Structure of an HIF-1α-pVHL Complex: Hydroxyproline Recognition in Signaling

See allHide authors and affiliations

Science  07 Jun 2002:
Vol. 296, Issue 5574, pp. 1886-1889
DOI: 10.1126/science.1073440


The ubiquitination of the hypoxia-inducible factor (HIF) by the von Hippel–Lindau tumor suppressor (pVHL) plays a central role in the cellular response to changes in oxygen availability. pVHL binds to HIF only when a conserved proline in HIF is hydroxylated, a modification that is oxygen-dependent. The 1.85 angstrom structure of a 20-residue HIF-1α peptide–pVHL–ElonginB–ElonginC complex shows that HIF-1α binds to pVHL in an extended β strand–like conformation. The hydroxyproline inserts into a gap in the pVHL hydrophobic core, at a site that is a hotspot for tumorigenic mutations, with its 4-hydroxyl group recognized by buried serine and histidine residues. Although the β sheet–like interactions contribute to the stability of the complex, the hydroxyproline contacts are central to the strict specificity characteristic of signaling.

The cellular response to oxygen is a central process in animal cells and figures prominently in the pathophysiology of several diseases, including cancer, cardiovascular disease, and stroke (1). This process is coordinated by the hypoxia-inducible factor (HIF) and its regulator, the von Hippel–Lindau tumor suppressor protein (pVHL) (2, 3). HIF is a heterodimeric transcription factor that activates the expression of genes involved in angiogenesis, erythropoiesis, energy metabolism, apoptosis, and/or proliferation in response to low-oxygen tension (hypoxia) conditions. pVHL inhibits HIF activity under normal oxygen conditions (normoxia) by targeting the HIF α subunits for polyubiquitination and proteasomal degradation (4–8). Under hypoxic conditions the HIF α subunits are not recognized by pVHL, and they consequently accumulate and dimerize with HIF-1β (2, 3). VHLis mutated in the von Hippel–Lindau cancer predisposition syndrome and in sporadic clear-cell renal carcinomas, and this is associated with constitutively high levels of HIF-1α and the development of highly vascularized tumors (2).

pVHL is the substrate-recognition subunit of a ubiquitin-protein ligase that also contains ElonginB, ElonginC, Cul2, and Rbx1 (VBC-CR complex) (6–9). VBC-CR and the related Skp1−Cul1−F-box (SCF) family of ubiquitin-protein ligases (10) catalyze the transfer of ubiquitin from a ubiquitin-conjugating enzyme to specific lysine residues on the substrate (11).

pVHL binding to HIF-1α is dependent on the hydroxylation of a core proline residue (Pro564) within the HIF-1α oxygen-dependent degradation domain (ODD) (12–14). This modification is carried out by recently identified HIF prolyl hydroxylases (HPHs) only in the presence of oxygen (15–17). A 20-residue HIF-1α ODD region (destruction sequence), conserved in animal orthologs and paralogs, is necessary and sufficient for hydroxylation by HPHs and for binding to pVHL (12, 13).

To investigate the targeting of HIF-1α by the VBC-CR ubiquitin-protein ligase and the basis of hydroxyproline recognition in intracellular signaling, we determined the 1.85 Å crystal structure of the pVHL–ElonginB–ElonginC (VBC) complex bound to the hydroxyproline-containing 20-residue destruction sequence of HIF-1α (Table 1 and fig. S1). The structure shows that a 15–amino acid portion of HIF-1α (residues 561 to 575) adopts an extended, β strand–like conformation (Fig. 1A). It binds pVHL in a bipartite manner, with two discontinuous HIF-1α segments interacting with a continuous site on pVHL. A six-residue NH2-terminal segment (residues 561 to 566; N segment) that is centered on Hyp564 (Hyp is the three-letter code for hydroxyproline), and a four-residue COOH-terminal segment (residues 571 to 574; C segment) are separated by a four–amino acid bulge that does not contact pVHL (Fig. 1B). Complex formation buries a total of 1960 Å2 surface area, 63% of which is due to the N segment. The NH2-terminal five amino acids (residues 556 to 560) of the peptide are disordered.

Figure 1

The HIF-1α destruction sequence binds the β domain of pVHL in an extended β strand–like conformation. (A) Schematic representation of the 15-residue portion of the HIF-1α destruction sequence bound to the β domain of pVHL in the pVHL–ElonginB–ElonginC complex. The portion of HIF-1α that adopts a continuous β-strand conformation is indicated by a wide arrow. HIF-1α is in blue, Hyp564 in yellow, pVHL in red, ElonginB in magenta, and ElonginC in green. C, COOH-terminus; N, NH2-terminus. The figures are prepared with MOLSCRIPT (27), GL_RENDER, and POVRAY (28). (B) Alignment of the first destruction sequence in the ODDs of HIF-1α orthologs and HIF-2α and HIF-3α paralogs, highlighting residues identical in seven of the nine sequences. The N and the C segments of human HIF-1α are indicated in red. The putative N and C segments of the second destruction sequences of HIF-1α and HIF-2α orthologs are aligned below. The reported second destruction sequence is 38 residues long, and only a 23-residue region that aligns with the first destruction sequence is shown. Dashes indicate gaps relative to the seven-residue spacing between the putative N and C segments of the second destruction sequence.

Table 1

Statistics from the crystallographic analysis. Details of the crystallization and structure determination are provided in the supplementary material (22). The statistics for the outermost shell, 1.92 to 1.85 Å, are shown in parentheses. rmsd, Root mean square deviations from ideal geometry and root mean square variation in the B-factor of bonded atoms.

View this table:

pVHL has two tightly coupled domains consisting of a ∼100–amino acid β sandwich (β domain), and a ∼35–amino acid three-helix cluster (α domain) (10). The α domain binds ElonginC, which, in association with ElonginB, recruits the VBC complex to Cul2-Rbx1 to form the ubiquitin-protein ligase (8–10). HIF-1α interacts exclusively with the β domain of pVHL. It binds alongside the β sandwich, making five backbone-backbone hydrogen bonds (Fig. 1A). The side of the pVHL β sandwich where HIF-1α binds has the hydrophobic core partially exposed. This exposed hydrophobic patch, together with several partially buried polar residues, makes up the binding site of the hydroxyproline.

The hydroxyproline has a central role in complex formation. It is nearly entirely buried, with 96% of its accessible surface area in a hypothetical free peptide covered by pVHL. The pyrrolidine ring inserts toward the partially exposed hydrophobic core of the pVHL β domain, making multiple van der Waals contacts with Trp88, Tyr98, and Trp117 of pVHL (Fig. 2A). The 4-hydroxyl group inserts farthest into pVHL and forms hydrogen bonds with the Nδ of His115 (2.7 Å) and the OH group of Ser111 (2.7 Å), both of which also form hydrogen bonds with other pVHL groups (Fig. 2A). His115 and Ser111 are partially solvent exposed in the apo-VBC structure (10) but become entirely buried on HIF-1α binding. If Pro564 were not hydroxylated, HIF-1α binding would result in the desolvation of His115 and Ser111 without the replacement of the lost interactions between these side chains and water. This would be an energetically unfavorable process. His115 and Ser111 are thus likely to be key determinants of the strict requirement for the hydroxylation of Pro564.

Figure 2

The contacts made by the N segment, and in particular by Hyp564, are central to the binding of HIF-1α to pVHL. (A) Closeup view of the HIF-1α N segment–pVHL contacts. The side chains of HIF-1α and pVHL are colored in light blue and yellow, respectively. The backbones of HIF-1α and pVHL are in medium blue and red, respectively. The dotted lines indicate hydrogen bonds between the Gln671, Tyr98 Oη, His110 NH, and His110 CO groups of pVHL, and the Leu562 NH, Hyp564 CO, Tyr565 NH, and Tyr565 CO groups of HIF-1α. In pVHL, only the structural elements that make up the HIF-1α binding site are shown for clarity. (B) Surface representation of pVHL colored according to the degree of conservation in the pVHL orthologs in Fig. 1B. Yellow indicates identity in five orthologs (labeled residues), light green in four, and dark green in three. The HIF-1α side chains are in light blue, and the backbone is in medium blue. The N segment is in an orientation similar to that of (A). The approximate boundaries of the N and C segments and of the bulge are indicated. (C) Closeup view of the bulge and the C-segment area of the HIF-1α peptide–pVHL complex. The 567 P-M-D-D 571 bulge sequence does not contact pVHL, but forms an intramolecular β-turn hydrogen bond (CO of Pro567 and NH of Asp569). The Asp570 side chain of HIF-1α is omitted for clarity.

The pVHL residues that interact with Hyp564 are highly conserved in the human, mouse, frog, fly, and worm pVHL orthologs (Fig. 2B). Trp88, Tyr98, His115, and Trp117 are among the 11 β domain residues that are invariant in the five orthologs, and Ser111 is replaced with a threonine in the frog, fly, and worm (fig. S2). Mutations of Tyr98, Ser111, and Trp117 of pVHL have been shown to abolish HIF-1α binding (18).

The backbone of HIF-1α in the vicinity of Hyp564 is held in place through an extensive network of hydrogen bonds involving both backbone and side-chain groups from pVHL (Fig. 2A). These interactions are likely to increase the specificity at the hydroxyproline-binding site of pVHL by limiting the conformational flexibility of the HIF-1α backbone.

Compared with Hyp564, the other N-segment residues make significantly fewer contacts. Among them, Ile566 makes the most contacts, interacting with Pro99 and Ile109 of pVHL. Met561 packs with Phe91 of pVHL, Leu562 with Tyr112and Arg69, Ala563 with Trp88, and Tyr565 with His110 (Fig. 2A). These side-chain contacts occur on the surface of the complex, however, and are unlikely to make a major contribution to specificity and affinity compared with Hyp564. This conclusion is consistent with the observations that mutation of Met561 (12) or Leu562 (12, 16) to alanine does not affect pVHL binding, and mutation of Ala563 to glycine (16) or Tyr567 to alanine (13, 16) reduces but does not abolish binding in the context of a hydroxylated peptide. These mutations prevent the hydroxylation of HIF-1α, however, suggesting that the overall high conservation of these residues (Fig. 1B) is due to their role in HPH binding (12, 16). Similarly, the conservation of residues 556 to 560 (Fig. 1B), which are disordered in the structure, may be due to their involvement in HPH binding. That HPH binding appears to involve more side chains of HIF-1α compared with pVHL binding (12, 13, 16,18) may reflect the reliance of pVHL on the presence of the relatively rare hydroxyproline amino acid for its recognition of HIF-1α. HPHs would have to rely solely on recognizing canonical amino acids.

The N segment is followed by Pro567, which directs the chain away from pVHL. Pro567 and the next three residues form a bulge, which, although visible in the electron density map, does not contact pVHL. The bulge could, in principle, be as short as two nonproline residues and also could be substantially longer and still allow the N and C segments to interact with pVHL (supporting online text). A second, independent destruction sequence within the HIF-1α ODD (19) may interact with pVHL in a similar mode if we assume seven instead of four residues in the bulge between its putative hydroxyproline-containing N-segment– and C-segment–like sequences (Fig. 1B).

The C segment following the bulge adopts a β-strand conformation and makes three backbone-backbone hydrogen bonds, as well as side-chain contacts to pVHL (Fig. 2C). The Asp571 side chain of HIF-1α hydrogen bonds with the Arg107 side chain of pVHL, Phe572 makes van der Waals contacts to Ile75 and Gly106, and Leu574contacts Cys77 and Arg79 (Fig. 2C). Overall, these contacts are solvent exposed on the surface of the complex, and the HIF-1α and pVHL residues involved are not as conserved as those in the N-segment portion of the interface (Figs. 1B, 2B, and fig. S2). This suggests that the C-segment side-chain contacts are not as important for pVHL binding.

This conclusion is supported by the distribution of VHL tumor–derived missense mutations (20), which tend to cluster at the Hyp564 binding site of pVHL. All of the five pVHL residues that contact Hyp564 are mutated at high frequencies (Fig. 3 and fig. S2). In fact, Tyr98, whose side chain is not important for the structural integrity of the β sandwich (10, 21) but which contacts both the pyrrolidine ring and backbone carbonyl group of Hyp564, is the second most frequently mutated amino acid of pVHL (the first one, Arg167, maps to the α-β interdomain interface). In contrast, the pVHL residues that contact the HIF-1α C segment are either not mutated or mutated at low frequencies (Fig. 3 and fig. S2).

Figure 3

The Hyp564 binding site is a hotspot of tumorigenic pVHL mutations. Surface representation of the pVHL β domain colored according to the frequency of tumorigenic missense mutations. The current universal VHL-mutation database contains 363 tumor-derived missense mutations, 210 of which map to the β domain residues 60 to 153. Mutation frequencies >4% are shown in yellow, <4% and >2% in orange, <2% and >1% in red. Mutations at residues that contact Hyp564 account for 21% of the β domain mutations, while those at residues that contact the C segment account for 2%, and these are limited to the Gly104-Thr105-Gly106 sequence that makes backbone-backbone hydrogen bonds to HIF-1α. The rest map to residues that have structural roles in the hydrophobic core or in turns between β strands.

To further investigate the relative importance of the N and C segments, we used isothermal titration calorimetry (ITC) to measure the affinities of the intact, N-segment and C-segment peptides for pVHL (22). We found that the intact peptide bound pVHL with a dissociation constant (K d) of 0.22 ± 0.04 μM; the N-segment peptide bound with one-third the affinity,K d of 0.61 ± 0.10 μM (Table 2 and fig. S3). We could not detect an interaction with the nonhydroxylated intact peptide under the same conditions. The C-segment peptide exhibited binding only at very high concentrations with a K d of 536 ± 57 μM (Table 2 and fig. S3).

Table 2

Thermodynamic parameters of HIF-1α−pVHL binding.

View this table:

We next investigated the minimal portion of the N segment required for pVHL binding with a far-Western assay (18, 22). Using a peptide lacking the C segment as reference (peptide N0 in Fig. 4 and fig. S4A), we found that deletion of the NH2-terminal residues that are disordered in the crystal structure (N1 to N3) and of Met561 (N4) had no significant effect. Deletion of Leu562 (N5) reduced binding, and further deletion of Ala563 (N6) eliminated binding, possibly as a cumulative effect of eliminating the contacts made by Met561, Leu562, and Ala563. COOH-terminal deletions of the bulge residues (N7 to N9) had no effect, but deletion of Ile566 (N10) drastically reduced binding, consistent with the more extensive contacts made by Ile566than those of Met561, Leu562, or Ala563. Using the same far-Western assay, we found that the only amino acid that could weakly substitute for the hydroxyproline was cysteine (fig. S4B).

Figure 4

Summary of pVHL far-Western assay with HIF-1α peptides lacking the C segment. The N-segment residues in the vicinity of Hyp564 have secondary roles in binding and specificity. Owing to variations in the amounts of the peptides on the filter, a more quantitative determination is not possible. X indicates hydroxyproline.

Taken together, the structural, thermodynamic, and mutational data indicate that the primary contribution to specificity and affinity is provided by the hydroxyproline side chain, with only secondary contributions from the other side chains in the N segment and an overall minor contribution to affinity from the entire C segment.

The mammalian HPHs have been suggested to contain a jelly roll β-barrel structure and active-site characteristics in common with other iron and 2-oxoglutarate-dependent oxygenases (17, 23, 24). In these enzymes (25), iron, 2-oxoglutarate, and substrates are bound in a spacious cavity, which is located between the two sheets of the jelly roll and is exposed at one side of the β barrel. Although topologically different, the architecture and position of this active-site cavity are reminiscent of the hydroxyproline-binding site of the pVHL β sandwich, leading to the model that HIF-1α may bind to HPHs by forming an intermolecular β sheet and inserting the proline side chain into the β-barrel cavity.

In contrast to extracellular proteins where hydroxyproline plays a structural role (26), the hydroxyproline in HIF-1α is a central determinant of signaling, and the structure of the HIF-1α−VBC complex reveals how it can be recognized with the strict specificity that is characteristic of signaling. The structure also provides a framework for the discovery of inhibitors that prevent HIF-1α degradation and promote therapeutic angiogenesis in cardiovascular disease and stroke.

  • * To whom correspondence should be addressed. E-mail: nikola{at}


View Abstract

Stay Connected to Science

Navigate This Article