The core challenge that MemBrain v2 addresses begins with segmentation. A cryo-electron tomogram is essentially a three-dimensional reconstruction assembled from a tilt series of two-dimensional projection images of a vitrified sample. Within that volumetric data, membranes appear as weak, noisy, curving sheets only a few nanometers thick, often distorted by the so-called missing wedge artifact that arises from physical limits on how far the sample can be tilted in the microscope. Early approaches to membrane segmentation relied on hand-crafted filters, edge detectors, and painstaking manual contouring, a process that could consume days of an expert’s time for a single tomogram and that varied substantially from one analyst to another. Such variability made large-scale, quantitative comparisons across cells, conditions, or organisms extremely difficult, effectively capping the throughput of the field.

MemBrain v2 builds on the deep-learning foundations laid by its predecessor, which introduced convolutional neural network models trained to recognize membrane density patterns that human eyes and classical algorithms routinely miss. The new version extends this philosophy across the entire analysis pipeline. Rather than providing a single segmentation model that researchers must then stitch into their own homemade workflows, the software packages the full sequence of steps: preprocessing of raw tomograms, automated segmentation of membranes, extraction of continuous membrane meshes from the segmented volumes, localization of membrane proteins relative to those meshes, and quantitative spatial statistics describing how proteins and curvature features are organized across the membrane surface. The result is a pipeline in which a user can move from a reconstructed tomogram to publication-ready membrane measurements with minimal manual intervention.

Technically, the segmentation engine in MemBrain v2 employs neural networks trained on curated, manually annotated datasets of tomograms spanning diverse biological contexts, from isolated organelles to intact cells prepared by focused ion beam milling. The models operate on three-dimensional patches of the tomogram, predicting per-voxel probabilities that a given location belongs to a membrane. Crucially, the developers have placed strong emphasis on generalization: the goal is that a network trained on one set of samples and imaging conditions will still perform reliably on tomograms acquired on a different microscope, with a different detector, or from a different biological system. This kind of cross-dataset robustness has been a persistent pain point in cryo-ET machine learning, where models frequently overfit to the specific noise characteristics and contrast regimes of their training data. Performance across heterogeneous datasets was therefore treated as a first-class evaluation criterion rather than an afterthought.

Once membranes are segmented, the software converts the voxel-level predictions into geometrically clean surface meshes, a step that is far less trivial than it sounds. Segmentation outputs are typically noisy at the pixel scale, containing holes, spurious protrusions, and disconnected fragments that would corrupt any downstream measurement of surface area, curvature, or protein spacing. MemBrain v2 incorporates methods to extract smooth, watertight representations of membrane surfaces, preserving fine topological features such as tight junctions, cristae junctions, organelle contact sites, and highly curved tubular structures. This geometric fidelity matters because many of the most interesting biological questions in modern cell biology hinge precisely on such features: how the folded inner membrane of a mitochondrion organizes its respiratory machinery, how the endoplasmic reticulum generates curved hubs that recruit certain proteins, or how viral replication compartments remodel host membranes into convoluted replication factories.

Perhaps the most consequential addition in version 2 is the integration of membrane protein localization into the same workflow. Subtomogram averaging has long allowed structural biologists to determine high-resolution structures of abundant, repetitive complexes, but mapping the broader landscape of membrane-associated proteins in situ, including those that are sparse, irregularly arranged, or conformationally heterogeneous, remains a formidable task. MemBrain v2 provides tools to detect and classify particles in the vicinity of segmented membranes, project their positions onto the extracted meshes, and compute spatial statistics that characterize their distribution: clustering tendencies, exclusion patterns, correlations with local membrane curvature, and nearest-neighbor relationships. By unifying segmentation and particle analysis in one software environment, the tool removes the error-prone handoffs between disconnected programs that have historically plagued multi-step cryo-ET analyses.

The significance of this integration extends well beyond convenience. Spatial organization of proteins on membranes is now understood to be a central regulatory principle in cell biology. The clustering of receptors can amplify signaling; the sorting of complexes into curvature-sensing domains can drive vesicle budding; the exclusion of certain proteins from contact sites can maintain organelle identity. Quantifying these patterns rigorously requires exactly the kind of joint membrane-and-protein analysis that MemBrain v2 enables. Instead of qualitative statements that a protein appears enriched on curved regions, researchers can now report statistically grounded measurements of enrichment, complete with the geometric context of the underlying membrane. This shift from descriptive to quantitative in situ analysis aligns cryo-electron tomography with the standards of rigor that fields such as light-sheet microscopy and single-cell genomics achieved years ago.

Accessibility has been a guiding concern throughout the design. Cryo-ET is a technique that sits at the intersection of biology, physics, and computer science, and many research groups possess deep biological questions but limited computational infrastructure. Earlier generations of analysis tools often assumed considerable programming expertise, requiring users to assemble pipelines from research code written by other laboratories, with little documentation and no guarantee of stability between versions. MemBrain v2 counters this with user-friendly interfaces, containerized installation options, thorough documentation, and standardized output formats designed to interoperate with the broader ecosystem of tomography software, including visualization platforms and subtomogram averaging packages. The developers have also made the trained models and processing tools openly available, consistent with a broader movement in structural biology toward open, reproducible computational methods that any laboratory can adopt, inspect, and extend.

The timing of such a tool is not accidental. Cryo-electron tomography is undergoing explosive growth, propelled by advances in direct electron detectors, automated data collection, focused ion beam milling of lamellae, and reconstruction algorithms that dramatically improve resolution and contrast. Large-scale initiatives now aim to image thousands of cellular volumes, generating data at a rate that makes manual analysis categorically impossible. In this environment, automated, reliable, generalizable analysis pipelines are not merely convenient; they are the precondition for the field’s ambition to build a quantitative atlas of cellular architecture. Tools like MemBrain v2 represent the computational counterpart to the hardware revolution, ensuring that the flood of tomographic data can be converted into biological insight rather than accumulating as unprocessed archives.

For the practicing researcher, the practical implications are immediate. A laboratory studying mitochondrial remodeling during stress, for example, could apply the pipeline across dozens of tomograms of control and treated cells, obtain consistent membrane segmentations, quantify changes in cristae morphology, map the redistribution of respiratory chain complexes, and test whether protein reorganization tracks with local membrane curvature, all within a unified framework. A virology group could chart how viral proteins reshape intracellular membranes as replication compartments mature. A neurobiologist could examine the nanoscale organization of synaptic vesicle clusters and presynaptic membrane specializations. In each case, the same underlying machinery of segmentation, mesh extraction, particle localization, and spatial statistics does the heavy lifting, freeing scientists to focus on experimental design and biological interpretation rather than on debugging bespoke image-analysis scripts.

As with any computational method, caveats remain. Deep-learning segmentations are probabilistic and can fail in unexpected regimes, particularly with extreme imaging conditions, unusual staining or labeling approaches, or membrane-like features such as dense protein coats that mimic lipid bilayers. Responsible use therefore still entails expert validation, and the developers emphasize that the tool is intended to accelerate and standardize expert workflows rather than eliminate judgment from the loop. Nevertheless, by demonstrating robust performance across diverse datasets and by wrapping the entire analysis chain into a coherent, accessible package, MemBrain v2 marks a meaningful step toward making the quantitative analysis of cellular membranes as routine as the imaging itself. If the trajectory of cryo-electron tomography continues on its current course, tools of this kind will become the standard infrastructure on which the next decade of in situ structural biology is built, transforming noisy three-dimensional images into a rigorous, comparative anatomy of the cell’s molecular geography.

Subject of Research: A deep-learning software pipeline for membrane segmentation and membrane protein spatial analysis in cryo-electron tomography

Article Title: MemBrain v2: an end-to-end tool for the analysis of membranes in cryo-electron tomography

Article References: Lamm, L., Zufferey, S., Zhang, H., Righetto, R. D., Waltz, F., Wietrzynski, W., Yamauchi, K. A., Burt, A., Liu, Y., Martinez-Sanchez, A., Ziegler, S., Isensee, F., Schnabel, J. A., Engel, B. D., & Peng, T. (2026). MemBrain v2: an end-to-end tool for the analysis of membranes in cryo-electron tomography. Nature Methods. https://doi.org/10.1038/s41592-026-03178-8

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03178-8

Keywords: cryo-electron tomography, MemBrain v2, membrane segmentation, deep learning, membrane proteins, spatial analysis, structural biology, cell biology, organelle architecture, image analysis, subtomogram averaging, Nature Methods

The article, led by Carsten Janke of Institut Curie and Université Paris-Saclay together with Anna Akhmanova of Utrecht University and an international consortium of sixteen colleagues from institutions spanning Europe and the United States, lays out a coordinated agenda for the microtubule field. Its authors span structural biology, biophysics, cell biology, neuroscience, parasitology and computational modelling, reflecting their central conviction that progress will come not from any single discipline but from a deliberate fusion of them. The paper is punctuated by the admission that some of the most basic questions remain unexplored, including how the properties and functions of microtubules are affected by the chemical marks known as tubulin post-translational modifications, by disease-related mutations in tubulin genes, or by variation in the microtubule lattice itself.

Microtubules are hollow tubes assembled from αβ-tubulin dimers that stack into protofilaments, typically thirteen of which align laterally to form the wall of the polymer. What makes these structures extraordinary is their behaviour: individual microtubules stochastically switch between phases of growth and rapid shrinkage, a phenomenon called dynamic instability, which was first described by Tim Mitchison and Marc Kirschner in 1984. A stabilizing cap of GTP-bound tubulin at the growing plus end controls this behaviour, and its loss triggers catastrophic depolymerization. Beyond pure polymerization, microtubules can self-repair by incorporating new tubulin along damaged stretches of lattice, a property revealed in studies showing that lattice defects actually induce self-renewal and protect the polymers from destruction by molecular motors. These discoveries have transformed the view of microtubules from static beams into dynamic, self-healing machines.

The Roadmap emphasizes that microtubule function is not determined by the polymer alone but by a dense layer of regulation encoded on the tubulin subunit itself. Enzymes of the tubulin tyrosine ligase-like family decorate the unstructured carboxy-terminal tails of tubulin with modifications such as tyrosination, detyrosination, acetylation, glutamylation and glycylation, creating what has become known as the tubulin code. Recent structural work has shown, for example, that enzymes involved in polyglutamylation recognize microtubules through a quadrivalent mechanism that links microtubule geometry to the generation of localized modification patterns, and that protofilament-specific nanopatterns of these marks tune the mechanics of beating cilia. The Roadmap argues that understanding how this chemical language is written, read and erased across cellular contexts is one of the field’s most pressing challenges, particularly because disruptions in the code are now firmly linked to neurodegeneration, ciliopathies and other disorders.

Nowhere is the clinical relevance more evident than in the nervous system. Neurons depend on microtubule tracks for the long-range transport of organelles, messenger RNAs and signalling endosomes along axons and dendrites, a process powered by kinesin and dynein motor proteins. Mutations in tubulin genes such as TUBA1A cause lissencephaly, a severe developmental brain malformation, by perturbing neuronal migration, and variants in a growing list of tubulin genes underlie a spectrum of neurodevelopmental disorders collectively known as tubulinopathies. Defective axonal transport is a shared hallmark of motor neuron diseases, and recent work has shown that excessive polyglutamylation of tubulin in the brain is sufficient to drive neurodegeneration by disrupting transport, while restoring the balance of these modifications can rescue affected neurons. The Roadmap positions a systems-level understanding of these processes as essential for translating molecular insight into therapies.

The heart and the immune system tell equally compelling stories. In beating cardiomyocytes, detyrosinated microtubules act as load-bearing elements that buckle under contraction and stiffen the cell, and suppressing this modification improves cardiac function in models of heart failure. In immune cells, microtubule networks orchestrate cell shape, migration and the directed release of lytic granules by natural killer cells, while platelets, whose tubulin repertoire is dominated by the β1-tubulin isotype, rely on precisely regulated microtubule assembly during their biogenesis; mutations in the TUBB1 gene cause congenital macrothrombocytopenia and thyroid dysgenesis. The Roadmap argues that these tissue-specific roles cannot be understood piecemeal and demand approaches that bridge the molecular properties of tubulin to the physiology of whole organs.

Technological innovation sits at the centre of the proposed strategy. Cryo-electron microscopy and cryo-electron tomography now reveal microtubules and their associated proteins at near-atomic and in situ resolution, while expansion microscopy and ultrastructure expansion microscopy allow researchers to visualize centrioles, cilia and mitotic spindles in whole cells and tissues with nanometre precision. In vitro reconstitution, a tradition stretching back to the pioneering work of Tim Mitchison and Marc Kirschner and continued in landmark experiments showing that purified microtubules and motors can self-organize into asters and vortices, remains indispensable for dissecting minimal systems. These experimental approaches are increasingly paired with computational and physical models that simulate spindle assembly, microtubule network organization and chromosome movement, and the Roadmap calls for tighter coupling between modelling and experiment, including the development of virtual cells powered by artificial intelligence.

Recombinant tubulin technology is highlighted as a particularly transformative advance. For decades, biochemists were limited to native tubulin purified from brain tissue, a mixture of isotypes and modifications that obscured cause and effect. The ability to produce functional human tubulin dimers in defined isotypes, with controlled post-translational modifications, has now enabled researchers to show directly that different isotypes confer distinct dynamic properties, that detyrosination tunes microtubule stability through selective recruitment of associated factors, and that systematic mutagenesis can map the functional landscape of disease-linked variants. Combined with deep mutational scanning and artificial-intelligence-driven phenotyping, the authors argue, these tools will allow the effects of every clinically observed tubulin mutation to be predicted and tested, a goal that seemed out of reach only a few years ago.

The Roadmap also looks beyond animal cells, drawing attention to the diversity of microtubule arrays across eukaryotes, from the cortical arrays that guide plant cell morphogenesis to the subpellicular arrays of trypanosomes and the specialized mitotic machinery of parasites. Comparative studies of parasites such as Plasmodium have revealed that adaptations in tubulin sequence generate distinct microtubule architectures, mechanics and drug susceptibilities, opening avenues for species-selective therapeutics. The authors contend that this evolutionary breadth is not a curiosity but a resource: organisms that build microtubules with unusual lattices, modifications or assembly mechanisms offer natural experiments that can illuminate principles hidden in familiar model systems.

Ultimately, the Roadmap is an invitation to think bigger. Its authors conclude that the microtubule cytoskeleton will continue to inspire scientists for decades precisely because so many fundamental questions remain open: how microtubule arrays are organized and diversified in different cell types, how the tubulin code is orchestrated across space and time, how mechanical forces reshape and stabilize microtubule lattices in living cells, and how all of this integrates into the physiology of tissues and organisms in health, ageing and disease. By bridging the atomic structure of the tubulin dimer with the behaviour of mitotic spindles, migrating neurons and beating hearts, the field aims to transform its fragmented molecular knowledge into a coherent systems-level picture, one that could reshape how disorders ranging from neurodegeneration and cancer to ciliopathies and heart failure are understood and treated.

The historical depth of the field is worth appreciating. Microtubules were first visualized in the late nineteenth and early twentieth centuries, but their protein building block was only identified in the late 1960s, when colchicine-binding assays led to the isolation of what was soon named tubulin, and amino-acid analysis of sperm flagella confirmed it as the universal subunit of the polymer. The discovery of dynamic instability two decades later established that these polymers are fundamentally nonequilibrium structures, a insight that continues to shape how spindle assembly and chromosome movement are understood today.

Equally important is the cast of regulatory proteins that act on microtubules. Microtubule-associated proteins such as tau can condense on the lattice in regulated phases, altering how motors and severing enzymes engage the polymer, while enzymes that cut microtubules generate new seeds and reshape networks. Proteins like doublecortin, which is mutated in human brain malformations, illustrate how even lattice geometry itself can be read selectively, recognizing only specific protofilament numbers. These layered interactions, spanning motors, maps, severing enzymes and modifying enzymes, form the mechanistic vocabulary that any systems-level account of the cytoskeleton will ultimately need to integrate.

Subject of Research: Systems-level integration of microtubule cytoskeleton structure, regulation and function in physiology and disease

Article Title: Towards a systems-level view of the microtubule cytoskeleton and its functions in physiology and disease

Article References: Janke, C., Akhmanova, A., Bartolini, F., Bodakuntla, S., Del Bene, F., Hamel, V., Mitchison, T. J., Müller-Reichert, T., Nédélec, F., Nogales, E., Pigino, G., Roll-Mecak, A., Schiavo, G., Surrey, T., & Lansky, Z. (2026). Towards a systems-level view of the microtubule cytoskeleton and its functions in physiology and disease. Nature Reviews Molecular Cell Biology. https://doi.org/10.1038/s41580-026-01011-w

Image Credits: AI Generated

DOI: 10.1038/s41580-026-01011-w

Keywords: microtubules, tubulin code, post-translational modifications, dynamic instability, axonal transport, tubulinopathies, ciliopathies, neurodegeneration, mitotic spindle, cryo-electron tomography, in vitro reconstitution, systems biology

The findings, published Sept. 10 in the journal Science, are significant not just for what they reveal about the architecture of the cilium, but also for what they promise for patients. Between 20 and 30 percent of people with primary ciliary dyskinesia, or PCD, currently receive a genetic diagnosis that leaves their disease unexplained — no known genetic mutation, no clear molecular culprit. This new work offers at least some of those patients, and their physicians, a long-sought answer, and points to fresh molecular targets for the development of therapies in a field where treatment options remain painfully limited.

Primary ciliary dyskinesia is a rare, inherited condition that arises when the cilia that line the respiratory tract, the inner ear, the reproductive organs, and other tissues fail to beat in the coordinated rhythm needed to move fluids and particles along. When these microscopic machinery components break down, mucus accumulates in the lungs and sinuses, fluid builds up in the middle ear, and infection follows infection, often leading to chronic respiratory illness, scarring, and permanent lung damage. Roughly half of patients also experience abnormal organ placement or incomplete organ formation during embryonic development, because motile cilia also play a crucial role in establishing the left-right asymmetry of the body plan. The disease is usually diagnosed in childhood, typically after parents and pediatricians notice a relentless cycle of congestion, coughing, and ear infections that fails to respond to conventional treatment.

To develop PCD, most people must inherit two copies of a disease-causing gene, one from each parent. More than 50 such genes have been identified to date, yet a substantial fraction of cases — as many as three in ten — remain genetically unexplained. For those patients, the new study offers a previously unexplored corner of the cilium as a place to look: the transition zone.

The transition zone sits at the base of the cilium, where the organelle emerges from the cell body. It is, in essence, a molecular gatekeeper — a structured region that controls what proteins and other cargo are allowed to pass from the cell into the cilium’s main shaft. Structural abnormalities in the transition zone had already been linked to ciliopathies involving non-motile cilia, but until now, the mutations known to cause PCD had been traced to a different part of the organelle, the axoneme, the beating core of the cilium. The transition zone, in the context of motile cilia, had largely escaped scrutiny.

That changed thanks to a technological marriage. Haixia Zhou, a research fellow in the laboratory of Alan Brown, professor of biological chemistry and molecular pharmacology at Harvard Medical School, applied a cutting-edge imaging technique called cryo-focused ion beam milling-assisted electron tomography, or cryo-FIB-ET, to cultured cells derived from the lining of the human respiratory tract. The technique is a sophisticated evolution of cryo-electron microscopy, or cryo-EM, in which biological samples are flash-frozen in a glassy, vitrified state that preserves them in near-native form. In conventional cryo-EM, samples must be extraordinarily thin — on the order of 100 to 250 nanometers — for the electron beam to penetrate them and produce interpretable images. Most biological structures, including cilia, are far too thick. Cryo-FIB milling solves this problem by using a focused beam of ions to carve the frozen sample into ultrathin slices, shaving away material until the region of interest is thin enough for the microscope’s electrons to pass through.

The researchers then captured high-resolution snapshots of the transition zone from multiple angles and used advanced computational reconstruction, augmented with artificial intelligence tools, to assemble them into three-dimensional atomic models. The resulting images resolved the transition zone down to the sub-nanometer scale — fine enough to identify individual proteins and trace how they interlock with one another.

What the team found was a remarkably intricate piece of molecular engineering. Nine distinct protein types populate the transition zone, and four of them assemble into complexes that physically link together the cilium’s major structural elements — long, tube-like scaffolds called doublet microtubules that run the length of the organelle and provide the rigid framework against which the ciliary beat is generated. These linker complexes, the researchers discovered, act as the cross-braces of the transition zone, holding its architecture together and preserving its function as a selective gate.

The genetic leap came from the clinic. Heymut Omran, co-senior author of the study and a pediatrician at University Children’s Hospital Münster who treats children with ciliopathies, brought to the collaboration patients whose PCD had resisted genetic explanation for years. Working with co-first author Lea Terbeck and colleagues in Münster, along with collaborators at the University of Geneva and Boston University, the team identified mutations in the genes encoding two of the transition-zone linker proteins: ECT2L and DZANK1. Neither gene had previously been implicated in PCD.

When the researchers examined cilia carrying these mutations, the consequences were unmistakable. The linker complexes that hold the transition zone together were disrupted, and the cilia adopted abnormal shapes — some developing bulbous, swollen tips. More critically, the cilia lost their ability to beat in synchrony. In the airways, where millions of cilia must pulse in coordinated waves to push mucus upward and out of the lungs, this loss of coordination is catastrophic. Mucus stagnates, microbes colonize, and the cycle of infection and inflammation that defines PCD begins.

The researchers suspect the underlying mechanism is subtle but profound: because the transition zone acts as a gate into the cilium’s main body, mutations in ECT2L and DZANK1 may allow the wrong proteins to enter — or prevent the right ones from doing so — corrupting the cilium’s internal composition and, with it, its capacity for coordinated motion. In this view, PCD arising from transition-zone defects is not simply a failure of the cilium’s motor machinery, but a failure of the molecular customs office that governs what that machinery is made of.

The work deepens fundamental understanding of how motile cilia are built and function, and it calls scientists’ attention to a region of the organelle that has been comparatively neglected in studies of motile-cilia disease. It also expands the roster of genes known to cause PCD, immediately improving the diagnostic landscape. For the 20 to 30 percent of patients whose PCD has been a genetic mystery, testing for mutations in ECT2L, DZANK1, and other transition-zone genes now becomes a concrete possibility — and with a genetic diagnosis often comes clarity about prognosis, guidance for reproductive counseling, and eligibility for emerging targeted therapies.

Beyond diagnosis, the newly identified proteins offer researchers concrete targets for drug development. Current treatments for PCD can only slow disease progression; nothing yet exists to halt or reverse it. Knowing precisely which molecular linkers fail, and seeing their structures at atomic resolution, gives structural biologists and medicinal chemists a starting point for designing interventions — whether small molecules that stabilize defective linkers, gene therapies that restore functional copies of the affected genes, or other strategies yet to be imagined.

For Alan Brown, the study exemplifies a broader methodological promise. “This is an example of how combining in situ structural biology with genetics can address fundamental questions and illuminate the causes of human disease,” he said. By imaging proteins in their native cellular environment rather than in isolation, and by anchoring those structural observations to real patients with real mutations, the team has demonstrated a template for future discoveries — one in which the transition zone, long a blind spot in motile-cilia biology, now stands revealed in all its molecular complexity, and in which a new chapter opens in the effort to understand, diagnose, and eventually treat one of medicine’s most stubborn rare diseases.

The research was conducted at the Harvard Cryo-EM Center for Structural Biology at HMS, which acquired cryo-FIB-ET capabilities in 2022, with contributions from the HMS Electron Microscopy Core Facility and the SBGrid Consortium. The authors also extended their thanks to the individuals with PCD and their families who participated in the study — the patients whose cells and genetic samples made this molecular portrait of the transition zone possible.

Cryo-electron tomography works by recording a series of two-dimensional electron microscope images while a specimen is tilted through a range of angles. Computational methods then combine those projections into a three-dimensional volume, or tomogram. Because biological samples are vulnerable to radiation damage, the electron dose must be limited, leaving the resulting images noisy and incomplete. The geometry of the experiment also creates a missing wedge: some orientations cannot be recorded because of physical restrictions on how the specimen and microscope are positioned. These limitations make objects appear distorted and complicate automated recognition. A particle-picking algorithm must distinguish genuine macromolecular complexes from features produced by noise, overlapping structures, uneven contrast and reconstruction artefacts.

The challenge described by Aaron Peck, John Hutchings, Jason Schwartz and colleagues focuses on that recognition task. Kaggle competitions are online data-science contests in which participants receive a defined computational problem, training data and an evaluation framework, then develop and compare predictive models. In a biological imaging setting, such a competition can expose the same dataset to a broad range of strategies, including classical image-processing pipelines, template-matching systems and deep-learning models. Rather than relying on a single laboratory’s preferred software or assumptions, a challenge can reveal which approaches generalize across samples and which perform well only under narrowly controlled conditions. Its value lies not simply in identifying a winning algorithm, but in making the hidden decisions behind model development visible.

Particle picking is often treated as a preliminary technical step, but its consequences extend through the rest of structural analysis. The coordinates selected by a picker may be used to extract subtomograms, align molecular views, classify structural states and calculate averaged reconstructions. False positives add irrelevant material to those calculations, while false negatives remove useful particles and can bias the final dataset toward the easiest examples to detect. Errors can be especially serious when the goal is to identify rare conformations, transient assemblies or particles embedded in crowded cellular landscapes. A detector that appears accurate when judged by a simple score may still distort downstream biological interpretation if its mistakes are unevenly distributed across orientations, sizes or molecular states.

The technical difficulty arises partly because cryo-ET data do not resemble ordinary photographs. Contrast depends on specimen thickness, defocus, ice conditions and the local environment. Particles may be partially obscured, tilted relative to the imaging plane or distributed at different depths within a volume. Their apparent shapes can change as a result of the microscope’s contrast-transfer function and the reconstruction method used to create the tomogram. In addition, a model trained on one type of macromolecular complex may learn visual shortcuts that fail on another. Deep neural networks can identify subtle patterns that are difficult to encode manually, but they also require representative training data and careful validation. If annotations are incomplete or inconsistent, a model may be penalized for finding legitimate particles that were never labeled.

A competition framework can help expose these issues by forcing participants to work against common data and evaluation rules. It also creates an opportunity to compare methods under conditions that are closer to the diversity of real cryo-ET experiments than a single benchmark may be. At the same time, competition scores require careful interpretation. A metric calculated on a held-out dataset measures performance on that dataset’s distribution; it does not automatically establish that a method will work on a different microscope, specimen preparation protocol or biological target. The most useful lessons may therefore concern experimental design as much as algorithmic architecture: how labels are produced, how training and test sets are separated, how detection thresholds are chosen and how performance is assessed across different structural contexts.

The report’s significance is tied to this broader question of reproducibility in computational cryo-EM. Automated particle picking promises to reduce the manual labor required to analyze tomograms, but automation is only scientifically useful when users can understand its failure modes. A robust workflow should make it possible to inspect detections, adjust operating points and retain information about uncertainty rather than presenting every prediction as equally reliable. In practice, researchers may need different balances between precision and recall. When particle abundance is high, rejecting contaminants may be the priority; when a target is rare, recovering as many plausible candidates as possible may matter more, even if manual review is required later. No single score captures all of these scientific priorities.

The Kaggle experience also highlights the importance of shared benchmarks for a rapidly evolving field. Cryo-ET is expanding from purified samples toward increasingly complex cellular and tissue contexts, where the visual background can be as informative—and as challenging—as the particle itself. Algorithms developed for isolated, high-contrast targets may not transfer directly to crowded tomograms. Standardized challenges can encourage the creation of better annotations, clearer evaluation protocols and software that is easier for biologists to deploy. They can also reveal where more data are needed, particularly for unusual particle orientations, damaged specimens and targets that occur at low abundance. Such resources may ultimately matter as much as any individual model.

The central lesson from the study is that particle picking should be understood as a systems problem rather than a contest to produce the highest isolated detection score. Image formation, reconstruction, annotation, model training, threshold selection and downstream structure determination are linked stages in one pipeline. Improvements at one stage can be undermined by weaknesses at another, while a method that performs modestly on a benchmark may prove valuable if it is transparent, adaptable and efficient in laboratory practice. By documenting what emerged from a community challenge, Peck and colleagues place the technical problem in a wider context: reliable three-dimensional molecular discovery depends not only on more powerful microscopes and larger neural networks, but also on carefully designed tests that show when computational tools can be trusted.

A new report in Nature Plants combines in situ cryo-electron tomography with high-resolution single-particle cryo-EM to reveal how chloroplast ribosomes in Chlamydomonas reinhardtii are reshaped by a small-subunit extension encoded in chloroplast DNA. The study targets both native membrane-associated ribosome states and functional tRNA-containing conformations, linking ultrastructure to translation-relevant dynamics.

For the in situ component, tilt-series data were retrieved from EMPIAR-1183026 and acquired on a Titan Krios G4 (300 kV) with an energy filter (10 eV slit) and a Falcon 4i detector. Dose-symmetric tilt schemes covered ±60° with 2–3° steps, while target focus varied in defined increments to maximize reconstruction quality. Raw dose-fractionated EER movies were motion-corrected and CTF-estimated, then reconstructed using AreTomo, followed by Icecream denoising on odd/even frame reconstructions.

Particle detection relied on 3D template matching (pyTOM-match-pick) after binning and low-pass filtering a ribosome reference from SPA. Candidate positions were imported into RELION for multiple rounds of 3D classification without alignment, removing false positives such as membrane segments. Successive rounds of CTF refinement, Bayesian polishing, and re-extraction at higher sampling improved map quality to ~6 Å, with focused refinements yielding sub-6 Å structures for major ribosomal regions.

To probe functional and spatial heterogeneity, the team further classified particles at different binning levels to isolate membrane-bound ribosomes and distinct tRNA states. Membrane-bound particles were separated via masked, alignment-free 3D classification, producing a final ~9.29 Å reconstruction. For translation-state analysis, masked classification with high T separated six tRNA-related conformational states, setting the stage for state-specific architectural comparisons.

Complementing tomography, the study purified chloroplast ribosomes from 2 liters of cultured Chlamydomonas and prepared grids for cryo-EM at both 200 kV (Glacios) and 300 kV (Titan Krios). Data were collected in EPU, processed independently in cryoSPARC, and merged at the final stages. Resolution for the SSU extension class reached ~2.6–2.7 Å after refinement strategies that included reference-based motion correction and local/global CTF refinement.

Model building used ModelAngelo de novo sequencing fed into AlphaFold2-based model retrieval, then rigid-body fitting into cryo-EM maps and iterative refinement in COOT with PHENIX real_space_refine. rRNA models were adapted from bacterial templates, while ions, ligands, and RNA modifications were curated by homology and manual inspection.

Finally, a dedicated polysome workflow represented ribosome contacts as graph components using Euclidean KDTree neighbor searches and SO(3) orientation constraints, accepting polysomes by both distance windows (200–320 nm) and angular similarity thresholds. Membrane orientation was quantified by extracting membrane normals from fitted membrane density and tracking deviations on tangent-plane projections.

Subject of Research:

Chloroplast ribosome structural reshaping by a small-subunit extension in Chlamydomonas

Article Title:

Chloroplast-encoded small subunit extensions reshape the Chlamydomonas chlororibosome.

Article References:

Waltz, F., Lehner, P.A., Van der Stappen, P. et al. Chloroplast-encoded small subunit extensions reshape the Chlamydomonas chlororibosome. Nat. Plants (2026). https://doi.org/10.1038/s41477-026-02361-1

Keywords:

cryo-ET, chloroplast ribosome, SSU extension, tRNA states, polysomes, membrane-bound ribosomes, cryo-EM, Titan Krios