The Bio‐Hermes Study: Biomarker database developed to investigate blood‐based and digital biomarkers in community‐based, diverse populations clinically screened for Alzheimer's disease

Abstract INTRODUCTION Alzheimer's disease (AD) trial participants are often screened for eligibility by brain amyloid positron emission tomography/cerebrospinal fluid (PET/CSF), which is inefficient as many are not amyloid positive. Use of blood‐based biomarkers may reduce screen failures. METHODS We recruited 755 non‐Hispanic White, 115 Hispanic, 112 non‐Hispanic Black, and 19 other minority participants across groups of cognitively normal (n = 417), mild cognitive impairment (n = 312), or mild AD (n = 272) participants. Plasma amyloid beta (Aβ)40, Aβ42, Aβ42/Aβ40, total tau, phosphorylated tau (p‐tau)181, and p‐tau217 were measured; amyloid PET/CSF (n = 956) determined amyloid positivity. Clinical, blood biomarker, and ethnicity/race differences associated with amyloid status were evaluated. RESULTS Greater impairment, older age, and carrying an apolipoprotein E (apoE) ε4 allele were associated with greater amyloid burden. Areas under the receiver operating characteristic curve for amyloid status of plasma Aβ42/Aβ40, p‐tau181, and p‐tau217 with amyloid positivity were ≥ 0.7117 for all ethnoracial groups (p‐tau217, ≥0.8128). Age and apoE ε4 adjustments and imputation of biomarker values outside limit of quantitation provided small improvement in predictive power. DISCUSSION Blood‐based biomarkers are highly associated with amyloid PET/CSF results in diverse populations enrolled at clinical trial sites. Highlights Amyloid beta (Aβ)42/Aβ40, phosphorylated tau (p‐tau)181, and p‐tau 217 blood‐based biomarkers predicted brain amyloid positivity. P‐tau 217 was the strongest predictor of brain amyloid positivity. Biomarkers from diverse ethnic, racial, and clinical cohorts predicted brain amyloid positivity. Community‐based populations have similar Alzheimer's disease (AD) biomarker levels as other populations. A prescreen process with blood‐based assays may reduce the number of AD trial screen failures.

• P-tau 217 was the strongest predictor of brain amyloid positivity.
• Biomarkers from diverse ethnic, racial, and clinical cohorts predicted brain amyloid positivity.
• Community-based populations have similar Alzheimer's disease (AD) biomarker levels as other populations.
• A prescreen process with blood-based assays may reduce the number of AD trial screen failures.

BACKGROUND
2][3] A robust screening process that can identify patients with a high probability to randomize into AD therapeutic research trials would greatly enhance the ability to conduct and reduce the time needed to complete clinical trials.
AD is characterized by the accumulation of two protein aggregates in the brain: extracellular deposits of amyloid beta (Aβ)-containing plaques and intraneuronal aggregates of misfolded tau protein. 4Amyloid tracers are now available to detect the presence of amyloid plaques in the brain during a positron emission tomography (PET) scan. 5,6Additionally, it has been shown that Aβ levels in cerebrospinal fluid (CSF) can be a valid reflection of brain amyloid deposits. 7merous AD clinical trials, particularly those targeting either Aβ or amyloid plaques have used amyloid PET scans and/or CSF measures as an inclusion criterion for enrollment.While amyloid PET tracers have been shown to be very accurate in detecting brain amyloid deposits, these scans are costly and impose a significant patient burden.][10] Additionally, there is substantial interest in blood-based biomarkers reflecting two other critical aspects of AD pathology: tau tangles and neurodegeneration.The amyloid/tau/neurodegeneration (A/T/N) framework has been proposed as goal for developing biomarkers of each of these three aspects of AD pathology. 2veral clinical studies have been conducted evaluating the ability of various blood-based biomarkers to identify AD.These studies have identified Aβ40, Aβ42, the Aβ42/Aβ40 ratio (Aβ42/Aβ40), tau, and several species of phosphorylated tau (p-tau) as good candidates. 8,9,11,12ny of the studies investigating the clinical utility of these biomarkers have used clinical groups identified for other purposes, such as autopsy studies, 13,14 genetic studies, 15,16 or epidemiological studies. 17,18While informative, these studies may not adequately reflect the kinds of patients typically enrolled in therapeutic studies conducted at clinical trial sites and often do not have the racial and ethnic diversity char-acteristics of the US population. 19Data from the Centers for Disease Control (CDC) indicates that the prevalence of AD is higher among non-Hispanic Blacks and Hispanics than among non-Hispanic Whites in the United States 20 ; at the same time, data from the IDEAS study 21 indicate that the rate of amyloid-positive PET scans is lower among non-Hispanic Blacks and Hispanics than among non-Hispanic Whites in the United States.Thus, any evaluation of biomarkers should assess whether there are differences by race and ethnicity.The study also aimed to include a significant representation of racial and ethnic minority participants.While blood samples and digital data were obtained to allow assessment of many biomarkers, this current report presents data on the most used blood biomarkers, Aβ40, Aβ42, Aβ42/Aβ40, total tau (t-tau), p-tau181, and p-tau217.
For the great majority of participants, brain amyloid was assessed with amyloid PET; at one site where PET was not available, CSF was analyzed.

Screening procedures and defined cohorts
Participants who met eligibility criteria were further identified as belonging to one of the three clinical cohorts representing the range of cognitive function typically included in therapeutic or prevention tri-  24 ; and in the investigator's judgment, no evidence of functional decline based on the Functional Activities Questionnaire (FAQ) score/study partner report (the equivalent of a score of 0-8). 25tential participants with exceptionally good memory scores on the RAVLT (> 1.5 standard deviation [SD] above age-and race-adjusted mean) were excluded from the study because of the very low likelihood of amyloid positivity in this group.A similar screening procedure was used in the A4 study 3 because of the great expense involved in completing brain amyloid PET scans on people with a very low likelihood of entering a therapeutic trial.

Allocation Follow-Up
Analysis

Study visits
Prior to any study procedures, an informed consent was obtained from all study participants.The protocol was reviewed and approved by Advarra, a central institutional review board (Reference Number Pro00046018).The study is registered on ClinicalTrials.govNCT04733989.

Visit 1 (cognitive testing and blood sampling)
During Visit 1, contact and demographic information were collected, vital signs, height, medical history (including known family history of AD), and concomitant medications were recorded.Cognitive assessments were completed in the following order: MMSE, RAVLT with delayed recall, FAQ, and the GDS.The FAQ and GDS could be administered during the RAVLT test between the initial learning trials and the delayed recall period.(Other cognitive assessments were conducted after the GDS during Visit 1; however, results will be presented in future publications.)Blood samples for biomarker analyses and required lab tests were also collected during Visit 1.

Visit 2 (imaging or CSF collection)
At Visit 2 participants received a brain amyloid PET scan at a designated imaging center OR if the site did not have local access to a qualified PET imaging center, a lumbar puncture for CSF collection was performed.

Visit 3 and follow-up (review of brain amyloid PET/CSF results and blood collection)
During Visit 3, additional blood samples were collected from participants.Participants also met with their site PI or other appropriate study staff to review their brain amyloid assessment results (unless the participant declined this disclosure).For all participants with brain amyloid positive results, an additional contact occurred within 24 to 72 hours postdisclosure by telephone or via a virtual visit.This followup contact evaluated the participant's overall well-being using the GDS and provided any warranted referral.

2.4
Blood biospecimen collection and management 2. 4 Only the blood assay results of the lipid panel were returned to the PI for review to determine any clinically significant abnormal lab-oratory findings and/or the need for participant follow-up to address non-study-related health issues.

Brain amyloid PET scan
Brain amyloid PET scans were conducted at a designated imaging facility near each site.Prior to the approximate 20-minute scan, the US Food and Drug Administration (FDA)-approved tracer, Amyvid, florbetapir F-18 (18F-AV-45), was injected intravenously with a target per the label dosage and administration instructions. 29To improve consistency of amyloid PET scan interpretations, all locally developed images were uploaded into an electronic imaging portal that were accessible by IXICO's network of specialists trained to interpret brain florbetapir PET images.Hence, all brain amyloid PET scans were centrally read by an expert trained in the manufacturer read process and following a process in which the reader had visibility to a subject's standardized uptake value ratio (SUVR) value but made the final determination according to manufacturer standards.Scan results were disseminated to each research site.Results of the scans were discussed with the participants at Visit 3 by a trained medical professional.

CSF collection
One research site centrally serving and recruiting from a largely Hispanic population in South Texas did not have local access to a qualified PET imaging center.For participants at this site a CSF sample was collected and analyzed for Aβ40, Aβ42, and Aβ42/Aβ40 by Quest Diagnostics, a CLIA-certified laboratory.
The CSF sample was acquired via a lumbar puncture performed by a qualified physician.Prior to the procedure, a coagulation panel and computerized tomography (CT) scan of the brain were reviewed to rule out any contraindications.CSF sampling was performed between 8:00 am and 12:00 pm to minimize diurnal variation of CSF parameters.CSF samples were sent to Quest Diagnostics' CLIA central lab for analysis and the results were disseminated to the site.The results of the CSF analysis were discussed with the participants at Visit 3 by a trained medical professional.

General statistical considerations
The primary purpose of this study is to determine the relationship of blood and digital biomarker measurements to the presence of amyloid plaques in the brain identified through brain amyloid PET or CSF analysis.(Digital biomarker data will be presented in future publications.) The The ability of blood-based biomarkers to predict amyloid positivity was assessed using logistic regression analyses as implemented in SAS proc logistic.ROC curves and associated AUC were calculated using the pROC package and proc logistic in R and SAS, respectively.Differences in AUCs were compared using the method of DeLong et al. 31 Logistic regression was performed with amyloid PET as the outcome or dependent variable and biomarkers as the independent or predictor variable.A secondary logistic regression analysis was performed with age and apoE ε4 status (carrier/non-carrier) as covariates.Subgroup analyses by ethnicity were also performed with and without covariate adjustment.For categorical variables (clinical cohort, sex, and apoE ε4 carrier status), chi square or Cochran-Mantel-Haenszel statistics were used to evaluate the relationship to brain amyloid status.
Due to the Bio-Hermes Study's robust database, it is anticipated that additional statistical analysis will be conducted by study partners, and with appropriate other partners with expertise in artificial intelligence and/or machine learning to possibly identify a predictive multivariate biomarker relationship or algorithm to improve prediction capabilities of amyloid PET.

Secondary analyses
Exploratory analyses were performed to identify any potential differences related to URP in biomarker thresholds and their relationship to brain amyloid positivity status.Differences in biomarker thresholds between groups defined by race and ethnicity were evaluated with logistic regression and AUC standard errors were calculated as described above.Some participants had biomarker values that were below the lower limit of quantitation (<LLOQ) or the lower limit of detection (<LOD) for the assay used.This was most common for measures of p-tau181 and p-tau217, for which 100 and 169 participants, respectively, had values below the LLOQ or LOD.In those cases, a random value was imputed between zero and the LLOQ or LOD because those participants were known to have very low values.There were seven values above the upper limit of quantitation (ULOQ) for p-tau217.In those cases, a value equal to the ULOQ was assigned.There was no significant association of sex with amyloid status in the population as a whole or in any of the clinical groups.Education level was also not related to amyloid status for the group as a whole or for any of the clinical groups.The proportion of more females than males in the Bio-Hermes

Participant characteristics (total population)
Study is also similar to that observed in most therapeutic trials. 33Education levels were generally high and comparable to those observed in many therapeutic trials. 19Cognitive and functional measures were generally worse in participants who were amyloid positive compared to those who were amyloid negative across the total study population (P < 0.0001 for MMSE, RAVLT, and FAQ).The differences for the MMSE and RAVLT related to amyloid status were greater in the mild AD group than in either the CN or the MCI group.The FAQ scores were slightly but significantly higher (more impaired) among amyloidpositive participants in the CN (P = 0.0083) and MCI (P < 0.0001) but not in the mild AD (P = 0.4864) groups.A strong relationship of apoE ε4 status to amyloid positivity was observed overall and in each of the clinical groups; in all groups the prevalence of apoE ε4 carriers was significantly higher among amyloid-positive than among amyloid-negative participants (P < 0.0001 in each case).
In summary, these results show strong relationships of amyloid positivity with older age and with a higher prevalence of apoE ε4 carriers across all clinical groups.Notes: Values indicated to be below the lower limit of quantitation were assigned an imputed value between 0 and the lower limit of quantitation.Values above the upper limit of quantitation were assigned the value at the upper limit of quantitation.P values were generated using a standard t test.Abbreviations: Aβ, amyloid beta; p-tau, phosphorylated tau; SD, standard deviation; t-tau, total tau.Notes: Values indicated to be below the lower limit of quantitation were assigned an imputed value between 0 and the lower limit of quantitation.Values above the upper limit of quantitation were assigned the value at the upper limit of quantitation.P values were generated using a standard t test.Abbreviations: Aβ, amyloid beta; p-tau, phosphorylated tau; SD, standard deviation.

TA B L E 1
Figure 2 presents the ROC curves for the three most promising biomarkers: p-tau181, p-tau217, and Aβ42/Aβ40.ROC curves are provided for the Bio-Hermes population as a whole and for the three race and ethnic groups separately.For the total population the ROC curve for p-tau 217 was significantly better at predicting amyloid status than those for Aβ42/Aβ40 and p-tau 181 (P < 0.0001) but the latter two groups were not different (P = 0.18).Because all the biomarkers' AUCs using adjusted and imputed values were > 0.8000, and in most instances > 0.8500, ROC curve results used AUC data adjusted for age and apoE ε4 status and imputed values are included for p-tau181 and p-tau217.The following are the number of imputed values for each group, total population (n = 269), non-Hispanic White (n = 179), Hispanic (n = 44), and non-Hispanic Black (n = 44).The specific values for sensitivity and specificity depend upon the cutoff chosen and optimal cutoffs will vary depending upon tolerance for false positives and false negatives.For the total population, and for each of the subgroups, the ROC curve for p-tau217 was above those for the other two biomarkers but all have ROC curves sufficient to provide good predictive power for brain amyloid status.

DISCUSSION
The Bio-Hermes Study was designed to provide a platform dataset to enable evaluation of blood-based and digital biomarkers for their ability to identify persons who are brain amyloid positive in a racially and ethnically diverse group of persons screened at clinical trial recruiting sites.The aim of the current analysis was to evaluate widely used blood-based biomarkers in predicting brain amyloid positivity.
Results indicate that Aβ42/Aβ40 ratio, p-tau181, and p-tau217 are good predictors of brain amyloid positivity in this clinical trial-ready population and suggest that further separate evaluation of biomarkers for Hispanic or non-Hispanic Black participants may be useful.
Of these promising biomarkers, p-tau217 was numerically superior for the entire population and for each race and ethnic group examined separately; however, all showed ROC curves adjusted for age and apoE ε4 (with no imputation) with AUCs ≥ 0.8120 relative to brain amyloid positivity.Some values of both plasma p-tau181 and p-tau217 were below the LLOQ, and some p-tau217 values were greater than the ULOQ; using a simple imputation scheme improved the AUCs for both Notes: N values reflect lab assays available for analyses.Values indicated to be below the lower limit of quantitation were assigned an imputed value between 0 and the lower limit of quantitation.Values above the upper limit of quantitation were assigned the value at the upper limit of quantitation.P values were generated using logistic regression as implemented in SAS proc logistic.Abbreviations: Aβ, amyloid beta; apoE, apolipoprotein E; AUC, area under the receiver operating characteristic curve; p-tau; phosphorylated tau; SE, standard error.
F I G U R E 2 ROC curves for prediction of brain PET or CSF amyloid positivity adjusted for age and apoE ε4 status with imputed values.Aβ, amyloid beta; apoE, apolipoprotein E; CSF, cerebrospinal fluid; PET, positron emission tomography; p-tau, phosphorylated tau; ROC, receiver operating characteristic.
analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication.
The primary objective of the Bio-Hermes Study was to evaluate the ability of several promising blood-based and digital biomarkers to reflect the presence of brain amyloid in participants enrolled at clinical trial sites using recruitment procedures similar to those used in AD therapeutic drug studies.Participants in the Bio-Hermes Study had clinical characteristics similar to those enrolled in clinical trials of disease-modifying treatments and, because multiple biomarkers were obtained, the predictive value of biomarkers alone or in combination can be evaluated.The data collected in this study should facilitate the use of biomarkers as prescreening tools to identify trial participants likely to have amyloid deposits in the brain as measured by PET or CSF, thus avoiding time consuming and costly screen failures.
Population general demographics by clinical status.
relationship of clinical and demographic variables and potential blood-based biomarkers to brain amyloid was determined for participants in the three study cohorts (CN, MCI, and mild AD).Because studies show the rate of amyloid PET positivity in persons > age 65 with nominal significance of at least P < 0.01 are noted in the results discussion.For ROC curves, area under the ROC curves (AUCs) were also calculated and adjusted values were used to assess the impact of age and apoE ε4 status (carrier vs. non-carrier) because of the strong relationship of age and apoE ε4 status to amyloid positivity.

Table 1
presents the demographic variables, clinical measures, and apoE ε4 results for each of the clinical cohorts separated by brain amyloid status along with results for the total Bio-Hermes Study population.Across the three clinical cohorts the rate of amyloid positivity increased with greater cognitive impairment, from 21% (84 of 400) for CN participants to 34.7% (103 of 297) for participants with MCI to 60.2% (156 of 259) in the mild AD group (P < 0.001).For the total population enrolled (N = 1001), brain amyloid-positive participants were older (P < 0.0001) than amyloid-negative participants; the amyloid relationship with older age was clear in the CN group and the MCI group and marginal in the mild AD group (P < 0.0001 for CN, P = 0.0007 for MCI, P = 0.0490 for mild AD groups).Overall mean age (years) was in the early 70s similar to the mean age of participants enrolled in most therapeutic research studies.

-
Notes: P values for age, education, MMSE, RAVLT, and FAQ were generated using a t test; P values for sex and apoE ε4 status were generated using a chi-square test.P values not calculated for individual ethnoracial groups due to small populations.Abbreviations: ApoE, apolipoprotein E; FAQ, Functional Activities Questionnaire; MMSE, Mini-Mental State Examination; RAVLT, Rey Auditory Verbal Learning Test; SD, standard deviation.Clinical status and blood-based biomarkers.
Ethnicity and race demographics.Notes: P values for age, education, MMSE, and FAQ for individuals not in the non-Hispanic White populations were generated using a standard t test vs. the non-Hispanic White population; P values for sex and apoE status were generated using a chi-square test.P values for amyloid positivity rates were generated by logistic regression using age, sex, and education as covariates.P values were not calculated for the other cohort due to the small population.Abbreviations: ApoE, apolipoprotein E; CSF, cerebrospinal fluid; FAQ, Functional Activities Questionnaire; MMSE, Mini-Mental State Examination; PET, brain positron emission tomography; SD, standard deviation.Ethnicity and race and select blood-based biomarkers.
TA B L E 5 AUC values across populations with no imputation and imputation values and unadjusted and adjusted for age and apoE ε4.