Harnessing team science in dementia research: Insights from the Alzheimer's disease research group in South Carolina

Abstract

Alzheimer's disease and related dementias (ADRD) are complex and rapidly growing public health challenges that require integrative, collaborative approaches. To meet this need, the Alzheimer's Disease Research Center (ADRC) of South Carolina, a state-funded partnership across multiple institutions, has embraced a team science model that brings together expertise from four key disciplines: neuroimaging-neurology, health sciences, molecular biology, and engineering. This paper highlights the work at the University of South Carolina (USC) and how each discipline contributes uniquely yet collaboratively within a recurring loop model framework. Extending from USC's Cancer Prevention and Control Program, the framework guides scientific growth through four iterative phases: discovery, development, delivery, and dissemination. From identifying cellular and molecular biomarkers to applying neuroimaging for early diagnosis, utilizing wearable technologies for real-time monitoring, and analyzing statewide data to understand caregiver burden and health inequities, each group contributes to a comprehensive and translational research cycle. By supporting structured mentorship, cross-disciplinary pilot projects, and shared infrastructure, the ADRC initiatives advance research that is both scientifically rigorous and equity-focused. This approach provides a valuable framework for advancing ADRD research and can inform other institutions aiming to tackle similarly complex health issues from discovery through dissemination.

Introduction

Alzheimer's disease and related dementias (ADRD) present a significant and rising public health challenge as the global population ages.^1,2 Within the United States (US), ADRD ranks as the seventh leading cause of death.^2,3 Approximately 6.9 million Americans aged 65 and older are living with ADRD.^4,5 This number is projected to nearly double, reaching 12.7 million by 2050 if no medical breakthroughs occur to prevent, slow, or cure the disease. ³ Alzheimer's disease (AD), the most common form of dementia, accounts for 60–80% of all dementia cases. ⁶ This progressive neurological disorder impairs memory, cognitive functions, and behavior. ⁷ Other dementias include vascular dementia, Lewy body dementia, frontotemporal dementia, and mixed dementia.^8,9 There is often pathological overlap, as vascular brain pathology frequently coexists with Alzheimer's-type pathology. ⁹

The economic impact of ADRD is daunting, with the total cost of care estimated at $345 billion in 2024, and expected to rise to $1 trillion by 2050. ¹⁰ These cost-of-care projections are based on direct healthcare costs associated with ADRD treatment as well as indirect costs, such as loss in productivity, diminished quality of life, and increasing dependence on unpaid family caregivers and long-term care, which are not included.^11,12 The overall direct medical costs for patients with ADRD are $196 billion, with an additional $254 billion for caregiver time. ¹³ On average, a 65-year-old in the US faces a total of $138,000 in long-term care costs over their lifetime due to ADRD, with families covering about half of these costs out-of-pocket.^10,14 As of 2023, unpaid caregivers provided 18.4 billion hours of care for patients with ADRD, valued at nearly $350 billion.^10,11 Caregiving burdens significantly impact caregiver health, employment, and financial stability.^6,15

The urgency to address ADRD stems from multiple factors; currently, there is no cure for ADRD, and treatments primarily manage symptoms. ¹⁶ Recently, the US Food and Drug Administration (FDA) granted full approval to two anti-amyloid drugs for treating early-stage AD: lecanemab in 2023 and donanemab in 2024. ¹⁷ The benefits versus risks of these drugs are being studied, and the reported improvement with these drugs may not be perceptible to the patient, their family, and their physician. ¹⁸ These advances emphasize the importance of early detection and accurate classification, as current interventions may slow but do not reverse the progression of the disease. Health disparities further exacerbate the toll of ADRD. ¹⁹ ADRD disproportionately affects adults from ethnically diverse groups (such as African American, Hispanic/Latino, and Native American populations) due to inequities in diagnosis, access to healthcare, continuity of care, insurance status, and education levels.^12,20,21

The University of South Carolina (USC) Alzheimer's disease research group

USC contributes to a statewide initiative to address ADRD through the South Carolina Alzheimer's Disease Research Center (ADRC) with funding from the South Carolina Department of Health and Human Services to Clemson University, the Medical University of South Carolina, and the University of South Carolina. ⁵ The ADRC of South Carolina offers a unique opportunity given the relatively high rate of dementia and medical need juxtaposed with the leading scientific centers of the state sharing research and clinical tools. Moreover, the geography of SC provides accessibility to researchers, community members, patients, and research participants to travel to other centers with relative ease. The state-funded ADRC also emphasizes the importance of data-driven education and outreach. They are especially attentive to engaging with ethnically, racially, and geographically diverse communities, including improving the recruitment of African-American and rural-dwelling adults into ADRD research to better understand and address the elevated risk of the disease in these populations.^20–22 The center's mission extends to translating research findings into practical improvements in diagnosis, care, and prevention, with a commitment to advancing research and addressing the needs of diverse communities in South Carolina. ⁵

Leveraging a framework for ADRD from cancer prevention and control research

The USC Alzheimer's Disease Research Group within the SC-ADRC adopted the recurring loop model from cancer prevention and control to present the evidence-based and comprehensive approach to its ADRD research from discovery through dissemination. ²³ Team members studying chronic disease have implemented the model for their cancer-focused initiatives. The model has four main phases (Figure 1): discovery, development, delivery, and dissemination, and each phase plays a critical role in the research process and the ultimate impact of the research on targeted communities. An innovative feature of this model is its iterative nature, allowing researchers to revisit earlier phases at any point in the process. This flexibility enables continuous refinement and improvement of interventions based on new evidence, community feedback, and changing needs. The model also emphasizes community engagement and multidisciplinary collaboration throughout the research process, ensuring the work remains relevant and impactful for the intended population. It is for these critical reasons—the iterative nature, opportunity for further refinement, emphasis on community engagement, and experience with the model from prior work on cancer prevention and control—that the team embraced this guiding model.

Figure 1.

Recurring loop model.

This model has been effectively used to support cancer research and offers a valuable framework for ADRD research. ²⁴ The recurring loop model focuses on the continuous process of improving healthcare outcomes through a cyclical approach. It includes four key components: the first component is discovery; this phase involves generating new knowledge or insights, typically through research or scientific inquiry. It includes basic research, clinical studies, epidemiological investigations, and systematic reviews. The goal of this phase is to generate research evidence that will inform future healthcare interventions or policies. The second component is development, which follows the previous step; research evidence is leveraged to develop practical solutions, such as new treatments, interventions, or technologies. The goal of this phase is to translate discoveries into viable products, therapies, or practices that can be tested in real-world settings. The third delivery component focuses on implementing and applying the evidence-informed solutions to healthcare practices or public health systems. In this phase, new treatments or interventions are introduced into practice through clinical trials, pilot programs, or integration into existing systems. The goal of this component is to ensure that the evidence-informed solutions are accessible, effective, and widely adopted by healthcare providers and patients. The last component, dissemination, refers to the translation of knowledge or intervention outcomes to a broader audience, such as other healthcare practitioners, policymakers, and the general public. This can include journal publications, research briefs, workshops, conferences, and media outreach. The dissemination phase aims to facilitate the widespread adoption and application of new knowledge and solutions across diverse populations and regions.^25–27

Overview of the multidisciplinary ADRD work and application of the recurring loop model

Four disciplines at USC (Neuroimaging-Neurology, Health Sciences, Molecular Biology, and Engineering) have joined together to conduct ADRD research to enhance early diagnosis, treatment, and support for persons with ADRD and caregivers. We describe this work and how the recurring loop model guides the initiatives. Our research group generates hypotheses from the cellular level to the community level and embraces team science by sharing ideas and forming research collaborations. While each of us works on different aspects of ADRD, we are open to learning from each other. For example, while biomarkers and brain imaging have made rapid gains, we need to better understand how to communicate those findings and how the increased knowledge impacts the individual and family. Likewise, drug therapies and lifestyle modifications that may prevent or slow the progression of the disease require translation to clinical specialties that can implement clinical trials. Team science requires looking at the big picture and reflecting on the impact of our findings through the lens of other research specialties; for example, findings from a drug trial conducted by microbiologists may have unexpected clinical implications identified by the nursing team. ²⁸ Our team's work prioritizes mentorship with active involvement of trainees (graduate students, postdoctoral fellows, early-career faculty) and more established scholars. We also endorse community members affected by ADRD or who lead community organizations to join our team to inform our methods and assist us in disseminating our findings. For example, a community-based organization recommended and supported dissemination of data and policy briefs to legislators and community organizations. This type of collaborative work has been successful due to team members’ long-term partnerships with community and state agency partners and organizations and having established trust and clear roles and communication. Building equitable partnerships requires time and authentic engagement to ensure that the goals and expectations of both researchers and community partners are aligned. ²⁹

Regular meetings, zoom calls, a longer retreat, and emails in between meetings help with clear, consistent communication and engagement of team members. We also have a shared document in which we track progress and deliverables. Finally, the group also promotes and shares infrastructure such as a comprehensive, statewide ADRD registry, intervention data, neuroimaging core, and molecular equipment and resources that help sustain cross-college and interdisciplinary collaboration.

Neuroimaging-Neurology group

The Neuroimaging-Neurology team uses advanced brain imaging techniques to study changes in the brain associated with aging and dementia, aiming to detect signs of neurodegeneration or neural dysfunction. This group has successfully leveraged imaging approaches such as magnetic resonance imaging (MRI), diffusion tensor imaging (DTI), and arterial spin labelling (ASL) in a variety of populations to study how in vivo small-vessel integrity, network efficiency, and neurodegeneration relate to behaviors affected by dementia, such as memory and language.^30–36

The Neuroimaging-Neurology team's overarching goal is to identify the neural mechanisms related to cognitive resilience or decline, with particular emphasis on partially modifiable risk factors such as cardiovascular and metabolic health, sleep, physical activity, hearing, and psychosocial stress. These modifiable factors contribute to nearly half of all dementia cases worldwide. ³⁷ By synergistically incorporating health-related risk pathways and advanced neuroimaging techniques, we can attempt to identify individuals at risk for developing dementia while also discovering actionable intervention targets. Embedding brain imaging into a Discovery → Development → Delivery → Dissemination pipeline can guide the neuroimaging initiative in isolation, as well as create reciprocal touchpoints with our sister teams.

Mapping to the recurring loop model: Neuroimaging-neurology group

Discovery

The Neuroimaging-Neurology group contributes to the discovery arm of the translational loop by investigating how systemic health, lifestyle, and psychosocial factors are reflected in the brain. Our Aging Brain Cohort, ³⁸ an ongoing cross-sectional and longitudinal study of brain health across the adult lifespan, currently has multimodal MRI scans and well-characterized biomedical data for over 400 adults aged 20–80 years. Each participant contributes high-resolution structural MRI, diffusion MRI, resting-state functional MRI, and ASL MR perfusion scans, along with behavioral and biomedical phenotyping. This includes cognitive testing (e.g., Montreal Cognitive Assessment ³⁹ ), National Institutes of Health Cognitive Toolbox, ⁴⁰ cardiovascular and physical activity profiles, hearing and vision testing, venous blood for omics (e.g., DNA methylation, neuropathological and inflammatory markers), and 5-year longitudinal follow-up for a subset of adults aged 65 or older. This rich dataset allows us to move towards more comprehensive modeling of factors that influence brain health.

Recent discoveries from our group highlight the power of this approach. 1) Network-specific atrophy mediates age-related cognitive decline. Reduced grey matter volume in frontal and parietal association cortices explained the effect of chronological age on multiple cognitive subdomains. ³¹ 2) Sensory health matters. ³⁰ Age-related hearing loss was associated with accelerated ‘brain age’ (e.g., worse brain health). Further, this decline in brain health mediated the relationship between hearing loss and cognitive decline, suggesting mechanistic links between sensory loss, brain health, and cognitive decline in older adults. 3) Cerebrovascular health is a silent driver of cognitive decline.^33,35 White matter hyperintensity (WMH) burden, a sign of cerebral small vessel dysfunction, mediated the relationship between age and cognitive status throughout the adult lifespan. High cholesterol was associated with increased WMH load, suggesting that lipid control, even in early or middle adulthood, is vital for maintaining brain health and cognition. 4) Societal factors have neural correlates. Lower socioeconomic status predicted accelerated brain aging even after accounting for other health and lifestyle factors, suggesting actionable links between public policy and neural resilience.^35,38 5) Individuals express different patterns of cortical vulnerability, and these patterns relate to cognitive status and sensorimotor performance across the lifespan. ⁴¹ Further, epigenetic clocks, derived from peripheral blood, are associated with structural vulnerability in cortices critical for cognition. ³² This finding highlights the potential of easily accessible biomarkers to track brain health and cognitive aging. Together, these discoveries link multilevel risk factors to quantifiable neural mechanisms and generate candidate targets for modification, such as sensory rehabilitation, lipid management, cardiometabolic fitness, and socio-environmental support, that can inform the development and delivery phases of our translational mission.

Development

The Development phase will marry Discovery with the Neuroimaging-Neurology team's proven record of developing open-source, easy-to-use tools for researchers and clinicians, with the overarching goal of aiding prognosis and prevention. We will aim to train and validate predictive tools that flag individuals who are on high-risk trajectories towards cognitive decline before overt symptoms emerge. Our comprehensive approach in assessing determinants of brain health, as outlined in the Discovery phase, ensures that these tools will be informed by holistic health factors with a focus on identifying modifiable risk pathways at the individual level, advancing towards precision medicine. ⁴² While precision medicine may take multiple forms such as use of genetic analyses, the group's primary approach to precision medicine is on brain imaging.

Our group has developed multiple open-source, point-and-click software packages that lower the barrier to using advanced image analytics and have been widely adopted by researchers and clinicians. MRIcro, MRIcron, ⁴³ MRIcroGL, ⁴⁴ Surfice ⁴⁵ and the browser-based NiiVue have become standards in the field for medical image viewing, with thousands of downloads and active clinical use (https://github.com/neurolabusc). Visualization tools such as these are vital at every stage of brain imaging, including inspecting the brains of individual patients, verifying successful data processing, and viewing group-level results after quantitative hypothesis testing. We have also developed software that allows for statistical modeling of the relationships between multimodal brain imaging and behavior, a primary focus of the development phase. NiiStat (https://github.com/neurolabuscNiiStat) supports quantitative lesion-symptom mapping and connectomics, allowing for causal analyses of the associations between region-specific brain damage or disconnections and variables of interest, such as cognition or cardiovascular factors. The Graphical Brain Association Tool (gBAT ^{31 )}, supports mediation analyses between brain and behavior, allowing researchers to test specific path-level hypotheses about how relationships between two variables (e.g., chronological age and cognitive status) are mediated by neuroimaging features (e.g., grey matter volume). A natural next step for these analytic tools is to leverage big data to pretrain deployable predictive models that can assess patient-level risk of cognitive decline and neurodegeneration. These tools also support education and outreach efforts; for example, our brain2print.org platform uses AI to segment widely available T1-weighted MRI scans into 3D-printable models, helping students, clinicians, and the public better understand and appreciate brain structure. ⁴⁶

These platforms, which blend neuroimaging-neurology visualization and statistical analyses, demonstrate that we are well-positioned to build the next generation of hybrid neuro-health tools that will leverage rich neuroimaging data, in combination with comprehensive biomedical assessments and population-level insights, to generate interpretable, individualized risk assessments along with bespoke health recommendations for at-risk patients. This ongoing effort embodies the Development phase of the loop, turning discovery-driven insights into accessible technologies that may ultimately be embedded in future research and routine care.

Delivery

Our Delivery phase centers on getting the tools and knowledge we create into the hands of clinicians, researchers, and public-health partners securely, transparently, and at no cost. We have already released widely used, open-source packages for visualization (MRIcro, MRIcron, NiiVue) and quantitative analysis (NiiStat, gBAT), accompanied by permissive licenses, user manuals, and example datasets. Going forward, new software modules or analytic pipelines that emerge from the Development loop will follow the same model: public repositories, containerized builds for easy installation, and detailed tutorials suitable for research and clinical audiences. To ensure that our discovery-driven markers translate beyond our cohort, we will train predictive models internally and evaluate them on independent datasets through collaborations with other imaging centers and, where feasible, in clinical settings that routinely acquire MRI. Performance metrics, model weights, and documentation will be disseminated openly so outside groups can replicate, critique, and improve our work. By continuing to release robust software, curated datasets, and transparent models, and by inviting the community to test and refine these resources, we aim to make our neuro-health tools accessible, effective, and widely adopted.

Dissemination

We will disseminate our findings and resources through the traditional scientific channels that have the widest reach in our field. Peer-reviewed manuscripts and accompanying preprints will report methodological advances, validation studies, and clinical evaluations, ensuring rapid access and rigorous vetting by the neuroimaging-neurology community. Results will be presented at national and international conferences, such as annual meetings focused on neuroimaging, neurology, and cognitive aging, where platform and poster sessions provide opportunities for direct engagement and feedback from researchers and clinicians.

In parallel, software packages or analytic pipelines we develop will be released in openly accessible code repositories (e.g., GitHub) under permissive licenses, accompanied by clear documentation, example datasets, and step-by-step tutorials. When feasible, we will offer workshops to guide new users through workflows, leveraging the conference venues noted above or virtual formats to reach a broad audience. By combining publications, conference presentations, and well-supported code releases, we aim to make our tools and insights readily discoverable, reproducible, and usable by researchers and practitioners.

Health sciences group: Public health, nursing, and social work

The Health Sciences group engages in research and activities to generate new knowledge that aligns with the recurring loop model. These activities span from identifying community and caregiver needs to quantifying the burden of ADRD in SC, identifying disparities, and determining novel predictors of ADRD. Public Health, Nursing, and Social Work researchers focus on improving the well-being of caregivers by studying their health and stress levels, as well as using statewide data to understand broader trends in aging populations. The group studies population-level data from SC's Alzheimer's Disease Registry – the oldest and most comprehensive statewide dementia registry in the US.^{4,8,15,47–50} This data supports policymakers and healthcare providers in making informed decisions to aid caregivers and enhance healthcare services for people living with dementia, eventually, deepening insights into aging populations.

Mapping to the recurring loop model: Health sciences group

Project 1

Discovery

Caregivers’ health and well-being are uniquely influenced by the duration, nature, and intensity of their caregiving role and further determined by the intersecting factors of age, gender, race, income, education, their place of residence (urban/rural) that differentially impact caregivers’ physical, mental, social, financial wellbeing outcomes.^51–53 Caregivers are at risk for increased financial strain, ill health, greater health service utilization, and decline in physical and social activities. ⁵⁴

Existing research on caregivers typically addresses isolated physical, mental, social, economic issues and at a single point in time.^55,56 Our central premise is that one's experience with caregiving and the degree to which it impacts caregiver health is determined by broader life circumstances, such as caregiver's gender, age, race, income, education, and the physical environment (urban/rural) creating potential health inequities and outcomes among caregivers as a function of privilege, oppression, and differential access to resources and opportunities.^51–53,57

Using an ecological momentary assessment research design, ⁵⁸ the Health Sciences group's first pilot project is investigating the feasibility and usability of mobile technology (digital watch, smartphone, and data collection app) to measure daily caregiving activities, physical, mental, social, and financial aspects of health and wellbeing among caregivers of individuals with ADRD. A data science analytical approach will assess the differential experiences of caregivers over time. We are documenting reported facilitators and challenges to using continuous-long-term mobile digital data collection with this population and developing strategies to refine this novel method of data collection.

Development

The ultimate goal and intention of this work is to generate ‘digital markers’ of unpaid caregivers’ health and well-being and to generate predictive models of health care to anticipate the need for care among unpaid caregivers. This takes the form of investigating the associations between caregiving activities, and the physical, mental, social and financial health outcomes among unpaid caregivers of diverse age, race, gender education, income and place of residence to generate digital markers to inform new models of care that will enable proactive and anticipatory health care practices. The mobile digital data collection strategy supporting continuous long-term data collection will form the basis of the unpaid caregiver health and well-being data registry.

Delivery

The potential benefits of leveraging ‘big data’ with data science methods allow for the development of new and novel models of health care that predict care needs and enable proactive care services as opposed to reactive or crisis care responses. Using the data from unpaid caregivers, we want to predict or anticipate a range of possible caregiver services and supports tailored to their specific needs. This evidence will be used to inform the development of affordable and scalable strategies to support proactive and anticipatory care for unpaid caregivers and mitigate the need for crisis care responses for unpaid caregivers and/or the person with ADRD. Currently, to address the need for caregiver training, the public health group has implemented the nationally registered Dementia Dialogues program that provides the most recent and practical evidence-based information to both formal and informal caregivers on how to care for individuals with ADRD. ⁵⁹ This program is both instructor-led and self-paced at no cost to participants, which ensures that developed solutions are effectively integrated into healthcare and public health systems for long-term impact.

Project 2

Discovery

The purpose of the second pilot study is to investigate the burden of risk for cognitive decline among US adults and assess the ADRD risk profile of unpaid caregivers relative to non-caregivers. We hypothesize that US unpaid caregivers will have a higher risk profile for ADRD than non-caregivers.

With an aging population in both the US and Canada, the demographics of family/friend caregivers are shifting. The US is seeing an increasing proportion of younger adults caring for older family members, as well as those caring for older family members and a younger generation of children or grandchildren simultaneously. ⁶⁰ Researchers are calling these caregivers “sandwich generation caregivers”, as they are sandwiched between two generations in need of care. Most of these caregivers fall between the ages of 35 and 64 and are more likely than other caregivers to be balancing formal employment with informal caregiving responsibilities. A 2020 survey found that 58% of caregivers fall into this age group, with an average caregiver age of 49.4 years. ⁵¹ In the US, the proportion of young caregivers is only expected to grow as the country's population ages.^61,62 Unpaid caregivers who are employed are “at risk” for developing caregiver stress (e.g., physical, financial, and emotional). High financial strain is associated with poorer physical and mental health, greater work-life conflict, increased workplace absenteeism, lower job satisfaction, and a higher number of visits to the emergency room and hospital. There are significant costs to unpaid caregivers related to altered work patterns, loss of career advancement, and increased use of healthcare system services.^63,64

Concernedly, the latest 2024 Lancet Commission report identified factors associated with higher risk for cognitive decline that included: quality of education, depression, smoking, excessive alcohol, hearing impairment, uncorrected visual impairment, conditions such as traumatic brain injury, diabetes, obesity, high blood pressure and cholesterol, physical inactivity, social isolation, and air pollution.^65,66 Many of these factors are exacerbated and/or consequential to the role of unpaid caregiving. The Lancet Commission also estimated that dementia attributable to modifiable risk factors has increased from 40% to 45% since 2020. The public health impact of prevention strategies targeted at the 14 ADRD risk factors is substantial with an estimated 50% reduction of all dementias. However, the potential for prevention appears most advantageous if applied early and then benefits continue throughout the life course.^65,66 Understanding the profile and pervasiveness of the risk allows for targeted prevention strategies to mitigate ADRD later in life. ⁶⁶

Development

A cross-sectional analysis of Behavioral Risk Factor Surveillance System (BRFSS) data from 2023 will generate findings related to the frequency and composition of risk factors (alternatively, preventive factors) of ADRD among American adults and determine differences in the frequency/composition of risk factors among unpaid caregivers vs. non-caregivers.

Delivery

The findings from this research will generate insight into the cognitive health care of unpaid caregivers at the intersection of age, gender, duration of caregiving, personal health, and the health status of their family/friend receiving care. The long-term goal is to generate an interactive tool to personalize risk and identify ADRD prevention strategies for the general public and targeted high-risk populations.

Dissemination for both projects

The findings from both pilot projects described will be shared in multiple ways directed to academic and clinical-community audiences. This includes academic publications, presentations, and working with clinicians and primary care clinics to support the development of new models of care and practice. The group also produces yearly ADRD fact sheets to inform and educate members of the South Carolina community on the burden and distribution of ADRD at the state and county levels. These fact sheets have also informed the passage of legislation such as the Senate Bill 570 to train police and fire officials engaging with people with ADRD, and the Senate Bill 526 to increase training in ADRD for clinicians. ⁴⁷

Molecular biology group

Molecular Biology scholars are developing brain organoids and mouse models to 1) understand the underlying cellular and molecular mechanisms of mixed dementia (MD) (i.e., AD and vascular dementia (VaD) comorbidity) and 2) test novel preventive and therapeutic agents for MD. They are particularly interested in creating model systems that mimic the human blood-brain barrier and that inform the means whereby brain cells (like neurons and microglia) sense and respond to vascular dysfunction. By using these organoid models, scholars will learn more about how dysfunction in the interactions between cerebral microvascular endothelial cells, microglia, and neurons lead to MD. The work with organoids will be complemented by the use of mouse models, including the 3xTg-AD mouse harboring human PS1_M146V, APP_Swe, and tau_P301L transgenes that recapitulate salient features of AD. ⁶⁷ Combining the 3xTg-AD mouse model of AD with bilateral carotid artery stenosis (BCAS^68,69), allows for the study of cerebral hypoperfusion on AD pathology and cognitive impairment. Together, these approaches will facilitate an understanding of the underlying cellular and molecular mechanisms of MD. In addition, this group is using cell culture approaches involving microglia, astrocytes, cerebral microvascular endothelial cells, and neurons – cells involved in MD—to screen natural compounds for the discovery of potential therapeutic agents for further testing in the organoid and mouse model systems. Overall, this research is key to identifying molecular targets in the brain for the prevention and treatment of MD. Taking advantage of the mechanistic insights gained through these studies, these model systems allow for testing of drugs that could potentially slow down, stop, or even reverse the progression of dementia by acting on the identified potential molecular targets.

Mapping to the recurring loop model: Molecular biology group

Discovery

The molecular biology group is engaging in research that is consistent with discovery in the recurring loop model. MD presents unique therapeutic challenges due to its complex pathogenesis. Indeed, both the neurodegenerative and vascular pathologies need to be concurrently addressed to effectively slow the progression of MD. Using the above-mentioned organoid and mouse models, they seek to identify molecules and signaling pathways in cerebral microvascular endothelial cells, microglia, and neurons that contribute to the pathogenesis of MD. Identification of such therapeutic targets will be further studied for causation using gene manipulation strategies (e.g., CRISPR gene editing and transgenic mouse models). Following the establishment of therapeutic targets, therapeutic strategies will be pursued under the development stage.

Development

On the basis of the mechanism-targeting discovery described above, the molecular biology group is implementing screening of a natural compound library to identify compounds that modulate molecules and/or signaling pathways involved in the pathogenesis of MD. To enhance efficiency, a high-throughput cell culture approach is the initial strategy. Microglia, cerebral microvascular endothelial cells, astrocytes, and neurons—cells involved in MD—will be treated with natural compounds for efficacy in targeted candidate molecules and signaling pathways driving MD. Promising therapeutics will then be tested for efficacy in organoid and/or mouse models, where disease progression and behavior can be assessed. Natural products that show potential in mitigating MD will move to the next stage in the recurring loop model (i.e., Delivery).

Delivery

The molecular biology group leaders will partner with AcePre, LLC, a biotechnology start-up company, that is focused on the development of natural compounds as preventive and/or therapeutic agents for diseases such as AD. The molecular biology group is currently supported by an R41 grant (R41AG087769) awarded to AcePre, LLC to develop a novel strategy to deliver N-acetylcysteine for AD treatment. Several more AD-related SBIR/STTR proposals are in various stages of progress. These grant mechanisms support preclinical testing of the compounds identified during the Development Stage for assessment of their pharmacokinetics, toxicology, and efficacy using small and large animal models. This approach provides a viable path for moving these promising therapeutic strategies to the clinic.

Dissemination

Regarding dissemination, the molecular biology group will publish their findings on drug development, including assessments of drug safety, pharmacokinetics, and efficacy. This is standard practice for their work on natural compound development.^70–73 Natural compounds that meet the threshold for safety and efficacy in pre-clinical studies will move towards clinical development. For this, the molecular biology group will submit Investigational New Drug (IND) applications to the FDA. As soon as the IND is approved, clinical trials will be carried out to test the prospective drugs. This is a multi-year process that will involve integration and collaboration with the Neuroimaging and Health Sciences groups. The end goal is to bring new drugs to the clinic for the prevention and treatment of AD and MD.

Engineering group

The Engineering team is creating innovative technologies to detect cognitive decline early, prior to significant damage. Engineering research related to ADRC focuses on developing advanced technologies for the early detection of ADRD in efforts to improve early intervention and patient outcomes. Early detection is crucial because it can help doctors start treatments before significant damage occurs and assist researchers in selecting subjects for clinical trials. The work of the engineering group is crucial for enhancing detection methods and care plans.

Mapping to the recurring loop model: Engineering group

Discovery

Integrating artificial intelligence (AI) with biomarker analysis holds great potential to enhance the accuracy and speed of diagnosis by identifying complex patterns across different biomarkers. AI and predictive modeling are increasingly used to predict various diseases by analyzing complex datasets, including medical images, blood biomarkers, genomic data, and clinical records. These AI-driven approaches leverage machine learning algorithms to identify patterns and make predictions about disease onset, progression, and outcomes. For instance, AI models have been applied in the early detection of AD using protein biomarkers, such as amyloid-β (Aβ) and tau. ⁷⁴ Machine learning algorithms, including deep learning and support vector machines, have also been utilized to predict the progression of neurodegenerative diseases by analyzing neuroimaging data. ⁷⁵ In oncology, AI has shown promise in predicting cancer risk and treatment outcomes. A study by Esteva et al. ⁷⁶ demonstrated the use of convolutional neural networks to predict skin cancer from dermatological images, achieving performance comparable to expert dermatologists. Similarly, AI models have been employed in predicting cardiovascular diseases, such as using deep learning to analyze electronic health records to predict the likelihood of heart failure or strokes. ⁷⁷ These applications highlight the potential of AI and predictive modeling to revolutionize disease prediction, offering more accurate, early-stage diagnostics, personalized treatment plans, and better healthcare outcomes.

AI holds the potential to significantly advance AD research by enabling diagnostics that are not currently possible with single-biomarker measurements. Recent studies demonstrate that AI, specifically machine learning and deep learning, can uncover complex patterns in large datasets that are beyond the scope of traditional methods.^78,79 These techniques have shown promise in identifying early neurodegenerative changes in imaging data and genetic markers, as well as predicting disease progression.^74,75 While there have been extensive efforts to use these advances, the integration of multi-modal data to develop biomarker-based diagnostics remains underexplored. This approach not only advances the capacity of AI to handle and analyze diverse data but also enhances our understanding of the complex biological mechanisms underpinning AD and ADRDs, which could lead to more effective diagnostic and therapeutic strategies. These advancements are crucial for improving diagnosis and prediction in the context of neurodegenerative diseases, with the potential to refine clinical decision-making and improve patient outcomes through more precise and personalized treatment options.

Development

Development of the AI approach will leverage data from both deceased patients diagnosed with AD and ADRD and living patients at risk for AD who have been followed longitudinally. Data available from these subjects can comprise biological samples (blood, cerebrospinal fluid, and ocular fluid) alongside medical records. From the deceased patients, protein biomarkers Aβ_40/42, Tau, P-tau 181/217/231, neurofilament light chain (NfL), inflammatory cytokines) can be measured,^75,80–82 and genotyping to identify risk genes (APOE4, clusterin, TREM2) can be performed.^80–82 From the living patients, information can be garnered about ADRD-associated medical screening (including cognitive exams, brain imaging), diagnosis of related conditions (cardiovascular disease, diabetes, traumatic brain injury), and lifestyle choices (diet, exercise, smoking, alcohol consumption). This variety of information represents multi-modal data, including quantitative, categorical, text, images, etc., that can be analyzed collectively via AI to identify relationships and perform predictions.

Three different machine learning approaches, Random Forest, Bayesian Networks, and Artificial Neural Networks, will be applied to analyze the multi-modal data. Random Forest will be utilized for its ability to handle complex datasets and identify biomarkers that contribute most significantly to diagnosis. Bayesian Networks, which are useful for modeling dependencies between various factors, will be employed to integrate biomarkers and genetic factors to predict the likelihood of AD. Lastly, Artificial Neural Networks, capable of handling multi-modal data, will be utilized to uncover intricate patterns and identify those associated with AD. Performance of the models will be evaluated using cross-validation techniques and independent validation sets. Hyperparameters will be optimized, and the models will be fine-tuned iteratively to maximize key performance metrics, including accuracy, precision, and recall.

Delivery

We envision the deployment of a modular, multi-modal software platform designed to provide diagnostic support. A comprehensive system could capture structured data such as patients’ lab tests, medications, and related diagnoses, as well as unstructured data such as imaging data or medical record text. The extracted information from different modalities will be integrated into a fusion module in which AI elements will determine the probability and severity of AD diagnosis. Prediction results will be displayed to physicians through a secure password-protected interface. Our modular approach will ensure interpretability of the prediction results by 1) allowing clinicians to observe which input module contributed the most to the prediction and 2) enabling each module to be interpreted for important features. The utility of this software platform will also allow expansion of the dataset by providing a decision support tool capable of collecting data from a network of AD clinicians.

Dissemination

The software platform could be patented and made available to physicians and clinical researchers, to aid in patient care and subject selection, respectively. Toward the former, this envisioned tool can potentially improve patient treatment options and outcomes and assist physicians and patients with critical decision-making. Toward the latter, this envisioned tool has the potential to enable clinical trials on subjects that present earlier in the disease progression, thus allowing assessment of the increased effectiveness of early intervention. Additionally, the created dataset will contribute to a larger national or international database, including those supported by the Alzheimer's Disease Data Initiative (ADDI). We also aim to explore potential collaboration with specific ADDI subgroups focused on sensory module development, with the goal of incorporating sensor perception metrics into the suite of AD biomarkers used for early diagnosis.

Conclusions

The USC Alzheimer's Disease Research Group is a strong example of how a multidisciplinary approach can meaningfully address the challenges of ADRD. Using the recurring loop model, we have created a dynamic, integrated process that drives both scientific progress and practical impact in dementia research and care. Each of our core research areas—Neuroimaging-Neurology, Health Sciences, Molecular Biology, and Engineering—contributes in unique but interconnected ways. Neuroimaging helps us uncover how cognitive decline unfolds, while our Health Sciences team explores caregiver wellbeing and systemic issues through public health initiatives. Molecular Biology provides insight into cellular mechanisms and works to develop new treatments, and our Engineering team supports early detection and predictive modeling through advanced technologies. We also prioritize community and state partner engagement in our work to ensure our research benefits all populations and helps reduce ADRD-related health disparities. While each of the groups conducts important dissemination of their research findings through professional outlets such as academic journals and scientific conferences, a future initiative will involve having the health sciences group be a formal liaison for more publicly facing dissemination practices with which they have expertise such as plain-language resource development and community health events, in collaboration with community partners.

All team-based work comes with challenges, especially when bringing people together across disciplines and communities. Sustaining the infrastructure to support this work will require extramural center-level funding, institutional commitment, and resources to maintain and enhance interdisciplinary and engagement. Such support is critical to provide stability for retaining core investigators, and administrative and lab support. With the group's strong foundation, continued commitment, and leveraged funding, we are uniquely positioned to conduct long-term comprehensive, interdisciplinary, and applied research on ADRD. We can play a key role in reducing the burden among individuals, families, and communities across the state and beyond.