Blood-Based Circular RNAs Enable Early and Accurate Alzheimer’s Disease Diagnosis

Blood-Based Circular RNAs Enable Early and Accurate Alzheimer’s Disease Diagnosis | 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 Blood-Based Circular RNAs Enable Early and Accurate Alzheimer’s Disease Diagnosis Carlos Cruchaga, Bridget Phillips, Jessie Sanford, Menghan Liu, and 20 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8023085/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Detection of Alzheimer’s disease (AD) prior to the development of clinical symptoms is critical due to new treatments for symptomatic AD. Circular RNAs (circRNAs) are highly stable non-coding RNAs enriched in brain that can cross the blood-brain-barrier and are associated with AD. Using two large and independent whole-blood transcriptomic datasets, we identified 34 circRNAs associated with clinical AD status. A single-cutoff predictive model including these 34 circRNAs was comparable to plasma pTau217 in classifying AD based on clinical or AT status, that replicated in the independent dataset. The circRNAs (Hazard Ratio (HR): 2.92) outperformed pTau217 (HR: 1.81) and amyloid-PET when predicting progression to symptomatic AD. CircRNA levels began diverging 4–8 years before symptom onset, enabling preclinical risk stratification. The circRNA model was specific for AD showing low predictive power for PD, DLB, and FTD. These results establish blood circRNAs as robust biomarkers in AD diagnosis and disease progression. Biological sciences/Neuroscience/Diseases of the nervous system/Alzheimer's disease Biological sciences/Biological techniques/Sequencing/RNA sequencing Health sciences/Biomarkers/Predictive markers Biological sciences/Computational biology and bioinformatics/Machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 INTRODUCTION Circular RNAs (circRNAs) are single-stranded RNA back-spliced with the 3’ and 5’ ends connected with a covalently closed loop. CircRNAs are more stable than linear transcripts, with double the half-life of linear RNA, ability to cross the blood-brain-barrier, and high expression in the mammalian brain 1 – 2 . The high stability and tissue-specific regulation of circRNAs suggest their potential utility as blood-based biomarkers. Alzheimer’s disease (AD) is the leading cause of dementia 3 , characterized by amyloid-beta (Aβ) and tau proteins aggregation in the brain paired with synaptic dysfunction and neuronal death. The pathological changes of AD can occur decades before noticeable cognitive decline 7 and detection of AD pathology in cognitively unimpaired individuals may enable interventions that delay or prevent irreversible neurodegeneration. Individuals with dementia have a median survival rate of five years from diagnosis 8 . Because mortality increases with disease progression severity 9 , delaying disease progression could lower mortality. Biomarker-confirmed diagnosis of AD has traditionally utilized biomarkers that measure Aβ, tau, and phosphorylated tau (p-tau) in cerebrospinal fluid (CSF) or amyloid positron emission tomography (PET). However, collection of CSF by lumbar puncture is perceived as invasive and PET is expensive 4 , limiting clinical use. AD biomarkers are continuing to evolve with promising biomarkers in plasma such as pTau217 in preclinical stages 5 – 6 . Plasma pTau217 is strongly associated with amyloid pathology, which starts accumulating many years before onset of AD symptoms, and therefore elevated plasma pTau217 is not strongly associated with clinical AD symptoms. Plasma concentrations of eMTBR-tau243 are associated with tau pathology 10 . However, additional biomarkers that are not just biomarker of AD pathology and capture overall Alzheimer’s disease and symptoms are needed to detect clinical AD and monitor neurodegeneration during this new age of amyloid plaque treatments. Some circRNAs have been shown to be differentially expressed in the brain and blood of individuals with symptomatic AD compared to cognitively normal controls 11 – 13 and thus are candidates for biomarkers of AD symptoms 14 . With high abundance in the brain and enrichment in synapses, circRNAs have been studied in multiple brain regions relevant to AD such as the parietal cortex 11 and hippocampus 12 . Brain circRNAs can differentiate dementia severity and neurodegenerative disease comorbidities. Dube et al . 11 utilized presymptomatic AD and Puri et al . 12 examined AD subtypes with DLB (dementia with Lewy body) and VaD (vascular dementia) pathology. AD-associated circRNAs in blood have been identified using microarray (PMBCs 15 , blood 16 , plasma 17 ) and RNA-seq 13 . Ren et al . 13 reported a small number of circRNA in blood that were associated with AD, but this study was performed in a small dataset (total sample size = 40), and illustrated the potential of blood circRNAs as AD biomarkers. Altogether, identifying circRNAs that differentiate symptomatic AD from cognitively normal individuals in blood allows for the examination of circRNA with AD diagnostic potential. Here, we analyzed cross-sectional blood samples from two large Alzheimer’s disease cohorts, the Knight Alzheimer Disease Research Center (ADRC) and the Anti-Amyloid Treatment in Asymptomatic Alzheimer Disease (A4) dataset, to identify and replicate circRNAs that were associated with AD clinical status AT-stages as well as progression to symptomatic AD. These models were leveraged to develop and validate novel predictive models of AD diagnosis and disease progression. RESULTS Study design To identify blood circRNAs associated with AD, we generated RNA-seq from 816 cognitive unimpaired (CU) and 405 AD cases covering the entire AD continuum. (Fig. 1 ; Table 1 , Supplementary Table 1 ). A total of 717 samples also had CSF Aβ and tau (pTau181) measurements, 776 had amyloid-PET, and 915 had plasma pTau217. We used multiple circRNA bioinformatic tools to perform high-quality and robust circRNA calls and expression levels ( Supplementary Table 2 ). Next, we analyzed if the blood circRNA levels were associated with clinical AD. Blood circRNAs were considered significant if they passed multiple-testing correction of clinical AD status regression using high-quality circRNA read counts. Table 1 Summary demographics of the Knight-ADRC cohorts Blood n Age Avg. (Q1-3) % Male Knight ADRC 1,221 74.09 (69–81) 46.27 AD 405 76.78 (72–83) 50.62 CU 816 72.75 (67–79) 44.12 Number of samples, mean and interquartile range Q1 to Q3 of age at blood draw and % males for the AD and CU (cognitive unimpaired) blood samples in Knight ADRC Next, we developed predictive model for clinical AD status using the associated circRNAs and benchmarked the exact model against Amyloid-PET status and full biomarker-confirmed (AT) status based on Amyloid-PET and CSF Aβ and tau levels. As up to 30% of the CU present with amyloid pathology and later develop disease, we leveraged the longitudinal clinical data to also determine if the circRNA model can identify those who progress to symptomatic AD using a Cox regression and the time in which circRNAs changes in relation to clinical disease onset. To benchmark our model against other established biomarkers, we performed comparison against plasma pTau217, Amyloid-PET or CSF biomarkers using a covariate-adjusted model. We finally analyzed if integrating several types of blood-based-biomarkers (circRNAs and pTau217) lead to better predictive power for progression to symptomatic AD. Previous studies have shown that there may be differences in the predictive power of biomarkers between males and females as well as APOE4 carrier status and different ancestries. For this reason, we also performed sensitivity analyses stratified by sex and APOE , as well as to test the model across ancestries. To determine if the model was specific to AD or captures neurodegeneration in general, we tested the model in additional PD, FTD, and DLB samples. Lastly, replication was performed in the A4 dataset, an independent cohort. As only blood-RNA-seq was available for baseline samples from participants who were CU, we replicated the association of the circRNA model against biomarker-confirmed AD status and AD progression. Identification of Blood circRNAs Associated with AD To identify circRNAs associated with AD in blood, we analyzed the levels of circRNAs with clinical AD status in the AD cases (n = 405) and CU (n = 829) from the Knight ADRC. A total of 1,601 circRNAs passed stringent QC in both DCC 18 and CIRI2 19 . Of the 203 circRNAs with nominal significant association ( P < 0.05) with clinical status, using DCC counts, 35 circRNA transcripts passed False Discovery Rate (FDR) correction ( Supplementary Fig. 1 ), including circDNAJC6 ( P = 1.88×10 − 08 ), circMBOAT2 ( P = 3.80×10 − 04 ), and circPICALM ( P = 1.23×10 − 03 ; Supplementary Table 3 ). In order to confirm our results were robust, we performed sensitivity analyses using CIRI2 counts ( Supplementary Table 4 ) on the 203 circRNA found to be associated with AD using DCC. The effect size of those 203 circRNA when using CIRI2 or DCC were highly correlated (R 2 = 0.78, P < 0.001), as well as between the individual circRNA counts ( Supplementary Fig. 1 ). In addition, 34 of the 35 DCC circRNA remained FDR significant with consistent direction and high effect size correlation ( Supplementary Table 5 ). Thus, a total of 34 circRNA transcripts in blood passed FDR correction using high-quality circRNA read counts from two independent circRNA tools and all 34 were up-regulated in AD cases. AD Diagnostic Accuracy Using circRNAs in Blood We analyzed the diagnostic accuracy of the 34 circRNAs associated with AD. To balance the number of AD cases and CU, we performed undersampling of the blood dataset over 100 iterations ( Supplementary Table 6 ). For clinical dementia status, based on clinical diagnosis, the 34 blood circRNAs showed an AUC (AUC = 0.769; Fig. 2 a, Supplementary Table 7 ) higher than the baseline models that included age at blood draw, sex, median TIN, and number of APOE4 alleles (AUC = 0.634). The circRNA model was not significantly different to the plasma pTau217 alone (AUC = 0.790). Moreover, combining the 34 circRNAs with plasma pTau217 (710 CU, 205 AD) led to the best predictive model with an AUC of 0.865, which is significantly better than those of circRNA or pTau217 alone ( P < 0.05). As there have been several studies 20 – 21 indicating the risk of developing AD is different between males and females, sex-stratified analyses were performed. The circRNA model showed similar predictive power in both sexes (female AUC = 0.778, male AUC = 0.736; Supplementary Tables 8–9 ). Next, we stratified by APOE as APOE4 is the strongest genetic risk factor of AD 3 . The APOE -stratified analyses of clinical status showed similar predictive ability with 0.724 and 0.693 for APOE4 + and APOE4 − respectively ( Supplementary Tables 10–11 ). As CU individuals may present with AD pathology, the most recent biomarker studies have focused on identifying biomarkers to distinguish amyloid positive (A+) from those who are amyloid negative (A-). Therefore, we next examined the circRNA predictive ability to identify individuals with brain amyloidosis based on Amyloid-PET (520 A-, 256 A+). The same circRNA model (same circRNAs, weights and cut off) showed an AUC of 0.733 (Fig. 2 b), improved up to AUC 0.868 when integrating with pTau217 (Fig. 2 d). As amyloid PET only captures one of the main AD pathologies, we also tested if the circRNA model could distinguish between CU who are biomarker negative (AT-) and AD cases who are biomarker positive (AT+) based on CSF Aβ42 and ptau181. The 34 blood circRNAs were able to differentiate CSF amyloid positivity (AUC = 0.739) based on Aβ42 (408 A-, 309 A+), as well as those that were T + vs T- based on CSF pTau181 levels (AUC = 0.693) or tau-PET (292 T-, 233 T+; AUC = 0.641). The circRNA model showed an AUC (AUC = 0.846) for AT- vs AT + biomarker-confirmed samples, which was comparable to that of plasma pTau217 (AUC = 0.877; Fig. 2 c, Supplementary Fig. 2–3 , Supplementary Table 12 ). Furthermore, combining the circRNA model with pTau217 model led to an AUC of 0.937. Likewise, the sex-stratified analyses using AT status showed robust AUC for females (AUC = 0.814) and males (AUC = 0.887). Moreover, the circRNA models showed an AUC of 0.762 in APOE4 − and 0.812 in APOE4 + . Thus, the 34 blood circRNAs had higher predictive ability using biomarker-confirmed status compared to clinical status alone and amyloid biomarker positivity in brain (Fig. 2 d). The blood circRNA predictive model is robust across ancestries To determine if the circRNA model can also be applied to samples from diverse genetic backgrounds, we tested the same model in European (EUR; n = 900) and African (AFR; n = 92) individuals and an additional 35 from diverse backgrounds (African American (AFAM; n = 9) Admixed American (AMR; n = 8), East Asian (EAS; n = 1), South Asian (SAS; n = 1), and multiple ethnicities (n = 9)) based on genetic information. The model run in clinical cases of genetically-defined EUR-background (637 CU, 341 AD) showed similar predictive ability (AUC = 0.748) to the model across ancestries (AUC = 0.769) and clinical cases of genetically-defined AFR-background (AUC = 0.765; Fig. 3 a, Supplementary Table 13 ). As the numbers of individuals in each of the other ancestries were small, we combined all non-EUR individuals ( n = 127). This circRNA model showed an AUC of 0.829 ( Supplementary Table 14 ). The model was also comparable across ancestries in biomarker-confirmed cases (AUC = 0.850), in which the EUR-background (AUC = 0.848) showed similar predictive ability to the AFR-background (0.816). Altogether, the circRNA model showed robust AUC across ancestries. The blood circRNA model is AD-specific To determine whether the 34 circRNA model is specific to AD, we also tested the same model in non-AD dementias (276 PD, 26 DLB, and 11 FTD) using the same cognitive unimpaired participants (816 CU). The 34 circRNA model showed very low AUC when applied to non-AD dementias (PD AUC = 0.440, PD + DLB + FTD AUC = 0.450, PD + DLB AUC = 0.449, PD + FTD AUC = 0.432; Fig. 3 b, Supplementary Fig. 4 ). We further examined if there was any circRNA that was nominally significant in all neurodegenerative diseases (AD, PD, DLB, and FTD) when compared to controls ( Supplementary Fig. 5 ). There were 431 circRNAs that showed an association with either AD or PD, but their effect size correlation was relatively low (R 2 = 0.40). Of these 431 shared circRNAs, 16 circRNA transcripts (15 genes), such as circDNAJC6 , circPICALM , and circMBOAT2 , shared between AD and PD. All 16 of these circRNA were up-regulated in both AD and PD. When comparing across all diseases there were three ( circRBM23 , circEPB41 , and circNUP54 ) of the 34 circRNAs that are part of the predictive models that where at least nominal associated with all of them. These results suggest the most of the circRNA included on the predictive model are AD-specific which could explain the low predictive power in other diseases. Progression to Symptomatic AD As clinical data was available showing that several individuals included 78 participants that progressed from CU to symptomatic AD after blood collection, we performed survival analyses to determine if the 34 blood circRNA could also predict progression to symptomatic AD ( Supplementary Table 15 ). Using a cox regression model, the circRNA model showed a hazard ratio (HR) of 2.92 (95% CI: 1.63–5.23), which was significantly higher of that of pTau217 alone (HR = 1.81, 95% CI: 1.11–2.94; P = 0.002; Fig. 4 a). Similar results were found when analyzed progression to symptomatic AD within five years, with the circRNA model leading to an AUC of 0.870 which was significantly higher ( P = 1.86×10 − 05 ) than the one for pTau217 alone (AUC = 0.676; Fig. 4 b, Supplementary Fig. 6 ). The sex and APOE stratified analyses also showed high AUCs (AUC > 0.80) that were significantly higher than pTau217 ( P < 0.05). Thus, the 34 blood circRNAs predicted AD progression better than the blood biomarker pTau217. Next, we analyzed whether combining pTau217 with the circRNA model can further improve the identification of individuals who progress to symptomatic AD. For these analyses, we compared the individuals who were negative for both biomarkers (pTau217 and the 34 circRNAs; n = 411) to those who were positive for both ( n = 77), those who were positive for pTau217 but negative for the circRNA model ( n = 146), and the opposite pattern ( n = 54; Fig. 4 a). Our analyses indicated that individuals positive for both biomarkers indeed progressed faster than any of the previous models (HR = 4.83, 95% CI: 2.19–10.68). In general, only 15% of individuals negative for both progressed to AD, compared to 84% who progressed to AD and were positive for both. In addition, these analyses identified two intermediate groups: one with medium-low risk, defined by those positive for pTau217 but negative for circRNA (HR = 3.25, 95% CI: 1.36–7.76) from which 41% progress to AD within 5 years; and one with a medium-high risk, from which 74% progress to AD (HR = 3.96, 95% CI: 1.50-10.48; Fig. 4 b). Determining the time of circRNA change in relationship to the disease Pseudo-trajectories of samples progressing to symptomatic AD We used survival modeling to calculate the time to onset (TTO) of the AD progressors in order to infer when the overall 34 circRNA model changes in relation to clinical onset. TTO was calculated by subtracting age at onset (AAO) from age at blood draw, in which, for example, TTO of -6 refers to a sample’s blood draw being 6 years before clinical AD onset. Wee created bins based on TTO with two-year intervals, starting from “-8 to -6” ( n = 5), “-6 to -4” ( n = 20), “-4 to -2” ( n = 33), and “-2 to 0” years ( n = 18). Compared with biomarker negative CU (CU AT-) individuals that do not progress to AD and the different TTO bin, the circRNAs had significant changes starting at the “-4 to -2” TTO range and continuing closer to onset (Fig. 4 c). Moreover, TTO subsets closer to onset showed higher overall circRNA values compared to TTO subsets further away from onset. Altogether, these observations suggest the there is a linear and consistent increase of the overall circRNA levels that starts in the presymptomatic phase around 4 to 2 years before onset an continue increasing until symptomatic AD. We next examined the association of the overall circRNA model with predicting progression of dementia severity based on CDR® 22 (Clinical Dementia Rating). Within the samples with no cognitive impairment at time of blood draw, the blood circRNAs showed an AUC of 0.781 in differentiating between samples from participants who did not progress (CDR = 0; n = 737) and from participants who progressed to cognitive impairment (CDR > 0; n = 42) by the last clinical visit ( Supplementary Fig. 7 ). Altogether, overall level of the 34 circRNAs were associated with dementia severity. Changes of circRNA levels in relation to cognitive changes The previous analyses using longitudinal clinical data suggest that the circRNA model captures AD changes around 5 years before onset. In order to examine this further, we analyzed how the circRNA model changes across the spectrum of memory decline using cross-sectional data. In these analyses, we examined the association of the blood circRNAs with dementia severity based on CDR 22 . Among the blood sample donors, 1,014 individuals had CDR data (CDR = 0; n = 779, CDR = 0.5; n = 115, CDR = 1; n = 84, CDR = 2; n = 33, and CDR = 3; n = 3). Differentiating between participants with no cognitive impairment (CDR = 0) and those in very early disease stage (CDR = 0.5) had an AUC of 0.721 ( Supplementary Fig. 8 ). Thus, these results suggest blood circRNAs capture early disease stage changes with more significant differentiation from CDR 0 to 0.5 compared to later disease stages. Replication of blood circRNA prediction models Orthogonal replication using CIRI2 counts All the prediction models so far used the circRNA counts based on DCC 18 . We next examined whether the DCC-based models had predictive ability using read counts from another independent bioinformatic tool: CIRI2 19 . Overall, using either circRNA tool had similar predictive ability and prediction of clinical AD status using CIRI2 (AUC = 0.720) was comparable to the DCC-based model (AUC = 0.769; Supplementary Fig. 9 ). For biomarker-confirmed status analyses, the performance of the model using CIRI2 (AUC = 0.796), using same weights and cut-off, and the DCC counts (AUC = 0.850) were also similar. Likewise, prediction of progression to symptomatic AD using CIRI2 (AUC = 0.752) was not different from the DCC counts (AUC = 0.870). Following progression prediction, survival analyses were also similar between CIRI2 (HR = 2.63) and DCC (HR = 2.92; P = 0.007). Altogether, using normalized read counts from either circRNA quantification tool did not significantly change the predictive power of the 34 circRNA model, and retraining or cut-off calculation was not needed. Independent replication in the A4 dataset To replicate the blood circRNA model in an independent AD cohort, we mined the existing RNA-seq data derived from whole blood RNA samples from the A4 dataset ( Supplementary Table 16–17 ). Other than amyloid-PET (n = 1,767), the A4 study included plasma pTau217 (n = 680) and longitudinal clinical data. All but one participant was recruited as CU. Progression to symptomatic AD was defined based on changes of CDR score between the time of blood draw and last visit. Of the participants that had plasma pTau217 at baseline, 97 progressed to symptomatic AD within 5 years. The circRNA model showed robust prediction of amyloid-PET positivity (457 A+, 1,310 A-) using either circRNA counts from DCC (AUC = 0.692) and CIRI2 (AUC = 0.664; Supplementary Fig. 10–11 ). As for the Knight ADRC dataset, the circRNA model showed even higher AUC (AUC = 0.715) for biomarker status (A-T- vs A + T). Moreover, the 34 circRNA model also replicated for progression to symptomatic AD showing an HR of 2.93, which was significantly higher than the pTau217 model (HR: 1.87). No significant differences were found when using DCC or CIRI2 counts (Fig. 5 a-b). We also found that combining the circRNA model with pTau217 led to better prediction power for both amyloid PET status and especially for progression. Altogether, the prediction model created using the Knight ADRC cohort replicated in the A4 dataset for brain amyloidosis and progression to symptomatic AD models, highlighting model generalizability and applicability to AD diagnosis in blood. DISCUSSION Circular RNAs have become potential novel candidates for predicting disease status due to their high stability, enrichment, and specificity across tissues 1 – 2 . Identifying circRNAs with differential expression in blood allows for clinically relevant prediction models to be created for pinpointing possible AD biomarkers. Here, we identified 34 blood circRNA transcripts associated with clinical and biomarker-confirmed status, and, more importantly, associated with progression from cognitively unimpaired to symptomatic AD. The robust replications in the A4 clinical dataset confirm the robustness of the circRNA models that are also significantly higher that plasma pTau217 when predicting progression to symptomatic AD, establishing blood circRNAs as a novel and potential non-invasive, high precision tool for early AD diagnosis and progression monitoring in clinical practice. Prior circRNAs studies have been largely focused in the brain 11 – 12 due to their high abundance and synapse enrichment. While these studies have spanned multiple AD-relevant brain regions relevant to AD including the parietal cortex 11 and hippocampus 12 , the brain is not readily accessible. Comparatively, studies on circRNAs in whole blood from AD participants are sparse but promising 16 . A recent study 13 indicates that circRNAs could potentially be used to predict clinical status confirmed by CSF biomarkers (pTau181/Aβ42), even though the study was performed in a small dataset (20 AD, 20 CU). Yet, such discriminatory power in clinical AD status may not fully reflect circRNAs ability to detect AD at detection of earlier stages and presymptomatic stages AD and additional studies in larger studies are needed. Recent biomarker development studies have shifted towards assessing amyloid accumulation rather than clinical status to identify asymptomatic individuals with underlying AD pathology. Plasma pTau217 is widely regarded as the leading plasma-based biomarker of AD in preclinical stages 5 – 6 and has been shown to be, indeed, a marker of amyloid pathology that changes 15–20 years prior to clinical onset. However, pTau217 can be unreliable as a biomarkers in patients treated with anti-amyloid antibody therapies, as it may normalize with brain Aβ plaques removal without corresponding cognitive improvements 23 . Biomarkers independent of Aβ and tau pathology are essential to monitor overall neurodegeration in this therapeutic area. Our 34 blood circRNAs appear to capture overall AD disease biology and not just pathology. Interestingly, the host genes of some of the circRNA part of the model such as circDNAJC6 is involved in synaptic function, while APP (amyloid precursor protein) is regulated by circCTCF and circRANBP9 host genes. On top of this, two of our 34 circRNA biomarkers predictive of AD are derived from the gene PICALM , which is a GWAS-significant AD risk gene associated with both Aβ, tau, APOE4 function and synaptic dysfunction and microglia-derived neuroinflammation. Thus, when we evaluated the accuracy of circRNA for biomarker-defined status, the model generated a robust AUC of 0.850 in biomarker confirmed (AT+) patients, and 0.733 for amyloid-PET positivity, demonstrating strong discriminatory power comparable to the current standard of care modalities. In addition to direct comparison with CSF and brain imaging biomarkers, we benchmarked our model against the AD blood biomarker: pTau217. Plasma pTau217 changes 15–20 years prior to clinical onset and models with better prediction of AD onset are required to determine progression to symptomatic AD 24 . Notably, the blood circRNA model had higher predictive ability of AD progression in 5 years (AUC = 0.870) compared to plasma pTau217 alone (AUC = 0.676). These findings highlight circRNA’s ability to capture dynamic progression signals which other pathology-focused biomarkers may miss. Prediction of progression to symptomatic AD was even further improved after combining the pTau217 and circRNA models. This combined model was able to distinguish non-progressors (negative for both biomarkers) from high progressors (both positive), with 84% of the positive group progressing to AD compared to 15% of the non-progressors. Given that plasma pTau217 predominantly tracks amyloid pathology, there is a pressing need for more biomarkers which extend beyond amyloidosis to detect and separate progression of cognitive impairment from decreased pathology due to anti-amyloid antibody treatments. Our results strongly support the use of blood-based circRNAs as non-invasive, scalable biomarkers for detecting overall AD pathology, with robust performance that surpasses gold standards in predicting progression to symptomatic AD. Beyond the ability to act as diagnostic biomarkers, the blood circRNAs predicted progression to symptomatic AD in CU individuals at the time of blood collection. In fact, progression analyses further reveal that circRNA levels changes between 4–8 years prior to disease onset, closer to symptom emergence. These results establish the 34 identified circRNAs as a novel and comprehensive approach for accurate AD diagnosis, with superior progression prediction and AD specificity against current gold standards like PET and pTau217, ultimately advancing early detection and monitoring of AD. METHODS Datasets The Knight Alzheimer Disease Research Center (Knight ADRC) and the PD MARS study at Washington University in St. Louis were used in the analyses. The Knight ADRC is supported by the NIH (National Institutes of Health) and conducts prospective studies on memory and aging for the treatment and prevention of AD. Study eligibility includes participants with asymptomatic or mild dementia at enrollment and the age of 45 or older. The MAP study involves the longitudinal collection of biofluids (plasma, cerebrospinal fluid), neuroimaging, and annual clinical assessments. This study utilized bulk blood RNA-seq data comprising 3,670 blood samples from 2,573 individuals. The Knight ADRC study included 405 AD cases, 816 CU (cognitively unimpaired), and the AD-related dementias (ADRD) including 1 PD (Parkinson’s disease) case, 17 DLB (dementia with Lewy body) cases, and 11 FTD (frontotemporal dementia) cases.. The PD MARS study, also supported by the NIH, supplied data and samples for the remaining 275 PD cases and 9 DLB cases included in this study. The mean age at blood draw were similar between individuals with AD and CU individuals with the average ages of 76.8 and 74.5 respectively. The Anti-Amyloid Treatment in Asymptomatic Alzheimer Disease (A4) study included individuals with evidence of amyloid accumulation based on amyloid-PET. The A4 study is a clinical trial to examine if anti-amyloid treatment slows down the rate of cognitive decline. The study enrolled CU individuals and longitudinally evaluated cognition based on neuropsychological tests. The A4 study comprises longitudinal measurements of amyloid-PET and plasma pTau217. Blood RNA-seq was available for CU samples at baseline and included 1,797 samples from 1,797 individuals. This cohort included a subset of the A4 clinical trial cohort of amyloid-PET positive asymptomatic individuals that were followed up tot 5 years for conversion to dementia/symptomatic AD, defined based on changes of CDR ( n = 190), a subset of the LEARN amyloid-PET negative cohort followed up for the same timeframe ( n = 128), as well as additional patients initially screened and not selected in the above cohorts, a subset of which had longitudinal data as a result of extension studies ( n = 358) followed up to 7 years. This cohort also included baseline blood draw compared to last visit CU ( n = 469), progressors ( n = 97) with available pTau217 data for comparison. Within the totality of this cohort (n = 1,767), the 1,797 CU individuals included 1,796 with CDR of 0 at blood draw and 1 individual with CDR of 0.5. Between these samples, the average age at blood draw of individuals with no cognitive impairment was 71.4 and 82 for very early disease stage. RNA sequencing This study generated 151-nt, paired-end, rRNA and globin depleted, RNA-seq data from whole blood donated by participants in the Knight ADRC MAP study. All the participants consented to blood donation and clinical analysis. RNA was purified from whole blood using the Maxwell RSC simplyRNA blood kit. RNA samples were extracted from whole blood in Paxgene tubes and transferred using Biomek. Globin and rRNA depletion was performed using FastSelect, RIN was quantified by TapeStation 4200, and RNA was then sequenced on an Illumina NovaSeq 6000 at the MGI at Washington University in St. Louis. Alzheimer disease traits The AD phenotype examined included clinically determined status. Samples with neuropath AD status are defined based on Clinical Dementia Rating® 22 (CDR®) score greater than 0 and CERAD (Consortium to Establish a Registry for Alzheimer’s Disease) 25 categorizing the sample as AD. Cognitive unimpaired (CU) is defined as “low probability AD” or “not AD” neuropath status and samples having both a CDR 22 score of 0 and Braak 26 tau (neurofibrillary tangle) stage under III. Phenotype processing Principal components analysis was performed to generate genetic ancestry covariates using PLINK 27 and genetic variant array data. Genetic PC1 and 2 were used to determine genetic ancestry to keep European samples. Related and duplicate samples were removed using cryptic relatedness through identity by descent (IBD; PIHAT ≥ 0.25) in PLINK. The median values of TIN (Transcript Integrity Number) were used to measure RNA quality. Compared to RIN (RNA Integrity Number) being calculated per sample, TIN is measured per transcript and was calculated using tin.py from the RSeQC package 28 with GENCODE v33 as reference. The blood cohort from the Knight ADRC included 3,656 blood samples. Samples with the covariates age at blood draw, sex, median TIN, batch, number of APOE4 alleles and AD clinical status were kept, with 1,221 individuals that passed QC. RNA detection and filtering Circular RNAs (circRNAs) were identified using DCC 18 and CIRI2 19 to reduce the false-positivity rate of circRNA prediction tools, defined by the fraction of RNase R sensitive circRNAs 29 – 30 . Alignment of raw sequencing reads to the GRCh38 human reference genome assembly for DCC was performed using STAR v.2.7.8a 31 in chimeric alignment mode. The DCC count parameter was increased (-Nr 5 5) for the blood cohort to output circRNAs with a minimum of 5 counts in 5 samples and account for the computational burden of the larger cohort ( n = 3,656). Alignment of reads for CIRI2 was performed using BWA-MEM 32 with the recommended minimum alignment score (-T) of 19. Linearly aligned and chimerically aligned reads were compared and backsplices with a minimum circ:linear read ratio of 0.1 in at least 3 samples were kept. CircRNA transcripts were used for downstream analyses if they contained at least 1 read count in at least 25% of the samples in both DCC and CIRI2. The blood samples from the Knight ADRC ( n = 1,221) were used to identify blood circRNA, in which 2,069 DCC circRNA and 1,871 CIRI2 circRNA passed count filtering. The blood cohort had 1,601 circRNA identified by both DCC and CIRI2. Transcript-level read counts were normalized for each cohort based on library size and sequencing depth using variance stabilizing transformation (VST) from the DESeq2 35 R package. Linear transcript identification The linear RNA-Seq pipeline included raw sequence data quality control (QC) checks using FastQC 33 , STAR 31 alignment, Picard (Broad Institute) summary statistics, and transcript quantification using Salmon 34 . Picard 2.27.4 FastqToSam converted the fastqc files to unmapped.bam and Picard RevertSam then cleared alignment information attributes in the bam file to allow STAR to process the file as paired-end. STAR 31 version 2.7.8a was used to align paired-end reads with the GRCH38 genome index. After using samtools sort, Picard Collect-RNA-Seq-Metrics, Collect-Alignment-Summary-Metrics, and Mark-Duplicates was run on the sorted STAR output file. Salmon was used to quantify the linear genes in quasi-mapping-based alignment mode and genes with less than 10 counts in less than 90% of individuals were filtered out. Lastly, linear RNA counts were normalized per cohort based on library size and sequencing depth using VST from the DESeq2 35 R package. Differential expression and correlation analyses Differential expression (DE) analysis was performed on the cross-sectional blood samples and was adjusted by age at blood draw, sex, batch, and median TIN. Negative binominal family logistic regression of AD vs CU (cognitive unimpaired) status was performed using DESeq2 35 and the DE results were FDR corrected with a 0.05 threshold. The model design for blood was circRNA counts ~ age + sex + batch + median TIN + case-CU status. Diagnostic models Logistic regression models of the top differentially expressed circRNAs were created using blood circRNAs, including age at blood draw, sex, and number of APOE4 alleles as covariates. Samples with covariates (age, sex, APOE4 ) and AD phenotype information were included in the models. Diagnostic utility of the models were analyzed by ROC curve and AUC analysis using the R packages pROC 36 and ROCR 37 . The dependent variable for all models was diagnosis and had AD cases for positive events and cognitive unimpaired for negative events. Prediction models were performed with 100 iterations of undersampling to balance the sample sizes of the cases and CU per phenotype. The weights and cutoff of the AD vs CU status model were determined using the glm function and model training was performed using the predict function. Youden’s index from pROC 36 was used to optimize and derive the threshold for calculating metrics such as accuracy, sensitivity, and specificity. These weights and cut-off were then used when testing the model performance against Amyloid-PET, CSF biomarker or AT status. Due to the different statistical framework, progression to symptomatic AD used the weights obtained from the Cox regression model (see below) Undersampling was performed for the full model and was not performed in the subsequent stratification by sex and APOE4 status. For the sex or APOE4 -stratified analysis, the same weights and cutoff as the full model was tested in individuals were used. The overall circRNA counts for A4 were lower compared to that of the Knight ADRC and a lower threshold cutoff was utilized in A4 progression analyses (Supplementary Fig. 11). Prediction probability of the circRNA model was calculated using the weights and cutoff of the AD vs CU clinical status model. The progression to symptomatic AD model used separate probability values and cutoff derived from the Cox regression model. Biomarker negative samples were determined by the prediction probability value being less than the progression cutoff and biomarker positive samples had a probability value greater than the progression cutoff. To infer the ancestry specificity of the blood circRNA models, we performed the AD prediction models using population subsets with the 1000 Genomes Project as a reference. Of the blood samples with genetic information, the PCA identified 990 EUR, 92 AFR, 16 AfAM, 9 Admixture, 8 AMR, 1 EAS, and 1 SAS sample. AD progression analyses For the progression to symptomatic AD, Cox proportional hazard was performed using the R package survival and Kaplan Meier (KM) curves were plotted via the R package survminer. The Cox proportional hazard model included age at draw and sex as the covariates and circRNA counts per circRNA. The status variable included 610 samples from participants who remained cognitive unimpaired at last visit (coded as binary variable 1 in the model) and 68 samples from participants who converted to AD by last visit (variable 2 in model). The time variable was calculated by subtracting the age at last visit from age at blood draw, ranging from 0 to 5 years. Declarations Competing interests CC has received research support from GSK and EISAI. The funders of the study had no role in the collection, analysis, or interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. CC is a member of the advisory board of Vivid Genomics and Circular Genomics and owns stocks in these companies. CC and BP have invention disclosures for the circRNA models for AD diagnosis. Acknowledgements We would like to thank the study participants. This work was supported by grants from the National Institutes of Health (R01AG044546 (CC), P01AG003991 (CC), RF1AG053303 (CC), RF1AG058501 (CC), U01AG058922 (CC), the Chan Zuckerberg Initiative (CC), the Michael J. Fox Foundation (CC), and the Alzheimer’s Association Zenith Fellows Award (ZEN-22-848604, awarded to CC). The PD MARS study sample collection was supported by the American Parkinson disease association, the Barnes-Jewish hospital foundation including the Elliot Stein family fund, the Paula and Rodger Riney fund, and the Jansky-Bander fund and NIH (NINDS, NIA) NS075321. The RNA-seq in A4 work was supported by the DP2 AG082342 (RB) and R01 AG079142 (RB). This work was supported by access to equipment made possible by the Hope Center for Neurological Disorders, the NeuroGenomics and Informatics Center (NGI: https://neurogenomics.wustl.edu/ ) and the Departments of Neurology and Psychiatry at Washington University School of Medicine. The recruitment and clinical characterization of research participants at Washington University were supported by NIH P30AG066444 (DMH), P01AG03991 (JCM), and P01AG026276 (JCM). Data availability Knight ADRC sequencing data is available to approved investigators through https://knightadrc.wustl.edu/data-request-form/ . A4 sequencing data is available to approved investigators through https://vmacdata.org/vmap/data-requests . Code availability Publicly available software was used for all analyses. The software included PLINK 27 for IBD and genetic ancestry, STAR v.2.7.8a 31 and BWA-MEM 32 for alignment, DCC 18 and CIRI2 19 to identify circRNAs, and linear mRNA RNA-Seq pipeline tools (FastQC 33 , STAR 31 , Picard 2.27.4 (Broad Institute), Salmon 34 ). Differential expression analysis was performed using the DESeq2 35 R package and diagnostic utility of logistic regression models were analyzed using the R packages pROC 36 and ROCR 37 . Cox proportional hazard was performed using the survival R package and Kaplan Meier curves were plotted using the survminer R package. References Rybak-Wolf A, Stottmeister C, Glažar P, Jens M, Pino N, Giusti S, Hanan M, Behm M, Bartok O, Ashwal-Fluss R, Herzog M, Schreyer L, Papavasileiou P, Ivanov A, Öhman M, Refojo D, Kadener S, Rajewsky N. Circular RNAs in the Mammalian Brain Are Highly Abundant, Conserved, and Dynamically Expressed. Mol Cell . 2015;58(5):870–85. doi: 10.1016/j.molcel.2015.03.027 . Jeck WR, Sorrentino JA, Wang K, Slevin MK, Burd CE, Liu J, Marzluff WF, Sharpless NE. Circular RNAs are abundant, conserved, and associated with ALU repeats. RNA . 2013;19(2):141–57. doi: 10.1261/rna.035667.112 . Scheltens P, De Strooper B, Kivipelto M, Holstege H, Chételat G, Teunissen CE, Cummings J, van der Flier WM. Alzheimer's disease. Lancet . 2021;397(10284):1577–1590. doi: 10.1016/S0140- 6736(20)32205–4. Florean I, Penolazzi B, Menichelli A, Pastore M, Cattaruzza T, Mazzon G, Manganotti P. Using the ATN system as a guide for the neuropsychological assessment of Alzheimer's disease. J Clin Exp Neuropsychol . 2021;43(9):926–943. doi: 10.1080/13803395.2022.2036327 . Ashton NJ, Puig-Pijoan A, Milà-Alomà M, Fernández-Lebrero A, García-Escobar G, González-Ortiz F, et al. Plasma and CSF biomarkers in a memory clinic: Head-to-head comparison of phosphorylated tau immunoassays. Alzheimers Dement . 2023;19(5):1913–1924. doi: 10.1002/alz.12841 . Janelidze S, Bali D, Ashton NJ, Barthélemy NR, Vanbrabant J, Stoops E, Vanmechelen E, He Y, Dolado AO, Triana-Baltzer G, Pontecorvo MJ, Zetterberg H, Kolb H, Vandijck M, Blennow K, Bateman RJ, Hansson O. Head-to-head comparison of 10 plasma phospho-tau assays in prodromal Alzheimer's disease. Brain . 2023;146(4):1592–1601. doi: 10.1093/brain/awac333 . Villemagne VL, Burnham S, Bourgeat P, Brown B, Ellis KA, Salvado O, Szoeke C, Macaulay SL, Martins R, Maruff P, Ames D, Rowe CC, Masters CL; Australian Imaging Biomarkers and Lifestyle (AIBL) Research Group. Amyloid β deposition, neurodegeneration, and cognitive decline in sporadic Alzheimer's disease: a prospective cohort study. Lancet Neurol . 2013;12(4):357–67. doi: 10.1016/S1474-4422(13)70044-9 . Joling KJ, Janssen O, Francke AL, Verheij RA, Lissenberg-Witte BI, Visser PJ, van Hout HPJ. Time from diagnosis to institutionalization and death in people with dementia. Alzheimers Dement . 2020;16(4):662–671. doi: 10.1002/alz.12063 . Crowell V, Reyes A, Zhou SQ, Vassilaki M, Gsteiger S, Gustavsson A. Disease severity and mortality in Alzheimer's disease: an analysis using the U.S. National Alzheimer's Coordinating Center Uniform Data Set. BMC Neurol . 2023;23(1):302. doi: 10.1186/s12883-023-03353-w . Horie K, Salvadó G, Koppisetti RK, Janelidze S, Barthélemy NR, He Y, Sato C, Gordon BA, Jiang H, Benzinger TLS, Stomrud E, Holtzman DM, Mattsson-Carlgren N, Morris JC, Palmqvist S, Ossenkoppele R, Schindler SE, Hansson O, Bateman RJ. Plasma MTBR-tau243 biomarker identifies tau tangle pathology in Alzheimer's disease. Nat Med . 2025;31(6):2044–2053. doi: 10.1038/s41591-025-03617-7 . Dube U, Del-Aguila JL, Li Z, Budde JP, Jiang S, Hsu S, Ibanez L, Fernandez MV, Farias F, Norton J, Gentsch J, Wang F; Dominantly Inherited Alzheimer Network (DIAN); Salloway S, Masters CL, Lee JH, Graff-Radford NR, Chhatwal JP, Bateman RJ, Morris JC, Karch CM, Harari O, Cruchaga C. An atlas of cortical circular RNA expression in Alzheimer disease brains demonstrates clinical and pathological associations. Nat Neurosci . 2019;22(11):1903–1912. doi: 10.1038/s41593-019-0501-5 . Puri S, Hu J, Sun Z, Lin M, Stein TD, Farrer LA, Wolozin B, Zhang X. Identification of circRNAs linked to Alzheimer's disease and related dementias. Alzheimers Dement . 2023. doi: 10.1002/alz.12960 . Ren Z, Chu C, Pang Y, Cai H, Jia L. A circular RNA blood panel that differentiates Alzheimer's disease from other dementia types. Biomark Res . 2022;10(1):63. doi: 10.1186/s40364-022-00405-0 . Huang A, Zheng H, Wu Z, Chen M, Huang Y. Circular RNA-protein interactions: functions, mechanisms, and identification. Theranostics. 2020;10(8):3503–3517. doi: 10.7150/thno.42174 . Liu L, Chen X, Chen YH, Zhang K. Identification of Circular RNA hsa_Circ_0003391 in Peripheral Blood Is Potentially Associated With Alzheimer's Disease. Front Aging Neurosci . 2020;12:601965. doi: 10.3389/fnagi.2020.601965 . Li Y, Lv Z, Zhang J, Ma Q, Li Q, Song L, Gong L, Zhu Y, Li X, Hao Y, Yang Y. Profiling of differentially expressed circular RNAs in peripheral blood mononuclear cells from Alzheimer's disease patients. Metab Brain Dis . 2020;35(1):201–213. doi: 10.1007/s11011-019-00497-y . Piscopo P, Manzini V, Rivabene R, Crestini A, Le Pera L, Pizzi E, Veroni C, Talarico G, Peconi M, Castellano AE, D'Alessio C, Bruno G, Corbo M, Vanacore N, Lacorte E. A Plasma Circular RNA Profile Differentiates Subjects with Alzheimer's Disease and Mild Cognitive Impairment from Healthy Controls. Int J Mol Sci . 2022;23(21):13232. doi: 10.3390/ijms232113232 . Cheng J, Metge F, Dieterich C. Specific identification and quantification of circular RNAs from sequencing data. Bioinformatics . 2016;32(7):1094–6. doi: 10.1093/bioinformatics/btv656 . Gao Y, Zhang J, Zhao F. Circular RNA identification based on multiple seed matching. Brief Bioinform . 2018;19(5):803–810. doi: 10.1093/bib/bbx014 . Belloy ME, Andrews SJ, Le Guen Y, Cuccaro M, Farrer LA, Napolioni V, Greicius MD. APOE Genotype and Alzheimer Disease Risk Across Age, Sex, and Population Ancestry. JAMA Neurol . 2023;80(12):1284–1294. doi: 10.1001/jamaneurol.2023.3599 . Aggarwal NT, Mielke MM. Sex Differences in Alzheimer's Disease. Neurol Clin . 2023;41(2):343–358. doi: 10.1016/j.ncl.2023.01.001 . Morris JC. The Clinical Dementia Rating (CDR): current version and scoring rules. Neurology . 1993;43(11):2412–4. doi: 10.1212/wnl.43.11.2412-a . Egan MF, Kost J, Voss T, Mukai Y, Aisen PS, Cummings JL, Tariot PN, Vellas B, van Dyck CH, Boada M, Zhang Y, Li W, Furtek C, Mahoney E, Harper Mozley L, Mo Y, Sur C, Michelson D. Randomized Trial of Verubecestat for Prodromal Alzheimer's Disease. N Engl J Med . 2019;380(15):1408–1420. doi: 10.1056/NEJMoa1812840 . Palmqvist S, Janelidze S, Quiroz YT, Zetterberg H, Lopera F, Stomrud E, Su Y, Chen Y, Serrano GE, Leuzy A, Mattsson-Carlgren N, Strandberg O, Smith R, Villegas A, Sepulveda-Falla D, Chai X, Proctor NK, Beach TG, Blennow K, Dage JL, Reiman EM, Hansson O. Discriminative Accuracy of Plasma Phospho-tau217 for Alzheimer Disease vs Other Neurodegenerative Disorders. JAMA . 2020;324(8):772–781. doi: 10.1001/jama.2020.12134 . Mirra SS, Heyman A, McKeel D, Sumi SM, Crain BJ, Brownlee LM, Vogel FS, Hughes JP, van Belle G, Berg L. The Consortium to Establish a Registry for Alzheimer's Disease (CERAD). Part II. Standardization of the neuropathologic assessment of Alzheimer's disease. Neurology . 1991;41(4):479–86. doi: 10.1212/wnl.41.4.479 . Braak H, Braak E. Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol . 1991;82(4):239–59. doi: 10.1007/BF00308809 . Purcell S, Neale B, Todd-Brown K, et al. PLINK: a toolset for whole-genome association and population-based linkage analysis. American Journal of Human Genetics . 2007;81(3):559–75 Wang L, Nie J, Sicotte H, Li Y, Eckel-Passow JE, Dasari S, Vedell PT, Barman P, Wang L, Weinshiboum R, Jen J, Huang H, Kohli M, Kocher JP. Measure transcript integrity using RNA-seq data. BMC Bioinformatics . 2016;17:58. doi: 10.1186/s12859-016-0922-z . Hansen TB, Venø MT, Damgaard CK, Kjems J. Comparison of circular RNA prediction tools. Nucleic Acids Res . 2016;44(6):e58. doi: 10.1093/nar/gkv1458 . Hansen TB. Improved circRNA Identification by Combining Prediction Algorithms. Front Cell Dev Biol . 2018;6:20. doi: 10.3389/fcell.2018.00020 . Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, Gingeras TR. STAR: ultrafast universal RNA-seq aligner. Bioinformatics . 2013;29(1):15–21. doi: 10.1093/bioinformatics/bts635 . Li H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. 2013; arXiv:1303.3997v2. Andrews S. FastQC: A Quality Control Tool for High Throughput Sequence Data [Online]. 2010. Available online at: http://www.bioinformatics.babraham.ac.uk/projects/fastqc/ . Patro R, Duggal G, Love MI, Irizarry RA, Kingsford C. Salmon provides fast and bias-aware quantification of transcript expression. Nat Methods . 2017;14(4):417–419. doi: 10.1038/nmeth.4197 . Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol . 2014;15(12):550. doi: 10.1186/s13059-014-0550-8 . Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez JC, Müller M. pROC: an open- source package for R and S + to analyze and compare ROC curves. BMC Bioinformatics . 2011;12:77. doi: 10.1186/1471-2105-12-77 . Sing T, Sander O, Beerenwinkel N, Lengauer T. ROCR: visualizing classifier performance in R. Bioinformatics . 2005;21(20):3940–1. doi: 10.1093/bioinformatics/bti623 . Additional Declarations Yes there is potential Competing Interest. CC has received research support from GSK and EISAI. The funders of the study had no role in the collection, analysis, or interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. CC is a member of the advisory board of Vivid Genomics and Circular Genomics and owns stocks in these companies. CC and BP have invention disclosures for the circRNA models for AD diagnosis. Supplementary Files 08CircPredictionSupp.Figures.docx Supplementary Figures RSCruchaga.pdf Reporting Summary Cite Share Download PDF Status: Under Review 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-8023085","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":553834450,"identity":"e5422240-fe5d-4f58-8f5b-75f83ad9f52b","order_by":0,"name":"Carlos Cruchaga","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0klEQVRIiWNgGAWjYLACxgYgwQ4iGCyAOIFYLTwHQKolSNEikUCkFv72s88kfu6ws+uf+cZM8ucPCQZ+9hwDvFokzqSbSfaeSU6ecTvHTJoHaItkzxv8Wgwk2NikGduYkxlu526TBjnM4AYBW6Ba6pPlb57dJvkDqMWeSC2H7Qxu8G6TADnMQIKgX9KYLXvbjicYnsn/bM2TJsEjceZZAV4t/O3HGG/8bKu2lzt+LPHmDxsbOf725A14tcBAYgOUwUOUchCwJ1rlKBgFo2AUjDwAAOeLPrOIz3gYAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0002-0276-2899","institution":"Washington University, School of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Carlos","middleName":"","lastName":"Cruchaga","suffix":""},{"id":553834451,"identity":"76588296-535a-46ac-b6e9-93a22ade1860","order_by":1,"name":"Bridget Phillips","email":"","orcid":"https://orcid.org/0000-0002-5261-689X","institution":"Washington University, School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Bridget","middleName":"","lastName":"Phillips","suffix":""},{"id":553834452,"identity":"afb93ecf-1990-44b5-b5e5-74bc35e5078c","order_by":2,"name":"Jessie Sanford","email":"","orcid":"https://orcid.org/0000-0001-5992-3027","institution":"Washington University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Jessie","middleName":"","lastName":"Sanford","suffix":""},{"id":553834453,"identity":"644b3618-0cb5-4247-b203-dc1f44477e98","order_by":3,"name":"Menghan Liu","email":"","orcid":"https://orcid.org/0009-0008-9902-5983","institution":"Washington University in St. Louis","correspondingAuthor":false,"prefix":"","firstName":"Menghan","middleName":"","lastName":"Liu","suffix":""},{"id":553834454,"identity":"32725c03-084f-4ddd-b76a-53d0a562510e","order_by":4,"name":"Matthew Johnson","email":"","orcid":"","institution":"Washington University in St Louis","correspondingAuthor":false,"prefix":"","firstName":"Matthew","middleName":"","lastName":"Johnson","suffix":""},{"id":553834455,"identity":"fc085024-6000-4b73-8e41-eeafd425120f","order_by":5,"name":"Katherine Gong","email":"","orcid":"https://orcid.org/0009-0008-7614-9545","institution":"Washington University, School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Katherine","middleName":"","lastName":"Gong","suffix":""},{"id":553834456,"identity":"c3e5e5e7-f936-4f8d-a2b8-f08968007c34","order_by":6,"name":"Kristy Bergmann","email":"","orcid":"","institution":"Washington University in St. Louis","correspondingAuthor":false,"prefix":"","firstName":"Kristy","middleName":"","lastName":"Bergmann","suffix":""},{"id":553834457,"identity":"89142ad6-fad7-463e-b45d-f8f78072fc3c","order_by":7,"name":"Joseph Lowery","email":"","orcid":"https://orcid.org/0009-0009-0318-7716","institution":"Washington University, School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Joseph","middleName":"","lastName":"Lowery","suffix":""},{"id":553834458,"identity":"f48b41de-ffb5-4983-997e-b85550310244","order_by":8,"name":"Allison Flynn","email":"","orcid":"","institution":"Washington University, School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Allison","middleName":"","lastName":"Flynn","suffix":""},{"id":553834459,"identity":"e86042fd-87f6-414d-8300-03052edc3b8a","order_by":9,"name":"William Brock","email":"","orcid":"","institution":"Washington University, School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"William","middleName":"","lastName":"Brock","suffix":""},{"id":553834460,"identity":"fb85fc18-5ade-42c9-ae4f-10d7bdb7e32f","order_by":10,"name":"Brenda Sanchez Montejo","email":"","orcid":"","institution":"Washington University, School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Brenda","middleName":"Sanchez","lastName":"Montejo","suffix":""},{"id":553834461,"identity":"fb33a615-7a99-460b-9f26-64f624cc905a","order_by":11,"name":"Nicholas Sykora","email":"","orcid":"https://orcid.org/0009-0008-1203-8627","institution":"Washington University, School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Nicholas","middleName":"","lastName":"Sykora","suffix":""},{"id":553834462,"identity":"fd594bd4-2432-4c79-a398-f9f44f8757f1","order_by":12,"name":"John Budde","email":"","orcid":"https://orcid.org/0000-0001-7778-7276","institution":"Washington University in St. Louis","correspondingAuthor":false,"prefix":"","firstName":"John","middleName":"","lastName":"Budde","suffix":""},{"id":553834463,"identity":"ef18c38c-a4aa-4c16-a042-6735c695856b","order_by":13,"name":"Nikolaos Mellios","email":"","orcid":"","institution":"Circular Genomics Inc","correspondingAuthor":false,"prefix":"","firstName":"Nikolaos","middleName":"","lastName":"Mellios","suffix":""},{"id":553834464,"identity":"70f83fb1-81a4-4368-810b-dc1207fa0cbc","order_by":14,"name":"Timothy Hohman","email":"","orcid":"https://orcid.org/0000-0002-3377-7014","institution":"Vanderbilt University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Timothy","middleName":"","lastName":"Hohman","suffix":""},{"id":553834465,"identity":"05cee477-1667-4d1d-8e35-94988545eed5","order_by":15,"name":"Vaibhav Janve","email":"","orcid":"","institution":"Vanderbilt University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Vaibhav","middleName":"","lastName":"Janve","suffix":""},{"id":553834466,"identity":"cc823f00-337e-4a3a-bf97-f1d1c01feaf1","order_by":16,"name":"Reisa Sperling","email":"","orcid":"https://orcid.org/0000-0003-1535-6133","institution":"Harvard Aging Brain Study, Department of Neurology, Massachusetts General Hospital, Boston, MA 02114;Center for Alzheimer Research and Treatment, Department of Neurology, Brigham and Women’s","correspondingAuthor":false,"prefix":"","firstName":"Reisa","middleName":"","lastName":"Sperling","suffix":""},{"id":553834467,"identity":"2bc583ce-ac6a-4e9c-b4ba-26da0fff3f0c","order_by":17,"name":"Rachel Buckley","email":"","orcid":"https://orcid.org/0000-0002-5356-5537","institution":"Harvard Medical School","correspondingAuthor":false,"prefix":"","firstName":"Rachel","middleName":"","lastName":"Buckley","suffix":""},{"id":553834468,"identity":"7325e238-7d8c-4eb7-899e-f212973982b8","order_by":18,"name":"Mabel Seto","email":"","orcid":"","institution":"Brigham and Women's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Mabel","middleName":"","lastName":"Seto","suffix":""},{"id":553834469,"identity":"d1d1c739-83e5-4241-9d9b-3c2a75fd02d2","order_by":19,"name":"John Morris","email":"","orcid":"","institution":"Washington University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"John","middleName":"","lastName":"Morris","suffix":""},{"id":553834470,"identity":"ec7d2f5a-25b9-4098-a608-1e5e2b619251","order_by":20,"name":"Joel Perlmutter","email":"","orcid":"https://orcid.org/0000-0003-3263-8917","institution":"Washington University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Joel","middleName":"","lastName":"Perlmutter","suffix":""},{"id":553834471,"identity":"d837b20c-7808-43f8-8fdb-fdfec11cbe83","order_by":21,"name":"Paul Kotzbauer","email":"","orcid":"","institution":"Washington University, School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Paul","middleName":"","lastName":"Kotzbauer","suffix":""},{"id":553834472,"identity":"3f09ddb3-37f0-4b0a-87c0-7037dda043ca","order_by":22,"name":"Richard Perrin","email":"","orcid":"https://orcid.org/0000-0002-3443-7716","institution":"Washington University in St. Louis","correspondingAuthor":false,"prefix":"","firstName":"Richard","middleName":"","lastName":"Perrin","suffix":""},{"id":553834473,"identity":"706d982e-74cf-49a8-88cf-95ca2612112c","order_by":23,"name":"Laura Ibanez","email":"","orcid":"","institution":"Washington University in St. Louis","correspondingAuthor":false,"prefix":"","firstName":"Laura","middleName":"","lastName":"Ibanez","suffix":""}],"badges":[],"createdAt":"2025-11-04 00:40:49","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8023085/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8023085/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":97767303,"identity":"a6992726-84a1-4ac4-974c-53fd12c8f8a3","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"tif","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1148000,"visible":true,"origin":"","legend":"","description":"","filename":"03CircPredictionFigures1.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/10814a2b59c1db08a620c5f2.tif"},{"id":97897665,"identity":"290a1999-c606-4d6e-90ef-421817b1313c","added_by":"auto","created_at":"2025-12-10 15:38:04","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96439,"visible":true,"origin":"","legend":"","description":"","filename":"02CircADPredictionManuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/00de3e79c5990b5be73710b2.docx"},{"id":97897040,"identity":"c16a6926-8b75-400f-8a18-d4bc23205594","added_by":"auto","created_at":"2025-12-10 15:37:22","extension":"tif","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6661798,"visible":true,"origin":"","legend":"","description":"","filename":"04CircPredictionFigures2.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/4b1df45398ff910a3c3ba19c.tif"},{"id":97767312,"identity":"ca1cec82-d8cd-41c4-85ec-8297510f2f6c","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4826878,"visible":true,"origin":"","legend":"","description":"","filename":"05CircPredictionFigures3.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/3f15590c56e571dc336aa2ed.tif"},{"id":97767315,"identity":"05ae3aea-8cda-4dac-a153-27377bcf0486","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"tif","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5492210,"visible":true,"origin":"","legend":"","description":"","filename":"06CircPredictionFigures4.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/207f0b99f51158201513e4e7.tif"},{"id":97767324,"identity":"6c42604c-b4b4-4255-aded-0049f98d0664","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"tif","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4402404,"visible":true,"origin":"","legend":"","description":"","filename":"07CircPredictionFigures5.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/c3e652e1bd9226fbbfbb0a08.tif"},{"id":97767316,"identity":"7c28b9d8-9bae-460a-b772-b81f1f37142a","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"json","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":21677,"visible":true,"origin":"","legend":"","description":"","filename":"NMEDA146843A.json","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/4a74eafdf29c7743c1384491.json"},{"id":97896382,"identity":"99d4077e-d4bd-4ea2-8c97-0f35c175eed3","added_by":"auto","created_at":"2025-12-10 15:36:28","extension":"docx","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3569272,"visible":true,"origin":"","legend":"","description":"","filename":"08CircPredictionSupp.Figures.docx","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/9ce4565255c6ace11c38f2d0.docx"},{"id":97897134,"identity":"171aa3ea-839f-4352-bdee-fb8b3682c8b0","added_by":"auto","created_at":"2025-12-10 15:37:28","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":153463,"visible":true,"origin":"","legend":"","description":"","filename":"NMEDA146843A1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/deb8badbaee2e697685597a7.xml"},{"id":97896142,"identity":"583ac936-10d2-440a-b1f4-82b10865652b","added_by":"auto","created_at":"2025-12-10 15:35:57","extension":"tif","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1148000,"visible":true,"origin":"","legend":"","description":"","filename":"03CircPredictionFigures1.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/70a83b72d58aed053ecbdbc7.tif"},{"id":97767326,"identity":"a22843f1-ab60-496d-8d4c-ac987f7501df","added_by":"auto","created_at":"2025-12-09 07:22:28","extension":"tif","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6661798,"visible":true,"origin":"","legend":"","description":"","filename":"04CircPredictionFigures2.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/c154eee9d06b5fc4d7f48edf.tif"},{"id":97767328,"identity":"57d990a6-6303-484c-9f29-38ae9d0b2889","added_by":"auto","created_at":"2025-12-09 07:22:28","extension":"tif","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4826878,"visible":true,"origin":"","legend":"","description":"","filename":"05CircPredictionFigures3.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/567116296b415e26003609dc.tif"},{"id":97767321,"identity":"1b6c2911-05fa-447d-a459-76350baea2d3","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"tif","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5492210,"visible":true,"origin":"","legend":"","description":"","filename":"06CircPredictionFigures4.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/98a3a4dc48978a05ed272d01.tif"},{"id":97767329,"identity":"c58a421f-0c7e-4d34-841a-fca853a378ab","added_by":"auto","created_at":"2025-12-09 07:22:28","extension":"tif","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4402404,"visible":true,"origin":"","legend":"","description":"","filename":"07CircPredictionFigures5.tif","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/23ab6b0f218c52c866fd51ec.tif"},{"id":97897080,"identity":"dd8d48a7-2c65-4b7e-949e-127e5350dc54","added_by":"auto","created_at":"2025-12-10 15:37:25","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":163887,"visible":true,"origin":"","legend":"","description":"","filename":"Online03CircPredictionFigures1.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/3bce75710d45b3097a980d44.png"},{"id":97767310,"identity":"df1495bf-0ccc-417e-94db-0dbd2914c60f","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":463097,"visible":true,"origin":"","legend":"","description":"","filename":"Online04CircPredictionFigures2.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/3bc687cfa6243e1069c6ef04.png"},{"id":97767318,"identity":"520fa74c-12ed-49b5-a4a1-3cff79efa8e4","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":315309,"visible":true,"origin":"","legend":"","description":"","filename":"Online05CircPredictionFigures3.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/304c892e50e2ca97f37021db.png"},{"id":97767319,"identity":"dc92b33f-48b1-4c07-a672-f80385919c6b","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":380911,"visible":true,"origin":"","legend":"","description":"","filename":"Online06CircPredictionFigures4.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/ef155d9125f0b5c8688c24f8.png"},{"id":97767325,"identity":"72ae6f2f-7956-422d-9ee0-de183cddb0f3","added_by":"auto","created_at":"2025-12-09 07:22:28","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":291644,"visible":true,"origin":"","legend":"","description":"","filename":"Online07CircPredictionFigures5.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/019e74cee4f5ec408b6aea1a.png"},{"id":97767323,"identity":"04a8b624-441d-435d-bae1-d8a29782d6ca","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"xml","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":150392,"visible":true,"origin":"","legend":"","description":"","filename":"NMEDA146843A1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/5f1e7f47a4c62341a71c7a40.xml"},{"id":97767330,"identity":"1d459aec-c0ca-45e3-82b1-1a0ad1737bb2","added_by":"auto","created_at":"2025-12-09 07:22:28","extension":"html","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":167646,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/dfa00541d702a15c065b0e93.html"},{"id":97767302,"identity":"5724980f-3f50-4ba2-b744-728bf0ab3555","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":4343525,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStudy design. \u003c/strong\u003eSchematic of the cognitively unimpaired (CU) and AD case groups within the logistic models predicting AD status of Knight ADRC blood samples. The AD-related diseases included PD (Parkinson’s disease), DLB (dementia with Lewy body), and FTD (frontotemporal dementia). Model replication was performed in the A4 (Anti-Amyloid Treatment in Asymptomatic Alzheimer Disease) dataset.\u003c/p\u003e","description":"","filename":"03CircPredictionFigures1.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/a0c866cc3916e85fc0b6eee2.png"},{"id":97897473,"identity":"c1dda003-904d-46bd-99a1-b242a2b1b27b","added_by":"auto","created_at":"2025-12-10 15:37:51","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":22315284,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClinical and amyloid pathology of AD. \u003c/strong\u003eROC curve analysis using the 34 circRNAs with status as (\u003cstrong\u003ea\u003c/strong\u003e) AD vs CU (405 AD, 816 CU), (\u003cstrong\u003eb\u003c/strong\u003e) Amyloid-PET positivity (256 A+, 520 A-), and (\u003cstrong\u003ec\u003c/strong\u003e) biomarker-confirmed status (79 AT+ AD, 252 AT- CU). (\u003cstrong\u003ed\u003c/strong\u003e) Whisker plot of AD status and Amyloid-PET positivity prediction using CSF Aβ42/pTau181, Amyloid-PET, plasma pTau217, and the 34 blood circRNAs. The circRNA models included library-size normalized counts of the top status circRNAs and the covariates (age at draw, sex, median TIN, number of \u003cem\u003eAPOE4\u003c/em\u003e alleles). CircRNA models were compared to known AD biomarkers alone.\u003c/p\u003e","description":"","filename":"04CircPredictionFigures2.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/53e8acb552907f45b667f26f.png"},{"id":97767307,"identity":"0654e7d5-1b58-4e4d-8807-28787702dc95","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":15400630,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eValidation of the prediction models. \u003c/strong\u003e(\u003cstrong\u003ea\u003c/strong\u003e) ROC curve analysis predicting AD status in European (\u003cem\u003en\u003c/em\u003e=978; 341 AD, 637 CU), African (\u003cem\u003en\u003c/em\u003e=92; 14 AD, 78 CU), and MIX (\u003cem\u003en\u003c/em\u003e=127; 37 AD, 90 CU) samples. MIX combines the 92 AFR, 16 AfAM, 9 Admixture, 8 AMR, 1 EAS, and 1 SAS blood samples. Populations were determined using 1000 genomes as a reference. (\u003cstrong\u003eb\u003c/strong\u003e)\u003cstrong\u003e \u003c/strong\u003eROC curve analysis of logistic models predicting AD status in Knight ADRC blood using AD-related dementias (816 CU; 405 AD, 276 PD, 26 DLB, 11 FTD). The models included library-size normalized counts of the 34 status circRNAs and the covariates (age at draw, sex, median TIN, number of \u003cem\u003eAPOE4\u003c/em\u003e alleles).\u003c/p\u003e","description":"","filename":"05CircPredictionFigures3.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/3e9c2a89fdf4500f45e48e69.png"},{"id":97767313,"identity":"bde190c7-4551-45ea-9680-09339d1620bd","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":19283027,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBlood circRNAs were more predictive of AD progression than pTau217. \u003c/strong\u003e(\u003cstrong\u003ea\u003c/strong\u003e) ROC curve analysis of logistic models predicting AD progression within 5-years in Knight ADRC blood CU samples (\u003cem\u003en\u003c/em\u003e=688; 78 progressors, 610 CU) using the 34 blood circRNAs and plasma pTau217 positivity. The predictor+covariate models included circRNA counts or pTau217 and the covariates (age at draw, sex, median TIN, number of \u003cem\u003eAPOE4\u003c/em\u003ealleles). (\u003cstrong\u003eb\u003c/strong\u003e) Kaplan Meier curves of AD progression using 34 blood circRNAs and pTau217 positivity. (\u003cstrong\u003ec\u003c/strong\u003e) Violin plot of prediction values over TTO (time to onset) subsets from estimated onset in AD progressors compared to biomarker-confirmed status samples. TTO was calculated based on the age at onset (AAO) subtracted from age at blood draw. Dotted line of the 0.342 AUC threshold for AD vs CU classification using the 34 blood circRNAs.\u003c/p\u003e","description":"","filename":"06CircPredictionFigures4.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/7dd750c8214b636bcd071762.png"},{"id":97896781,"identity":"8cd8ee94-e1e2-48f2-b57e-463cda6294a8","added_by":"auto","created_at":"2025-12-10 15:37:04","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":12847723,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eReplication of prediction models. \u003c/strong\u003e(\u003cstrong\u003ea\u003c/strong\u003e) Whisker plot of AD prediction using Knight ADRC (WU) DCC and A4 DCC of the 34 blood circRNAs compared to plasma pTau217 in A4. (\u003cstrong\u003eb\u003c/strong\u003e) Whisker plot of AD prediction using Knight ADRC CIRI2 and A4 CIRI2 of the 34 blood circRNAs compared to pTau217 in A4. The circRNA models included library-size normalized counts of the top status circRNAs and the covariates (age at draw, sex, median TIN, number of \u003cem\u003eAPOE4\u003c/em\u003e alleles).\u003c/p\u003e","description":"","filename":"07CircPredictionFigures5.png","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/7bd17bd27ff2713374a9b3bf.png"},{"id":98421381,"identity":"61409d16-ac1d-483b-aa24-f8739e5cc8de","added_by":"auto","created_at":"2025-12-17 16:26:59","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":70486682,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/427b74d5-7dd5-494b-b913-bb774e107918.pdf"},{"id":97767306,"identity":"585e0219-af9f-480e-89c5-cd021b218532","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":3569272,"visible":true,"origin":"","legend":"Supplementary Figures","description":"","filename":"08CircPredictionSupp.Figures.docx","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/608c42a5f57632bdd45fe84e.docx"},{"id":97767305,"identity":"df70cb32-8993-46e6-bf49-824f1eb39120","added_by":"auto","created_at":"2025-12-09 07:22:27","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":2675852,"visible":true,"origin":"","legend":"Reporting Summary","description":"","filename":"RSCruchaga.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8023085/v1/87abd1527f7842c74300ecec.pdf"}],"financialInterests":"\u003cb\u003eYes\u003c/b\u003e there is potential Competing Interest.\nCC has received research support from GSK and EISAI. The funders of the study had no role in the collection, analysis, or interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. CC is a member of the advisory board of Vivid Genomics and Circular Genomics and owns stocks in these companies. CC and BP have invention disclosures for the circRNA models for AD diagnosis.","formattedTitle":"Blood-Based Circular RNAs Enable Early and Accurate Alzheimer’s Disease Diagnosis","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eCircular RNAs (circRNAs) are single-stranded RNA back-spliced with the 3\u0026rsquo; and 5\u0026rsquo; ends connected with a covalently closed loop. CircRNAs are more stable than linear transcripts, with double the half-life of linear RNA, ability to cross the blood-brain-barrier, and high expression in the mammalian brain\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. The high stability and tissue-specific regulation of circRNAs suggest their potential utility as blood-based biomarkers.\u003c/p\u003e\u003cp\u003eAlzheimer\u0026rsquo;s disease (AD) is the leading cause of dementia\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e, characterized by amyloid-beta (Aβ) and tau proteins aggregation in the brain paired with synaptic dysfunction and neuronal death. The pathological changes of AD can occur decades before noticeable cognitive decline\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e and detection of AD pathology in cognitively unimpaired individuals may enable interventions that delay or prevent irreversible neurodegeneration. Individuals with dementia have a median survival rate of five years from diagnosis\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. Because mortality increases with disease progression severity\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, delaying disease progression could lower mortality.\u003c/p\u003e\u003cp\u003eBiomarker-confirmed diagnosis of AD has traditionally utilized biomarkers that measure Aβ, tau, and phosphorylated tau (p-tau) in cerebrospinal fluid (CSF) or amyloid positron emission tomography (PET). However, collection of CSF by lumbar puncture is perceived as invasive and PET is expensive\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e, limiting clinical use. AD biomarkers are continuing to evolve with promising biomarkers in plasma such as pTau217 in preclinical stages\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. Plasma pTau217 is strongly associated with amyloid pathology, which starts accumulating many years before onset of AD symptoms, and therefore elevated plasma pTau217 is not strongly associated with clinical AD symptoms. Plasma concentrations of eMTBR-tau243 are associated with tau pathology\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. However, additional biomarkers that are not just biomarker of AD pathology and capture overall Alzheimer\u0026rsquo;s disease and symptoms are needed to detect clinical AD and monitor neurodegeneration during this new age of amyloid plaque treatments.\u003c/p\u003e\u003cp\u003eSome circRNAs have been shown to be differentially expressed in the brain and blood of individuals with symptomatic AD compared to cognitively normal controls\u003csup\u003e\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e and thus are candidates for biomarkers of AD symptoms\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. With high abundance in the brain and enrichment in synapses, circRNAs have been studied in multiple brain regions relevant to AD such as the parietal cortex\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e and hippocampus\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Brain circRNAs can differentiate dementia severity and neurodegenerative disease comorbidities. Dube \u003cem\u003eet al\u003c/em\u003e.\u003csup\u003e11\u003c/sup\u003e utilized presymptomatic AD and Puri \u003cem\u003eet al\u003c/em\u003e.\u003csup\u003e12\u003c/sup\u003e examined AD subtypes with DLB (dementia with Lewy body) and VaD (vascular dementia) pathology. AD-associated circRNAs in blood have been identified using microarray (PMBCs\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, blood\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e, plasma\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e) and RNA-seq\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Ren \u003cem\u003eet al\u003c/em\u003e.\u003csup\u003e13\u003c/sup\u003e reported a small number of circRNA in blood that were associated with AD, but this study was performed in a small dataset (total sample size\u0026thinsp;=\u0026thinsp;40), and illustrated the potential of blood circRNAs as AD biomarkers. Altogether, identifying circRNAs that differentiate symptomatic AD from cognitively normal individuals in blood allows for the examination of circRNA with AD diagnostic potential.\u003c/p\u003e\u003cp\u003eHere, we analyzed cross-sectional blood samples from two large Alzheimer\u0026rsquo;s disease cohorts, the Knight Alzheimer Disease Research Center (ADRC) and the Anti-Amyloid Treatment in Asymptomatic Alzheimer Disease (A4) dataset, to identify and replicate circRNAs that were associated with AD clinical status AT-stages as well as progression to symptomatic AD. These models were leveraged to develop and validate novel predictive models of AD diagnosis and disease progression.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStudy design\u003c/h2\u003e\u003cp\u003eTo identify blood circRNAs associated with AD, we generated RNA-seq from 816 cognitive unimpaired (CU) and 405 AD cases covering the entire AD continuum. (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e; Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, \u003cb\u003eSupplementary Table\u0026nbsp;1\u003c/b\u003e). A total of 717 samples also had CSF Aβ and tau (pTau181) measurements, 776 had amyloid-PET, and 915 had plasma pTau217. We used multiple circRNA bioinformatic tools to perform high-quality and robust circRNA calls and expression levels (\u003cb\u003eSupplementary Table\u0026nbsp;2\u003c/b\u003e). Next, we analyzed if the blood circRNA levels were associated with clinical AD. Blood circRNAs were considered significant if they passed multiple-testing correction of clinical AD status regression using high-quality circRNA read counts.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSummary demographics of the Knight-ADRC cohorts\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBlood\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u003cem\u003en\u003c/em\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAge Avg. (Q1-3)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e% Male\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eKnight ADRC\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e1,221\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e74.09 (69\u0026ndash;81)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e46.27\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e405\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e76.78 (72\u0026ndash;83)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e50.62\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCU\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e816\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e72.75 (67\u0026ndash;79)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e44.12\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e\u003cp\u003eNumber of samples, mean and interquartile range Q1 to Q3 of age at blood draw and % males for the AD and CU (cognitive unimpaired) blood samples in Knight ADRC\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eNext, we developed predictive model for clinical AD status using the associated circRNAs and benchmarked the exact model against Amyloid-PET status and full biomarker-confirmed (AT) status based on Amyloid-PET and CSF Aβ and tau levels. As up to 30% of the CU present with amyloid pathology and later develop disease, we leveraged the longitudinal clinical data to also determine if the circRNA model can identify those who progress to symptomatic AD using a Cox regression and the time in which circRNAs changes in relation to clinical disease onset. To benchmark our model against other established biomarkers, we performed comparison against plasma pTau217, Amyloid-PET or CSF biomarkers using a covariate-adjusted model. We finally analyzed if integrating several types of blood-based-biomarkers (circRNAs and pTau217) lead to better predictive power for progression to symptomatic AD.\u003c/p\u003e\u003cp\u003ePrevious studies have shown that there may be differences in the predictive power of biomarkers between males and females as well as \u003cem\u003eAPOE4\u003c/em\u003e carrier status and different ancestries. For this reason, we also performed sensitivity analyses stratified by sex and \u003cem\u003eAPOE\u003c/em\u003e, as well as to test the model across ancestries. To determine if the model was specific to AD or captures neurodegeneration in general, we tested the model in additional PD, FTD, and DLB samples.\u003c/p\u003e\u003cp\u003eLastly, replication was performed in the A4 dataset, an independent cohort. As only blood-RNA-seq was available for baseline samples from participants who were CU, we replicated the association of the circRNA model against biomarker-confirmed AD status and AD progression.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eIdentification of Blood circRNAs Associated with AD\u003c/h3\u003e\n\u003cp\u003eTo identify circRNAs associated with AD in blood, we analyzed the levels of circRNAs with clinical AD status in the AD cases (n\u0026thinsp;=\u0026thinsp;405) and CU (n\u0026thinsp;=\u0026thinsp;829) from the Knight ADRC. A total of 1,601 circRNAs passed stringent QC in both DCC\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e and CIRI2\u003csup\u003e19\u003c/sup\u003e. Of the 203 circRNAs with nominal significant association (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) with clinical status, using DCC counts, 35 circRNA transcripts passed False Discovery Rate (FDR) correction (\u003cb\u003eSupplementary Fig.\u0026nbsp;1\u003c/b\u003e), including \u003cem\u003ecircDNAJC6\u003c/em\u003e (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.88\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;08\u003c/sup\u003e), \u003cem\u003ecircMBOAT2\u003c/em\u003e (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3.80\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;04\u003c/sup\u003e), and \u003cem\u003ecircPICALM\u003c/em\u003e (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.23\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;03\u003c/sup\u003e; \u003cb\u003eSupplementary Table\u0026nbsp;3\u003c/b\u003e).\u003c/p\u003e\u003cp\u003eIn order to confirm our results were robust, we performed sensitivity analyses using CIRI2 counts (\u003cb\u003eSupplementary Table\u0026nbsp;4\u003c/b\u003e) on the 203 circRNA found to be associated with AD using DCC. The effect size of those 203 circRNA when using CIRI2 or DCC were highly correlated (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.78, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), as well as between the individual circRNA counts (\u003cb\u003eSupplementary Fig.\u0026nbsp;1\u003c/b\u003e). In addition, 34 of the 35 DCC circRNA remained FDR significant with consistent direction and high effect size correlation (\u003cb\u003eSupplementary Table\u0026nbsp;5\u003c/b\u003e). Thus, a total of 34 circRNA transcripts in blood passed FDR correction using high-quality circRNA read counts from two independent circRNA tools and all 34 were up-regulated in AD cases.\u003c/p\u003e\n\u003ch3\u003eAD Diagnostic Accuracy Using circRNAs in Blood\u003c/h3\u003e\n\u003cp\u003eWe analyzed the diagnostic accuracy of the 34 circRNAs associated with AD. To balance the number of AD cases and CU, we performed undersampling of the blood dataset over 100 iterations (\u003cb\u003eSupplementary Table\u0026nbsp;6\u003c/b\u003e). For clinical dementia status, based on clinical diagnosis, the 34 blood circRNAs showed an AUC (AUC\u0026thinsp;=\u0026thinsp;0.769; Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea, \u003cb\u003eSupplementary Table\u0026nbsp;7\u003c/b\u003e) higher than the baseline models that included age at blood draw, sex, median TIN, and number of \u003cem\u003eAPOE4\u003c/em\u003e alleles (AUC\u0026thinsp;=\u0026thinsp;0.634). The circRNA model was not significantly different to the plasma pTau217 alone (AUC\u0026thinsp;=\u0026thinsp;0.790). Moreover, combining the 34 circRNAs with plasma pTau217 (710 CU, 205 AD) led to the best predictive model with an AUC of 0.865, which is significantly better than those of circRNA or pTau217 alone (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eAs there have been several studies\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e indicating the risk of developing AD is different between males and females, sex-stratified analyses were performed. The circRNA model showed similar predictive power in both sexes (female AUC\u0026thinsp;=\u0026thinsp;0.778, male AUC\u0026thinsp;=\u0026thinsp;0.736; \u003cb\u003eSupplementary Tables\u0026nbsp;8\u0026ndash;9\u003c/b\u003e). Next, we stratified by \u003cem\u003eAPOE\u003c/em\u003e as \u003cem\u003eAPOE4\u003c/em\u003e is the strongest genetic risk factor of AD\u003csup\u003e3\u003c/sup\u003e. The \u003cem\u003eAPOE\u003c/em\u003e-stratified analyses of clinical status showed similar predictive ability with 0.724 and 0.693 for \u003cem\u003eAPOE4\u003c/em\u003e\u003csup\u003e+\u003c/sup\u003e and \u003cem\u003eAPOE4\u003c/em\u003e\u003csup\u003e\u0026minus;\u003c/sup\u003e respectively (\u003cb\u003eSupplementary Tables\u0026nbsp;10\u0026ndash;11\u003c/b\u003e).\u003c/p\u003e\u003cp\u003eAs CU individuals may present with AD pathology, the most recent biomarker studies have focused on identifying biomarkers to distinguish amyloid positive (A+) from those who are amyloid negative (A-). Therefore, we next examined the circRNA predictive ability to identify individuals with brain amyloidosis based on Amyloid-PET (520 A-, 256 A+). The same circRNA model (same circRNAs, weights and cut off) showed an AUC of 0.733 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb), improved up to AUC 0.868 when integrating with pTau217 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed).\u003c/p\u003e\u003cp\u003eAs amyloid PET only captures one of the main AD pathologies, we also tested if the circRNA model could distinguish between CU who are biomarker negative (AT-) and AD cases who are biomarker positive (AT+) based on CSF Aβ42 and ptau181. The 34 blood circRNAs were able to differentiate CSF amyloid positivity (AUC\u0026thinsp;=\u0026thinsp;0.739) based on Aβ42 (408 A-, 309 A+), as well as those that were T\u0026thinsp;+\u0026thinsp;vs T- based on CSF pTau181 levels (AUC\u0026thinsp;=\u0026thinsp;0.693) or tau-PET (292 T-, 233 T+; AUC\u0026thinsp;=\u0026thinsp;0.641). The circRNA model showed an AUC (AUC\u0026thinsp;=\u0026thinsp;0.846) for AT- vs AT\u0026thinsp;+\u0026thinsp;biomarker-confirmed samples, which was comparable to that of plasma pTau217 (AUC\u0026thinsp;=\u0026thinsp;0.877; Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec, \u003cb\u003eSupplementary Fig.\u0026nbsp;2\u0026ndash;3\u003c/b\u003e, \u003cb\u003eSupplementary Table\u0026nbsp;12\u003c/b\u003e). Furthermore, combining the circRNA model with pTau217 model led to an AUC of 0.937. Likewise, the sex-stratified analyses using AT status showed robust AUC for females (AUC\u0026thinsp;=\u0026thinsp;0.814) and males (AUC\u0026thinsp;=\u0026thinsp;0.887). Moreover, the circRNA models showed an AUC of 0.762 in \u003cem\u003eAPOE4\u003c/em\u003e\u003csup\u003e\u0026minus;\u003c/sup\u003e and 0.812 in \u003cem\u003eAPOE4\u003c/em\u003e\u003csup\u003e+\u003c/sup\u003e. Thus, the 34 blood circRNAs had higher predictive ability using biomarker-confirmed status compared to clinical status alone and amyloid biomarker positivity in brain (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed).\u003c/p\u003e\n\u003ch3\u003eThe blood circRNA predictive model is robust across ancestries\u003c/h3\u003e\n\u003cp\u003eTo determine if the circRNA model can also be applied to samples from diverse genetic backgrounds, we tested the same model in European (EUR; n\u0026thinsp;=\u0026thinsp;900) and African (AFR; n\u0026thinsp;=\u0026thinsp;92) individuals and an additional 35 from diverse backgrounds (African American (AFAM; n\u0026thinsp;=\u0026thinsp;9) Admixed American (AMR; n\u0026thinsp;=\u0026thinsp;8), East Asian (EAS; n\u0026thinsp;=\u0026thinsp;1), South Asian (SAS; n\u0026thinsp;=\u0026thinsp;1), and multiple ethnicities (n\u0026thinsp;=\u0026thinsp;9)) based on genetic information. The model run in clinical cases of genetically-defined EUR-background (637 CU, 341 AD) showed similar predictive ability (AUC\u0026thinsp;=\u0026thinsp;0.748) to the model across ancestries (AUC\u0026thinsp;=\u0026thinsp;0.769) and clinical cases of genetically-defined AFR-background (AUC\u0026thinsp;=\u0026thinsp;0.765; Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea, \u003cb\u003eSupplementary Table\u0026nbsp;13\u003c/b\u003e). As the numbers of individuals in each of the other ancestries were small, we combined all non-EUR individuals (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;127). This circRNA model showed an AUC of 0.829 (\u003cb\u003eSupplementary Table\u0026nbsp;14\u003c/b\u003e). The model was also comparable across ancestries in biomarker-confirmed cases (AUC\u0026thinsp;=\u0026thinsp;0.850), in which the EUR-background (AUC\u0026thinsp;=\u0026thinsp;0.848) showed similar predictive ability to the AFR-background (0.816). Altogether, the circRNA model showed robust AUC across ancestries.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eThe blood circRNA model is AD-specific\u003c/h3\u003e\n\u003cp\u003eTo determine whether the 34 circRNA model is specific to AD, we also tested the same model in non-AD dementias (276 PD, 26 DLB, and 11 FTD) using the same cognitive unimpaired participants (816 CU). The 34 circRNA model showed very low AUC when applied to non-AD dementias (PD AUC\u0026thinsp;=\u0026thinsp;0.440, PD\u0026thinsp;+\u0026thinsp;DLB\u0026thinsp;+\u0026thinsp;FTD AUC\u0026thinsp;=\u0026thinsp;0.450, PD\u0026thinsp;+\u0026thinsp;DLB AUC\u0026thinsp;=\u0026thinsp;0.449, PD\u0026thinsp;+\u0026thinsp;FTD AUC\u0026thinsp;=\u0026thinsp;0.432; Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb, \u003cb\u003eSupplementary Fig.\u0026nbsp;4\u003c/b\u003e).\u003c/p\u003e\u003cp\u003eWe further examined if there was any circRNA that was nominally significant in all neurodegenerative diseases (AD, PD, DLB, and FTD) when compared to controls (\u003cb\u003eSupplementary Fig.\u0026nbsp;5\u003c/b\u003e). There were 431 circRNAs that showed an association with either AD or PD, but their effect size correlation was relatively low (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.40). Of these 431 shared circRNAs, 16 circRNA transcripts (15 genes), such as \u003cem\u003ecircDNAJC6\u003c/em\u003e, \u003cem\u003ecircPICALM\u003c/em\u003e, and \u003cem\u003ecircMBOAT2\u003c/em\u003e, shared between AD and PD. All 16 of these circRNA were up-regulated in both AD and PD. When comparing across all diseases there were three (\u003cem\u003ecircRBM23\u003c/em\u003e, \u003cem\u003ecircEPB41\u003c/em\u003e, and \u003cem\u003ecircNUP54\u003c/em\u003e) of the 34 circRNAs that are part of the predictive models that where at least nominal associated with all of them. These results suggest the most of the circRNA included on the predictive model are AD-specific which could explain the low predictive power in other diseases.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eProgression to Symptomatic AD\u003c/h2\u003e\u003cp\u003eAs clinical data was available showing that several individuals included 78 participants that progressed from CU to symptomatic AD after blood collection, we performed survival analyses to determine if the 34 blood circRNA could also predict progression to symptomatic AD (\u003cb\u003eSupplementary Table\u0026nbsp;15\u003c/b\u003e). Using a cox regression model, the circRNA model showed a hazard ratio (HR) of 2.92 (95% CI: 1.63\u0026ndash;5.23), which was significantly higher of that of pTau217 alone (HR\u0026thinsp;=\u0026thinsp;1.81, 95% CI: 1.11\u0026ndash;2.94; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.002; Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eSimilar results were found when analyzed progression to symptomatic AD within five years, with the circRNA model leading to an AUC of 0.870 which was significantly higher (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.86\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;05\u003c/sup\u003e) than the one for pTau217 alone (AUC\u0026thinsp;=\u0026thinsp;0.676; Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb, \u003cb\u003eSupplementary Fig.\u0026nbsp;6\u003c/b\u003e). The sex and \u003cem\u003eAPOE\u003c/em\u003e stratified analyses also showed high AUCs (AUC\u0026thinsp;\u0026gt;\u0026thinsp;0.80) that were significantly higher than pTau217 (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Thus, the 34 blood circRNAs predicted AD progression better than the blood biomarker pTau217.\u003c/p\u003e\u003cp\u003eNext, we analyzed whether combining pTau217 with the circRNA model can further improve the identification of individuals who progress to symptomatic AD. For these analyses, we compared the individuals who were negative for both biomarkers (pTau217 and the 34 circRNAs; \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;411) to those who were positive for both (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;77), those who were positive for pTau217 but negative for the circRNA model (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;146), and the opposite pattern (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;54; Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). Our analyses indicated that individuals positive for both biomarkers indeed progressed faster than any of the previous models (HR\u0026thinsp;=\u0026thinsp;4.83, 95% CI: 2.19\u0026ndash;10.68). In general, only 15% of individuals negative for both progressed to AD, compared to 84% who progressed to AD and were positive for both. In addition, these analyses identified two intermediate groups: one with medium-low risk, defined by those positive for pTau217 but negative for circRNA (HR\u0026thinsp;=\u0026thinsp;3.25, 95% CI: 1.36\u0026ndash;7.76) from which 41% progress to AD within 5 years; and one with a medium-high risk, from which 74% progress to AD (HR\u0026thinsp;=\u0026thinsp;3.96, 95% CI: 1.50-10.48; Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb).\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eDetermining the time of circRNA change in relationship to the disease\u003c/h3\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003ePseudo-trajectories of samples progressing to symptomatic AD\u003c/h2\u003e\u003cp\u003eWe used survival modeling to calculate the time to onset (TTO) of the AD progressors in order to infer when the overall 34 circRNA model changes in relation to clinical onset. TTO was calculated by subtracting age at onset (AAO) from age at blood draw, in which, for example, TTO of -6 refers to a sample\u0026rsquo;s blood draw being 6 years before clinical AD onset. Wee created bins based on TTO with two-year intervals, starting from \u0026ldquo;-8 to -6\u0026rdquo; (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;5), \u0026ldquo;-6 to -4\u0026rdquo; (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;20), \u0026ldquo;-4 to -2\u0026rdquo; (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;33), and \u0026ldquo;-2 to 0\u0026rdquo; years (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;18). Compared with biomarker negative CU (CU AT-) individuals that do not progress to AD and the different TTO bin, the circRNAs had significant changes starting at the \u0026ldquo;-4 to -2\u0026rdquo; TTO range and continuing closer to onset (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec). Moreover, TTO subsets closer to onset showed higher overall circRNA values compared to TTO subsets further away from onset. Altogether, these observations suggest the there is a linear and consistent increase of the overall circRNA levels that starts in the presymptomatic phase around 4 to 2 years before onset an continue increasing until symptomatic AD.\u003c/p\u003e\u003cp\u003eWe next examined the association of the overall circRNA model with predicting progression of dementia severity based on CDR\u0026reg;\u003csup\u003e22\u003c/sup\u003e (Clinical Dementia Rating). Within the samples with no cognitive impairment at time of blood draw, the blood circRNAs showed an AUC of 0.781 in differentiating between samples from participants who did not progress (CDR\u0026thinsp;=\u0026thinsp;0; \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;737) and from participants who progressed to cognitive impairment (CDR\u0026thinsp;\u0026gt;\u0026thinsp;0; \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;42) by the last clinical visit (\u003cb\u003eSupplementary Fig.\u0026nbsp;7\u003c/b\u003e). Altogether, overall level of the 34 circRNAs were associated with dementia severity.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eChanges of circRNA levels in relation to cognitive changes\u003c/h2\u003e\u003cp\u003eThe previous analyses using longitudinal clinical data suggest that the circRNA model captures AD changes around 5 years before onset. In order to examine this further, we analyzed how the circRNA model changes across the spectrum of memory decline using cross-sectional data. In these analyses, we examined the association of the blood circRNAs with dementia severity based on CDR\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Among the blood sample donors, 1,014 individuals had CDR data (CDR\u0026thinsp;=\u0026thinsp;0; \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;779, CDR\u0026thinsp;=\u0026thinsp;0.5; \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;115, CDR\u0026thinsp;=\u0026thinsp;1; \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;84, CDR\u0026thinsp;=\u0026thinsp;2; \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;33, and CDR\u0026thinsp;=\u0026thinsp;3; \u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3). Differentiating between participants with no cognitive impairment (CDR\u0026thinsp;=\u0026thinsp;0) and those in very early disease stage (CDR\u0026thinsp;=\u0026thinsp;0.5) had an AUC of 0.721 (\u003cb\u003eSupplementary Fig.\u0026nbsp;8\u003c/b\u003e). Thus, these results suggest blood circRNAs capture early disease stage changes with more significant differentiation from CDR 0 to 0.5 compared to later disease stages.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eReplication of blood circRNA prediction models\u003c/h2\u003e\u003cdiv id=\"Sec13\" class=\"Section3\"\u003e\u003ch2\u003eOrthogonal replication using CIRI2 counts\u003c/h2\u003e\u003cp\u003eAll the prediction models so far used the circRNA counts based on DCC\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. We next examined whether the DCC-based models had predictive ability using read counts from another independent bioinformatic tool: CIRI2\u003csup\u003e19\u003c/sup\u003e. Overall, using either circRNA tool had similar predictive ability and prediction of clinical AD status using CIRI2 (AUC\u0026thinsp;=\u0026thinsp;0.720) was comparable to the DCC-based model (AUC\u0026thinsp;=\u0026thinsp;0.769; \u003cb\u003eSupplementary Fig.\u0026nbsp;9\u003c/b\u003e). For biomarker-confirmed status analyses, the performance of the model using CIRI2 (AUC\u0026thinsp;=\u0026thinsp;0.796), using same weights and cut-off, and the DCC counts (AUC\u0026thinsp;=\u0026thinsp;0.850) were also similar. Likewise, prediction of progression to symptomatic AD using CIRI2 (AUC\u0026thinsp;=\u0026thinsp;0.752) was not different from the DCC counts (AUC\u0026thinsp;=\u0026thinsp;0.870). Following progression prediction, survival analyses were also similar between CIRI2 (HR\u0026thinsp;=\u0026thinsp;2.63) and DCC (HR\u0026thinsp;=\u0026thinsp;2.92; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.007). Altogether, using normalized read counts from either circRNA quantification tool did not significantly change the predictive power of the 34 circRNA model, and retraining or cut-off calculation was not needed.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eIndependent replication in the A4 dataset\u003c/h2\u003e\u003cp\u003eTo replicate the blood circRNA model in an independent AD cohort, we mined the existing RNA-seq data derived from whole blood RNA samples from the A4 dataset (\u003cb\u003eSupplementary Table\u0026nbsp;16\u0026ndash;17\u003c/b\u003e). Other than amyloid-PET (n\u0026thinsp;=\u0026thinsp;1,767), the A4 study included plasma pTau217 (n\u0026thinsp;=\u0026thinsp;680) and longitudinal clinical data. All but one participant was recruited as CU. Progression to symptomatic AD was defined based on changes of CDR score between the time of blood draw and last visit. Of the participants that had plasma pTau217 at baseline, 97 progressed to symptomatic AD within 5 years.\u003c/p\u003e\u003cp\u003eThe circRNA model showed robust prediction of amyloid-PET positivity (457 A+, 1,310 A-) using either circRNA counts from DCC (AUC\u0026thinsp;=\u0026thinsp;0.692) and CIRI2 (AUC\u0026thinsp;=\u0026thinsp;0.664; \u003cb\u003eSupplementary Fig.\u0026nbsp;10\u0026ndash;11\u003c/b\u003e). As for the Knight ADRC dataset, the circRNA model showed even higher AUC (AUC\u0026thinsp;=\u0026thinsp;0.715) for biomarker status (A-T- vs A\u0026thinsp;+\u0026thinsp;T).\u003c/p\u003e\u003cp\u003eMoreover, the 34 circRNA model also replicated for progression to symptomatic AD showing an HR of 2.93, which was significantly higher than the pTau217 model (HR: 1.87). No significant differences were found when using DCC or CIRI2 counts (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea-b). We also found that combining the circRNA model with pTau217 led to better prediction power for both amyloid PET status and especially for progression. Altogether, the prediction model created using the Knight ADRC cohort replicated in the A4 dataset for brain amyloidosis and progression to symptomatic AD models, highlighting model generalizability and applicability to AD diagnosis in blood.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eCircular RNAs have become potential novel candidates for predicting disease status due to their high stability, enrichment, and specificity across tissues\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Identifying circRNAs with differential expression in blood allows for clinically relevant prediction models to be created for pinpointing possible AD biomarkers. Here, we identified 34 blood circRNA transcripts associated with clinical and biomarker-confirmed status, and, more importantly, associated with progression from cognitively unimpaired to symptomatic AD. The robust replications in the A4 clinical dataset confirm the robustness of the circRNA models that are also significantly higher that plasma pTau217 when predicting progression to symptomatic AD, establishing blood circRNAs as a novel and potential non-invasive, high precision tool for early AD diagnosis and progression monitoring in clinical practice.\u003c/p\u003e\u003cp\u003ePrior circRNAs studies have been largely focused in the brain\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e due to their high abundance and synapse enrichment. While these studies have spanned multiple AD-relevant brain regions relevant to AD including the parietal cortex\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e and hippocampus\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e, the brain is not readily accessible. Comparatively, studies on circRNAs in whole blood from AD participants are sparse but promising\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. A recent study\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e indicates that circRNAs could potentially be used to predict clinical status confirmed by CSF biomarkers (pTau181/Aβ42), even though the study was performed in a small dataset (20 AD, 20 CU). Yet, such discriminatory power in clinical AD status may not fully reflect circRNAs ability to detect AD at detection of earlier stages and presymptomatic stages AD and additional studies in larger studies are needed.\u003c/p\u003e\u003cp\u003eRecent biomarker development studies have shifted towards assessing amyloid accumulation rather than clinical status to identify asymptomatic individuals with underlying AD pathology. Plasma pTau217 is widely regarded as the leading plasma-based biomarker of AD in preclinical stages\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e and has been shown to be, indeed, a marker of amyloid pathology that changes 15\u0026ndash;20 years prior to clinical onset. However, pTau217 can be unreliable as a biomarkers in patients treated with anti-amyloid antibody therapies, as it may normalize with brain Aβ plaques removal without corresponding cognitive improvements\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. Biomarkers independent of Aβ and tau pathology are essential to monitor overall neurodegeration in this therapeutic area. Our 34 blood circRNAs appear to capture overall AD disease biology and not just pathology. Interestingly, the host genes of some of the circRNA part of the model such as \u003cem\u003ecircDNAJC6\u003c/em\u003e is involved in synaptic function, while \u003cem\u003eAPP\u003c/em\u003e (amyloid precursor protein) is regulated by \u003cem\u003ecircCTCF\u003c/em\u003e and \u003cem\u003ecircRANBP9\u003c/em\u003e host genes. On top of this, two of our 34 circRNA biomarkers predictive of AD are derived from the gene \u003cem\u003ePICALM\u003c/em\u003e, which is a GWAS-significant AD risk gene associated with both Aβ, tau, \u003cem\u003eAPOE4\u003c/em\u003e function and synaptic dysfunction and microglia-derived neuroinflammation. Thus, when we evaluated the accuracy of circRNA for biomarker-defined status, the model generated a robust AUC of 0.850 in biomarker confirmed (AT+) patients, and 0.733 for amyloid-PET positivity, demonstrating strong discriminatory power comparable to the current standard of care modalities.\u003c/p\u003e\u003cp\u003eIn addition to direct comparison with CSF and brain imaging biomarkers, we benchmarked our model against the AD blood biomarker: pTau217. Plasma pTau217 changes 15\u0026ndash;20 years prior to clinical onset and models with better prediction of AD onset are required to determine progression to symptomatic AD\u003csup\u003e24\u003c/sup\u003e. Notably, the blood circRNA model had higher predictive ability of AD progression in 5 years (AUC\u0026thinsp;=\u0026thinsp;0.870) compared to plasma pTau217 alone (AUC\u0026thinsp;=\u0026thinsp;0.676). These findings highlight circRNA\u0026rsquo;s ability to capture dynamic progression signals which other pathology-focused biomarkers may miss. Prediction of progression to symptomatic AD was even further improved after combining the pTau217 and circRNA models. This combined model was able to distinguish non-progressors (negative for both biomarkers) from high progressors (both positive), with 84% of the positive group progressing to AD compared to 15% of the non-progressors.\u003c/p\u003e\u003cp\u003eGiven that plasma pTau217 predominantly tracks amyloid pathology, there is a pressing need for more biomarkers which extend beyond amyloidosis to detect and separate progression of cognitive impairment from decreased pathology due to anti-amyloid antibody treatments. Our results strongly support the use of blood-based circRNAs as non-invasive, scalable biomarkers for detecting overall AD pathology, with robust performance that surpasses gold standards in predicting progression to symptomatic AD. Beyond the ability to act as diagnostic biomarkers, the blood circRNAs predicted progression to symptomatic AD in CU individuals at the time of blood collection. In fact, progression analyses further reveal that circRNA levels changes between 4\u0026ndash;8 years prior to disease onset, closer to symptom emergence. These results establish the 34 identified circRNAs as a novel and comprehensive approach for accurate AD diagnosis, with superior progression prediction and AD specificity against current gold standards like PET and pTau217, ultimately advancing early detection and monitoring of AD.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003cdiv id=\"Sec17\" class=\"Section3\"\u003e\u003ch2\u003eDatasets\u003c/h2\u003e\u003cp\u003eThe Knight Alzheimer Disease Research Center (Knight ADRC) and the PD MARS study at Washington University in St. Louis were used in the analyses. The Knight ADRC is supported by the NIH (National Institutes of Health) and conducts prospective studies on memory and aging for the treatment and prevention of AD. Study eligibility includes participants with asymptomatic or mild dementia at enrollment and the age of 45 or older. The MAP study involves the longitudinal collection of biofluids (plasma, cerebrospinal fluid), neuroimaging, and annual clinical assessments. This study utilized bulk blood RNA-seq data comprising 3,670 blood samples from 2,573 individuals. The Knight ADRC study included 405 AD cases, 816 CU (cognitively unimpaired), and the AD-related dementias (ADRD) including 1 PD (Parkinson\u0026rsquo;s disease) case, 17 DLB (dementia with Lewy body) cases, and 11 FTD (frontotemporal dementia) cases.. The PD MARS study, also supported by the NIH, supplied data and samples for the remaining 275 PD cases and 9 DLB cases included in this study. The mean age at blood draw were similar between individuals with AD and CU individuals with the average ages of 76.8 and 74.5 respectively.\u003c/p\u003e\u003cp\u003eThe Anti-Amyloid Treatment in Asymptomatic Alzheimer Disease (A4) study included individuals with evidence of amyloid accumulation based on amyloid-PET. The A4 study is a clinical trial to examine if anti-amyloid treatment slows down the rate of cognitive decline. The study enrolled CU individuals and longitudinally evaluated cognition based on neuropsychological tests. The A4 study comprises longitudinal measurements of amyloid-PET and plasma pTau217. Blood RNA-seq was available for CU samples at baseline and included 1,797 samples from 1,797 individuals. This cohort included a subset of the A4 clinical trial cohort of amyloid-PET positive asymptomatic individuals that were followed up tot 5 years for conversion to dementia/symptomatic AD, defined based on changes of CDR (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;190), a subset of the LEARN amyloid-PET negative cohort followed up for the same timeframe (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;128), as well as additional patients initially screened and not selected in the above cohorts, a subset of which had longitudinal data as a result of extension studies (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;358) followed up to 7 years. This cohort also included baseline blood draw compared to last visit CU (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;469), progressors (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;97) with available pTau217 data for comparison. Within the totality of this cohort \u003cem\u003e(n\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1,767), the 1,797 CU individuals included 1,796 with CDR of 0 at blood draw and 1 individual with CDR of 0.5. Between these samples, the average age at blood draw of individuals with no cognitive impairment was 71.4 and 82 for very early disease stage.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003eRNA sequencing\u003c/h2\u003e\u003cp\u003eThis study generated 151-nt, paired-end, rRNA and globin depleted, RNA-seq data from whole blood donated by participants in the Knight ADRC MAP study. All the participants consented to blood donation and clinical analysis. RNA was purified from whole blood using the Maxwell RSC simplyRNA blood kit. RNA samples were extracted from whole blood in Paxgene tubes and transferred using Biomek. Globin and rRNA depletion was performed using FastSelect, RIN was quantified by TapeStation 4200, and RNA was then sequenced on an Illumina NovaSeq 6000 at the MGI at Washington University in St. Louis.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003eAlzheimer disease traits\u003c/h2\u003e\u003cp\u003eThe AD phenotype examined included clinically determined status. Samples with neuropath AD status are defined based on Clinical Dementia Rating\u0026reg;\u003csup\u003e22\u003c/sup\u003e (CDR\u0026reg;) score greater than 0 and CERAD (Consortium to Establish a Registry for Alzheimer\u0026rsquo;s Disease)\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e categorizing the sample as AD. Cognitive unimpaired (CU) is defined as \u0026ldquo;low probability AD\u0026rdquo; or \u0026ldquo;not AD\u0026rdquo; neuropath status and samples having both a CDR\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e score of 0 and Braak\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e tau (neurofibrillary tangle) stage under III.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003ePhenotype processing\u003c/h2\u003e\u003cp\u003ePrincipal components analysis was performed to generate genetic ancestry covariates using PLINK\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e and genetic variant array data. Genetic PC1 and 2 were used to determine genetic ancestry to keep European samples. Related and duplicate samples were removed using cryptic relatedness through identity by descent (IBD; PIHAT\u0026thinsp;\u0026ge;\u0026thinsp;0.25) in PLINK. The median values of TIN (Transcript Integrity Number) were used to measure RNA quality. Compared to RIN (RNA Integrity Number) being calculated per sample, TIN is measured per transcript and was calculated using tin.py from the RSeQC package\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e with GENCODE v33 as reference. The blood cohort from the Knight ADRC included 3,656 blood samples. Samples with the covariates age at blood draw, sex, median TIN, batch, number of APOE4 alleles and AD clinical status were kept, with 1,221 individuals that passed QC.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec21\" class=\"Section2\"\u003e\u003ch2\u003eRNA detection and filtering\u003c/h2\u003e\u003cp\u003eCircular RNAs (circRNAs) were identified using DCC\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e and CIRI2\u003csup\u003e19\u003c/sup\u003e to reduce the false-positivity rate of circRNA prediction tools, defined by the fraction of RNase R sensitive circRNAs\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. Alignment of raw sequencing reads to the GRCh38 human reference genome assembly for DCC was performed using STAR v.2.7.8a\u003csup\u003e31\u003c/sup\u003e in chimeric alignment mode. The DCC count parameter was increased (-Nr 5 5) for the blood cohort to output circRNAs with a minimum of 5 counts in 5 samples and account for the computational burden of the larger cohort (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3,656). Alignment of reads for CIRI2 was performed using BWA-MEM\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e with the recommended minimum alignment score (-T) of 19. Linearly aligned and chimerically aligned reads were compared and backsplices with a minimum circ:linear read ratio of 0.1 in at least 3 samples were kept. CircRNA transcripts were used for downstream analyses if they contained at least 1 read count in at least 25% of the samples in both DCC and CIRI2.\u003c/p\u003e\u003cp\u003eThe blood samples from the Knight ADRC (\u003cem\u003en\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1,221) were used to identify blood circRNA, in which 2,069 DCC circRNA and 1,871 CIRI2 circRNA passed count filtering. The blood cohort had 1,601 circRNA identified by both DCC and CIRI2. Transcript-level read counts were normalized for each cohort based on library size and sequencing depth using variance stabilizing transformation (VST) from the DESeq2\u003csup\u003e35\u003c/sup\u003e R package.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec22\" class=\"Section2\"\u003e\u003ch2\u003eLinear transcript identification\u003c/h2\u003e\u003cp\u003eThe linear RNA-Seq pipeline included raw sequence data quality control (QC) checks using FastQC\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e, STAR\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e alignment, Picard (Broad Institute) summary statistics, and transcript quantification using Salmon\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. Picard 2.27.4 FastqToSam converted the fastqc files to unmapped.bam and Picard RevertSam then cleared alignment information attributes in the bam file to allow STAR to process the file as paired-end. STAR\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e version 2.7.8a was used to align paired-end reads with the GRCH38 genome index. After using samtools sort, Picard Collect-RNA-Seq-Metrics, Collect-Alignment-Summary-Metrics, and Mark-Duplicates was run on the sorted STAR output file. Salmon was used to quantify the linear genes in quasi-mapping-based alignment mode and genes with less than 10 counts in less than 90% of individuals were filtered out. Lastly, linear RNA counts were normalized per cohort based on library size and sequencing depth using VST from the DESeq2\u003csup\u003e35\u003c/sup\u003e R package.\u003c/p\u003e\u003cdiv id=\"Sec23\" class=\"Section3\"\u003e\u003ch2\u003eDifferential expression and correlation analyses\u003c/h2\u003e\u003cp\u003eDifferential expression (DE) analysis was performed on the cross-sectional blood samples and was adjusted by age at blood draw, sex, batch, and median TIN. Negative binominal family logistic regression of AD vs CU (cognitive unimpaired) status was performed using DESeq2\u003csup\u003e35\u003c/sup\u003e and the DE results were FDR corrected with a 0.05 threshold. The model design for blood was circRNA counts\u0026thinsp;~\u0026thinsp;age\u0026thinsp;+\u0026thinsp;sex\u0026thinsp;+\u0026thinsp;batch\u0026thinsp;+\u0026thinsp;median TIN\u0026thinsp;+\u0026thinsp;case-CU status.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec24\" class=\"Section2\"\u003e\u003ch2\u003eDiagnostic models\u003c/h2\u003e\u003cp\u003eLogistic regression models of the top differentially expressed circRNAs were created using blood circRNAs, including age at blood draw, sex, and number of \u003cem\u003eAPOE4\u003c/em\u003e alleles as covariates. Samples with covariates (age, sex, \u003cem\u003eAPOE4\u003c/em\u003e) and AD phenotype information were included in the models. Diagnostic utility of the models were analyzed by ROC curve and AUC analysis using the R packages pROC\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e and ROCR\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. The dependent variable for all models was diagnosis and had AD cases for positive events and cognitive unimpaired for negative events. Prediction models were performed with 100 iterations of undersampling to balance the sample sizes of the cases and CU per phenotype. The weights and cutoff of the AD vs CU status model were determined using the glm function and model training was performed using the predict function. Youden\u0026rsquo;s index from pROC\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e was used to optimize and derive the threshold for calculating metrics such as accuracy, sensitivity, and specificity. These weights and cut-off were then used when testing the model performance against Amyloid-PET, CSF biomarker or AT status. Due to the different statistical framework, progression to symptomatic AD used the weights obtained from the Cox regression model (see below) Undersampling was performed for the full model and was not performed in the subsequent stratification by sex and \u003cem\u003eAPOE4\u003c/em\u003e status. For the sex or \u003cem\u003eAPOE4\u003c/em\u003e-stratified analysis, the same weights and cutoff as the full model was tested in individuals were used. The overall circRNA counts for A4 were lower compared to that of the Knight ADRC and a lower threshold cutoff was utilized in A4 progression analyses (Supplementary Fig.\u0026nbsp;11).\u003c/p\u003e\u003cp\u003ePrediction probability of the circRNA model was calculated using the weights and cutoff of the AD vs CU clinical status model. The progression to symptomatic AD model used separate probability values and cutoff derived from the Cox regression model. Biomarker negative samples were determined by the prediction probability value being less than the progression cutoff and biomarker positive samples had a probability value greater than the progression cutoff.\u003c/p\u003e\u003cp\u003eTo infer the ancestry specificity of the blood circRNA models, we performed the AD prediction models using population subsets with the 1000 Genomes Project as a reference. Of the blood samples with genetic information, the PCA identified 990 EUR, 92 AFR, 16 AfAM, 9 Admixture, 8 AMR, 1 EAS, and 1 SAS sample.\u003c/p\u003e\u003cdiv id=\"Sec25\" class=\"Section3\"\u003e\u003ch2\u003eAD progression analyses\u003c/h2\u003e\u003cp\u003eFor the progression to symptomatic AD, Cox proportional hazard was performed using the R package survival and Kaplan Meier (KM) curves were plotted via the R package survminer. The Cox proportional hazard model included age at draw and sex as the covariates and circRNA counts per circRNA. The status variable included 610 samples from participants who remained cognitive unimpaired at last visit (coded as binary variable 1 in the model) and 68 samples from participants who converted to AD by last visit (variable 2 in model). The time variable was calculated by subtracting the age at last visit from age at blood draw, ranging from 0 to 5 years.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eCompeting interests\u003c/h2\u003e\u003cp\u003eCC has received research support from GSK and EISAI. The funders of the study had no role in the collection, analysis, or interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. CC is a member of the advisory board of Vivid Genomics and Circular Genomics and owns stocks in these companies. CC and BP have invention disclosures for the circRNA models for AD diagnosis.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e\u003cp\u003eWe would like to thank the study participants.\u003c/p\u003e\u003cp\u003eThis work was supported by grants from the National Institutes of Health (R01AG044546 (CC), P01AG003991 (CC), RF1AG053303 (CC), RF1AG058501 (CC), U01AG058922 (CC), the Chan Zuckerberg Initiative (CC), the Michael J. Fox Foundation (CC), and the Alzheimer\u0026rsquo;s Association Zenith Fellows Award (ZEN-22-848604, awarded to CC). The PD MARS study sample collection was supported by the American Parkinson disease association, the Barnes-Jewish hospital foundation including the Elliot Stein family fund, the Paula and Rodger Riney fund, and the Jansky-Bander fund and NIH (NINDS, NIA) NS075321. The RNA-seq in A4 work was supported by the DP2 AG082342 (RB) and R01 AG079142 (RB).\u003c/p\u003e\u003cp\u003eThis work was supported by access to equipment made possible by the Hope Center for Neurological Disorders, the NeuroGenomics and Informatics Center (NGI: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://neurogenomics.wustl.edu/\u003c/span\u003e\u003cspan address=\"https://neurogenomics.wustl.edu/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and the Departments of Neurology and Psychiatry at Washington University School of Medicine.\u003c/p\u003e\u003cp\u003e The recruitment and clinical characterization of research participants at Washington University were supported by NIH P30AG066444 (DMH), P01AG03991 (JCM), and P01AG026276 (JCM).\u003c/p\u003e\u003ch2\u003eData availability\u003c/h2\u003e\u003cp\u003eKnight ADRC sequencing data is available to approved investigators through \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://knightadrc.wustl.edu/data-request-form/\u003c/span\u003e\u003cspan address=\"https://knightadrc.wustl.edu/data-request-form/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. A4 sequencing data is available to approved investigators through \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://vmacdata.org/vmap/data-requests\u003c/span\u003e\u003cspan address=\"https://vmacdata.org/vmap/data-requests\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\u003ch2\u003eCode availability\u003c/h2\u003e\u003cp\u003ePublicly available software was used for all analyses. The software included PLINK\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e for IBD and genetic ancestry, STAR v.2.7.8a\u003csup\u003e31\u003c/sup\u003e and BWA-MEM\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e for alignment, DCC\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e and CIRI2\u003csup\u003e19\u003c/sup\u003e to identify circRNAs, and linear mRNA RNA-Seq pipeline tools (FastQC\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e, STAR\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e, Picard 2.27.4 (Broad Institute), Salmon\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e). Differential expression analysis was performed using the DESeq2\u003csup\u003e35\u003c/sup\u003e R package and diagnostic utility of logistic regression models were analyzed using the R packages pROC\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e and ROCR\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. Cox proportional hazard was performed using the survival R package and Kaplan Meier curves were plotted using the survminer R package.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eRybak-Wolf A, Stottmeister C, Glažar P, Jens M, Pino N, Giusti S, Hanan M, Behm M, Bartok O, Ashwal-Fluss R, Herzog M, Schreyer L, Papavasileiou P, Ivanov A, \u0026Ouml;hman M, Refojo D, Kadener S, Rajewsky N. Circular RNAs in the Mammalian Brain Are Highly Abundant, Conserved, and Dynamically Expressed. \u003cem\u003eMol Cell\u003c/em\u003e. 2015;58(5):870\u0026ndash;85. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.molcel.2015.03.027\u003c/span\u003e\u003cspan address=\"10.1016/j.molcel.2015.03.027\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJeck WR, Sorrentino JA, Wang K, Slevin MK, Burd CE, Liu J, Marzluff WF, Sharpless NE. Circular RNAs are abundant, conserved, and associated with ALU repeats. \u003cem\u003eRNA\u003c/em\u003e. 2013;19(2):141\u0026ndash;57. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1261/rna.035667.112\u003c/span\u003e\u003cspan address=\"10.1261/rna.035667.112\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eScheltens P, De Strooper B, Kivipelto M, Holstege H, Ch\u0026eacute;telat G, Teunissen CE, Cummings J, van der Flier WM. Alzheimer's disease. \u003cem\u003eLancet\u003c/em\u003e. 2021;397(10284):1577\u0026ndash;1590. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S0140-\u003c/span\u003e\u003cspan address=\"10.1016/S0140-\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e 6736(20)32205\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFlorean I, Penolazzi B, Menichelli A, Pastore M, Cattaruzza T, Mazzon G, Manganotti P. Using the ATN system as a guide for the neuropsychological assessment of Alzheimer's disease. \u003cem\u003eJ Clin Exp Neuropsychol\u003c/em\u003e. 2021;43(9):926\u0026ndash;943. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1080/13803395.2022.2036327\u003c/span\u003e\u003cspan address=\"10.1080/13803395.2022.2036327\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAshton NJ, Puig-Pijoan A, Mil\u0026agrave;-Alom\u0026agrave; M, Fern\u0026aacute;ndez-Lebrero A, Garc\u0026iacute;a-Escobar G, Gonz\u0026aacute;lez-Ortiz F, et al. Plasma and CSF biomarkers in a memory clinic: Head-to-head comparison of phosphorylated tau immunoassays. \u003cem\u003eAlzheimers Dement\u003c/em\u003e. 2023;19(5):1913\u0026ndash;1924. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/alz.12841\u003c/span\u003e\u003cspan address=\"10.1002/alz.12841\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJanelidze S, Bali D, Ashton NJ, Barth\u0026eacute;lemy NR, Vanbrabant J, Stoops E, Vanmechelen E, He Y, Dolado AO, Triana-Baltzer G, Pontecorvo MJ, Zetterberg H, Kolb H, Vandijck M, Blennow K, Bateman RJ, Hansson O. Head-to-head comparison of 10 plasma phospho-tau assays in prodromal Alzheimer's disease. \u003cem\u003eBrain\u003c/em\u003e. 2023;146(4):1592\u0026ndash;1601. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/brain/awac333\u003c/span\u003e\u003cspan address=\"10.1093/brain/awac333\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVillemagne VL, Burnham S, Bourgeat P, Brown B, Ellis KA, Salvado O, Szoeke C, Macaulay SL, Martins R, Maruff P, Ames D, Rowe CC, Masters CL; Australian Imaging Biomarkers and Lifestyle (AIBL) Research Group. Amyloid β deposition, neurodegeneration, and cognitive decline in sporadic Alzheimer's disease: a prospective cohort study. \u003cem\u003eLancet Neurol\u003c/em\u003e. 2013;12(4):357\u0026ndash;67. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S1474-4422(13)70044-9\u003c/span\u003e\u003cspan address=\"10.1016/S1474-4422(13)70044-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJoling KJ, Janssen O, Francke AL, Verheij RA, Lissenberg-Witte BI, Visser PJ, van Hout HPJ. Time from diagnosis to institutionalization and death in people with dementia. \u003cem\u003eAlzheimers Dement\u003c/em\u003e. 2020;16(4):662\u0026ndash;671. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/alz.12063\u003c/span\u003e\u003cspan address=\"10.1002/alz.12063\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCrowell V, Reyes A, Zhou SQ, Vassilaki M, Gsteiger S, Gustavsson A. Disease severity and mortality in Alzheimer's disease: an analysis using the U.S. National Alzheimer's Coordinating Center Uniform Data Set. \u003cem\u003eBMC Neurol\u003c/em\u003e. 2023;23(1):302. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12883-023-03353-w\u003c/span\u003e\u003cspan address=\"10.1186/s12883-023-03353-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHorie K, Salvad\u0026oacute; G, Koppisetti RK, Janelidze S, Barth\u0026eacute;lemy NR, He Y, Sato C, Gordon BA, Jiang H, Benzinger TLS, Stomrud E, Holtzman DM, Mattsson-Carlgren N, Morris JC, Palmqvist S, Ossenkoppele R, Schindler SE, Hansson O, Bateman RJ. Plasma MTBR-tau243 biomarker identifies tau tangle pathology in Alzheimer's disease. \u003cem\u003eNat Med\u003c/em\u003e. 2025;31(6):2044\u0026ndash;2053. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41591-025-03617-7\u003c/span\u003e\u003cspan address=\"10.1038/s41591-025-03617-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDube U, Del-Aguila JL, Li Z, Budde JP, Jiang S, Hsu S, Ibanez L, Fernandez MV, Farias F, Norton J, Gentsch J, Wang F; Dominantly Inherited Alzheimer Network (DIAN); Salloway S, Masters CL, Lee JH, Graff-Radford NR, Chhatwal JP, Bateman RJ, Morris JC, Karch CM, Harari O, Cruchaga C. An atlas of cortical circular RNA expression in Alzheimer disease brains demonstrates clinical and pathological associations. \u003cem\u003eNat Neurosci\u003c/em\u003e. 2019;22(11):1903\u0026ndash;1912. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41593-019-0501-5\u003c/span\u003e\u003cspan address=\"10.1038/s41593-019-0501-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePuri S, Hu J, Sun Z, Lin M, Stein TD, Farrer LA, Wolozin B, Zhang X. Identification of circRNAs linked to Alzheimer's disease and related dementias. \u003cem\u003eAlzheimers Dement\u003c/em\u003e. 2023. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/alz.12960\u003c/span\u003e\u003cspan address=\"10.1002/alz.12960\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRen Z, Chu C, Pang Y, Cai H, Jia L. A circular RNA blood panel that differentiates Alzheimer's disease from other dementia types. \u003cem\u003eBiomark Res\u003c/em\u003e. 2022;10(1):63. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40364-022-00405-0\u003c/span\u003e\u003cspan address=\"10.1186/s40364-022-00405-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang A, Zheng H, Wu Z, Chen M, Huang Y. Circular RNA-protein interactions: functions, mechanisms, and identification. Theranostics. 2020;10(8):3503\u0026ndash;3517. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7150/thno.42174\u003c/span\u003e\u003cspan address=\"10.7150/thno.42174\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu L, Chen X, Chen YH, Zhang K. Identification of Circular RNA hsa_Circ_0003391 in Peripheral Blood Is Potentially Associated With Alzheimer's Disease. \u003cem\u003eFront Aging Neurosci\u003c/em\u003e. 2020;12:601965. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fnagi.2020.601965\u003c/span\u003e\u003cspan address=\"10.3389/fnagi.2020.601965\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi Y, Lv Z, Zhang J, Ma Q, Li Q, Song L, Gong L, Zhu Y, Li X, Hao Y, Yang Y. Profiling of differentially expressed circular RNAs in peripheral blood mononuclear cells from Alzheimer's disease patients. \u003cem\u003eMetab Brain Dis\u003c/em\u003e. 2020;35(1):201\u0026ndash;213. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11011-019-00497-y\u003c/span\u003e\u003cspan address=\"10.1007/s11011-019-00497-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePiscopo P, Manzini V, Rivabene R, Crestini A, Le Pera L, Pizzi E, Veroni C, Talarico G, Peconi M, Castellano AE, D'Alessio C, Bruno G, Corbo M, Vanacore N, Lacorte E. A Plasma Circular RNA Profile Differentiates Subjects with Alzheimer's Disease and Mild Cognitive Impairment from Healthy Controls. \u003cem\u003eInt J Mol Sci\u003c/em\u003e. 2022;23(21):13232. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/ijms232113232\u003c/span\u003e\u003cspan address=\"10.3390/ijms232113232\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCheng J, Metge F, Dieterich C. Specific identification and quantification of circular RNAs from sequencing data. \u003cem\u003eBioinformatics\u003c/em\u003e. 2016;32(7):1094\u0026ndash;6. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/bioinformatics/btv656\u003c/span\u003e\u003cspan address=\"10.1093/bioinformatics/btv656\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGao Y, Zhang J, Zhao F. Circular RNA identification based on multiple seed matching. \u003cem\u003eBrief Bioinform\u003c/em\u003e. 2018;19(5):803\u0026ndash;810. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/bib/bbx014\u003c/span\u003e\u003cspan address=\"10.1093/bib/bbx014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBelloy ME, Andrews SJ, Le Guen Y, Cuccaro M, Farrer LA, Napolioni V, Greicius MD. APOE Genotype and Alzheimer Disease Risk Across Age, Sex, and Population Ancestry. \u003cem\u003eJAMA Neurol\u003c/em\u003e. 2023;80(12):1284\u0026ndash;1294. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jamaneurol.2023.3599\u003c/span\u003e\u003cspan address=\"10.1001/jamaneurol.2023.3599\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAggarwal NT, Mielke MM. Sex Differences in Alzheimer's Disease. \u003cem\u003eNeurol Clin\u003c/em\u003e. 2023;41(2):343\u0026ndash;358. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ncl.2023.01.001\u003c/span\u003e\u003cspan address=\"10.1016/j.ncl.2023.01.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMorris JC. The Clinical Dementia Rating (CDR): current version and scoring rules. \u003cem\u003eNeurology\u003c/em\u003e. 1993;43(11):2412\u0026ndash;4. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1212/wnl.43.11.2412-a\u003c/span\u003e\u003cspan address=\"10.1212/wnl.43.11.2412-a\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEgan MF, Kost J, Voss T, Mukai Y, Aisen PS, Cummings JL, Tariot PN, Vellas B, van Dyck CH, Boada M, Zhang Y, Li W, Furtek C, Mahoney E, Harper Mozley L, Mo Y, Sur C, Michelson D. Randomized Trial of Verubecestat for Prodromal Alzheimer's Disease. \u003cem\u003eN Engl J Med\u003c/em\u003e. 2019;380(15):1408\u0026ndash;1420. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1056/NEJMoa1812840\u003c/span\u003e\u003cspan address=\"10.1056/NEJMoa1812840\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePalmqvist S, Janelidze S, Quiroz YT, Zetterberg H, Lopera F, Stomrud E, Su Y, Chen Y, Serrano GE, Leuzy A, Mattsson-Carlgren N, Strandberg O, Smith R, Villegas A, Sepulveda-Falla D, Chai X, Proctor NK, Beach TG, Blennow K, Dage JL, Reiman EM, Hansson O. Discriminative Accuracy of Plasma Phospho-tau217 for Alzheimer Disease vs Other Neurodegenerative Disorders. \u003cem\u003eJAMA\u003c/em\u003e. 2020;324(8):772\u0026ndash;781. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jama.2020.12134\u003c/span\u003e\u003cspan address=\"10.1001/jama.2020.12134\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMirra SS, Heyman A, McKeel D, Sumi SM, Crain BJ, Brownlee LM, Vogel FS, Hughes JP, van Belle G, Berg L. The Consortium to Establish a Registry for Alzheimer's Disease (CERAD). Part II. Standardization of the neuropathologic assessment of Alzheimer's disease. \u003cem\u003eNeurology\u003c/em\u003e. 1991;41(4):479\u0026ndash;86. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1212/wnl.41.4.479\u003c/span\u003e\u003cspan address=\"10.1212/wnl.41.4.479\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBraak H, Braak E. Neuropathological stageing of Alzheimer-related changes. \u003cem\u003eActa Neuropathol\u003c/em\u003e. 1991;82(4):239\u0026ndash;59. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/BF00308809\u003c/span\u003e\u003cspan address=\"10.1007/BF00308809\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePurcell S, Neale B, Todd-Brown K, et al. PLINK: a toolset for whole-genome association and population-based linkage analysis. \u003cem\u003eAmerican Journal of Human Genetics\u003c/em\u003e. 2007;81(3):559\u0026ndash;75\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang L, Nie J, Sicotte H, Li Y, Eckel-Passow JE, Dasari S, Vedell PT, Barman P, Wang L, Weinshiboum R, Jen J, Huang H, Kohli M, Kocher JP. Measure transcript integrity using RNA-seq data. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e. 2016;17:58. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12859-016-0922-z\u003c/span\u003e\u003cspan address=\"10.1186/s12859-016-0922-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHansen TB, Ven\u0026oslash; MT, Damgaard CK, Kjems J. Comparison of circular RNA prediction tools. \u003cem\u003eNucleic Acids Res\u003c/em\u003e. 2016;44(6):e58. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/nar/gkv1458\u003c/span\u003e\u003cspan address=\"10.1093/nar/gkv1458\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHansen TB. Improved circRNA Identification by Combining Prediction Algorithms. \u003cem\u003eFront Cell Dev Biol\u003c/em\u003e. 2018;6:20. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fcell.2018.00020\u003c/span\u003e\u003cspan address=\"10.3389/fcell.2018.00020\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, Gingeras TR. STAR: ultrafast universal RNA-seq aligner. \u003cem\u003eBioinformatics\u003c/em\u003e. 2013;29(1):15\u0026ndash;21. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/bioinformatics/bts635\u003c/span\u003e\u003cspan address=\"10.1093/bioinformatics/bts635\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. 2013; arXiv:1303.3997v2.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAndrews S. FastQC: A Quality Control Tool for High Throughput Sequence Data [Online]. 2010. Available online at: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.bioinformatics.babraham.ac.uk/projects/fastqc/\u003c/span\u003e\u003cspan address=\"http://www.bioinformatics.babraham.ac.uk/projects/fastqc/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePatro R, Duggal G, Love MI, Irizarry RA, Kingsford C. Salmon provides fast and bias-aware quantification of transcript expression. \u003cem\u003eNat Methods\u003c/em\u003e. 2017;14(4):417\u0026ndash;419. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/nmeth.4197\u003c/span\u003e\u003cspan address=\"10.1038/nmeth.4197\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLove MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. \u003cem\u003eGenome Biol\u003c/em\u003e. 2014;15(12):550. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13059-014-0550-8\u003c/span\u003e\u003cspan address=\"10.1186/s13059-014-0550-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRobin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez JC, M\u0026uuml;ller M. pROC: an open- source package for R and S\u0026thinsp;+\u0026thinsp;to analyze and compare ROC curves. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e. 2011;12:77. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/1471-2105-12-77\u003c/span\u003e\u003cspan address=\"10.1186/1471-2105-12-77\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSing T, Sander O, Beerenwinkel N, Lengauer T. ROCR: visualizing classifier performance in R. \u003cem\u003eBioinformatics\u003c/em\u003e. 2005;21(20):3940\u0026ndash;1. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/bioinformatics/bti623\u003c/span\u003e\u003cspan address=\"10.1093/bioinformatics/bti623\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"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":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8023085/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8023085/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDetection of Alzheimer\u0026rsquo;s disease (AD) prior to the development of clinical symptoms is critical due to new treatments for symptomatic AD. Circular RNAs (circRNAs) are highly stable non-coding RNAs enriched in brain that can cross the blood-brain-barrier and are associated with AD. Using two large and independent whole-blood transcriptomic datasets, we identified 34 circRNAs associated with clinical AD status. A single-cutoff predictive model including these 34 circRNAs was comparable to plasma pTau217 in classifying AD based on clinical or AT status, that replicated in the independent dataset. The circRNAs (Hazard Ratio (HR): 2.92) outperformed pTau217 (HR: 1.81) and amyloid-PET when predicting progression to symptomatic AD. CircRNA levels began diverging 4\u0026ndash;8 years before symptom onset, enabling preclinical risk stratification. The circRNA model was specific for AD showing low predictive power for PD, DLB, and FTD. These results establish blood circRNAs as robust biomarkers in AD diagnosis and disease progression.\u003c/p\u003e","manuscriptTitle":"Blood-Based Circular RNAs Enable Early and Accurate Alzheimer’s Disease Diagnosis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-09 07:22:22","doi":"10.21203/rs.3.rs-8023085/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-medicine","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"nm","sideBox":"Learn more about [Nature Medicine](http://www.nature.com/nm/)","snPcode":"","submissionUrl":"","title":"Nature Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Research","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"976a515c-3980-4e11-8230-ecb96c50e3e2","owner":[],"postedDate":"December 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":58932019,"name":"Biological sciences/Neuroscience/Diseases of the nervous system/Alzheimer's disease"},{"id":58932020,"name":"Biological sciences/Biological techniques/Sequencing/RNA sequencing"},{"id":58932021,"name":"Health sciences/Biomarkers/Predictive markers"},{"id":58932022,"name":"Biological sciences/Computational biology and bioinformatics/Machine learning"}],"tags":[],"updatedAt":"2026-05-12T18:45:20+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-09 07:22:22","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8023085","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8023085","identity":"rs-8023085","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}