Effects of Inhibitors on Hsp90's Conformational Dynamics, Cochaperone and Client Interactions

Abstract

The molecular chaperone and heat-shock protein Hsp90 has become a central target in anti-cancer therapy. Nevertheless, the effect of Hsp90 inhibition is still not understood at the molecular level, preventing a truly rational drug design. Here we report on the effect of the most prominent drug candidates, namely, radicicol, geldanamycin, derivatives of purine, and novobiocin, on Hsp90's characteristic conformational dynamics and the binding of three interaction partners. Unexpectedly, the global opening and closing transitions are hardly affected by Hsp90 inhibitors. Moreover, we find no significant changes in the binding of the cochaperones Aha1 and p23 nor of the model substrate Δ131Δ. This holds true for competitive and allosteric inhibitors. Therefore, direct inhibition mechanisms affecting only one molecular interaction are unlikely. We suggest that the inhibitory action observed in vivo is caused by a combination of subtle effects, which can be used in the search for novel Hsp90 inhibition mechanisms.

Keywords: FRET; chaperone Hsp90; inhibitor; protein conformational dynamics; single molecule.

Figure 1

(A) Illustration of the smFRET experiment revealing the effect of drug candidates on Hsp90's conformational dynamics. Single molecule FRET between the attached fluorescent dyes allows us to distinguish open and closed conformations of Hsp90. To follow one molecule for minutes, the dimers are stabilized by a C-terminal zipper motif (see Experimental Section) and immobilized on a polyethylene glycol (PEG) passivated coverslip using biotin-neutravidin coupling. Fluorescence intensities of individual dyes are recorded by total internal reflection (TIR) fluorescence microscopy. Schematic laser rays are depicted as green lines. The evanescent excitation intensity is shown in fading green. See example data in Figure 2A. (B) Solution structure of the open conformation of Hsp90[14] in gray with nucleotides as colored spheres. The binding sites of the competitive (green circles), and c-terminal Novobiocin-derived inhibitors[10] (green square) are indicated.

Figure 2

(A) smFRET trajectories show conformational dynamics in the presence of the indicated inhibitor. Fluorescence intensities of individual FRET donor and acceptor dyes coupled to Hsp90 are shown as green and orange lines, respectively. The resulting FRET efficiency (E) is shown in black. Hsp90's closed and open conformations are indicated as white and colored overlays, respectively, given by the Viterbi path [22]. (B) FRET histograms (normalized to unity) show that open conformations (low FRET) prevail under all conditions, except for the non-hydrolysable nucleotide analogue AMP-PNP, which stabilizes the closed conformation. Blue ribbons highlight the expected FRET efficiencies of the indicated open and closed conformations. n(ADP)=107, n(geldanamycin)=108, n(radicicol)=142, n(PU-H71)=123, n(KU-32)=65, n(AMP-PNP)=104. On average, the traces were 150-200 frames long, corresponding to 40 seconds. For clarity we omitted the directly excited acceptor trace. It can be found in Supplementary Figure S3, along with an AMP-PNP example trace.

Figure 3

Kinetic model of Hsp90’s conformational dynamics. (A) Schematic consensus model under all tested inhibitor conditions, as well as ADP. (B) Quantitative rate constants and confidence intervals of the transitions in (A). Gray areas represent the 95% confidence intervals in the presence of ADP. Significantly different rate constants were found with AMP-PNP (black), but none of the inhibitors changed the kinetics in a similar way. (C) Same data as in (B) plotted in a linear scale.

Figure 4

Hsp90 inhibitors show no significant effect on the binding of the cochaperones Aha1 and p23 or the model client Δ131Δ in fluorescence anisotropy experiments. The fluorescence anisotropy of labeled (A) Aha1, (B) p23 or (C) Δ131Δ is shown in absence or presence of the indicated inhibitor, or DMSO as a control. Error bars are the SD of the 31 data points that were taken for each sample.