TDP-43’s next misprocessing class is circular

In Short. Cryptic circRNAs are the next TDP-43 misprocessing class after cryptic linear exons. They are more stable, a biomarker-class claim before any drug claim, and UNC13A hotspots bridge them to the existing ASO story. Both objects are still preprints.

---

The first public lesson about TDP-43 failure in amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) was linear. When the protein leaves the neuronal nucleus, splice sites that should stay silent turn on, and cells insert cryptic exons into otherwise ordinary messenger RNAs. Those linear mistakes became the teachable footprint of TDP-43 nuclear loss. Two independent bioRxiv preprints posted in the same window now say the same failure mode also makes circles.

TDP-43 is an RNA-binding protein whose nuclear depletion and cytoplasmic mislocalisation define most ALS and roughly half of FTD. In the nucleus it polices splicing and related RNA processing steps. Take it away, and the transcriptome fills with events that healthy neurons barely make. Cryptic exons were the first class of those events to enter the ALS and FTD vocabulary. Drug talk already orbits them. UNC13A, a strong genetic risk locus in both diseases, carries a disease-linked cryptic exon that splice-switching antisense oligonucleotides (ASOs) aim to suppress.

Circular RNAs are a different kind of transcript. They form when the spliceosome joins a downstream splice donor back to an upstream acceptor, closing a loop with no free ends. That closed topology makes them harder for cellular nucleases to chew up than ordinary linear RNAs. Until these papers, TDP-43’s public story was still mostly about linear mis-splicing. The new claim is that nuclear TDP-43 loss also deregulates back-splicing, so neurons start producing circular RNAs that were absent, or nearly absent, under physiological conditions.

Both groups call that class cryptic circRNAs, and both define the word from methods rather than marketing. Dario Dattilo, Flaminia Pellegrini, Pietro Fratta and colleagues at University College London and the Francis Crick Institute treat cryptic circRNAs as molecules detected after TDP-43 knockdown and absent in matched controls in human neuronal models, mirroring how cryptic linear events are defined. In New York Genome Center ALS Consortium bulk RNA-seq (1,863 CNS samples across regions), they further call a circRNA cryptic when it appears in fewer than 5% of control samples yet is enriched in ALS-TDP tissue (bioRxiv). Chengzhang Zhu, Shuying Sun and colleagues at Johns Hopkins use a closely related rule. Cryptic circRNAs are those that contain cryptic exons, emerge de novo after TDP-43 loss, or both (bioRxiv). Some circles incorporate cryptic exons. Others back-splice only annotated exons that become newly available once TDP-43 is gone. Either way, the object is not a familiar circRNA merely going up a little. It is a circular transcript that the healthy baseline barely admits.

The UCL and Crick paper pushes the biomarker angle hardest. Four pathology-linked cryptic circles (from UNC13A, BCAR1, DIP2C and DSCAM) were measured in post-mortem frontal cortex with a rolling-circle reverse-transcription step followed by back-splice-junction TaqMan assays. The tissue set was healthy controls (n=21), FTD with TDP-43 pathology (n=56, split as GRN n=17, C9ORF72 n=14, sporadic FTLD-TDP n=25), and FTD without TDP-43 pathology (FTD-FUS, n=12). Each marker alone separated FTD-TDP from the combined control-plus-non-TDP group at AUC 0.90 to 0.98. A four-marker logistic model reached AUC 0.99. That is a strong tissue discrimination claim on this cohort. It is not yet a clinical diagnostic. The paper does not describe assay operators as blinded to diagnosis for the ROC analysis, and it does not report an independent held-out validation set. Honest reading stops at near-complete separation of labelled frontal-cortex groups under rolling-circle amplification, not at a ready blood test.

Johns Hopkins supplies the complementary physical claim. After TDP-43 depletion in human i3Neurons, candidate cryptic circRNAs decay more slowly than matched linear isoforms under transcription shutoff, and they accumulate as differentiated neurons age with constitutive knockdown. In Johns Hopkins Brain Resource Center temporal cortex, several candidates rise in C9ORF72-associated ALS/FTD (n=15) and sporadic FTD (n=5) versus non-neurological controls (n=15). A different subset pattern appears in Alzheimer’s disease temporal cortex (n=8), consistent with TDP-43 proteinopathy spanning disorders without implying one universal circle signature. BaseScope imaging of back-splice junctions (quantified blind to disease condition) backs enrichment in ALS/FTD tissue. The circles look like durable molecular leftovers of a splicing failure that linear cryptic transcripts, often cleared by nonsense-mediated decay, struggle to leave behind.

UNC13A is where the new class meets the existing ASO story. Both papers recover a cryptic circular isoform from the locus. The UCL group shows it is co-regulated with the linear cryptic exon. Delete the genomic region that makes the cryptic exon, or treat neurons (and a human UNC13A BAC mouse under TDP-43 knockdown) with CE-targeting ASOs, and the circle falls with the linear cryptic transcript. That is target-engagement language, not efficacy language. An ASO that “fixes the splice” at UNC13A may also clear the circle. The circle may therefore become a more stable engagement readout than the linear cryptic exon itself. It does not yet prove that clearing either species slows ALS or FTD.

Causal language needs the same restraint. In knockdown models, TDP-43 loss is sufficient to induce cryptic circRNAs. Binding of TDP-43 in flanking introns supports a direct repressive role at some loci. In human post-mortem tissue the claim is association with TDP-43 pathology and enrichment where cryptic burden is high. Neither preprint shows that cryptic circRNAs are the toxins that kill motor neurons. They report a misprocessing class and a detection advantage, not a pathogenic mechanism closed for drug design.

So the grammar of TDP-43 failure has a second chapter. Cryptic linear exons taught that nuclear loss invents splice products healthy cells refuse. Cryptic circRNAs are the next class in that same failure mode. They are more stable, easier to amplify with rolling-circle chemistry, and already touching the UNC13A hotspot that ASO programmes know. Hold them as a biomarker-class claim before any therapy claim. Both objects remain preprints, not approved diagnostics or medicines. Fixing the linear splice may still be the right first drug verb. Measuring the circle may change how tightly that verb can be watched.■