Covalent attachment of pyridoxal-phosphate derivatives to 14-3-3 proteins

Small-molecule inhibitors of 14-3-3 protein–protein interactions could serve as valuable chemical biology tools and starting points for drug development. The article by Zhao et al. (1) described FOBISIN101 (1) (Fig. 1A), a pyridoxal-phosphate (PLP) derivative that inhibits 14-3-3 protein–protein interactions. The authors provided a crystal structure of 14-3-3ζ with PLP covalently bound to K120 [Protein Data Bank (PDB) ID code 3RDH] lacking the p-amino-benzoate moiety. As a mechanism, they propose X-ray–induced cleavage of the N=N bond. They suggest this mechanism could serve as a radiation-triggered anticancer prodrug concept.

Fig. 1.

(A) PLP derivatives 1 (FOBISIN101) and 2. The PLP part is highlighted in red. (B) Deconvoluted mass spectra of 14-3-3ζ incubated with FOBISIN101. A mixture of 14-3-3ζ (1 mM) with FOBISIN101 (5 mM) in the presence of tris(2-carboxyethyl)phosphine (25 mM) was incubated overnight and analyzed by using ESI MS after 1:50 dilution with water (blue). An aliquot of the initial mixture was also immediately diluted (1:50 in water) followed by overnight incubation and ESI MS analysis (orange). The resulting mass spectra were deconvoluted by using Mag-Tran. Red marks show the calculated masses for 14-3-3ζ with zero to four FOBISIN101 moieties attached via imine linkage. (C) Crystal structure of 2 covalently bound to K122 of 14-3-3σ at 1.8 Å resolution. [Blue mesh, simulated annealing composite-omit map calculated with PHENIX (5) at 1.3 σ contour level; green mesh, mF_o-DF_c density map at 3.0 σ contour level.] The mF_o-DF_c density map has been calculated after protein refinement without ligand bias. (D and E) Crystal structure of 1 covalently bound to K122 of 14-3-3σ at 1.65 Å resolution (meshes and coloring according to C). The mF_o-DF_c density map in D has been calculated directly after molecular replacement using 14-3-3σ as search model without any refinement. The mF_o-DF_c density map in E has been calculated after protein refinement without ligand bias. Software used was XDS for data processing, PHASER for molecular replacement, REFMAC and Coot for refinement, and PyMOL for figures.

Table 1.

Data collection and refinement statistics

Compound	PLP derivative 1 (FOBISIN101)	PLP derivative 2
Crystal parameters
Cell dimensions
a, b, c, Å	82.25, 112.29, 62.73	82.22, 112.14, 62,66
α, β, γ, °	90.0, 90.0, 90.0	90.0, 90.0, 90.0
Space group	C2221	C2221
Data collection
Wavelength, Å	1.5418	1.5418
Resolution, Å	19.60–1.65 (1.75–1.65)	45.54–1.80 (2.0–1.8)
Measured reflections	191,999 (21,100)	157,300 (26,966)
Unique reflections	34,969 (5,475)	26,966 (7,058)
Completeness	99.1 (97.6)	99.0 (97.0)
Redundancy	9.1 (6.4)	5.8 (3.8)
I/σ(I)	20.05 (4.57)	35.76 (16.03)
Rmeas, %	6.3 (38.6)	4.9 (9.0)
Refinement
Resolution, Å	19.60–1.65 (1.69–1.65)	45.54–1.80 (1.85–1.80)
No. of atoms	2,511	2,548
Rwork, %	14.44 (18.40)	13.56 (17.30)
Rfree, %	19.13 (25.40)	18.81 (22.60)
rmsd bond lengths, Å	0.021	0.017
rmsd bond angles, °	1.972	1.584
Averaged B-factors, Å2
Total	19.60	17.63
Ligand without lysine moiety	32.24	21.38
Residues in Ramachandran plot
Favored regions, %	98.2	96.0
Additional allowed regions, %	1.8	4.0
Generously allowed regions, %	0.0	0.0
Disallowed regions, %	0.0	0.0

Data for outermost resolution shells are in parentheses. rmsds are from ideal geometry values. R_meas, redundancy-independent R-factor (intensities).

Our research on 14-3-3 inhibitors also revealed a related PLP derivative (Fig. 1A). By means of soaking, we solved a crystal structure of 14-3-3σ with this compound to 1.8 Å resolution (crystallization details in ref. 2; PDB ID code 3U9X; Fig. 1C), whereby PLP is covalently bound to a lysine as well (K122, corresponding to K120 of 14-3-3ζ). A likely mechanism involves hydrolysis of the imine bond to a PLP intermediate followed by imine formation with the amine group of K122. The resulting N6-PLP-L-lysine residue is well documented in the literature and is currently present in 229 PDB structures [as of October 20, 2011; Proteomics Standards Initiative Modifications Ontology (PSI-MOD) 128]. It plays a key role in most—if not all—of the numerous PLP-dependent enzymes (3). Furthermore, in plasma, more than 95% of circulating PLP is reversibly linked to a lysine residue in human serum albumin, serving as a reservoir and mediating PLP transport. Thus, we arrived at the conclusion that the crystal structure described by Zhao et al. (1) deviated from the common mode of action of this compound class.

We wondered why FOBISIN101, which already features the reactive aldehyde, should reside in the phosphate binding pocket until X-ray exposure, rather than directly form the imine as well. To investigate this, we obtained a sample of FOBISIN101, soaked it into our 14-3-3σ crystals, and solved the complex structure to 1.65 Å resolution (Fig. D and E). The refined model reveals two important features: first, the substitution pattern of the PLP moiety bound to K122 unambiguously supports our imine hypothesis, and second, we found interpretable density for the p-amino-benzoate moiety despite its high flexibility. Furthermore, the density hints at a second molecule bound to nearby K49. By using electrospray ionization (ESI) MS, we also verified imine formation between 14-3-3ζ and FOBISIN101 in solution (Fig. 1B). Overnight incubation [1 mM 14-3-3ζ, 5 mM FOBISIN101, 25 mM tris(2-carboxyethyl)phosphine] followed by dilution reveals two-, three-, and four-compound molecules attached to the protein, whereas dilution followed by incubation shows the protein carrying zero, one, and two modifications. We thus speculate that the fragmentation observed in MALDI-MS might be a laser-induced artifact, as the absorption maximum of PLP is very close to the laser wavelength (4).

We wish to note that we do not challenge the results and FOBISIN101 as 14-3-3 inhibiting molecule in general. However, our findings unfortunately imply that the proposed radiation-triggered mechanism of the FOBISIN101 action seems to be unlikely.