Alzheimer's Research & Therapy

Alzheimer's disease is characterized by progressive cognitive decline and early cerebrovascular dysfunction, including impaired neurovascular coupling (NVC) and reduced cerebral blood flow (CBF). Tau pathology is a major driver of these deficits, yet therapeutic strategies targeting tau-induced neurovascular dysfunction remain limited. The M1 muscarinic acetylcholine receptor (M1 mAChR) is a promising therapeutic target because of its critical role in cognition. We previously demonstrated that pharmacological activation of M1 mAChR improves cognitive function and neuronal survival in amyloid-based Alzheimer's disease mouse models through sex-specific mechanisms. However, whether M1 mAChR activation restores tau-mediated NVC deficits remains unknown. P301S mice were used as a model of tauopathy. Cognitive function was evaluated using the novel object recognition and Morris water maze tests, and NVC was assessed by measuring whisker stimulation-induced changes in CBF using laser speckle contrast imaging. Following baseline measurements, mice received an acute intraperitoneal injection of VU0486846, a selective M1 mAChR positive allosteric modulator (3 mg/kg), and CBF responses were reassessed over time. P301S tau mice exhibited impaired recognition and spatial memory functions, associated with reduced whisker stimulation-induced increase in CBF, indicative of impaired NVC response, while acute treatment with VU0486846 reversed these changes in NVC. This rescuing effect of VU0486846 was observed earlier in female tau mice compared to males, suggesting a sex-biased effect of M1 mAChR modulation. These findings demonstrate that M1 mAChR positive allosteric modulation reverses tau-induced neurovascular dysfunction, supporting M1 mAChR activation as a promising disease-modifying approach for Alzheimer's disease. The earlier improvement observed in females further suggests that therapeutic efficacy is influenced by biological sex.

Alzheimer's disease (AD) is characterized by substantial clinical and biological heterogeneity, with diverse neuroanatomical, cognitive, clinical, functional, and genetic risk profiles that are poorly captured by current diagnostic frameworks. Identifying reproducible disease dimensions is critical for understanding divergent pathophysiological mechanisms and designing future stratified therapeutic interventions. Herein, we applied a semi-supervised machine learning clustering framework and robust reproducibility validation strategies to structural MRI data from 1,757 participants, comprising 1,108 cognitively normal controls (CN) and 649 patients with mild cognitive impairment (MCI) or dementia due to AD. We identified two MRI-based neuroanatomical dimensions of AD: Dimension 1 (n = 393; prevalence ~ 61%) and Dimension 2 (n = 256; prevalence ~ 39%). Dimension 1 exhibited widespread cortical and subcortical atrophy, including involvement of the hippocampus, amygdala, parahippocampal, temporal, frontal, and occipital regions, while Dimension 2 showed relatively preserved brain patterns. The two dimensions were not statistically different in demographic characteristics, including age, sex, race, and education (all p > 0.05). Compared to Dimension 2, Dimension 1 had more severe cognitive and clinical impairment, greater functional impairment, elevated APOE {varepsilon}4 carrier burden, and higher polygenic risk for AD (p < 0.05). Analysis of individual-level summary neuroanatomical signature expression scores, quantifying each subject's continuous position along two disease dimensions, revealed that Dimension 1 signature expression correlated strongly with cognitive impairment, clinical severity, functional impairment, and genetic risk, while Dimensional 2 signature expression showed relatively weaker associations. These results reveal that AD encompasses at least two biologically distinct dimensions identifiable from structural MRI profiles and their distinct associations with cognitive, clinical, functional, and genetic profiles. These findings may aid in AD patient stratification for designing targeted therapeutic approaches in the future.

Limbic-predominant age-related TDP-43 encephalopathy neuropathologic change (LATE-NC) frequently co-occurs with Alzheimer's disease (AD) pathology and has been studied mainly in older populations. We examined its frequency and expression in 196 individuals with autopsy-confirmed intermediate-to-high AD neuropathologic change, enriched for younger-onset cases, integrating neuropathology with antemortem clinical, cognitive and neuroimaging data. LATE-NC occurred in 31% overall and 19% of individuals with symptom onset [&le;]65 years. Its frequency increased with older age at symptom onset, and its presence was associated with greater episodic memory impairment and smaller amygdala and hippocampal volumes. Among individuals with LATE NC, older age at symptom onset was associated with memory-first presentation and greater medial temporal involvement, including higher inferior-to-medial temporal metabolism ratios on FDG PET. Despite its lower frequency, LATE-NC co-pathology should still be considered in early-onset AD, where selected clinical features and neuroimaging markers described in older populations may be less applicable.

In addition to the well described beta-amyloid plates accumulation and tau hyper-phosphorylation, Alzheimers disease (AD) is accompanied by major changes in gene expression. Herein, we aimed at revealing master transcription factors (TFs) responsible for gene expression changes during AD progression. For this, we have used human brain organoids (BORGs) harbouring AD-related genetic mutations (APP-Swedish, PSEN1-M146V), which were traced over multiple time-points by bulk and spatially-resolved transcriptomics. By reconstructing gene regulatory networks (GRNs) that recapitulate BORG development, we have identified a subset of 110 AD-specific master TFs, and for 75 of them we retrieved KLF5 and/or KLF8 binding motifs within their promoters. Furthermore, 64 of the AD-specific TFs found in BORGs are significantly over-expressed on AD human patients samples, confirming the relevance of these factors beyond the context of the familial genetic mutations. Finally, we have demonstrated that this AD-specific regulome is at least partially controlled by the aberrant CREB3L2-ATF4 heterodimer previously described as being induced by the beta-amyloid plates deposition. Overall, these findings reconstitute the regulome behind the progression of AD and highlights key TFs as potential druggable targets for the treatment of the disease.

Background Fluid biomarkers enable the demonstration of the biological effects of novel therapies in Alzheimers disease (AD). However, longitudinal biomarker data are sparse and sample size calculations for fluid biomarkers are often lacking. Here, we provided longitudinal CSF and plasma AD biomarkers measured in samples collected in a placebo arm in a 1.5-year phase 2b trial, allowing us to study natural trajectories, required sample sizes and heterogeneity in early AD clinical trials. Methods We studied individuals from the placebo group (MCI due to AD (n=65) and AD dementia (n=41)) of the T-817MA trial (NCT04191486) with positive CSF AD biomarkers (mean age=69(7) years, Female=63%). Longitudinal biomarker changes in CSF (A{beta}42, A{beta}40, A{beta}42/40, pTau181, pTau217, NFL, tTau, YKL40, NRGN, ABL1, CHIT1, CLEC5A, ITGB2, MMP10, SDC4, SPON2, THBD) and plasma biomarkers (A{beta}42, A{beta}40, A{beta}42/40, pTau181, pTau217, NFL, GFAP) were analyzed with linear mixed-effect models. Required sample size estimates for predefined treatment effects were generated. Lastly, we investigated the influence of between person variability in biomarker change by simulating a randomized clinical trial (1:1) 10000 times, and assessed the group differences at 1.5 years. Findings Fourteen biomarkers changed over time, with the largest annual changes observed for plasma pTau217 (+9.8%), CSF MMP10 (+7.1%), and CSF NFL (+6.9%), and CSF A{beta}40 by (-4.0%), CSF pTau217 (-3.0%), and CSF NRGN (-2.5%). To show a 30% change, similar to biomarker effects of approved AD drugs, almost all markers required less than 45 patients per trial arm. To reach normalized levels, established CSF markers required lower sample sizes than plasma markers. The effects of heterogeneity over time were approximately twice as large in plasma compared to CSF. Interpretation These findings offer insights into the biomarker trajectories and power in early AD, supporting more informed endpoint selection and forming a frame of reference for the interpretation of treatment effects in clinical trials.

Background: The Centiloid (CL) scale standardizes global amyloid PET quantification and is widely used to define amyloid positivity. As a global summary measure, however, CL may not fully reflect the regional distribution of amyloid deposition, which can carry additional prognostic information about the rate of cognitive decline. Objective: To develop and externally validate a fixed, regional amyloid PET composite score that complements CL for predicting cognitive decline in Alzheimer's disease. Methods: The Regional Amyloid PET Score (RAPS) was derived from 82 FreeSurfer regions using machine learning with bootstrap stability selection to predict the rate of change in CDR-Sum of Boxes (CDR-SB) in 433 amyloid-positive ADNI [18F]florbetapir participants. The fixed nine-region weights were applied without retraining in a cross-tracer ADNI [18F]florbetaben subset (N = 71; largely overlapping the discovery participants) and two external validation cohorts, NACC SCAN (N = 1531; four tracers) and OASIS-3 (N = 428). Results: RAPS comprised nine regions. In ADNI, RAPS correlated more strongly with CDR-SB slope than CL and showed higher discrimination of rapid decliners (AUC 0.813 vs 0.713). Performance was directionally consistent across validation cohorts; in NACC SCAN, RAPS and CL independently predicted clinical progression. Cross-cohort meta-analysis of the three independent cohorts supported incremental discrimination beyond CL (pooled {Delta}AUC +0.066; I2 = 0%). Conclusions: RAPS, a fixed regional amyloid PET-derived score, may complement CL for prognostic stratification in Alzheimer's disease research.

Abstract Importance Recently, increasing studies have demonstrated that blood p-tau 217, the most promising diagnostic biomarker for Alzheimer's disease (AD), is increased and originated from muscle damage in amyotrophic lateral sclerosis (ALS) patients. These findings suggested that blood total p-tau 217 may partly derive from muscle damage. Thus, there is an urgent need for identifying blood brain-derived p-tau 217, but not blood total p-tau 217 which may contain muscle-derived p-tau 217, and other brain-derived biomarkers in order to reduce the potential peripheral interference in its adoption in assisting early diagnosis in AD patients. Objective To explore whether serum brain-derived p-tau 217 is a more promising diagnostic biomarker and has less peripheral interference than total p-tau 217 for AD patients in a large multicentre cohort. To examine whether serum synaptic vesicle glycoprotein 2A (SV2A) is a potential biomarker for assessing brain damage in AD patients. To examine whether serum p-tau 217 is a specific lower motor neuron (LMN) damage biomarker for ALS patients. Design, Setting, and Participants This cross-sectional study was conducted in 3 independent cohorts and a total of 1198 participants, including 325 AD patients, 235 ALS patients, 289 LMN disease controls (LMNDCs), 145 dementia controls (DDCs), and 204 cognitive intact healthy controls (CIHCs). Main Outcomes and Measures Serum brain-derived p-tau 217, total p-tau 217, SV2A, and NfL were measured based on single molecular detection technique. Results Serum brain-derived p-tau 217 was significantly increased in AD patients compared to ALS patients, DDCs, LMNDCs and CIHCs, while serum p-tau 217 was significantly increased in both ALS patients and AD patients compared to DDCs, LMNDCs and CIHCs after familywise error correction (p < 0.05). Serum SV2A was significantly decreased in AD patients than in other groups. Moreover, area under the curve for serum brain-derived p-tau 217 in differentiating AD from other groups were 0.927-0.954. Conclusions and Relevance Our findings suggest that serum brain-derived p-tau 217 and SV2A are more specific diagnostic biomarkers for reflecting brain damage and may not be disturbed by peripheral damage for AD patients. Moreover, we suggest that serum p-tau 217 is a specific LMN damage biomarker for ALS patients. Keywords: AD, ALS, p-tau 217, brain-derived p-tau 217, SV2A

Neuropsychiatric symptoms (NPS) are increasingly recognized as critical components of the disease progression in Alzheimer's disease (AD), yet their relationship with neurodegeneration remain poorly characterized. We investigated the multivariate relationships between structural MRI (sMRI)-based regional neurodegenerative biomarkers and NPS using the Alzheimer's Disease Neuroimaging Initiative (ADNI) and Knight Alzheimer Disease Research Center (Knight-ADRC) cohorts (N = 1,756). The machine learning regression models were compared for NPS prediction, and the best-performing deep neural network (named NPSNet) was integrated with three feature-importance methods: SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and Layer-wise Relevance Propagation (LRP). To establish a known ground truth, we introduced predefined regional perturbations into semi-simulated data and tested whether NPSNetSHAP, NPSNetLIME, and NPSNetLRP could recover them. The NPSNet strongly predicted NPS scores (pooled Spearman {rho} = 0.927, p < 2.2 x 10-3; fold-wise {rho} = 0.912-0.963) and NPSNetSHAP recovered all 100% perturbed regions, compared with 90% for NPSNetLIME and 40% for NPSNetLRP. In the experimental data (N = 1,756), the NPSNet produced the highest held-out correlation ({rho} = 0.387, p = 1.7 x 10-{superscript 1}), exceeding gradient boosting ({rho} = 0.301), support vector regression ({rho} = 0.266), and others ({rho} < 0.266). NPSNetSHAP identified individual-level regional contribution patterns relevant to NPS predictions. Comparing NPSNetSHAP attributions between cognitively normal (CN) and mild cognitive impairment (MCI)/AD groups revealed distributed multivariate neurodegenerative signatures of NPS, with the largest differences between CN and AD participants. This study introduces an explainable deep learning framework for identifying distributed, individualized neurodegeneration signatures of NPS burden across the AD continuum.

INTRODUCTION: Individuals with Alzheimer's disease (AD) exhibit substantial heterogeneity in the severity of vascular changes and neurodegeneration, making the identification of biologically meaningful subtypes across the AD spectrum critical for improving prognosis and tailoring interventions. We aimed to determine whether data-driven subtyping of individuals at risk for or diagnosed with AD, based on magnetic resonance imaging (MRI), can identify distinct trajectories of disease progression and clinical outcomes. METHODS: We analyzed baseline biomarkers and longitudinal clinical and cognitive assessments from the PREVENT-AD, CIMA-Q, ALLFTD, and ADNI cohorts, including 1396 participants. Regional atrophy and White Matter Hyperintensity (WMH) burden were derived from T1-weighted MRI. The SuStaIn algorithm was applied to baseline data to infer individuals' subtypes and stages based on MRI biomarker profiles. The resulting subtypes were compared in terms of baseline molecular biomarker levels and vascular risk factors, longitudinal cognitive decline trajectories, and clinical and neuropathological outcomes. RESULTS: Two distinct subtypes were identified across all cohorts: one characterized by early parahippocampal and cortical atrophy, and a second by initially elevated WMH volumes. Both subtypes showed abnormal tau and amyloid levels in- and ex-vivo, and subtype two exhibited increased vascular risk factors. Longitudinally, the atrophy-first subtype exhibited steeper cognitive decline than the WMH-first subtype. Accordingly, the atrophy-first subtype carried a higher risk of conversion to AD than the WMH-first group. DISCUSSION: These findings, replicated in four diverse and independent datasets, support imaging-based subtyping as a scalable and clinically translatable approach for capturing heterogeneity in AD pathophysiology and associated cognitive trajectories.

Open-source deep learning models promise scalable quantification of MRI-visible perivascular spaces (PVS), an emerging marker of glymphatic dysfunction in Alzheimer's disease (AD). Whether these models remain valid when applied to heterogeneous AD cohorts and preserve ground-truth derived biological associations has not been established. We benchmarked five open-source nnU-Net-based PVS segmentation models (nnU-Net T1, nnU-Net T1+FLAIR, mcPVS-Net, MedNet-PVS, ADNI-PVS-Net) against manual labels in 60 multi-site ADNI-3 participants balanced across cognitively unimpaired, mild cognitive impairment, and AD dementia, and in 12 publicly available scans from the VAscular Lesions object-level and segmentatiOn (VALDO) challenge. We evaluated overall, voxel, object, and lesion-wise Dice; stratified performance by region, diagnosis, scanner, PVS severity, and white matter hyperintensity (WMH) burden; and tested whether model-derived PVS burden reproduced ground-truth derived associations with AD-related biomarkers. Despite high internal performance, all models showed a specialist-generalist trade-off: the AD-trained ADNI-PVS-Net achieved highest internal performance (overall Dice = 0.58 {+/-} 0.14) but degraded most externally (Dice [&le;] 0.11), while MedNet-PVS and mcPVS-Net generalized externally (Dice 0.27-0.30) but systematically misestimated PVS burden. Models were relatively robust across scanner and diagnosis but inconsistent across PVS severity, and some misclassified up to 32% of WMH as PVS. Notably, only AD-trained models reproduced the expected PVS-amyloid biomarker relationship; other models returned a null result, despite comparable or superior Dice scores. These findings show that internal segmentation accuracy and external generalizability alone do not guarantee downstream inferential validity. Careful model selection and cohort-specific validation against manual labels should precede clinical inference in AD.

Accurate prediction of progression from mild cognitive impairment (MCI) to Alzheimer's disease (AD) is important for prognosis, patient management, and clinical trial enrollment. Cognitive and functional assessments are routinely used in memory clinics, but their relative predictive value remains unclear. We sought to identify which assessments are most predictive of 24-month progression from MCI to AD. We analyzed 2,430 participants with baseline MCI from the Alzheimer's Disease Neuroimaging Initiative (ADNI) who were classified by 24-month progression to AD. Extreme Gradient Boosting (XGBoost) models were trained using repeated stratified 5-fold cross-validation with 10 repetitions. We compared demographic and genetic variables, global cognitive measures (MMSE, ADAS-Cog13, CDR-SB, MoCA), episodic memory, executive function, functional status, and Everyday Cognition (ECog) questionnaires. The baseline clinical model (age, sex, education, APOE {varepsilon}4 status) achieved an area under the receiver operating characteristic curve (AUC) of 0.692. Episodic memory showed the highest predictive performance (AUC = 0.915), followed by the Functional Activities Questionnaire (AUC = 0.913). Combining episodic memory, functional assessment, and executive function achieved the best performance (AUC = 0.943, sensitivity = 0.857, specificity = 0.889). Among individual memory measures, Logical Memory Delayed Recall achieved the highest standalone performance (AUC = 0.896), whereas RAVLT Learning provided minimal incremental value. Episodic memory demonstrated the strongest predictive performance among the individual assessment domains evaluated of 24-month progression from MCI to AD, with functional assessment providing substantial complementary value. Streamlined assessment batteries emphasizing episodic memory and functional status may improve efficient risk stratification in memory clinics and AD clinical trials.

Identifying progressive dementia during its prodromal stage is a challenging yet critical task for healthcare providers. Traditional clinical assessments are often not sensitive enough to detect warning signs of cognitive decline. Machine learning (ML) methods have shown promise in modeling clinical trajectories associated with neurodegenerative disorders such as Alzheimer's disease (AD). Validation of these models' forecasting abilities, however, is widely lacking in the literature. To address this gap, we established a longitudinal cohort from the National Alzheimer's Coordinating Center (n = 15,727) comprising individuals who were cognitively normal (CN) or had mild cognitive impairment (MCI) at baseline. Cognitive status and clinical observations collected over up to 10 visits were used to train several ML classifiers. We employed the gradient-boosted tree ensemble method XGBoost with a custom feature-engineering pipeline. On non-imputed data, our models achieved an AUC-ROC of 0.9447 [0.9417-0.9476] and PR-AUC of 0.8554 [0.8497-0.8611] for the CN to MCI task, and an AUC-ROC of 0.9059 [0.8983-0.9134] and PR-AUC of 0.9214 [0.9156-0.9272] for the MCI to AD task. In a lead-time forecasting analysis spanning mean follow-up periods of 56.1 months (range, 9.1-214.1) for CN to MCI and 42.7 months (range, 9.1-126.1) for MCI to AD, the non-imputed models detected future progression before diagnosis in 29.58% [28.29%-30.87%] of CN to MCI progressors and 22.41% [21.36%-23.46%] of MCI to AD progressors. Among these individuals, the mean lead time was 33.04 months [32.52-33.56] and 17.95 months [17.71-18.19], respectively. False alarm rates among non-progressors were 12.6% [11.5%-13.7%] and 20.0% [18.4%-21.6%], respectively. These findings suggest that longitudinal machine learning models can identify future cognitive decline before diagnosis, indicating clinical utility for early dementia risk stratification.

INTRODUCTION: Voice data combined with large language models may detect cognitive impairment, yet an interpretable framework has not been formally established. We developed and validated a scalable and interpretable framework, the Neurocognitive Speech Taxonomy (NST). METHODS: NST maps 314 voice features to 7 neurocognitive domains derived from neuropsychology, speech pathology, neurology and related fields. The domains capture features including articulatory precision, cognitive-linguistic, executive fluency & planning, phonation & laryngeal control, prosodic modulation, lexical-semantic, and morphosyntactic complexity. We evaluated NST structural stability, clinical validity and racial-educational equity in three cohorts (total N = 1,479). Cohort data included demographic and neuropsychological evaluations and AD biomarker status. We tested whether NST scores distinguish cognitive status (unimpaired vs impaired) and AD-biomarker status, track cognitive change, and correlate with AD neuroimaging measures. RESULTS: NST showed stable cross-cohort structure (Mantel r = 0.71-0.77) and differentiated MCI (d = -0.39) and AD biomarker status (d = -0.23). NST tracked longitudinal cognitive change, correlated with hippocampal volume (r up to 0.28, P < .001), and showed smaller racial and educational disparities relative to standard cognitive testing (NST d = 0.03-0.33, MoCA d = 0.37-0.72). DISCUSSION: NST provides a validated neurocognitive construct that facilitates interpretation of AI-driven voice biomarkers of AD. This study was not registered in a public trials registry.

Neuroinflammation is a principal driver of brain atrophy and cognitive decline in Alzheimers disease (AD). Here we show that laromestrocel, a mesenchymal stem cell therapy, inhibits progressive brain inflammation in subjects (n=49) with mild AD, with the strongest effect seen in core AD brain regions: the left hippocampus (25 million cells (M)x4 monthly dose group; p<0.001, N=11, and 100Mx4 group; p<0.001; N=10), the left temporal cortex (25Mx4 group; p=0.007; N=11) and left parietal cortex (25Mx4, week 26; p=0.042; N=10), and the medulla (25Mx4 group and 100Mx4; p=0.007; N=11 and N=10) after 39 weeks compared with placebo (N=9). Reduced neuroinflammation correlated with reduced brain atrophy (hippocampus: R=-0.326; p=0.040; N=40) and improvements in clinical scores, particularly cognitive function (hippocampus x MoCA: R=-0.401; p=0.011; N=40). Treated subjects also exhibited reduced neurogranin (25Mx4 group; p=0.023; N=11) in blood plasma, consistent with reduced synaptic loss. Finally, blood biomarker analysis supported a convergent multimodal mechanism of action implicating peripheral immune engagement. Together these results show that long-lasting anti-neuroinflammatory effects of laromestrocel mitigate AD brain inflammation and are associated with markedly reduced brain tissue destruction and improved clinical function.

INTRODUCTION: The locus coeruleus (LC) is one of the earliest sites of Alzheimers disease (AD) pathophysiological changes. LC magnetic resonance imaging (MRI) intensity may reflect effects of AD pathophysiology on the LC, providing a biomarker of early-stage AD. This study examined relationships among MRI-derived LC intensity, plasma phosphorylated tau 217 (pTau217), and memory in older adults without dementia. METHODS: Independently living adults aged 59-89 years (N = 163) underwent neuropsychological assessment and T1-weighted fast spin-echo MRI. We quantified a bilateral LC contrast ratio relative to a pontine reference region, averaged across axial slices. Plasma pTau217 was measured using the Quanterix ALZpath p-Tau217 Advantage PLUS assay. For a subset (n = 127), a memory composite score was calculated. RESULTS: After adjustment for age, sex, and APOE4 carrier status, lower LC intensity was associated with higher plasma pTau217 concentrations (B = 0.17, P = .02). Each 1-SD decrease in LC intensity was associated with 53% higher odds of pTau217 positivity (P = .03). In models additionally adjusted for education, a significant LC x pTau217 interaction for memory (B = 0.18, P = .009) indicated that lower LC intensity was associated with worse memory at low (P = .01), but not high (P = .64) pTau217 levels. DISCUSSION: These findings support MRI-derived LC intensity as a marker of early AD-related processes and suggest that LC intensity may provide complementary information not captured by plasma pTau217 alone. Longitudinal studies are warranted to determine how these relationships change across the early stages of AD.

INTRODUCTION: Neurodegenerative research identifies biomarkers to confirm presence of disease and inform about risk for clinical symptoms. Expert guidance advises caution about disclosing individual research results (IRR), but participant interest remains high even when IRR may not inform individual prognosis. Existing studies of stakeholder attitudes emphasize Alzheimer's disease (AD) biomarkers. We explore participant attitudes toward IRR from the United States CADASIL Consortium (USCC), an observational study of Cerebral Autosomal Dominant Arteriopathy with Subcortical Infarcts and Leukoencephalopathy (CADASIL), the most heritable form of vascular dementia. METHODS: Since CADASIL research participant attitudes are unstudied and AD-focused guidelines for IRR may not generalize to populations with dominantly inherited conditions, we surveyed USCC participants using three 5-point Likert items and one open-ended question. Descriptive statistics were analyzed for Likert items. The distribution of responses to one item was directly compared to an AD participant survey. Open-ended responses underwent qualitative content analysis. RESULTS: We received 152 responses. The highest-rated reason to return IRR was "learn about my disease and its predicted course". The highest-rated IRR were imaging/MRI scans and cognitive testing. Hypothetical negative outcomes were rated as a little to somewhat concerning. USCC respondents rated reasons to return IRR higher than AD counterparts, with statistically significant differences for seven of eight items. In open-ended responses, the most frequent code was "IRR return will help improve my health and well-being". DISCUSSION: Most respondents expressed support for disclosure upon participant request. These findings could inform IRR guidance for CADASIL and other disorders and investigations of personal utility.

MicroRNAs (miRNAs) are implicated in Alzheimer's disease (AD), but research has focused on individual miRNAs and direct targets. Existing approaches to miRNA regulation in AD identify associations rather than causal effects, and few methods estimate multi-stage chains from miRNAs through target genes to target transcription factor (TF) cascades. We developed the LSD pipeline (LASSO-SEM-DoWhy), integrating LASSO feature selection, multi-stage structural equation modeling, and DoWhy causal inference to identify and validate miRNA causal pathways in AD. Applying LSD to six blood miRNA and brain mRNA datasets, we identified four LSD-validated miRNAs (miR-30d-5p, miR-92a-3p, miR-296-5p, miR-193a-5p) as AD biomarkers, achieving >86% ROC accuracy in an independent validation cohort. Several miRNAs with no significant direct association with AD showed significant effects when estimated through their target networks, while others significant in direct analysis were not supported at the network level, underscoring the value of network-level analysis. Extending to the TF layer revealed complete miRNA [-&gt;] targets [-&gt;] TF cascades [-&gt;] AD causal chains, with HMGA1, NKX2-3, and PRRX2 as key intermediaries. Confirmed classic pathways converge primarily on tau pathology and synaptic dysfunction. miRNA effects were largely age-independent, suggesting miRNAs act as early initiators of AD pathogenesis. Beyond AD, the LSD pipeline provides a generalizable framework for uncovering causal regulatory mechanisms in other diseases.

Background. Deep-learning models predict mild cognitive impairment (MCI)-to-Alzheimer's disease (AD) conversion from structural MRI with high accuracy. However, they are typically validated within a single cohort and judged on discrimination alone. Whether their risk scores are biologically grounded, and whether they generalise to independent data, remains unclear. Methods. We applied the ADNI-trained Temporal Adaptive Fusion Network (TAF-Net), without retraining, to 101 MCI participants from OASIS-3 (25 converters, 76 stable) and tested whether its conversion-risk scores were grounded in independent structural, functional, and cognitive markers of AD. Regional atrophy rates were derived from longitudinal FreeSurfer, cognition from longitudinal MMSE and CDR-Sum-of-Boxes, and baseline resting-state functional connectivity from fMRIPrep, in structural and functional subsamples of 41 and 40 participants. Associations used rank-based statistics with false-discovery-rate correction. Results. Higher TAF-Net risk tracked faster atrophy in medial-temporal AD-signature regions but not in AD-spared cortex, an anatomically specific coupling that survived adjustment for global atrophy. In external validation, risk discriminated converters (AUC = 0.72), comparable to native atrophy and strongly concordant with it; atrophy statistically accounted for the model's prognostic signal. Discrimination transferred but calibration did not: the two lowest tertiles of risk were assigned near-zero probability yet converted at 15%. Risk also tracked the multi-year rate of cognitive decline and, at baseline, was associated with reduced within-network functional-connectivity integrity, concentrated in salience and default-mode hubs; longitudinal functional analyses were underpowered. Conclusions. A conversion model trained on one cohort produced risk scores that, in an independent cohort, were grounded in the structural, functional, and cognitive hallmarks of Alzheimer's disease, supporting biological validity and external generalisation. The score, however, largely re-expresses the neurodegenerative substrate captured by structural atrophy. Rank ordering transferred across cohorts but absolute risk did not, so the score requires recalibration before its values can be interpreted as individual probabilities.

Alzheimer's Disease (AD) remains a leading cause of cognitive decline with no known cure, motivating the development of therapies that slow neurodegeneration. Rapamycin, an FDA-approved inhibitor of the mammalian target of rapamycin (mTOR) pathway, has demonstrated promising anti-aging and neuroprotective effects. However, characterizing its treatment effects and identifying the biological factors that contribute to treatment response remain challenging because of complex interactions across multiple biological systems and the limited availability of patient data. In this work, we propose a three-stage multimodal deep learning framework called TreatmentFormer for predicting rapamycin treatment status from heterogeneous biomedical data including both brain imaging data and tabular data (e.g., microbiome profiles, blood-based biomarkers, cerebral blood flow measurements, and clinical variables (e.g., gender, age, and body mass index)). First, a Random Forest-based feature selection module reduces noise in high-dimensional tabular data while preserving representation across modalities. Second, modality-specific encoders map imaging and tabular inputs into a shared latent space via self-supervised contrastive learning, enabling alignment across modalities. Finally, a transformer-based architecture integrates these representations to capture cross-modal interactions and perform treatment classification. Evaluated on a cohort of 23 participants with baseline and post-treatment timepoints, TreatmentFormer achieves an average prediction accuracy of 71.25\% across 10 independent test runs. Despite the challenges of small sample size and heterogeneous data, the model demonstrates stable and consistent performance. Post hoc SHAP-based feature analysis further identifies key biomarkers associated with treatment response, particularly within blood-based and inflammatory modalities. These findings demonstrate that combining feature selection with multimodal representation learning provides a promising and robust approach for modeling treatment effects in small-sample biomedical studies. Importantly, this framework may have significant implications for clinical research and medical applications by identifying the biological features and quantitative measurements that drive individual responses to rapamycin. Such insights could facilitate the development of predictive biomarkers, improve patient stratification, and ultimately inform future approaches to AD diagnosis and therapeutic development.

Cellular mechanisms in early Alzheimers disease (AD) remain poorly understood despite generation of extensive observational multiomic data. Weighted integration across diverse datasets can prioritise convergent mechanisms and targets. We conducted a systematic review to identify datasets profiling cell-type-specific changes in early AD. We integrated multiomic findings using Meta-Analysis by Information Content (MAIC), a computational method which combines genomic data from diverse experimental sources. Electron transport chain dysfunction in mitochondria emerged as a shared pathway in neurons, microglia, and astrocytes, and Wnt signalling was a prominent astrocyte-specific signature. Pathological neuro-glial signalling mechanisms included APP-CD74, APOE-SORL1, and WNT-FZD/LRP6. Integration with pharmacological data identified therapeutics already under investigation in AD, such as guanfacine, and candidates not previously prioritised, such as eltrombopag and encorafenib. We present a systematic analysis of cell-type-specific changes in early AD to prioritise pathways for investigation and therapeutics for repurposing.