Analysis was similar to Uversky et al. (2000), the differences being (i) that we considered not the net charge, but the total fraction of charged residues and (ii) we used a more strictly defined hydrophobicity scale that is not biased by globular protein structures. For each protein sequence, the mean fraction of charged residues was determined by counting D, E, K, and R and dividing this sum by the sequence length. Mean hydrophobicity was calculated according to a scale based on partitioning of Nα-acetyl-amino acid amides between 1-octanol and water at neutral pH (given in Table 2 in Fauchere and Pliska, 1983). For clarity, we re-scaled their numbers linearly to range between 0 and 1, and thus used the following parameters: (R, 0); (K, 0.006); (D, 0.012); (E, 0.113); (N, 0.132); (Q, 0.242); (S, 0.298); (G, 0.31); (H, 0.35); (T, 0.39); (A, 0.405); (P, 0.46); (Y, 0.604); (V, 0.684); (M, 0.687); (C, 0.782); (L, 0.831); (F, 0.859); (I, 0.862); (W, 1). The brightness in the heat maps reflects the number of proteins in a given regime of the plot. (A) 667 intrinsically disordered protein (IDP) regions were extracted from the DisProt database (Sickmeier et al., 2007) and analysed as described above. Note their wide distribution in the plot, their high content of charges residues and low hydrophobicity. (B) Analysis of 34,551 non-redundant protein sequence entries from the PDB (Bernstein et al., 1978), representing mostly globular, folded proteins. Note that these sequences are on average less charged and considerably more hydrophobic than the bulk of IDPs. (C) Analysis of the 666 identified Nup98 FG domains (excluding the GLEBS domain). Despite also being intrinsically disordered, they strongly cluster in a very narrow region with extremely very low charge density and a hydrophobicity very close to globular proteins. The few outliers with slightly less hydrophobicity represent NQ-rich sequences and reflect the facts (i) that N and Q are more hydrophilic than other typical inter-FG spacer residues (A, T, S, G, P) and (ii) probably that NQ-rich stretches contribute to cohesiveness by conferring very hydrophilic (cross-β) contacts (Ader et al., 2010). (D) For direct comparison, plots A (in red), B (in blue), and C (in green) were overlaid in a single plot.