No abstract

Epidemiological and experimental studies suggest that a disease-aggravating neuroinflammatory process is present at preclinical stages of Alzheimer’s disease. Given that individuals with Down syndrome are at increased genetic risk of Alzheimer’s disease and therefore develop the spectrum of Alzheimer’s neuropathology in a uniform manner, they constitute an important population to study the evolution of neuroinflammation across the Alzheimer’s continuum. Therefore, in this cross-sectional study, we characterized the brain inflammatory profile across the lifespan of individuals with Down syndrome. Microglial morphology and inflammatory cytokine expression were analysed by immunohistochemistry and electrochemiluminescent-based immunoassays in the frontal cortex from foetuses to adults with Down syndrome and control subjects (16 gestational weeks to 64 years), totalling 127 cases. Cytokine expression in mixed foetal primary cultures and hippocampus of adults with Down syndrome, as well as the effects of sex on cytokine expression were also analysed. A higher microglial soma size-to-process length ratio was observed in the frontal cortex of children and young adults with Down syndrome before the development of full-blown Alzheimer’s pathology. Moreover, young adults with Down syndrome also displayed increased numbers of rod-like microglia. Increased levels of interleukin-8 and interleukin-10 were observed in children with Down syndrome (1–10 years; Down syndrome n = 5, controls n = 10) and higher levels of interleukin-1β, interleukin-1α, interleukin-6, interleukin-8, interleukin-10, interleukin-15, eotaxin-3, interferon gamma-induced protein 10, macrophage-derived chemokine, and macrophage inflammatory protein-beta, were found in young adults with Down syndrome compared to euploid cases (13–25 years, Down syndrome n = 6, controls n = 24). Increased cytokine expression was also found in the conditioned media of mixed cortical primary cultures from second trimester foetuses with Down syndrome (Down syndrome n = 7, controls n = 7). Older adults with Down syndrome (39–68 years, Down syndrome n = 22, controls n = 16) displayed reduced levels of interleukin-10, interleukin-12p40, interferon-gamma and tumour necrosis factor-alpha. Microglia displayed larger somas and shorter processes. Moreover, an increase in dystrophic microglia and rod-like microglia aligning to neurons harbouring tau pathology were also observed. Sex stratification analyses revealed that females with Down syndrome had increased interleukin-6 and interleukin-8 levels compared to males with Down syndrome. Finally, multivariate projection methods identified specific cytokine patterns among individuals with Down syndrome. Our findings indicate the presence of an early and evolving neuroinflammatory phenotype across the lifespan in Down syndrome, a knowledge that is relevant for the discovery of stage-specific targets and for the design of possible anti-inflammatory trials against Alzheimer’s disease in this population.

One reason pretraining on self-supervised linguistic tasks is effective is that it teaches models features that are helpful for language understanding. However, we want pretrained models to learn not only to represent linguistic features, but also to use those features preferentially during fine-turning. With this goal in mind, we introduce a new English-language diagnostic set called MSGS (the Mixed Signals Generalization Set), which consists of 20 ambiguous binary classification tasks that we use to test whether a pretrained model prefers linguistic or surface generalizations during finetuning. We pretrain RoBERTa models from scratch on quantities of data ranging from 1M to 1B words and compare their performance on MSGS to the publicly available RoBERTa BASE . We find that models can learn to represent linguistic features with little pretraining data, but require far more data to learn to prefer linguistic generalizations over surface ones. Eventually, with about 30B words of pretraining data, RoBERTa BASE does demonstrate a linguistic bias with some regularity. We conclude that while self-supervised pretraining is an effective way to learn helpful inductive biases, there is likely room to improve the rate at which models learn which features matter.

(1) Background: Recent studies have reported elevated risks of multiple cancers in the World Trade Center (WTC) affected community members (also called WTC “Survivors”). The large variety of WTC-cancers created a need to develop a comprehensive cancer database. This paper describes the development of a pan-cancer database at the WTC Environmental Health Center (EHC) Data Center. (2) Methods: A new REDCap-based pan-cancer database was created using the pathology reports and available biomarker data of confirmed cancer cases after review by a cancer epidemiologist, a pathologist, physicians and biostatisticians. (3) Results: The WTC EHC pan-cancer database contains cancer characteristics and emerging biomarker information for cancers of individuals enrolled in the WTC EHC and diagnosed after 11 September 2001 and up to 31 December 2019 obtained from WTC EHC clinical records, pathological reports and state cancer registries. As of 31 December 2019, the database included 3440 cancer cases with cancer characteristics and biomarker information. (4) Conclusions: This evolving database represents an important resource for the scientific community facilitating future research about the etiology, heterogeneity, characteristics and outcomes of cancers and comorbid mental health conditions, cancer economics and gene–environment interaction in the unique population of WTC survivors.