CNS neurons are endowed with the ability to recover from cytotoxic insults associated with the accumulation of proteinaceous aggregates in mouse models of polyglutamine disease, but the cellular mechanism underlying this phenomenon is unknown. Here, we show that autophagy is essential for the elimination of aggregated forms of mutant huntingtin and ataxin-1 from the cytoplasmic but not nuclear compartments. Human orthologs of yeast autophagy genes, molecular determinants of autophagic vacuole formation, are recruited to cytoplasmic but not nuclear inclusion bodies in vitro and in vivo. These data indicate that autophagy is a critical component of the cellular clearance of toxic protein aggregates and may help to explain why protein aggregates are more toxic when directed to the nucleus.autophagy ͉ huntingtin ͉ Huntington's disease P olyglutamine (polyQ) expansion disorders, such as Huntington's disease (HD) and spinocerebellar ataxia type 1, are caused by the production of mutant proteins, huntingtin (Htt) and ataxin-1 (Atx1), respectively, that adopt cytotoxic, nonnative, oligomeric, or aggregation-prone conformations because of the presence of long homopolymeric tracts of glutamine (1). The length of polyQ tracts in normal humans is polymorphic but always below a threshold of 35-40. PolyQ repeats above this threshold are invariably associated with disease, with a strong inverse correlation between repeat length and age-of-onset of neurological disease (2). These findings, together with studies on Htt fragments expressed in cell culture (3) or polyQ peptides in vitro (4), support a model in which glutamine repeats above the threshold adopt a nonnative conformation that is highly prone to self-associate into high-molecular-weight, stable aggregates. These aggregates accumulate in nuclear or cytoplasmic inclusion bodies (IBs) that are invariably associated with end-stage neurodegenerative disease in patients and animal models (5, 6). Although IBs are probably not directly responsible for cellular toxicity (5) and may even contribute to cytoprotection (7,8), the toxicity of mutant polyQ-expanded proteins appears to depend strongly on the cellular compartment in which they accumulate (9, 10). Experimental redirection of Atx1, normally a nuclear protein, to the cytoplasm, drastically reduces toxicity and IB formation in a mouse model of spinocerebellar ataxia type 1 (11). Conversely, the addition of a nuclear localization sequence to the N-terminal polyQ-containing fragment of Htt, which normally partitions between the nucleus and cytoplasm, directs it to the nucleus and increases its toxicity (12, 13). These experiments have been interpreted to indicate that the primary target of polyQ toxicity is in the nucleus. However, it also is possible that both the nucleus and cytoplasm contain targets that are equally vulnerable to these toxic proteins, but that the cytoplasm may be better endowed to neutralize or destroy the proteotoxic agent.Although protein aggregates are highly insoluble (14), animal models using condit...