Abstract

TABLE OF CONTENTS GRAPHIC

Open in a separate window

INTRODUCTION

The Ras/Raf/MEK/ERK pathway involves the transduction of extracellular growth signals to the nucleus to regulate events in cell proliferation and differentiation. This pathway is frequently affected in tumor formation through the overexpression of growth factor receptors and activating mutations in Ras and Raf kinase are common events. Considerable efforts in drug discovery have been invested and have in recent years paid some dividends. In particular Raf kinases (ARAF, BRAF and Raf-1/C are known members) are considered as attractive therapeutic targets ^{1, 2}. Of these, BRAF is the major activating kinase for MEK/ERK and as a result is probably the most frequently mutated kinase in cancers including melanoma, hairy cell leukemia and colorectal carcinomas among other tumor types ^{3, 4}. A breakthrough in the treatment of malignant melanomas has been achieved in the approval of vemurafenib, a BRAF inhibitor initially producing dramatic responses in treated patients and which targets a constitutively active BRAF mutant (V600E). These drugs target the signal transduction pathways stimulated by binding of growth factors to their receptors that then result in activation of Ras proteins. Oncogenic Ras signaling occurs in about 30% of all human cancers and triggers homo-or hetero-dimerization of Raf-kinases that is critical for several aspects of signal propagation through downstream MEK and ERK kinases^{5, 6}. Despite intense efforts, pharmacologic inhibition of RAS proteins themselves and inhibition of their downstream effector kinases has so far been unsuccessful in treating RAS-driven tumors. Despite the dramatic initial response rates of vemurafenib in BRAF mutant melanoma patients, drug resistance and secondary neoplasms emerge in treated patients thereby dampening the initial enthusiasm for this approach ^{7, 8}. Further investigation into the mechanisms driving these clinical complications has provided considerable insights and determined that a major cause results from “paradoxical MEK/ERK signaling” by the same mechanisms precluding the use of these drugs in Ras-driven tumors. These studies have demonstrated that while vemurafenib inhibits BRAF^V600E very potently, in the context of WT BRAF (in both homodimers and BRAF/C-Raf heterodimers) and activating Ras mutations, leads to kinase activation of the other partner in the dimer thereby stimulating the downstream pathway rather than inhibiting it ^9–11. Allosteric transactivation of a catalytically competent RAF protomer by a drug-bound BRAF molecule requires an intact dimer interface¹². This resistance pathway therefore requires further efforts to complement inhibition of the mutant V600E kinase with other ways of inhibiting downstream signaling. Despite clinical success, the emergence of resistants tumors necessitates continued investigation and drug discovery efforts around the Ras/Raf/MEK/ERK pathway. Combination of MEK inhibitors with approved BRAF drugs has been shown to be an effective strategy and has resulted in the recent approval of trametinib to treat BRAF mutated melanomas¹³. While a significant improvement, MEK inhibitors have some toxicity issues and thus further advances are required. ATP-competitive Raf inhibitors that induce paradoxical ERK activation must not be used to treat RAS mutant tumors^{12, 14}. A recent preclinical study has shown that targeting the complete Raf node phenocopies the growth inhibiting effects of removing the oncogenic driver, mutant Ras¹⁵. A new class of inhibitors that take the dynamic interplay of Raf-isoforms by dimerization and feedback loops into consideration would therefore be beneficial and this requires a detailed understanding of BRAF and its homo and heterodimerization and effects on downstream signaling. In this study, based on elucidation of the residues in the DIF important for dimer formation, the requirement for transactivation for an intact dimer interface^{12, 16} and published structural information on the BRAF dimer¹⁷, peptides were designed that successfully bind to BRAF and furthermore act to abrogate downstream signaling of ERK. These can be classified as type IV kinase inhibitors i.e. those that bind and inhibit allosterically at sites distant from the catalytic cleft¹⁸. The structure activity relationship of DIF peptides has been defined through computational analysis, alanine scanning and testing of these in an intrinsic tryptophan fluorescence (ITF) assay measuring direct binding to BRAF. Based on activities of the linear peptides and the observed loop structure at the dimer interface, highly potent cyclic peptides that mimic and stabilize the bioactive conformation have been generated. Macrocyclic drug discovery has in recent years become an area of interest especially in targeting protein-protein interactions ^19–22. Macrocycles (MCs) typically go beyond the rule of 5 for orally available drugs especially with regard to allowing high MW compounds^{20, 23}. This enables more extensive coverage of the larger interfaces of PPIs. The lead BRAF DIF inhibitor macrocycles therefore represent tool compounds to probe how Raf dimerization events contribute to propagation of signals in the Ras/Raf/MEK/ERK pathway, while providing a scaffold for making drug-like cyclic peptides with favorable pharmacological properties.

RESULTS

Structure guided design and optimization of peptidic inhibitors of BRAF dimerization

The crystal structure of the homodimer of BRAF kinase (in complex with an ATP competitive inhibitor) has recently been solved²⁴. Furthermore, the dimerization of Raf kinases has been shown to involve a central cluster of residues known as the dimer interface (DIF). In particular, highly conserved basic residues within the DIF (R509 in BRAF) play a critical role in promoting dimerization²⁵ and DIF mutants incorporating the R509H mutation, either singly or in combination with the L515G and M517W substitutions, prevent mutant Ras and vemurafenib-induced paradoxical MEK/ERK signaling¹². Based on this information, the crystal structure of the BRAF homodimer (PDB ID: 4E26) was examined to determine if peptide sequences isolated from the DIF could potentially block the protein-protein interaction involved in BRAF homo and heterodimerization events. It was observed that the residues in the dimer interface occur in a looped turn structure (Figure 1) and present a relatively compact binding interface with the other BRAF monomer (and presumably similar interactions in heterodimers with other RAF kinases).

Open in a separate window

Figure 1.

Structure of the DIF peptide bound to BRAF (PDB ID 4E26).

The two BRAF monomers are shown as white and yellow surfaces respectively and the DIF sequence is represented by a cyan ribbon and blue carbon atoms. The ATP binding site is filled with a kinase inhibitor in both monomers for perspective.

Coincidentally, a study was published shortly after the initial peptide design, describing the cellular expression and effects on BRAF signalling of a 19-residue BRAF peptide, comprising residues 503–521 of BRAF¹⁶. To validate the approach of blocking the dimer interface through this sequence, DIF peptides were designed in cell permeable form as TAT fusions. These included TAT-pep1 (BRAF 503–521, prior study¹⁶), TAT-pep17 (BRAF 504–518, loop forming residues from DIF contact surface) and TAT-6alaNC3 (residues contacting the other monomer mutated to alanine). As shown (Supplementary Figure), results from a colony forming assay using EGF dependent MCF-10A cells as a preliminary screen for anti-proliferative activity demonstrated that TAT-pep1 completely inhibited growth at 2 μM. The shorter BRAF sequence, TAT-pep17 had somewhat greater antiproliferative potency with complete inhibition at the slightly lower dose of 1.5 μM and the negative control sequence had little effect on the colony forming ability of these cells (supplementary figure 1). Next, we tested the TAT-peptides on the human melanoma cell line Sbcl2. This cell line does not carry BRAF mutations and is driven by the Q61K gain-of-function mutation in NRAS, an alternative and ERK pathway-activating event to BRAF mutations in melanoma¹² We first asked whether the melanoma cells efficiently take up the TAT peptides. To this end, we used TAT peptides with an N-terminal 5-Carboxyfluorescein (5-FAM) label. As shown in Figure 2A, melanoma cultures exposed to 3.6 μM of both peptides accumulate fluorescently labelled peptides, while cells exposed to the lower concentration only display a diffuse autofluorescence. Interestingly, the cells treated with the biologically active FAM-TAT-Pep17 not only form smaller colonies, but also tend to display a round morphology, while control peptide exposed cells attached to the substrate are more elongated. The dose dependent effect of TAT-Pep17 on proliferation was also evident in a colony formation assay (Figure 2B). Similar results were obtained for the unlabelled versions of the TAT-peptides (Supplementary Figure 2). In summary, these data show that both labelled and unlabelled TAT-peptides matching to the BRAF DIF impair proliferation, while the control TAT-6alaNC3 peptides only show effects at higher concentrations.

Open in a separate window

Figure 2.

FAM-TAT-Pep17 impairs the growth of NRAS^Q61K driven human Sbcl2 melanoma cells.

(A) FAM labeled peptides are taken up by melanoma cells. Sbcl2 cells were plated in tissue culture vessels (6-well format) and grown in the presence of the indicated concentrations of FAM labeled TAT peptides. Medium with freshly added peptides were changed every 3 to 4 days. Shown are micrographs taken two weeks after seeding. Cells incubated with the higher concentration accumulate fluorescently labeled peptides, while cells exposed to the lower concentration only display a diffuse autofluorescence. Note that cells treated with the biologically active FAM-TAT-Pep17 not only form smaller colonies, but also tend to display a round morphology, while control peptide exposed cells attached to the substrate are more elongated. (B) Five thousand Sbcl2 cells were seeded onto 6-well plates and grown in the presence of the indicated peptide concentrations for two weeks. Medium with freshly added peptides were changed every 3 to 4 days. Cells were stained with Giemsa solution. Shown is a representative result from two independent biological replicates with comparable outcome

Having achieved cellular validation for on target inhibitory activity of BRAF dimerization, an intrinsic tryptophan fluorescence (ITF) assay was developed to quantify binding of these peptides to BRAF. A tryptophan residue (Trp450) directly contacts R509 thereby providing the impetus to use this assay format to quantify binding. Since there is no competitor involved in this assay, the readout is the Kd for direct interaction between the peptide and the BRAF construct (Tables 1, ​,2,2, and ​and3,3, see supplementary Figure 1 for an example Kd curve). For this assay, a truncated BRAF comprised of the kinase domain in addition to the dimerization interface was expressed and purified. Peptide 1 comprising BRAF residues 503–521 and shown above to inhibit cellular proliferation as a TAT fusion, was found to have a Kd of 3.8 μM thus confirming affinity for the BRAF monomer¹⁶. Two negative control peptides based on mutation of residues shown to be important for dimer interface in the endogenously expressed BRAF mutants¹² were tested. As expected, the triple mutant (2; R509H, L515G, M517W) had no detectable binding. Following confirmation of the binding of the initial peptide and lack of affinity of the negative control, a peptide library (Table 1) was designed around the initial sequence and synthesised using solid phase peptide synthesis. Computationally designed mutations were incorporated to probe roles of specific residues in binding to BRAF and affecting dimer formation. The aim of this library was to probe the binding determinants and effects of truncation with the eventual goal of generating cyclic peptides that stabilize the loop structure observed in the dimeric crystal structure.

Table 1.

Structure-activity relationship of BRAF Dimer Interface Peptides

ID	BRAF Residues	Sequence	Kd (μM)
1	503–521	GVLRKTRHVNILLFMGYST	3.84 ±0.32
2	503–521 R509H, L515G, M517W	GVLRKTHHVNILGFWGYST	NB
3	scrambled	GRINKGRHTFLLVVMTYSL	2.96 ± 0.18
4	503–521 L505A	GVARKTRHVNILLFMGYST	3.89 ±0.53
5	503–521 R506E	GVLEKTRHVNILLFMGYST	1.09 ±0.29
6	503–521 R506L	GVLLKTRHVNILLFMGYST	0.54 ±0.11
7	503–521 T508D	GVLRKDRHVNILLFMGYST	2.2 ±0.83
8	503–521 T508A	GVLRKARHVNILLFMGYST	2.8 ±0.29
9	503–521 H510F	GVLRKTRFVNILLFMGYST	NB
10	503–521 V511A	GVLRKTRHANILLFMGYST	4.75 ±1.7
11	503–521 L514A	GVLRKTRHVNIALFMGYST	9.8 ±1.6
12	503–521 L515I	GVLRKTRHVNILIFMGYST	4.1 ±1.1
13	503–521 L515homoleucine	GVLRKTRHVNIL{HL}FMGYST	1.25 ±0.36
14	503–521 F515D	GVLRKTRHVNILLDMGYST	NB
15	503–518	GVLRKTRHVNILLFMG	1.88 ±0.36

Open in a separate window

Table 2.

Ala scan and Truncation studies of BRAF Dimer Interface

ID	BRAF Residues	Sequence	Kd (μM)
16	504–517	VLRKTRHVNILLFM	5.75 ± 1.2
17	504–518	VLRKTRHVNILLFMG	0.13 ±0.040
18	504–518	VLRKTRHVNILLFMG-NH2	0.48 ±0.091
19	504–518	Ac-VLRKTRHVNILLFMG	0.8 ±0.083
20	504–518 L515homoleucine	VLRKARHVNIL{HL}FMG	0.49 ±0.16
21	504–518 L505A	VARKTRHVNILLFMG	0.45 ±0.03
22	504–518 R506A	VLAKTRHVNILLFMG	0.36 ±0.03
23	504–518 K507A	VLRATRHVNILLFMG	ND
24	504–518 R509A	VLRKTAHVNILLFMG	2.4 ±0.35
25	504–518 H510A	VLRKTRAVNILLFMG	2.7 ±0.4
26	504–518 N512A	VLRKTRHVAILLFMG	NB
27	504–518 l513A	VLRKTRHVNALLFMG	2.69± 0.35
28	504–518 L514A	VLRKTRHVNIALFMG	1.02 ±0.14
29	504–518 F516A	VLRKTRHVNILLAMG	0.57 ±0.08
30	504–518 M517A	VLRKTRHVNILLFAG	0.54 ±0.15
31	505–518	LRKTRHVNILLFMG	0.19 ±0.13
32	505–518 R506L	LLKTRHVNI LLFMG	0.55 ±0.105

Open in a separate window

Table 3.

Structure-activity of cyclic BRAF Dimer Interface Peptides

ID	BRAF Residues	Sequence	Kd (μM)
33	504–517 cyclo (505, 516) L505C, F516C	VCRKTRHVNILLCM	0.36 ±0.32
34	505–518T5080, N512A, l513E	LRKORHVAELLFMG	NB
35	505–518 cyclo(508, 513) T5080, l513E	LRKORHVNELLFMG	0.78±0.10
36	505–518 cyclo(508, 513) T5080, N512A, l513E	LRKORHVAELLFMG	0.46±0.04
37	505–518 cyclo(508, 513) 1508K, I513E	LRKKRHVNELLFMG	1.89±0.33
38	505–518 cyclo(508, 513) T508K, N512A, l513E	LRKKRHVAELLFMG	0.061±0.01

Open in a separate window

A scrambled peptide (3, 2.96 μM) was previously reported as a negative control with no effect on BRAF functional activity¹⁶. In contrast, however, when tested in the ITF assay, was shown to have similar and even slightly improved binding compared to the native BRAF sequence. Alignment of its sequence with BRAF shows that the key DIF interface residues are preserved in this peptide. Peptide 4 revealed that L505 plays little role in binding to the DIF as evidenced by an equipotent Kd of the alanine replacement peptide (compared to 1). From the crystal structure (4E26) it was observed that the first arginine of the sequence (R506) potentially disfavors binding due its role in making an adjacent salt bridge less energetically favorable. Synthesis and testing of the R506E mutant confirmed this hypothesis as shown by the lowered Kd value (5, 1.09 μM). Removing the charge in the R506L mutant was even more beneficial for inhibiting the dimer in the doubling of activity and 8-fold (6, 0.54 μM) relative to the native BRAF sequence. Replacement of T507 with both Asp (7) and Ala (8) resulted in a marginal increase in binding as indicated by the slightly reduced Kd values of these peptides. H510F had a complete loss of affinity.

Two hydrophobic residues (10, V511; 11, L514) were also shown to contribute to the affinity of the WT BRAF as evidenced by decreased binding of the alanine mutants, but more so in the L514A context. The conservative mutations at L515 included Ile and homoleucine and these were shown to have minimal effect and a significant increase in potency respectively (12, L515I, 4.1 μM; 13, L515homoLeu,1.25 μM). Computationally, the F516D mutation (14) was predicted to increase binding however the opposite effect was observed with no binding being detected using the intrinsic tryptophan fluorescence assay.

Truncation of the C-terminal tripeptide, YST to generate BRAF 503–518 resulted in a 2-fold increase in binding affinity (15, 1.88 μM) relative to BRAF 503–521. Further truncation of the N-terminal glycine from this molecule led to a more dramatic increase of almost 30-fold (Table 2; 17, 0.13 μM). Replacement of the C-terminal carboxylate with an amide group (18, 0.48 μM) resulted in a significant loss of affinity (3–4-fold). Furthermore, acetylation of the N-terminal amine (19, 0.8 μM) led to an even greater drop-off in binding affinity. With the observed 30-fold decrease in Kd for N and C-terminally truncated peptide (17, 504–518) relative to the initial 503–521 sequence tested (1, 3.84 μM), it was determined that this was an ideal scaffold to further investigate the contributions of each residue to binding through alanine scanning mutagenesis (Table 2). Results from testing of L505A (21, 0.45 μM), R506A (22, 0.36 μM), F516A (29, 0.57 μM), M517A (30, 0.54 μM), revealed that these positions are relatively insensitive to substitution compared to others including R509A (24, 2.4 μM), H510A (25, 2.7 μM), N512A (26, no binding) and I513A (27, 2.7 μM) which underwent a more than 10-fold potency drop-off. L514A (30, 1.02 μM) had a significant but less drastic decrease in affinity. After completion of the alanine scan, peptides were synthesized performing additional truncations from both the N and C-terminal ends. Truncation of the N-terminal valine producing BRAF 505–518 resulted in essentially equipotent binding affinity (31, 0.19 μM) relative to 17 while removal of the C-terminal glycine from 15 had a much more consequential effect and led to a steep drop off in activity (16, 5.75 μM). Relative to 1 however, it can be seen that truncation of the initial DIF peptide from both N and C-terminus by 1 and 4 residues respectively led to compromised binding (16, 5.75 μM). These studies suggest that BRAF residues 505–518 are the minimal recognition motif for potent dimer breaking activity.

Design of cyclized BRAF peptides with enhanced binding affinity.

From the crystal structure of the BRAF dimer, it was observed that residues 505–516 are in close proximity and that the sequence between these residues forms a loop without the side chains of the termini contributing to the protein-protein interaction (Figure 1). This observation suggested that a cyclic peptide could be generated to rigidify the peptide, stabilize this loop conformation and thereby decrease the entropic cost of binding in turn generating more potent inhibitors relative to the linear sequences. Recently there has been considerable attention given to the development of macrocyclic compounds as potential drug leads and where cyclization has been shown to increase potency, impart cell permeability and furthermore improve metabolic stability¹⁹.

To obtain proof of concept for macrocyclic BRAF inhibitors, the disulfide cyclized molecule was synthesized incorporating cysteine residues replacing L505 and F516 which are close in the crystal structure and whose side chains when replaced with alanine resulted in diminution but not total loss of activity. Testing of this cyclic peptide showed that it underwent a 15-fold increase in binding (Table 3, 33, 0.36 μM) relative to similar linear sequence (16) therefore confirming hypothesis for constraining the peptide sequence. In addition to these points of cyclization (504–518, Figure 4C), another potential connection point for a covalent restraint is between the side chains of position 508 and 513 which occur as Thr and Ile respectively in the native sequence. Analysis of the crystal structure suggests that these residues also are in close proximity and do not contribute directly to binding. An orthogonal cyclization strategy was designed for introduction of a lactam bridge between these residues through use of highly acid sensitive side chain protection. Synthesis of cyclo 508–513 T508O, I513E revealed that substantial activity was observed for the cyclic molecule (35, 0.78 μM) whereas no detectable binding was obtained for its linear counterpart (34) which contained the same mutations but not in a lactam bridge. Interestingly substitution of N512 for alanine resulted in a potency increase for the non-native substitution (36, 0.46 μM). Expansion of the lactam bridge by one methylene through use of lysine instead of ornithine in the orthogonal protection strategy resulted in a potency decrease in the N512 context (37, 1.89 μM) but a substantial increase in the N512A cyclic version (38, 0.061 μM).

Open in a separate window

Figure 4.

Structure of Peptide 38 bound to BRAF (From PDB ID 4E26).

A. The major binding residue R509 (magenta), smaller (508–513, cyan), and larger macrocyclization sites (505–516, yellow and orange residues replaced with cysteines in peptide 33) are highlighted. Other important residues for BRAF interaction are labeled including H510, A512 and L515. B. Close in view of the arginine handshake motif that is provides for a major part of the affinity of the BRAF dimer. R509 (peptide) forms an arginine handshake (anti-parallel binding mode) with Arg509 (BRAF) as observed from the crystal structure. The charge-charge repulsion of the two guanidinium groups is offset by the interaction of the positive charge with the negative charge on the C-terminal end of the α-C helix created by the helix dipole. The ATP competitive kinase inhibitor (from 4E26) is shown as a space filling representation to indicate proximity of the DIF to the catalytic site. C. Close in view of the bound conformation of peptide 38 (side chains with green carbons, one letter residue codes) with BRAF (side chains with yellow carbons, three letter residue codes) and illustrating the specific stabilizing non-bonded interactions. Pi-cation, H-bond and salt bridge interactions are shown as light brown, green and dark brown dashed lines respectively.

Isothermal Titration Calorimetry (ITC) of BRAF binding of DIF Peptides.

As confirmation of binding affinity in an alternate format and to investigate the thermodynamics of binding for DIF peptides to BRAF, ITC experiments were carried out for three peptides shown to interact with BRAF through ITF measurements (See supplementary Figure 2 for a Kd curve generated by ITC). Peptide 8 (ITF Kd = 2.8 μM) was shown to be somewhat less potent by ITC however the Kd result was variable (Kd = 14.9 ± 10.8 μM; ΔH = −34.8 kJ/mol; ΔS= −28.4 J/mol·K). Peptide 17 (ITF Kd = 0.13 μM) had a comparable affinity as measure by ITC with a K_d found to be 0.35 ± 0.17 μM. As expected for a flexible linear peptide (also for 8) with high entropy in the free state, the binding is primarily enthalpy driven (ΔH=−199 kJ/mol) and the entropy term is unfavorable for the overall free energy (ΔS=−567J/mol·K). The cyclic derivative, 36 also had a comparable Kd as measured by ITC (K_d = 0.31 ± 0.16 μM) to that obtained with the ITF assay (K_d = 0.46 μM). As expected and in line with the rationale for generating cyclic peptides, binding of the macrocyclic peptide is driven more by entropic factors than enthalpic (ΔH=−9.41 kJ/mol and ΔS= 92.05 J/mol·K).

DIF Peptides block paradoxical activation of BRAF induced by vemurafenib.

As described above, a major driving force for the development of DIF inhibitors is to provide an alternate therapeutic strategy to overcome the observed resistance to approved BRAF inhibitors. Drugs such as vemurafenib cause “paradoxical activation” of the MEK/ERK pathway in RAS mutant cells, thereby contributing to primary and acquired BRAF inhibitor resistance^{10, 11, 26, 27}. Since this resistance mechanism requires an intact dimer interface¹², the ability of DIF peptides (Peptide 1, Table 1 and FAM-TAT-Peptide17) were tested for their ability to overcome signaling downstream of BRAF induced by the ATP competitive inhibitor vemurafenib (PLX 4032). As shown previously for NRASQ61K driven human Sbcl2 melanoma cells in the absence of DIF targeting peptides^{12, 28}, BRAF inhibition induces paradoxical MEK/ERK pathway activation in control peptide treated Sbcl2 cells, as reflected by increased levels of phosphorylated MEK (pMEK) and ERK (pERK). When introduced through electroporation, 1 effected a dose dependent inhibition of paradoxical MEK/ERK activation induced by this drug in a NRAS^Q61K mutant melanoma cells. Due to the inefficiencies of electroporation, relatively high concentrations were required however almost complete inhibition of vemurafenib induced phosphorylation of MEK and ERK was observed at 300 μM peptide. To test the effects of the TAT-DIF peptide, shown to strongly inhibit the growth of the melanoma cell line (Figure 1), and to transfect DIF inhibitory peptides in a more gentle manner, Sbcl2 cells were treated with FAM-TAT-Pep17 and FAM-TAT-Pep6alaNC3. Interestingly, this paradoxical signalling effect was absent in FAM-TAT-Pep17 treated cells, and these also showed a reduction in RAF-1 levels. This is of particular interest as RAF-1 plays an important role as a MEK activator in NRAS driven melanoma, (including the Sbcl2 model)²⁹, and represents an important dimerization partner for BRAF in models of paradoxical ERK pathway activation by either drug-bound or kinase-dead BRAF molecules^{12, 30}. Furthermore, to monitor an important biological endpoint of MEK/ERK signaling in melanoma cells, we assessed the expression and phosphorylation of FRA1, the product of the immediate early gene FOSL1. FOSL1 has a well-documented role as an ERK target gene and phosphorylation of its product, FRA1 at S265 (pFRA1), serves as an excellent read-out for long-term persistence effects of ERK activity³¹. As shown in Figure 3, vemurafenib induced FRA1 expression and phosphorylation in control cells but not in FAM-TAT-Pep17 treated cells. In summary these data conclusively show that introduction of peptides into Sblc2 melanoma cells quenches vemurafenib induced paradoxical ERK pathway activity.

Open in a separate window

Figure 3.

Peptide 1 and FAM-TAT-Pep17 impair vemurafenib induced paradoxical ERK pathway signaling

induced by vemurafenib (PLX4032) in NRAS^Q61K-mutant SBCl2 melanoma cells. A. Cells were electroporated with the BioRad GenePulser XCellTM in the presence of the indicated concentrations of peptide 1¹⁶. Following recovery at 37 °C for 30 min, the cells were treated with 1 μM PLX4032 for 1 h or DMSO as a vehicle control. Subsequently, the cells were harvested, lysed using RIPA buffer and analyzed by Western blotting using the indicated antibodies as described previously¹². B. NRAS^Q61K mutant human Sbcl2 melanoma cells were incubated with 3.60 μM of FAM-TAT-Pep6AlaNC3 (control) or FAM-TAT-Pep17 for three days. Four hours prior to harvest, the cells were treated with 1 μM vemurafenib (PLX4032) or the same volume of DMSO as vehicle control. RIPA buffer lysates were subjected to Western blotting using the indicated antibodies. Detection of HSP90 serves as a representative loading control. Note that vemurafenib upregulates the expression and phosphorylation of MEK, ERK and its target FRA1 in FAM-TAT-Pep6AlaNC3 treated control cells, while this response is not observed in the presence of FAM-TAT-Pep17. Shown is one representative experiment out of two independent biological replicates with comparable outcome.

DISCUSSION

The results from the design and Kd evaluation of BRAF DIF peptide inhibitors, demonstration of their antiproliferative activity and subsequent proof of concept in their ability to block downstream activation of MEK/ERK promoted by vemurafenib, confirms the potential of this strategy for next generation BRAF inhibitors. By monitoring the expression and phosphorylation of the transcriptional target and protein substrate of ERK, FRA1, the observation has been made that DIF peptide inhibitors impair a critical node in melanoma pathobiology, since FRA1 plays an important role in NRAS and BRAF driven melanoma^32,33. Targeting the dimer interface through such type IV kinase inhibitors should also be an effective strategy against non-V600E BRAF point mutants or BRAF fusion proteins that require an intact dimer interface and are intrinsically vemurafenib resistant as a result of their increased homodimerization and/or mutations potentially interfering with drug uptake^{35, 36}. Indeed, genetic approaches using the R509H mutation revealed an essential role for the paradoxical activity of the kinase-dead BRAF^D594A mutant (PMID 22510884). BRAF mutations affecting D594, a residue essential for catalysis, represent the third most common class of BRAF mutants in cancer (PMID 20141835) and these and other kinase-inactivating or impairing mutations even outnumber the canonical V600E substitution in tumor entities such as non-small cell lung cancer (PMIDs 30653256, 29540830). Likewise, the R509H mutation impairs the transforming potential of intermediate activity BRAF mutants found in cancer and RASopathy patients such as Q257R and F595L (PMID: 22510884) and oncogenic BRAF fusion proteins (PMID 23533272), which are frequently encountered in human cancer due to advances in sequencing technologies (PMID 29540830). Insensitivity to ATP competitive inhibitors makes treatment of these tumors problematic and therefore new modalities such a dimerization inhibitors would provide new therapeutic options.

Having demonstrated in cellular studies the importance and relevance of the DIF for RAF kinase drug discovery, the detailed structure-activity relationship information obtained from testing of the peptide libraries provides key insights into the minimal recognition motif and the binding determinants of dimer blocking. This information will be of profound importance in the optimization of binding affinity, metabolic stability and drug-like properties of the peptidic inhibitors in both the linear and cyclic contexts and lead to a drug-like macrocyclic RAF kinase dimer breaker. Furthermore, the SAR data can be interpreted in structural terms through examination of the protein-protein interactions present in the dimeric BRAF (Figure 4A-​-C,C, PDB ID 4E26) and the extrapolation of these to the peptide-protein context. In the following discussion, residues of the BRAF peptide are given as one letter amino acid codes while those of the BRAF protein are described by 3 letter residue codes. Firstly, in the BRAF 504–518 context, the L505A mutation leads to a potency decrease not through loss of binding contacts but more likely through disruption of a stabilizing intramolecular interactions with F516 explaining the almost 4-fold potency loss of peptide 21 relative to 17. R506 when mutated to alanine resulted in a 3-fold loss of potency and which can be rationalized by absence of contacts to Asp448 (Figure 4C). The alanine side chain however has additional interactions with F516 offsetting the loss of the ion pairing interaction. The K507A peptide proved to be highly insoluble and therefore could not be tested. In the crystal structures, K507 is adjacent to Asp448 however does not appear to have a close ionic interaction and therefore its replacement should have minimal impact. T508 was demonstrated in the larger peptide context to be relatively insensitive to substitution in that its replacement with both Asp and Ala led to a mild increase in binding affinity. Both the T508 backbone amide NH and its side chain hydroxyl makes a bifurcated H-bond to the backbone carbonyl of L505 and therefore the loss of the side chain H-bond may result in an increase in strength of the backbone-backbone interaction.

As expected and validating the mutagenesis studies with the full-length BRAF¹², R509 in the peptide inhibitor context is very sensitive to mutation since R509A has an 18-fold loss of activity. R509 has an unusual binding motif since it is buried and forms an antiparallel interaction with the corresponding residue in the other monomer of the dimeric structure (Figure 4B and ​andC).C). The charge-charge repulsion of this motif is probably offset by side chain ionic interactions with the helix dipole of the other monomer. If the charges are stabilized on the two guanidine nitrogens (since a guanidinium group has three resonance structures; one from each monomer or from the peptide-monomer complex) that interact with the helix dipole, then the repulsive forces between the two arginine residues would be minimized due to the distance. R509 also interacts with Trp450 of BRAF through a pi-cation interaction (Figure 4C) further illustrating the contributions of this residue.

H510 makes an extensive network of intermolecular H-bonds (with the backbone carbonyl of His447 of the other BRAF momomer) and intramolecular hydrogen bonds (with N512 sidechain stabilizing a reverse turn) and therefore explaining its similar potency loss to R509 when mutated to alanine. The complete loss of binding for the N512A substitution is harder to rationalize in structural terms since it only has intramolecular contacts as opposed to H510 which also has intermolecular interactions. BRAF residues 510–513 (HVNI) make up a type II β-turn and therefore it is likely that the alanine substitution has more of a conformational effect. Statistical analyses of β-turns indicate that asparagine is among the most frequently occurring residues at the i+2 position (His is the i position) and alanine is one of the least probable³⁴. It is thus likely that in the linear context, the His-Ala substitution is quite destabilizing for the turn structure and in turn has a significant impact on the presentation of the binding determinants and entropy of binding. I513 (the i+3 residue in the turn) has few protein contacts but possesses important sidechain-sidechain interactions to both T508 and H510. The loss of these interactions in the alanine replacement peptide and the significant loss in binding affinity underscore the importance of the β-turn conformation to efficient binding to BRAF in the linear peptides.

From the crystal structure, it appears that L514 makes neither intra nor intermolecular contacts in line with a more moderate potency decrease after modification to alanine. Mutation probably results in increased rotational entropy of the alanine containing peptide disfavoring nucleation of the β-turn structure. The importance of L515 to binding is underscored by an extensive network of interactions with the lipophilic portion of the side chain of Arg509 and with the backbone of His510. Arg509 of the BRAF monomer receptor forms a bridge over the pocket that the cognate R509 residue from the peptide inserts into. L515 also has moderate Van der Waals contacts with M517 thereby providing intramolecular conformational stabilization. Further to this, synthesis and testing of F516A suggests a less involved role of the aromatic side chain in binding to BRAF. The 3-fold drop in affinity was consistent with the interactions of the sidechain both intramolecularly (as mentioned above with L505 and R506) and intermolecularly with Arg509 of the monomer. M517 alanine substitution led to a loss comparable to that induced by F516 mutation. This was explained by numerous contacts to Arg509 of the monomer and to His510 of the BRAF protein.

As described above, truncation studies were included in the peptide library design since optimization of molecular weight is a key parameter defining drug-likeness and oral bioavailability in beyond rule of 5 (bRo5) space^{20, 23} (For a direct comparison of the peptides in the truncation series see Supplementary Table 1). Truncation of BRAF 503–521 from both the N and C-termini resulted in an optimized sequence (17, 504–518) with a sub-micromolar Kd. The 30-fold potency enhancement can be surmised from the BRAF dimer crystal structure and likely results from optimal positioning of the C-terminal carboxylate (i.e. when G518 is the C-terminal residue) for interaction with the R506 sidechain to form a putative cyclic structure stabilized by an ion pairing interaction. Removal of the charge by replacement with a C-terminal amide (18) and the resulting decrease in binding provides additional evidence for this. The 10-fold lower affinity of the Gly N-terminal extension (G503, 15 compared with 17) is harder to explain in structural terms however it appears that in this sequence the free amine of the N-terminus (positively charged) is positioned to compete with R506 for binding to Asp448. This alternative binding mode may in turn promote conformations that are sub-optimal in terms of interactions of the rest of the peptide with BRAF. The peptide with the N-terminal valine (31) removed had similar potency to 16 further confirming that this residue does not in itself contribute significantly to binding but that the side chain to head salt bridge interaction promotes a favorable bound conformation.

As described above in the initial rationale for the project, the existence of a loop structure in the dimer interface provided the impetus for stabilization of a potential peptidic inhibitor through a covalent restraint. Due to the β-turn structure, the BRAF sequence 505–516 loops back on itself with Leu505 and Phe516 being close to each other in space. The observation that these residues do not contribute significant intermolecular interactions makes them ideal candidates to generate a cyclic peptide. Peptide 33 has a disulfide bridge replacing these two residues which led to a 16-fold increase in affinity as measured by the Kd value compared to 16, the native BRAF linear counterpart. This confirms that macrocyclic peptides stabilizing the loop conformation provide a beneficial contribution to the binding free energy and this occurs putatively through decrease in the entropic cost of binding. With disordered linear peptides, their-TΔS contribution to binding is positive when the molecule assumes a more rigid bound conformation (ΔS = low S_bound-high S_free) thereby overall making ΔG more negative. Cyclization generally lowers the S_free value by paying the entropic penalty during the cyclization reaction. Further to the 505–516 cyclized molecule, another cyclic variant was generated by incorporating a lactam (cyclic amide) bridge between residues 508 and 513 which also are close in space in the crystal structure. This would generate a smaller 6 residue cyclic peptide compared to the larger 12 residue 505–516 macrocycle. While T508 makes minimal contributions to binding, I513 substitution led to a 20-fold decrease in affinity due to the loss of intramolecular contacts. Decreased interactions in replacing the hydrophobic side chain with glutamate should be compensated however through the formation of the covalent restraint since this would make the intramolecular contacts redundant. The results obtained for the six residue (and with the 12mer) macrocycles fully validate the initial design rationale in generating cyclic molecules in that highly potent ligands for BRAF have been obtained and thus have the potential for further development as next generation therapeutics.

The data obtained for testing the 4 cyclic 508–513 variants provides insights into the conformational determinants required for optimal binding to BRAF and for macrocyclic drug discovery of this cancer target. It is apparent from the Kd results (Table 3) that the ornithine derivatives are binding in a suboptimal conformation since they are less potent than the linear derivatives (peptides 35, 36). The extra methylene in the lysine to glutamate cyclized molecule (38, Figure 4A) provides the necessary flexibility to promote more favorable non-bonded interactions with BRAF. It is therefore the best framework for optimization and occurs when the native asparagine residue at position 512 is exchanged for an alanine. The Kd of 61nM for Peptide 38 (K508, N512A, E513) provides strong conformation for the macrocycle approach in that a cyclic DIF inhibitor was the most potent compound identified. In contrast, in the native context (N512), the lysine cyclized molecule 37 (K508, E513) was a weaker binder than either of 35 (O508, E513) and 36 (O508, N512A, E513), a somewhat anomalous result. A possible explanation is that residue 512, as discussed above, is likely crucial to stabilization of the β-turn (HVNI), therefore should have a significant effect on the overall cyclic conformation and how tightly the molecule binds. Since the N512A exchange in the linear context led to loss of binding, computational analysis was performed for the cyclic peptides and suggested that the alanine substitution in the peptide 38 context allows more flexibility for the turn structure which may be necessary for optimal binding given the covalent bridge between residues 508 and 513 (Figure 4A).

The results obtained from the ITF assay were validated using isothermal titration calorimetry and confirmed that the binding affinities for three peptides compared favorably between the two methods. ITC data also corroborated the macrocyclic design hypothesis in that the linear peptides had favorable enthalpy of binding and the cyclic version had beneficial entropic contributions to ΔG as expected. Results for 35 were in line with the observation that the reduced affinity for the ornithine cyclized version is due to an unfavorable conformation, leading to suboptimal interactions with the BRAF pocket and with the result that adding an extra methylene group to the bridge alleviates this leading to significantly increased binding affinity.

CONCLUSIONS

The preliminary validation for the approach of targeting the dimerization interface of BRAF (and heterodimers with other RAF kinases) through peptide inhibitors has been successful. Despite the clinical success of BRAF inhibitors like vemurafenib in treating metastatic melanoma, resistance has emerged through “paradoxical MEK/ERK signaling” where transactivation of one protomer occurs as a result of drug inhibition of the other partner in the activated dimer. Through a structure-guided approach, linear and macrocyclic peptides targeting the dimerization interface have been identified and shown to bind with high affinity to BRAF. The lead cyclic molecule is a type IV kinase inhibitor and represents a promising scaffold for macrocyclic drug discovery where further ring stabilization, truncation, N-methylation and incorporation of non-natural amino acids have been shown to improve drug-likeness and pharmacological properties. Furthermore, DIF peptides efficiently exhibit anti-proliferative activity and inhibit paradoxical signalling in malignant melanoma cells stimulated by vemurafenib as evidenced by decreased levels of phosphoMEK/ERK and downregulation of the ERK target gene FOSL1. Targeting the dimer interface through type IV BRAF kinase inhibitors provides a new strategy to target non-V600E BRAF point mutants or BRAF fusion proteins since these require an intact dimer interface, are vemurafenib resistant and therefore pose a significant challenge. Overall DIF inhibitors show considerable promise for further development as next generation RAF kinase inhibitors as anti-tumor therapeutics.

EXPERIMENTAL SECTION

Supplementary Material

ACKNOWLEDGEMENTS

ABBREVIATIONS USED

Footnotes

REFERENCES