The expansion of an unstable CAG repeat causes spinocerebellar ataxia type 1 (SCA1) and several other neurodegenerative diseases. How polyglutamine expansions render the resulting proteins toxic to neurons, however, remains elusive. Hypothesizing that long polyglutamine tracts alter gene expression, we found certain neuronal genes involved in signal transduction and calcium homeostasis sequentially downregulated in SCA1 mice. These genes were abundant in Purkinje cells, the primary site of SCA1 pathogenesis; moreover, their downregulation was mediated by expanded ataxin-1 and occurred before detectable pathology. Similar downregulation occurred in SCA1 human tissues. Altered gene expression may be the earliest mediator of polyglutamine toxicity.

PQBP-1 was isolated on the basis of its interaction with polyglutamine tracts. In this study, using in vitro and in vivo assays, we show that the association between ataxin-1 and PQBP-1 is positively influenced by expanded polyglutamine sequences. In cell lines, interaction between the two molecules induces apoptotic cell death. As a possible mechanism underlying this phenomenon, we found that mutant ataxin-1 enhances binding of PQBP-1 to the C-terminal domain of RNA polymerase II large subunit (Pol II). This reduces the level of phosphorylated Pol II and transcription. Our results suggest the involvement of PQBP-1 in the pathology of spinocerebellar ataxia type 1 (SCA1) and support the idea that modified transcription underlies polyglutamine-mediated pathology.

Multi-hop reasoning is an effective approach for query answering (QA) over incomplete knowledge graphs (KGs). The problem can be formulated in a reinforcement learning (RL) setup, where a policy-based agent sequentially extends its inference path until it reaches a target. However, in an incomplete KG environment, the agent receives low-quality rewards corrupted by false negatives in the training data, which harms generalization at test time. Furthermore, since no golden action sequence is used for training, the agent can be misled by spurious search trajectories that incidentally lead to the correct answer. We propose two modeling advances to address both issues: (1) we reduce the impact of false negative supervision by adopting a pretrained onehop embedding model to estimate the reward of unobserved facts; (2) we counter the sensitivity to spurious paths of on-policy RL by forcing the agent to explore a diverse set of paths using randomly generated edge masks. Our approach significantly improves over existing path-based KGQA models on several benchmark datasets and is comparable or better than embedding-based models.