Dependence of deacetylation rate on substrate concentration for two representative peptides catalyzed by HDAC6, measured using the acetate assay. The initial velocity for each substrate concentration was determined from a linear regression of a time course consisting of a minimum of three timepoints, standard error is shown. The kinetic parameters are determined from a nonlinear least square fit of the Michaelis–Menten equation to the data and are listed in Table 1. Black squares: example of data which met the criteria to produce accurate k_cat/K_M values. Green circles: example of overfit data resulting in calculation of a lower limit for k_cat/K_M, due to the K_M being lower than the detection limit of the assay.

Performance of the calibrated protocol on different datasets (see Supplementary Table S2). (A) Correlation: Predicted vs. measured activities on D-TRAINING (dots, blue: substrates, red: non-substrates); D-CAPPED (yellow triangles); and D-TEST (grey squares) datasets. (B) Binary distinction: ROC curves on D-TRAINING (magenta), D-EXTENDED (cyan) and D-HPLC (brown). (C) Performance of protocol on D-HPLC, a dataset measured using HPLC (from Kutil et al.³). In (A) and (C), a dashed horizontal line denotes the cutoff dividing measured substrates and non-substrates (k_cat/K_M=10⁴ M⁻¹ s⁻¹). Note the different scales in (A) and (C). The cutoffs dividing predicted substrates from not substrates are indicated as dotted vertical line for the loose cutoff (reweighted score: − 1105) and a dotted-dashed vertical line for the strict cutoff (reweighted score: − 1118).

Application of the calibrated protocol to the acetylome to detect novel potential HDAC6 substrates. (A) Blue dots and yellow dots show the distribution of scores for acetylated peptides in the D-TRAINING set and the acetylome peptides (as annotated in the PhosphoSitePlus database), respectively. Boxplots are drawn between the 1st and 3rd quartiles of the data and whiskers extend by 1.5 times the interquartile range. (B, C) Sequence logos of (B) substrates from the D-EXTENDED set, and (C) top 100 peptides predicted by our protocol.

Different datasets of experimentally determined HDAC6 substrates show minor overlap and agreement. Comparison of substrate specificities of different experimental datasets, based on their sequence logos and correlation of substrate activities. (A) Plot of substrate activities for peptides shared between the D-13MER³ and D-SILAC¹⁴ datasets. (blue, red: predicted substrates and non-substrates by our protocol; triangles indicate peptides with H/L ratio > 3 in the SILAC experiment, or measured intensity > 2 in the peptide array, for which the values have been capped to fit into the plot; dashed lines indicate cutoffs for defining substrates and non-substrates; S, NS: substrates and non-substrates, respectively) (B–F) Sequence logos made with PSSMSearch⁸⁹ using its default background containing human sequences (B–D), and with Two Sample Logo⁹⁰ using the non-substrate peptides as background for each dataset (E–G). (B, E) D-13MER, (C, F) D-SILAC, and (D, G) D-3MER datasets.

Key differences between HDAC6 and HDAC8 in the coordination of the substrate. The active sites of HDAC6 (receptor from PDB ID 6WSJ, substrate from 5EFN for visualization purposes, orange) and HDAC8 (PDB ID 2V5W, teal) are overlaid. The loops that distinguish the two HDACs are highlighted (orange and teal), together with conformations generated for modeling substrate activity (grey). Residues S531 (HDAC6) and D101 (HDAC8) that coordinate hydrogen bonding (light blue: HDAC6, yellow: HDAC8) to the backbone of the acetylated substrate lysine residue, are also highlighted, as well as positions W459 (HDAC6) and Y100 (HDAC8). The hydroxamic acid inhibitors are shown in sticks, and the catalytic Zn2+ ions are shown as a white sphere. See Text for more details.