Several neurodegenerative diseases are triggered by proteins containing a polyglutamine (polyQ) stretch expanded beyond a critical threshold. Among these, ataxin-3 (AT3) is the causative agent of spinocerebellar ataxia type-3. We expressed three authentic AT3 variants in Escherichia coli: one normal (AT3-Q24), one expanded (AT3-Q55) and one truncated immediately upstream of the polyQ (AT3-291Δ). Then, based on growth rate reduction, we quantified protein toxicity. We show that AT3-Q55 and -291Δ strongly reduced the growth rate in the early stages (2-4 h), unlike AT3-Q24. This correlated well with the appearance of soluble cytosolic oligomers, but not with the amount of insoluble protein in inclusion bodies (IBs). The impact of AT3-291Δ on cell growth suggests an intrinsic toxicity of the AT3 fragment. Besides the typical Fourier Transform Infrared Spectroscopy (FTIR) signal for intermolecular β-sheets, the expanded form displayed an additional infrared signature, which was assigned to glutamine side-chain hydrogen bonding and associated with SDS-insoluble fibrils. The elongation of the latter was monitored by Atomic Force Microscopy (AFM). This mirrors the well-known in vitro two-step aggregation pattern of expanded AT3. We also demonstrated that final aggregates of strains expressing expanded or truncated AT3 play a protective role against toxicity. Furthermore, our findings suggest that the mechanisms of toxicity are evolutionarily conserved.