White, Charles L.

AbstractBackgroundPost‐mortem assessment remains the only option and a gold standard for a definitive diagnosis of Parkinson’s disease (PD). The antemortem diagnosis is challenging due to a variable clinical presentation and the abundance of mimicking conditions, obscuring the clinical picture and delaying the treatment. It is well established that while PD pathologically manifests in central nervous system with aggregation of α‐synuclein as Lewy bodies and Lewy neurites, this pathology is also found in the peripheral nervous system. Peripheral Lewy‐type synucleinopathy (LTS) is present in early PD, suggesting its utility as a diagnostic and prognostic biomarker. We have previously confirmed that detection of LTS in submandibular gland (SMG) biopsies is sensitive (.75) and specific (.90) for early PD. However, the assessment by a neuropathologist of multiple levels of such biopsy is laborious and time‐consuming.MethodHere work we applied two Convolutional Neural Networks (CNN) for detection of LTS on 283 digital whole slide images (WSI) from 95 unique SMG biopsies. A total number of 8,386 LTS were annotated following the interrater reliability study with kappa=.7. We used transfer learning to train a CNN model to classify image patches (40‐by‐40 pixels at 20x) with and without presence of LTS objects.ResultThe trained CNNs showed the following performance on image patches: sensitivity: .982/.988, specificity: .998/.999, precision: .852/.950, accuracy: .995/.998, and F‐1 score .912/.968 for ResNet30/InceptionV4 respectively. On test WSI the sensitivity in detecting LTS was .963/.971 respectively.ConclusionThis work is first to demonstrate the practical utility of an CNN for screening for LTS, which can further translate into a practical screening tool facilitating the antemortem diagnosis of PD. We further plan to develop neural networks for accurate and precise detection and quantification of LTS in antemortem SMG biopsies. This, altogether, would offer a screening, confirmatory, prognostic, and quantitative tool for clinical assessment of early PD.

AbstractBackgroundDigital pathology and machine learning (ML) workflows have paradigm shifting potential especially in the field of Alzheimer’s disease and related disorders. However, many institutions/centers may not have access to these technologies. To provide benchmark data, a survey was developed and distributed to National Institutes of Health’s Alzheimer’s Disease Centers (ADCs) in the United States.MethodSurvey questions covered topics such as: infrastructure (i.e. type of digital slide scanners used), funding sources (i.e. how the scanner was funded), and data management/storage (i.e. size of digital files). After development, review, and approval by the ADC digital pathology working group, the survey was distributed via email to 35 past and current ADC directors and/or ADC neuropathology core leaders. Participation in the survey was completely voluntary and answers did not contain any personal data.ResultThe survey generated over a 90% response rate, and the majority of those who completed the survey were neuropathology core leaders; 81% stated their ADC had access to a digital slide scanner, most common brand being Aperio/Leica. One third of respondents stated there was a fee for service to utilize the scanner. For digital pathology and/or ML resources, 40% of respondents stated none are supported in any way by their ADC. To cover the purchase and operation of the scanner, 50% stated they had institutional support. Many were unsure of the approximate average scanned file size of digital images (37%) and total amount of storage space that files occupied (50%). Many (75%) were aware of other departments at their institution working with digital pathology and/or ML, but a similar percentage was unaware of multi‐university or industry partnerships.ConclusionThese results demonstrate many ADCs have access to a digital slide scanner and had institutional support to cover the purchase. However, further investigation is needed to understand hurdles and barriers for implementing both digital pathology and ML workflows aiding in standardized methods across ADCs.

Missense mutations in LRRK2 (leucine-rich repeat kinase 2) are a major cause of PD (Parkinson's disease). Several antibodies against LRRK2 have been developed, but results using these polyclonal antibodies have varied widely leading to conflicting conclusions. To address this challenge, the Michael J. Fox Foundation for Parkinson's Research generated a number of monoclonal antibodies targeting epitopes across the LRRK2 protein. In the present paper, we report optimized protocols and results for ten monoclonal antibodies for immunoblotting, immunohistochemistry, immunoprecipitation and kinase activity assays, in rat, mouse and human brain tissue. Several efficacious antibodies were identified, but results demonstrate that the mouse monoclonal N241A/34 is suitable for most applications, with the best overall rabbit monoclonal antibody being c41-2. These antibodies produced a dominant band of the expected size via immunoblotting and a lack of labelling in tissue derived from LRRK2-knockout animals under optimized conditions. A significant proportion of LRRK2 protein localizes to insoluble fractions and no evidence of truncated LRRK2 protein was detected in any fraction from rodent or human tissues. An assay was developed for the robust detection of LRRK2 kinase activity directly from frozen mouse and human brain tissue, but precipitous declines in activity were observed that corresponded to increasing post-mortem intervals and processing times. Finally, we demonstrate the highest levels of brain-localized LRRK2 in the striatum, but note differential expression patterns between rat and mouse in both striatum and cortex. Anti-LRRK2 monoclonal antibodies that are unlimited in availability together with the proposed standardized protocols should aid in the definition of LRRK2 function in both health and disease.