The protein ataxin-3 is responsible for Machado-Joseph disease/spinocerebellar ataxia type 3, a neurodegenerative disorder caused by the presence of an expanded polyglutamine tract. A previous investigation [Bevivino, A. E., and Loll, P. J. (2001) Proc. Natl. Acad. Sci. U.S.A. 98, 11955-11960] showed that a nonexpanded ataxin-3 (Q27) was fully soluble, whereas an expanded form (Q78) gave rise to amyloid fibrils. Here, we report investigations on three forms of ataxin-3 (i.e., human nonexpanded (Q26), moderately expanded (Q36) ataxins-3, and the murine protein (Q6)). Far-UV circular dichroism spectra at room temperature were substantially similar, with a relatively high helical content. On heating to 96 degrees C, human Q26 and murine proteins did not display large structural changes, nor did they undergo any precipitation, which highlights their amazing heat-resistance. In contrast, human Q36 ataxin-3 underwent a progressive increase in the beta-sheet and a concomitant decrease in helical content when the temperature was shifted from 37 to 80 degrees C, followed by the irreversible formation of aggregates above 80 degrees C. They were shown to consist of amyloid fibrils, as supported by both electron microscopy images and the typical spectral shift displayed by Congo red when it was added to the protein at growing temperatures. We also found that protein precipitation could be prevented by mixing the dye with Q36 ataxin-3 prior to heating, which also confirms that the precipitates do represent authentic amyloid fibrils. In contrast, other compounds structurally related to Congo red did not exert significant effects. Our observations suggest that the temperature of the observed transition is inversely related to the length of the expansion. Finally, we suggest that antiamyloidogenic compounds might be selected on the basis of their ability to block or retard human Q36 ataxin-3 precipitation on heat-treatment.