic equilibrium.

Cholesterol-Recognition Motifs in Membrane Proteins.

The impact of cholesterol on the structure and function of membrane proteins was recognized several decades ago, but the molecular mechanisms underlying these effects have remained elusive. There appear to be multiple mechanisms by which cholesterol interacts with proteins. A complete understanding of cholesterol-sensing motifs is still undergoing refinement. Initially, cholesterol was thought to exert only non-specific effects on membrane fluidity. It was later shown that this lipid could specifically interact with membrane proteins and affect both their structure and function. In this article, we have summarized and critically analyzed our evolving understanding of the affinity, specificity and stereoselectivity of the interactions of cholesterol with membrane proteins. We review the different computational approaches that are currently used to identify cholesterol binding sites in membrane proteins and the biochemical logic that governs each type of site, including CRAC, CARC, SSD and amphipathic helix motifs. There are physiological implications of these cholesterol-recognition motifs for G-protein coupled receptors (GPCR) and ion channels, in membrane trafficking and membrane fusion (SNARE) proteins. There are also pathological implications of cholesterol binding to proteins involved in neurological disorders (Alzheimer, Parkinson, Creutzfeldt-Jakob) and HIV fusion. In each case, our discussion is focused on the key molecular aspects of the cholesterol and amino acid motifs in membrane-embedded regions of membrane proteins that define the physiologically relevant crosstalk between the two. Our understanding of the factors that determine if these motifs are functional in cholesterol binding will allow us enhanced predictive capabilities.

The lipid habitats of neurotransmitter receptors in brain.

Neurotransmitter receptors, the macromolecules specialized in decoding the chemical signals encrypted in the chemical signaling mechanism in the nervous system, occur either at the somatic cell surface of chemically excitable cells or at specialized subcellular structures, the synapses. Synapses have lipid compositions distinct from the rest of the cell membrane, suggesting that neurotransmitter receptors and their scaffolding and adaptor protein partners require specific lipid habitats for optimal operation. In this review we discuss some paradigmatic cases of neurotransmitter receptor-lipid interactions, highlighting the chemical nature of the intervening lipid species and providing examples of the receptor mechanisms affected by interaction with lipids. The focus is on the effects of cholesterol, glycerophospholipids and covalent fatty acid acylation on neurotransmitter receptors. We also briefly discuss the role of lipid phase states involving lateral heterogeneities of the host membrane known to modulate membrane transport, protein sorting and signaling. Modulation of neurotransmitter receptors by lipids occurs at multiple levels, affecting a wide span of activities including their trafficking, sorting, stability, residence lifetime at the cell surface, endocytosis, and recycling, among other important functional properties at the synapse.

Relevance of CARC and CRAC Cholesterol-Recognition Motifs in the Nicotinic Acetylcholine Receptor and Other Membrane-Bound Receptors.

Cholesterol is a ubiquitous neutral lipid, which finely tunes the activity of a wide range of membrane proteins, including neurotransmitter and hormone receptors and ion channels. Given the scarcity of available X-ray crystallographic structures and the even fewer in which cholesterol sites have been directly visualized, application of in silico computational methods remains a valid alternative for the detection and thermodynamic characterization of cholesterol-specific sites in functionally important membrane proteins. The membrane-embedded segments of the paradigm neurotransmitter receptor for acetylcholine display a series of cholesterol consensus domains (which we have coined "CARC"). The CARC motif exhibits a preference for the outer membrane leaflet and its mirror motif, CRAC, for the inner one. Some membrane proteins possess the double CARC-CRAC sequences within the same transmembrane domain. In addition to in silico molecular modeling, the affinity, concentration dependence, and specificity of the cholesterol-recognition motif-protein interaction have recently found experimental validation in other biophysical approaches like monolayer techniques and nuclear magnetic resonance spectroscopy. From the combined studies, it becomes apparent that the CARC motif is now more firmly established as a high-affinity cholesterol-binding domain for membrane-bound receptors and remarkably conserved along phylogenetic evolution.

Phylogenetic conservation of protein-lipid motifs in pentameric ligand-gated ion channels.

Using the crosstalk between the nicotinic acetylcholine receptor (nAChR) and its lipid microenvironment as a paradigm, this short overview analyzes the occurrence of structural motifs which appear not only to be conserved within the nAChR family and contemporary eukaryotic members of the pentameric ligand-gated ion channel (pLGIC) superfamily, but also extend to prokaryotic homologues found in bacteria. The evolutionarily conserved design is manifested in: 1) the concentric three-ring architecture of the transmembrane region, 2) the occurrence in this region of distinct lipid consensus motifs in prokaryotic and eukaryotic pLGIC and 3) the key participation of the outer TM4 ring in conveying the influence of the lipid membrane environment to the middle TM1-TM3 ring and this, in turn, to the inner TM2 channel-lining ring, which determines the ion selectivity of the channel. The preservation of these constant structural-functional features throughout such a long phylogenetic span likely points to the successful gain-of-function conferred by their early acquisition. This article is part of a Special Issue entitled: Lipid-protein interactions.

Transbilayer asymmetry and sphingomyelin composition modulate the preferential membrane partitioning of the nicotinic acetylcholine receptor in Lo domains.

We have previously shown that the intact nicotinic acetylcholine receptor (AChR) lacks preference for Lo domains when reconstituted in a sphingomyelin (SM), cholesterol (Chol) and POPC (1:1:1) model system (Bermúdez V, Antollini SS, Fernández-Nievas GA, Aveldaño MI, Barrantes FJ. J. Lipid Res. 2010; 51: 2629-2641). Here, we have furthered our studies by characterizing the influence of different lipid host compositions on the distribution of purified AChR reconstituted in two model systems (POPC:Chol, 1:1 and POPC:Chol:SM, 1:1:1), involving a) different SM species (porcine brain SM (bSM), 16:0-SM, 18:0-SM or 24:1-SM); or b) induced transbilayer asymmetry, resulting from enrichment in bSM in the external hemilayer. AChR distribution was evaluated by fluorescence resonance energy transfer efficiency between the AChR intrinsic fluorescence and Laurdan or dehydroergosterol fluorescence, and by analyzing the distribution of AChR in detergent-resistant and detergent-soluble fractions (1% Triton X-100, 4 °C). bSM-induced transbilayer asymmetry or the presence of 16:0-SM and/or 18:0-SM (unlike bSM or 24:1-SM) resulted in the preferential partitioning of AChR in Lo domains, suggesting that the localization of AChR in ordered domains strongly depends on the characteristics of the host lipid membrane, and in particular on the sphingolipid composition and transbilayer asymmetry.

Prenatal restraint stress decreases the expression of alpha-7 nicotinic receptor in the brain of adult rat offspring.

Prenatal stress (PS) strongly impacts fetal brain development and function in adulthood. In normal aging and Alzheimer's disease, there is hypothalamic-pituitary-adrenal axis dysfunction and loss of cholinergic neurons and neuronal nicotinic acetylcholine receptors (nAChRs). This study investigated whether prenatal restraint stress affects nAChR expression in the brain of adult offspring. For PS, pregnant dams were placed in a plastic restrainer for 45 min, three times daily during the last week of pregnancy; controls were undisturbed. Male offspring were analyzed at postnatal day (PND) 60 (n = 4 rats per group). Western blot (WB) and fluorescence microscopy showed that PS decreased α7-AChR subunit expression (∼50%) in the frontal cortex in the adult offspring. PS decreased significantly the number of α7-AChR-expressing cells in the medial prefrontal cortex (by ∼25%) and in the sensory-motor cortex (by ∼20%) without affecting the total cell number in those areas. No alterations were found in the hippocampus by quantitative polymerase chain reaction (qPCR), or WB analysis, but a detailed fluorescence microscopy analysis showed that PS affected α7-AChR mainly in the CA3 and dentate gyrus subfields: PS decreased α7-AChR subunit expression by ∼25 and ∼30%, respectively. Importantly, PS decreased the number of α7-AChR-expressing cells and the total cell number (by ∼15 and 20%, respectively) in the dentate gyrus. PS differently affected α4-AChR: PS impaired its mRNA expression in the frontal cortex (by ∼50%), without affecting protein levels. These results demonstrate that disturbances during gestation produce long-term alterations in the expression pattern of α7-AChR in rat brain.

Cognitive synaptopathy: synaptic and dendritic spine dysfunction in age-related cognitive disorders.

Cognitive impairment is a leading component of several neurodegenerative and neurodevelopmental diseases, profoundly impacting on the individual, the family, and society at large. Cognitive pathologies are driven by a multiplicity of factors, from genetic mutations and genetic risk factors, neurotransmitter-associated dysfunction, abnormal connectomics at the level of local neuronal circuits and broader brain networks, to environmental influences able to modulate some of the endogenous factors. Otherwise healthy older adults can be expected to experience some degree of mild cognitive impairment, some of which fall into the category of subjective cognitive deficits in clinical practice, while many neurodevelopmental and neurodegenerative diseases course with more profound alterations of cognition, particularly within the spectrum of the dementias. Our knowledge of the underlying neuropathological mechanisms at the root of this ample palette of clinical entities is far from complete. This review looks at current knowledge on synaptic modifications in the context of cognitive function along healthy ageing and cognitive dysfunction in disease, providing insight into differential diagnostic elements in the wide range of synapse alterations, from those associated with the mild cognitive changes of physiological senescence to the more profound abnormalities occurring at advanced clinical stages of dementia. I propose the term "cognitive synaptopathy" to encompass the wide spectrum of synaptic pathologies associated with higher brain function disorders.

A supervised graph-based deep learning algorithm to detect and quantify clustered particles.

Considerable efforts are currently being devoted to characterizing the topography of membrane-embedded proteins using combinations of biophysical and numerical analytical approaches. In this work, we present an end-to-end (i.e., human intervention-independent) algorithm consisting of two concatenated binary Graph Neural Network (GNNs) classifiers with the aim of detecting and quantifying dynamic clustering of particles. As the algorithm only needs simulated data to train the GNNs, it is parameter-independent. The GNN-based algorithm is first tested on datasets based on simulated, albeit biologically realistic data, and validated on actual fluorescence microscopy experimental data. Application of the new GNN method is shown to be faster than other currently used approaches for high-dimensional SMLM datasets, with the additional advantage that it can be implemented on standard desktop computers. Furthermore, GNN models obtained via training procedures are reusable. To the best of our knowledge, this is the first application of GNN-based approaches to the analysis of particle aggregation, with potential applications to the study of nanoscopic particles like the nanoclusters of membrane-associated proteins in live cells.