Novel mutation in hexokinase 2 confers resistance to 2-deoxyglucose by altering protein dynamics

Glucose is central to many biological processes, serving as an energy source and a building block for biosynthesis. After glucose enters the cell, hexokinases convert it to glucose-6-phosphate (Glc-6P) for use in anaerobic fermentation, aerobic oxidative phosphorylation, and the pentose-phosphate pathway. We here describe a genetic screen in Saccharomyces cerevisiae that generated a novel spontaneous mutation in hexokinase-2, hxk2G238V, that confers resistance to the toxic glucose analog 2-deoxyglucose (2DG). Wild-type hexokinases convert 2DG to 2-deoxyglucose-6-phosphate (2DG-6P), but 2DG-6P cannot support downstream glycolysis, resulting in a cellular starvation-like response. Curiously, though the hxk2G238V mutation encodes a loss-of-function allele, the affected amino acid does not interact directly with bound glucose, 2DG, or ATP. Molecular dynamics simulations suggest that Hxk2G238V impedes sugar binding by altering the protein dynamics of the glucose-binding cleft, as well as the large-scale domain-closure motions required for catalysis. These findings shed new light on Hxk2 dynamics and highlight how allosteric changes can influence catalysis, providing new structural insights into this critical regulator of carbohydrate metabolism. Given that hexokinases are upregulated in some cancers and that 2DG and its derivatives have been studied in anti-cancer trials, the present work also provides insights that may apply to cancer biology and drug resistance.


G238V mutation impacts glucose phosphorylation and ATP hydrolysis
Hxk2 G238V adversely impacts both glucose phosphorylation and ATP hydrolysis (Table 2), as measured in enzymatic assays of Hxk2 function (see Materials and Methods). The Michaelis-Menton kinetic plots are provided for three replicate experiments. Below each plot, we show the Km and Vmax values, where the Vmax is determined based on arbitrary units of enzyme added from yeast protein extracts. These data support the values given in Table 2. Note that the Hxk2 G238V mutant has consistently higher Km values for both glucose and ATP than the Hxk2 wild-type protein. While 2DG can be phosphorylated by wild-type Hxk2, it also has a higher Km and reduced Vmax compared to glucose. Together these data demonstrate that the hxk2 G238V allele is hypomorphic, unable to phosphorylate glucose as effectively as Hxk2 and less able to use ATP. The data in these graphs support the enzymatic activity reported in Table 2. Three replicate experiments for each assay were preformed, and the results of each individual assay are presented here. In each case, total protein extracts were made from hxk1∆ hxk2∆ glk1∆ cells containing either WT Hxk2 (panels A, C and E) or Hxk2-G238V (panels D and B) and used in enzymatic assays where NADPH production is a readout for Hxk2 enzymatic function, as described in the Materials and Methods. (A) WT Hxk2 enzyme kinetics of glucose turnover were measured using 0.1 µl of yeast protein extract, which corresponds to 0.032 au of enzyme. (B) Hxk2-G238V enzyme kinetics of glucose turnover were measured using 1.0 µl of yeast protein extract, which corresponds to 0.311 au of enzyme. (C) WT Hxk2 enzyme kinetics of ATP turnover were measured using 0.2 µl of yeast protein extract, which corresponds to 0.064 au of enzyme. (D) Hxk2-G238V enzyme kinetics of ATP turnover were measured using 1.0 µl of yeast protein extract, which corresponds to 0.064 au of enzyme. (E) WT Hxk2 kinetics of 2DG turnover were measured using 0.1 µl of yeast protein extract, which corresponds to 0.032 au of enzyme. A similar assay with Hxk2-G238V and 2DG did not yield values above the lower limits of detection for this assay.

Location
Residue apo ΔB-factor holo ΔB-factor β9/β10 β-hairpin I231 The differences in RMSF and DCC values (ΔRMSF and ΔDCC, respectively), followed by the corresponding raw WT and Hxk2 G238V values in parentheses. Negative and positive ΔRMSF values suggest Hxk2 G238V has increased or decreased the flexibility of the corresponding residue, respectively. Similarly, negative and positive ΔDCC values suggest that Hxk2 G238V has increased or decreased the degree of correlation between the motions of G/V238 and the corresponding residue, respectively. For convenience, we also show the differences in calculated B-factors, derived from the RMSF values using the equation ! = (8 " 3 ⁄ ) ! " (see reference [4]). In all cases, values that deviate from the respective means by more than two standard deviations are shown in bold. All values are rounded to the nearest hundredth. See also S1 Table.

Large-scale domain-closure dynamics
Our simulations ran long enough to sample both open (apo, unbound) and closed (holo, glucose bound) conformations (Fig D in S1 Text). To assess the impact of Hxk2 G238V on the shape and volume of the enzymatic cleft, we used the POVME2 algorithm [7,8] to analyze the cleft geometries of trajectory frames spaced 100 ps apart. We used a POVME grid spacing of 1.0 Å, with inclusion and contiguous-pocket-seed spheres both centered on the bound glucose molecule (radii of 14.0 Å and 2.0 Å, respectively). To speed the calculations, we did not discard points that fell outside the protein-encompassing convex hull, nor did we consider hydrogen atoms. All other POVME parameters were identical to those given in the example file included in the POVME2 download.
We found that the Hxk2 G238V cleft volume was generally lower than that of WT Hxk2, both in the apo and holo (glucose-bound) states (p-value < 0.001 in both cases per a two sample t-test; Fig E in S1 Text). In both cases, the standard deviations associated with the Hxk2 G238V simulations (both apo and holo) were greater than the corresponding standard deviations associated with the WT Hxk2 simulations.

Ensuring the simulation had equilibrated
We aligned trajectory frames taken every ten ps to the corresponding first frame and calculated the backbone-heavy-atom RMSD. Plotting RMSD values over simulation time suggested that the simulations had not fully equilibrated during the beginning portions of the production runs (Fig F in S1 Text). We discarded the initial five ns preequilibrated portions of each simulation. Subsequent analyses focused on the remaining portions.   Table B in S1 Text similarly provides average RMSF/B-factor values across the 12 simulations.