For twenty years the Huntington's field has spoken about "huntingtin lowering" as if huntingtin were one thing. It is not. When the CAG repeat in exon 1 of the HTT gene grows long, the cell's processing machinery increasingly stops reading the gene early, at a cryptic stop signal buried in the first intron, and produces a short transcript called HTT1a. That transcript makes a stubby protein carrying the whole expanded repeat, and it clumps. The longer the repeat grows inside a neuron over a lifetime, the more of this short, sticky version gets made. So a drug that lowers huntingtin might be lowering the long harmless one, the short toxic one, or both, depending entirely on where along the message it happens to bind. Until this year nobody had run the head to head experiment.

On 18 March 2026, Science Translational Medicine published two papers that did. In the first, Aikaterini Smaragdi Papadopoulou, Gillian Bates and colleagues designed small interfering RNAs against 24 sites inside mouse Htt intron 1, upstream of the first cryptic polyA site at 680 base pairs, screened them in a HeLa reporter assay, and picked two winners: siRNAs 486 and 634, which cut the Htt1a signal to 11.6 percent and 8.3 percent of untreated levels. They then put those against 10150, an established and frankly more potent siRNA that hits full-length Htt and leaves Htt1a alone. Both were injected into the brain ventricles of zQ175 knock-in mice at two months of age or at six, with brains examined at six or ten months. The full-length drug won on potency; it lowered its target harder. The HTT1a drug won on the thing that matters. Treatment with 634/486 at two months left the dentate gyrus with no inclusion bodies at all at six months and only faint nuclear staining at ten, and it did more to hold back transcriptional dysregulation. The authors' own summary is unshowy: targeting Htt1a was more effective at delaying aggregation than targeting full-length Htt, despite the greater potency of the full-length drug.

The companion paper, from Robert Bragg, Jeff Carroll and the same UCL group, arrived at the same place by a different road. Working in heterozygous HttQ111 mice, which carry a humanised stretch of intron 1 on the mutant allele, they built an antisense oligonucleotide aimed at that mutant-specific sequence. Because the target sits upstream of the cryptic polyadenylation signals, the ASO knocked down mutant full-length huntingtin and HTT1a together. Against a pan-allele ASO that lowers both wild-type and mutant full-length protein while sparing HTT1a, the mutant-selective one, in the authors' words, "essentially eliminated aggregate formation" and gave marked protection against transcriptional dysregulation. Two labs, two modalities, two knock-in models, one conclusion: the short transcript is doing a disproportionate share of the damage, and drugs that miss it are leaving the arsonist in the building.

This is a mouse result, and it needs saying clearly. Both studies measured molecular and neuropathological endpoints: aggregate load in hippocampal fields, gene expression signatures. Neither measured whether treated mice thought better, moved better or lived longer, and neither involved a human being. But the argument has an uncomfortable fit with what has actually happened in clinics. Tominersen, the drug that carried the field's hopes furthest, binds downstream in the message and lowers full-length HTT while leaving HTT1a untouched, a point the Bates group makes explicitly when listing which agents hit which transcript. On 9 July 2026 Roche told the HD community it was discontinuing tominersen after GENERATION HD2 showed no clinical benefit despite doing exactly what it was designed to do biologically, and was also stopping its first-in-human allele-selective ASO, RG6496. The small molecule splicing modulators, branaplam and PTC518, also act on the full-length transcript. Meanwhile AMT-130, uniQure's gene therapy, targets exon 1 and so lowers both isoforms; on 17 June 2026 the company said the FDA had agreed that three-year phase 1/2 data could serve as the primary basis for an accelerated approval filing, which it planned to submit in the third quarter of this year. That correlation is not proof. It is one plausible reading of a very expensive decade, and these two papers are the first to test it experimentally rather than argue it on a slide.

For a family sitting with a diagnosis, the practical meaning is narrower than the headlines will suggest and more useful. Nothing here changes what is available to anyone today, and nothing here is a treatment. What it changes is the question worth asking about any trial a relative is offered or considering: not just "does it lower huntingtin", but "which transcript does it lower, and how do you know". That is a fair question to put to a trial coordinator, and after March it is a question the field can no longer wave away. It also reframes the CAG instability story that has dominated the last three years. If somatic expansion drives onset because expansion drives HTT1a production, then blocking repeat growth and clearing HTT1a are two routes at the same target, and a drug that does neither may lower a great deal of protein to very little effect.

The honest caveat is that no HTT1a-directed drug exists for people. The mouse siRNAs went in by direct injection into the ventricles, the mouse ASO exploited a humanised sequence engineered into that particular model, and the long-term consequences of removing HTT1a specifically, in a brain that has been making it for forty years, are simply unknown. What the field has gained is not a candidate. It is a specification.