Families arriving at a primary progressive aphasia clinic often bring an old story with them: someone in this house could never spell. A brother held back a year. A father who read slowly all his life and became an engineer anyway. Then, decades later, the words start going, and someone asks whether the two things are connected. The question is not naive. It has been in the literature since 2008, when Emily Rogalski, Sandra Weintraub and Marsel Mesulam at Northwestern reported in Archives of Neurology on self-reported learning disability among 699 people in their registry, of whom 108 had PPA, 154 had amnestic Alzheimer's dementia, 84 had behavioural variant frontotemporal dementia and 353 were controls. Personal and family history of learning disability was significantly higher in the PPA group than in the others (P<.001). Childhood learning difficulty has been the one risk factor most PPA information sheets actually name ever since, and for nearly two decades it has had no genetic test behind it at all.
A preprint posted to medRxiv on 13 October 2025, and not yet peer reviewed, is the first attempt to put a number on that hunch using DNA rather than recollection. Sterre de Boer, Lianne Reus, Yolande Pijnenburg, Alexander Santillo and colleagues took 3,407 genotyped participants from the Amsterdam Dementia Cohort: 969 controls, 66 with semantic variant PPA, 30 with non-fluent variant PPA, 38 with logopenic variant PPA, 51 with right temporal variant frontotemporal dementia, 17 with PPA not otherwise specified, 230 with behavioural variant FTD, 1,750 with Alzheimer's dementia and 256 with dementia with Lewy bodies. For each person they built a polygenic score for developmental dyslexia, using summary statistics from the 2022 Nature Genetics genome-wide association study of 51,800 adults reporting a dyslexia diagnosis and 1,087,070 controls, computed with LDpred2 and standardised. This is a cross-sectional genetic comparison of diagnostic groups, not a trial and not a follow-up study: one blood sample, one score, one snapshot.
The result went against the authors' own hypothesis. They expected the dyslexia score to be highest in logopenic variant PPA, the subtype most often linked to Alzheimer's pathology and to phonological trouble that superficially resembles dyslexia. Instead it was highest in the semantic variant, the syndrome in which the meanings of words drain away while speech stays fluent. And when the model showed a significant diagnosis by sex interaction, the team split the analysis. Within men, dyslexia polygenic scores were significantly higher in semantic variant PPA than in every other diagnosis and than controls (all p-values below 0.05), with the single exception of the non-fluent variant. Within women, nothing reached significance. Unstratified, the pattern looked like the male one, except that logopenic PPA was no longer clearly separated; in a sensitivity analysis that dropped the adjustment for education, non-fluent PPA and logopenic PPA both stopped separating too.
Hold that finding at arm's length for a moment and look at what carries it. Sixty-six people with semantic variant PPA, then divided by sex, are a very small pair of groups to hang a sex-specific claim on, and the honest reading of "significant in men, not in women" is that the study cannot yet distinguish a real biological difference from a difference in statistical power. The exposure side has its own softness: the reference genome-wide association study rests on adults who reported having been diagnosed with dyslexia, and who ever got that diagnosis, in which decade and in which school system, is not evenly distributed. The authors themselves float the plainest explanation, that more of the svPPA patients may simply have had dyslexia, which would be interesting in its own right rather than a nuisance. And the causal question has already been tested another way: a 2023 Mendelian randomisation analysis in Neurology, using 42 dyslexia variants against semantic PPA (308 cases, 616 controls) and agrammatic PPA (269 cases, 538 controls), found no association between dyslexia and any PPA subtype.
So what does a family do with this? Nothing, and that is the point worth saying out loud in a prevention section. A polygenic score is an average over a group, not a verdict on a person; there is no clinic that offers this test, no threshold that means anything for an individual, and nothing here that anyone could have avoided, treated, or done differently at school or since. Dyslexia is not being described as a cause of dementia, and dementia is not being described as the destination of a childhood reading difficulty. What the preprint offers instead is a candidate explanation for an old observation: that the language network in some brains may be built slightly differently from the start, and that the same architecture that made reading hard at eight may be the tissue that gives way first at sixty-five. Vulnerability, not exposure.
For people already living with a semantic variant diagnosis, that framing has a quiet use. It moves the question away from what did I do and towards what was always true of this brain, which is a different kind of answer to sit with. It also sharpens what researchers should ask next: replication in larger PPA cohorts, in ancestries beyond the mostly European samples these scores are built on, with sex included as a question rather than an afterthought, and with peer review done. Until then this is one preprint from one Dutch cohort, and it should be read as the beginning of an argument.