A TDP-43 phase-modifier map that needs an organoid to stay honest

What the work claims

TDP-43 is an RNA-binding protein that forms cytoplasmic inclusions in the large majority of amyotrophic lateral sclerosis cases and a substantial fraction of frontotemporal dementia, and its transition from liquid-like assemblies to solid aggregates is thought to track its toxicity. The paper claims to map the cellular processes that govern that transition. Using a colorectal cancer cell line engineered to express an RNA-binding-defective, acetylation-mimetic TDP-43 mutant that forms roughly twenty to thirty nuclear droplets called anisosomes, the authors run a 1280-compound chemical screen and a genome-wide small interfering RNA screen across 21,404 genes, then read out changes in anisosome number and dynamics.¹

The screens converge on RNA splicing, protein translation, the HSP90 and proteasome proteostasis network, and nucleocytoplasmic transport as phase regulators. The most developed claim is that inhibiting nuclear export, via the exportin XPO1, keeps TDP-43 in a liquid, RNA-dependent state and mitigates its solidification, and that the same intervention lowers pathogenic phosphorylated TDP-43 in a human brain organoid carrying a disease-associated mutation. This is a primary screening study with a mechanistic follow-up and a single organoid validation, and it should be weighted accordingly: strong on breadth of hits, thinner on disease proof.

How it works

The assay is built on a sensitized construct. The acetylation-mimetic mutant lacks RNA binding and rapidly demixes into anisosomes, spherical nuclear bodies with a shell of TDP-43 around a chaperone-rich core, which gives a countable, high-content imaging phenotype. The chemical screen recovered proteasome inhibition, which produced fewer but enlarged, irregular puncta consistent with earlier reports, along with HSP90 inhibitors, deubiquitinase inhibitors, several kinase inhibitors, and exportin inhibitors. The genetic screen independently flagged nuclear transport regulators including XPO1 and nuclear pore components.

The nuclear-export thread is the one the authors pursue. Blocking XPO1 with leptomycin B or the inhibitor KPT-276 reduced the number of anisosomes but increased their size while keeping them mobile by fluorescence recovery measurements, and a semi-permeabilized cell assay showed the enlarged condensates were stabilized in an RNA-dependent way, since an RNase treatment dissolved them. Conversely, overexpressing XPO1 drove TDP-43 into cytoplasmic, gel-like puncta. The interpretation is that low export activity holds TDP-43 in a protective liquid state, while high export activity pushes it toward solid, cytoplasmic species. The steelman is that two orthogonal screens plus an in vitro reconstitution point at the same pathway, which is a strong internal-consistency argument that nuclear transport genuinely modulates TDP-43 phase behavior.

The capstone is a human iPSC-derived brain organoid carrying an endogenous disease-associated TDP-43 mutation. Homozygous mutant organoids treated with a low dose of KPT-276 for 35 days showed reduced phosphorylated TDP-43 puncta volume and a rescue of nuclear TDP-43, with no change in total TDP-43, relative to controls. That the total protein was unchanged while the phosphorylated species fell is the encouraging part, because the effect lands on the pathological form rather than on TDP-43 abundance, and it moves the mechanism out of a cancer-cell surrogate and into human neural tissue.

Where a skeptic should push

The load-bearing assumption is that anisosome count in this construct is a meaningful proxy for disease-relevant aggregation. The authors are commendably candid that anisosomes are seen mostly in the nucleus under overexpression and that their existence in humans and relevance to the disease are unclear. The screening model is a non-neuronal colorectal cancer line expressing an artificial, RNA-binding-dead, acetylation-mimetic mutant, while the disease lesion is a cytoplasmic, phosphorylated, insoluble inclusion in neurons. That is a different compartment, cell type, and physical state from what the screen scores.

There is a subtler problem that cuts to the interpretation. Nuclear-export inhibition scores as a hit because it reduces the number of anisosomes, yet it makes each one larger, and boosting export instead produces cytoplasmic gel. So the same axis produces opposite-looking outcomes depending on whether you count condensates, measure their size, or ask where they end up. Calling the low-export state protective presumes that liquid equals benign and solid equals toxic, which is a reasonable hypothesis but is not demonstrated here by any measure of neuronal survival or function. The organoid experiment also shifts the ground: it uses a different mutation than the screen, in the homozygous state chosen to make the phosphorylated signal detectable, whereas human disease mutations are heterozygous and dominant, and the endpoint is a biomarker rather than neurodegeneration. Finally, XPO1 inhibitors of this class are clinically usable but toxic, so a therapeutic-window question hangs over any repurposing, and a 35-day exposure to an inhibitor of an essential export receptor invites a specificity concern, since chronic global export blockade can lower a phosphorylated species through broad proteostatic or cell-state changes rather than through a TDP-43-specific route. The authors are also careful to note that the drop in phosphorylated TDP-43 was not shown to correspond to fewer insoluble aggregates, so even the organoid result is a biomarker shift rather than a demonstrated reduction of the aggregated species. The honest reading is a broad modifier map with one encouraging human-tissue biomarker signal, not an established drug mechanism.

What phase modifiers mean for ALS drug screens

For organoid models of human organs and the drug discovery built on them, the constructive contribution is a modifier map with a built-in arbiter. The hits, RNA splicing, translation, HSP90 and proteasome proteostasis, and nuclear export, are a menu of druggable nodes for TDP-43 proteinopathies, which span most of ALS, a large share of frontotemporal dementia, and a subset of Alzheimer cases. Pairing that map with a human iPSC brain organoid that reports phosphorylated TDP-43 gives a disease-context endpoint that a metered organoid screen can adopt, rather than relying on a surrogate condensate count alone.

The non-obvious implication is a methodological warning that generalizes well beyond this protein. The study is an unintended demonstration that the endpoint you choose can invert your hit list. A screen scored only on condensate number would flag proteasome inhibition, with bortezomib, and HSP90 inhibition as hits, because both cut anisosome count, yet both leave fewer but enlarged, immobile, gel-like puncta, arguably a worse material state, while nuclear-export inhibition also cuts the count but keeps the condensates liquid. Count, size, and material state can therefore rank the same compound in opposite directions, and the manipulation that reduces nuclear condensates can, when reversed, drive TDP-43 into cytoplasmic gel. A phase-separation screen that optimizes a single metric can select compounds that merely reorganize pathology into fewer, larger, or differently localized species without touching the neurotoxic axis. The disease-relevant readout, phosphorylated TDP-43 in human neural tissue, is what a count-only screen cannot supply on its own, which is why the organoid endpoint, or at minimum a multi-parameter readout combining count, mobility, and localization, is needed to keep such a screen honest. In this paper the organoid is consistent with the nuclear-export hit rather than overturning it, but it is exactly the kind of disease-anchored arbiter a count-only screen would need to catch a compound pointing the wrong way.

The genuine threat is that the most reliable modifiers are core housekeeping pathways. Reducing aggregation by inhibiting splicing, translation, or the proteasome runs straight into a therapeutic-window wall in post-mitotic neurons, which tolerate such insults poorly. So the map is more valuable as mechanism than as a ready drug list, and reading it as a set of shovel-ready targets would be the optimistic error. Its best use is to prioritize interventions, like tuning nuclear transport, that might be titratable, and to insist that the organoid endpoint, not the cancer-cell count, decides which ones are real.

The bottom line

Established: converging chemical and genetic screens identify splicing, translation, proteostasis, and nuclear export as modifiers of TDP-43 phase behavior in a sensitized cell model, and XPO1 inhibition lowers phosphorylated TDP-43 in a human brain organoid carrying a disease mutation. Hypothesis: that keeping TDP-43 liquid by limiting nuclear export is neuroprotective, and that any of these modifiers is a viable drug target. What would confirm the direction is a functional neuronal readout, survival or network activity, moving with the phospho-TDP-43 signal in a heterozygous organoid at a tolerable dose. What would break it is finding that condensate metrics and disease species diverge, so that a screen optimizing one worsens the other. A genuinely useful map, kept honest only by the organoid at its end.