Researchers at Emory University, the University of Kentucky, the National Institutes of Health and the Mayo Clinic report that the amount of cryptic exon inclusion in hippocampal tissue can be used to sort donated brains into low, intermediate and high molecular subtypes that cut across the standard neuropathologic labels for Alzheimer's disease and for limbic-predominant age-related TDP-43 encephalopathy (LATE). The study, by Adam N. Trautwig and colleagues, was published in Cell Reports on 18 March 2026. It is postmortem work: hippocampal tissue from 91 people in the University of Kentucky Alzheimer's Disease Research Center community-based autopsy cohort was selected across four groups (21 controls, 25 with Alzheimer's neuropathologic change, 35 with both Alzheimer's change and LATE, and 10 with LATE alone), and 90 cases passed quality control, yielding 9,537 measured proteins.
The protein TDP-43 normally sits in the cell nucleus and suppresses the inclusion of so-called cryptic exons, stretches of sequence that do not belong in the finished message. When TDP-43 leaves the nucleus, those exons slip in, and the resulting transcripts are degraded or truncated. The team measured cryptic exons in eight TDP-43-regulated transcripts (STMN2, UNC13A, ELAVL3, KALRN, ARHGAP32, CAMK2B, PFKP and SYT7) and combined them into a single score. Cases with both Alzheimer's change and LATE had the highest levels. The eight measures correlated more strongly with each other than with phosphorylated TDP-43, the aggregated form that pathologists stain for, and the authors report cases with high aggregate pathology but low cryptic exon scores, and the reverse. The subtypes were largely independent of amyloid and tau burden, and the score differed sharply between subtypes (p = 2.7e-30).
Mapping the proteome against the score, the authors compiled a list of 183 TDP-43-regulated cryptic exon transcripts from two published RNA sequencing datasets and found 29 of the corresponding proteins reduced or negatively correlated in high-score cases, against five that moved the other way. Canonical STMN2, ELAVL3 and KALRN were among those significantly lower, as were the kinesins KIF3A and KIF21A. Networks built around STMN2 and KALRN overlapped with proteins that fell when TDP-43 was knocked down by about 90 percent in human induced pluripotent stem cell derived neurons, a laboratory model run in quadruplicate per group, not in people. Mini-Mental State Examination scores recorded before death were lower in the high subtype than in the low subtype.
The counterpoint is the design. This is a cross-sectional autopsy study of one brain region in one regional cohort, and the link between cryptic exon burden and cognitive scores is an association measured at a single moment, at the end of life, not a demonstration that mis-splicing caused the decline. The cell work supports the direction of the mechanism but cannot substitute for evidence in living people. Nothing in the paper is a test, a treatment, or a diagnosis.