Most eukaryote molecular phylogenies have been based on small-subunit ribosomal RNA as its database includes the most species, but serious problems have been encountered that can make these trees misleading. More recent studies using concatenated protein sequences have increased the data per organism, reducing misleading signals from a single sequence, but taxon sampling is limited. To increase the database of protein-coding genes we sequenced the cytosolic form of heat-shock protein Hsp90 from a broad variety of previously unsampled eukaryote groups: protozoan flagellates (phyla Choanozoa, Apusozoa, Cercozoa) and all three groups of chromists (Cryptophyta, Heterokonta, Haptophyta). Gamma-corrected distance trees robustly show three groups: bacterial sequences are sister to all eukaryote sequences, which are cleanly subdivided into the cytosolic sequences and a clade comprising the chloroplast and endoplasmic reticulum (ER) Hsp90 sequences. The eukaryote cytosolic sequences comprise a robust opisthokont clade (animals/Choanozoa/fungi), a bikont clade, and an amoebozoan branch. However their topology is not robust. When the cytosolic sequences are rooted using only the ER/chloroplast clade as outgroup the amoebozoan Dictyostelium is sister to the opisthokonts forming a unikont clade in the distance tree. Congruence of this tree with that for concatenated mitochondrial proteins suggests that the root of the eukaryote tree is between unikonts and bikonts. Gamma-corrected maximum likelihood analyses of cytosolic sequences alone (519 unambiguously aligned amino acid positions) show bikonts as a clade, as do least-squares distance trees, but with other distance methods and parsimony the sole amoebozoan species branches weakly within bikonts. Choanozoa are clearly sisters to animals. Some major bikont groups (e.g. green plants, alveolates, Euglenozoa) are consistently recovered, but others (e.g. discicristates, chromalveolates) appear only in some trees; the backbone of the bikont subtree is not resolved, the position of groups represented only by single sequences being particularly unclear. Although single-gene trees will probably never resolve these uncertainties, the congruence of Hsp90 trees with other data is greater than for most other molecules and further taxon sampling of this molecule is recommended.
