Huth, A. | Search bioRxiv

bioRxiv2018

Antigen-specific T-cell receptor signatures of cytomegalovirus infection

Cytomegalovirus (CMV) is a prevalent human pathogen. The virus cannot be eliminated from the body, but is kept in check by CMV-specific T cells. Patients with an insufficient T-cell response, such as transplant recipients, are at high risk of developing CMV disease. However, the CMV-specific T-cell repertoire is complex, and is not yet clear which T cells protect best against virus reactivation and disease. Here we present a highly resolved characterization of CMV-specific CD8+ T cells based on enrichment by specific peptide stimulation and mRNA sequencing of their T-cell receptor {beta} chains (TCR{beta}). Our analysis included recently identified T-cell epitopes restricted through HLA-C, whose presentation is resistant to viral immunomodulation, and well-studied HLA-B-restricted epitopes. In 8 healthy virus carriers, we identified a total of 1052 CMV-specific TCR{beta} chains. HLA-C-restricted, CMV-specific TCR{beta} clonotypes the ex vivo T-cell response, and contributed the highest-frequency clonotype of the entire repertoire in 2 of 8 donors. We analyzed sharing and similarity of CMV-specific TCR{beta} sequences and identified 63 public or related sequences belonging to 17 public TCR{beta} families. In our cohort and in an independent cohort of 352 donors, the cumulative frequency of these public TCR{beta} family members was a highly discriminatory indicator of carrying both CMV infection and the relevant HLA type. Based on these findings, we propose CMV-specific TCR{beta} signatures as a biomarker for an antiviral T-cell response to identify patients in need of treatment and to guide future development of immunotherapy.

immunology↗

bioRxiv2018

Incorporating Context into Language Encoding Models for fMRI

Language encoding models help explain language processing in the human brain by learning functions that predict brain responses from the language stimuli that elicited them. Current word embedding-based approaches treat each stimulus word independently and thus ignore the influence of context on language understanding. In this work, we instead build encoding models using rich contextual representations derived from an LSTM language model. Our models show a significant improvement in encoding performance relative to state-of-the-art embeddings in nearly every brain area. By varying the amount of context used in the models and providing the models with distorted context, we show that this improvement is due to a combination of better word embeddings learned by the LSTM language model and contextual information. We are also able to use our models to map context sensitivity across the cortex. These results suggest that LSTM language models learn high-level representations that are related to representations in the human brain.

neuroscience↗