Aβ-related Cerebrospinal Fluid Proteins in Alzheimer’s Disease Reflect Gene Expression Levels Found in The Healthy Cortex

Aβ-related Cerebrospinal Fluid Proteins in Alzheimer’s Disease Reflect Gene Expression Levels Found in The Healthy Cortex | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Aβ-related Cerebrospinal Fluid Proteins in Alzheimer’s Disease Reflect Gene Expression Levels Found in The Healthy Cortex Pedro Rosa-Neto, Seyyed Ali Hosseini, Yi-Ting Wang, Nesrine Rahmouni, and 33 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8428920/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract A growing body of evidence indicates that CSF proteomic signatures shift with increasing brain amyloidosis in Alzheimer’s disease (AD). However, it remains unknown whether protein profiles within cortical regions that are vulnerable to early amyloid-beta (Aβ) deposition contribute to, or predict, CSF Aβ-related protein measures. To address this question, we examined 220 cognitively unimpaired (CU) and cognitively impaired (CI) older participants from the TRIAD cohort who had Aβ- and tau-PET scans, MRI, and CSF NULISA™seq CNS panel data. Aβ-related hierarchical clustering identified 12 clusters corresponding to biologically interpretable gene ontology processes, supported by bootstrap resampling, silhouette analysis, and dynamic tree cutting. Clusters’ composite scores for tau-markers, neuronal-injury, and APOE4 correlated strongly with global neocortical Aβ-PET SUVR (p < 0.001) and showed elevated odds ratios (OR) of Aβ positivity (ORs: 10.4, 2.4, and 4.0, respectively). The Aβ42 cluster, as expected, was inversely associated with brain Aβ burden and predicted reduced Aβ positivity (OR = 0.13). Voxel-wise analyses revealed distinct spatial signatures for each cluster, such as associations in cortical regions (tau-markers, neuronal-injury, and APOE4 clusters), white matter (axonal metabolic injury cluster), and periventricular areas (synaptic signaling and cellular response clusters). The magnitude of Aβ-related cortical associations and CSF protein clusters correlated with the magnitude of regional mRNA expression derived from the Allen Brain Atlas. Our findings support the notion that CSF protein expression reflects underlying regional mRNA expression in cortical regions vulnerable to AD pathophysiology. Biological sciences/Neuroscience/Diseases of the nervous system/Alzheimer's disease Biological sciences/Neuroscience/Neural ageing Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Cerebrospinal fluid (CSF) is in direct communication with the brain and therefore reflects its biological state, offering valuable in vivo insight into the progression of dementia 1 . Advances in proteomic and imaging biomarkers hold promise for personalized approaches by expanding our understanding of Alzheimer’s disease (AD) biology 2 . The advent of ultrasensitive antibody-based proteomic platforms, such as nucleic acid-linked immune-sandwich assay sequencing (NULISA™seq), enables a more comprehensive characterization of AD pathophysiology across the disease continuum 3 . In fact, CSF protein alterations occurring in AD converge on brain regions with a high burden of amyloid-beta (Aβ) and tau aggregates 4 . However, one should consider whether these changes merely represent downstream responses to the spread of Aβ or might also reflect features of specific cortical regions 5 . To address this question, we deployed regional cortical transcriptomics profiles from publicly available datasets to enable assessments of differential susceptibility or resilience to disease processes at the population level 6,7 . Here, we aimed to integrate neuroimaging, clinical data, and fluid biomarker proteomes by hierarchical clustering to identify and validate multi-modal biomarkers that reflect different stages of Aβ pathology. Additionally, we assessed the relationships between the population-based, spatially resolved gene expression profiles and CSF protein signature clusters relevant to Aβ 8 . We tested the hypothesis that cortical transcripts from brain regions vulnerable to Aβ aggregation are associated with Aβ-related protein signatures in AD. Results The study framework depicted in Fig. 1 provides an overview of the data analytical procedures. It starts with patient preparation for Aβ- and tau-PET and MRI scanning, as well as lumbar puncture for CSF collection, which was processed using the NULISA™seq CNS Panel to measure 120 neurodegenerative disorder-associated proteins. The next step involved registering PET images to their native MRI space and measuring global neocortical Aβ. This was followed by hierarchical clustering, subsequent validation processes, and gene ontology (GO) analysis to capture nuanced patterns in the CSF data related to Aβ aggregation in the brain. Finally, composite scores for biomarkers within the same cluster (identified by their distance to Aβ-PET) were calculated, validated, and statistical analyses were conducted. To identify biomarker profiles reflecting Aβ pathology, we performed hierarchical clustering of the proteins. The resulting clustering dendrogram is shown in Fig. 2a. Biomarkers with pairwise distances below 20 were considered part of the same cluster. We found 12 clusters from 120 different biomarkers, as shown in Table 2. To assess how well proteins tend to co-cluster consistently against perturbations introduced via sampling, we created a 120 × 120 matrix for bootstrapping co-membership stability with 1,000 resamples (Fig. 2b). The resulting structure revealed that the majority of cluster boundaries were highly reproducible, with several biologically coherent modules, including those comprising cluster 1, cluster 4, and cluster 7, tending towards perfect stability (co-membership > 0.95), while others, with more variable signaling pathways, tend towards lower but still visible stability. To quantify the degree of internal separation between the protein clusters in the CSF, we computed silhouette widths for each protein in the 12-cluster solution derived from Ward’s hierarchical clustering (Fig. 3a). The global silhouette scores revealed that they are rather moderate on average (mean silhouette score = 0.255), as expected for proteins undergoing common expression in CSF. Despite this relatively moderate degree of global separation, several clusters demonstrated clearly positive silhouette distributions, indicating well-structured cluster cores, whereas others displayed mixed or negative silhouettes, consistent with diffuse or cross-pathway biological roles. To determine whether the predefined tree-cut height (distance = 20) imposed artificial cluster boundaries, we compared the fixed 12-cluster solution with a data-driven set of cluster partitions generated by scanning cluster numbers between k = 8 and k = 24 (Extended Data Table 1). Silhouette scores increased gradually with larger k, reaching a maximum at k = 20 (silhouette = 0.264), while similarity to the fixed 12-cluster solution (as measured by the adjusted Rand index (ARI)) decreased from ARI = 1.00 at k = 12 to ARI = 0.78 at k = 20. This pattern indicates that the 12-cluster solution lies at a stable structural plateau of the dendrogram, while higher-k decompositions introduce finer-grained subdivisions without altering the core biological modules. Next, using GO analysis, we attributed these 12 clusters to tau markers (cluster 1), neuronal injury (cluster 2), axonal metabolic injury (cluster 3), synaptic signaling (cluster 4), cellular response (cluster 5), Aβ42 (cluster 6), cell differentiation (cluster 7), immune (I) (cluster 8), carbohydrate derivative metabolic processes (cluster 9), choroid plexus (cluster 10), immune (II) (cluster 11) and apolipoprotein-E4 (APOE4) (cluster 12) (Fig. 3b & Extended Data Table 2). To analyze the association of the clusters with brain Aβ accumulation, we computed for each cluster a composite score of the proteins included in the cluster. We then correlated them with global neocortical Aβ SUVR and compared them based on Aβ status (Fig. 4). We found that, in Aβ-positive participants, the composite scores for the tau markers, neuronal injury, immune (II), and APOE4 clusters were higher, while the composite score for the Aβ42 cluster was lower. The other clusters did not show differences based on Aβ status. Additionally, tau markers, neuronal injury, axonal metabolic injury, immune (II), and APOE4 clusters were positively correlated, whereas the Aβ42 cluster was negatively correlated with neocortical Aβ SUVR. Next, we conducted a linear model voxel-wise analysis to investigate the impact of each cluster on Aβ load in older adults (>65 years; N = 165), corrected for age and sex (Fig. 5a). The tau markers and neuronal injury clusters showed strong to moderate positive associations with Aβ load over the cortex. The axonal metabolic injury cluster showed a positive association with Aβ load in the white matter, and the synaptic signaling and cellular response clusters showed positive associations within the ventricles, and not the brain parenchyma. The Aβ42 cluster was associated with Aβ load in the cortex. The cell differentiation and immune (I) clusters did not show associations in the voxel-wise analysis. The carbohydrate derivative metabolic processes cluster showed a positive association with Aβ load in the thalamus and brainstem regions. The choroid plexus marker cluster showed positive correlations within the ventricles and cisterns compatible with the choroid plexus. The immune (II) cluster showed associations in the hippocampal/parahippocampal regions and the lateral temporal cortices. The APOE4 cluster showed a strong positive association with Aβ load over the cortex. Next, we asked whether the CSF protein clusters are associated with the brain-region specific transcriptional profiles. We computed the voxel-wise gene expression intensity of each cluster by averaging the mRNA expression intensity of proteins within the same cluster derived from Allen Human Brain Atlas of healthy brain tissue (Fig. 5b). The analysis indicated that the tau markers, APOE4, Aβ42, axonal metabolic injury, and neuronal injury clusters displayed the strongest mean expression in the brain, ranked from highest to lowest. To investigate the relationship between gene expression intensity and the regional associations with Aβ load, for each cluster, we computed a mask of the brain regions that showed associations between the cluster’s composite score and Aβ load, averaged the β-values in the mask, and applied it to the corresponding mRNA expression intensity map. Thereby, we were able to analyze whether the brain-region specific mRNA profiles are associated with the Aβ-associated CSF proteomes. Notably, we detected a correlation between the regional mRNA-expression intensity and the Aβ-related cortical associations with CSF protein clusters (Fig. 5c). We next used logistic regressions to verify the risk of Aβ positivity based on the cluster composite scores. Our results revealed that the tau markers (OR = 10.42, CI [5.21, 20.83]), neuronal injury (OR = 2.40, CI [1.54, 3.75]), and APOE4 (OR = 3.97, CI [2.58, 6.11]) clusters were risk factors for Aβ positivity. Furthermore, the Aβ42 cluster (OR = 0.13, CI [0.08, 0.24]) was inversely associated with Aβ positivity while the other clusters were not (Fig. 5d & Extended Data Table 3). Finally, we aimed to analyze the correlations between the different clusters. Our results revealed distinct correlation patterns among the biomarker clusters (Fig. 6a & Extended Data Table 4). The tau markers cluster was correlated with all other clusters except the cell differentiation, axonal metabolic injury, and immune (II) clusters. The neuronal injury cluster showed correlations with all clusters except the cellular response, Aβ42, axonal metabolic injury, and immune (II) clusters. The axonal metabolic injury cluster was correlated with all other clusters except the Aβ42, choroid plexus marker, and immune (II) clusters. The synaptic signaling cluster demonstrated correlations with all other clusters. The cellular response cluster was correlated with all clusters except the neuronal injury cluster. The Aβ42 cluster was correlated with all clusters except the neuronal injury, axonal metabolic injury, and choroid plexus marker clusters. The cell differentiation cluster was correlated with all clusters except the tau markers, Aβ42, choroid plexus marker, and APOE4 clusters. The immune (I) cluster was correlated with all clusters except the choroid plexus marker marker cluster. The carbohydrate derivative metabolic processes cluster showed correlations with all clusters. The choroid plexus marker cluster was only correlated with the synaptic signaling, cellular response, and carbohydrate derivative metabolic processes clusters. The immune (II) cluster was correlated only with the synaptic signaling, Aβ42, cell differentiation, immune (I), and carbohydrate derivative metabolic processes clusters. The APOE4 cluster was correlated with all clusters except the cellular response, cell differentiation, choroid plexus marker, and immune (II) clusters. Finally, we observed a similar trend of correlation of each clusters’ composite score over Aβ stages (Fig. 6b). Discussion In this study, a multidimensional clustering approach was designed to examine the topographic association between Aβ-related CSF biomarkers and cortical gene expression profiles. First, we aimed to identify multimodal biomarker signatures capturing distinct stages of Aβ pathology. We further sought to delineate how spatially resolved gene-expression profiles relate to these Aβ-related protein clusters. To establish the reliability of these protein modules, we performed a comprehensive robustness assessment, comprising resampling with the 1,000 bootstrap iterations. We also conducted silhouette analysis and dynamic tree-cut comparisons to ensure that clusters were biologically meaningful and reproducible patterns rather than sampling artifacts or arbitrary threshold-based dendrogram points. Our approach captured distinct biological processes, including tau pathology, neuronal injury, axonal metabolic deficits, synaptic signaling, cellular response, cell differentiation, carbohydrate derivative metabolic process, and two immune clusters designated (I) and (II). Importantly, some clusters correlated with Aβ-PET burden, highlighting potential markers of disease severity and risk. Although Aβ aggregation has been conceptualized as a global cortical event, our results showed that the associations between CSF biomarker clusters and Aβ burden are regionally specific, rather than globally distributed across the brain. The strongest associations with Aβ burden were observed in clusters related to tau phosphorylation, neuronal injury, and APOE4, and occurred in precisely those brain regions in which the same underlying molecular machinery is highly expressed. This points to an important role of pre-existing transcriptional topology in mediating regional susceptibility to Aβ deposition. These clusters appeared across several Aβ staging cortical regions but were largely absent in areas with low baseline gene expression of the respective biological pathways. This suggests that inherent regional biology (not just the disease process) influences where insoluble Aβ builds up. In addition, our data support the framework that region-specific transcriptional profiles might also explain the resilience and vulnerability to AD 6,7 . Hierarchical clustering grouped 120 CSF biomarkers into 12 clusters based on their patterns of co-expression. The tight correlation of expression clusters with Aβ load highlights the dominance of Aβ pathology in AD. Our clustering framework was shown to be robust by conducting extensive validation analyses. Resampling using 1,000 bootstrap iterations, silhouette profiling analyses, and dynamic tree cuts validated that our identified modules are indeed representative of stable protein subcommunities as opposed to threshold-driven or sampling-dependent artifacts. GO analysis further validated the clustering by assigning biological themes to the identified clusters. However, in some cases, the nomenclature was modified to better fit the context of AD pathophysiology and improve interpretability. Our data-driven approach corroborates that tau-phosphorylation and neuronal injury clusters are correlated with global neocortical Aβ SUVR and replicates the concept that tau phosphorylation and neuronal injury are activated as Aβ accumulates, potentially enhancing neurodegeneration. This is in line with the literature demonstrating that tau markers rise with Aβ and cause downstream synaptic dysfunction, inflammation 9,10 , and cognitive decline 11 . Our results also showed high intercorrelations between clusters, suggesting interrelated pathological pathways. The observation that the tau markers cluster correlated with most other clusters is consistent with the broad implication of tau in AD pathology 12 . The neuronal injury cluster also correlated with most clusters, consistent with the known interaction between neuronal damage and inflammatory processes 13 . Subsequent logistic regression analysis identified tau markers and neuronal injury as predictors for Aβ positivity. Notably, the axonal metabolic injury cluster exhibited a positive correlation with global neocortical Aβ SUVR, highlighting the importance of energy homeostasis for brain health. These findings are in line with previous transcriptomics and functional studies that causally linked an impaired glucose metabolism with AD progression 14,15 . Together, our data-driven approach supported the well-known associations between tau burden or neuronal injury and Aβ deposition 16 . The immune (II) cluster highlights the dominance of C-reactive protein and shows a positive association with global neocortical Aβ SUVR, predominantly observed in later stages of Aβ deposition. CRP is a large molecule (115 kD) that possibly reaches the brain via a leaky blood-brain barrier (BBB), validating the role of systemic inflammation in modulating neurodegeneration 17 . The correlation patterns support the notion that AD may be powered by an intricate interdependence among Aβ accretion, tau pathology, metabolic disturbances, and neuroinflammatory responses, each uniquely contributing to the disease. The axonal metabolic injury and immune (II) clusters were also positively associated voxel-wise with Aβ-PET, indicating the involvement of damaged metabolic processes and systemic inflammation in Aβ-associated pathology 18 . Inter-correlation analysis showed that axonal metabolic injury correlated with most other clusters, suggesting metabolic dysfunction as an early 19-21 and central aspect of AD pathophysiology 22-24 . Astrocytes play a pivotal role in metabolism as the primary regulators of cerebral glucose uptake, lactate shuttling, and metabolic support to neurons. Furthermore, astrocytes also form the metabolic backbone that couples synaptic activity to energy demand 25,26 . Given that astrocytes are also key mediators of neuroinflammatory responses 27 , and the axonal metabolic injury cluster is tightly connected with the immune (II) cluster, our results highlight the intersection between inflammation and metabolic changes. Interestingly, the synaptic signaling and cellular response clusters showed region-restricted associations, with strong positive associations in ventricular regions but not in the cortex, supporting the notion that impaired CSF dynamics or dysfunctional clearance mechanisms contribute to Aβ deposition in periventricular spaces 28 . These results align with recent evidence that glymphatic cortical aggregation is an essential contributor to Aβ clearance failure in AD 29 . Additionally, correlation among clusters further supports that inflammation-associated periventricular pathologies are key players in AD progression 30 . The Aβ42 cluster also exhibited high negative correlation with global neocortical Aβ SUVR, corroborating the well-known relation between CSF Aβ42 and brain fibrillar Aβ plaques in the brain 31-33 . The negligible correlation between the cell differentiation and immune (I) clusters and Aβ-PET suggests that either low cortical expression or the association with AD pathophysiology occurs via mechanisms independent of regional Aβ deposition. For example, interferon signaling is associated with tau pathology in AD 34,35 , and neurodegeneration in AD 36 , but not with cortical Aβ aggregates. The carbohydrate derivative metabolic processes cluster correlated with nearly all clusters, indicating the central role of the astrocytic response and lipid metabolism 37-39 . The choroid plexus-related cluster showed minimal correlation with other CSF protein clusters, suggesting a distinct and specific role in AD progression. Subsequent logistic regression analysis identified the APOE4 cluster as a predictor for Aβ positivity. APOE4 is the strongest and most well-established genetic risk factor for AD, influencing Aβ aggregation and clearance 40 . The strong statistical correlations between the Aβ-PET burden and gene expression clusters further suggested that these molecular networks likely play an active role in disease development. Intriguingly, single-cell sequencing and spatial RNA-sequencing studies have shown that brain-region-specific transcriptomes reflect vulnerability to AD pathologies and neurodegeneration 6,7,41 . For example, cortical regions with high transcriptional expression are predominantly implicated in synaptic function 42 , lipid metabolism 43 , or activation of the innate immune response 44 , where Aβ deposition occurs earliest and is most severe 45 . Conversely, regions with low expression within those transcriptional circuits remain resistant to pathological change even in advanced disease 46 . One of the major strengths of this study is that our recovered clusters represented meaningful data structures rather than being assigned at random to dendrogram slices. An analysis of cluster robustness for stability indices, silhouette, and dynamic tree cut indices established that our 12-cluster solution identified stable inference units for expression patterns. Second, beyond looking at the pure associations of clusters with Aβ load, we demonstrated how spatially resolved gene-expression profiles in healthy brain, are related to these Aβ-relevant protein clusters. Third, we explored the inter-correlation of these clusters as well, which may be relevant to other processes in AD. Thus, our findings support the framework that the brain-region specific vulnerabilities may predict the disease progression, making them potential therapeutic targets for disease-modifying therapies. These results emphasize that AD progression emerges from an interaction between pathology and pre-existing biological terrain, and highlight the importance of targeting region-specific molecular vulnerabilities in early-intervention strategies. Limitations Our study bears several limitations that must be carefully considered. First, as a single-center study, the replication of our findings in other populations, particularly those with more heterogeneous and larger sample sizes, is required. Secondly, the NULISA™seq panel is a relatively novel biomarker measurement tool, therefore ongoing verification of the biomarkers across larger, multi-ethnic, and longitudinal samples will be useful in ascertaining their usefulness in clinical utilities. Third, the participants in the TRIAD cohort are not representative of the diversity worldwide. This limits extrapolation to other genetic backgrounds, lifestyle determinants, and environmental exposures from other populations. Finally, our study systematically excluded participants with comorbidities, especially those with neurological conditions. Although this was required to minimize confounding factors, it restricts the generalizability of our results to people with complex diseases. Several neurodegenerative diseases are characterized by aberrant protein aggregation and can potentially affect the parameters examined in our study. Subsequent research should enroll patients with comorbid neurological illness to ascertain how such conditions affect the biomarkers and imaging parameters evaluated in our investigation. Future studies should comprise longitudinal analysis to determine if these gene expression changes happen prior to or as a result of Aβ deposition, which could provide insights into the earliest molecular drivers of AD. Conclusion Overall, our study provides a multidimensional framework of AD pathophysiology. We identified and validated biological CSF modules with clear relevance to Aβ pathology that may be useful for diagnosis and individualized disease monitoring in early stages of the disease. These CSF signatures strongly correlated with brain-region specific transcriptional profiles mirroring the accumulation of Aβ and therefore suggesting that brain-region specific vulnerabilities may predict the different CSF modules. Clinically, this study lays the groundwork for improved biomarkers and personalized treatments, from blood-based tests that identify at-risk individuals to multi-drug therapies targeting both Aβ and downstream network disruptions. Methods Participants We recruited 420 participants from the Translational Biomarkers of Aging and Dementia (TRIAD) cohort, including 267 CU older adults, 59 individuals with MCI due to AD, and 94 individuals with AD dementia. A total of 32 individuals were excluded due to a cognitive impairment not due to AD (e.g., frontotemporal dementia or Aβ-PET negative results). Additional exclusion criteria included inadequately managed systemic conditions, active substance abuse, recent head trauma, major surgery, or contraindications for medical imaging. Participants in the CU group had a Clinical Dementia Rating (CDR) score of 0. Those with MCI had a CDR score of 0.5 47 . Participants diagnosed with AD dementia had a CDR score of 1 or 2 48 . Among the 420 participants, 384 underwent [ 18 F]MK6240 PET imaging, 392 underwent [ 18 F]AZD4694 PET imaging, and 403 underwent MRI. A subset of participants (N = 225) had NULISA™seq CNS CSF Panel incorporating 120 biomarkers. To ensure all participants had all imaging modalities and the CNS CSF NULISA™seq panel, the final sample size was 220 participants including, 142 CU older adults, and 78 CI, including 34 mild MCI due to AD and 44 dementia due to AD. Participants were recruited from the community or the McGill Centre for Studies in Aging memory clinic between November 2016 and March 2024. The study was approved by the Montreal Neurological Institute PET Working Committee and the Douglas Mental Health University Institute Research Ethics Board. Written informed consent was obtained from all participants. This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. Imaging Acquisition and Reconstruction A 3 Tesla Siemens MAGNETOM MRI scanner with a 32-channel head coil was used to acquire T1-weighted images via an Ultrafast Gradient Echo 3D sequence (isotropic 1 mm voxels, TR: 2300 ms, TE: 2.96 ms, FoV: 256 mm, flip angle: 9°). [ 18 F]AZD4694 and [ 18 F]MK6240 PET scans were performed using a Siemens high-resolution research tomograph dedicated to brain imaging. Each participant received a radioactive dose of 5-7 mCi for each PET scan. Tau-PET images were acquired 90-110 minutes post-injection of [ 18 F]MK6240, with four frames (4 × 300 s) 49 . Aβ PET images were collected 40-70 minutes post-injection of [ 18 F]AZD4694, with six frames (6 × 600 s). Image reconstruction for both PET tracers was performed using a sequential subset expectation-maximization algorithm on a 4D volume. To correct for motion, dead time, decay, as well as random and scattered coincidences, a 6-minute transmission scan with a rotating ¹³⁷Cs point source was conducted after each PET acquisition. PET Image Processing PET images were initially linearly registered to the T1-weighted MRI space before undergoing both linear and nonlinear registration to the Alzheimer’s Disease Neuroimaging Initiative (ADNI) standard template using ANTs (version 2.2.0). After transformation into ADNI space, the images were spatially smoothed to an 8 mm full width at half maximum resolution for the Modulation Transfer Function curve. To minimize meningeal spillover into adjacent brain regions and enhance signal localization, a stripped meningeal mask, generated with FSL (version 6.0.2), was applied to the PET images 50 . The standardized uptake value ratio (SUVR) for [ 18 F]MK6240 PET was calculated using the inferior cerebellum gray matter as the reference region 51 . For [ 18 F]AZD4694 PET, the entire cerebellum gray matter served as the reference region 52 . The global neocortical [ 18 F]AZD4694 SUVR composite encompassed the neocortex, including the precuneus, prefrontal, orbitofrontal, parietal, temporal, and cingulate cortices, with a positivity threshold of SUVR > 1.55, as previously reported 52 . The summary [ 18 F]MK6240 SUVR composite was derived from the temporal meta-region of interest (ROI), which included the entorhinal cortex, amygdala, hippocampus, and the parahippocampal, fusiform, lingual, inferior temporal, and middle temporal gyri 53 . Cortical Aβ burden was further characterized using the four-stage cortical Aβ PET staging framework as described previously 54 , applying the corresponding regional staging masks to derive individual Aβ progression stages. NULISA TM seq assay CSF samples were processed in the NULISA™seq CNS Panel for 120 neurodegenerative disorder-associated proteins. Next-generation sequencing data were processed with the NULISA™seq algorithm (Alamar Biosciences) and subsequently underwent intra- and inter-plate normalization, as previously reported 3 . Control-normalized counts were log2 transformed and expressed as NULISA™seq Protein Quantification Units, which served as the primary measurement in this study. Further descriptions of the NULISA™seq assay, data handling, and normalization are available 55 . Gene expression quantification The messenger ribonucleic acid (mRNA) expression intensity of each cluster was computed by averaging the regional mRNA expression values of all biomarkers included in that cluster. The magnitude of regional mRNA expression reflects the baseline transcriptional activity of each gene across anatomically defined regions of the healthy human brain, as quantified in the Allen Human Brain Atlas. The processed gene expression data were obtained from the open-source database (https://www.meduniwien.ac.at/neuroimaging/mRNA.html) as previously elaborated 56,57 . For the cluster composed of multiple phospho-tau biomarkers (cluster 1: phosphorylated tau (Ptau)181, Ptau217, and Ptau231), the expression of MAPT was used as the representative regional transcript, and for the APOE4 cluster (cluster 12), APOE mRNA expression values were used directly, while all remaining clusters used the averaged regional expression of their respective biomarker genes. Statistical analyses Hierarchical clustering Hierarchical clustering was conducted on CNS CSF biomarkers of CI and CU older adults after covariate adjustments. The analysis was executed using Python software integrated with pandas, scipy, matplotlib and scikit-learn library tools. Each biomarker underwent statistical adjustments through a multiple linear regression model using age, binarized sex (0 for male and 1 for female), and temporal meta-ROI tau SUVR as independent variables in order to minimize the influence of confounding variables. For each biomarker, the residuals were derived from the subtraction of predicted values from the observed values. These residuals representing the covariate-adjusted biomarker values were used for the following steps. This ensured that each biomarker was only related to Aβ pathology after correcting for all covariates. Hierarchical clustering was then performed on the covariate-adjusted biomarker values to identify patterns in the data. The adjusted values were used as the input for computing pairwise Euclidean distances between them, which were processed through the Ward linkage method from the Python package scipy . Ward's method minimizes the total within-cluster variance, meaning at each step, the biomarkers with the smallest increase in the overall variance when merged are combined. The linkage function generated a linkage matrix for hierarchical cluster encoding. The dendrogram created by the dendrogram function from scipy function displayed the clustering procedure through visual representation based on the linkage matrix data. The dendrogram displays both the sequence of cluster merges along with the distances used for cluster combination. By examining the dendrogram, clusters of biomarkers with similar adjusted profiles were identified. This grouping helps in understanding how different biomarkers relate to one another after correcting for the effects of the covariates. The final clusters were determined by selecting a threshold on the dendrogram, effectively cutting the tree at a specified level of similarity. Biomarkers with pairwise distances below 20 58 were considered part of the same cluster. Each cluster then underwent quantification to identify biological patterns based on the selected biomarkers. As the biomarkers in one specific cluster were systematically similar, for each biomarker, Z-scores were calculated, and the composite score for each cluster was then computed per subject, representing the cluster variation as a single marker, using the following formula: where N represents the number of biomarkers in each cluster. Cluster robustness and validation For analyzing the stability and internal validity of the protein clusters in the CSF, we applied a three-step robustness framework comprising: (i) Bootstrap resampling, (ii) Silhouette analysis, and (iii) dynamic tree cutting. Bootstrap cluster stability Cluster reproducibility was assessed using 1,000 bootstrap resamples. In each iteration, participants were sampled with replacement, biomarker residual calculation was adjusted for age, gender, and temporal meta-ROI tau SUVR, and hierarchical clustering using Ward’s linkage was repeated. For each pair of proteins, we computed the probability of each protein pair being in the same cluster for each resample, resulting in a 120 × 120 matrix for protein pair cluster stability. The stability scores varied between 0 and 1, with scores approaching 0 for pairs rarely observed in the same cluster or approaching 1 for pairs always observed in the same cluster. Silhouette analysis Cluster separation was evaluated using silhouette widths, computed for each protein using Euclidean distances in the residualized biomarker space. Silhouette values range from -1 to 1, where larger indices are desirable and lead to assignment to one cluster rather than to another for those with negative indices. The silhouette index for all proteins averaged together created a global assessment for how well-formed cluster assignments are. Comparison between dynamic tree cuts To assess whether the predefined clustering dendrogram height threshold (height = 20) had introduced non-biologically relevant boundaries, the primary clustering solution was contrasted with a data-driven clustering solution derived using dynamic tree-cutting approach. This approach to Clustering relies on adapting to dendrogram topology and local patterns of density to identify candidate clusters (ranging from 8 to 24). This enabled identification of cluster solutions that best aligned with the dendrogram’s intrinsic structure while assessing whether the height-based cut introduced non-biological discontinuities. Differences between solutions were examined to ensure that the final 12-cluster configuration reflected biologically meaningful and robust subdivisions. Gene Ontology GO was performed as a functional enrichment analysis using the R package ClusterProfiler, with default parameters 1. The results were adjusted for multiple comparisons (FDR) using the Benjamini-Hochberg method. Using the “GOSlim” package 2, we clustered the top GO terms by semantic similarity and represented the similarity matrices in an alluvial plot using the R package “ggalluvial”. Each cluster was named by summarizing how strongly each theme would appear in each cluster in terms of count and p-value. Subsequently, the names were adapted for the context of AD. Statistical testing and multiple-comparison correction All statistical analyses were performed in Python (pandas, scipy, statsmodels). Group comparisons of cluster composite scores between Aβ-positive and Aβ-negative participants were assessed using Mann-Whitney U tests, with FDR correction (Benjamini-Hochberg) applied across all 12 clusters. Associations among clusters and global neocortical Aβ SUVR were evaluated using Spearman rank correlations, with FDR correction applied to the correlation matrices. Neuroimaging voxel-based analysis Voxel-based linear regression was performed to examine the relationship between each clusters’ composite score and Aβ-PET imaging using MATLAB VoxelStats 59 , with adjustments for age, and sex. Multiple comparisons were controlled using the random field theory method 60 , with a significance threshold of P < 0.001 and were visualized as a T-map on the anatomical MRI template. Additionally, the regions showing an association between each cluster’s composite score and Aβ-PET (T-map) were compared with the corresponding regional mRNA expression levels from the healthy brain. Mean values were extracted from both maps to quantify the relationship between Aβ-cluster associations and baseline gene expression patterns. All tests were two-tailed, and statistical significance was defined as q < 0.05 after multiple-comparison correction unless otherwise specified. Forest Plot Analysis For each cluster’s composite scores, a logistic regression model was fitted using Aβ status as the outcome variable and the corresponding cluster composite score as the predictor, to estimate the risk of Aβ positivity. The odds ratio (OR) and 95% confidence interval were extracted from each model to quantify the association, and FDR correction was applied across the 12 models. Declarations Data availability All requests for raw and analyzed data and other materials will be promptly reviewed by McGill university to verify if the request is subject to any intellectual property or confidentiality obligations. Anonymized data will be shared upon request to the study’s senior author from a qualified academic investigator for the sole purpose of replicating the procedures and results presented in this article. Any data and materials that can be shared will be released via a material transfer agreement. Data are not publicly available due to information that could compromise the privacy of research participants. Related documents, including study protocol and informed consent forms, can similarly be made available upon request. Acknowledgment We would like to express our gratitude to our participant volunteers and their families for their participation in this study. We thank the staff, research nurses, psychometrist and neurologist at the McGill Centre for Studies in Aging for their contribution. We thank the radiographers and technicians at the McConnell Brain Imaging Centre and The Neuro (Montreal Neurological Institute-Hospital) for their role in imaging data acquisition. Funding This research is supported by an anonymous donor, the European Union’s Horizon Europe research and innovation programme under grant agreement no. 101053962, the Weston Brain Institute, Canadian Institutes of Health Research (CIHR) (MOP-11-51-31; RFN 152985, 159815, 162303), Canadian Consortium of Neurodegeneration and Aging (CCNA; MOP-11-51-31-team 1), the Alzheimer’s Association (NIRG-12-92090 and NIRP-12-259245), Brain Canada Foundation (CFI Project 34874, 33397), the Fonds de Recherche du Québec-Santé (FRQS; Chercheur Boursier, 2020-VICO-279314) and the Colin J. Adair Charitable Foundation. Y.-T.W. received the FRQS doctoral award. H.Z. is a Wallenberg Scholar and a Distinguished Professor at the Swedish Research Council. Unrelated to the work presented in this paper, H.Z. reports additional grant support from the Swedish Research Council (#2023-00356, #2022-01018 and #2019-02397), Swedish State Support for Clinical Research (#ALFGBG-71320), the Alzheimer Drug Discovery Foundation (ADDF), USA (#201809-2016862), the AD Strategic Fund and the Alzheimer’s Association (#ADSF-21-831376-C, #ADSF-21-831381-C, #ADSF-21-831377-C and #ADSF-24-1284328-C), the Bluefield Project, Cure Alzheimer’s Fund, the Olav Thon Foundation, the Erling-Persson Family Foundation, Familjen Rönströms Stiftelse, Stiftelsen för Gamla Tjänarinnor, Hjärnfonden, Sweden (#FO2022-0270), the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement no. 860197 (MIRIADE), the European Union Joint Programme-Neurodegenerative Disease Research (JPND2021-00694), the National Institute for Health and Care Research at University College London (UCL) Hospitals Biomedical Research Centre and the UK Dementia Research Institute at UCL (UKDRI-1003). K.B. is supported by the Swedish Research Council (#2017-00915 and #2022-00732), the Swedish Alzheimer Foundation (#AF-930351, #AF-939721 and #AF-968270), Hjärnfonden, Sweden (#FO2017-0243 and #ALZ2022-0006), the Swedish state under the agreement between the Swedish government and the County Councils, the ALF-agreement (an agreement between central government and seven regions on physician education and clinical research) (#ALFGBG-715986 and #ALFGBG-965240), the European Union Joint Program for Neurodegenerative Disorders (JPND2019-466-236), the Alzheimer’s Association 2021 Zenith Award (ZEN-21-848495) and the Alzheimer’s Association 2022-2025 Grant (SG-23-1038904 QC). M.S.W. is supported by the Else-Kröner-Fresenius Foundation (2023_EKMS.03), the German Research Foundation (WO 2835/1-1), and the Corona Foundation (S0199/10110/2025). Competing Interest Outside the work presented in this paper, P.R.-N. provides consultancy services for Roche, Cerveau Radiopharmaceuticals, Lilly, Eisai, Pfizer and Novo Nordisk. He also serves as a clinical trial investigator for Biogen, Novo Nordisk and Biogen. M.S.W. receives honoraria from Lilly for educational lectures outside the scope of this manuscript. S.G. is a member of the scientific advisory boards of Alzheon, AmyriAD, Eisai Canada, Enigma USA, Lilly Canada, Medesis, Okutsa Canada, Roche Canada and TauRx. He is a member of the editorial board of JPAD and of the Neurotorium. He has given lectures under the auspices of Biogen Canada and Lundbeck Korea. H.Z. has served at scientific advisory boards and/or as a consultant for Abbvie, Acumen, Alector, Alzinova, ALZpath, Amylyx, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognito Therapeutics, CogRx, Denali, Eisai, Enigma, LabCorp, Merck Sharp & Dohme, Merry Life, Nervgen, Novo Nordisk, Optoceutics, Passage Bio, Pinteon Therapeutics, Prothena, Quanterix, Red Abbey Labs, reMYND, Roche, Samumed, ScandiBio Therapeutics AB, Siemens Healthineers, Triplet Therapeutics, and Wave, has given lectures sponsored by Alzecure, BioArctic, Biogen, Cellectricon, Fujirebio, LabCorp, Lilly, Novo Nordisk, Oy Medix Biochemica AB, Roche, and WebMD, is a co-founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program, and is a shareholder of CERimmune Therapeutics (outside submitted work). K.B. has served as a consultant and on advisory boards for Acumen, ALZPath, BioArctic, Biogen, Eisai, Julius Clinical, Lilly, Novartis, Ono Pharma, Prothena, Roche Diagnostics and Siemens Healthineers; has served at data monitoring committees for Julius Clinical and Novartis; has given lectures, produced educational materials and participated in educational programmes for Biogen, Eisai and Roche Diagnostics; and is a co-founder of BBS in Gothenburg AB, which is a part of the GU Ventures Incubator Program. The remaining authors have no conflicts of interest to report related to this work. References Blennow, K., Hampel, H., Weiner, M. & Zetterberg, H. Cerebrospinal fluid and plasma biomarkers in Alzheimer disease. Nature Reviews Neurology 6 , 131-144 (2010). Rayaprolu, S. , et al. Systems-based proteomics to resolve the biology of Alzheimer’s disease beyond amyloid and tau. Neuropsychopharmacology 46 , 98-115 (2021). Feng, W. , et al. NULISA: a proteomic liquid biopsy platform with attomolar sensitivity and high multiplexing. Nature communications 14 , 7238 (2023). Puchades, M. , et al. Proteomic studies of potential cerebrospinal fluid protein markers for Alzheimer's disease. Molecular brain research 118 , 140-146 (2003). Tijms, B.M. , et al. Pathophysiological subtypes of Alzheimer’s disease based on cerebrospinal fluid proteomics. Brain 143 , 3776-3792 (2020). Mathys, H. , et al. Single-cell multiregion dissection of Alzheimer’s disease. Nature 632 , 858-868 (2024). Green, G.S. , et al. Cellular communities reveal trajectories of brain ageing and Alzheimer’s disease. Nature 633 , 634-645 (2024). Kitani, A., Matsui, Y. & Initiative, A.s.D.N. Integrative network analysis reveals novel moderators of Aβ-Tau interaction in Alzheimer's disease. Alzheimer's Research & Therapy 17 , 70 (2025). Li, K. , et al. Synaptic dysfunction in Alzheimer’s disease: Aβ, tau, and epigenetic alterations. Molecular neurobiology 55 , 3021-3032 (2018). Chen, X. , et al. Microglia-mediated T cell infiltration drives neurodegeneration in tauopathy. Nature 615 , 668-677 (2023). Zhang, H. , et al. Interaction between Aβ and tau in the pathogenesis of Alzheimer's disease. International journal of biological sciences 17 , 2181 (2021). Tolnay & Probst. Tau protein pathology in Alzheimer’s disease and related disorders. Neuropathology and applied neurobiology 25 , 171-187 (1999). Amor, S., Puentes, F., Baker, D. & Van Der Valk, P. Inflammation in neurodegenerative diseases. Immunology 129 , 154-169 (2010). Santiago, J.A., Bottero, V. & Potashkin, J.A. Transcriptomic and network analysis highlight the association of diabetes at different stages of Alzheimer’s disease. Frontiers in Neuroscience 13 , 1273 (2019). Minhas, P.S. , et al. Restoring hippocampal glucose metabolism rescues cognition across Alzheimer’s disease pathologies. Science 385 , eabm6131 (2024). Minter, M.R., Taylor, J.M. & Crack, P.J. The contribution of neuroinflammation to amyloid toxicity in Alzheimer's disease. Journal of neurochemistry 136 , 457-474 (2016). Sankowski, R., Mader, S. & Valdés-Ferrer, S.I. Systemic inflammation and the brain: novel roles of genetic, molecular, and environmental cues as drivers of neurodegeneration. Frontiers in cellular neuroscience 9 , 28 (2015). Obici, L., Perfetti, V., Palladini, G., Moratti, R. & Merlini, G. Clinical aspects of systemic amyloid diseases. Biochimica et Biophysica Acta (BBA)-Proteins and Proteomics 1753 , 11-22 (2005). Cai, H. , et al. Metabolic dysfunction in Alzheimer's disease and related neurodegenerative disorders. Current Alzheimer Research 9 , 5-17 (2012). Minoshima, S., Cross, D., Thientunyakit, T., Foster, N.L. & Drzezga, A. 18F-FDG PET imaging in neurodegenerative dementing disorders: insights into subtype classification, emerging disease categories, and mixed dementia with copathologies. Journal of nuclear medicine 63 , 2S-12S (2022). Che, P. , et al. Cerebral perfusion is correlated with cerebral metabolism and amyloid deposition in Alzheimer’s disease. Translational Psychiatry 15 , 189 (2025). Daulatzai, M.A. Cerebral hypoperfusion and glucose hypometabolism: Key pathophysiological modulators promote neurodegeneration, cognitive impairment, and Alzheimer's disease. Journal of neuroscience research 95 , 943-972 (2017). Kumar, V., Kim, S.-H. & Bishayee, K. Dysfunctional glucose metabolism in Alzheimer’s disease onset and potential pharmacological interventions. International Journal of Molecular Sciences 23 , 9540 (2022). Mosconi, L., Pupi, A. & De Leon, M.J. Brain glucose hypometabolism and oxidative stress in preclinical Alzheimer's disease. Annals of the new York Academy of Sciences 1147 , 180-195 (2008). Bonvento, G. & Bolanos, J.P. Astrocyte-neuron metabolic cooperation shapes brain activity. Cell metabolism 33 , 1546-1564 (2021). Takahashi, S. Neuroprotective function of high glycolytic activity in astrocytes: common roles in stroke and neurodegenerative diseases. international journal of molecular sciences 22 , 6568 (2021). Colombo, E. & Farina, C. Astrocytes: key regulators of neuroinflammation. Trends in immunology 37 , 608-620 (2016). Thomas, J.H. Fluid dynamics of cerebrospinal fluid flow in perivascular spaces. Journal of the Royal Society Interface 16 , 20190572 (2019). Harrison, I.F. , et al. Impaired glymphatic function and clearance of tau in an Alzheimer’s disease model. Brain 143 , 2576-2593 (2020). Keil, S.A., Jansson, D., Braun, M. & Iliff, J.J. Glymphatic dysfunction in Alzheimer’s disease: A critical appraisal. Science 389 , eadv8269 (2025). Fagan, A.M. , et al. Decreased cerebrospinal fluid Aβ42 correlates with brain atrophy in cognitively normal elderly. Annals of Neurology: Official Journal of the American Neurological Association and the Child Neurology Society 65 , 176-183 (2009). Janelidze, S. , et al. CSF Aβ42/Aβ40 and Aβ42/Aβ38 ratios: better diagnostic markers of Alzheimer disease. Annals of clinical and translational neurology 3 , 154-165 (2016). Bateman, R.J. , et al. Human amyloid-β synthesis and clearance rates as measured in cerebrospinal fluid in vivo. Nature medicine 12 , 856-861 (2006). Sanford, S.A. , et al. The type‐I interferon response potentiates seeded tau aggregation and exacerbates tau pathology. Alzheimer's & Dementia 20 , 1013-1025 (2024). Udeochu, J.C. , et al. Tau activation of microglial cGAS–IFN reduces MEF2C-mediated cognitive resilience. Nature neuroscience 26 , 737-750 (2023). Jorfi, M. , et al. Infiltrating CD8+ T cells exacerbate Alzheimer’s disease pathology in a 3D human neuroimmune axis model. Nature neuroscience 26 , 1489-1504 (2023). Zhang, X. , et al. Astrocytic functions and lipid metabolism: Correlations and therapeutic targets in Alzheimer's disease and glioblastoma. Clinical and Translational Discovery 4 , e287 (2024). Preman, P. , et al. APOE from astrocytes restores Alzheimer’s Aβ-pathology and DAM-like responses in APOE deficient microglia. EMBO molecular medicine 16 , 3113-3141 (2024). Haney, M.S. , et al. APOE4/4 is linked to damaging lipid droplets in Alzheimer’s disease microglia. Nature 628 , 154-161 (2024). Serrano-Pozo, A., Das, S. & Hyman, B.T. APOE and Alzheimer's disease: advances in genetics, pathophysiology, and therapeutic approaches. The Lancet Neurology 20 , 68-80 (2021). Piwecka, M., Rajewsky, N. & Rybak-Wolf, A. Single-cell and spatial transcriptomics: deciphering brain complexity in health and disease. Nature Reviews Neurology 19 , 346-362 (2023). Liang, W.S. , et al. Gene expression profiles in anatomically and functionally distinct regions of the normal aged human brain. Physiological genomics 28 , 311-322 (2007). Yang, D. , et al. Lipid metabolism and storage in neuroglia: role in brain development and neurodegenerative diseases. Cell & bioscience 12 , 106 (2022). Rivest, S. Regulation of innate immune responses in the brain. Nature Reviews Immunology 9 , 429-439 (2009). Castillo, E. , et al. Comparative profiling of cortical gene expression in Alzheimer’s disease patients and mouse models demonstrates a link between amyloidosis and neuroinflammation. Scientific reports 7 , 17762 (2017). Grothe, M.J. , et al. Molecular properties underlying regional vulnerability to Alzheimer’s disease pathology. Brain 141 , 2755-2771 (2018). Petersen, R.C. Mild cognitive impairment as a diagnostic entity. Journal of internal medicine 256 , 183-194 (2004). McKhann, G.M. , et al. The diagnosis of dementia due to Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimer's & dementia 7 , 263-269 (2011). Pascoal, T.A. , et al. In vivo quantification of neurofibrillary tangles with [18 F] MK-6240. Alzheimer's research & therapy 10 , 1-14 (2018). Pascoal, T.A. , et al. 18F-MK-6240 PET for early and late detection of neurofibrillary tangles. Brain 143 , 2818-2830 (2020). Jack Jr, C.R. , et al. Defining imaging biomarker cut points for brain aging and Alzheimer's disease. Alzheimer's & Dementia 13 , 205-216 (2017). Therriault, J. , et al. Determining amyloid-β positivity using 18F-AZD4694 PET imaging. Journal of Nuclear Medicine 62 , 247-252 (2021). Chen, C.D. , et al. Comparing tau PET visual interpretation with tau PET quantification, cerebrospinal fluid biomarkers, and longitudinal clinical assessment. Journal of Alzheimer's Disease 93 , 765-777 (2023). Collij, L.E. , et al. Multitracer model for staging cortical amyloid deposition using PET imaging. Neurology 95 , e1538-e1553 (2020). Wang, Y.-T. , et al. Identify biological Alzheimer’s disease using a novel nucleic acid–linked protein immunoassay. Brain Communications 7 , fcaf004 (2025). Komorowski, A. , et al. Association of protein distribution and gene expression revealed by PET and post-mortem quantification in the serotonergic system of the human brain. Cerebral Cortex 27 , 117-130 (2017). Gryglewski, G. , et al. Spatial analysis and high resolution mapping of the human whole-brain transcriptome for integrative analysis in neuroimaging. Neuroimage 176 , 259-267 (2018). Winder, Z. , et al. Hierarchical clustering analyses of plasma proteins in subjects with cardiovascular risk factors identify informative subsets based on differential levels of angiogenic and inflammatory biomarkers. Frontiers in Neuroscience 14 , 84 (2020). Mathotaarachchi, S. , et al. VoxelStats: a MATLAB package for multi-modal voxel-wise brain image analysis. Frontiers in neuroinformatics 10 , 20 (2016). Taylor, J.E. & Worsley, K.J. Detecting sparse signals in random fields, with an application to brain mapping. Journal of the American Statistical Association 102 , 913-928 (2007). Tables Tables are available in the Supplementary Files section. Additional Declarations Yes there is potential Competing Interest. These interests are listed in a paragraph at the end of the manuscript. Supplementary Files Extendeddata.docx Extended_data Table1.docx Table 1 Table2.docx Table 2 Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8428920","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":599368309,"identity":"72afa87f-0aef-4500-92e4-cefc4dfada83","order_by":0,"name":"Pedro Rosa-Neto","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3UlEQVRIiWNgGAWjYHACAyjNfABISMjwkaCFLQGkhYeNBC08YAZhLebszds+fGC4Jyfffubrhh+/LIBamB9+wKfFsudY8cwZDMXGBmdyt93s7QM5jM1YAq+rbuQYM/MwJCRuYMjddpuxB6SFhwG/lvtvIFrm9795BtPC/AO/LTwQLQ03cthuM/wAa2HDa4tlT1ox4wyDBGODG8/MbvY2ALUws5lZ4NNizn54M8OHigQ5+f7kZzd+/KmT42dvfnwDr8OQSAYGxjYgwYxPPZJiKPhDQPkoGAWjYBSMSAAAMVo/+O202nwAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0001-9116-1376","institution":"McGill Univeristy","correspondingAuthor":true,"prefix":"","firstName":"Pedro","middleName":"","lastName":"Rosa-Neto","suffix":""},{"id":599368310,"identity":"e3c567ac-18fc-4304-9622-9871872364a4","order_by":1,"name":"Seyyed Ali Hosseini","email":"","orcid":"https://orcid.org/0000-0001-7542-7541","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Seyyed","middleName":"Ali","lastName":"Hosseini","suffix":""},{"id":599368311,"identity":"cfd1ad17-5558-470d-b8f5-177cc65caa66","order_by":2,"name":"Yi-Ting Wang","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Yi-Ting","middleName":"","lastName":"Wang","suffix":""},{"id":599368312,"identity":"0feb457b-89e2-4d3e-a1d1-353fbe0d1482","order_by":3,"name":"Nesrine Rahmouni","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Nesrine","middleName":"","lastName":"Rahmouni","suffix":""},{"id":599368313,"identity":"f75f98be-79ac-4666-9385-fca7a6244486","order_by":4,"name":"Etienne Aumont","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Etienne","middleName":"","lastName":"Aumont","suffix":""},{"id":599368314,"identity":"318f8866-436a-4456-91c8-1339adb0c528","order_by":5,"name":"Arthur Macedo","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Arthur","middleName":"","lastName":"Macedo","suffix":""},{"id":599368315,"identity":"b74051f3-938b-416c-93ae-b362b597308a","order_by":6,"name":"Ilaria Pola","email":"","orcid":"","institution":"University of Gothenburg","correspondingAuthor":false,"prefix":"","firstName":"Ilaria","middleName":"","lastName":"Pola","suffix":""},{"id":599368316,"identity":"ce5e197c-9584-42d1-a4c1-3f3d9813bc6d","order_by":7,"name":"Tevy Chan","email":"","orcid":"https://orcid.org/0009-0001-9134-5949","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Tevy","middleName":"","lastName":"Chan","suffix":""},{"id":599368317,"identity":"47284d2a-e262-4b7e-a151-0420f02fe4ee","order_by":8,"name":"Gleb Bezgin","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Gleb","middleName":"","lastName":"Bezgin","suffix":""},{"id":599368318,"identity":"84ddf196-adb1-4057-9625-2f17c8b555f2","order_by":9,"name":"Brandon Hall","email":"","orcid":"https://orcid.org/0009-0007-4578-2033","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Brandon","middleName":"","lastName":"Hall","suffix":""},{"id":599368319,"identity":"cefe3ba5-e665-4a7e-bb5e-6eccfdda3e10","order_by":10,"name":"Aurélie Lebrun","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Aurélie","middleName":"","lastName":"Lebrun","suffix":""},{"id":599368320,"identity":"5b704ed2-8feb-4116-95fc-a487ca5e238b","order_by":11,"name":"Jaime Fernandez-Arias","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Jaime","middleName":"","lastName":"Fernandez-Arias","suffix":""},{"id":599368321,"identity":"62968db7-0645-4883-a8fe-c86d9e6cd3e8","order_by":12,"name":"Kely Quispialaya-Socualaya","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Kely","middleName":"","lastName":"Quispialaya-Socualaya","suffix":""},{"id":599368322,"identity":"98ab1375-e2ba-4a84-915a-cbba331c3c31","order_by":13,"name":"Yansheng Zheng","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Yansheng","middleName":"","lastName":"Zheng","suffix":""},{"id":599368323,"identity":"12163856-e625-4ad7-8aaa-4565d05f9c17","order_by":14,"name":"Marina Goncalves","email":"","orcid":"https://orcid.org/0000-0002-5769-5427","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Marina","middleName":"","lastName":"Goncalves","suffix":""},{"id":599368324,"identity":"f4ce9425-557f-42dc-8a98-1a5f3910cb1a","order_by":15,"name":"Juliana Godoy","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Juliana","middleName":"","lastName":"Godoy","suffix":""},{"id":599368325,"identity":"4cdf91e7-460f-459c-9709-9fa18c5cdbad","order_by":16,"name":"Stuart Mitchell","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Stuart","middleName":"","lastName":"Mitchell","suffix":""},{"id":599368326,"identity":"acce76ac-f76f-40ef-9401-b1e037bb7286","order_by":17,"name":"Delphine Oliva-Lopez","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Delphine","middleName":"","lastName":"Oliva-Lopez","suffix":""},{"id":599368327,"identity":"9aaeb786-1598-494c-9126-3d525fced459","order_by":18,"name":"Rodrigo Paes","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Rodrigo","middleName":"","lastName":"Paes","suffix":""},{"id":599368328,"identity":"b9d728fb-2f42-4f78-8940-098e3c8e7b24","order_by":19,"name":"Gabriela Baldasso","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Gabriela","middleName":"","lastName":"Baldasso","suffix":""},{"id":599368329,"identity":"37d3ad63-8231-4824-8a6a-4dd241f01a7b","order_by":20,"name":"Firoza Lussier","email":"","orcid":"","institution":"University of Pittsburgh","correspondingAuthor":false,"prefix":"","firstName":"Firoza","middleName":"","lastName":"Lussier","suffix":""},{"id":599368330,"identity":"d7a8b7b8-ba3e-47fb-a30c-2d9439a0bbd1","order_by":21,"name":"Robert Hopewell","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Robert","middleName":"","lastName":"Hopewell","suffix":""},{"id":599368331,"identity":"9a85ca46-6e1b-4a9d-bace-f80035375e31","order_by":22,"name":"Sanjeev Chawla","email":"","orcid":"","institution":"University of Pennsylvania","correspondingAuthor":false,"prefix":"","firstName":"Sanjeev","middleName":"","lastName":"Chawla","suffix":""},{"id":599368332,"identity":"56921f1d-929c-4dd1-bca1-4a9bd676f17e","order_by":23,"name":"Vladimir Fonov","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Vladimir","middleName":"","lastName":"Fonov","suffix":""},{"id":599368333,"identity":"bb46ba94-3da9-46c3-b324-c8e1e6af3388","order_by":24,"name":"Gassan Massarweh","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Gassan","middleName":"","lastName":"Massarweh","suffix":""},{"id":599368334,"identity":"7f0c3708-2d15-4c48-a6ce-2200611ace7a","order_by":25,"name":"Jean-Paul Soucy","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Jean-Paul","middleName":"","lastName":"Soucy","suffix":""},{"id":599368335,"identity":"aa87f25c-69c8-4514-8d11-0f2e2420dd10","order_by":26,"name":"Thomas Karikari","email":"","orcid":"https://orcid.org/0000-0003-1422-4358","institution":"University of Pittsburgh","correspondingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"","lastName":"Karikari","suffix":""},{"id":599368336,"identity":"a6b0fe0d-eaae-463b-8799-c730286d746f","order_by":27,"name":"Andréa Benedet","email":"","orcid":"","institution":"University of Gothenburg","correspondingAuthor":false,"prefix":"","firstName":"Andréa","middleName":"","lastName":"Benedet","suffix":""},{"id":599368337,"identity":"8edbbb09-a8fe-4834-847d-5f3cec1924e9","order_by":28,"name":"Nicholas Ashton","email":"","orcid":"https://orcid.org/0000-0002-3579-8804","institution":"Banner Sun Health Research Institute","correspondingAuthor":false,"prefix":"","firstName":"Nicholas","middleName":"","lastName":"Ashton","suffix":""},{"id":599368338,"identity":"e3118718-7bdd-452e-82f2-4cbd654886f1","order_by":29,"name":"Henrik Zetterberg","email":"","orcid":"","institution":"University of Gothenburg","correspondingAuthor":false,"prefix":"","firstName":"Henrik","middleName":"","lastName":"Zetterberg","suffix":""},{"id":599368339,"identity":"97dfbf9c-9fd3-4778-9716-7b4c6c3d741e","order_by":30,"name":"Jesse Klostranec","email":"","orcid":"","institution":"McGill Unviersity","correspondingAuthor":false,"prefix":"","firstName":"Jesse","middleName":"","lastName":"Klostranec","suffix":""},{"id":599368340,"identity":"a3a59aad-4ef6-476a-bc22-71410231074c","order_by":31,"name":"Maxime Montembeault","email":"","orcid":"https://orcid.org/0000-0002-9369-0498","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Maxime","middleName":"","lastName":"Montembeault","suffix":""},{"id":599368341,"identity":"8459563b-054c-46d8-8789-c470c36909c2","order_by":32,"name":"Paolo Vitali","email":"","orcid":"https://orcid.org/0000-0001-8953-1542","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"Paolo","middleName":"","lastName":"Vitali","suffix":""},{"id":599368342,"identity":"fefa30d5-c442-4063-8b7b-d03cb940092c","order_by":33,"name":"Kaj Blennow","email":"","orcid":"https://orcid.org/0000-0002-1890-4193","institution":"University of Gothenburg","correspondingAuthor":false,"prefix":"","firstName":"Kaj","middleName":"","lastName":"Blennow","suffix":""},{"id":599368343,"identity":"ec570cfd-cc80-4a19-b765-597347dab14c","order_by":34,"name":"Tharick Pascoal","email":"","orcid":"https://orcid.org/0000-0001-9057-8014","institution":"University of Pittsburgh","correspondingAuthor":false,"prefix":"","firstName":"Tharick","middleName":"","lastName":"Pascoal","suffix":""},{"id":599368344,"identity":"678d8a11-7c65-4b76-b403-2be051901246","order_by":35,"name":"D. Louis Collins","email":"","orcid":"","institution":"McGill University","correspondingAuthor":false,"prefix":"","firstName":"D.","middleName":"Louis","lastName":"Collins","suffix":""},{"id":599368345,"identity":"a7d0c1c9-0cea-49eb-99cd-cf766a52e5c8","order_by":36,"name":"Marcel Woo","email":"","orcid":"https://orcid.org/0000-0002-1306-2708","institution":"University Medical Center Hamburg-Eppendorf","correspondingAuthor":false,"prefix":"","firstName":"Marcel","middleName":"","lastName":"Woo","suffix":""}],"badges":[],"createdAt":"2025-12-23 00:56:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8428920/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8428920/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104178642,"identity":"3b561170-ba7a-4b9e-b192-a6dda4747d07","added_by":"auto","created_at":"2026-03-08 16:58:21","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2833087,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSchematic representation of the study design.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study integrated Aβ-PET imaging, CSF biomarker quantification, and statistical and clustering analyses to derive data-driven CSF biomarker clusters and examine their association with global neocortical Aβ burden. Fig. 1 was created in BioRender. Goncalves, M. (2025) https://biorender.com/au0bbl5\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/0b6b1caae6f6764ecda190a1.png"},{"id":104178637,"identity":"9b10e329-ed5b-4a8b-9006-8aa438d84b1c","added_by":"auto","created_at":"2026-03-08 16:58:21","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1222312,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHierarchical clustering dendrogram of CNS CSF biomarkers and robustness of protein modules.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ea) The dendrogram illustrates the hierarchical clustering of CSF biomarkers. The dendrogram visually displays similarities among biomarkers, with highly clustered biomarkers (lower on the y-axis) showing greater similarity and more distantly connected biomarkers showing weaker associations. Dendrogram Structure: Single biomarkers are marked by vertical lines, and horizontal lines denote the merging of clusters. Similar features (based on the distance measure) are merged at lower positions in the dendrogram, and less similar features are merged at higher positions. Distance Metric: The y-axis indicates the dissimilarity or distance between clusters. Shorter horizontal lines indicate more similarity. Two features or clusters are linked to a height on the y-axis that indicates their dissimilarity, with greater differences indicated by taller links. clusters: The biomarkers are divided into groups by a horizontal cut-off line (at threshold distance of 20). In each cluster, biomarkers within the selected threshold are grouped similar to one another.\u003c/p\u003e\n\u003cp\u003eb) A 1,000-iteration bootstrap resampling analysis was performed to evaluate the reproducibility of the hierarchical clustering solution. Within each iteration, participants were resampled with replacement, biomarker residuals were recalculated, and clustering was repeated. The heatmap reveals how often pairs of proteins cluster together within each iteration of resampling, ranging from 0 (never clustered together) to 1 (always clustered together). The stable areas shown on the diagonal reveal well-defined (clustered) areas, including areas centered on the clusters 1, 4, and 7.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/26c9d6fc02757d08eff36b2e.png"},{"id":104178639,"identity":"12c6c240-f246-4932-af46-30d1d0ae128e","added_by":"auto","created_at":"2026-03-08 16:58:21","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1195962,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInternal cohesion and biological characterization of CSF protein modules.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ea) The silhouette scores for each protein are calculated to assess within-cluster homogeneity and between-cluster heterogeneity. Boxplots are provided to analyze how scores are distributed for each cluster. This provides insight into how proteins are distributed within their respective cluster scores, and how they are distinguished between various clusters. The dashed horizontal line represents the global mean silhouette score. The highest scores are identified in cluster 1, 4, and 7, indicating well-separated and internally consistent modules.\u003c/p\u003e\n\u003cp\u003eb) Alluvial plot visualizing the link of each cluster with Gene Ontology Biological Process (GO Slim). Line thickness reflects the number of pathways contributing to the biological terms.\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/3d6a6f8f949dfd29d84ab38b.png"},{"id":104178643,"identity":"4f17c99b-f4e0-4d5f-9f99-55c30ba5aaf5","added_by":"auto","created_at":"2026-03-08 16:58:21","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":3415469,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCSF clusters for tau, inflammation, APOE and Aβ have the highest correlation with Aβ load.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorrelation with global neocortical AβSUVR and differences based on Aβ status of each cluster’s composite score. Cluster composite scores were compared between Aβ-positive and Aβ-negative participants. Boxplots show the median and interquartile range for each cluster, with individual participant values overlaid. Statistical differences were assessed using Mann-Whitney U tests, and Benjamini-Hochberg FDR correction was applied across the 12 clusters. Clusters 1, 2, 11, and 12 showed higher values in Aβ-positive individuals, while cluster 6 was lower. No differences were observed for the remaining clusters after FDR correction.\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/eb96bd449c89f0c505b20dd1.png"},{"id":104178644,"identity":"5790651a-5fa5-48cb-80aa-da38b34726a8","added_by":"auto","created_at":"2026-03-08 16:58:21","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":5937564,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eRegional associations between Aβ and CSF protein clusters are predicted by mRNA expression.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ea) Voxel-wise analysis between Aβ-PET and each cluster’s composite score (CS) in older adults (\u0026gt;65 years; N = 165), covariates (COV) : age and sex.\u003c/p\u003e\n\u003cp\u003eb) Mean mRNA expression intensity (EI) for genes corresponding to each cluster.\u003c/p\u003e\n\u003cp\u003ec) Relationship between the mRNA expression intensity and association of each cluster with Aβ-PET.\u003c/p\u003e\n\u003cp\u003ed) Forest Plot showing the odds ratios (OR) and confidence intervals (CI) for 12 clusters' composite scores associated with Aβ positivity. Each horizontal line represents the CI for a factor, with the red dot indicating the OR. The vertical dashed line at OR = 1 represents the null effect. Factors with CI that do not cross this line are statistically significant. ​​Odds Ratio (OR): OR \u0026gt; 1: The factor increases the odds of being Aβ positivity (risk factor). OR \u0026lt; 1: The factor decreases the odds of being Aβ positivity (protective factor). OR = 1: No association. The odds ratio represents the likelihood of Aβ positivity for a one-unit increase in the composite score, while the confidence interval indicates the precision of this estimate.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/1f132dc31652a28972f65eea.png"},{"id":104178641,"identity":"5646222c-eb6d-4986-8dd9-7b2b89a2979c","added_by":"auto","created_at":"2026-03-08 16:58:21","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":2825764,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe CSF protein clusters show specific inter-correlations.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ea) Spearman rank correlations were computed among the 12 cluster composite scores to characterize shared biological relationships. Benjamini-Hochberg false discovery rate (FDR) correction was applied across the correlation matrix.\u003c/p\u003e\n\u003cp\u003eb) Correlations between each cluster composite score and global neocortical Aβ SUVR are shown across Aβ staging groups, illustrating how cluster-Aβ associations strengthen with increasing Aβ burden. Together, these findings highlight coordinated activation of tau, inflammatory, metabolic, and APOE-related pathways in individuals with greater Aβ pathology.\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/43b394f1e2ceac74448e1078.png"},{"id":106728013,"identity":"76ec57f7-01e5-43c3-8892-dc72e14fda23","added_by":"auto","created_at":"2026-04-12 18:41:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":18173062,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/baf1583e-67af-401c-9732-9022395ef43a.pdf"},{"id":104178636,"identity":"7ff9726a-0ac8-4777-b6d2-7b70ba4486b5","added_by":"auto","created_at":"2026-03-08 16:58:21","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":42328,"visible":true,"origin":"","legend":"Extended_data","description":"","filename":"Extendeddata.docx","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/fba3975e9b0a2fe1f802067a.docx"},{"id":104178640,"identity":"595ec982-29b9-49ee-a375-36a9d80f485f","added_by":"auto","created_at":"2026-03-08 16:58:21","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":18283,"visible":true,"origin":"","legend":"Table 1","description":"","filename":"Table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/1f75baeb55d2d1c70dc72cbf.docx"},{"id":104404033,"identity":"bf2df6e4-be46-4de8-9970-581572f8665e","added_by":"auto","created_at":"2026-03-11 12:19:37","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":39324,"visible":true,"origin":"","legend":"Table 2","description":"","filename":"Table2.docx","url":"https://assets-eu.researchsquare.com/files/rs-8428920/v1/1782e68e88d76ddba2ca6a9e.docx"}],"financialInterests":"\u003cb\u003eYes\u003c/b\u003e there is potential Competing Interest.\nThese interests are listed in a paragraph at the end of the manuscript.","formattedTitle":"\u003cp\u003eAβ-related Cerebrospinal Fluid Proteins in Alzheimer’s Disease Reflect Gene Expression Levels Found in The Healthy Cortex\u003c/p\u003e","fulltext":[{"header":"Introduction","content":"\u003cp\u003eCerebrospinal fluid (CSF) is in direct communication with the brain and therefore reflects its biological state, offering valuable \u003cem\u003ein vivo\u003c/em\u003e insight into the progression of dementia \u003csup\u003e1\u003c/sup\u003e. \u0026nbsp;Advances in proteomic and imaging biomarkers hold promise for personalized approaches by expanding our understanding of Alzheimer\u0026rsquo;s disease (AD) biology \u003csup\u003e2\u003c/sup\u003e. The advent of ultrasensitive antibody-based proteomic platforms, such as nucleic acid-linked immune-sandwich assay sequencing (NULISA\u0026trade;seq), enables a more comprehensive characterization of AD pathophysiology across the disease continuum \u003csup\u003e3\u003c/sup\u003e. In fact, CSF protein alterations occurring in AD converge on brain regions with a high burden of amyloid-beta (A\u0026beta;) and tau aggregates \u003csup\u003e4\u003c/sup\u003e. However, one should consider whether these changes merely represent downstream responses to the spread of A\u0026beta; or might also reflect features of specific cortical regions \u003csup\u003e5\u003c/sup\u003e. \u0026nbsp;To address this question, we deployed regional cortical transcriptomics profiles from publicly available datasets to enable assessments of differential susceptibility or resilience to disease processes at the population level \u003csup\u003e6,7\u003c/sup\u003e. \u0026nbsp;Here, we aimed to integrate neuroimaging, clinical data, and fluid biomarker proteomes by hierarchical clustering to identify and validate multi-modal biomarkers that reflect different stages of A\u0026beta; pathology. \u0026nbsp;Additionally, we assessed the relationships between the population-based, spatially resolved gene expression profiles and CSF protein signature clusters relevant to A\u0026beta; \u003csup\u003e8\u003c/sup\u003e. \u0026nbsp;We tested the hypothesis that cortical transcripts from brain regions vulnerable to A\u0026beta; aggregation are associated with A\u0026beta;-related protein signatures in AD.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThe study framework depicted in Fig. 1 provides an overview of the data analytical procedures. It starts with patient preparation for A\u0026beta;- and tau-PET and MRI scanning, as well as lumbar puncture for CSF collection, which was processed using the NULISA\u0026trade;seq CNS Panel to measure 120 neurodegenerative disorder-associated proteins. The next step involved registering PET images to their native MRI space and measuring global neocortical A\u0026beta;. This was followed by hierarchical clustering, subsequent validation processes, and gene ontology (GO) analysis to capture nuanced patterns in the CSF data related to A\u0026beta; aggregation in the brain. Finally, composite scores for biomarkers within the same cluster (identified by their distance to A\u0026beta;-PET) were calculated, validated, and statistical analyses were conducted.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo identify biomarker profiles reflecting A\u0026beta; pathology, we performed hierarchical clustering of the proteins. The resulting clustering dendrogram is shown in Fig. 2a. Biomarkers with pairwise distances below 20 were considered part of the same cluster. We found 12 clusters from 120 different biomarkers, as shown in Table 2. To assess how well proteins tend to co-cluster consistently against perturbations introduced via sampling, we created a 120 \u0026times; 120 matrix for bootstrapping co-membership stability with 1,000 resamples (Fig. 2b). The resulting structure revealed that the majority of cluster boundaries were highly reproducible, with several biologically coherent modules, including those comprising cluster 1, cluster 4, and cluster 7, tending towards perfect stability (co-membership \u0026gt; 0.95), while others, with more variable signaling pathways, tend towards lower but still visible stability.\u003c/p\u003e\n\u003cp\u003eTo quantify the degree of internal separation between the protein clusters in the CSF, we computed silhouette widths for each protein in the 12-cluster solution derived from Ward\u0026rsquo;s hierarchical clustering (Fig. 3a). The global silhouette scores revealed that they are rather moderate on average (mean silhouette score = 0.255), as expected for proteins undergoing common expression in CSF. Despite this relatively moderate degree of global separation, several clusters demonstrated clearly positive silhouette distributions, indicating well-structured cluster cores, whereas others displayed mixed or negative silhouettes, consistent with diffuse or cross-pathway biological roles.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo determine whether the predefined tree-cut height (distance = 20) imposed artificial cluster boundaries, we compared the fixed 12-cluster solution with a data-driven set of cluster partitions generated by scanning cluster numbers between k = 8 and k = 24 (Extended Data Table 1). Silhouette scores increased gradually with larger k, reaching a maximum at k = 20 (silhouette = 0.264), while similarity to the fixed 12-cluster solution (as measured by the adjusted Rand index (ARI)) decreased from ARI = 1.00 at k = 12 to ARI = 0.78 at k = 20. This pattern indicates that the 12-cluster solution lies at a stable structural plateau of the dendrogram, while higher-k decompositions introduce finer-grained subdivisions without altering the core biological modules.\u003c/p\u003e\n\u003cp\u003eNext, using GO analysis, we attributed these 12 clusters to tau markers (cluster 1), neuronal injury (cluster 2), axonal metabolic injury (cluster 3), synaptic signaling (cluster 4), cellular response (cluster 5), A\u0026beta;42 (cluster 6), cell differentiation (cluster 7), immune (I) (cluster 8), carbohydrate derivative metabolic processes (cluster 9), choroid plexus (cluster 10), immune (II) (cluster 11) and apolipoprotein-E4 (APOE4) (cluster 12) (Fig. 3b \u0026amp; Extended Data Table 2).\u003c/p\u003e\n\u003cp\u003eTo analyze the association of the clusters with brain A\u0026beta; accumulation, we computed for each cluster a composite score of the proteins included in the cluster. We then correlated them with global neocortical A\u0026beta; SUVR and compared them based on A\u0026beta; status (Fig. 4). We found that, in A\u0026beta;-positive participants, the composite scores for the tau markers, neuronal injury, immune (II), and APOE4 clusters were higher, while the composite score for the A\u0026beta;42 cluster was lower. The other clusters did not show differences based on A\u0026beta; status. Additionally, tau markers, neuronal injury, axonal metabolic injury, immune (II), and APOE4 clusters were positively correlated, whereas the A\u0026beta;42 cluster was negatively correlated with neocortical A\u0026beta; SUVR.\u003c/p\u003e\n\u003cp\u003eNext, we conducted a linear model voxel-wise analysis to investigate the impact of each cluster on A\u0026beta; load in older adults (\u0026gt;65 years; N = 165), corrected for age and sex (Fig. 5a). The tau markers and neuronal injury clusters showed strong to moderate positive associations with A\u0026beta; load over the cortex. The axonal metabolic injury cluster showed a positive association with A\u0026beta; load in the white matter, and the synaptic signaling and cellular response clusters showed positive associations within the ventricles, and not the brain parenchyma. The A\u0026beta;42 cluster was associated with A\u0026beta; load in the cortex. The cell differentiation and immune (I) clusters did not show associations in the voxel-wise analysis. The carbohydrate derivative metabolic processes cluster showed a positive association with A\u0026beta; load in the thalamus and brainstem regions. The choroid plexus marker cluster showed positive correlations within the ventricles and cisterns compatible with the choroid plexus. The immune (II) cluster showed associations in the hippocampal/parahippocampal regions and the lateral temporal cortices. The APOE4 cluster showed a strong positive association with A\u0026beta; load over the cortex.\u003c/p\u003e\n\u003cp\u003eNext, we asked whether the CSF protein clusters are associated with the brain-region specific transcriptional profiles. We computed the voxel-wise gene expression intensity of each cluster by averaging the mRNA expression intensity of proteins within the same cluster derived from Allen Human Brain Atlas of healthy brain tissue (Fig. 5b). The analysis indicated that the tau markers, APOE4, A\u0026beta;42, axonal metabolic injury, and neuronal injury clusters displayed the strongest mean expression in the brain, ranked from highest to lowest. To investigate the relationship between gene expression intensity and the regional associations with A\u0026beta; load, for each cluster, we computed a mask of the brain regions that showed associations between the cluster\u0026rsquo;s composite score and A\u0026beta; load, averaged the \u0026beta;-values in the mask, and applied it to the corresponding mRNA expression intensity map. Thereby, we were able to analyze whether the brain-region specific mRNA profiles are associated with the A\u0026beta;-associated CSF proteomes. Notably, we detected a correlation between the regional mRNA-expression intensity and the A\u0026beta;-related cortical associations with CSF protein clusters \u0026nbsp;(Fig. 5c).\u003c/p\u003e\n\u003cp\u003eWe next used logistic regressions to verify the risk of A\u0026beta; positivity based on the cluster composite scores. Our results revealed that the tau markers (OR = 10.42, CI [5.21, 20.83]), neuronal injury (OR = 2.40, CI [1.54, 3.75]), and APOE4 (OR = 3.97, CI [2.58, 6.11]) clusters were risk factors for A\u0026beta; positivity. Furthermore, the A\u0026beta;42 cluster (OR = 0.13, CI [0.08, 0.24]) was inversely associated with A\u0026beta; positivity while the other clusters were not \u0026nbsp;(Fig. 5d \u0026amp; Extended Data Table 3).\u003c/p\u003e\n\u003cp\u003eFinally, we aimed to analyze the correlations between the different clusters. Our results revealed distinct correlation patterns among the biomarker clusters (Fig. 6a \u0026amp; Extended Data Table 4). The tau markers cluster was correlated with all other clusters except the cell differentiation, axonal metabolic injury, and immune (II) \u0026nbsp;clusters. The neuronal injury cluster showed correlations with all clusters except the cellular response, A\u0026beta;42, axonal metabolic injury, and immune (II) \u0026nbsp;clusters. The axonal metabolic injury cluster was correlated with all other clusters except the A\u0026beta;42, choroid plexus marker, and immune (II) \u0026nbsp;clusters. The synaptic signaling cluster demonstrated correlations with all other clusters. The cellular response cluster was correlated with all clusters except the neuronal injury cluster. The A\u0026beta;42 cluster was correlated with all clusters except the neuronal injury, axonal metabolic injury, and \u0026nbsp; choroid plexus marker clusters. The cell differentiation cluster was correlated with all clusters except the tau markers, A\u0026beta;42, choroid plexus marker, and \u0026nbsp;APOE4 clusters. The immune (I) cluster was correlated with all clusters except the choroid plexus marker marker cluster. The carbohydrate derivative metabolic processes cluster showed correlations with all clusters. The choroid plexus marker cluster was only correlated with the synaptic signaling, cellular response, and \u0026nbsp;carbohydrate derivative metabolic processes clusters. The immune (II) cluster was correlated only with the synaptic signaling, A\u0026beta;42, cell differentiation, \u0026nbsp;immune (I), and carbohydrate derivative metabolic processes clusters. The APOE4 cluster was correlated with all clusters except the cellular response, cell differentiation, choroid plexus marker, and immune (II) \u0026nbsp;clusters. Finally, we observed a similar trend of correlation of each clusters\u0026rsquo; composite score over A\u0026beta; stages (Fig. 6b).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, a multidimensional clustering approach was designed to examine the topographic association between A\u0026beta;-related CSF biomarkers and cortical gene expression profiles. First, we aimed to identify multimodal biomarker signatures capturing distinct stages of A\u0026beta; pathology. We further sought to delineate how spatially resolved gene-expression profiles relate to these A\u0026beta;-related protein clusters. To establish the reliability of these protein modules, we performed a comprehensive robustness assessment, comprising resampling with the 1,000 bootstrap iterations. We also conducted silhouette analysis and dynamic tree-cut comparisons to ensure that clusters were biologically meaningful and reproducible patterns rather than sampling artifacts or arbitrary threshold-based dendrogram points. Our approach captured distinct biological processes, including tau pathology, neuronal injury, axonal metabolic deficits, synaptic signaling, cellular response, cell differentiation, carbohydrate derivative metabolic process, and two immune clusters designated (I) and (II). Importantly, some clusters correlated with A\u0026beta;-PET burden, highlighting potential markers of disease severity and risk. Although A\u0026beta; aggregation has been conceptualized as a global cortical event, our results showed that the associations between CSF biomarker clusters and A\u0026beta; burden are regionally specific, rather than globally distributed across the brain. The strongest associations with A\u0026beta; burden were observed in clusters related to tau phosphorylation, neuronal injury, and APOE4, and occurred in precisely those brain regions in which the same underlying molecular machinery is highly expressed. This points to an important role of pre-existing transcriptional topology in mediating regional susceptibility to A\u0026beta; deposition. These clusters appeared across several A\u0026beta; staging cortical regions but were largely absent in areas with low baseline gene expression of the respective biological pathways. This suggests that inherent regional biology (not just the disease process) influences where insoluble A\u0026beta; builds up. In addition, our data support the framework that region-specific transcriptional profiles might also explain the resilience and vulnerability to AD \u003csup\u003e6,7\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eHierarchical clustering grouped 120 CSF biomarkers into 12 clusters based on their patterns of co-expression. The tight correlation of expression clusters with A\u0026beta; load highlights the dominance of A\u0026beta; pathology in AD. Our clustering framework was shown to be robust by conducting extensive validation analyses. Resampling using 1,000 bootstrap iterations, silhouette profiling analyses, and dynamic tree cuts validated that our identified modules are indeed representative of stable protein subcommunities as opposed to threshold-driven or sampling-dependent artifacts. GO analysis further validated the clustering by assigning biological themes to the identified clusters. However, in some cases, the nomenclature was modified to better fit the context of AD pathophysiology and improve interpretability.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur data-driven approach corroborates that tau-phosphorylation and neuronal injury clusters are correlated with global neocortical A\u0026beta; SUVR and replicates the concept that tau phosphorylation and neuronal injury are activated as A\u0026beta; accumulates, potentially enhancing neurodegeneration. This is in line with the literature demonstrating that tau markers rise with A\u0026beta; and cause downstream synaptic dysfunction, inflammation \u003csup\u003e9,10\u003c/sup\u003e, and cognitive decline \u003csup\u003e11\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur results also showed high intercorrelations between clusters, suggesting interrelated pathological pathways. The observation that the tau markers cluster correlated with most other clusters is consistent with the broad implication of tau in AD pathology \u003csup\u003e12\u003c/sup\u003e. The neuronal injury cluster also correlated with most clusters, consistent with the known interaction between neuronal damage and inflammatory processes \u003csup\u003e13\u003c/sup\u003e. Subsequent logistic regression analysis identified tau markers and neuronal injury as predictors for A\u0026beta; positivity. Notably, the axonal metabolic injury cluster exhibited a positive correlation with global neocortical A\u0026beta; SUVR, highlighting the importance of energy homeostasis for brain health. These findings are in line with previous transcriptomics and functional studies that causally linked an impaired glucose metabolism with AD progression \u003csup\u003e14,15\u003c/sup\u003e. Together, our data-driven approach supported the well-known associations between tau burden or neuronal injury and A\u0026beta; deposition \u003csup\u003e16\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eThe immune (II) cluster highlights the dominance of C-reactive protein and shows a positive association with global neocortical A\u0026beta; SUVR, predominantly observed in later stages of A\u0026beta; deposition. \u0026nbsp;CRP is a large molecule (115 kD) that possibly reaches the brain via a leaky blood-brain barrier (BBB), \u0026nbsp;validating the role of systemic inflammation in modulating neurodegeneration \u003csup\u003e17\u003c/sup\u003e. The correlation patterns support the notion that AD may be powered by an intricate interdependence among A\u0026beta; accretion, tau pathology, metabolic disturbances, and neuroinflammatory responses, each uniquely contributing to the disease.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe axonal metabolic injury and immune (II) \u0026nbsp;clusters were also positively associated voxel-wise with A\u0026beta;-PET, indicating the involvement of damaged metabolic processes and systemic inflammation in A\u0026beta;-associated pathology \u003csup\u003e18\u003c/sup\u003e. Inter-correlation analysis showed that axonal metabolic injury correlated with most other clusters, suggesting metabolic dysfunction as an early \u0026nbsp;\u003csup\u003e19-21\u003c/sup\u003e and central aspect of AD pathophysiology \u003csup\u003e22-24\u003c/sup\u003e. Astrocytes play a pivotal role in metabolism as the primary regulators of cerebral glucose uptake, lactate shuttling, and metabolic support to neurons. Furthermore, astrocytes also form the metabolic backbone that couples synaptic activity to energy demand \u003csup\u003e25,26\u003c/sup\u003e. Given that astrocytes are also key mediators of neuroinflammatory responses \u003csup\u003e27\u003c/sup\u003e, and the axonal metabolic injury cluster is tightly connected with the immune (II) cluster, our results highlight the intersection between inflammation and metabolic changes.\u003c/p\u003e\n\u003ch3\u003eInterestingly, the synaptic signaling and cellular response clusters showed region-restricted associations, with strong positive associations in ventricular regions but not in the cortex, supporting the notion that impaired CSF dynamics or dysfunctional clearance mechanisms contribute to A\u0026beta; deposition in periventricular spaces \u003csup\u003e28\u003c/sup\u003e. These results align with recent evidence that glymphatic cortical aggregation is an essential contributor to A\u0026beta; clearance failure in AD \u003csup\u003e29\u003c/sup\u003e. Additionally, correlation among clusters further supports that inflammation-associated periventricular pathologies are key players in AD progression \u003csup\u003e30\u003c/sup\u003e.\u0026nbsp;\u003c/h3\u003e\n\u003cp\u003eThe A\u0026beta;42 cluster also exhibited high negative correlation with global neocortical A\u0026beta; SUVR, corroborating the well-known relation between CSF A\u0026beta;42 and brain fibrillar A\u0026beta; plaques in the brain \u003csup\u003e31-33\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eThe negligible correlation between the cell differentiation and immune (I) clusters and A\u0026beta;-PET suggests that either low cortical expression or the association with AD pathophysiology occurs via mechanisms independent of regional A\u0026beta; deposition. For example, interferon signaling is associated with tau pathology in AD \u003csup\u003e34,35\u003c/sup\u003e, and neurodegeneration in AD \u003csup\u003e36\u003c/sup\u003e, but not with cortical \u0026nbsp;A\u0026beta; aggregates. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe carbohydrate derivative metabolic processes cluster correlated with nearly all clusters, indicating the central role of the astrocytic response and lipid metabolism \u003csup\u003e37-39\u003c/sup\u003e. The choroid plexus-related cluster showed minimal correlation with other CSF protein clusters, suggesting a distinct and specific role in AD progression. Subsequent logistic regression analysis identified the APOE4 cluster as a predictor for A\u0026beta; positivity. APOE4 is the strongest and most well-established genetic risk factor for AD, influencing A\u0026beta; aggregation and clearance \u003csup\u003e40\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eThe strong statistical correlations between the A\u0026beta;-PET burden and gene expression clusters further suggested that these molecular networks likely play an active role in disease development. Intriguingly, single-cell sequencing and spatial RNA-sequencing studies have shown that brain-region-specific transcriptomes reflect vulnerability to AD pathologies and neurodegeneration \u003csup\u003e6,7,41\u003c/sup\u003e. For example, cortical regions with high transcriptional expression are predominantly implicated in synaptic function \u003csup\u003e42\u003c/sup\u003e, lipid metabolism \u003csup\u003e43\u003c/sup\u003e, or activation of the innate immune response \u003csup\u003e44\u003c/sup\u003e, where A\u0026beta; deposition occurs earliest and is most severe \u003csup\u003e45\u003c/sup\u003e. Conversely, regions with low expression within those transcriptional circuits remain resistant to pathological change even in advanced disease \u003csup\u003e46\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eOne of the major strengths of this study is that our recovered clusters represented meaningful data structures rather than being assigned at random to dendrogram slices. An analysis of cluster robustness for stability indices, silhouette, and dynamic tree cut indices established that our 12-cluster solution identified stable inference units for expression patterns. Second, beyond looking at the pure associations of clusters with A\u0026beta; load, we demonstrated how spatially resolved gene-expression profiles in healthy brain, are related to these A\u0026beta;-relevant protein clusters. Third, we explored the inter-correlation of these clusters as well, which may be relevant to other processes in AD. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThus, our findings support the framework that the brain-region specific vulnerabilities may predict the disease progression, making them potential therapeutic targets for disease-modifying therapies. These results emphasize that AD progression emerges from an interaction between pathology and pre-existing biological terrain, and highlight the importance of targeting region-specific molecular vulnerabilities in early-intervention strategies.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eLimitations\u003c/h2\u003e\n\u003cp\u003eOur study bears several limitations that must be carefully considered. First, as a single-center study, the replication of our findings in other populations, particularly those with more heterogeneous and larger sample sizes, is required. Secondly, the NULISA\u0026trade;seq panel is a relatively novel biomarker measurement tool, therefore ongoing verification of the biomarkers across larger, multi-ethnic, and longitudinal samples will be useful in ascertaining their usefulness in clinical utilities. Third, the participants in the TRIAD cohort are not representative of the diversity worldwide. This limits extrapolation to other genetic backgrounds, lifestyle determinants, and environmental exposures from other populations. Finally, our study systematically excluded participants with comorbidities, especially those with neurological conditions. Although this was required to minimize confounding factors, it restricts the generalizability of our results to people with complex diseases. Several neurodegenerative diseases are characterized by aberrant protein aggregation and can potentially affect the parameters examined in our study. Subsequent research should enroll patients with comorbid neurological illness to ascertain how such conditions affect the biomarkers and imaging parameters evaluated in our investigation. Future studies should comprise longitudinal analysis to determine if these gene expression changes happen prior to or as a result of A\u0026beta; deposition, which could provide insights into the earliest molecular drivers of AD.\u0026nbsp;\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOverall, our study provides a multidimensional framework of AD pathophysiology. We identified and validated biological CSF modules with clear relevance to A\u0026beta; pathology that may be useful for diagnosis and individualized disease monitoring in early stages of the disease. These CSF signatures strongly correlated with brain-region specific transcriptional profiles mirroring the accumulation of A\u0026beta; and therefore suggesting that brain-region specific vulnerabilities may predict the different CSF modules. Clinically, this study lays the groundwork for improved biomarkers and personalized treatments, from blood-based tests that identify at-risk individuals to multi-drug therapies targeting both A\u0026beta; and downstream network disruptions.\u003c/p\u003e\n"},{"header":"Methods","content":"\u003ch2\u003eParticipants\u003c/h2\u003e\n\u003cp\u003eWe recruited 420 participants from the Translational Biomarkers of Aging and Dementia (TRIAD) cohort, including 267 CU older adults, 59 individuals with MCI due to AD, and 94 individuals with AD dementia. A total of 32 individuals were excluded due to a cognitive impairment not due to AD (e.g., frontotemporal dementia or A\u0026beta;-PET negative results). Additional exclusion criteria included inadequately managed systemic conditions, active substance abuse, recent head trauma, major surgery, or contraindications for medical imaging. Participants in the CU group had a Clinical Dementia Rating (CDR) score of 0. Those with MCI had a CDR score of 0.5 \u003csup\u003e47\u003c/sup\u003e. Participants diagnosed with AD dementia had a CDR score of 1 or 2 \u003csup\u003e48\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eAmong the 420 participants, 384 underwent [\u003csup\u003e18\u003c/sup\u003eF]MK6240 PET imaging, 392 underwent [\u003csup\u003e18\u003c/sup\u003eF]AZD4694 PET imaging, and 403 underwent MRI. A subset of participants (N = 225) had NULISA\u0026trade;seq CNS CSF Panel incorporating 120 biomarkers. To ensure all participants had all imaging modalities and the CNS CSF NULISA\u0026trade;seq panel, the final sample size was 220 participants including, 142 CU older adults, and 78 CI, including 34 mild MCI due to AD and 44 dementia due to AD. Participants were recruited from the community or the McGill Centre for Studies in Aging memory clinic between November 2016 and March 2024. The study was approved by the Montreal Neurological Institute PET Working Committee and the Douglas Mental Health University Institute Research Ethics Board. Written informed consent was obtained from all participants. This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.\u003c/p\u003e\n\u003ch2\u003eImaging Acquisition and Reconstruction\u003c/h2\u003e\n\u003cp\u003eA 3 Tesla Siemens MAGNETOM MRI scanner with a 32-channel head coil was used to acquire T1-weighted images via an Ultrafast Gradient Echo 3D sequence (isotropic 1 mm voxels, TR: 2300 ms, TE: 2.96 ms, FoV: 256 mm, flip angle: 9\u0026deg;). [\u003csup\u003e18\u003c/sup\u003eF]AZD4694 and [\u003csup\u003e18\u003c/sup\u003eF]MK6240 PET scans were performed using a Siemens high-resolution research tomograph dedicated to brain imaging. Each participant received a radioactive dose of 5-7 mCi for each PET scan. Tau-PET images were acquired 90-110 minutes post-injection of [\u003csup\u003e18\u003c/sup\u003eF]MK6240, with four frames (4 \u0026times; 300 s) \u003csup\u003e49\u003c/sup\u003e. A\u0026beta; PET images were collected 40-70 minutes post-injection of [\u003csup\u003e18\u003c/sup\u003eF]AZD4694, with six frames (6 \u0026times; 600 s). Image reconstruction for both PET tracers was performed using a sequential subset expectation-maximization algorithm on a 4D volume. To correct for motion, dead time, decay, as well as random and scattered coincidences, a 6-minute transmission scan with a rotating \u0026sup1;\u0026sup3;⁷Cs point source was conducted after each PET acquisition.\u003c/p\u003e\n\u003ch2\u003ePET Image Processing\u003c/h2\u003e\n\u003cp\u003ePET images were initially linearly registered to the T1-weighted MRI space before undergoing both linear and nonlinear registration to the Alzheimer\u0026rsquo;s Disease Neuroimaging Initiative (ADNI) standard template using ANTs (version 2.2.0). After transformation into ADNI space, the images were spatially smoothed to an 8 mm full width at half maximum resolution for the Modulation Transfer Function curve. To minimize meningeal spillover into adjacent brain regions and enhance signal localization, a stripped meningeal mask, generated with FSL (version 6.0.2), was applied to the PET images \u003csup\u003e50\u003c/sup\u003e. The standardized uptake value ratio (SUVR) for [\u003csup\u003e18\u003c/sup\u003eF]MK6240 PET was calculated using the inferior cerebellum gray matter as the reference region \u003csup\u003e51\u003c/sup\u003e. For [\u003csup\u003e18\u003c/sup\u003eF]AZD4694 PET, the entire cerebellum gray matter served as the reference region \u003csup\u003e52\u003c/sup\u003e. The global neocortical [\u003csup\u003e18\u003c/sup\u003eF]AZD4694 SUVR composite encompassed the neocortex, including the precuneus, prefrontal, orbitofrontal, parietal, temporal, and cingulate cortices, with a positivity threshold of SUVR \u0026gt; 1.55, as previously reported \u003csup\u003e52\u003c/sup\u003e. The summary [\u003csup\u003e18\u003c/sup\u003eF]MK6240 SUVR composite was derived from the temporal meta-region of interest (ROI), which included the entorhinal cortex, amygdala, hippocampus, and the parahippocampal, fusiform, lingual, inferior temporal, and middle temporal gyri \u003csup\u003e53\u003c/sup\u003e. Cortical A\u0026beta; burden was further characterized using the four-stage cortical A\u0026beta; PET staging framework as described previously \u003csup\u003e54\u003c/sup\u003e, applying the corresponding regional staging masks to derive individual A\u0026beta; progression stages.\u003c/p\u003e\n\u003ch2\u003eNULISA\u003csup\u003e TM\u003c/sup\u003eseq assay\u003c/h2\u003e\n\u003cp\u003eCSF samples were processed in the NULISA\u0026trade;seq CNS Panel for 120 neurodegenerative disorder-associated proteins. Next-generation sequencing data were processed with the NULISA\u0026trade;seq algorithm (Alamar Biosciences) and subsequently underwent intra- and inter-plate normalization, as previously reported \u003csup\u003e3\u003c/sup\u003e. Control-normalized counts were log2 transformed and expressed as NULISA\u0026trade;seq Protein Quantification Units, which served as the primary measurement in this study. Further descriptions of the NULISA\u0026trade;seq assay, data handling, and normalization are available \u003csup\u003e55\u003c/sup\u003e.\u003c/p\u003e\n\u003ch2\u003eGene expression quantification\u003c/h2\u003e\n\u003cp\u003eThe messenger ribonucleic acid (mRNA) expression intensity of each cluster was computed by averaging the regional mRNA expression values of all biomarkers included in that cluster. The magnitude of regional mRNA expression reflects the baseline transcriptional activity of each gene across anatomically defined regions of the healthy human brain, as quantified in the Allen Human Brain Atlas. The processed gene expression data were obtained from the open-source database (https://www.meduniwien.ac.at/neuroimaging/mRNA.html) as previously elaborated \u003csup\u003e56,57\u003c/sup\u003e. For the cluster composed of multiple phospho-tau biomarkers (cluster 1: phosphorylated tau (Ptau)181, Ptau217, and Ptau231), the expression of MAPT was used as the representative regional transcript, and for the APOE4 cluster (cluster 12), APOE mRNA expression values were used directly, while all remaining clusters used the averaged regional expression of their respective biomarker genes.\u003c/p\u003e\n\u003ch2\u003eStatistical analyses\u003c/h2\u003e\n\u003ch2\u003eHierarchical clustering\u003c/h2\u003e\n\u003cp\u003eHierarchical clustering was conducted on CNS CSF biomarkers of CI and CU older adults after covariate adjustments. The analysis was executed using Python software integrated with pandas, scipy, matplotlib and scikit-learn library tools. Each biomarker underwent statistical adjustments through a multiple linear regression model using age, binarized sex (0 for male and 1 for female), and temporal meta-ROI tau SUVR as independent variables in order to minimize the influence of confounding variables. For each biomarker, the residuals were derived from the subtraction of predicted values from the observed values. These residuals representing the covariate-adjusted biomarker values were used for the following steps. This ensured that each biomarker was only related to A\u0026beta; pathology after correcting for all covariates.\u003c/p\u003e\n\n\u003cp\u003eHierarchical clustering was then performed on the covariate-adjusted biomarker values to identify patterns in the data. The adjusted values were used as the input for computing pairwise Euclidean distances between them, which were processed through the Ward linkage method from the Python package \u003cem\u003escipy\u003c/em\u003e. Ward\u0026apos;s method minimizes the total within-cluster variance, meaning at each step, the biomarkers with the smallest increase in the overall variance when merged are combined. The linkage function generated a linkage matrix for hierarchical cluster encoding. The dendrogram created by the dendrogram function from scipy function displayed the clustering procedure through visual representation based on the linkage matrix data. The dendrogram displays both the sequence of cluster merges along with the distances used for cluster combination.\u003c/p\u003e\n\n\u003cp\u003eBy examining the dendrogram, clusters of biomarkers with similar adjusted profiles were identified. This grouping helps in understanding how different biomarkers relate to one another after correcting for the effects of the covariates. The final clusters were determined by selecting a threshold on the dendrogram, effectively cutting the tree at a specified level of similarity. Biomarkers with pairwise distances below 20 \u003csup\u003e58\u003c/sup\u003e were considered part of the same cluster. Each cluster then underwent quantification to identify biological patterns based on the selected biomarkers.\u003c/p\u003e\n\n\u003cp\u003eAs the biomarkers in one specific cluster were systematically similar, for each biomarker, Z-scores were calculated, and the composite score for each cluster was then computed per subject, representing the cluster variation as a single marker, using the following formula:\u003c/p\u003e\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/p\u003e\n\u003cp\u003ewhere N represents the number of biomarkers in each cluster. \u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eCluster robustness and validation\u003c/h2\u003e\n\u003cp\u003eFor analyzing the stability and internal validity of the protein clusters in the CSF, we applied a three-step robustness framework comprising: (i) Bootstrap resampling, (ii) Silhouette analysis, and (iii) dynamic tree cutting.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Bootstrap cluster stability\u003c/p\u003e\n\u003cp\u003eCluster reproducibility was assessed using 1,000 bootstrap resamples. In each iteration, participants were sampled with replacement, biomarker residual calculation was adjusted for age, gender, and temporal meta-ROI tau SUVR, and hierarchical clustering using Ward\u0026rsquo;s linkage was repeated. For each pair of proteins, we computed the probability of each protein pair being in the same cluster for each resample, resulting in a 120 \u0026times; 120 matrix for protein pair cluster stability. The stability scores varied between 0 and 1, with scores approaching 0 for pairs rarely observed in the same cluster or approaching 1 for pairs always observed in the same cluster.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Silhouette analysis\u003c/p\u003e\n\u003cp\u003eCluster separation was evaluated using silhouette widths, computed for each protein using Euclidean distances in the residualized biomarker space. Silhouette values range from -1 to 1, where larger indices are desirable and lead to assignment to one cluster rather than to another for those with negative indices. The silhouette index for all proteins averaged together created a global assessment for how well-formed cluster assignments are.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Comparison between dynamic tree cuts\u003c/p\u003e\n\u003cp\u003eTo assess whether the predefined clustering dendrogram height threshold (height = 20) had introduced non-biologically relevant boundaries, the primary clustering solution was contrasted with a data-driven clustering solution derived using dynamic tree-cutting approach. This approach to Clustering relies on adapting to dendrogram topology and local patterns of density to identify candidate clusters (ranging from 8 to 24). This enabled identification of cluster solutions that best aligned with the dendrogram\u0026rsquo;s intrinsic structure while assessing whether the height-based cut introduced non-biological discontinuities. Differences between solutions were examined to ensure that the final 12-cluster configuration reflected biologically meaningful and robust subdivisions.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Gene Ontology\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eGO was performed as a functional enrichment analysis using the R package ClusterProfiler, with default parameters 1. The results were adjusted for multiple comparisons (FDR) using the Benjamini-Hochberg method. Using the \u0026ldquo;GOSlim\u0026rdquo; package 2, we clustered the top GO terms by semantic similarity and represented the similarity matrices in an alluvial plot using the R package \u0026ldquo;ggalluvial\u0026rdquo;. Each cluster was named by summarizing how strongly each theme would appear in each cluster in terms of count and p-value. Subsequently, the names were adapted for the context of AD. \u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eStatistical testing and multiple-comparison correction\u003c/h2\u003e\n\u003cp\u003eAll statistical analyses were performed in Python (pandas, scipy, statsmodels). Group comparisons of cluster composite scores between A\u0026beta;-positive and A\u0026beta;-negative participants were assessed using Mann-Whitney U tests, with FDR correction (Benjamini-Hochberg) applied across all 12 clusters. Associations among clusters and global neocortical A\u0026beta; SUVR were evaluated using Spearman rank correlations, with FDR correction applied to the correlation matrices.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eNeuroimaging voxel-based analysis\u003c/h2\u003e\n\u003cp\u003eVoxel-based linear regression was performed to examine the relationship between each clusters\u0026rsquo; composite score and A\u0026beta;-PET imaging using MATLAB VoxelStats \u003csup\u003e59\u003c/sup\u003e, with adjustments for age, and sex. Multiple comparisons were controlled using the random field theory method \u003csup\u003e60\u003c/sup\u003e, with a significance threshold of P \u0026lt; 0.001 and were visualized as a T-map on the anatomical MRI template. Additionally, the regions showing an \u0026nbsp;association between each cluster\u0026rsquo;s composite score and A\u0026beta;-PET (T-map) were compared with the corresponding regional mRNA expression levels from the healthy brain. Mean values were extracted from both maps to quantify the relationship between \u0026nbsp; A\u0026beta;-cluster associations and baseline gene expression patterns. \u0026nbsp;All tests were two-tailed, and statistical significance was defined as q \u0026lt; 0.05 after multiple-comparison correction unless otherwise specified.\u003c/p\u003e\n\u003ch2\u003eForest Plot Analysis\u003c/h2\u003e\n\u003cp\u003eFor each cluster\u0026rsquo;s composite scores, a logistic regression model was fitted using A\u0026beta; status as the outcome variable and the corresponding cluster composite score as the predictor, to estimate the risk of A\u0026beta; positivity. The odds ratio (OR) and 95% confidence interval were extracted from each model to quantify the association, and FDR correction was applied across the 12 models.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eData availability\u003c/p\u003e\n\u003cp\u003eAll requests for raw and analyzed data and other materials will be promptly reviewed by McGill university to verify if the request is subject to any intellectual property or confidentiality obligations. Anonymized data will be shared upon request to the study\u0026rsquo;s senior author from a qualified academic investigator for the sole purpose of replicating the procedures and results presented in this article. Any data and materials that can be shared will be released via a material transfer agreement. Data are not publicly available due to information that could compromise the privacy of research participants. Related documents, including study protocol and informed consent forms, can similarly be made available upon request.\u003c/p\u003e\n\u003cp\u003eAcknowledgment\u003c/p\u003e\n\u003cp\u003eWe would like to express our gratitude to our participant volunteers and their families for their participation in this study. We thank the staff, research nurses, psychometrist and neurologist at the McGill Centre for Studies in Aging for their contribution. We thank the radiographers and technicians at the McConnell Brain Imaging Centre and The Neuro (Montreal Neurological Institute-Hospital) for their role in imaging data acquisition.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Funding\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis research is supported by an anonymous donor, the European Union\u0026rsquo;s Horizon Europe research and innovation programme under grant agreement no. 101053962, the Weston Brain Institute, Canadian Institutes of Health Research (CIHR) (MOP-11-51-31; RFN 152985, 159815, 162303), Canadian Consortium of Neurodegeneration and Aging (CCNA; MOP-11-51-31-team 1), the Alzheimer\u0026rsquo;s Association (NIRG-12-92090 and NIRP-12-259245), Brain Canada Foundation (CFI Project 34874, 33397), the Fonds de Recherche du Qu\u0026eacute;bec-Sant\u0026eacute; (FRQS; Chercheur Boursier, 2020-VICO-279314) and the Colin J. Adair Charitable Foundation. Y.-T.W. received the FRQS doctoral award. H.Z. is a Wallenberg Scholar and a Distinguished Professor at the Swedish Research Council. Unrelated to the work presented in this paper, H.Z. reports additional grant support from the Swedish Research Council (#2023-00356, #2022-01018 and #2019-02397), Swedish State Support for Clinical Research (#ALFGBG-71320), the Alzheimer Drug Discovery Foundation (ADDF), USA (#201809-2016862), the AD Strategic Fund and the Alzheimer\u0026rsquo;s Association (#ADSF-21-831376-C, #ADSF-21-831381-C, #ADSF-21-831377-C and #ADSF-24-1284328-C), the Bluefield Project, Cure Alzheimer\u0026rsquo;s Fund, the Olav Thon Foundation, the Erling-Persson Family Foundation, Familjen R\u0026ouml;nstr\u0026ouml;ms Stiftelse, Stiftelsen f\u0026ouml;r Gamla Tj\u0026auml;narinnor, Hj\u0026auml;rnfonden, Sweden (#FO2022-0270), the European Union\u0026rsquo;s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement no. 860197 (MIRIADE), the European Union Joint Programme-Neurodegenerative Disease Research (JPND2021-00694), the National Institute for Health and Care Research at University College London (UCL) Hospitals Biomedical Research Centre and the UK Dementia Research Institute at UCL (UKDRI-1003). K.B. is supported by the Swedish Research Council (#2017-00915 and #2022-00732), the Swedish Alzheimer Foundation (#AF-930351, #AF-939721 and #AF-968270),\u003c/p\u003e\n\u003cp\u003eHj\u0026auml;rnfonden, Sweden (#FO2017-0243 and #ALZ2022-0006), the Swedish state under the agreement between the Swedish government and the County Councils, the ALF-agreement (an agreement between central government and seven regions on physician education and clinical research) (#ALFGBG-715986 and #ALFGBG-965240), the European Union Joint Program for Neurodegenerative Disorders (JPND2019-466-236), the Alzheimer\u0026rsquo;s Association 2021 Zenith Award (ZEN-21-848495) and the Alzheimer\u0026rsquo;s Association 2022-2025 Grant (SG-23-1038904 QC). M.S.W. is supported by the Else-Kr\u0026ouml;ner-Fresenius Foundation (2023_EKMS.03), the German Research Foundation (WO 2835/1-1), and the Corona Foundation (S0199/10110/2025).\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;Competing Interest\u003c/p\u003e\n\u003cp\u003eOutside the work presented in this paper, P.R.-N. provides consultancy services for Roche, Cerveau Radiopharmaceuticals, Lilly, Eisai, Pfizer and Novo Nordisk. He also serves as a clinical trial investigator for Biogen, Novo Nordisk and Biogen. M.S.W. receives honoraria from Lilly for educational lectures outside the scope of this manuscript. S.G. is a member of the scientific advisory boards of Alzheon, AmyriAD, Eisai Canada, Enigma USA, Lilly Canada, Medesis, Okutsa Canada, Roche Canada and TauRx. He is a member of the editorial board of JPAD and of the Neurotorium. He has given lectures under the auspices of Biogen Canada and Lundbeck Korea. H.Z. has served at scientific advisory boards and/or as a consultant for Abbvie, Acumen, Alector, Alzinova, ALZpath, Amylyx, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognito Therapeutics, CogRx, Denali, Eisai, Enigma, LabCorp, Merck Sharp \u0026amp; Dohme, Merry Life, Nervgen, Novo Nordisk, Optoceutics, Passage Bio, Pinteon Therapeutics, Prothena, Quanterix, Red Abbey Labs, reMYND, Roche, Samumed, ScandiBio Therapeutics AB, Siemens Healthineers, Triplet Therapeutics, and Wave, has given lectures sponsored by Alzecure, BioArctic, Biogen, Cellectricon, Fujirebio, LabCorp, Lilly, Novo Nordisk, Oy Medix Biochemica AB, Roche, and WebMD, is a co-founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program, and is a shareholder of CERimmune Therapeutics (outside submitted work). K.B. has served as a consultant and on advisory boards for Acumen, ALZPath, BioArctic, Biogen, Eisai, Julius Clinical, Lilly, Novartis, Ono Pharma, Prothena, Roche Diagnostics and Siemens Healthineers; has served at data monitoring committees\u003c/p\u003e\n\u003cp\u003efor Julius Clinical and Novartis; has given lectures, produced educational materials and participated in educational programmes for Biogen, Eisai and Roche Diagnostics; and is a co-founder of BBS in Gothenburg AB, which is a part of the GU Ventures Incubator Program. The remaining authors have no conflicts of interest to report related to this work.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBlennow, K., Hampel, H., Weiner, M. \u0026amp; Zetterberg, H. Cerebrospinal fluid and plasma biomarkers in Alzheimer disease. \u003cem\u003eNature Reviews Neurology\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 131-144 (2010).\u003c/li\u003e\n\u003cli\u003eRayaprolu, S.\u003cem\u003e, et al.\u003c/em\u003e Systems-based proteomics to resolve the biology of Alzheimer\u0026rsquo;s disease beyond amyloid and tau. \u003cem\u003eNeuropsychopharmacology\u003c/em\u003e \u003cstrong\u003e46\u003c/strong\u003e, 98-115 (2021).\u003c/li\u003e\n\u003cli\u003eFeng, W.\u003cem\u003e, et al.\u003c/em\u003e NULISA: a proteomic liquid biopsy platform with attomolar sensitivity and high multiplexing. \u003cem\u003eNature communications\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 7238 (2023).\u003c/li\u003e\n\u003cli\u003ePuchades, M.\u003cem\u003e, et al.\u003c/em\u003e Proteomic studies of potential cerebrospinal fluid protein markers for Alzheimer\u0026apos;s disease. \u003cem\u003eMolecular brain research\u003c/em\u003e \u003cstrong\u003e118\u003c/strong\u003e, 140-146 (2003).\u003c/li\u003e\n\u003cli\u003eTijms, B.M.\u003cem\u003e, et al.\u003c/em\u003e Pathophysiological subtypes of Alzheimer\u0026rsquo;s disease based on cerebrospinal fluid proteomics. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e143\u003c/strong\u003e, 3776-3792 (2020).\u003c/li\u003e\n\u003cli\u003eMathys, H.\u003cem\u003e, et al.\u003c/em\u003e Single-cell multiregion dissection of Alzheimer\u0026rsquo;s disease. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e632\u003c/strong\u003e, 858-868 (2024).\u003c/li\u003e\n\u003cli\u003eGreen, G.S.\u003cem\u003e, et al.\u003c/em\u003e Cellular communities reveal trajectories of brain ageing and Alzheimer\u0026rsquo;s disease. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e633\u003c/strong\u003e, 634-645 (2024).\u003c/li\u003e\n\u003cli\u003eKitani, A., Matsui, Y. \u0026amp; Initiative, A.s.D.N. Integrative network analysis reveals novel moderators of A\u0026beta;-Tau interaction in Alzheimer\u0026apos;s disease. \u003cem\u003eAlzheimer\u0026apos;s Research \u0026amp; Therapy\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 70 (2025).\u003c/li\u003e\n\u003cli\u003eLi, K.\u003cem\u003e, et al.\u003c/em\u003e Synaptic dysfunction in Alzheimer\u0026rsquo;s disease: A\u0026beta;, tau, and epigenetic alterations. \u003cem\u003eMolecular neurobiology\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 3021-3032 (2018).\u003c/li\u003e\n\u003cli\u003eChen, X.\u003cem\u003e, et al.\u003c/em\u003e Microglia-mediated T cell infiltration drives neurodegeneration in tauopathy. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e615\u003c/strong\u003e, 668-677 (2023).\u003c/li\u003e\n\u003cli\u003eZhang, H.\u003cem\u003e, et al.\u003c/em\u003e Interaction between A\u0026beta; and tau in the pathogenesis of Alzheimer\u0026apos;s disease. \u003cem\u003eInternational journal of biological sciences\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 2181 (2021).\u003c/li\u003e\n\u003cli\u003eTolnay \u0026amp; Probst. Tau protein pathology in Alzheimer\u0026rsquo;s disease and related disorders. \u003cem\u003eNeuropathology and applied neurobiology\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, 171-187 (1999).\u003c/li\u003e\n\u003cli\u003eAmor, S., Puentes, F., Baker, D. \u0026amp; Van Der Valk, P. Inflammation in neurodegenerative diseases. \u003cem\u003eImmunology\u003c/em\u003e \u003cstrong\u003e129\u003c/strong\u003e, 154-169 (2010).\u003c/li\u003e\n\u003cli\u003eSantiago, J.A., Bottero, V. \u0026amp; Potashkin, J.A. Transcriptomic and network analysis highlight the association of diabetes at different stages of Alzheimer\u0026rsquo;s disease. \u003cem\u003eFrontiers in Neuroscience\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 1273 (2019).\u003c/li\u003e\n\u003cli\u003eMinhas, P.S.\u003cem\u003e, et al.\u003c/em\u003e Restoring hippocampal glucose metabolism rescues cognition across Alzheimer\u0026rsquo;s disease pathologies. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e385\u003c/strong\u003e, eabm6131 (2024).\u003c/li\u003e\n\u003cli\u003eMinter, M.R., Taylor, J.M. \u0026amp; Crack, P.J. The contribution of neuroinflammation to amyloid toxicity in Alzheimer\u0026apos;s disease. \u003cem\u003eJournal of neurochemistry\u003c/em\u003e \u003cstrong\u003e136\u003c/strong\u003e, 457-474 (2016).\u003c/li\u003e\n\u003cli\u003eSankowski, R., Mader, S. \u0026amp; Vald\u0026eacute;s-Ferrer, S.I. Systemic inflammation and the brain: novel roles of genetic, molecular, and environmental cues as drivers of neurodegeneration. \u003cem\u003eFrontiers in cellular neuroscience\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 28 (2015).\u003c/li\u003e\n\u003cli\u003eObici, L., Perfetti, V., Palladini, G., Moratti, R. \u0026amp; Merlini, G. Clinical aspects of systemic amyloid diseases. \u003cem\u003eBiochimica et Biophysica Acta (BBA)-Proteins and Proteomics\u003c/em\u003e \u003cstrong\u003e1753\u003c/strong\u003e, 11-22 (2005).\u003c/li\u003e\n\u003cli\u003eCai, H.\u003cem\u003e, et al.\u003c/em\u003e Metabolic dysfunction in Alzheimer\u0026apos;s disease and related neurodegenerative disorders. \u003cem\u003eCurrent Alzheimer Research\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 5-17 (2012).\u003c/li\u003e\n\u003cli\u003eMinoshima, S., Cross, D., Thientunyakit, T., Foster, N.L. \u0026amp; Drzezga, A. 18F-FDG PET imaging in neurodegenerative dementing disorders: insights into subtype classification, emerging disease categories, and mixed dementia with copathologies. \u003cem\u003eJournal of nuclear medicine\u003c/em\u003e \u003cstrong\u003e63\u003c/strong\u003e, 2S-12S (2022).\u003c/li\u003e\n\u003cli\u003eChe, P.\u003cem\u003e, et al.\u003c/em\u003e Cerebral perfusion is correlated with cerebral metabolism and amyloid deposition in Alzheimer\u0026rsquo;s disease. \u003cem\u003eTranslational Psychiatry\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 189 (2025).\u003c/li\u003e\n\u003cli\u003eDaulatzai, M.A. Cerebral hypoperfusion and glucose hypometabolism: Key pathophysiological modulators promote neurodegeneration, cognitive impairment, and Alzheimer\u0026apos;s disease. \u003cem\u003eJournal of neuroscience research\u003c/em\u003e \u003cstrong\u003e95\u003c/strong\u003e, 943-972 (2017).\u003c/li\u003e\n\u003cli\u003eKumar, V., Kim, S.-H. \u0026amp; Bishayee, K. Dysfunctional glucose metabolism in Alzheimer\u0026rsquo;s disease onset and potential pharmacological interventions. \u003cem\u003eInternational Journal of Molecular Sciences\u003c/em\u003e \u003cstrong\u003e23\u003c/strong\u003e, 9540 (2022).\u003c/li\u003e\n\u003cli\u003eMosconi, L., Pupi, A. \u0026amp; De Leon, M.J. Brain glucose hypometabolism and oxidative stress in preclinical Alzheimer\u0026apos;s disease. \u003cem\u003eAnnals of the new York Academy of Sciences\u003c/em\u003e \u003cstrong\u003e1147\u003c/strong\u003e, 180-195 (2008).\u003c/li\u003e\n\u003cli\u003eBonvento, G. \u0026amp; Bolanos, J.P. Astrocyte-neuron metabolic cooperation shapes brain activity. \u003cem\u003eCell metabolism\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 1546-1564 (2021).\u003c/li\u003e\n\u003cli\u003eTakahashi, S. Neuroprotective function of high glycolytic activity in astrocytes: common roles in stroke and neurodegenerative diseases. \u003cem\u003einternational journal of molecular sciences\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 6568 (2021).\u003c/li\u003e\n\u003cli\u003eColombo, E. \u0026amp; Farina, C. Astrocytes: key regulators of neuroinflammation. \u003cem\u003eTrends in immunology\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 608-620 (2016).\u003c/li\u003e\n\u003cli\u003eThomas, J.H. Fluid dynamics of cerebrospinal fluid flow in perivascular spaces. \u003cem\u003eJournal of the Royal Society Interface\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 20190572 (2019).\u003c/li\u003e\n\u003cli\u003eHarrison, I.F.\u003cem\u003e, et al.\u003c/em\u003e Impaired glymphatic function and clearance of tau in an Alzheimer\u0026rsquo;s disease model. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e143\u003c/strong\u003e, 2576-2593 (2020).\u003c/li\u003e\n\u003cli\u003eKeil, S.A., Jansson, D., Braun, M. \u0026amp; Iliff, J.J. Glymphatic dysfunction in Alzheimer\u0026rsquo;s disease: A critical appraisal. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e389\u003c/strong\u003e, eadv8269 (2025).\u003c/li\u003e\n\u003cli\u003eFagan, A.M.\u003cem\u003e, et al.\u003c/em\u003e Decreased cerebrospinal fluid A\u0026beta;42 correlates with brain atrophy in cognitively normal elderly. \u003cem\u003eAnnals of Neurology: Official Journal of the American Neurological Association and the Child Neurology Society\u003c/em\u003e \u003cstrong\u003e65\u003c/strong\u003e, 176-183 (2009).\u003c/li\u003e\n\u003cli\u003eJanelidze, S.\u003cem\u003e, et al.\u003c/em\u003e CSF A\u0026beta;42/A\u0026beta;40 and A\u0026beta;42/A\u0026beta;38 ratios: better diagnostic markers of Alzheimer disease. \u003cem\u003eAnnals of clinical and translational neurology\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, 154-165 (2016).\u003c/li\u003e\n\u003cli\u003eBateman, R.J.\u003cem\u003e, et al.\u003c/em\u003e Human amyloid-\u0026beta; synthesis and clearance rates as measured in cerebrospinal fluid in vivo. \u003cem\u003eNature medicine\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 856-861 (2006).\u003c/li\u003e\n\u003cli\u003eSanford, S.A.\u003cem\u003e, et al.\u003c/em\u003e The type‐I interferon response potentiates seeded tau aggregation and exacerbates tau pathology. \u003cem\u003eAlzheimer\u0026apos;s \u0026amp; Dementia\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 1013-1025 (2024).\u003c/li\u003e\n\u003cli\u003eUdeochu, J.C.\u003cem\u003e, et al.\u003c/em\u003e Tau activation of microglial cGAS\u0026ndash;IFN reduces MEF2C-mediated cognitive resilience. \u003cem\u003eNature neuroscience\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 737-750 (2023).\u003c/li\u003e\n\u003cli\u003eJorfi, M.\u003cem\u003e, et al.\u003c/em\u003e Infiltrating CD8+ T cells exacerbate Alzheimer\u0026rsquo;s disease pathology in a 3D human neuroimmune axis model. \u003cem\u003eNature neuroscience\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 1489-1504 (2023).\u003c/li\u003e\n\u003cli\u003eZhang, X.\u003cem\u003e, et al.\u003c/em\u003e Astrocytic functions and lipid metabolism: Correlations and therapeutic targets in Alzheimer\u0026apos;s disease and glioblastoma. \u003cem\u003eClinical and Translational Discovery\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, e287 (2024).\u003c/li\u003e\n\u003cli\u003ePreman, P.\u003cem\u003e, et al.\u003c/em\u003e APOE from astrocytes restores Alzheimer\u0026rsquo;s A\u0026beta;-pathology and DAM-like responses in APOE deficient microglia. \u003cem\u003eEMBO molecular medicine\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 3113-3141 (2024).\u003c/li\u003e\n\u003cli\u003eHaney, M.S.\u003cem\u003e, et al.\u003c/em\u003e APOE4/4 is linked to damaging lipid droplets in Alzheimer\u0026rsquo;s disease microglia. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e628\u003c/strong\u003e, 154-161 (2024).\u003c/li\u003e\n\u003cli\u003eSerrano-Pozo, A., Das, S. \u0026amp; Hyman, B.T. APOE and Alzheimer\u0026apos;s disease: advances in genetics, pathophysiology, and therapeutic approaches. \u003cem\u003eThe Lancet Neurology\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 68-80 (2021).\u003c/li\u003e\n\u003cli\u003ePiwecka, M., Rajewsky, N. \u0026amp; Rybak-Wolf, A. Single-cell and spatial transcriptomics: deciphering brain complexity in health and disease. \u003cem\u003eNature Reviews Neurology\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 346-362 (2023).\u003c/li\u003e\n\u003cli\u003eLiang, W.S.\u003cem\u003e, et al.\u003c/em\u003e Gene expression profiles in anatomically and functionally distinct regions of the normal aged human brain. \u003cem\u003ePhysiological genomics\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e, 311-322 (2007).\u003c/li\u003e\n\u003cli\u003eYang, D.\u003cem\u003e, et al.\u003c/em\u003e Lipid metabolism and storage in neuroglia: role in brain development and neurodegenerative diseases. \u003cem\u003eCell \u0026amp; bioscience\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 106 (2022).\u003c/li\u003e\n\u003cli\u003eRivest, S. Regulation of innate immune responses in the brain. \u003cem\u003eNature Reviews Immunology\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 429-439 (2009).\u003c/li\u003e\n\u003cli\u003eCastillo, E.\u003cem\u003e, et al.\u003c/em\u003e Comparative profiling of cortical gene expression in Alzheimer\u0026rsquo;s disease patients and mouse models demonstrates a link between amyloidosis and neuroinflammation. \u003cem\u003eScientific reports\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 17762 (2017).\u003c/li\u003e\n\u003cli\u003eGrothe, M.J.\u003cem\u003e, et al.\u003c/em\u003e Molecular properties underlying regional vulnerability to Alzheimer\u0026rsquo;s disease pathology. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e141\u003c/strong\u003e, 2755-2771 (2018).\u003c/li\u003e\n\u003cli\u003ePetersen, R.C. Mild cognitive impairment as a diagnostic entity. \u003cem\u003eJournal of internal medicine\u003c/em\u003e \u003cstrong\u003e256\u003c/strong\u003e, 183-194 (2004).\u003c/li\u003e\n\u003cli\u003eMcKhann, G.M.\u003cem\u003e, et al.\u003c/em\u003e The diagnosis of dementia due to Alzheimer\u0026rsquo;s disease: Recommendations from the National Institute on Aging-Alzheimer\u0026rsquo;s Association workgroups on diagnostic guidelines for Alzheimer\u0026apos;s disease. \u003cem\u003eAlzheimer\u0026apos;s \u0026amp; dementia\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 263-269 (2011).\u003c/li\u003e\n\u003cli\u003ePascoal, T.A.\u003cem\u003e, et al.\u003c/em\u003e In vivo quantification of neurofibrillary tangles with [18 F] MK-6240. \u003cem\u003eAlzheimer\u0026apos;s research \u0026amp; therapy\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 1-14 (2018).\u003c/li\u003e\n\u003cli\u003ePascoal, T.A.\u003cem\u003e, et al.\u003c/em\u003e 18F-MK-6240 PET for early and late detection of neurofibrillary tangles. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e143\u003c/strong\u003e, 2818-2830 (2020).\u003c/li\u003e\n\u003cli\u003eJack Jr, C.R.\u003cem\u003e, et al.\u003c/em\u003e Defining imaging biomarker cut points for brain aging and Alzheimer\u0026apos;s disease. \u003cem\u003eAlzheimer\u0026apos;s \u0026amp; Dementia\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 205-216 (2017).\u003c/li\u003e\n\u003cli\u003eTherriault, J.\u003cem\u003e, et al.\u003c/em\u003e Determining amyloid-\u0026beta; positivity using 18F-AZD4694 PET imaging. \u003cem\u003eJournal of Nuclear Medicine\u003c/em\u003e \u003cstrong\u003e62\u003c/strong\u003e, 247-252 (2021).\u003c/li\u003e\n\u003cli\u003eChen, C.D.\u003cem\u003e, et al.\u003c/em\u003e Comparing tau PET visual interpretation with tau PET quantification, cerebrospinal fluid biomarkers, and longitudinal clinical assessment. \u003cem\u003eJournal of Alzheimer\u0026apos;s Disease\u003c/em\u003e \u003cstrong\u003e93\u003c/strong\u003e, 765-777 (2023).\u003c/li\u003e\n\u003cli\u003eCollij, L.E.\u003cem\u003e, et al.\u003c/em\u003e Multitracer model for staging cortical amyloid deposition using PET imaging. \u003cem\u003eNeurology\u003c/em\u003e \u003cstrong\u003e95\u003c/strong\u003e, e1538-e1553 (2020).\u003c/li\u003e\n\u003cli\u003eWang, Y.-T.\u003cem\u003e, et al.\u003c/em\u003e Identify biological Alzheimer\u0026rsquo;s disease using a novel nucleic acid\u0026ndash;linked protein immunoassay. \u003cem\u003eBrain Communications\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, fcaf004 (2025).\u003c/li\u003e\n\u003cli\u003eKomorowski, A.\u003cem\u003e, et al.\u003c/em\u003e Association of protein distribution and gene expression revealed by PET and post-mortem quantification in the serotonergic system of the human brain. \u003cem\u003eCerebral Cortex\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 117-130 (2017).\u003c/li\u003e\n\u003cli\u003eGryglewski, G.\u003cem\u003e, et al.\u003c/em\u003e Spatial analysis and high resolution mapping of the human whole-brain transcriptome for integrative analysis in neuroimaging. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e176\u003c/strong\u003e, 259-267 (2018).\u003c/li\u003e\n\u003cli\u003eWinder, Z.\u003cem\u003e, et al.\u003c/em\u003e Hierarchical clustering analyses of plasma proteins in subjects with cardiovascular risk factors identify informative subsets based on differential levels of angiogenic and inflammatory biomarkers. \u003cem\u003eFrontiers in Neuroscience\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 84 (2020).\u003c/li\u003e\n\u003cli\u003eMathotaarachchi, S.\u003cem\u003e, et al.\u003c/em\u003e VoxelStats: a MATLAB package for multi-modal voxel-wise brain image analysis. \u003cem\u003eFrontiers in neuroinformatics\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 20 (2016).\u003c/li\u003e\n\u003cli\u003eTaylor, J.E. \u0026amp; Worsley, K.J. Detecting sparse signals in random fields, with an application to brain mapping. \u003cem\u003eJournal of the American Statistical Association\u003c/em\u003e \u003cstrong\u003e102\u003c/strong\u003e, 913-928 (2007). \u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8428920/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8428920/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"A growing body of evidence indicates that CSF proteomic signatures shift with increasing brain amyloidosis in Alzheimer’s disease (AD). However, it remains unknown whether protein profiles within cortical regions that are vulnerable to early amyloid-beta (Aβ) deposition contribute to, or predict, CSF Aβ-related protein measures. To address this question, we examined 220 cognitively unimpaired (CU) and cognitively impaired (CI) older participants from the TRIAD cohort who had Aβ- and tau-PET scans, MRI, and CSF NULISA™seq CNS panel data. Aβ-related hierarchical clustering identified 12 clusters corresponding to biologically interpretable gene ontology processes, supported by bootstrap resampling, silhouette analysis, and dynamic tree cutting. Clusters’ composite scores for tau-markers, neuronal-injury, and APOE4 correlated strongly with global neocortical Aβ-PET SUVR (p \u003c 0.001) and showed elevated odds ratios (OR) of Aβ positivity (ORs: 10.4, 2.4, and 4.0, respectively). The Aβ42 cluster, as expected, was inversely associated with brain Aβ burden and predicted reduced Aβ positivity (OR = 0.13). Voxel-wise analyses revealed distinct spatial signatures for each cluster, such as associations in cortical regions (tau-markers, neuronal-injury, and APOE4 clusters), white matter (axonal metabolic injury cluster), and periventricular areas (synaptic signaling and cellular response clusters). The magnitude of Aβ-related cortical associations and CSF protein clusters correlated with the magnitude of regional mRNA expression derived from the Allen Brain Atlas. Our findings support the notion that CSF protein expression reflects underlying regional mRNA expression in cortical regions vulnerable to AD pathophysiology.","manuscriptTitle":"Aβ-related Cerebrospinal Fluid Proteins in Alzheimer’s Disease Reflect Gene Expression Levels Found in The Healthy Cortex","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-08 16:58:16","doi":"10.21203/rs.3.rs-8428920/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c79148a9-bd01-43b3-87c2-c0c878a40421","owner":[],"postedDate":"March 8th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":63828264,"name":"Biological sciences/Neuroscience/Diseases of the nervous system/Alzheimer's disease"},{"id":63828265,"name":"Biological sciences/Neuroscience/Neural ageing"}],"tags":[],"updatedAt":"2026-04-11T08:56:01+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-08 16:58:16","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8428920","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8428920","identity":"rs-8428920","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}