Astrocytes are crucial to brain homeostasis, yet their changes along the spatiotemporal progression of Alzheimer’s disease (AD) neuropathology remain unexplored. Here we performed single-nucleus RNA sequencing of 628,943 astrocytes from five brain regions representing the stereotypical progression of AD pathology across 32 donors spanning the entire normal aging to severe AD continuum. We mapped out several unique astrocyte subclusters that exhibited varying responses to neuropathology across the AD-vulnerable neural network (spatial axis) or AD pathology stage (temporal axis). The proportion of homeostatic, intermediate and reactive astrocytes changed only along the spatial axis, whereas two other subclusters changed along the temporal axis. One of these, a trophic factor-rich subcluster, declined along pathology stages, whereas the other increased in the late stage but returned to baseline levels in the end stage, suggesting an exhausted response with chronic exposure to neuropathology. Our study underscores the complex dynamics of astrocytic responses in AD.

Background: Neuronal damage in acute CNS injuries and chronic neurodegenerative diseases is invariably accompanied by an astrocyte reaction in both mice and humans. However, whether and how the nature of the CNS insult—acute versus chronic—influences the astrocyte response, and whether astrocyte transcriptomic changes in these mouse models faithfully recapitulate the astrocyte reaction in human diseases remains to be elucidated. We hypothesized that astrocytes set off different transcriptomic programs in response to acute versus chronic insults, besides a shared “pan-injury” signature common to both types of conditions, and investigated the presence of these mouse astrocyte signatures in transcriptomic studies from human neurodegenerative diseases. Methods: We performed a meta-analysis of 15 published astrocyte transcriptomic datasets from mouse models of acute injury (n = 6) and chronic neurodegeneration (n = 9) and identified pan-injury, acute, and chronic signatures, with both upregulated (UP) and downregulated (DOWN) genes. Next, we investigated these signatures in 7 transcriptomic datasets from various human neurodegenerative diseases. Results: In mouse models, the number of UP/DOWN genes per signature was 64/21 for pan-injury and 109/79 for acute injury, whereas only 13/27 for chronic neurodegeneration. The pan-injury-UP signature was represented by the classic cytoskeletal hallmarks of astrocyte reaction (Gfap and Vim), plus extracellular matrix (i.e., Cd44, Lgals1, Lgals3, Timp1), and immune response (i.e., C3, Serping1, Fas, Stat1, Stat2, Stat3). The acute injury-UP signature was enriched in protein synthesis and degradation (both ubiquitin-proteasome and autophagy systems), intracellular trafficking, and anti-oxidant defense genes, whereas the acute injury-DOWN signature included genes that regulate chromatin structure and transcriptional activity, many of which are transcriptional repressors. The chronic neurodegeneration-UP signature was further enriched in astrocyte-secreted extracellular matrix proteins (Lama4, Cyr61, Thbs4), while the DOWN signature included relevant genes such as Agl (glycogenolysis), S1pr1 (immune modulation), and Sod2 (anti-oxidant). Only the pan-injury-UP mouse signature was clearly present in some human neurodegenerative transcriptomic datasets. Conclusions: Acute and chronic CNS injuries lead to distinct astrocyte gene expression programs beyond their common astrocyte reaction signature. However, caution should be taken when extrapolating astrocyte transcriptomic findings from mouse models to human diseases.

Aims: Reactive astrocytes in Alzheimer's disease (AD) have traditionally been demonstrated by increased glial fibrillary acidic protein (GFAP) immunoreactivity; however, astrocyte reaction is a complex and heterogeneous phenomenon involving multiple astrocyte functions beyond cytoskeletal remodelling. To better understand astrocyte reaction in AD, we conducted a systematic review of astrocyte immunohistochemical studies in post‐mortem AD brains followed by bioinformatics analyses on the extracted reactive astrocyte markers.Methods: NCBI PubMed, APA PsycInfo and WoS‐SCIE databases were interrogated for original English research articles with the search terms ‘Alzheimer's disease’ AND ‘astrocytes.’ Bioinformatics analyses included protein–protein interaction network analysis, pathway enrichment, and transcription factor enrichment, as well as comparison with public human ‐omics datasets.Results: A total of 306 articles meeting eligibility criteria rendered 196 proteins, most of which were reported to be upregulated in AD vs control brains. Besides cytoskeletal remodelling (e.g., GFAP), bioinformatics analyses revealed a wide range of functional alterations including neuroinflammation (e.g., IL6, MAPK1/3/8 and TNF), oxidative stress and antioxidant defence (e.g., MT1A/2A, NFE2L2, NOS1/2/3, PRDX6 and SOD1/2), lipid metabolism (e.g., APOE, CLU and LRP1), proteostasis (e.g., cathepsins, CRYAB and HSPB1/2/6/8), extracellular matrix organisation (e.g., CD44, MMP1/3 and SERPINA3), and neurotransmission (e.g., CHRNA7, GABA, GLUL, GRM5, MAOB and SLC1A2), among others. CTCF and ESR1 emerged as potential transcription factors driving these changes. Comparison with published ‐omics datasets validated our results, demonstrating a significant overlap with reported transcriptomic and proteomic changes in AD brains and/or CSF.Conclusions: Our systematic review of the neuropathological literature reveals the complexity of AD reactive astrogliosis. We have shared these findings as an online resource available at www.astrocyteatlas.org.

Background: Electronic health records (EHRs) with large sample sizes and rich information offer great potential for dementia research, but current methods of phenotyping cognitive status are not scalable. Objective: The aim of this study was to evaluate whether natural language processing (NLP)–powered semiautomated annotation can improve the speed and interrater reliability of chart reviews for phenotyping cognitive status. Methods: In this diagnostic study, we developed and evaluated a semiautomated NLP-powered annotation tool (NAT) to facilitate phenotyping of cognitive status. Clinical experts adjudicated the cognitive status of 627 patients at Mass General Brigham (MGB) health care, using NAT or traditional chart reviews. Patient charts contained EHR data from two data sets: (1) records from January 1, 2017, to December 31, 2018, for 100 Medicare beneficiaries from the MGB Accountable Care Organization and (2) records from 2 years prior to COVID-19 diagnosis to the date of COVID-19 diagnosis for 527 MGB patients. All EHR data from the relevant period were extracted; diagnosis codes, medications, and laboratory test values were processed and summarized; clinical notes were processed through an NLP pipeline; and a web tool was developed to present an integrated view of all data. Cognitive status was rated as cognitively normal, cognitively impaired, or undetermined. Assessment time and interrater agreement of NAT compared to manual chart reviews for cognitive status phenotyping was evaluated. Results: NAT adjudication provided higher interrater agreement (Cohen κ=0.89 vs κ=0.80) and significant speed up (time difference mean 1.4, SD 1.3 minutes; P<.001; ratio median 2.2, min-max 0.4-20) over manual chart reviews. There was moderate agreement with manual chart reviews (Cohen κ=0.67). In the cases that exhibited disagreement with manual chart reviews, NAT adjudication was able to produce assessments that had broader clinical consensus due to its integrated view of highlighted relevant information and semiautomated NLP features. Conclusions: NAT adjudication improves the speed and interrater reliability for phenotyping cognitive status compared to manual chart reviews. This study underscores the potential of an NLP-based clinically adjudicated method to build large-scale dementia research cohorts from EHRs.

There are over 7,000 rare diseases, some affecting 3,500 or fewer patients in the US. Due to clinicians’ limited experience with such diseases and the heterogeneity of clinical presentations, approximately 70% of individuals seeking a diagnosis remain undiagnosed. Deep learning has demonstrated success in aiding the diagnosis of common diseases. However, existing approaches require labeled datasets with thousands of diagnosed patients per disease. We present SHEPHERD, a few shot learning approach for multi-faceted rare disease diagnosis. SHEPHERD performs deep learning over a knowledge graph enriched with rare disease information and is trained primarily on simulated rare disease patients. We demonstrate SHEPHERD’s effectiveness across diverse diagnostic tasks, performing causal gene discovery, retrieving “patients-like-me”, and characterizing novel disease presentations, using real-world cohorts from the Undiagnosed Diseases Network ( N = 465), MyGene2 ( N = 146), and the Deciphering Developmental Disorders Study ( N = 1, 431). SHEPHERD demonstrates the potential of knowledge-guided deep learning to accelerate rare disease diagnosis.