Outer nuclear layer thinning as an in vivo biomarker for discriminating probable FTLD-tau from probable FTLD-TDP with PET-supported subtyping.

Abstract

Background and objectives: Outer nuclear layer (ONL) thinning has been identified in frontotemporal lobar degeneration (FTLD); however, its utility for distinguishing the subtypes of FTLD-tauopathy (FTLD-tau) and TDP-43 proteinopathy (FTLD-TDP) remains unknown. We investigated whether ONL thickness provides a subtype-informative retinal signal for differentiating PET-supported probable FTLD-tau (pFTLD-tau) from probable FTLD-TDP (pFTLD-TDP) in vivo.

Methods: Patients clinically diagnosed with FTLD were subtyped into pFTLD-tau and pFTLD-TDP groups based on multimodal PET and clinical criteria. Normal controls (NCs) were cognitively unimpaired on standardized testing and clinical evaluation. Macular images were acquired using swept-source OCT. A custom deep learning algorithm segmented the retina into eight sublayers. The thickness of each retinal sublayer was assessed across the eight sectors of the Early Treatment Diabetic Retinopathy Study (ETDRS) grid. Retinal thickness differences were analyzed using generalized estimating equations, and exploratory discrimination models were evaluated using age- and sex-adjusted stepwise logistic regression with apparent and bootstrap optimism-corrected AUCs reported. Exploratory partial correlation analysis was conducted to examine the associations between ONL thickness and cognitive scores.

Results: A total of 86 participants were included (21 pFTLD-tau, 27 pFTLD-TDP and 38 NCs). Widespread ONL thinning was observed in pFTLD-tau (Cohen's d= -0.753 to -1.268 vs. controls; -0.666 to -1.069 vs. pFTLD-TDP; all FDR-adjusted P < 0.05), while ONL in pFTLD-TDP remained preserved. A model combining retinal nerve fiber layer (RNFL), ONL, and myoid-ellipsoid zone (MEZ) thickness showed exploratory discrimination for differentiating pFTLD-tau from pFTLD-TDP (apparent AUC, 0.922; optimism-corrected AUC, 0.866). The outer thickness model yielded higher AUC estimates than the inner thickness model (0.884/0.835 vs. 0.713/0.630), and the individual ONL model showed moderate exploratory discrimination (0.808/0.765). ONL thickness was correlated with cognitive scores in pFTLD-tau (partial r = 0.433-0.483; all P < 0.05), whereas corresponding associations in pFTLD-TDP did not reach statistical significance.

Discussion: ONL thinning was preferentially observed in pFTLD-tau and contributed to exploratory discrimination between PET-supported probable FTLD subtypes. These findings suggest that ONL thickness may provide complementary, noninvasive information for probable FTLD subtype stratification, with potential to facilitate therapeutic trial enrollment and personalized management. Future studies incorporating neuropathological confirmation and fluid biomarkers are warranted to validate these findings.