In vitro reconstitutions suggest a general model for paradoxical activation of ARAF, BRAF, and CRAF by diverse RAF inhibitor types that does not rely on negative allostery

Reviewer #1 (Public review):

Summary:

Tkacik et al describe their efforts to reconstitute and biochemically characterize ARAF, BRAF, and CRAF proteins and measure their ability to be paradoxically activated by current clinical and preclinical RAF inhibitors. Paradoxical activation of MAPK signaling is a major clinical problem plaguing current RAF inhibitors, and the mechanisms are complex and relatively poorly understood. The authors utilize their preparations of purified ARAF, BRAF, and CRAF kinase domains to measure paradoxical activation by type I and type II inhibitors, utilizing MEK protein as the substrate, and show that CRAF is activated in a similar fashion to BRAF, whereas ARAF appears resistant to activation. These data are analyzed using a simple cooperativity model with the goal of testing whether paradoxical activation involves negative cooperativity between RAF dimer binding sites, as has been previously reported. The authors conclude that it does not. They also test activation of B- and CRAF isoforms prepared in their full-length autoinhibited states and show that under the conditions of their assays, activation by inhibitors is not observed. In a particularly noteworthy part of the paper, the authors show that mutation of the N-terminal acidic (NtA) motif of ARAF and CRAF to match that of BRAF enhances paradoxical activation of CRAF and dramatically restores paradoxical activation of ARAF, which is not activated at all in its WT form, indicating a clear role for the NtA motif in the paradoxical activation mechanism. Additional experiments use mass photometry to measure BRAF dimer induction by inhibitors. The mass photometry measurements are a relatively novel way of achieving this, and the results are qualitatively consistent with previous studies that tracked BRAF dimerization in response to inhibitors using other methods. Overall, the paper establishes that WT CRAF is paradoxically activated by the same inhibitors that activate BRAF, and that ARAF contains the latent potential for activation that appears to be controlled by its NtA motif. The biochemical activation data for BRAF are qualitatively consistent with previous work.

Strengths:

While previous studies have put forward detailed molecular mechanisms for paradoxical activation of BRAF, comparatively little is known about the degree to which ARAF and CRAF are prone to this problem, and relatively little biochemical data of any sort are available for ARAF. Seen in this light, the current work should be considered of substantial potential significance for the RAF signaling field and for efforts to understand paradoxical activation and design new inhibitors that avoid it.

Weaknesses:

There are, unfortunately, some significant flaws in the data analysis and fitting of the RAF activation data that render the primary conclusion of the paper about the detailed activation mechanism, namely that it does not involve negative cooperativity between active sites, unjustified. This claim is made repeatedly throughout the manuscript, including in the title. Unfortunately, their data analysis approach is overly simplistic and does not probe this question thoroughly. This is the primary weakness of the study and should be addressed. A full biochemical modeling approach that accurately captures what is happening in the experiment needs to be applied in order for detailed inferences to be drawn about the mechanism beyond just the observation of activation.

The authors' analysis of their RAF:MEK "monomer" paradoxical activation data (Figures 1, 3, and Tables 1, 2) suffers from two fundamental flaws that render the resulting AC50/IC50 and cooperativity (Hill) parameters essentially uninterpretable. Without explaining or justifying their choice, the authors use a two-phase cooperative binding model from GraphPad Prism to fit their activation/inhibition data. This model is intended to describe cooperative ligand binding to multiple coupled sites within a preformed receptor assembly, and does not provide an adequate description of what is happening in this complicated experiment. Specifically, it has two fundamental flaws when applied to the analysis in question:

(a) It does not account for ligand depletion effects that occur with high-affinity drugs, and that profoundly affect the shapes of the dose-response curves, which are what are being fit

The chosen model is one of a class of ligand-binding models that are derived by assuming that the free ligand concentration is effectively equal to the total ligand concentration. Under these conditions, binding curves have a characteristic steepness, and the presence of cooperativity can be inferred from changes in this steepness as described by a Hill coefficient. However, many RAF inhibitors, including most of the type II inhibitors in this study, bind to the dimerized forms of at least one of the RAF isoforms with ultra-high affinity in the picomolar range (particularly apparent in Figure 1 with LY inhibiting BRAF). Under these conditions, the model assumption is not valid. Instead, binding occurs in the high-affinity regime in which the drug titrates the receptor and effectively all the added drug molecules bind, so there is hardly any free ligand (see e.g. Jarmoskaite and Herschlag eLife 2020 for a full description of this "titration" regime). The shapes of the curves under these conditions reflect the total amount of RAF protein (and to some extent drug affinity), rather than the presence of cooperativity. Fitting dose response curves with the chosen model under these conditions will result in conflating binding affinity and protein concentration with cooperativity.

(b) It does not model the RAF monomer-dimer equilibrium, which is dramatically modulated by drug binding, rendering the results RAF-concentration dependent in a manner not accounted for by the analysis.

The chosen analysis model also fails to consider the monomer-dimer equilibrium of RAF. This has two ramifications. Since drug binding is coupled to dimerization to a very strong degree, the observed apparent affinities of drug binding (reflected in AC50 and IC50 values) are functions of the concentration of RAF molecules used in the experiment. Since dimerization affinities are likely different for ARAF, BRAF, and CRAF, the measured AC50 values also cannot be compared between isoforms. This concentration dependence is not addressed by the authors. A related issue is that the model assumes drug binding occurs to two coupled sites on preformed dimers, not to a mixture of monomers and dimers. "Cooperativity" parameters determined in this manner will reflect the shifting monomer-dimer equilibrium rather than the cooperativity within dimers. Additionally, the inhibition side of the activation/inhibition curves is driven by binding of the drug to the single remaining site on the dimer, not to two coupled sites, and so one cannot determine cooperativity values for this process in this manner.

As a result of both of these issues, the parameters reported in the tables do not correctly reflect cooperativity and cannot be used to infer the presence or absence of negative cooperativity between RAF dimer subunits. To address these major issues, the authors would need to apply a data analysis/fitting procedure that correctly models the biochemical interactions occurring in the sample, including both the monomer-dimer equilibrium and how this equilibrium is coupled to drug binding, such as that developed in e.g., Kholodenko Cell Reports 2015. Alternatively, the authors should remove the statements claiming a lack of negative cooperativity from the manuscript and alter the title to reflect this.

Some other points to consider

(1) The observation that ARAF is not activated by type II inhibitors is interesting. A detailed comparison of the activation magnitudes between inhibitors and between A-, B-, and CRAF is hampered by the arbitrary baseline signal in the assay, which arises from a non-zero FRET ratio in the absence of any RAF activity. The authors might consider background correcting their data using a calibration curve constructed using MEK samples of known degrees of phosphorylation, so that they can calculate turnover numbers and fold activation values rather than an increase over baseline. This will likely reveal that the activation effects are more substantial than they appear against the high background signal.

(2) The authors note that full-length autoinhibited 14-3-3-bound RAF monomers are not activated by type I and II inhibitors. However, since this process involves the formation of a RAF dimer from two monomers, the process would also be expected to be concentration dependent, and the authors have only investigated this at a single protein concentration. Since disassembly of the autoinhibited state must also occur before dimerization, it might be expected to be kinetically disfavored as well. Have the authors tested this?

(3) ATP concentration modulates activation. While this is an interesting observation, some of this analysis suffers from the same issue discussed above, of not considering high-affinity binding effects. For instance, LY is not affected by ATP concentration in their data (Figure 4D), but this is easily explained as being due to its very tight binding affinity, resulting in titration of the receptor and the shape of the inhibition curve reflecting the amount of RAF kinase in the experiment and not the effective Kd or IC50 value.

Reviewer #2 (Public review):

This manuscript by Tkacik et al. uses in vitro reconstituted systems to examine paradoxical activation across RAF isoforms and inhibitor classes. The authors conclude that paradoxical activation can be explained without invoking negative allostery and propose a general model in which ATP displacement from an "open monomer" promotes dimerization and activation. The biochemical work is technically sound, and the systematic comparison across RAF paralogs (along with mutational/functional analysis) across inhibitor classes is a strength.

However, the central mechanistic conclusions are overgeneralized relative to the experimental systems, and several key claims, particularly the dismissal of negative allostery and the proposed unifying model in Figure 6, are not directly supported by the data presented. Most importantly, the absence of RAS, membranes, and relevant regulatory context fundamentally limits the physiological relevance of several conclusions, especially regarding the current clinical type I.5 RAF inhibitors and paradoxical activation.

Overall, this is a potentially valuable biochemical study, but the manuscript would benefit from more restrained interpretation, clearer framing of scope, and revisions to the model and title to better reflect what is actually tested.

(1) A central issue is that the biochemical system lacks RAS, membranes, 14-3-3 and endogenous regulatory factors that are known to be required for paradoxical RAF and MAPK activation in cells. As previous work has repeatedly shown and the authors also acknowledge, paradoxical activation by RAF inhibitors is RAS-dependent in cells, and this dependence presumably explains why full-length autoinhibited RAF complexes are refractory to activation in the authors' assays.

Importantly, the absence of paradoxical activation by type I.5 inhibitors in this system is therefore not mechanistically informative. Type I.5 inhibitors (e.g., vemurafenib, dabrafenib, encorafenib), but not Paradox Breakers (e.g., plixorafenib), robustly induce paradoxical activation in cells because binding of the inhibitor to inactive cytosolic RAF monomer promotes a conformational change that drives RAF recruitment to RAS in the membrane, promoting dimerization. The inability of the type 1.5 inhibitor to suppress the newly formed dimers is the basis of the pronounced paradoxical activation in cells. In the absence of RAS and membrane recruitment, failure to observe paradoxical activation in vitro does not distinguish between competing mechanistic models.

As a result, conclusions regarding inhibitor class differences, and especially the generality of the proposed model, should be substantially tempered.

(2) The authors argue that their data argue against negative allostery as a central feature of paradoxical activation. However, the presented data do not directly test negative allostery, nor do they exclude it. The biochemical assays do not recreate the cellular context in which negative allostery has been inferred. Further, structural data showing asymmetric inhibitor occupancy in RAF dimers cannot be dismissed on the basis of alternative symmetric structures alone, particularly given the dynamic nature of RAF dimers in cells.

Most importantly, negative allostery was proposed to explain paradoxical activation by Type I.5 RAF inhibitors, yet these inhibitors do not paradoxically activate in the assays presented here. The absence of paradoxical activation in this system, therefore, cannot be used to argue against a mechanism that is specifically invoked to explain cellular behavior not recapitulated by the assay.

(3) The model presented in Figure 6 is conceptually possible but remains speculative. Key elements of the model, including RAS engagement, membrane recruitment, 14-3-3 rearrangements, and the involvement of cellular kinases and phosphatases, are explicitly absent from the experimental system. Accordingly, the model is not tested by the data presented and should not be framed as a validated or general mechanism. The figure and accompanying text should be clearly labeled as a working or conceptual model rather than a mechanistically supported conclusion.

(4) The manuscript states that type I.5 inhibitors do not induce paradoxical activation in the biochemical assay because their C-helix-out binding mode disfavors dimerization. While this is true in isolation, it overlooks the well-established fact that type I.5 inhibitors (with the exception of paradox breakers) clearly promote RAS-dependent RAF dimerization in cells. This distinction is critical and should be explicitly acknowledged when interpreting the in vitro findings.

(5) The title suggests a general mechanism for paradoxical activation across RAF isoforms and inhibitor classes, whereas the data primarily address type I and type II inhibitors acting on isolated kinase-domain monomers. A more accurate framing would avoid the term "general" and confine the conclusions to C-helix-in (type I/II) RAF inhibitors in a reduced biochemical context.

Reviewer #3 (Public review):

Summary:

Tkacik et al. systematically characterized all three RAF kinase isoforms in vitro with all three types of RAF inhibitors (Type I, I1/2, and II) to investigate the mechanism underlying paradoxical activation.

In this study, the authors reconstituted heterodimers of A-, B-, and C-RAF kinase domains bound to non-phosphorylable MEK1 (SASA), mimicking the monomeric auto-inhibited state of RAF. These "RAF monomers" were tested for MEK phosphorylation with an increasing concentration of all three types of RAF inhibitors (Type I, I1/2, and II). This study is reminiscent of a previous study of the same team measuring RAF kinase activity in the presence of all three types of inhibitors in the context of dimeric RAF isoforms stabilized by 14-3-3 proteins (Tkacik et al 2025 JBC). RAF monomers had little to no activity at low concentrations of inhibitors (consistent with their "monomeric state"). Addition of type I1/2 inhibitor did not induce paradoxical activation as, in this context, they do not induce RAF dimerization required for activation, as observed by MP. Addition of type I and type II inhibitors led to paradoxical activation consistent with the RAF dimerization induced by these inhibitors, as observed by MP. Interestingly, type II inhibitors induced activation only for B- and C-RAF and not A-RAF.

At high concentrations of type II inhibitors, kinase activity is inhibited with a strong or weak positive cooperativity for BRAF and CRAF, respectively. This observation is very similar to what the authors previously observed with their dimeric RAF system. Interestingly, when the NtA motif is modified by phosphomimetic mutations in A- and C-Raf, basal kinase activity is stronger, but most importantly, inhibitor-induced paradoxical activation is much stronger with both type I and II inhibitors. This demonstrates that mutation of the NtA motif of ARAF and CRAF sensitized them to paradoxical activation by type II inhibitors.

The authors also tested the effect of ATP in the paradoxical activation observed in their RAF "monomer" system. As previously published in their assay with 14-3-3 stabilized dimeric RAF, the authors observed an expected shift of the IC50 with Type I inhibitors, while Type II inhibitors seem to behave as a non-competitive inhibitor. The authors next reconstituted the MAP kinase pathway (with RAF monomers at the top of the phosphorylation cascade) to test paradoxical activation amplification. Again, Type I1/2 inhibitors did not induce paradoxical activation, while Type I and II inhibitors did. The authors tested the inhibitors with FL auto-inhibited RAF/MEK/14-3-3 complexes, where, contrary to the "RAF monomers" experiments, FL B- and C-RAF were not paradoxically activated but were inhibited by all three types of inhibitors.

Overall, Tkacik et al. tackle an important question in the field for which definitive experiments and thorough biochemical investigation to understand the molecular mechanisms for the inhibitor-induced paradoxical activation are still missing, and of high importance for future drug development.

Strengths:

The biochemical experiments here are rigorously executed, and the results obtained are highly informative in the field to decipher the intricate mechanisms of RAF activation and inhibitor-induced paradoxical activation.

Weaknesses:

The interpretation of the results in the context of the current state of the art is ambiguous and raises questions about the relevance of introducing a new model for inhibitor-induced paradoxical activation, particularly since the findings presented here do not clearly contradict established paradigms. I believe some clarification and precision are required.

Main comments:

(1) Figure 2:

The authors comment on the expected greater increase (for a cascade assay) in the magnitude of ERK phosphorylation compared to what was observed for MEK phosphorylation. However, this observation might be reflective of the stoichiometries used in the assay, with 40 times more MEK compared to RAF concentration (250nm vs 6nM), which might favour pERK vs pMEK.

- The authors should clarify their rationale for the protein concentration used in this assay and explain how protein stoichiometry was taken into account for the interpretation of their results.

- In addition, the authors should justify comparing pMEK and pERK TR-FRET values when different anti-phospho antibodies were used. Antibodies may have distinct binding affinities for their epitopes. Could this not lead to differences in FRET signal amplitudes that complicate direct comparison?

(2) Supplementary Figure 2:

The author mentioned that the inhibitors did not activate the FL auto-inhibited RAF complexes; however, they did inhibit the TR-FRET signal.

- Can the authors comment on the origin of the observed basal activity? Would the authors expect self-release of the RAF kinase protein from the auto-inhibited state in the absence of RAS, leading to dimerization and activation? Alternatively, do the inhibitors at low-concentration relieve the auto-inhibited state, thereby driving dimerization and activation?

- Did the author test the addition of RAS protein in their in vitro system to determine whether "soluble" RAS is sufficient to release the protective interactions with RBD/CRD/14-3-3 and lead to inhibitor-induced paradoxical activation of FL RAF?

(3) Figure 5B:

The authors said that the Kd values obtained from their MP assay are consistent with prior studies of RAF homodimerization and RAF:MEK heterodimerization. While this is true from the previous studies of RAF:MEK interaction by BLI (performed from the same team), the Kd of isolated RAF kinase homodimerization has been measured around ~30µM by AUC in the cited ref (24,27 & 37).

- The authors should discuss the discrepancy between their Kd of homodimerization and the reported Kd values in the literature. At the concentration used for MP, it is surprising to observe RAF dimerization while the Kd of homodimerization has been measured at ~30µM (in the absence of MEK).

- Would the authors expect the presence of MEK to influence the homodimerization affinity for the isolated KD?

(4) Conclusions:

Several times in the introduction and the conclusion, the authors suggest that the negative allostery model (where "inhibitor binding to one protomer of the dimer promotes an active but inhibitor-resistant conformation in the other") is a model that applies to all types of RAF inhibitors (I, I1/2, and II).

However, from my understanding and all the references cited by the authors, this model only applies to type I1/2 inhibitors, where indeed the aC IN conformation in the second (inhibitor-free) protomer of the RAF dimer might be incompatible with the type I1/2 inhibitors inducing aC OUT conformation. The type I and type II inhibitors are aC IN inhibitors and are expected to bind both protomers from RAF dimers with similar affinities. Therefore, the negative allostery model does not apply to the type I and type II inhibitors. The difference in the mechanism of action of inhibitors is even used to explain the difference in the concentration range in which inhibitor-induced activation is observed in cells. The description of the state of the art in this study is confusing and does not help to properly understand their argumentation to revise the established model for paradoxical RAF activation.

- Can the authors clarify their analysis of the state of the art on the different mechanisms of action for the paradoxical activation of RAF by the different types of RAF inhibitors?

5) Conclusions:

"Our results suggest that negative allostery (or negative cooperativity) is not a requisite feature of paradoxical activation. The type I and type II inhibitors studied here induce RAF dimers and exhibit paradoxical activation but do so without evidence of negative cooperativity, nor do they appear to inhibit intentionally engineered RAF dimers with negative cooperativity (25). Indeed, type II inhibitors exhibit apparent positive cooperativity while type I inhibitors are non-cooperative inhibitors of RAF dimers (25)."

- Can the authors explain how results on the paradoxical activation induced by type I and type II inhibitors inform or challenge a model that specifically applies to type I1/2 inhibitors?

The authors often refer to their previous study (reference 25), where they tested the inhibition of all three types of inhibitors with engineered RAF dimers. While I agree with the authors that in reference 25 the Type I and type II inhibitors inhibit RAF dimers without exhibiting negative cooperativity (as expected from the literature and the current model), the authors did observe some negative cooperativity for Type I1/2 inhibitors in their study most particularly for the type I1/2 PB (with hill slope ranging from -0.4 to -0.9, indicative of negative cooperativity).
While the observations that type II inhibitors display positive cooperativity is both novel and very interesting, from what I understand the results from thakick et al 2025 and the current study appear more in line with the current paradigm in the field (which describe paradoxical activation with negative cooperativity for type I1/2 inhibitors and no negative cooperativity for the Type I and II inhibitors) rather than disapproving of the current model and supporting for a new model.

- In this context, can the authors clarify how their results challenge the current model for paradoxical activation?

(6) Conclusions:

The authors describe the JAB34 experiment from Poulikakos et al. 2010 to conclude that "While this experiment cleanly demonstrates inhibitor-induced transactivation of RAF dimers, it is important to recognize that the differential inhibitor sensitivity of the two subunits in this experiment is artificial - it is engineered rather than induced by inhibitor binding as the negative allostery model proposes."

Indeed, the JAB34 experiment demonstrated the inhibitor-induced transactivation, but the Poulikakos et al. 2010 study does not discuss differential inhibitor sensitivity. The negative allostery model was proposed later by poulikakos team in other papers (Yao et al 2015 and Karoulia et al, 2016), in which JAB34 was not used.

- Can the authors clarify how the JAB34 experiments question differential inhibitor sensitivity?

(7) Conclusions:

"Considering that the conformation required for binding of type I.5 inhibitors destabilizes RAF dimers, it is unclear how an inhibitor binding to one protomer would be able to transmit an allosteric change to the opposite protomer, if that inhibitor's binding causes the existing dimer to dissociate."

- The authors should comment on whether 14-3-3 proteins might overcome negative regulation by type I1/2 inhibitors, similar to what has been shown for ATP, which acts as a dimer breaker like type I1/2 inhibitors.

(8) Conclusions:

"Furthermore, the complex effects of type I.5 inhibitors on dimer stability and the clear resistance of active RAF dimers to these inhibitors complicates interpretation of inhibition data - weak or incomplete inhibition of an enzyme can be difficult to discern from true negative cooperativity (43). As we discuss below, the clear resistance of RAF dimers to type I.5 inhibitors is alone sufficient to explain their ineffective inhibition during paradoxical activation, without invoking negative allostery."

- The authors should explain how they reconcile this statement and their proposal of a new model that does not rely on negative allostery with their previous findings showing negative cooperativity for RAF dimer inhibition with type I1/2 inhibitors.

(9) Conclusions:

Here, the authors propose a new universal model to explain paradoxical activation of RAF by all types of RAF inhibitors:
" Our findings here, in light of structural studies of RAF complexes and prior cellular investigations of paradoxical activation, lead us to a model for paradoxical activation that does not rely on negative allostery and is consistent with activation by diverse inhibitor classes. In this model, the open monomer complex is the target of inhibitor-induced paradoxical activation (Figure 6). Binding of ATP to the RAF active site stabilizes the inactive conformation of the open monomer, which disfavors dimerization. Displacement of ATP by an ATP-competitive inhibitor, irrespective of class, alters the relative N- and C-lobe orientations of the kinase to promote dimerization (30, 35). Once dimerized, inhibitor dissociation from one or both sides of the dimer would allow phosphorylation and activation of MEK."

From my understanding, the novelty of this new model is twofold: a) the open monomer is the target of the inhibitor-induced paradoxical activation and b) once dimerized, inhibitor dissociation from one or both sides of the dimer would allow phosphorylation and activation of MEK.

Novelty a) implies, as the authors stated, that "Inhibitor-induced activation and inhibition act on distinct species - activation on the open monomer and inhibition on the 14-3-3-stabilized dimer". The authors should explain what they mean by "activation of the open monomer", while only RAF dimers are catalytically active (except for BRAF V600E mutant)?

For novelty b), the authors should explain more clearly what experimental results support this new model.

Author Response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

Tkacik et al describe their efforts to reconstitute and biochemically characterize ARAF, BRAF, and CRAF proteins and measure their ability to be paradoxically activated by current clinical and preclinical RAF inhibitors. Paradoxical activation of MAPK signaling is a major clinical problem plaguing current RAF inhibitors, and the mechanisms are complex and relatively poorly understood. The authors utilize their preparations of purified ARAF, BRAF, and CRAF kinase domains to measure paradoxical activation by type I and type II inhibitors, utilizing MEK protein as the substrate, and show that CRAF is activated in a similar fashion to BRAF, whereas ARAF appears resistant to activation. These data are analyzed using a simple cooperativity model with the goal of testing whether paradoxical activation involves negative cooperativity between RAF dimer binding sites, as has been previously reported. The authors conclude that it does not. They also test activation of B- and CRAF isoforms prepared in their full-length autoinhibited states and show that under the conditions of their assays, activation by inhibitors is not observed. In a particularly noteworthy part of the paper, the authors show that mutation of the N-terminal acidic (NtA) motif of ARAF and CRAF to match that of BRAF enhances paradoxical activation of CRAF and dramatically restores paradoxical activation of ARAF, which is not activated at all in its WT form, indicating a clear role for the NtA motif in the paradoxical activation mechanism. Additional experiments use mass photometry to measure BRAF dimer induction by inhibitors. The mass photometry measurements are a relatively novel way of achieving this, and the results are qualitatively consistent with previous studies that tracked BRAF dimerization in response to inhibitors using other methods. Overall, the paper establishes that WT CRAF is paradoxically activated by the same inhibitors that activate BRAF, and that ARAF contains the latent potential for activation that appears to be controlled by its NtA motif. The biochemical activation data for BRAF are qualitatively consistent with previous work.

Strengths:

While previous studies have put forward detailed molecular mechanisms for paradoxical activation of BRAF, comparatively little is known about the degree to which ARAF and CRAF are prone to this problem, and relatively little biochemical data of any sort are available for ARAF. Seen in this light, the current work should be considered of substantial potential significance for the RAF signaling field and for efforts to understand paradoxical activation and design new inhibitors that avoid it.

Weaknesses:

There are, unfortunately, some significant flaws in the data analysis and fitting of the RAF activation data that render the primary conclusion of the paper about the detailed activation mechanism, namely that it does not involve negative cooperativity between active sites, unjustified. This claim is made repeatedly throughout the manuscript, including in the title. Unfortunately, their data analysis approach is overly simplistic and does not probe this question thoroughly. This is the primary weakness of the study and should be addressed. A full biochemical modeling approach that accurately captures what is happening in the experiment needs to be applied in order for detailed inferences to be drawn about the mechanism beyond just the observation of activation.

The authors' analysis of their RAF:MEK "monomer" paradoxical activation data (Figures 1, 3, and Tables 1, 2) suffers from two fundamental flaws that render the resulting AC50/IC50 and cooperativity (Hill) parameters essentially uninterpretable. Without explaining or justifying their choice, the authors use a two-phase cooperative binding model from GraphPad Prism to fit their activation/inhibition data. This model is intended to describe cooperative ligand binding to multiple coupled sites within a preformed receptor assembly, and does not provide an adequate description of what is happening in this complicated experiment. Specifically, it has two fundamental flaws when applied to the analysis in question:

(a) It does not account for ligand depletion effects that occur with high-affinity drugs, and that profoundly affect the shapes of the dose-response curves, which are what are being fit 

The chosen model is one of a class of ligand-binding models that are derived by assuming that the free ligand concentration is effectively equal to the total ligand concentration. Under these conditions, binding curves have a characteristic steepness, and the presence of cooperativity can be inferred from changes in this steepness as described by a Hill coefficient. However, many RAF inhibitors, including most of the type II inhibitors in this study, bind to the dimerized forms of at least one of the RAF isoforms with ultra-high affinity in the picomolar range (particularly apparent in Figure 1 with LY inhibiting BRAF). Under these conditions, the model assumption is not valid. Instead, binding occurs in the high-affinity regime in which the drug titrates the receptor and effectively all the added drug molecules bind, so there is hardly any free ligand (see e.g. Jarmoskaite and Herschlag eLife 2020 for a full description of this "titration" regime). The shapes of the curves under these conditions reflect the total amount of RAF protein (and to some extent drug affinity), rather than the presence of cooperativity. Fitting dose response curves with the chosen model under these conditions will result in conflating binding affinity and protein concentration with cooperativity.

(b) It does not model the RAF monomer-dimer equilibrium, which is dramatically modulated by drug binding, rendering the results RAF-concentration dependent in a manner not accounted for by the analysis.

The chosen analysis model also fails to consider the monomer-dimer equilibrium of RAF. This has two ramifications. Since drug binding is coupled to dimerization to a very strong degree, the observed apparent affinities of drug binding (reflected in AC50 and IC50 values) are functions of the concentration of RAF molecules used in the experiment. Since dimerization affinities are likely different for ARAF, BRAF, and CRAF, the measured AC50 values also cannot be compared between isoforms. This concentration dependence is not addressed by the authors. A related issue is that the model assumes drug binding occurs to two coupled sites on preformed dimers, not to a mixture of monomers and dimers. "Cooperativity" parameters determined in this manner will reflect the shifting monomer-dimer equilibrium rather than the cooperativity within dimers. Additionally, the inhibition side of the activation/inhibition curves is driven by binding of the drug to the single remaining site on the dimer, not to two coupled sites, and so one cannot determine cooperativity values for this process in this manner.

As a result of both of these issues, the parameters reported in the tables do not correctly reflect cooperativity and cannot be used to infer the presence or absence of negative cooperativity between RAF dimer subunits. To address these major issues, the authors would need to apply a data analysis/fitting procedure that correctly models the biochemical interactions occurring in the sample, including both the monomer-dimer equilibrium and how this equilibrium is coupled to drug binding, such as that developed in e.g., Kholodenko Cell Reports 2015. Alternatively, the authors should remove the statements claiming a lack of negative cooperativity from the manuscript and alter the title to reflect this.

The bell-shaped dose response model that we employed models the sum of two dose-response curves – one that activates and one that inhibits. That is a simple way of capturing the essence of paradoxical activation -- the superposition of drug-induced activation at low inhibitor concentrations with inhibition at higher concentrations. That said, we agree completely with the reviewer that the model does not capture the complexity of what is happening in the experiment. We worked extensively with the Kholodenko model (which we implemented in Kintek Explorer), which accounts for the effect of drug on the monomer/dimer equilibrium and for the affinity of drug for each protomer of a dimer (and can therefore model positive or negative cooperativity as well as non-cooperative binding). We could obtain excellent fits with this model with positive cooperativity – perhaps not surprising considering that this is a 12 parameter model – with reasonable Kd values for drug binding and monomer/dimer equilibrium. However, we ultimately chose not to include this analysis when we realized that the fits were not at steady-state. The underlying Kon and Koff rates for the reasonable Kd’s for monomer/dimer formation were unreasonably slow. We could also obtain superficially reasonable fits with negative or non-cooperative binding, but close inspection revealed that they did not accurately fit the steepness of the inhibition phase of the dose-response curves for type II inhibitors. Even the Kholodenko model does not capture all the key aspects of our experiment. Perhaps most notably competition with ATP, the effect of ATP on the monomer dimer equilibrium, and the divergent conformations of the kinase required for binding ATP vs a type II inhibitor. We put some effort into explicitly including ATP in the model, but quickly decided that it was beyond our modeling expertise (and it also was not feasible to implement in Kintek explorer). In the end, we settled on the bell-shaped dose-response model because it was the simplest model that fit the data. We expect to include a supplemental figure/note in the revised manuscript to discuss our work with the Kholodenko model. We will also acknowledge the limitations of the bell-shaped dose response model.

This reviewer is also concerned that the steepness of the inhibition phase of the curves may be the result of enzyme-titration with these tight-binding inhibitors, rather than a result of positive cooperativity. We are reasonably sure that this is not the case. The shape of these curves and the IC50/AC50 values obtained is relatively insensitive to enzyme concentration, and we will include additional data in our revision to demonstrate this. Also, the steep hill slopes are unique to the type II inhibitors, which require a distinct inactive conformation of the kinase. Type I inhibitor SB590885 is similarly potent to the type II inhibitors, but does not exhibit this effect. If we were simply titrating enzyme, we would expect to see this with SB590885 as well.

Also, we will clarify in the revised manuscript that our interpretation of positive cooperativity of inhibition by type II inhibitors is also supported by our prior work with 14-3-3-bound RAF dimers (Tkacik et al, JBC 2025). This is a much simpler experiment, as dimers are pre-formed. We have now done a thorough study of the effect of enzyme concentration on the IC₅₀ and apparent cooperativity in dimer inhibition, which we will include in our revised manuscript. These experiments confirm that we are not in a regime where we are titrating enzyme.

As an aside, with respect to models that incorporate free inhibitor concentration, we did try to fit our 14-3-3-bound dimer inhibition data (in Tkacik et al, JBC 2025) with the Morrison equation for tight-binding inhibitors, which does take into account free ligand concentration. The fits were not reasonable with type II inhibitors, at least in part due to the non-ATP-competitive behavior of the type II drugs. Also the Morrison equation does not model cooperativity.

Some other points to consider

(1) The observation that ARAF is not activated by type II inhibitors is interesting. A detailed comparison of the activation magnitudes between inhibitors and between A-, B-, and CRAF is hampered by the arbitrary baseline signal in the assay, which arises from a non-zero FRET ratio in the absence of any RAF activity. The authors might consider background correcting their data using a calibration curve constructed using MEK samples of known degrees of phosphorylation, so that they can calculate turnover numbers and fold activation values rather than an increase over baseline. This will likely reveal that the activation effects are more substantial than they appear against the high background signal.

We will explore this for our revision.

(2) The authors note that full-length autoinhibited 14-3-3-bound RAF monomers are not activated by type I and II inhibitors. However, since this process involves the formation of a RAF dimer from two monomers, the process would also be expected to be concentration dependent, and the authors have only investigated this at a single protein concentration. Since disassembly of the autoinhibited state must also occur before dimerization, it might be expected to be kinetically disfavored as well. Have the authors tested this?

Good points. We have carried out this experiment at more than one enzyme concentration and differing reaction times, and also failed to see activation. However, we have not systematically explored either variable.

(3) ATP concentration modulates activation. While this is an interesting observation, some of this analysis suffers from the same issue discussed above, of not considering high-affinity binding effects. For instance, LY is not affected by ATP concentration in their data (Figure 4D), but this is easily explained as being due to its very tight binding affinity, resulting in titration of the receptor and the shape of the inhibition curve reflecting the amount of RAF kinase in the experiment and not the effective Kd or IC50 value.

As discussed above, we’ve convinced ourselves that we are not simply titrating enzyme. It occurred to us that such an effect could explain both the steepness of the inhibition curves with LY and other type II inhibitors and the apparent ATP-insensitivity. Our studies of concentration-dependence and the correlation of this effect with the type II binding mode argue against this possibility.

Finally, as an overarching comment to this Reviewer and the others, we understand well that our enzyme inhibition studies (here and in Tkacik 2025) do not rise to the level of a formal demonstration of cooperative ligand binding. We envision a future study in which we could address this directly, perhaps by using single molecule fluorescence to observe on/off rates for binding of fluorescently tagged inhibitors to immobilized RAF dimers. (This is clearly beyond the scope of the present work).

Reviewer #2 (Public review):

This manuscript by Tkacik et al. uses in vitro reconstituted systems to examine paradoxical activation across RAF isoforms and inhibitor classes. The authors conclude that paradoxical activation can be explained without invoking negative allostery and propose a general model in which ATP displacement from an "open monomer" promotes dimerization and activation. The biochemical work is technically sound, and the systematic comparison across RAF paralogs (along with mutational/functional analysis) across inhibitor classes is a strength.

However, the central mechanistic conclusions are overgeneralized relative to the experimental systems, and several key claims, particularly the dismissal of negative allostery and the proposed unifying model in Figure 6, are not directly supported by the data presented. Most importantly, the absence of RAS, membranes, and relevant regulatory context fundamentally limits the physiological relevance of several conclusions, especially regarding the current clinical type I.5 RAF inhibitors and paradoxical activation.

Overall, this is a potentially valuable biochemical study, but the manuscript would benefit from more restrained interpretation, clearer framing of scope, and revisions to the model and title to better reflect what is actually tested.

(1) A central issue is that the biochemical system lacks RAS, membranes, 14-3-3 and endogenous regulatory factors that are known to be required for paradoxical RAF and MAPK activation in cells. As previous work has repeatedly shown and the authors also acknowledge, paradoxical activation by RAF inhibitors is RAS-dependent in cells, and this dependence presumably explains why full-length autoinhibited RAF complexes are refractory to activation in the authors' assays.

Importantly, the absence of paradoxical activation by type I.5 inhibitors in this system is therefore not mechanistically informative. Type I.5 inhibitors (e.g., vemurafenib, dabrafenib, encorafenib), but not Paradox Breakers (e.g., plixorafenib), robustly induce paradoxical activation in cells because binding of the inhibitor to inactive cytosolic RAF monomer promotes a conformational change that drives RAF recruitment to RAS in the membrane, promoting dimerization. The inability of the type 1.5 inhibitor to suppress the newly formed dimers is the basis of the pronounced paradoxical activation in cells. In the absence of RAS and membrane recruitment, failure to observe paradoxical activation in vitro does not distinguish between competing mechanistic models.

As a result, conclusions regarding inhibitor class differences, and especially the generality of the proposed model, should be substantially tempered.

We will emphasize the limitations of our highly simplified experimental system in the revised manuscript, and temper some of our interpretations. And while the lack of membranes/RAS/14-3-3 in our system and the lack of observed PA with type I.5 inhibitors is a limitation of our study, we disagree that it renders our study of type I.5 inhibitors mechanistically uninformative. As seen here and consistent with prior studies, the binding mode of these compounds disfavors formation of the kinase dimer. While this may be overcome by 14-3-3 binding and other effects in the cellular context, it reflects a fundamental mechanistic difference as compared with type I and type II inhibitors, which also exhibit paradoxical activation.

(2) The authors argue that their data argue against negative allostery as a central feature of paradoxical activation. However, the presented data do not directly test negative allostery, nor do they exclude it. The biochemical assays do not recreate the cellular context in which negative allostery has been inferred. Further, structural data showing asymmetric inhibitor occupancy in RAF dimers cannot be dismissed on the basis of alternative symmetric structures alone, particularly given the dynamic nature of RAF dimers in cells.

Most importantly, negative allostery was proposed to explain paradoxical activation by Type I.5 RAF inhibitors, yet these inhibitors do not paradoxically activate in the assays presented here. The absence of paradoxical activation in this system, therefore, cannot be used to argue against a mechanism that is specifically invoked to explain cellular behavior not recapitulated by the assay.

To be clear, we are not dismissing the possibility of negative cooperativity. And we do not think of our model as an alternative to the negative cooperativity model – rather it is a generalization that can account for paradoxical activation by diverse inhibitor classes, irrespective of positive, negative or non-cooperative modes of inhibition. We will emphasize these points in the revised manuscript.

If negative allostery were a requisite feature of PA, we would not expect to see PA with type II inhibitors. As discussed in our response to Reviewer 1, we see clear evidence of positively cooperative inhibition of 14-3-3-bound RAF dimers by type II inhibitors (Tkacik JBC 2025) and in the present study, we find clear paradoxical activation by type II inhibitors (and there are many reports in the literature of PA by type II inhibitors in cellular contexts).

(3) The model presented in Figure 6 is conceptually possible but remains speculative. Key elements of the model, including RAS engagement, membrane recruitment, 14-3-3 rearrangements, and the involvement of cellular kinases and phosphatases, are explicitly absent from the experimental system. Accordingly, the model is not tested by the data presented and should not be framed as a validated or general mechanism. The figure and accompanying text should be clearly labeled as a working or conceptual model rather than a mechanistically supported conclusion.

We will revise the text to more clearly reflect that this is a working model, and importantly, that it is based on a large literature in this area in addition to the relevant experimental work in this manuscript.

(4) The manuscript states that type I.5 inhibitors do not induce paradoxical activation in the biochemical assay because their C-helix-out binding mode disfavors dimerization. While this is true in isolation, it overlooks the well-established fact that type I.5 inhibitors (with the exception of paradox breakers) clearly promote RAS-dependent RAF dimerization in cells. This distinction is critical and should be explicitly acknowledged when interpreting the in vitro findings.

We will explicitly make this point in the revised manuscript.

(5) The title suggests a general mechanism for paradoxical activation across RAF isoforms and inhibitor classes, whereas the data primarily address type I and type II inhibitors acting on isolated kinase-domain monomers. A more accurate framing would avoid the term "general" and confine the conclusions to C-helix-in (type I/II) RAF inhibitors in a reduced biochemical context.

As noted above, and in our response to Reviewer 3 below, we will clarify the contribution of data in present manuscript to the model and that it is based more broadly on the literature on PA and our insights into RAF structure and regulation. We will also revise the title to avoid the implication that the model arises mainly from the experimental data in the manuscript.

Reviewer #3 (Public review):

Summary:

Tkacik et al. systematically characterized all three RAF kinase isoforms in vitro with all three types of RAF inhibitors (Type I, I1/2, and II) to investigate the mechanism underlying paradoxical activation.

In this study, the authors reconstituted heterodimers of A-, B-, and C-RAF kinase domains bound to non-phosphorylable MEK1 (SASA), mimicking the monomeric auto-inhibited state of RAF. These "RAF monomers" were tested for MEK phosphorylation with an increasing concentration of all three types of RAF inhibitors (Type I, I1/2, and II). This study is reminiscent of a previous study of the same team measuring RAF kinase activity in the presence of all three types of inhibitors in the context of dimeric RAF isoforms stabilized by 14-3-3 proteins (Tkacik et al 2025 JBC). RAF monomers had little to no activity at low concentrations of inhibitors (consistent with their "monomeric state"). Addition of type I1/2 inhibitor did not induce paradoxical activation as, in this context, they do not induce RAF dimerization required for activation, as observed by MP. Addition of type I and type II inhibitors led to paradoxical activation consistent with the RAF dimerization induced by these inhibitors, as observed by MP. Interestingly, type II inhibitors induced activation only for B- and C-RAF and not A-RAF.

At high concentrations of type II inhibitors, kinase activity is inhibited with a strong or weak positive cooperativity for BRAF and CRAF, respectively. This observation is very similar to what the authors previously observed with their dimeric RAF system. Interestingly, when the NtA motif is modified by phosphomimetic mutations in A- and C-Raf, basal kinase activity is stronger, but most importantly, inhibitor-induced paradoxical activation is much stronger with both type I and II inhibitors. This demonstrates that mutation of the NtA motif of ARAF and CRAF sensitized them to paradoxical activation by type II inhibitors.

The authors also tested the effect of ATP in the paradoxical activation observed in their RAF "monomer" system. As previously published in their assay with 14-3-3 stabilized dimeric RAF, the authors observed an expected shift of the IC50 with Type I inhibitors, while Type II inhibitors seem to behave as a non-competitive inhibitor. The authors next reconstituted the MAP kinase pathway (with RAF monomers at the top of the phosphorylation cascade) to test paradoxical activation amplification. Again, Type I1/2 inhibitors did not induce paradoxical activation, while Type I and II inhibitors did. The authors tested the inhibitors with FL auto-inhibited RAF/MEK/14-3-3 complexes, where, contrary to the "RAF monomers" experiments, FL B- and C-RAF were not paradoxically activated but were inhibited by all three types of inhibitors.

Overall, Tkacik et al. tackle an important question in the field for which definitive experiments and thorough biochemical investigation to understand the molecular mechanisms for the inhibitor-induced paradoxical activation are still missing, and of high importance for future drug development.

Strengths:

The biochemical experiments here are rigorously executed, and the results obtained are highly informative in the field to decipher the intricate mechanisms of RAF activation and inhibitor-induced paradoxical activation.

Weaknesses:

The interpretation of the results in the context of the current state of the art is ambiguous and raises questions about the relevance of introducing a new model for inhibitor-induced paradoxical activation, particularly since the findings presented here do not clearly contradict established paradigms. I believe some clarification and precision are required.

While our model does not conflict with established paradigms (because it can allow for negative cooperativity) our experimental findings (here and in Tkacik et al JBC 2025) are in conflict with the negative allostery model. We will work to clarify this in the revised manuscript.

Main comments:

(1) Figure 2:

The authors comment on the expected greater increase (for a cascade assay) in the magnitude of ERK phosphorylation compared to what was observed for MEK phosphorylation. However, this observation might be reflective of the stoichiometries used in the assay, with 40 times more MEK compared to RAF concentration (250nm vs 6nM), which might favour pERK vs pMEK.

The authors should clarify their rationale for the protein concentration used in this assay and explain how protein stoichiometry was taken into account for the interpretation of their results.

The Reviewer makes a good point, the concentrations and ratios chosen are expected to make a substantial difference in observed amplification. We intended this experiment more as a qualitative demonstration of cascade amplification and will clarify this in the revised manuscript.

In addition, the authors should justify comparing pMEK and pERK TR-FRET values when different anti-phospho antibodies were used. Antibodies may have distinct binding affinities for their epitopes. Could this not lead to differences in FRET signal amplitudes that complicate direct comparison?

Also a good point, we will note this limitation in the revised manuscript.

(2) Supplementary Figure 2:

The author mentioned that the inhibitors did not activate the FL auto-inhibited RAF complexes; however, they did inhibit the TR-FRET signal.

Can the authors comment on the origin of the observed basal activity? Would the authors expect self-release of the RAF kinase protein from the auto-inhibited state in the absence of RAS, leading to dimerization and activation? Alternatively, do the inhibitors at low-concentration relieve the auto-inhibited state, thereby driving dimerization and activation?

We think that the baseline activity that is being inhibited is due to low concentrations of active dimer in our autoinhibited state preparations.

Did the author test the addition of RAS protein in their in vitro system to determine whether "soluble" RAS is sufficient to release the protective interactions with RBD/CRD/14-3-3 and lead to inhibitor-induced paradoxical activation of FL RAF?

We did not, but we’ve thought about it. We expect that soluble RAS would not be activating. We have previously carried our extensive studies of BRAF activation by soluble vs. farnesylated RAS in a membrane environment (liposomes) and observed partial activation in the latter (Park et al, Nature Communications 2023).

(3) Figure 5B:

The authors said that the Kd values obtained from their MP assay are consistent with prior studies of RAF homodimerization and RAF:MEK heterodimerization. While this is true from the previous studies of RAF:MEK interaction by BLI (performed from the same team), the Kd of isolated RAF kinase homodimerization has been measured around ~30µM by AUC in the cited ref (24,27 & 37).

The authors should discuss the discrepancy between their Kd of homodimerization and the reported Kd values in the literature. At the concentration used for MP, it is surprising to observe RAF dimerization while the Kd of homodimerization has been measured at ~30µM (in the absence of MEK).

We will cite/discuss these differences in our revised manuscript.

Would the authors expect the presence of MEK to influence the homodimerization affinity for the isolated KD?

Perhaps, but likely only modestly. We do not think this explains the discrepancy noted above.

(4) Conclusions:

Several times in the introduction and the conclusion, the authors suggest that the negative allostery model (where "inhibitor binding to one protomer of the dimer promotes an active but inhibitor-resistant conformation in the other") is a model that applies to all types of RAF inhibitors (I, I1/2, and II).

However, from my understanding and all the references cited by the authors, this model only applies to type I1/2 inhibitors, where indeed the aC IN conformation in the second (inhibitor-free) protomer of the RAF dimer might be incompatible with the type I1/2 inhibitors inducing aC OUT conformation. The type I and type II inhibitors are aC IN inhibitors and are expected to bind both protomers from RAF dimers with similar affinities. Therefore, the negative allostery model does not apply to the type I and type II inhibitors. The difference in the mechanism of action of inhibitors is even used to explain the difference in the concentration range in which inhibitor-induced activation is observed in cells. The description of the state of the art in this study is confusing and does not help to properly understand their argumentation to revise the established model for paradoxical RAF activation.

We will work to clarify these complicated issues in the revised manuscript. While the reviewer is correct that the negative allostery model was developed in the context of Type 1.5 inhibitors, there are many examples in the literature of it being used to explain PA by type I and type II inhibitors as well.

Can the authors clarify their analysis of the state of the art on the different mechanisms of action for the paradoxical activation of RAF by the different types of RAF inhibitors?

We’ll try!

5) Conclusions:

"Our results suggest that negative allostery (or negative cooperativity) is not a requisite feature of paradoxical activation. The type I and type II inhibitors studied here induce RAF dimers and exhibit paradoxical activation but do so without evidence of negative cooperativity, nor do they appear to inhibit intentionally engineered RAF dimers with negative cooperativity (25). Indeed, type II inhibitors exhibit apparent positive cooperativity while type I inhibitors are non-cooperative inhibitors of RAF dimers (25)."

Can the authors explain how results on the paradoxical activation induced by type I and type II inhibitors inform or challenge a model that specifically applies to type I1/2 inhibitors?

As noted above, the negative allostery model has also been widely applied irrespective of inhibitor type (rightly or wrongly). Essentially any review or discussion of the topic will explain in one way or another how inhibitor binding to one side of a dimer leaves the opposite side active but resistant to inhibitor. Our model is agnostic with respect to cooperativity of inhibition – essentially we are pointing out a simple circumstance that seems to have been lost in the focus on negative allostery. Paradoxical activation is a result of drug action on RAF monomers, while inhibition is a result of drug action on RAF dimers. Because these are distinct molecular species/complexes, they can be expected to differ in their affinity for RAF inhibitors, irrespective of type. Because binding of ATP in the active site of RAF monomers stabilizes the inactive monomeric state, displacing ATP can promote activation/dimerization. For any inhibitor that is more potent at displacing ATP from a monomer that from an active dimer, we could expect to observe a window of paradoxical activation.

The authors often refer to their previous study (reference 25), where they tested the inhibition of all three types of inhibitors with engineered RAF dimers. While I agree with the authors that in reference 25 the Type I and type II inhibitors inhibit RAF dimers without exhibiting negative cooperativity (as expected from the literature and the current model), the authors did observe some negative cooperativity for Type I1/2 inhibitors in their study most particularly for the type I1/2 PB (with hill slope ranging from -0.4 to -0.9, indicative of negative cooperativity).

Correct! Although we do note the caveat that weak inhibition can also give rise to apparent negative cooperativity.

While the observations that type II inhibitors display positive cooperativity is both novel and very interesting, from what I understand the results from thakick et al 2025 and the current study appear more in line with the current paradigm in the field (which describe paradoxical activation with negative cooperativity for type I1/2 inhibitors and no negative cooperativity for the Type I and II inhibitors) rather than disapproving of the current model and supporting for a new model. 

In this context, can the authors clarify how their results challenge the current model for paradoxical activation?

While the difference in binding modes and structural effects of type I.5 vs type I and type II inhibitors are well known in the field, we do not know of any work that suggests paradoxical activation arises from anything other than negative allostery. As one example to the contrary, Rasmussen et al. observe allosteric coupling asymmetry in binding of type II inhibitors to BRAF and attribute the observed paradoxical activation to “induction of dimers with one inhibited and one catalytically active subunit” (Rasmussen et al., Elife 2024). They also studied type I inhibitors in this work, but did not observe paradoxical activation.

(6) Conclusions:

The authors describe the JAB34 experiment from Poulikakos et al. 2010 to conclude that "While this experiment cleanly demonstrates inhibitor-induced transactivation of RAF dimers, it is important to recognize that the differential inhibitor sensitivity of the two subunits in this experiment is artificial - it is engineered rather than induced by inhibitor binding as the negative allostery model proposes."

Indeed, the JAB34 experiment demonstrated the inhibitor-induced transactivation, but the Poulikakos et al. 2010 study does not discuss differential inhibitor sensitivity. The negative allostery model was proposed later by poulikakos team in other papers (Yao et al 2015 and Karoulia et al, 2016), in which JAB34 was not used.

Can the authors clarify how the JAB34 experiments question differential inhibitor sensitivity?

Good point, we neglected to discuss the Yao and Karoulia papers and will do so in our revised manuscript.

(7) Conclusions:

"Considering that the conformation required for binding of type I.5 inhibitors destabilizes RAF dimers, it is unclear how an inhibitor binding to one protomer would be able to transmit an allosteric change to the opposite protomer, if that inhibitor's binding causes the existing dimer to dissociate."

The authors should comment on whether 14-3-3 proteins might overcome negative regulation by type I1/2 inhibitors, similar to what has been shown for ATP, which acts as a dimer breaker like type I1/2 inhibitors.

Certainly we expect that they will, and we will discuss this in our revised manuscript.

(8) Conclusions:

"Furthermore, the complex effects of type I.5 inhibitors on dimer stability and the clear resistance of active RAF dimers to these inhibitors complicates interpretation of inhibition data - weak or incomplete inhibition of an enzyme can be difficult to discern from true negative cooperativity (43). As we discuss below, the clear resistance of RAF dimers to type I.5 inhibitors is alone sufficient to explain their ineffective inhibition during paradoxical activation, without invoking negative allostery." 

The authors should explain how they reconcile this statement and their proposal of a new model that does not rely on negative allostery with their previous findings showing negative cooperativity for RAF dimer inhibition with type I1/2 inhibitors.

As discussed above and in responses to other Reviewers, we do not exclude negative cooperativity for Type I.5 inhibitors. That said, we are skeptical, even in light of our own findings of apparent negative cooperativity by type 1.5 compounds, due in part to the caveats the reviewer highlights above.

(9) Conclusions:

Here, the authors propose a new universal model to explain paradoxical activation of RAF by all types of RAF inhibitors:

" Our findings here, in light of structural studies of RAF complexes and prior cellular investigations of paradoxical activation, lead us to a model for paradoxical activation that does not rely on negative allostery and is consistent with activation by diverse inhibitor classes. In this model, the open monomer complex is the target of inhibitor-induced paradoxical activation (Figure 6). Binding of ATP to the RAF active site stabilizes the inactive conformation of the open monomer, which disfavors dimerization. Displacement of ATP by an ATP-competitive inhibitor, irrespective of class, alters the relative N- and C-lobe orientations of the kinase to promote dimerization (30, 35). Once dimerized, inhibitor dissociation from one or both sides of the dimer would allow phosphorylation and activation of MEK."

From my understanding, the novelty of this new model is twofold: a) the open monomer is the target of the inhibitor-induced paradoxical activation and b) once dimerized, inhibitor dissociation from one or both sides of the dimer would allow phosphorylation and activation of MEK.

Novelty a) implies, as the authors stated, that "Inhibitor-induced activation and inhibition act on distinct species - activation on the open monomer and inhibition on the 14-3-3-stabilized dimer". The authors should explain what they mean by "activation of the open monomer", while only RAF dimers are catalytically active (except for BRAF V600E mutant)?

We will clarify – by activation we mean promoting conversion of the open monomer to a dimer.

For novelty b), the authors should explain more clearly what experimental results support this new model.

We will more explicitly detail how our results here as well as prior work in the field support this model.