At the heart of this pursuit are induced pluripotent stem cells, or IPSCs. These remarkable cells hold the capacity to develop into any of the thousands of cell types found in the human body. Unlike embryonic stem cells derived from early embryos, IPSCs are reprogrammed from adult cells, circumventing certain ethical concerns while retaining the vital plasticity necessary to model human development. Scientists have used IPSCs to generate three-dimensional structures called embryoid models that mimic aspects of early human embryos under laboratory conditions.

Creating these three-dimensional embryoid structures itself posed significant technical challenges. The models needed to recreate the spatial and temporal complexity of early embryos closely enough to capture the emergence of distinct cell types and tissue layers. Although successful generation of such models was a critical first step, understanding the dynamic sequence of developmental events that occur within these heterogeneous cell communities remained elusive. Traditional methods relying on snapshot analyses or averaging data across many samples often masked the intricate variability and nuanced transitions fundamental to development.

Addressing these challenges, researchers led by Jianping Fu, Ph.D., at the University of Michigan Medical School, have innovatively employed artificial intelligence to unlock the hidden dynamics of these human embryoid models. Dr. Fu and his team recognized that despite the impressive complexity of stem cell–derived models, their utility was hindered by their intrinsic heterogeneity—random fluctuations in cell types, shapes, and molecular profiles confound straightforward interpretation. This heterogeneity had long been a barrier to extracting meaningful developmental trajectories from static imaging data.

The team’s breakthrough came from integrating physics-informed neural networks, a specialized AI architecture inspired by recent advances in analyzing plant growth and disease progression. Initially proposed by Fu’s former graduate student Kejie Chen, Ph.D., this computational approach leverages known physical constraints to guide the learning process, enabling the model to infer continuous developmental processes from discrete imaging snapshots. This synergy of biological data and AI frameworks marks a profound methodological advancement in systems biology.

By applying this neural network to thousands of confocal fluorescent microscopy images—each capturing the spatial distribution of tissue structures and protein markers at defined time points—the AI reveals previously hidden features and developmental bifurcations. These bifurcations represent critical decision points wherein stem cells commit to divergent fates, giving rise to distinct cell types. Traditional imaging analyses often overlook such subtle transitions, but the AI’s capacity to integrate morphological and molecular signals concurrently facilitates a much deeper understanding of temporal developmental dynamics.

Crucially, this computational approach not only characterizes growth patterns and differentiation events more precisely but also quantifies protein expression changes with fine granularity. The team’s work showcases how AI-driven analysis can overcome human biases and the limited resolution of conventional assessments. Features such as transient signaling states or rare cell populations become discernible, expanding the granularity of developmental maps and opening new avenues for exploring embryogenesis.

The implications of these findings are far-reaching. Beyond fundamental biology, the AI tool sets a new precedent for high-throughput screening of embryoid cultures, which could accelerate drug discovery, toxicology testing, and regenerative medicine applications. Understanding the nuances of early development also lays the groundwork for unraveling the root causes of developmental disorders, miscarriages, and birth defects, where deviations from normal differentiation trajectories play a crucial role.

Looking toward the future, the integration of AI with bioengineered embryoid models paints an exciting picture. Such systems may enable the in silico generation of realistic embryoid images, providing a virtual testing ground to examine how genetic mutations, environmental factors, or pharmaceutical interventions influence development. These advances promise unbiased, predictive insights far beyond the capacity of traditional biological assays.

Dr. Fu emphasizes the transformative potential of combining data-intensive imaging techniques with AI frameworks, highlighting that this hybrid approach can yield a comprehensive exploration of human development through a lens that simultaneously captures complexity and coherence. This represents a vital leap forward in disentangling the layered orchestration of molecular and cellular events underpinning life’s earliest stages.

The study also democratizes embryoid research by providing analytical tools that can be adapted across laboratories and developmental systems, fostering a more integrative and computationally sophisticated life sciences community. As AI continues to permeate biological research, the marriage of stem cell modeling and machine learning emerges as a powerful blueprint for decoding the mysteries of human biology.

This remarkable fusion of stem cell biology and artificial intelligence not only illuminates the earliest chapters in human formation but also heralds a new era where computational precision meets biological complexity. The work done by Jianping Fu’s group epitomizes how interdisciplinary research can push the boundaries of knowledge, ultimately ushering in transformative biomedical breakthroughs.

Subject of Research: Human embryoid model development analyzed by artificial intelligence

Article Title: Deep manifold learning reveals hidden developmental dynamics of a human embryo model

News Publication Date: Not specified

Web References: DOI: 10.1126/sciadv.adr8901

References: Fu, J., Chen, K., Qin, K-R., Na, J., Gao, G., Yang, C., et al. (2024). Deep manifold learning reveals hidden developmental dynamics of a human embryo model. Science Advances.

Keywords: Developmental biology, Machine learning, Stem cells, Organ cultures

The PICALM gene, previously linked to Alzheimer’s risk through genome-wide association studies, has been primarily known for roles in endocytosis and membrane trafficking. However, this investigation extends its functional repertoire to lipid biology within microglia. Utilizing induced pluripotent stem cell (iPSC)-derived microglia (iMGs) harboring the PICALM risk and non-risk alleles, scientists observed a striking two- to sevenfold increase in intracellular LDs in risk allele carriers. This effect was visualized and quantified using BODIPY staining, a fluorescent probe targeting neutral lipids, underscoring a profound lipid metabolic disturbance.

Further delving into cholesterol metabolism, the study employed filipin staining to detect unesterified cholesterol and revealed elevated free cholesterol levels in microglia bearing the risk allele. Cholesterol esters, which are cholesterol derivatives stored in LDs alongside triacylglycerols (TGs), have been implicated in lipid droplet biogenesis. The lipid droplet accumulation in PICALM risk-allele microglia correlates with increased levels of these storage lipids, suggesting a link between genetic variation, cholesterol handling defects, and lipid storage-organelles proliferation.

Crucially, pharmacological manipulation demonstrated that inhibiting long-chain acyl-CoA synthetase via triacsin C leads to a significant reduction of LD formation. This finding not only confirms the nature of these droplets as bona fide lipid storage entities but also hints at enzymatic steps amenable to therapeutic intervention. Moreover, the involvement of lysosomal dysfunction in this process was suggested by altered expression of lysosomal genes, and microscopic analysis showed increased lysosome and LD colocalization, pointing to impaired lipid degradation pathways as a driver of lipid overload.

The researchers further interrogated whether PICALM expression levels directly mediate these lipid anomalies. Activation of PICALM expression via CRISPR activation (CRISPRa) in risk allele microglia normalized LD levels, whereas knocking down PICALM in non-risk microglia elevated LD formation. This dose-dependent effect solidifies PICALM’s pivotal role in maintaining lipid homeostasis within microglia and identifies reduced PICALM function as a mechanistic link to lipid droplet accumulation associated with Alzheimer’s risk.

Lipidomic profiling revealed a selective enrichment of triacylglycerols among the altered lipid species in PICALM risk-allele microglia, with over thirty TG species significantly elevated. This lipid signature mirrors the previously characterized lipid-laden microglia (LDAMs) and the lipid abnormalities seen in APOE4-associated microglial states, hallmark features of Alzheimer’s pathology. Restoration of PICALM expression reversed these lipidomic disruptions, suggesting therapeutic potential in targeting this pathway to rebalance microglial lipid metabolism.

Interestingly, although PICALM has been shown to facilitate lipid transfer between neurons and astrocytes in other systems such as Drosophila and rat astrocytes, similar lipid transfer assays in these human microglia models showed no difference between risk and non-risk allele carriers. This cell-type-specific distinction implies that PICALM’s impact on lipid droplet formation in microglia may operate independently of lipid uptake from neurons, instead reflecting intrinsic defects in lipid metabolism or degradation within these immune cells.

The implication of lysosomal dysregulation in the observed phenotype highlights the intersection of lipid storage and autophagic processes. Lysosomes, essential for the catabolism of complex lipids, appeared functionally compromised in PICALM risk microglia, potentially leading to lipid droplet accumulation. Given microglial roles in debris clearance and immune surveillance, such an intracellular metabolic imbalance may impair their neuroprotective functions and exacerbate neurodegeneration.

Taken together, these data present a novel pathological mechanism by which the PICALM Alzheimer’s risk allele predisposes specifically microglia to lipid metabolic dysfunction through promoting excessive lipid droplet accumulation and lysosomal perturbation. This insight adds a vital layer to the complex molecular etiology of Alzheimer’s, shifting focus onto the metabolic health and immune competency of resident brain macrophages.

The study’s use of CRISPR-based gene editing and activation provides robust causal evidence linking PICALM expression to cellular lipid phenotypes and offers compelling proof-of-concept for modulating this pathway therapeutically. By restoring PICALM function, it may be possible to reverse lipid droplet buildup, thereby normalizing microglial physiology and potentially mitigating downstream neuroinflammatory cascades contributing to Alzheimer’s disease progression.

Future research building on these findings will be instrumental in dissecting the precise molecular pathways downstream of PICALM that govern microglial lipid metabolism. Moreover, the distinct cell-specific effects observed underscore the complexity and heterogeneity of glial lipid handling in the brain, warranting careful contextual analysis in designing targeted interventions.

As Alzheimer’s disease continues to impose an immense societal and healthcare burden worldwide, insights such as these, which illuminate the metabolic vulnerabilities ingrained in genetic risk factors, pave the way for novel biomarker development and innovative therapeutic strategies. This study stands as a landmark in bridging genetic susceptibility to cellular lipid dysregulation within microglia, expanding our understanding of neurodegenerative disease pathophysiology.

In sum, unraveling how PICALM risk alleles disrupt microglial lipid homeostasis via lipid droplet pathology and lysosomal dysfunction offers an exciting and viral new perspective on Alzheimer’s disease mechanisms. These discoveries spotlight microglia lipid metabolism as a promising target for future interventions aimed at halting or reversing the devastating cognitive decline afflicting millions.

Subject of Research: Alzheimer’s disease; microglial lipid metabolism; genetic risk factors

Article Title: PICALM Alzheimer’s risk allele causes aberrant lipid droplets in microglia

Article References:
Kozlova, A., Zhang, S., Sudwarts, A. et al. PICALM Alzheimer’s risk allele causes aberrant lipid droplets in microglia. Nature (2025). https://doi.org/10.1038/s41586-025-09486-x

Image Credits: AI Generated

The impetus for developing such a system stems from the pressing need to adapt treatments to each individual’s genetic variation. Historically, certain ASOs have been crafted to skip problematic exons in genes tied to severe disorders such as Duchenne muscular dystrophy, which stems from mutations in the dystrophin gene. The principle is elegant: by adding a small piece of synthetic RNA that instructs cellular machinery to ignore a specific, mutation-ridden exon, the cell can then stitch together the rest of the protein transcript. While the resultant dystrophin may be shorter than normal, it often remains functional enough to preserve muscle integrity. Several of these ASOs have already garnered regulatory approval for human use. Yet, each patient may carry a slightly different mutation, necessitating a distinct “exon skip” or molecular strategy. Traditional research pipelines—culturing cells, verifying corrected protein production in test tubes, moving to animal models—can run for months or years. This timeline is untenable for children whose disease progresses rapidly, or for those who do not fit neatly into existing ASO designs. Means et al. recognized that cell-based models, if universally accessible and seamlessly scalable, could drastically speed up the process of matching patients with effective, custom-made RNA interventions.

In their recent publication, the authors describe their integrated approach, starting from a large-scale reprogramming pipeline that transforms standard patient blood samples into iPS cells. Conventional methods to derive iPS cells are often laborious, requiring extensive hands-on time, specialized growth media, and highly controlled lab environments. Moreover, creating iPS cells from multiple patients in parallel can be logistically daunting, with each cell line requiring unique sets of reagents, repeated validations, and round-the-clock monitoring. However, the new protocol consolidates all these steps into a streamlined process. Within two to three weeks, blood cells from up to a dozen different individuals can be simultaneously coaxed back to an embryonic-like, pluripotent state, then expanded in sufficient quantities to fill entire cryogenic storage libraries. The authors report that over a span of six months, they derived and banked nearly 300 iPS lines from a wide variety of patients, each line representing a distinct genetic background. Such reproducibility lays the foundation for exploring genetic diseases at scale, with each iPS cell line capturing the unique genome of an individual child.

From these iPS cells, the group then set about constructing three key tissue models: skeletal muscle cells, brain organoids, and heart organoids. Though organoids derived from pluripotent stem cells have captivated researchers for over a decade, their complexities are formidable. It is not enough to simply supply iPS cells with generic growth signals; one must carefully orchestrate each step of development—mimicking the precise cues that direct cells to form branching neural tubes or contractile cardiac tissues in vivo. Means et al. advanced these protocols in an accessible, cost-effective manner. Remarkably, within a matter of weeks, they obtained small but functional “mini-hearts” (cardiac organoids) that mimicked the rhythmic beating of actual heart tissue. Similarly, by exposing the iPS cells to specific differentiation factors, they generated brain organoids that show the early architecture of developing neural tissues. In parallel, they guided iPS cells down the path of the myogenic lineage to produce skeletal muscle cells. Each tissue type, once formed, recapitulated signature features of adult organ function—for instance, the organoids’ muscle cells contract in a coordinated pattern reminiscent of the beating heart.

Such replication of complex biology in a dish opens the door to a battery of functional tests. In healthy organoids, the structural protein dystrophin is robustly expressed in skeletal muscle and heart tissues. Meanwhile, an individual with Duchenne muscular dystrophy typically harbors a frameshift mutation that halts the production of dystrophin altogether, leading to fragile and easily damaged muscle cells. To explore whether their platform could genuinely test the efficacy of FDA-approved ASOs, the team employed iPS cells from a patient with the classical dystrophin mutation. They cultivated both heart organoids and skeletal muscle cells, confirming that these miniature tissues displayed negligible dystrophin levels and manifested weak, erratic contractions. Next, they administered an ASO clinically indicated for that particular exon skipping. Over the ensuing days, the researchers observed that the mini-hearts began contracting with a renewed vigor that matched or closely approximated organoids derived from healthy donors. Analysis of protein extracts confirmed that the treated tissues had begun producing a partially truncated, yet functional form of dystrophin. This reintroduction of dystrophin was enough to restore near-normal muscle contraction patterns, demonstrating that the tissue model faithfully mirrored the in vivo pathology—and, more importantly, that the ASO therapy was working at a cellular level. Subsequent tests with alternative ASOs aimed at different exons showcased similarly promising outcomes, reinforcing that this platform could handle a diverse range of patient mutations.

Such empirical demonstration cements the idea that a single set of protocols can handle an array of individualized ASOs, effectively bridging the gulf between academic research and personalized therapy. Instead of evaluating each prospective ASO in distinct animal models—a process that can be hampered by differences in species biology—clinicians and scientists can now see how human-derived tissues respond to treatments in a controlled, in vitro environment. This strategy offers deeper mechanistic insights, too. By analyzing gene and protein expression levels directly in the organoids, investigators can detect subtle changes in splicing patterns, gauge off-target effects, and confirm whether the reintroduced protein is properly localized. The system might thus serve as an early warning if a prospective ASO inadvertently disrupts other vital transcripts, or if the muscle fibers do not properly incorporate the new dystrophin. In short, rather than waiting for ambiguous, late-stage readouts in animal models or small human trials, one can refine and optimize promising therapies at the earliest possible juncture.

Still, the authors are forthright about the platform’s constraints. Although organoids display remarkable parallels to living tissues, they cannot perfectly emulate a fully mature organ. The immature state of the cells may mask issues that would surface in adult contexts, such as more complex immune interactions or changes in gene regulation that occur over years of development. In addition, these organoids represent only a handful of possible tissues—skeletal muscle, brain, and heart. Many diseases target entirely different organs: the liver, pancreas, or kidneys. While the protocols to create these additional organoids do exist, they often require specialized mediums, 3D scaffolding, or advanced bioreactor systems. Scaling those methods to the point that they can be seamlessly integrated into the present pipeline will take further ingenuity and technical expertise. Once accomplished, the expanded library of organoids could serve as a veritable compendium of patient-based disease models, pushing the boundary of how thoroughly we can test and refine RNA therapies.

Another factor is that certain mutations are so profoundly damaging that they might hinder normal development of the iPS-derived cells, potentially leading to inconsistent or incomplete organoid formation. In such cases, the question arises: how reliably can researchers interpret the therapeutic potential of an ASO if the underlying cells refuse to mature into the relevant tissues? This challenge emphasizes the importance of controlling genetic backgrounds. While the team currently obtains iPS lines from patients with widely diverse genetic profiles, a refined approach would leverage “isogenic controls.” That is, the same cell line can be genetically corrected at the targeted locus, allowing researchers to compare the corrected version directly to the original line. Alternatively, CRISPR-based methods could systematically introduce or repair specific mutations, thereby isolating the effect of each genetic alteration within the same baseline genome. Such comparative experiments would clarify whether an organoid’s abnormal phenotype is indeed the direct result of a single splicing defect or if it stems from background variants that are often present in real-world patients.

Variability across different batches of organoids remains another concern. Recreating identical mini-organs from the same patient line in distinct labs, or sometimes even at different times in the same lab, can be fraught with subtle changes in culture conditions—ranging from lot-to-lot variability in growth media to small fluctuations in temperature or pH. This can lead to discrepancies in organoid structure, viability, or functional readouts, making cross-experimental comparisons challenging. Means et al. address reproducibility by employing standardized protocols and carefully tracking each step of the differentiation process. Still, as other researchers adopt the pipeline, the field will likely converge on best practices for organoid generation, along with robust quality metrics that define exactly how an organoid’s “health” or maturity is measured. Such a consensus would expedite the broader deployment of these personalized models in labs worldwide.

Beyond these technical aspects, one must also consider the practical dimension: how do we convert the success of an in vitro organoid-based test into a clinically actionable therapy? In the case of Duchenne muscular dystrophy, approved ASOs already exist, but their clinical efficacy in halting disease progression remains partial at best. Delivery poses a stubborn challenge—achieving consistent distribution of ASOs throughout large muscle groups or across the blood–brain barrier can be difficult. Even if a mini-heart or mini-brain in a dish responds optimally to a given ASO, real human tissues could face obstacles related to blood supply, immune responses, or synergy with existing medications. Further research on targeted delivery vehicles, such as lipid nanoparticles or viral vectors, might integrate seamlessly with the organoid pipeline. Indeed, the pipeline could itself incorporate steps where experimental delivery systems are tested on organoids, evaluating not just the direct effect of the ASO but also whether it can penetrate deeper layers of cells or maintain stable expression over time.

Nevertheless, by substantially shortening the path from genotype to therapy evaluation, Means et al. have placed the realm of truly personalized medicine within closer reach. Suppose a child arrives at a genetic clinic with a newly characterized mutation in the dystrophin gene. Clinicians may send a blood sample for genomic sequencing and simultaneously launch the pipeline to convert those cells into iPS lines. Within a month or two, that child’s miniature muscles, hearts, or brain-like organoids could be grown. Scientists could then test multiple candidate ASOs, each designed to skip a relevant exon or correct the splicing error. Which agent triggers the best restoration of dystrophin? Which yields the fewest off-target consequences? The answers would emerge quickly, enabling a more confident path toward compassionate use or rapid trial enrollment—especially pressing for degenerative conditions that show early onset.

Equally promising is the pipeline’s potential utility beyond just muscular dystrophy. The principle of splicing-based interventions extends to a wide array of other disorders. Rare diseases resulting from splicing anomalies in CFTR (the cystic fibrosis transmembrane conductance regulator) or in genes underlying certain neurological conditions may all benefit from carefully crafted ASOs. Similarly, central nervous system disorders that hinge on a faulty transcript might be amenable to correction if the splicing machinery can be modulated in a targeted way. Integrating these additional disease models would require refining the organoid platform to recapitulate advanced lung or neuronal tissue, but the blueprint is there. Each success story in the pipeline paves the way to replicate or expand methods for a more diverse set of conditions.

Beyond the urgent demands of rare disease, these revelations also carry broad implications for how we conceptualize preclinical modeling. Traditional cell monolayers in a Petri dish, while still invaluable for certain assays, often fail to replicate the intricate architecture and mechanical signals of an actual tissue. Meanwhile, genetically modified animal models might not recapitulate the same splicing patterns or gene expression relevant to humans. Organoids offer an attractive middle ground, providing three-dimensional structure with human-specific biology, but they remain more straightforward to manipulate than a full living organism. By systematically applying organoids to test RNA therapies, the research community may unearth novel insights about how splicing is regulated in development, or how certain cell types respond to partial correction of a mutated gene.

Means et al. have presented a well-rounded approach to the next generation of personalized medicine, rooted in an accessible pipeline to create iPS cells from patient blood, differentiate them into relevant organoids, and administer custom RNA therapies. While challenges remain—immature organoid phenotypes, variations in batch quality, and difficulties translating to in vivo therapy—the potential to rapidly iterate and refine interventions is undeniable. This synergy between advanced stem-cell technology and molecular medicine could transform how we approach genetic disorders, particularly those that strike early in life and progress rapidly. The transition from bench to bedside may still demand further studies, but the conceptual leap has been made. A future in which each child with a devastating genetic condition can have a tailored splicing correction therapy, validated in their own tissue analogs before it even enters their bloodstream, no longer reads like science fiction. Rather, it stands on the horizon, beckoning us to continue refining and scaling up these extraordinary breakthroughs so that any child, no matter how rare or complex their mutation, can hope for a targeted and effective treatment.

Subject of Research: Development of a scalable platform using patient-derived mini-organs (organoids) for personalized RNA therapy testing
Article Title : A scalable system using mini-organs to test personalized RNA therapy
News Publication Date : 22 January 2025
Article Doi References : https://doi.org/10.1038/d41586-025-00078-3
Image Credits : Scienmag
Keywords : iPS cells, Organoids, Antisense oligonucleotides, Duchenne muscular dystrophy, Gene therapy, Personalized medicine, Stem-cell differentiation, Frameshift mutation