Crystal structure of the Retinoblastoma protein N-domain provides insight into tumor suppression, ligand interaction and holoprotein architecture

Summary

The retinoblastoma susceptibility protein, Rb, has a key role in regulating cell cycle progression via interactions involving the central 'pocket' and C-terminal regions. While the N-terminal domain of Rb is dispensable for this function, it is nonetheless strongly conserved, and harbours many missense mutations found in hereditary retinoblastoma, indicating that disruption of its function is oncogenic. The crystal structure of the Rb N-terminal domain (RbN), reveals a single globular entity formed by two cyclin-like folds. The intrinsic similarity of RbN to the A and B boxes of the Rb pocket domain suggests that Rb evolved through domain duplication. Structural and functional analysis provides insight into oncogenicity of mutations in RbN and identifies a unique phosphorylation-regulated site of protein interaction. Additionally, this analysis suggests a coherent conformation for the Rb holoprotein in which RbN and pocket domains directly interact, and which can be modulated through ligand binding and possibly Rb phosphorylation.

Introduction

Loss of signaling involving the retinoblastoma tumor suppressor protein (Rb) is common and important in cancer development. Several classes of tumor viruses inhibit Rb and mutations in the Rb gene are associated with oncogenic transformation (Classon and Harlow, 2002). Rb is a member of the ‘pocket’ protein family implicated in the regulation of cell proliferation. In its hypo-phosphorylated form, Rb interacts with and represses E2F/DRTF transcription factors, impeding the G1/S transition. During late G1, Rb is phosphorylated at multiple sites by Cyclin D/cdk4 and Cyclin E/cdk2 kinases, which abrogates Rb's repressive interaction with E2Fs and allows progression through the cell cycle (Mittnacht, 2005).

Besides its G1/S inhibitory function, Rb is involved in promotion of differentiation, prevention of cell death and control of tissue fate, via its ability to activate transcription factors such as ATF-2, MyoD, Runx2, C/EBP, glucocorticoid (GR) and androgen (AR) receptors, and to recruit SWI/SNF chromatin remodeling activity (Mittnacht, 2005). Rb’s ability to activate gene transcription, is regulated independently of its ability to inhibit E2F, by a mechanism involving the tripartite motif protein 27 / ret finger protein (TRIM27/RFP), which stabilizes the E1A-like inhibitor of differentiation (EID-1) (Krutzfeldt et al., 2005). EID-1, itself an Rb binding protein degraded in an Rb-dependent manner, inhibits p300/CBP in vitro and may interfere with activating chromatin modifications in vivo (Miyake et al., 2000).

Regulation of G1/S progression is primarily a function of the conserved central pocket (aa 379-792) and C-terminal region (aa 792-928) of Rb. The N-terminal region (aa 1-378), although well conserved amongst Rb orthologues and paralogues, has been studied far less. However, a significant number of mutations within this region are found in retinoblastomas, strongly implicating it in tumor suppression, and its integrity is critical for rescue of both the developmental defects and increased tumor susceptibility in Rb deficient mouse models (Goodrich, 2003).

A number of cellular proteins have been reported to interact with the Rb N-terminal region. Yeast two-hybrid studies identified MCM7 (Sterner et al., 1998), a component of the replication origin-recognition-complex (ORC), and p84N5/pThoc1 (Durfee et al., 1994), a death domain protein involved in mRNA splicing and transport (Li et al., 2005). Other identified partners include Sp1 (Udvadia et al., 1995), TFIID/TAF1 (Shao et al., 1997), and the interferon responsive protein p202 (Choubey and Lengyel, 1995). Most recently, transcriptional co-activators, ASC-2 (Goo et al., 2004) and GRIP-1/SRC-2 (Batsche et al., 2005) were found to associate with Rb’s N-terminal region, consistent with the recognized role of Rb in gene activation.

In order to establish a basis for the functional understanding of the Rb N-terminal region we determined the crystal structure of its protease resistant core, encompassing residues 40 to 355 of human Rb. The structure and associated analysis reveal novel and unexpected insight into the evolution of the pocket protein family and suggest mechanisms by which RbN contributes to tumor suppression. Importantly our associated functional analysis provides evidence for (i) a unique site of protein interaction in RbN, regulated by phosphorylation, (ii) a closed Rb holoprotein conformation in which the RbN and pocket domains interact directly, (iii) the modulation of this interaction through ligand binding.

Results

Structure determination

Attempts to generate crystals using the entire N-terminal region (aa 1-370) of human Rb were unsuccessful. To identify a sub-construct amenable to crystallization we employed partial proteolysis (Fig 1A; Supplement; Fig S-1). In line with previously published reports using full length Rb (Hensey et al., 1994), limited tryptic digestion of Rb residues 1-370 generated two fragments with approximate size of 24.7 and 10.9 kDa (Supplement; Fig S-1A). Mass spectrometry and N-terminal sequencing of these identified three polypeptides encompassing residues 46-251, 263-355 and 266-355 (not shown), indicating trimming from both termini and removal of an internal arginine (Arg)-rich linker (aa 251-266) connecting the 24.7 and 10.9 kDa fragments. The two fragments co-purified chromatographically indicating that they remained physically associated (Supplement; Fig S-1B). A refined RbN construct (aa 40-355) combined with limited tryptic digestion, reproducibly yielded diffracting crystals containing both fragments (Supplement; Fig S-1C). The structure was determined by single wavelength anomalous diffraction (SAD) and was refined to 2.0 Å resolution (Supplement; Table S-1).

Fig 1. The Rb N-terminal region consists of two cyclin folds.

A) Schematic representation of Rb domains (top) and the trypsin resistant RbN core (bottom). Arrows indicate sites of tryptic cleavage; protease resistant fragments are depicted in cyan and orange. Positioning of cyclin fold helices for fold A (dark blue) and fold B (red) is indicated. Putative phosphorylation sites are marked.

B) Ribbon representation of the RbN structure colored as in A). Structurally disordered or absent residues are indicated by dashed lines.

Overall structure

The RbN structure is a single globular entity consisting of two sequential cyclin-like folds: lobe A formed by helices α1, α2, α3, α4 and α5 and lobe B formed by the C-terminal 3.5 turns of α6, and helices α7, α8, α10 and α11 (Fig 1B). A C-terminal segment (aa 313-355) containing helices α12, η1 and α13 packs tightly onto helices α5, α6 and α7, and occupies the space between the two lobes, positioning the visible C-terminus (Arg355) between lobe A and B.

Both cyclin folds superimpose well (< 2Å rmsd between Cα atoms) with the canonical folds present in cyclins (Andersen et al., 1996; Brown et al., 1995) (Supplement; Fig S-2C) and TFIIB (Nikolov et al., 1995). The fold composition of RbN is reminiscent of the Rb pocket region, which also consists of two sequential cyclin folds (Lee et al., 1998). Structure-guided sequence alignment reveals an above-average degree of identity (17.75%) between equivalent residues of the B folds of RbN and Rb pocket (Supplement; Fig S-2B,C) indicating that these folds are remote homologues and suggesting that Rb probably has arisen through domain duplication from an ancestral cyclin fold pair.

While RbN and Rb pocket regions are both formed by a pair of cyclin folds, the means by which these pairs are joined and their juxtaposition, differ significantly (Fig 1B and Supplement; Fig S-2A). The RbN lobes are rigidly connected through a single, long central helix α6, which projects from the end of lobe A with its C-terminal half providing the first helix of lobe B. The hydrophobic cores of lobes A and B are consequently entirely separate, suggesting that they may fold independently of each other. This is in line with observations that Rb variants lacking all or part of lobe A stably accumulate in cells and can provide tumor suppressor activity (Sanchez-Sanchez et al., 2006; Xu et al., 1994). The cyclin fold arrangement also means that the interface that provides the E2F docking site in the Rb pocket (Lee et al., 2002; Xiao et al., 2003) is not recapitulated in RbN.

In contrast to lobe A, lobe B contains substantial extensions connecting helices α6 and α7, and helices α8 and α10 (Fig 1B). The connection between helices α6 and α7 forms a well-ordered hairpin-loop (aa 173-188) projecting from the main body of the protein. The conformation of this ‘Projection’ is stabilized by its symmetry-related counterpart in the crystal lattice (Fig 2C). The region connecting helix α8 to helix α10 (aa 230-271) forms an elaborate coil and helix α9, before running into the start of the proteolytically labile Arg-rich linker (aa 251-266) whose excision is required for crystallization.

Fig 2. RbN domain functional surfaces.

A) Surface representation of RbN colored according to sequence conservation in metazoa (human, mouse, chicken, frog, newt, trout, rainbow fish). Prominent conserved surface residues are indicated.

B), C) Crystal lattice contacts indicate sites for protein interaction. Residues involved in crystal-packing interactions are labeled. (B) Helix α13 of a symmetry related molecule packs through hydrophobic interactions onto the conserved patch CP2 (see A). (C) The crystal contact involving the Projection features a ‘handshake’ interaction motif involving exposed hydrophobic residues.

D), D’) Superposition of cyclin folds from RbN (N-A and N-B) with those of the Rb pocket (P-A and P-B) (1GUX). Structural alignments were produced by superposition of 40-57 residues within the cyclin fold forming helices (rmsd of Cα atoms < 2.00 Å for all four folds). For clarity, pair-wise superposition, between N-B and P-B (rmsd of Cα atoms = 1.51 Å, D), and between N-A and P-A (rmsd of Cα atoms = 2.00 Å, D’) are shown. Helices are colored as in Figure 2. Interaction partners involving the canonical protein binding ‘cyclin wedge’ in the four cyclin folds, including E7 peptide for P-B (D) and intramolecular helices α6 and α9 for N-A and P-A, respectively (D’), are indicated in yellow.

E) Conservation between the LxCxE binding surface and equivalent in N-B. Conserved residues in N-B iso-structural to those involved in LxCxE binding by P-B are in green. Remaining residues are colored as in A). Dotted red lines depict disordered areas near the cyclin wedge. Putative cdk phosphorylation sites S249, T252 are indicated.

F) Putative functional surfaces in RbN. Ribbon representation for the RbN structure in a color gradient from blue (aa 51) to red (aa 355). The relative position of individual functional surfaces is indicated.

The structure of RbN contradicts previous predictions from yeast two hybrid and bioinformatics studies (Bork et al., 1997; Hensey et al., 1994; Yamane et al., 2000; Yamane et al., 2001), which identified two independently folded sub-domains, the more amino-terminal of which was deemed to contain a BRCT-like fold. Our structure shows that the two segments generated by limited proteolysis are in fact intimately interacting parts of a single globular entity, so that interaction studies based on one or other of these in isolation must be viewed with considerable caution.

RbN domain functional surfaces

While the precise biochemical function of the Rb N-terminal region is not known, it displays distinctive patterns of sequence conservation, and mutation of this region is associated with familial retinoblastoma (see below), indicating important functional roles. Several lines of structural evidence identify potential functional surfaces on RbN.

Two clusters of highly conserved and surface-exposed residues are evident (Fig 2A). One comprises an extensive patch of predominantly polar residues including Lys122, Asp332, Arg334 and Asp340 (conserved patch (CP) 1) (Fig 2A, left). A second (CP2), on the opposite side, consists of a cluster of prominent hydrophobic residues, including Met208, Leu212, Val213, and Ile214 (Fig 2A, right), which interact with a hydrophobic helical segment in α13 from a symmetry-related molecule (Fig 2B).

A third site involves the Projection (aa 173-188), which contains a number of moderately conserved surface-exposed hydrophobic residues, including Leu174, Pro177, Ile181 and Ile185, that make a ‘handshake’ interaction with their symmetry equivalents in the crystal (Fig 2C). Both lattice contact sites (CP2 and the Projection) yield significant scores for functioning as interaction interfaces using the MSDPisa tool (p value = 0.055 and 0.084 respectively) (Krissinel and Henrick, 2005), supporting their potential involvement in protein-protein binding.

Finally, superimposition of RbN lobe B (N-B) with pocket B (P-B) identifies an unoccupied ‘cyclin wedge’ in N-B (Fig 2D). In known cyclin structures this area, formed by the third, fourth and fifth cyclin fold helix, mediates high affinity ligand interactions (Jeffrey et al., 1995, Nikolov et al., 1995). Most notably, in the context of P-B, it facilitates the binding of the LxCxE motif common to many Rb ligands (Lee et al., 1998). Comparison of the LxCxE binding site in the Rb pocket with the corresponding surface in N-B reveals considerable similarity. Several residues in the N-B cyclin wedge structurally equivalent to those that coordinate the LxCxE peptide in P-B show strong conservation (e.g. Lys228, Lys289, Tyr292) (Fig 2E). However, the N-B wedge is shorter than that in P-B providing insufficient room to accommodate a residue iso-structural to the glutamic acid in the LxCxE motif. The hydrophobic recess, which in the pocket accommodates the leucine, is deeper in the RbN cyclin wedge, potentially providing space for a larger hydrophobic residue. An unstructured loop (aa 301-311) connecting helices α11 and α12 most likely provides one edge of the hydrophobic recess . The Arg-rich linker, which is removed by proteolysis and hence not present in the structure (see above; Fig 1) lies at the opposite side and probably provides the other edge. This linker contains two known cyclin-dependent kinase phosphorylation sites, indicating a possible mechanism for regulating this putative binding site (see Fig 2E).

The area analogous to the cyclin wedge in lobe A is occupied by the N-terminal half of helix α6, and thus unavailable for further intermolecular interactions. This mirrors the situation in the Rb pocket A where the cyclin wedge likewise is engaged by an intramolecular helix (Fig 2D’).

In summary, the cumulative evidence suggests a minimum of four candidate functional surfaces within RbN any or all of which could be involved in protein-protein interactions (Fig 2F).

RbN domain alterations in cancer

A significant number of mutations that map to RbN have been detected in retinoblastoma patients. These include mutations predicting in-frame deletion of amino acids encoded by exons 4, 5, 7 and 9 (Dryja et al., 1993; Jakubowska et al., 2001; Lefevre et al., 2002; Sanchez et al., 2000) as well as missense mutations Glu72Gln, Glu137Asp, Ile185Thr, Leu220Val, Thr307Ile and Thr310Gly (Blanquet et al., 1993; Blanquet et al., 1995; Brichard et al., 2006; Liu et al., 1995; Lohmann et al., 1997).

The structure of RbN shows that protein segments encoded by exons 4, 7 and 9 form integral parts of the protein core (Fig 3A). Loss of these exons is likely to cause gross misfolding, which could trigger triage of the mutant protein into misfolding-associated protein degradation. In line with this, we found that mutant full length proteins with deletion of these exons display increased turnover when expressed in tissue culture cells, and some only accumulate to detectable levels following proteasome inhibition (Supplement; Fig S-3). Destabilization of the Rb holoprotein as a result of exon 4, 7 or 9 deletion provides a satisfactory explanation for the oncogenicity of these mutations. Significantly, we also observed increased turnover of an Rb construct lacking the N-terminal region (Rb∆N, aa 379-928), pointing to a general role of RbN in holoprotein stabilization.

Fig 3. Tumorigenic mutations in RbN.

A) Exon deletions and point mutations associated with cancer are shown. Deletions of Exon 4 (purple), Exon 7 (orange), Exon 9 (blue), and Exon 5 (green) are indicated. Residues associated with point mutations (red) are shown as stick models.

B) Conservation and accessibility of RbN residues. Grey bars indicate conservation among Rb homologues as in Figure 3A. Circles show accessibility of residues (red = highly accessible, black = buried). Dashed lines in the secondary structure assignments indicate residues are disordered or absent in the crystals. α-helices corresponding to the cyclin fold are labeled. Sites of cancer-associated mutations are colored as in A).

In contrast, loss of exon 5 is likely to be accommodated without gross deleterious effects on the overall fold. The residues encoded by this exon form two turns of the long central helix 6 and part of the Projection (Fig 3). Deletion of exon 5 removes the Projection, but otherwise is unlikely to disturb the continuity or hydrophobic core of RbN. Consistent with the limited structural effects predicted, an exon 5-deleted Rb protein accumulates in cells to similar levels as, and clears with a half-life indistinguishable from, wild-type Rb (Supplement; Fig S-3A). Furthermore, like wild-type RbN, but unlike the other exon deletion mutants, exon 5-deleted RbN can be expressed in E. coli and purified as a stable soluble protein, whose circular dichroism (CD) spectrum is indistinguishable from wild type (Supplement; Fig S-8). In consequence, the link between loss of exon 5 and cancer argues for a specific role of the Projection in tumor suppression. Of note, the retinoblastoma-associated missense mutation Ile185Thr maps within the Projection. Ile185 forms hydrophobic contacts with several residues including its own counterpart from a symmetry-related molecule in the lattice (Fig 2C). Mutation to threonine disrupts these contacts by introducing a polar side chain into an otherwise hydrophobic interaction, and may impact on protein function in a similar way as the loss of exon 5.

The structure provides no obvious explanation why missense alterations Glu72Gln and Glu137Asp might confer oncogenicity. Both are fully surface exposed, and the mutations would have no effect on local structure (Fig 3A), but might conceivably abrogate a protein contact involving the mutated residue. However, explicit evidence for involvement of this region in protein-protein interactions is not evident from the structure.

Thr307Ile and Glu310Gly mutations map within the disordered loop (aa 301- 311) between α11 and α12 (Fig 3)adjacent to the N-B cyclin wedge (refer to Fig 2E), and might affect the interaction of ligands. Leu220Val lies in the hydrophobic core of lobe B, at the interface of helices α8, α10 and α11 (Fig 3A), and the mutation generates a gap in the closely packed interface, which would perturb the relative orientation of these helices, with effects on the binding of potential cyclin wedge ligands.

Thus, while a number of cancer-associated RbN mutations clearly act through global destabilization of the Rb holoprotein, others generate subtle surface changes, most notably affecting the hydrophobic Projection and lobe B cyclin wedge. Both features are likely sites of protein-protein interaction (see above), which implicates these sites and their interactions in the tumor suppressor activity of Rb.

Protein interactions involving RbN

We assessed the in vitro interaction with RbN of a number of previously reported proteins, using purified components and GST affinity precipitation (Fig 4A). While we were unable to recapitulate binding of MCM7 (Sterner et al., 1998) (not shown), binding was seen for p84N5 (aa 379-675), (Durfee et al., 1994) and GRIP-1 (aa 1236-1392), (Batsche et al., 2005). We also confirmed an interaction between RbN and the rest of the Rb protein (Rb∆N, aa 379-928), previously suggested by yeast two-hybrid screen (Hensey et al., 1994). In addition we identified a novel interaction with the E1A-like inhibitor of differentiation, EID-1 (MacLellan et al., 2000; Miyake et al., 2000). This interaction was seen with either EID-1 or RbN fused to GST, and EID-1 bound RbN to the same degree as p/CAF (Supplement; Fig S-4A). A dissociation constant for the RbN – EID-1 interaction was determined as ~ 8 μM by isothermal titration calorimetry (ITC) (Supplement; Fig S-4B). Significantly, an Rb deletion construct lacking RbN showed reduced sensitivity to the inhibitory effects of EID-1 as determined by the ability of Rb to co-activate GR driven transcription (Supplement; Fig S-5), implying that the RbN – EID-1 interaction contributes to EID-1’s ability to modulate Rb activity in cells.

Fig 4. RbN interacts with EID-1.

In vitro protein-binding assays with purified human proteins. Immunoblots were probed with antibodies as indicated.

A) RbN interacts with p84N5, GRIP-1, EID-1, and Rb∆N. Aliquots of RbN were incubated with equivalent amounts of GST tagged proteins. GST and GST–p300 Bromo are controls.

B), B’) EID-1 interacts independently with RbN and the Rb pocket. Aliquots of EID-1 were incubated with equivalent amounts of GST tagged proteins or GST as a negative control. (B’) Schematic representation; dashed boxes indicate regions of interaction in Rb.

C) The Arg-rich linker is required for EID-1 binding to RbN. Aliquots of Rb 40-355 wt, Rb 40-355 polyG, Rb 40-355 ∆Ex5, Rb 40-355 CP2 were incubated with equivalent amounts of GST-EID-1.

D) Effect of RbN phophorylation on EID-1 binding. Antibody-FRET mediated complex detection. GST-RbN was phosphorylated by cyclin A/cdk2 (A/K2) or B/cdk1 (B/K1). Controls omitting enzyme or ATP were run in parallel. Assays used 2.5 nM GST-RbN and 300nM HisTrx-EID1. Data shown are net of background fluorescence transfer obtained with unfused HisTrx. The net background fluorescence transfer using HisTrx-EID-1 with GST instead of GST-RbN is shown. Error bars represent standard deviation form quadruplet samples. D’) Anti-phospho-Rb western blots. Rb preparations were analysed by immunoblot using phosphorylation site-selective and pan-Rb antibodies as indicated.

E) p130 N-terminal domain does not interact with EID-1. Aliquots of EID-1 or Rb pocket-FLAG were incubated with equivalent amounts of GST tagged proteins, GST as a negative control, or GST-RbN as a positive control.

Both Rb∆N and EID-1 precipitated RbN more readily than GRIP-1 or p84N5, while GST or an irrelevant GST protein did not bind RbN, confirming that the interactions are selective (Fig 4A). Taken together these data show that RbN facilitates interactions with an array of heterologous proteins, and can engage in homotypic interaction with a region C-terminal to RbN.

The interaction of EID-1 with RbN

While the RbN – EID-1 interaction is novel, an EID-1 segment containing an LxCxE motif has previously been shown shown to bind to an Rb pocket construct (379-885) in a yeast-two hybrid analysis (MacLellan et al., 2000; Miyake et al., 2000). We observed a similar interaction in vitro between EID-1 and the central Rb pocket (aa 379-792) (Fig 4B). Both independent interactions of EID-1 with RbN and Rb pocket were confirmed using dimethyl-superimidate (DMS) crosslinking and antibody-facilitated Foerster resonance excitation transmission (FRET) (Supplement; Fig S-6, S-7). Binding to both Rb pocket and RbN has been noted for other ligands, including GRIP-1, ASC-2, TAF1 and p202, and appears to be a common theme amongst Rb binding proteins (Batsche et al., 2005; Choubey and Lengyel, 1995; Goo et al., 2004; Shao et al., 1997).

In the current absence of a crystal structure for their complex, we have probed the interaction of EID-1 and RbN by targeted mutagenesis of putative functional surfaces of RbN. Specifically, we (i) mutated the highly conserved residues Leu212, Val213 and Ile214 to affect CP2, (ii) generated a construct corresponding to the oncogenic exon 5 deletion (aa 168-181) which compromises the Projection and (iii) replaced the Arg-rich linker (aa 250-263) with a six residue glycine-rich linker (polyG). Correct folding was confirmed for all mutants using CD spectroscopy (Supplement; Fig S-8). Mutations in CP1 significantly impaired solubility (not shown), suggesting that the mutant proteins did not fold correctly, and precluded their inclusion in the analysis.

Using GST affinity precipitation we found that EID-1 retained binding to RbN CP2 and exon 5 mutants, suggesting that these features are not required for the interaction (Fig 4C). In contrast, replacement of the Arg-rich linker led to complete loss of binding, strongly implicating this feature in the interaction with EID-1. The possible involvement of the adjacent cyclin wedge, was not confirmed, as targeted or tumour-derived point mutations that map to this feature did not abrogate EID-1 binding (Supplement; Fig 9A). However, similar to attempts to create mutants of the LxCxE binding cleft in the Rb pocket (Chan et al., 2001; Dahiya et al., 2000; Dick et al., 2000), it is possible that single residue exchange in the RbN cyclin wedge is insufficient to abrogate ligand binding. Our results do however indicate that the various cancer-associated mutants tested do not target EID-1 binding.

As the Arg-rich linker encompasses Thr 252 and Ser 249, two cdk phosphorylation sites, we asked whether phosphorylation might affect EID-1 binding to RbN. Using antibody-FRET we found that phosphorylation by either cyclin A/cdk2 or cyclin B/cdk1, both of which effectively phosphorylated the respective sites, abrogated RbN interaction with EID1, revealing a mechanism of regulation by cdk phosphorylation and corroborating the involvement of the linker in EID-1 binding (Fig 4D, D’). The Arg-rich linker and cdk phosphorylation sites in RbN are not conserved in the Rb homologues p107 and p130, and a p130 N-domain construct did not bind EID-1 (Fig 4E). In contrast to RbN, phosphorylation of the Rb pocket did not affect its interaction with EID-1 (Supplement; Fig S-10).

EID-1 regions interacting with Rb

To delineate the region(s) of EID-1 that interact with Rb, we probed an array of serially overlapping 25 mer peptides of the entire EID-1 sequence (Fig 5A), generated using SPOT™ technology (Frank and Overwin, 1996). EID-1 is a natively unfolded protein as verified by 1D NMR (Supplement; Fig S-11A) and therefore ideally suited for array analysis. Probing the array with RbN revealed two discontinuous footprints covering residues 142-169 and residues 152-180 of EID-1, respectively (Fig 5A and Supplement; Fig S-11B). When probed with the Arg-rich linker mutant (polyG) the second footprint, was absent, suggesting that residues within this sequence facilitate interaction with the linker. The Rb pocket displayed a different footprint, involving peptides spanning the central and acidic regions of EID-1, and a C-terminal cluster encompassing the EID-1 LxCxE motif, consistent with previous suggestions (MacLellan et al., 2000; Miyake et al., 2000) (Fig 5A). Notably, the RbN footprints were not bound by the Rb pocket, showing that the two regions engage non-overlapping segments of EID-1. Using a second peptide array, covering the two RbN interacting footprints on EID-1, but with serial Ala substitutions in single and multiple consecutive residues (Fig 5A’ and Supplement; S-11C) we defined a minimal motif of Phe166 and Val170 (FxxxV), plus a triplet of residues N-terminal to these (FIE), as essential for RbN interaction (Supplement; Fig S-11B, C).

Fig 5. Separate EID-1 regions interact with RbN and Rb pocket.

A) SPOT™ array of serially overlapping 25mer EID-1 peptides was probed with RbN wt, RbN polyG, or Rb pocket. Peptide stretches of EID-1 interacting with various Rb constructs are indicated in the schematics at the top right of individual panels.

A’) Identification of EID-1 residues critical for interaction with RbN. A SPOT™ array of 30mer EID-1 peptides (aa 139-168 (A1-C30) and aa 152-181 (D1-F30)) with serial Ala substitutions as indicated was probed with RbN wt. Single (rows A, D), double (B, E) or triple (C, F) substitutions were used as indicated.

B) Summary of EID-1 peptide interactions. Coloured residues denote amino acids crucial for interaction with RbN. The EID-1 LxCxE motif is underlined. Peptide designs based on the SPOT array results, and used for further study, are indicated.

C), C’) Association between GST-RbN (C) and GST-Rb pocket (C’) with EID-1 in the presence of various EID-1 peptides based on the SPOT assay results. Affinity precipitation with proteins as indicated. RbN-interacting peptide (aa 144-176, FxxxV), RbN-interacting mutant peptide (aa 144-176, Mutant), acidic stretch peptide (aa 86-107, Acidic), or LxCxE peptide (aa 172-187, LxCxE) was included in 30-fold molar excess over GST-Rb and EID-1.

Competition experiments provided verification of this site, with a peptide spanning the FxxxV motif, but not an Ala-substituted mutant (see Fig 5B), out-competing the interaction of EID-1 with RbN (Fig 5C and Supplement; Fig S-12A). Consistent with the selectivity of this motif, FxxxV peptide did not affect binding of EID-1 to the Rb pocket. Conversely, a peptide spanning the LxCxE motif competed binding of EID-1 to the Rb pocket, but did not affect its interaction with RbN (Fig 5C’ and Supplement; Fig S-12B). Together these results show that EID-1 utilises non-overlapping peptide motifs to interact simultaneously with RbN, via the Arg-rich linker, and with the Rb pocket, most probably via the LxCxE binding site.

The Rb homotypic interaction and its interplay with EID-1

Data presented above indicate a homotypic interaction between RbN and the remainder of the Rb protein, Rb∆N. Using constructs for different Rb subregions, we identified the central Rb pocket region (aa 379-792), excluding the spacer sequence located between the A and B subdomains, as sufficient for this interaction (Fig 6A, B, B’ and Supplement; Fig S-13).

Fig 6. The RbN domain interacts with the Rb pocket.

In vitro protein-binding assays. Immunoblots were probed with antibodies as indicated.

A), A’) RbN interacts with Rb pocket (aa 379-792) and Rb∆N (aa 379-928). Aliquots of RbN were incubated with equivalent amounts of GST tagged Rb pocket or GST as a negative control. (A’) Schematic; dashed boxes indicate regions of interaction in the Rb large pocket.

B), B’) RbN interacts with the Rb pocket independently of the pocket spacer region (aa 580-640). (B) Aliquots of RbN were incubated with equivalent amounts of GST tagged proteins as indicated. (B’) Aliquots of modified Rb pocket with the spacer removed by thrombin (∆ aa 589-636; RbP∆Sp) was incubated with equivalent amounts of GST tagged proteins as indicated.

C) RbN does not use CP2, Projection, or Arg-rich linker for its interaction with the Rb pocket. Aliquots of Rb 40-355 wt, Rb 40-355 CP2, Rb 40-355 ∆Ex5, Rb 40-355 polyG were incubated with equivalent amounts of GST-Rb 379-792.

D) Effect of RbN phosphorylation on pocket binding. Antibody FRET and data presentation was done as for Figure 5 using phosphorylated RbN as shown but FLAG-tagged Rb pocket in place of EID-1.

E), F), F’) Association between RbN domain and pocket in the presence of EID-1. Affinity precipitation with proteins as indicated. Full length EID-1 (aa 1-187) was added in equimolar amounts and 3-fold or 10-fold molar excess (E). EID-1 RbN-interacting peptide (aa 144-176, FxxxV), EID-1 acidic stretch peptide (aa 86-107, Acidic), or EID-1 LxCxE peptide (aa 172-187, LxCxE) was included at a 30-fold molar excess (F). LxCxE peptide was added in 1-fold, 3-fold, 9-fold, or 27-fold molar excess (F’).

To determine if any of the potential interaction surfaces of RbN are essential for binding to the Rb pocket, we tested the available RbN mutants using GST affinity precipitations. All mutants bound the Rb pocket equally well (Fig 6C and Supplement; Fig S-9B), including the Arg-rich linker mutant, which is unable to bind EID-1 (Supplement; Fig S-9A). This Arg-rich linker mutant was also defective in binding p84N5 and GRIP-1 (Supplement; Fig S-14), indicating that all these heterologous ligands interact with RbN in a similar way, but dissimilar to the Rb pocket. Furthermore, cdk phosphorylation of RbN, which affected its ability to bind EID-1 (see Fig 4D), did not affect its interaction with the Rb pocket, corroborating the distinct nature of these interactions and indicating that the RbN – Rb pocket interaction is not regulated by phosphorylation of RbN (Fig 6D). Similar results were obtained for phosphorylation of the Rb pocket, and coordinate phosphorylation of RbN and Rb pocket (Supplement; Fig S-15, S-16), neither of which measurably affected the interaction of these two parts of Rb with each other.

We also assessed the ability of the Rb pocket to associate with RbN in the presence of EID-1 (Fig 6E and Supplement; Fig S-17). These experiments provided collective evidence that EID-1 displaces RbN from the Rb pocket, in favour of binary interactions with RbN and the Rb pocket respectively (Supplement; Fig S-17). Intriguingly, this displacement could be recapitulated by the EID-1 derived LxCxE containing peptide (Fig 6F, F’), but not by the FxxxV peptide, which effectively competed for EID-1 binding to RbN (see Fig 5C).

Discussion

Previous structural work has concentrated on the central Rb pocket domain, which provides the binding sites for two of Rb’s best characterized protein ligands, the E2F-1 transcription factor and the papillomavirus E7 oncoprotein (Lee et al., 2002; Lee et al., 1998; Xiao et al., 2003). No experimental structural analysis of the N-terminal region has previously been reported. The structure presented here offers experimental insight into the fold and surface properties of this part of the retinoblastoma protein and provides a means for dissecting the mode of action and biological significance of this region from a firm structural basis.

Contrary to bioinformatics analyses suggesting the presence of a BRCT domain (Bork et al., 1997; Yamane et al., 2000), the crystal structure of RbN reveals an architecture closely resembling that of the Rb pocket domain. Like the Rb pocket, RbN is built around two tandemly arrayed cyclin folds. Initially identified in cyclin A (Brown et al., 1995) and subsequently in the basal transcription factor TFIIB (Nikolov et al., 1995), the cyclin fold is a known scaffold for protein-protein interactions. However, whereas cyclins and TFIIB contain one tandem repeat of this fold, Rb unusually contains not one but two tandem cyclin repeats. Threading analysis (not shown) shows this architecture is shared by the Rb-like pocket proteins p107 and p130.

As in other characterized cyclin fold proteins, the tandem RbN cyclin folds have diverged substantially. However, structural alignment of the ‘B’ folds from RbN and Rb pocket, reveals a degree of sequence identity between equivalent residues (17.75%), comparable to that between the cyclin folds within RbN itself (15.25%), cyclin A (15.63%) or TFIIB (17%) (Noble et al., 1997) (Supplement; Fig S-2C). The identity between structurally equivalent positions in the ‘A’ folds of RbN and Rb pocket is much lower (6.78%) suggesting that the conservation in B folds does not merely reflect common ancestry, but indicates a common selective pressure to retain structural and/or surface features. While most conserved residues in the RbN B fold are in the core, some are surface accessible, include several whose pocket equivalents participate in binding LxCxE peptide, arguing for functional retention of the corresponding surface in RbN.

As in other cyclin fold proteins, which offer multiple binding sites for ligands, several other putative functional surfaces can be identified in the RbN structure. Two of these are involved in lattice contacts in the crystal and might conceivably play a role in the reported oligomerisation of Rb (Hensey et al., 1994), although our own biochemical work using purified recombinant RbN has provided no evidence for this.

Substantial conservation and the demonstration of protein binding capacity have long emphasized the importance of RbN, while its biological role has remained ambiguous. Previous efforts to assign function to this region employed scanning deletion and insertion mutagenesis (Qian et al., 1992; Riley et al., 1997). Re-evaluation in light of the RbN crystal structure, indicates that mutants with loss-of-function in those studies invariably disrupted the overall structure of the protein, and are therefore not suited to reveal specific functions of this region. Our finding that structural disruption of RbN in vivo impacts on the stability of the entire protein resolves the conundrum whereby mutation in RbN results in greater loss of function than if it is totally ablated (Qian et al., 1992; Riley et al., 1997; Yang et al., 2002).

Our structure-guided analysis identifies the Arg-rich linker connecting cyclin fold helices 3 and 4 of the RbN B lobe as essential for interaction with several ligands, including the EID-1 inhibitor of histone acetylation and differentiation, and the previously recognized Rb interacting proteins p84N5 and GRIP-1. The requirement of this linker for binding different protein ligands identifies it as a key structural element, and the first in the RbN domain to which protein-binding activity can be assigned. Its functional significance is further supported by the presence of two cdk phosphorylation sites within it, and we have shown that phosphorylation of these sites abrogates these interactions, providing the first mechanistic insight as to how phosphorylation within RbN may contribute to its functional inactivation.

We have also identified an interaction between the RbN and pocket regions which together with the central positioning of the C-terminus of RbN, suggests a compact arrangement of the Rb holoprotein with RbN and pocket engaged in direct contact. This differs from the current model, in which RbN and pocket domains are considered as essentially separate structural and functional entities, loosely connected like beads on a string. Our knowledge as to topology of interaction and the relative juxtaposition of the RbN and pocket regions is currently limited, although our data clearly exclude involvement of the Arg-rich linker. The observation that the LxCxE containing peptide of EID-1 disrupts the interaction is intriguing, as binding of LxCxE ligands does not induce conformational changes in the Rb pocket (Lee et al., 1998). This suggests that EID-1 acts by steric interference, and in turn implicates the LxCxE binding cleft, or a site nearby, in facilitating the interaction of RbN with the Rb pocket. It should be noted that no LxCxE motif is present in RbN, excluding a canonical LxCxE engagement with the Rb pocket. However there is precedence for other types of interactions near or within the LxCxE docking cleft, including the phosphorylated Rb C-terminus, which is thought to engage via a cluster of basic residues lining the rim of the LxCxE binding cleft (Lee et al., 1998; Rubin et al., 2005), and the APC-associated protein cdh1, which binds Rb though a sequence within its WD40 repeat and is disrupted by some mutations that also disrupt LxCxE binding (Binne et al., 2007).

The interaction we observe between the N-terminal and pocket regions of Rb is not static, but can be modulated by binding of protein ligands. Indeed Rb may exist in two opposing states; a ‘closed’ conformation, in which RbN and Rb pocket interact and are in close proximity (Fig 7A), and an ‘open’ conformation in which this interaction is absent and proximity lost (Fig 7B, C). Our data suggest that the conformational switch from closed to open is facilitated by recruitment of EID-1 into the vicinity of the pocket where it disrupts the RbN - Rb pocket interaction most probably by interfacing with the LxCxE binding surface (Fig 7B).

Fig 7. Interaction with EID-1 causes conformational re-arrangement of Rb.

Model for ligand induced conformational response of Rb. (A) Full length Rb can exist in a conformation in which RbN and pocket directly interact. (B) Occupancy of the LxCxE binding site impairs association between RbN and Rb pocket resulting in the protein adopting an open conformation. (C) Although not experimentally tested here, phosphorylation of the Rb C-terminus could have a similar effect (Rubin et al., 2005).

Although not experimentally addressed here, this raises the possibility that other ligands might elicit similar conformational responses. These might include LxCxE containing ligands more generally, but also the phosphorylated C-terminal region of Rb itself, whose binding may functionally inactivate Rb by disrupting composite surfaces formed by association of RbN and Rb pocket regions (Fig 7C). Our observations may therefore represent a more general mechanism whereby ligands direct conformational changes in Rb that either cause functional inactivation, and/or orchestrate the assembly of specific Rb-containing complexes, and may provide a rational explanation for the prevailing evidence that Rb participates in functionally distinct multiprotein complexes made from non-overlapping sets of Rb ligands.

Experimental procedures

Crystallization and structure determination

Purified RbN (aa 40-355) was partially digested using trypsin (Promega), purified over a Resource Q and concentrated to 8 mg/ml. Crystals were grown at 4°C by hanging drop vapor diffusion, from mixtures containing equal volumes of protein solution and reservoir solution containing: 0.2 M Na acetate, 25% (w/v) PEG 4000, and 0.1 M Tris at pH 8.0, and flash frozen in mother-liquor made up to 25% (v/v) MPD. Diffraction data of SeMet substituted crystals were collected on ID29 and ID14 at ESRF and processed using MOSFLM and SCALA (Leslie, 2006). SAD phases were calculated in SOLVE/RESOLVE (Terwilliger and Berendzen, 1999) and the model refined with REFMAC5 (Murshudov et al., 1997). Model building used COOT (Emsley and Cowtan, 2004) and images were generated using PyMol (DeLano, 2002) and Chimera (Pettersen et al., 2004). Crystallographic statistics are presented in Supplemental Table 1. Coordinates and structure factors have been deposited in the Protein DataBank with PDB Code XXXX. Details on other methods can be found in the supplemental material.