Analyses of the oncogenic BRAFD594G variant reveal a kinase-independent function of BRAF in activating MAPK signaling

Class 3 mutations in B-Raf proto-oncogene, Ser/Thr kinase (BRAF), that result in kinase-impaired or kinase-dead BRAF have the highest mutation frequency in BRAF gene in lung adenocarcinoma. Several studies have reported that kinase-dead BRAF variants amplify mitogen-activated protein kinase (MAPK) signaling by dimerizing with and activating WT C-Raf proto-oncogene, Ser/Thr kinase (CRAF). However, the structural and functional principles underlying their activation remain elusive. Herein, using cell biology and various biochemical approaches, we established that variant BRAFD594G, a kinase-dead representative of class 3 mutation-derived BRAF variants, has a higher dimerization potential as compared with WT BRAF. Molecular dynamics simulations uncovered that the D594G substitution orients the αC-helix toward the IN position and extends the activation loop within the kinase domain, shifting the equilibrium toward the active, dimeric conformation, thus priming BRAFD594G as an effective allosteric activator of CRAF. We found that B/CRAF heterodimers are the most thermodynamically stable RAF dimers, suggesting that RAF heterodimers, and not homodimers, are the major players in determining the amplitude of MAPK signaling in cells. Additionally, we show that BRAFD594G:CRAF heterodimers bypass autoinhibitory P-loop phosphorylation, which might contribute to longer duration of MAPK pathway signaling in cancer cells. Last, we propose that the dimer interface of the BRAFD594G:CRAF heterodimer may represent a promising target in the design of novel anticancer therapeutics.

The mitogen-activated protein kinase (MAPK) 2 pathway is important in regulating cell proliferation and survival. The duration and amplitude of MAPK signaling are mainly controlled by the RAF Ser/Thr kinase family (ARAF, BRAF, and CRAF), that contain three conserved regions (CR) for membrane recruitment (CR1), N-terminal 14-3-3 regulation (CR2), and catalysis (CR3) (1). Among the three RAF isoforms, BRAF has the highest basal activity and is most frequently mutated in 7-8% of all human cancers (2). The high mutation propensity in cancer led to extensive interest in the development of BRAFspecific drugs.
WT BRAF requires RAS-initiated membrane recruitment and homo-or heterodimerization with other RAF isoforms for activation. Although all BRAF mutations trigger constitutive activation of MAPK signaling in cancer cells, they have distinct activation mechanisms and have been grouped into three main classes to reflect different biochemical properties: RAS-and dimer-independent (class 1), RAS-independent but dimer-dependent (class 2), and RAS-and dimer-dependent (class 3) (3). Class 1 mutations include the most common BRAF mutation, V600E, which accounts for ϳ90% of all BRAF mutations (2). The prominence of BRAF V600E led to Food and Drug Administration (FDA) approval of two ATP-competitive drugs, vemurafenib and dabrafenib (4). These ATP-competitive drugs elicit effective clinical outcomes for melanoma patients carrying the V600E mutation; however, paradoxical activation and drug resistance restrict their application to cancers driven by class 2 and 3 BRAF mutations (3,5,6).
Inactivating BRAF mutations, class 3, are currently the most common BRAF mutation in lung adenocarcinoma (7). Herein, we study a representative of class 3 BRAF mutations, BRAF D594G , which is the most frequent kinase-dead mutation. Asp-594 is part of the highly conserved DFG motif found in all kinases. The aliphatic residues Phe/Gly help to maintain the closed inactive conformation by occupying the ATP binding pocket, whereas the Asp residue is a key catalytic residue involved in Mg 2ϩ coordination for the phospho-transfer reaction. Therefore, the mutation of Asp-594 to Gly prevents Mg 2ϩ coordination, rendering BRAF catalytically inactive.
Intriguingly, the "kinase-dead" BRAF mutation has been validated as a driver of lung carcinoma (8,9), suggesting that BRAF has functions that are not limited to catalyzing MEK phosphorylation but extend to other roles that are independent of enzyme activity. In parallel, the FDA-approved BRAF inhibitors inhibit the catalytic activity of WT BRAF while boosting the noncatalytic function of BRAF to further up-regulate the MAPK signaling, which leads to secondary malignancies in clinic (5,6,10). This phenomenon, known as "paradoxical activation," is a major concern surrounding current BRAF inhibitors. Apparently, ATP-competitive inhibitors rearrange the structural elements of BRAF kinase to augment the noncatalytic functions, resulting in fundamentally different biological outcomes. Our current view of RAF regulation is mainly derived from mechanistic studies of enzymatically active RAF kinases by evaluating their catalytic capability. The biochemistry of noncatalytic BRAF functions is poorly understood.
The unknown noncatalytic behavior of BRAF generates therapeutic limitations of ATP-competitive RAF drugs, including paradoxical activation and drug resistance. It remains a challenge to directly target kinase-dead BRAF mutants with traditional ATP-competitive inhibitors. Novel approaches are urgently needed. Here, we integrate biophysical, biochemical, and computational approaches together with cellular assays to explore the noncatalytic function of BRAF in cancer. Additionally, our study sheds light on the activation and regulation mechanism of RAF kinase family under physiological and pharmacological conditions.

BRAF D594G is more prone to dimerization than WT BRAF
Dimerization through the kinase domain is required for WT BRAF and non-Val-600 BRAF mutations to function in normal and cancerous cells (8,11). Previously, Heidorn et al. (12) have shown that kinase-dead BRAF D594A dimerizes with CRAF when transiently transfected into D04 NRAS Q61L cancer cells, and Yao et al. (8) determined that BRAF D594G has a stronger binding potential than WT BRAF to RAS. To complement the previous cellular studies, we examined how this mutation affects the biochemical properties of BRAF in its purified form. Thus, we purified the catalytic domain (CD) of BRAF D594G from Escherichia coli through affinity and size-exclusion chromatography with a purity of Ͼ90% suitable for biophysical analyses (Fig. 1A). CD-BRAF D594G was subjected to velocity analytical ultracentrifugation (AUC) to determine the mass of macromolecules in solution. The majority of our protein sedimented with a sedimentation coefficient (s) of ϳ3.8 S, which relates to dimeric BRAF identified previously by Grasso et al. (13) (Fig. 1B). WT CD-BRAF and a small population of CD-BRAF D594G exist at ϳ2.6 s, which relates to monomeric BRAF, previously identified by our laboratory and others ( Fig.  1B) (13)(14)(15).
Next, we conducted molecular dynamics (MD) simulations on monomeric and dimeric WT BRAF and monomeric BRAF D594G to elucidate the structural basis for enhanced dimerization. The starting structures are WT monomeric (PDB entry 4WO5 (14)) and dimeric (PDB entry 4E26 (16)) kinase domain, respectively. The BRAF D594G initial structure is constructed by mutating the WT monomer crystal structure (PDB entry 4WO5 (14). The inhibitors in the crystal structure were removed. We did not include ATP molecules, because BRAF D594G mutation disables ATP binding (Fig. S1). One hallmark of BRAF activation is the movement of the ␣C-helix, with respect to orientation toward the C-lobe, from OUT (inactive) to IN (active) conformation. The production runs of the simulations were carried out for 200 -400 ns. The distances between the center of masses of the ␣C-helix and the C-lobe were measured as on average 30 Ϯ 0.44, 28.5 Ϯ 0.45, and 25.6 Ϯ 0.30 Å, for monomeric BRAF WT , monomeric BRAF D594G , and dimeric BRAF WT , respectively. This emphasizes that the ␣C-helix of monomeric BRAF D594G is closer to the C-lobe than the monomeric WT counterpart (Fig. 1, C and D). In the ␣C-helix-IN conformation, residues in the ␣C-helix stabilize RAF dimers through hydrophobic interactions (Arg-509 with Trp-450) and H-bonds (Arg-509 with Thr-508/Arg-506/Leu-505). Our previous MD simulations together with structures of monomeric BRAF kinase domain support that a fully extended activation loop stabilizes active BRAF, whereas a rigid secondary-structure element, termed helix AS-H1, maintains an inactive conformation of BRAF (14,17). The possibility of forming secondary structures, including 3 10 helix and ␣-helix, were summarized in Table 1. The MD simulations show that the activation loop of BRAF D594G is in a transition between the AS-H1 and extended loop, which is supported by the loss of the ␣-helical structure of the activation loop, whereas aa 598 -600 form a short 3 10 helix ( Fig. 1 (D and E) and Table 1). The formation of the 3 10 helix is a product of the torsional strain placed on the N terminus of the activation loop as the C terminus unwinds.
Together, these observations support that monomeric CD-BRAF D594G adopts a partially active conformation indicating higher dimerization potential than CD-BRAF WT , as demonstrated by the AUC data (Fig. 1D). Additionally, adopting the partially activated state instead of fully activated state minimizes BRAF D594G homodimerization that occurs when the ␣C-helix and activation loop are fully IN and extended, as evidenced by oncogenic BRAF V600E (17).

Dimerization potential correlates with the kinase activity of RAF dimers
It was previously established that BRAF:CRAF heterodimers are the most active RAF dimers in vitro (15). It remains elusive how this translates to ERK hyperactivation by BRAF D594G : CRAF. In light of this, we transfected CRAF alone or co-transfected BRAF:CRAF or BRAF D594G :CRAF into HEK293T cells and immunoprecipitated MBP-tagged CRAF using amylose resin to obtain either CRAF alone or BRAF:CRAF and BRAF D594G :CRAF heterodimers, respectively. We added the MBP tag to enhance the solubility of CRAF for our in vitro experiments. The heterodimers have much higher activity than CRAF in cells and in vitro (Fig. 2, A and B). The activity of BRAF D594G :CRAF is comparable with that of BRAF:CRAF. Thus, the kinase activity of BRAF is dispensable for transactivation of CRAF. Additionally, the observed high kinase activity rules out the possibility of forming stable RAF homodimers, as both BRAF D594G and CRAF homodimers have no or very minimal kinase activity. This is consistent with our previous studies on BRAF:CRAF heterodimers, which demonstrated that separately purified BRAF and CRAF homodimers form stable BRAF:CRAF heterodimers after mixing in vitro (15).

Oncogenic BRAF D594G reveals kinase-independent function
As mentioned previously, D594G is commonly found in lung cancer. To observe what effects lung cancer cells have on heterodimer activity, we performed a similar experiment in HTB-177 (KRAS Q61H ) lung cancer cells. For this experiment, we transfected CRAF, D594G:CRAF, and BRAF:CRAF in HTB-177 cells and monitored their activity (Fig. S2). As expected, Figure 1. Catalytic domain of BRAF D594G has increased dimerization in solution as compared with its WT counterpart. A, CD-BRAF D594G was purified and analyzed by Coomassie Blue-stained SDS-PAGE. The molecular weight marker (M) and CD-BRAF D594G are identified above the gel. B, purified CD-BRAF D594G and WT CD-BRAF were investigated using velocity AUC. Monomeric BRAF or dimeric BRAF is labeled above the image at ϳ2.5 or ϳ4 s, respectively. C, molecular dynamics trajectory signifying the center of mass distance from the ␣C-helix toward the C-lobe as a function of simulation time are shown for monomeric WT CD-BRAF (violet), monomeric CD-BRAF D594G (green), and dimeric WT CD-BRAF (blue). D, overlay of the representative snapshot from the MD simulations for monomeric WT CD-BRAF (violet), CD-BRAF D594G (green), and dimeric WT CD-BRAF (blue). The ␣C-helix and activation loop are highlighted with dark blue boxes. E, the stability of the activation loop for monomeric WT CD-BRAF, CD-BRAF D594G , and dimeric WT CD-BRAF was measured by looking at the root mean square deviation (RMSD), with respect to the average structure of monomeric WT CD-BRAF (featuring the AS-H1 helix throughout the whole trajectory). To understand why the BRAF:CRAF heterodimer forms readily, we applied MD simulations to investigate the atomistic details of BRAF:CRAF heterodimerization. We started with the dimeric CRAF (PDB entry 3OMV (6)) and dimeric BRAF (PDB entry 4E26 (16)) crystal structures with any bound inhibitors removed. The BRAF D594G :CRAF heterodimer is constructed by replacing one protomer of the BRAF dimer with a CRAF protomer (extracted from the CRAF dimer structure) via alignment between the incoming and outgoing protomer, as well as introduction of D594G mutation to the remaining BRAF protomer. The production simulations of BRAF:BRAF, BRAF D594G :CRAF, and CRAF:CRAF were carried out for 200 -300 ns. By comparing the gain or loss and the strength of H-bonds, we conclude that BRAF:CRAF heterodimers are thermodynamically more stable than the corresponding homodimers. The interprotomer interaction can be illustrated by analyzing the distance between BRAF:CRAF Arg-509/Arg-401 (H-bond donors) and their corresponding H-bonding partners on the other protomer (Fig. 2, C and D). The calculated distance (ϳ3 Å) for BRAF:CRAF and BRAF:CRAF dimers is ideal for H-bond formation between two protomers, but not for CRAF:CRAF dimer. In addition to the Arg-509 H-bonds, BRAF D594G :CRAF heterodimers have strong interprotomer interactions between the following donor/acceptor pairs: His-510/His-369, His-402/His-477, and Arg-398/Asp-448. The average number of H-bonds (Table 2) is the sum of the fraction of H-bonds that occur in the simulations during a period of 150 -200 ns. Thus, the BRAF D594G :CRAF heterodimer forms the strongest dimer, followed by BRAF:BRAF, and the CRAF: CRAF homodimer is the weakest of the three.
Together, the increased dimerization potential of the heterodimer is reflected by the increased BRAF:CRAF heterodimer activity in cellulo and in vitro. CRAF homodimer forms the weakest dimer, reinforcing the notion that CRAF has the lowest basal activity alone.

Oncogenic BRAF D594G reveals kinase-independent function The stability of BRAF D594G :CRAF heterodimers is 14-3-3-dependent
The formation of BRAF:CRAF heterodimers occurs in normal cells through RAS-induced membrane recruitment (5,6,18,19). The factors promoting dimerization of oncogenic BRAF D594G :CRAF remain controversial. In light of this, we created a series of BRAF D594G truncation constructs: ⌬N-tail (region containing the BRAF-specific region), ⌬RAS-binding domain (⌬RBD), ⌬CR2, and ⌬SSDD by truncating from the N-terminal toward the kinase domain to identify which regions of BRAF D594G are crucial for dimerization with CRAF ( Fig. 3A). We transiently transfected the BRAF D594G deletion constructs: ⌬N-tail, ⌬RAS-binding domain (⌬RBD), ⌬CR2, and ⌬SSDD into HEK293 cells with or without WT CRAF (Fig. 3B). The BRAF D594G constructs and CRAF have poor expression as single transfections, but when co-transfected together, the expression increases on average 9-and 2-fold for BRAF D594G and CRAF, respectively, suggesting that formation of BRAF D594G : CRAF protects both proteins from degradation. The observed change in expression of both BRAF and CRAF can be used as a tool to determine whether heterodimer pairs form in cells. All of the BRAF D594G constructs containing the SSDD motif were able to stabilize CRAF (Fig. 3B). It is not surprising that ⌬SSDD lost the ability to activate CRAF, as the SSDD motif is important for proper regulatory spine alignment, which is a prerequisite for dimerization and activation of RAF family members (13).
14-3-3 chaperone protein binding to pSer-729 of BRAF enhances homo-or heterodimerization of WT RAF (20,21). AMPK has been identified to phosphorylate the BRAF C-terminal 14-3-3-binding site, Ser-729 (21). To verify the effect of 14-3-3 on BRAF D594G :CRAF dimerization, we treated cells expressing RAF heterodimers with dorsomorphin 2HCl (compound C), a potent AMPK inhibitor. The phosphorylation of Raptor, a well-established protein substrate of AMPK, was decreased in the presence of compound C, verifying that AMPK is potently inhibited (Fig. 3C). Additionally, total CRAF, pCRAF, and pSer-729 were decreased in the context of the BRAF D594G :CRAF heterodimer (Fig. 3C). To further verify the role of 14-3-3 binding, we introduced the S729A mutation to mimic unphosphorylated BRAF D594G . Unfortunately, BRAF D594G/S729A was not expressed well in HEK293 cells, preventing more detailed investigation. This is consistent with the findings of Park et al. (22), who show that the mutation to S729A results in poor stability of purified BRAF from insect or mammalian cells. Regardless, the observed destabilization by the S729A mutation together with the compound C data suggest that 14-3-3 binding to pSer-729 stimulates BRAF D594G : CRAF heterodimerization and subsequent CRAF activation. Because phosphorylation of Ser-729 was not completely depleted by the AMPK inhibitor, these results suggest that there are other kinases, other than AMPK, that are responsible for Ser-729 phosphorylation. In parallel, AMPK inhibitor has little effect on pSer-729 and pCRAF-Ser-338 in the context of WT BRAF:CRAF heterodimers, suggesting that they have a distinct regulation mechanism.

BRAF D594G :CRAF heterodimers bypass autoinhibitory phosphorylation in cancer cells
We previously identified that BRAF WT can autophosphorylate its P-loop (Ser-465/Ser-467) to inactivate itself in the absence of native substrate MEK (15). By doing so, BRAF is subject to fine-tuned regulation to ensure rapid termination of RAF signaling. To determine whether BRAF:CRAF and BRAF D594G :CRAF use a similar mechanism, we incubated these kinases with ATP for 30 min to let autophosphorylation complete prior to quantifying their specific activity. The activity of BRAF:CRAF was decreased by ϳ40 -50% upon autophosphorylation ( Fig. 4A), suggesting that autophosphorylation of certain residues negatively regulates the kinase activity of BRAF. Intriguingly, no obvious change was detected for BRAF D594G : CRAF ( Fig. 4A), demonstrating that oncogenic mutations can bypass autophosphorylation to sustain pathway signaling. To further verify this finding, we monitored total BRAF autophosphorylation via 32 P autoradiograph. As shown in Fig. 4B, BRAF WT was able to autophosphorylate itself and MEK; however, BRAF D594G :CRAF could only phosphorylate BRAF D594G and MEK, and not CRAF (Fig. 4B). The lack of CRAF phosphorylation within BRAF D594G :CRAF dimers provides compelling evidence that RAF autophosphorylation occurs in trans, not in cis.
To further verify this model, we introduced the R509H mutation, a mutation known to disrupt the "side-to-side" dimer interface of BRAF, to BRAF V600E . As monomeric BRAF V600E/R509H is still active, our data clearly demonstrate that autophosphorylation requires an intact dimer interface and occurs as transrather than cis-autophosphorylation (Fig. 4, A and B). Our finding highlights the physiological significance of the "side-toside" dimer interface, other than allosteric trans-activation. Within a BRAF dimer, BRAF trans-phosphorylates both activation loop and inhibitory phosphorylation sites to fine-tune the overall BRAF activity (Fig. 4, A and B) (15,17).

Oncogenic BRAF D594G reveals kinase-independent function P-loop phosphorylation destabilizes BRAF dimers
Our mechanistic studies identified that P-loop phosphorylation negatively regulates the kinase activity of BRAF WT (17). We carried out MD simulations on dimeric BRAF WT kinase domain with P-loop phosphorylation (pSer-465/pSer-467) and compared it with the simulations of WT unphosphorylated dimer. Both simulations started from the same crystal structure (PDB entry 4E26 (16)). Our MD data demonstrate that phosphorylation of the P-loop disfavors dimer formation by disrupting multiple H-bonds in the dimer interface, whereas these H-bonds remained intact in the unphosphorylated dimer (Fig.  4C). In addition, the two phosphorylated protomers also moved apart from each other (black trajectory, Fig. 4C), relative to the unphosphorylated protomers (blue trajectory, Fig. 4C), further validating the distorted dimer configuration. Other dimer simulations, including WT dimer and P-loop phosphorylation with inhibitors and WT dimer with the R509H mutation, show consistent results of weakened dimeric interaction upon P-loop phosphorylation or introduction of the R509H mutation (Fig.  S3). Consistent with our MD simulations, co-immunoprecipitation experiments verified that introduction of the phosphomimetic mutation S465D/S467D (SS/DD) to the P-loop compromised the dimerization ability of BRAF, as shown in Fig. 4D. Together, we conclude that P-loop phosphorylation decreases the formation of BRAF dimers and most likely displaces the ␣C-helix to an OUT position that leads to an inactive conformation.

Inhibiting the catalytic activity of CRAF within the BRAF D594G :CRAF
The two FDA-approved BRAF drugs dabrafenib and vemurafenib have a positive clinical outcome for patients harboring oncogenic BRAF V600E , but are less effective toward non-Val- Oncogenic BRAF D594G reveals kinase-independent function 600 BRAF mutants (4,9). Other laboratories have suggested that targeting MEK or EGFR is a viable treatment option for class 3 BRAF tumors (8), but currently no RAF directed therapies are available. There are three types of BRAF inhibitors, which are grouped by how they orient the ␣C-helix or DFG motif (4). Type I inhibitors, such as SB-590885 (SB), lock BRAF into an active conformation with ␣C-helix-IN and DFG-IN. Type I 1 ⁄ 2 inhibitors, including dabrafenib, vemurafenib, PLX-4720, and PB-PLX7904 (PB), orient the ␣C-helix-OUT and DFG-IN conformation. Type II inhibitors, such as AZ-628 (AZ), adopt the ␣C-helix-IN and DFG-OUT conformation. The ␣C-helix-IN (type I/II) configuration allows for inhibitor binding to both protomers of BRAF with equal potency; however, type I 1 ⁄2 inhibitors lock the ␣C-helix-OUT configuration to sterically hinder occupancy for the second protomer, which leads to negative cooperativity. Although each category of inhibitor is effective at inhibiting BRAF V600E (17), they have very different binding mechanisms that regulate dimeric WT BRAF differently. WT BRAF displays drug resistance and paradoxical activation toward type I 1 ⁄2 inhibitors.
Here we evaluate the different categories of RAF inhibitors against BRAF D594G :CRAF heterodimers. Differential scanning fluorimetry (DSF) results show that all of the tested ATP-competitive inhibitors bind to CD-BRAF D594G (Fig. 5A), although BRAF D594G could not bind to ATP (Fig. S1). However, it remains unclear whether these inhibitors can trigger paradoxical activation in cells expressing BRAF D594G . Additionally, our MD simulations have shown that the D594G:CRAF heterodimer is in a different conformation than the CRAF homodimer; therefore, different drug response is expected between RAF homo-and heterodimers. To address these concerns, we applied cell-based assays to evaluate the potency of the selected inhibitors in HEK293 cells transiently transfected with BRAF D594G and CRAF. BRAF V600E and WT BRAF/CRAF were used as the controls (Fig. 5, B and C).
Pan-RAF inhibitors, such as AZ, are active against ARAF, BRAF, and CRAF with equal potency; thus, titrating different

Oncogenic BRAF D594G reveals kinase-independent function
concentrations of AZ showed inhibition at 0.1 M (Fig. 5D). The potency of AZ toward BRAF D594G :CRAF is not surprising, because this inhibitor has been shown to have low nanomolar potency toward CRAF in cell-free experiments (23). Additionally, Noeparast et al. (24) show a similar inhibition profile toward the D594G:CRAF heterodimer. PB is a paradox-breaking inhibitor designed to minimize paradoxical activation (25), but type I 1 ⁄ 2 inhibitors have much lower activity toward dimeric RAF, and no significant activity of PB toward the BRAF D594G : CRAF heterodimer was observed (Fig. 5E). Although AZ and SB target monomeric and dimeric BRAF with similar potency, SB only shows inhibition at the highest concentration tested toward BRAF D594G :CRAF (Fig. 5, D and F), most likely due to its low potency toward CRAF. We also tested PLX-4720, which is known to potently inhibit monomeric BRAF, and it only shows slight inhibition of the BRAF D594G :CRAF heterodimer at the highest concentration of drug used (Fig. 5G). AZ, a representative of type II pan-RAF inhibitors, exhibits inhibition of the BRAF D594G :CRAF heterodimer at the lowest concentrations tested. Additionally, Karoulia et al. (4) have identified that pan-RAF inhibitors have a slower off rate and dissociate the RAF/ MEK complex. These data suggest that pan-RAF inhibitors might be a good scaffold for rationally designed BRAF D594G : CRAF inhibitors.
Notably, none of the ATP-competitive inhibitors induced paradoxical activation even though binding of the DFG-IN inhibitors dabrafenib, PLX-4720, and SB shows slight enhancement of BRAF D594G :CRAF heterodimerization (Fig. 5H). On the contrary, WT BRAF:CRAF heterodimers were paradoxically activated by dabrafenib under the same conditions ( Fig.  5C) (24). Due to the correlation of BRAF D594G and P-loop autoinhibitory phosphorylation, we introduced an S465A/S467A (SS/AA) mutation to WT BRAF to prevent P-loop phosphorylation. No paradoxical activation was observed for BRAF SS/AA , suggesting that P-loop phosphorylation plays a role in paradoxical activation (Fig. 5I). Under this scenario, kinase-dead BRAF abolishes P-loop phosphorylation of catalytically active CRAF within the BRAF D594G :CRAF dimer, which also evades paradoxical activation. This model is consistent with the work of Holderfield et al. (26), who highlighted that ATP-competitive inhibitor relieves inhibitory P-loop phosphorylation of WT RAF to trigger paradoxical activation. Another possibility is that the BRAF D594G mutation already makes BRAF a perfect allosteric activator and inhibitor binding makes no further enhancement regarding its ability to transactivate CRAF (Fig.  5J).

Oncogenic BRAF D594G reveals kinase-independent function
inhibitor completely abolished MEK/ERK signaling at 1 M, which indicates that AZ can inhibit the MAPK pathway in a short time period. However, long exposure of 96 h still exhibited ϳ50% of cell growth at the highest concentration (5 M) of drug treated (Fig. 5L), suggesting that BRAF D594G :CRAF heterodimers display drug resistance toward ATP-competitive inhibitors. This is consistent with another group who showed that AZ alone cannot fully eliminate growth of cells harboring various RAS mutations (27). The discrepancy between the amount of drug required for MAPK inhibition and cancer cell death highlights that pan-RAF inhibitors might not be a viable therapeutic treatment for class 3 BRAF mutations.

RAF dimer breaker is active against BRAF D594G :CRAF
Recognizing that dimerization of BRAF D594G and CRAF is critical for both protein stability and kinase activity poises the dimer interface as a promising drug target. Previously, we developed a 10-mer peptide (Braftide) that shows efficient inhibition and degradation of WT BRAF and BRAF G469A both in vitro and in a variety of cells exhibiting either RAS dependence or independence, respectively (28). To extend the same study to BRAF D594G :CRAF heterodimers, we titrated 0 -100 M TAT-Braftide into HEK293 cells overexpressing the BRAF D594G : CRAF heterodimer (Fig. 6A). This peptide inhibitor shows not only inhibition of MAPK signaling, but also degradation of the BRAF D594G :CRAF heterodimer at as low as 25 M. We applied a similar experiment to WM3629 cancer cells (Fig. 6B). TAT-Braftide exhibits degradation of BRAF D594G and CRAF at concentrations as low as 2.5 M in WM3629 cells. We treated WM3629 cells with TAT-Braftide for 48 h to observe the activity of this peptide by cell viability assays, which showed dosedependent inhibition of cell growth and an EC 50 of 11.05 Ϯ 1.12 M (Fig. 6C). Together, these results highlight that BRAF D594G : CRAF heterodimers are still sensitive to dimer breakers, despite the fact that it has a stronger dimer interface than BRAF homodimers.

Discussion
In our study, we focused on BRAF D594G , a representative of kinase-dead BRAF mutants, to better understand the noncatalytic function of BRAF. Interestingly, protein kinases have evolved to have allosteric scaffolding properties, in addition to their inherent capability of catalyzing the transfer of the ␥-phosphate from ATP to tyrosine, serine, or threonine residues in the substrate proteins. Pseudokinases are an example of enzymes that have diverged to primarily function as scaffolds, such as kinase-impaired HER3 which allosterically activate other EGFR/HER family members through asymmetric dimerization (29). Another example is kinase-suppressor of RAS (KSR) which has been shown to function as a scaffolding protein to allosterically regulate RAF and MEK (30). BRAF kinase has both scaffold function and catalytic activity that are integrated to determine the final output of RAF signaling. Intriguingly, ATP-competitive inhibitors can differentially alter the two functions to trigger paradoxical activation, in which the scaffold function of BRAF is exaggerated to compensate the inhibited kinase activity, and the overall outcome is enhanced RAF signaling. However, it is hard to pinpoint which role is more significant than the other. The kinase-dead BRAF mutant provides a valuable tool to elucidate the allosteric consequence of mutations that are linked to cancer.
Our biophysical experiments and MD simulations show that BRAF D594G is more prone to dimerization than BRAF WT due to the positioning of the ␣C-helix toward the IN conformation. In addition, the activation loop of BRAF D594G is beginning to unwind from the AS-H1 loop toward full extension, which is a prerequisite for kinase activation. The extended activation loop stabilizes the regulatory spine (R-spine) and ␣C-helix into a conformation suitable for dimerization (30); thus, BRAF D594G is in a partially active, dimerization-competent conformation leading to enhanced allosteric function. Interestingly, the most recent cryo-EM structure of BRAF in complex with ATP analog

Oncogenic BRAF D594G reveals kinase-independent function
ATP␥S suggests that ATP binding is critical to maintain the autoinhibited state (22). Because BRAF D594G cannot bind ATP, this feature further stabilizes the active conformation of BRAF to support RAF dimerization.
Experiments with other pseudokinases (VRK, MLKL, STRAD, ERBB3, and JAK2) have emphasized that ATP binding can enhance allosteric activation of the associated partner for some pseudokinases (31)(32)(33)(34). However, BRAF D594G does not bind ATP but still functions as an allosteric activator. In addition, the A481F or A373F mutations, which interfere with ATP binding to BRAF or CRAF, respectively, have been shown to constitutively dimerize with and activate the RAF binding partner (35). These results combined with ours suggests that the alignment of the catalytic spine is not necessary for the scaffold function of BRAF.
Enhanced signaling required to drive MAPK pathway activation in cells is attained by either increased enzyme activity or reduced negative regulatory feedback effects (36). We show that BRAF D594G :CRAF heterodimers are more active than CRAF alone and have an activity that is comparable with that of the BRAF:CRAF heterodimer. This increase in activity could be attributed to BRAF D594G :CRAF heterodimers bypassing inhibitory P-loop autophosphorylation that negatively controls WT RAF activity in cells (17). The glycine-rich P-loop (Gly-X-Gly-X-X-Gly motif) is highly conserved among protein kinases and is known to position the ␣-/␤-phosphates of ATP for optimal ␥-phosphoryl transfer during catalysis. In addition, it is believed that the hydrophobic interactions between P-loop and activation loop stabilize the inactive configuration of the kinase (37). Protein kinases share a universal activation mechanism, in which conformational changes in three conserved structural motifs at the active site (the activation loop, the DFG motif, and the ␣C-helix) determine the transition from the inactive to the active conformation (38). The effect of P-loop phosphorylation has not been well-studied. Our MD simulation results demonstrate that phosphorylated P-loop disfavors the side-to-side dimer configuration, which in turn supports the nonproductive position of the ␣C-helix. Here, we have confirmed the existence of a cross-talk among the P-loop and the three motifs that modulates the conformational equilibrium of the BRAF kinase. These data unveil P-loop as an important structural feature involved in positioning the ␣C-helix, depending on its phosphorylation status. Given that the Ser-465/Ser-467 residues are highly conserved among the P-loop (Fig. S4, A and B), it is appealing to propose a similar role for the conserved Ser/Thr residues in other kinase families.
Previously, the molecular mechanism of RAF regulation has been mostly derived from structures of isolated kinase domains or fragments of other domains. The cryo-EM structures of fulllength BRAF-MEK1-14-3-3 complexes were solved recently (22), providing a more comprehensive structural characterization of RAF. In those structures, both autoinhibited and fully activated BRAF were captured, demonstrating that MEK1 and 14-3-3 are both involved in stabilizing BRAF in the active or inactive conformations (22). Our findings introduce an additional layer of complexity to the precise mechanism of BRAF activation. Our laboratory previously identified that the presence of MEK1 decreases overall BRAF autophosphorylation while increasing phosphorylation of the activation loop in vitro, suggesting a new allosteric role of MEK1, other than functioning as a substrate of RAF kinases (15). In light of those findings, we propose one model for RAF regulation under physiological conditions, as shown in Fig. 7. After RAS-initiated RAF dimerization, MEK binding primes the activation loop of RAF for trans-phosphorylation (15), which leads to maximal RAF activation and subsequent MEK phosphorylation. Upon dissociation of phosphorylated MEK, RAF trans-phosphorylates the P-loop to diminish its own kinase activity. Meanwhile, this negative regulation loop can be dampened by protein phosphatases to maintain the highest RAF activity. In this model, P-loop autophosphorylation is a self-sustained mechanism to self-limit its kinase activity in vivo.
Two main models have emerged to shed light on the molecular mechanism of paradoxical activation. The first model suggests that binding of RAF inhibitors increases dimerization of drug-bound RAF with inhibitor-free RAF, which results in RAS-dependent MEK/ERK activation (5). Similarly, Heidorn et al. (12) have related this model to how kinase-dead BRAF activates CRAF in a RAS-dependent manner. The second model identifies that RAF inhibitors activate MEK/ERK by relieving inhibitory RAF autophosphorylation (26). In our work, we provide evidence reconciling the two models of paradoxical activation. We show that inhibitory autophosphorylation occurs in trans, not in cis; thus, oncogenic BRAF D594G can evade this negative feedback loop by eliminating autophosphorylation of CRAF in the BRAF D594G :CRAF heterodimer due to the lack of enzyme activity. RAF ATP-competitive inhibitors do not dis- Oncogenic BRAF D594G reveals kinase-independent function play paradoxical activation in the context of BRAF D594G :CRAF heterodimer, providing direct evidence that relief of P-loop phosphorylation by RAF inhibitors is capable of activating MAPK signaling. Additionally, BRAF D594G and CRAF readily form a stable dimer; hence, RAF inhibitors could no longer boost the allosteric function of BRAF.
The increased frequency of non-Val-600 mutations detected in lung and other cancers highlights the urgency of targeting class 3 BRAF (39). Although there have been several proposed strategies to combat dimer-dependent BRAF mutations, such as targeting CRAF, MEK, and EGFR (3,9,24), there are currently no clinically approved drugs that are designed specifically to treat tumors harboring class 3 BRAF mutations. Since the approval of targeting BRAF V600E mutations with vemurafenib or dabrafenib in combination with MEK inhibitors, this has been a suggested strategy to treat other BRAF mutations. Combination therapy with RAF and MEK inhibitors in cell lines harboring class 3 BRAF mutations has been analyzed previously (8,24). These experiments show that MEK inhibitors alone or in combination with RAF inhibitors decrease MAPK pathway activation and induce cellular apoptosis. Although treatment with MEK inhibitors is promising, cytotoxicity is an inevitable concern; thus, more specific inhibitors are urgently needed. Our results on inhibiting phosphorylation of Ser-729 of BRAF make the C-terminal 14-3-3 binding site a druggable target for class 3 BRAF mutations. Similarly, Kondo et al. identified that targeting the ␣I-helix, at the 14-3-3-binding interface, could have improved therapeutic efficacy in dimer-dependent BRAF mutations (40).
Braftide activity toward BRAF D594G :CRAF heterodimers in cells suggests that allosteric inhibitors targeting the dimer interface could have an improved therapeutic index. In addition, we show that pan-RAF ATP-competitive inhibitors show initial inhibition of BRAF D594G :CRAF heterodimer activity, but prolonged exposure fails to abolish WM3629 cancer cell growth. Moreover, Montagut et al. (41) suggest that overexpression of CRAF, as we see with the co-transfection of BRAF D594G :CRAF, can lead to acquired resistance to pan-RAF inhibitors, suggesting that targeting the ATP-binding pocket of CRAF alone might not be the best therapeutic strategy. Hence, rationally designed drugs with improved selectivity are required to target CRAF or the dimer interface for class 3 BRAF mutations.
Altogether, our work complements the current understanding of kinase-impaired/kinase-dead BRAF mutations, while highlighting key biochemical properties of the most common kinase-dead mutation, BRAF D594G . These results identify how BRAF D594G :CRAF heterodimers sustain increased MEK/ERK activation in cancer cells, while exposing a new allosteric site, at the dimer interface, to target class 3 BRAF mutations.

Cell lines
WM3629 and HTB177 were purchased from Rockland and HTCC, respectively. HEK293 cells were a gift from Dr. Catherine Moore (University of the Sciences).

CD-BRAF purification
CD-BRAF-6xHIS was transformed into BL21 codon plus E. coli cells and grown at 37°C to an optical density of 600 nm of 0.6 -0.8, and then 1 mM isopropyl 1-thio-␤-D-galactopyranoside was added and the cells were induced at 16°C with stirring at 200 rpm. The harvested pellet was resuspended in lysis buffer (50 mM HEPES buffer, pH 7.4, 250 mM NaCl, 50 mM KCl, 10 mM imidazole, 10 mM ␤-mercaptoethanol, 10% glycerol, and protease inhibitor tablet (Sigma, catalog no. 11836170001)) and then lysed with a sonicator. The lysed suspension was centrifuged to separate the clarified supernatant, and then the supernatant was applied to pre-equilibrated nickel resin and incubated with rotation at 4°C for 1-2 h. After resin incubation, the superna-Oncogenic BRAF D594G reveals kinase-independent function tant/resin slurry was added to a gravity column and washed with wash buffer (50 mM HEPES, pH 7.4, 250 mM NaCl, 10 mM imidazole, and 10% glycerol). 50 -250 mM imidazole was added into low-salt buffer (50 mM HEPES, pH 7.4, 250 mM NaCl, and 10% glycerol) to elute the protein, which was pooled and concentrated, using a Millipore concentrator (#UFC803024), before injecting onto size-exclusion chromatography (column: Superdex 200 10/300 GL; #28-9909-44). The corresponding peaks were analyzed with SDS-PAGE and then pooled, concentrated, and aliquoted. These aliquots were flash-frozen in liquid nitrogen and stored at Ϫ80°C.

Analytical ultracentrifugation
The velocity AUC experiment was performed as described previously (42). Briefly, 10 -15 M WT CD-BRAF and CD-D594G were centrifuged at 42,000 r.p.m. for ϳ8 h while data were collected over the entire time frame. The results were analyzed with SEDFIT to calculate the continuous c(S) distribution, and data were graphed using Origin.

Molecular dynamics simulations
The molecular dynamics simulations were performed as described previously (17). In brief, the crystal structures that were acquired from the respective PDB files were cleaned up and solvated in a periodic box using TIP3P water molecules. The ff14SB (43) force field was used on the proteins. GPUimplemented Amber18 was utilized for each MD simulation (44). All systems were equilibrated using the same protocol for solvent minimization, heating, and NPT simulation at 1 atmosphere and 300 K. Production runs using the NVT system at 300 K were then applied for 200 -400 ns per system. The trajectories were then investigated with the cpptraj program in the Amber18 package for hydrogen bond interactions, distance, root mean square deviation, etc. as conveyed in the results sections describing Fig. 1, 2 and 4.

Transient transfection into mammalian cells
HEK293 cells were transiently transfected as described previously (17). In brief, 0.5-1 million HEK293 cells were plated onto 12-or 6-well plates, respectively, and then were incubated overnight until the cells reached 40 -60% confluence. DNA (0.5-2 g) was mixed with PEI-MAX in a 1:3 ratio. DNA:PEI-MAX was added to fresh DMEM on the HEK293 cells. The cells were washed with ice-cold PBS after 24 -48 h and then lysed in 4% SDS. The harvested lysate was applied to homogenizer columns, and the flow-through protein concentration was analyzed using a BCA kit (Pierce, #23225). 10 -40 g of lysates were used for Western blotting analyses.

BRAF:CRAF co-immunoprecipitation
For immunoprecipitation, 5 million HEK293 cells were plated onto a 10-cm dish and incubated overnight at 37°C with 5% CO 2 . Once the cells reached a confluence of 40 -60%, the DMEM was aspirated out, and fresh DMEM was added. A total of 10 g of DNA (5 g ϩ 5 g for co-transfections) was mixed with 30 g of PEI-MAX for 25 min and then added to the plated cells for 48 h at 37°C with 5% CO 2 . For inhibitor treatment, the cells were washed with PBS, and then DMEM containing DMSO (0.5% final) or inhibitor was added to the cells and incubated for 1 h at 37°C with 5% CO 2 . Cells were washed again with ice-cold PBS and then lysed in 1 ml of lysis buffer (50 mM HEPES, pH 7.4, 150 mM NaCl, 0.1% Nonidet P-40 (IGE-PAL630), 1 mM EDTA, 5% glycerol, 1 mM phenylmethylsulfonyl fluoride, 20 mM ␤-glycerolphosphate, 2.5 mM sodium pyrophosphate, and protease inhibitor) for 2 h with rotation at 4°C. The lysate was clarified, and 0.5-1 mg of lysate was applied to 10 l of pre-equilibrated amylose or FLAG resin for 2 h with rotation at 4°C. The resin was washed 5ϫ (5 min each wash) with lysis buffer and then resuspended in 2ϫ dilution buffer (25 mM HEPES, pH 7.4, 0.125 mg/ml BSA, and 300 mM NaCl) for kinase assays or quenched with 4ϫ loading dye to be analyzed by Western blotting.

Western blotting-based kinase assay
The kinase assay was performed as described previously (17). In brief, the purified proteins or co-immunoprecipitation products were diluted with 2ϫ dilution buffer and combined with 2ϫ mixture buffer (25 mM HEPES, pH 7.4, 20 mM MgCl 2 , 1 mM DTT, 50 mM ␤-glycerolphosphate, 0.2-2 M MEK (0.1-1 M final), and 2000 M ATP) for 5-10 min, unless another time is indicated at 30°C. The reaction was quenched using 4ϫ loading dye and then analyzed by SDS-PAGE immunoblotting. The nitrocellulose membranes were probed with the suitable primary and secondary antibodies and imaged using fluorescencebased approaches on a GE Typhoon imager. Analyses of the immunoblots were analyzed using ImageJ software.

In-gel radioactive assay
As described previously (17), purified or co-immunoprecipitated proteins were mixed with a 2ϫ mixture buffer (25 mM HEPES, pH 7.4, 20 mM MgCl 2 , 50 mM ␤glycerolphosphate, 1 mM DTT, 4 mM Na 3 VO 4 , 200 M ATP, and 1 Ci of radioactive ATP). A final concentration of 1.3 M MEK was added into the mixture buffer for tests that contained MEK and ATP. The reaction was incubated at 30 min at room temperature before being transferred to SDS-PAGE, dried, and imaged on a Typhoon imager.

Differential scanning fluorimetry
CD-D594G aliquots were diluted in DSF buffer (25 mM HEPES, pH 7.0, 150 mM NaCl) to a concentration of 5 M Oncogenic BRAF D594G reveals kinase-independent function (final). To a 384-well PCR plate (Thermo Scientific, #AB-1384), 15 l of CD-D594G, 4 l of SYPRO Orange (5ϫ final), and 1 l of inhibitor in DMSO (25 M final) were added. The plate was sealed with adhesive seal sheets (Thermo Scientific, #AB-1170) and then spun down and heated from 20 to 95°C on QuantStudio 7 Flex (Applied Biosystems) with a ramp rate of 0.015°C/s. The protein melt curves were produced, and the first derivative of the curve was used to analyze the data. The data were examined and plotted using Origin.

Cell viability
The cell viability assay was performed as described previously (28). In brief, WM3629 cells were plated onto a 96-well plate for 24 h before treatment. TAT-Braftide or AZ-628 were titrated on to the cells at the indicated concentrations for 48 or 96 h, respectively. The drug medium was switched out every 24 h, and then the WST reagent was added, in a 1:10 ratio, and incubated for 4 h at 37°C. Absorbance readings were measured for the plate at 450 nm.